MyArxiv
Robotics 46
☆ ViTa-Zero: Zero-shot Visuotactile Object 6D Pose Estimation ICRA 2025
Object 6D pose estimation is a critical challenge in robotics, particularly for manipulation tasks. While prior research combining visual and tactile (visuotactile) information has shown promise, these approaches often struggle with generalization due to the limited availability of visuotactile data. In this paper, we introduce ViTa-Zero, a zero-shot visuotactile pose estimation framework. Our key innovation lies in leveraging a visual model as its backbone and performing feasibility checking and test-time optimization based on physical constraints derived from tactile and proprioceptive observations. Specifically, we model the gripper-object interaction as a spring-mass system, where tactile sensors induce attractive forces, and proprioception generates repulsive forces. We validate our framework through experiments on a real-world robot setup, demonstrating its effectiveness across representative visual backbones and manipulation scenarios, including grasping, object picking, and bimanual handover. Compared to the visual models, our approach overcomes some drastic failure modes while tracking the in-hand object pose. In our experiments, our approach shows an average increase of 55% in AUC of ADD-S and 60% in ADD, along with an 80% lower position error compared to FoundationPose.
comment: Accepted by ICRA 2025
☆ Novel Demonstration Generation with Gaussian Splatting Enables Robust One-Shot Manipulation
Visuomotor policies learned from teleoperated demonstrations face challenges such as lengthy data collection, high costs, and limited data diversity. Existing approaches address these issues by augmenting image observations in RGB space or employing Real-to-Sim-to-Real pipelines based on physical simulators. However, the former is constrained to 2D data augmentation, while the latter suffers from imprecise physical simulation caused by inaccurate geometric reconstruction. This paper introduces RoboSplat, a novel method that generates diverse, visually realistic demonstrations by directly manipulating 3D Gaussians. Specifically, we reconstruct the scene through 3D Gaussian Splatting (3DGS), directly edit the reconstructed scene, and augment data across six types of generalization with five techniques: 3D Gaussian replacement for varying object types, scene appearance, and robot embodiments; equivariant transformations for different object poses; visual attribute editing for various lighting conditions; novel view synthesis for new camera perspectives; and 3D content generation for diverse object types. Comprehensive real-world experiments demonstrate that RoboSplat significantly enhances the generalization of visuomotor policies under diverse disturbances. Notably, while policies trained on hundreds of real-world demonstrations with additional 2D data augmentation achieve an average success rate of 57.2%, RoboSplat attains 87.8% in one-shot settings across six types of generalization in the real world.
comment: Published at Robotics: Science and Systems (RSS) 2025
☆ A New Semidefinite Relaxation for Linear and Piecewise-Affine Optimal Control with Time Scaling
We introduce a semidefinite relaxation for optimal control of linear systems with time scaling. These problems are inherently nonconvex, since the system dynamics involves bilinear products between the discretization time step and the system state and controls. The proposed relaxation is closely related to the standard second-order semidefinite relaxation for quadratic constraints, but we carefully select a subset of the possible bilinear terms and apply a change of variables to achieve empirically tight relaxations while keeping the computational load light. We further extend our method to handle piecewise-affine (PWA) systems by formulating the PWA optimal-control problem as a shortest-path problem in a graph of convex sets (GCS). In this GCS, different paths represent different mode sequences for the PWA system, and the convex sets model the relaxed dynamics within each mode. By combining a tight convex relaxation of the GCS problem with our semidefinite relaxation with time scaling, we can solve PWA optimal-control problems through a single semidefinite program.
☆ RUKA: Rethinking the Design of Humanoid Hands with Learning
Dexterous manipulation is a fundamental capability for robotic systems, yet progress has been limited by hardware trade-offs between precision, compactness, strength, and affordability. Existing control methods impose compromises on hand designs and applications. However, learning-based approaches present opportunities to rethink these trade-offs, particularly to address challenges with tendon-driven actuation and low-cost materials. This work presents RUKA, a tendon-driven humanoid hand that is compact, affordable, and capable. Made from 3D-printed parts and off-the-shelf components, RUKA has 5 fingers with 15 underactuated degrees of freedom enabling diverse human-like grasps. Its tendon-driven actuation allows powerful grasping in a compact, human-sized form factor. To address control challenges, we learn joint-to-actuator and fingertip-to-actuator models from motion-capture data collected by the MANUS glove, leveraging the hand's morphological accuracy. Extensive evaluations demonstrate RUKA's superior reachability, durability, and strength compared to other robotic hands. Teleoperation tasks further showcase RUKA's dexterous movements. The open-source design and assembly instructions of RUKA, code, and data are available at https://ruka-hand.github.io/.
comment: Website at https://ruka-hand.github.io/
☆ Long Range Navigator (LRN): Extending robot planning horizons beyond metric maps
A robot navigating an outdoor environment with no prior knowledge of the space must rely on its local sensing to perceive its surroundings and plan. This can come in the form of a local metric map or local policy with some fixed horizon. Beyond that, there is a fog of unknown space marked with some fixed cost. A limited planning horizon can often result in myopic decisions leading the robot off course or worse, into very difficult terrain. Ideally, we would like the robot to have full knowledge that can be orders of magnitude larger than a local cost map. In practice, this is intractable due to sparse sensing information and often computationally expensive. In this work, we make a key observation that long-range navigation only necessitates identifying good frontier directions for planning instead of full map knowledge. To this end, we propose Long Range Navigator (LRN), that learns an intermediate affordance representation mapping high-dimensional camera images to `affordable' frontiers for planning, and then optimizing for maximum alignment with the desired goal. LRN notably is trained entirely on unlabeled ego-centric videos making it easy to scale and adapt to new platforms. Through extensive off-road experiments on Spot and a Big Vehicle, we find that augmenting existing navigation stacks with LRN reduces human interventions at test-time and leads to faster decision making indicating the relevance of LRN. https://personalrobotics.github.io/lrn
comment: 10 pages, 9 figures
☆ Force and Speed in a Soft Stewart Platform
Many soft robots struggle to produce dynamic motions with fast, large displacements. We develop a parallel 6 degree-of-freedom (DoF) Stewart-Gough mechanism using Handed Shearing Auxetic (HSA) actuators. By using soft actuators, we are able to use one third as many mechatronic components as a rigid Stewart platform, while retaining a working payload of 2kg and an open-loop bandwidth greater than 16Hx. We show that the platform is capable of both precise tracing and dynamic disturbance rejection when controlling a ball and sliding puck using a Proportional Integral Derivative (PID) controller. We develop a machine-learning-based kinematics model and demonstrate a functional workspace of roughly 10cm in each translation direction and 28 degrees in each orientation. This 6DoF device has many of the characteristics associated with rigid components - power, speed, and total workspace - while capturing the advantages of soft mechanisms.
comment: Published at Robosoft 2025
☆ Uncertainty-Aware Trajectory Prediction via Rule-Regularized Heteroscedastic Deep Classification
Deep learning-based trajectory prediction models have demonstrated promising capabilities in capturing complex interactions. However, their out-of-distribution generalization remains a significant challenge, particularly due to unbalanced data and a lack of enough data and diversity to ensure robustness and calibration. To address this, we propose SHIFT (Spectral Heteroscedastic Informed Forecasting for Trajectories), a novel framework that uniquely combines well-calibrated uncertainty modeling with informative priors derived through automated rule extraction. SHIFT reformulates trajectory prediction as a classification task and employs heteroscedastic spectral-normalized Gaussian processes to effectively disentangle epistemic and aleatoric uncertainties. We learn informative priors from training labels, which are automatically generated from natural language driving rules, such as stop rules and drivability constraints, using a retrieval-augmented generation framework powered by a large language model. Extensive evaluations over the nuScenes dataset, including challenging low-data and cross-location scenarios, demonstrate that SHIFT outperforms state-of-the-art methods, achieving substantial gains in uncertainty calibration and displacement metrics. In particular, our model excels in complex scenarios, such as intersections, where uncertainty is inherently higher. Project page: https://kumarmanas.github.io/SHIFT/.
comment: 17 Pages, 9 figures. Accepted to Robotics: Science and Systems(RSS), 2025
☆ Imperative MPC: An End-to-End Self-Supervised Learning with Differentiable MPC for UAV Attitude Control
Modeling and control of nonlinear dynamics are critical in robotics, especially in scenarios with unpredictable external influences and complex dynamics. Traditional cascaded modular control pipelines often yield suboptimal performance due to conservative assumptions and tedious parameter tuning. Pure data-driven approaches promise robust performance but suffer from low sample efficiency, sim-to-real gaps, and reliance on extensive datasets. Hybrid methods combining learning-based and traditional model-based control in an end-to-end manner offer a promising alternative. This work presents a self-supervised learning framework combining learning-based inertial odometry (IO) module and differentiable model predictive control (d-MPC) for Unmanned Aerial Vehicle (UAV) attitude control. The IO denoises raw IMU measurements and predicts UAV attitudes, which are then optimized by MPC for control actions in a bi-level optimization (BLO) setup, where the inner MPC optimizes control actions and the upper level minimizes discrepancy between real-world and predicted performance. The framework is thus end-to-end and can be trained in a self-supervised manner. This approach combines the strength of learning-based perception with the interpretable model-based control. Results show the effectiveness even under strong wind. It can simultaneously enhance both the MPC parameter learning and IMU prediction performance.
comment: 14 pages, 3 figures, accepted by L4DC 2025
☆ RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins CVPR 2025
In the rapidly advancing field of robotics, dual-arm coordination and complex object manipulation are essential capabilities for developing advanced autonomous systems. However, the scarcity of diverse, high-quality demonstration data and real-world-aligned evaluation benchmarks severely limits such development. To address this, we introduce RoboTwin, a generative digital twin framework that uses 3D generative foundation models and large language models to produce diverse expert datasets and provide a real-world-aligned evaluation platform for dual-arm robotic tasks. Specifically, RoboTwin creates varied digital twins of objects from single 2D images, generating realistic and interactive scenarios. It also introduces a spatial relation-aware code generation framework that combines object annotations with large language models to break down tasks, determine spatial constraints, and generate precise robotic movement code. Our framework offers a comprehensive benchmark with both simulated and real-world data, enabling standardized evaluation and better alignment between simulated training and real-world performance. We validated our approach using the open-source COBOT Magic Robot platform. Policies pre-trained on RoboTwin-generated data and fine-tuned with limited real-world samples demonstrate significant potential for enhancing dual-arm robotic manipulation systems by improving success rates by over 70% for single-arm tasks and over 40% for dual-arm tasks compared to models trained solely on real-world data.
comment: CVPR 2025 Highlight. 22 pages. Project page: https://robotwin-benchmark.github.io/
☆ Adaptive Task Space Non-Singular Terminal Super-Twisting Sliding Mode Control of a 7-DOF Robotic Manipulator
This paper presents a new task-space Non-singular Terminal Super-Twisting Sliding Mode (NT-STSM) controller with adaptive gains for robust trajectory tracking of a 7-DOF robotic manipulator. The proposed approach addresses the challenges of chattering, unknown disturbances, and rotational motion tracking, making it suited for high-DOF manipulators in dexterous manipulation tasks. A rigorous boundedness proof is provided, offering gain selection guidelines for practical implementation. Simulations and hardware experiments with external disturbances demonstrate the proposed controller's robust, accurate tracking with reduced control effort under unknown disturbances compared to other NT-STSM and conventional controllers. The results demonstrated that the proposed NT-STSM controller mitigates chattering and instability in complex motions, making it a viable solution for dexterous robotic manipulations and various industrial applications.
comment: 10 pages, 8 figures
☆ Krysalis Hand: A Lightweight, High-Payload, 18-DoF Anthropomorphic End-Effector for Robotic Learning and Dexterous Manipulation
This paper presents the Krysalis Hand, a five-finger robotic end-effector that combines a lightweight design, high payload capacity, and a high number of degrees of freedom (DoF) to enable dexterous manipulation in both industrial and research settings. This design integrates the actuators within the hand while maintaining an anthropomorphic form. Each finger joint features a self-locking mechanism that allows the hand to sustain large external forces without active motor engagement. This approach shifts the payload limitation from the motor strength to the mechanical strength of the hand, allowing the use of smaller, more cost-effective motors. With 18 DoF and weighing only 790 grams, the Krysalis Hand delivers an active squeezing force of 10 N per finger and supports a passive payload capacity exceeding 10 lbs. These characteristics make Krysalis Hand one of the lightest, strongest, and most dexterous robotic end-effectors of its kind. Experimental evaluations validate its ability to perform intricate manipulation tasks and handle heavy payloads, underscoring its potential for industrial applications as well as academic research. All code related to the Krysalis Hand, including control and teleoperation, is available on the project GitHub repository: https://github.com/Soltanilara/Krysalis_Hand
☆ Taccel: Scaling Up Vision-based Tactile Robotics via High-performance GPU Simulation
Tactile sensing is crucial for achieving human-level robotic capabilities in manipulation tasks. VBTSs have emerged as a promising solution, offering high spatial resolution and cost-effectiveness by sensing contact through camera-captured deformation patterns of elastic gel pads. However, these sensors' complex physical characteristics and visual signal processing requirements present unique challenges for robotic applications. The lack of efficient and accurate simulation tools for VBTS has significantly limited the scale and scope of tactile robotics research. Here we present Taccel, a high-performance simulation platform that integrates IPC and ABD to model robots, tactile sensors, and objects with both accuracy and unprecedented speed, achieving an 18-fold acceleration over real-time across thousands of parallel environments. Unlike previous simulators that operate at sub-real-time speeds with limited parallelization, Taccel provides precise physics simulation and realistic tactile signals while supporting flexible robot-sensor configurations through user-friendly APIs. Through extensive validation in object recognition, robotic grasping, and articulated object manipulation, we demonstrate precise simulation and successful sim-to-real transfer. These capabilities position Taccel as a powerful tool for scaling up tactile robotics research and development. By enabling large-scale simulation and experimentation with tactile sensing, Taccel accelerates the development of more capable robotic systems, potentially transforming how robots interact with and understand their physical environment.
comment: 17 pages, 7 figures
☆ 3D-PNAS: 3D Industrial Surface Anomaly Synthesis with Perlin Noise
Large pretrained vision foundation models have shown significant potential in various vision tasks. However, for industrial anomaly detection, the scarcity of real defect samples poses a critical challenge in leveraging these models. While 2D anomaly generation has significantly advanced with established generative models, the adoption of 3D sensors in industrial manufacturing has made leveraging 3D data for surface quality inspection an emerging trend. In contrast to 2D techniques, 3D anomaly generation remains largely unexplored, limiting the potential of 3D data in industrial quality inspection. To address this gap, we propose a novel yet simple 3D anomaly generation method, 3D-PNAS, based on Perlin noise and surface parameterization. Our method generates realistic 3D surface anomalies by projecting the point cloud onto a 2D plane, sampling multi-scale noise values from a Perlin noise field, and perturbing the point cloud along its normal direction. Through comprehensive visualization experiments, we demonstrate how key parameters - including noise scale, perturbation strength, and octaves, provide fine-grained control over the generated anomalies, enabling the creation of diverse defect patterns from pronounced deformations to subtle surface variations. Additionally, our cross-category experiments show that the method produces consistent yet geometrically plausible anomalies across different object types, adapting to their specific surface characteristics. We also provide a comprehensive codebase and visualization toolkit to facilitate future research.
☆ Versatile, Robust, and Explosive Locomotion with Rigid and Articulated Compliant Quadrupeds
Achieving versatile and explosive motion with robustness against dynamic uncertainties is a challenging task. Introducing parallel compliance in quadrupedal design is deemed to enhance locomotion performance, which, however, makes the control task even harder. This work aims to address this challenge by proposing a general template model and establishing an efficient motion planning and control pipeline. To start, we propose a reduced-order template model-the dual-legged actuated spring-loaded inverted pendulum with trunk rotation-which explicitly models parallel compliance by decoupling spring effects from active motor actuation. With this template model, versatile acrobatic motions, such as pronking, froggy jumping, and hop-turn, are generated by a dual-layer trajectory optimization, where the singularity-free body rotation representation is taken into consideration. Integrated with a linear singularity-free tracking controller, enhanced quadrupedal locomotion is achieved. Comparisons with the existing template model reveal the improved accuracy and generalization of our model. Hardware experiments with a rigid quadruped and a newly designed compliant quadruped demonstrate that i) the template model enables generating versatile dynamic motion; ii) parallel elasticity enhances explosive motion. For example, the maximal pronking distance, hop-turn yaw angle, and froggy jumping distance increase at least by 25%, 15% and 25%, respectively; iii) parallel elasticity improves the robustness against dynamic uncertainties, including modelling errors and external disturbances. For example, the allowable support surface height variation increases by 100% for robust froggy jumping.
comment: 20 pages, 25 figures
☆ UncAD: Towards Safe End-to-end Autonomous Driving via Online Map Uncertainty
End-to-end autonomous driving aims to produce planning trajectories from raw sensors directly. Currently, most approaches integrate perception, prediction, and planning modules into a fully differentiable network, promising great scalability. However, these methods typically rely on deterministic modeling of online maps in the perception module for guiding or constraining vehicle planning, which may incorporate erroneous perception information and further compromise planning safety. To address this issue, we delve into the importance of online map uncertainty for enhancing autonomous driving safety and propose a novel paradigm named UncAD. Specifically, UncAD first estimates the uncertainty of the online map in the perception module. It then leverages the uncertainty to guide motion prediction and planning modules to produce multi-modal trajectories. Finally, to achieve safer autonomous driving, UncAD proposes an uncertainty-collision-aware planning selection strategy according to the online map uncertainty to evaluate and select the best trajectory. In this study, we incorporate UncAD into various state-of-the-art (SOTA) end-to-end methods. Experiments on the nuScenes dataset show that integrating UncAD, with only a 1.9% increase in parameters, can reduce collision rates by up to 26% and drivable area conflict rate by up to 42%. Codes, pre-trained models, and demo videos can be accessed at https://github.com/pengxuanyang/UncAD.
☆ Explainable Scene Understanding with Qualitative Representations and Graph Neural Networks
This paper investigates the integration of graph neural networks (GNNs) with Qualitative Explainable Graphs (QXGs) for scene understanding in automated driving. Scene understanding is the basis for any further reactive or proactive decision-making. Scene understanding and related reasoning is inherently an explanation task: why is another traffic participant doing something, what or who caused their actions? While previous work demonstrated QXGs' effectiveness using shallow machine learning models, these approaches were limited to analysing single relation chains between object pairs, disregarding the broader scene context. We propose a novel GNN architecture that processes entire graph structures to identify relevant objects in traffic scenes. We evaluate our method on the nuScenes dataset enriched with DriveLM's human-annotated relevance labels. Experimental results show that our GNN-based approach achieves superior performance compared to baseline methods. The model effectively handles the inherent class imbalance in relevant object identification tasks while considering the complete spatial-temporal relationships between all objects in the scene. Our work demonstrates the potential of combining qualitative representations with deep learning approaches for explainable scene understanding in autonomous driving systems.
comment: Workshop "Advancing Automated Driving in Highly Interactive Scenarios through Behavior Prediction, Trustworthy AI, and Remote Operations" @ 36th IEEE Intelligent Vehicles Symposium (IV)
☆ Approaching Current Challenges in Developing a Software Stack for Fully Autonomous Driving
Autonomous driving is a complex undertaking. A common approach is to break down the driving task into individual subtasks through modularization. These sub-modules are usually developed and published separately. However, if these individually developed algorithms have to be combined again to form a full-stack autonomous driving software, this poses particular challenges. Drawing upon our practical experience in developing the software of TUM Autonomous Motorsport, we have identified and derived these challenges in developing an autonomous driving software stack within a scientific environment. We do not focus on the specific challenges of individual algorithms but on the general difficulties that arise when deploying research algorithms on real-world test vehicles. To overcome these challenges, we introduce strategies that have been effective in our development approach. We additionally provide open-source implementations that enable these concepts on GitHub. As a result, this paper's contributions will simplify future full-stack autonomous driving projects, which are essential for a thorough evaluation of the individual algorithms.
comment: Accepted at IEEE IV 2025
☆ Trajectory Adaptation using Large Language Models
Adapting robot trajectories based on human instructions as per new situations is essential for achieving more intuitive and scalable human-robot interactions. This work proposes a flexible language-based framework to adapt generic robotic trajectories produced by off-the-shelf motion planners like RRT, A-star, etc, or learned from human demonstrations. We utilize pre-trained LLMs to adapt trajectory waypoints by generating code as a policy for dense robot manipulation, enabling more complex and flexible instructions than current methods. This approach allows us to incorporate a broader range of commands, including numerical inputs. Compared to state-of-the-art feature-based sequence-to-sequence models which require training, our method does not require task-specific training and offers greater interpretability and more effective feedback mechanisms. We validate our approach through simulation experiments on the robotic manipulator, aerial vehicle, and ground robot in the Pybullet and Gazebo simulation environments, demonstrating that LLMs can successfully adapt trajectories to complex human instructions.
comment: Accepted to CoRL LangRob workshop 2024
☆ Biasing the Driving Style of an Artificial Race Driver for Online Time-Optimal Maneuver Planning
In this work, we present a novel approach to bias the driving style of an artificial race driver (ARD) for online time-optimal trajectory planning. Our method leverages a nonlinear model predictive control (MPC) framework that combines time minimization with exit speed maximization at the end of the planning horizon. We introduce a new MPC terminal cost formulation based on the trajectory planned in the previous MPC step, enabling ARD to adapt its driving style from early to late apex maneuvers in real-time. Our approach is computationally efficient, allowing for low replan times and long planning horizons. We validate our method through simulations, comparing the results against offline minimum-lap-time (MLT) optimal control and online minimum-time MPC solutions. The results demonstrate that our new terminal cost enables ARD to bias its driving style, and achieve online lap times close to the MLT solution and faster than the minimum-time MPC solution. Our approach paves the way for a better understanding of the reasons behind human drivers' choice of early or late apex maneuvers.
☆ B*: Efficient and Optimal Base Placement for Fixed-Base Manipulators
B* is a novel optimization framework that addresses a critical challenge in fixed-base manipulator robotics: optimal base placement. Current methods rely on pre-computed kinematics databases generated through sampling to search for solutions. However, they face an inherent trade-off between solution optimality and computational efficiency when determining sampling resolution. To address these limitations, B* unifies multiple objectives without database dependence. The framework employs a two-layer hierarchical approach. The outer layer systematically manages terminal constraints through progressive tightening, particularly for base mobility, enabling feasible initialization and broad solution exploration. The inner layer addresses non-convexities in each outer-layer subproblem through sequential local linearization, converting the original problem into tractable sequential linear programming (SLP). Testing across multiple robot platforms demonstrates B*'s effectiveness. The framework achieves solution optimality five orders of magnitude better than sampling-based approaches while maintaining perfect success rates and reduced computational overhead. Operating directly in configuration space, B* enables simultaneous path planning with customizable optimization criteria. B* serves as a crucial initialization tool that bridges the gap between theoretical motion planning and practical deployment, where feasible trajectory existence is fundamental.
☆ Embodied Neuromorphic Control Applied on a 7-DOF Robotic Manipulator
The development of artificial intelligence towards real-time interaction with the environment is a key aspect of embodied intelligence and robotics. Inverse dynamics is a fundamental robotics problem, which maps from joint space to torque space of robotic systems. Traditional methods for solving it rely on direct physical modeling of robots which is difficult or even impossible due to nonlinearity and external disturbance. Recently, data-based model-learning algorithms are adopted to address this issue. However, they often require manual parameter tuning and high computational costs. Neuromorphic computing is inherently suitable to process spatiotemporal features in robot motion control at extremely low costs. However, current research is still in its infancy: existing works control only low-degree-of-freedom systems and lack performance quantification and comparison. In this paper, we propose a neuromorphic control framework to control 7 degree-of-freedom robotic manipulators. We use Spiking Neural Network to leverage the spatiotemporal continuity of the motion data to improve control accuracy, and eliminate manual parameters tuning. We validated the algorithm on two robotic platforms, which reduces torque prediction error by at least 60% and performs a target position tracking task successfully. This work advances embodied neuromorphic control by one step forward from proof of concept to applications in complex real-world tasks.
☆ A Genetic Approach to Gradient-Free Kinodynamic Planning in Uneven Terrains
This paper proposes a genetic algorithm-based kinodynamic planning algorithm (GAKD) for car-like vehicles navigating uneven terrains modeled as triangular meshes. The algorithm's distinct feature is trajectory optimization over a fixed-length receding horizon using a genetic algorithm with heuristic-based mutation, ensuring the vehicle's controls remain within its valid operational range. By addressing challenges posed by uneven terrain meshes, such as changing face normals, GAKD offers a practical solution for path planning in complex environments. Comparative evaluations against Model Predictive Path Integral (MPPI) and log-MPPI methods show that GAKD achieves up to 20 percent improvement in traversability cost while maintaining comparable path length. These results demonstrate GAKD's potential in improving vehicle navigation on challenging terrains.
☆ Autonomous Drone for Dynamic Smoke Plume Tracking
This paper presents a novel autonomous drone-based smoke plume tracking system capable of navigating and tracking plumes in highly unsteady atmospheric conditions. The system integrates advanced hardware and software and a comprehensive simulation environment to ensure robust performance in controlled and real-world settings. The quadrotor, equipped with a high-resolution imaging system and an advanced onboard computing unit, performs precise maneuvers while accurately detecting and tracking dynamic smoke plumes under fluctuating conditions. Our software implements a two-phase flight operation, i.e., descending into the smoke plume upon detection and continuously monitoring the smoke movement during in-plume tracking. Leveraging Proportional Integral-Derivative (PID) control and a Proximal Policy Optimization based Deep Reinforcement Learning (DRL) controller enables adaptation to plume dynamics. Unreal Engine simulation evaluates performance under various smoke-wind scenarios, from steady flow to complex, unsteady fluctuations, showing that while the PID controller performs adequately in simpler scenarios, the DRL-based controller excels in more challenging environments. Field tests corroborate these findings. This system opens new possibilities for drone-based monitoring in areas like wildfire management and air quality assessment. The successful integration of DRL for real-time decision-making advances autonomous drone control for dynamic environments.
comment: 7 pages, 7 figures
☆ A0: An Affordance-Aware Hierarchical Model for General Robotic Manipulation
Robotic manipulation faces critical challenges in understanding spatial affordances--the "where" and "how" of object interactions--essential for complex manipulation tasks like wiping a board or stacking objects. Existing methods, including modular-based and end-to-end approaches, often lack robust spatial reasoning capabilities. Unlike recent point-based and flow-based affordance methods that focus on dense spatial representations or trajectory modeling, we propose A0, a hierarchical affordance-aware diffusion model that decomposes manipulation tasks into high-level spatial affordance understanding and low-level action execution. A0 leverages the Embodiment-Agnostic Affordance Representation, which captures object-centric spatial affordances by predicting contact points and post-contact trajectories. A0 is pre-trained on 1 million contact points data and fine-tuned on annotated trajectories, enabling generalization across platforms. Key components include Position Offset Attention for motion-aware feature extraction and a Spatial Information Aggregation Layer for precise coordinate mapping. The model's output is executed by the action execution module. Experiments on multiple robotic systems (Franka, Kinova, Realman, and Dobot) demonstrate A0's superior performance in complex tasks, showcasing its efficiency, flexibility, and real-world applicability.
☆ Graph-based Path Planning with Dynamic Obstacle Avoidance for Autonomous Parking
Safe and efficient path planning in parking scenarios presents a significant challenge due to the presence of cluttered environments filled with static and dynamic obstacles. To address this, we propose a novel and computationally efficient planning strategy that seamlessly integrates the predictions of dynamic obstacles into the planning process, ensuring the generation of collision-free paths. Our approach builds upon the conventional Hybrid A star algorithm by introducing a time-indexed variant that explicitly accounts for the predictions of dynamic obstacles during node exploration in the graph, thus enabling dynamic obstacle avoidance. We integrate the time-indexed Hybrid A star algorithm within an online planning framework to compute local paths at each planning step, guided by an adaptively chosen intermediate goal. The proposed method is validated in diverse parking scenarios, including perpendicular, angled, and parallel parking. Through simulations, we showcase our approach's potential in greatly improving the efficiency and safety when compared to the state of the art spline-based planning method for parking situations.
☆ Crossing the Human-Robot Embodiment Gap with Sim-to-Real RL using One Human Demonstration
Teaching robots dexterous manipulation skills often requires collecting hundreds of demonstrations using wearables or teleoperation, a process that is challenging to scale. Videos of human-object interactions are easier to collect and scale, but leveraging them directly for robot learning is difficult due to the lack of explicit action labels from videos and morphological differences between robot and human hands. We propose Human2Sim2Robot, a novel real-to-sim-to-real framework for training dexterous manipulation policies using only one RGB-D video of a human demonstrating a task. Our method utilizes reinforcement learning (RL) in simulation to cross the human-robot embodiment gap without relying on wearables, teleoperation, or large-scale data collection typically necessary for imitation learning methods. From the demonstration, we extract two task-specific components: (1) the object pose trajectory to define an object-centric, embodiment-agnostic reward function, and (2) the pre-manipulation hand pose to initialize and guide exploration during RL training. We found that these two components are highly effective for learning the desired task, eliminating the need for task-specific reward shaping and tuning. We demonstrate that Human2Sim2Robot outperforms object-aware open-loop trajectory replay by 55% and imitation learning with data augmentation by 68% across grasping, non-prehensile manipulation, and multi-step tasks. Project Site: https://human2sim2robot.github.io
comment: 15 pages, 13 figures
☆ Acoustic Analysis of Uneven Blade Spacing and Toroidal Geometry for Reducing Propeller Annoyance
Unmanned aerial vehicles (UAVs) are becoming more commonly used in populated areas, raising concerns about noise pollution generated from their propellers. This study investigates the acoustic performance of unconventional propeller designs, specifically toroidal and uneven-blade spaced propellers, for their potential in reducing psychoacoustic annoyance. Our experimental results show that these designs noticeably reduced acoustic characteristics associated with noise annoyance.
comment: For paper website, see https://tubaa.dev/ . 5 pages, 6 figures. Manuscript originally completed on October 6, 2023 and revised on April 16, 2025
☆ UniPhys: Unified Planner and Controller with Diffusion for Flexible Physics-Based Character Control
Generating natural and physically plausible character motion remains challenging, particularly for long-horizon control with diverse guidance signals. While prior work combines high-level diffusion-based motion planners with low-level physics controllers, these systems suffer from domain gaps that degrade motion quality and require task-specific fine-tuning. To tackle this problem, we introduce UniPhys, a diffusion-based behavior cloning framework that unifies motion planning and control into a single model. UniPhys enables flexible, expressive character motion conditioned on multi-modal inputs such as text, trajectories, and goals. To address accumulated prediction errors over long sequences, UniPhys is trained with the Diffusion Forcing paradigm, learning to denoise noisy motion histories and handle discrepancies introduced by the physics simulator. This design allows UniPhys to robustly generate physically plausible, long-horizon motions. Through guided sampling, UniPhys generalizes to a wide range of control signals, including unseen ones, without requiring task-specific fine-tuning. Experiments show that UniPhys outperforms prior methods in motion naturalness, generalization, and robustness across diverse control tasks.
comment: Project page: https://wuyan01.github.io/uniphys-project/
♻ ☆ Learning Diverse Robot Striking Motions with Diffusion Models and Kinematically Constrained Gradient Guidance ICRA 2025
Advances in robot learning have enabled robots to generate skills for a variety of tasks. Yet, robot learning is typically sample inefficient, struggles to learn from data sources exhibiting varied behaviors, and does not naturally incorporate constraints. These properties are critical for fast, agile tasks such as playing table tennis. Modern techniques for learning from demonstration improve sample efficiency and scale to diverse data, but are rarely evaluated on agile tasks. In the case of reinforcement learning, achieving good performance requires training on high-fidelity simulators. To overcome these limitations, we develop a novel diffusion modeling approach that is offline, constraint-guided, and expressive of diverse agile behaviors. The key to our approach is a kinematic constraint gradient guidance (KCGG) technique that computes gradients through both the forward kinematics of the robot arm and the diffusion model to direct the sampling process. KCGG minimizes the cost of violating constraints while simultaneously keeping the sampled trajectory in-distribution of the training data. We demonstrate the effectiveness of our approach for time-critical robotic tasks by evaluating KCGG in two challenging domains: simulated air hockey and real table tennis. In simulated air hockey, we achieved a 25.4% increase in block rate, while in table tennis, we saw a 17.3% increase in success rate compared to imitation learning baselines.
comment: ICRA 2025
♻ ☆ Relevance for Human Robot Collaboration
Inspired by the human ability to selectively focus on relevant information, this paper introduces relevance, a novel dimensionality reduction process for human-robot collaboration (HRC). Our approach incorporates a continuously operating perception module, evaluates cue sufficiency within the scene, and applies a flexible formulation and computation framework. To accurately and efficiently quantify relevance, we developed an event-based framework that maintains a continuous perception of the scene and selectively triggers relevance determination. Within this framework, we developed a probabilistic methodology, which considers various factors and is built on a novel structured scene representation. Simulation results demonstrate that the relevance framework and methodology accurately predict the relevance of a general HRC setup, achieving a precision of 0.99, a recall of 0.94, an F1 score of 0.96, and an object ratio of 0.94. Relevance can be broadly applied to several areas in HRC to accurately improve task planning time by 79.56% compared with pure planning for a cereal task, reduce perception latency by up to 26.53% for an object detector, improve HRC safety by up to 13.50% and reduce the number of inquiries for HRC by 80.84%. A real-world demonstration showcases the relevance framework's ability to intelligently and seamlessly assist humans in everyday tasks.
comment: under review
♻ ☆ Perceive With Confidence: Statistical Safety Assurances for Navigation with Learning-Based Perception
Rapid advances in perception have enabled large pre-trained models to be used out of the box for transforming high-dimensional, noisy, and partial observations of the world into rich occupancy representations. However, the reliability of these models and consequently their safe integration onto robots remains unknown when deployed in environments unseen during training. To provide safety guarantees, we rigorously quantify the uncertainty of pre-trained perception systems for object detection and scene completion via a novel calibration technique based on conformal prediction. Crucially, this procedure guarantees robustness to distribution shifts in states when perception outputs are used in conjunction with a planner. As a result, the calibrated perception system can be used in combination with any safe planner to provide an end-to-end statistical assurance on safety in unseen environments. We evaluate the resulting approach, Perceive with Confidence (PwC), in simulation and on hardware where a quadruped robot navigates through previously unseen indoor, static environments. These experiments validate the safety assurances for obstacle avoidance provided by PwC. In simulation, our method reduces obstacle misdetection by $70\%$ compared to uncalibrated perception models. While misdetections lead to collisions for baseline methods, our approach consistently achieves $100\%$ safety. We further demonstrate reducing the conservatism of our method without sacrificing safety, achieving a $46\%$ increase in success rates in challenging environments while maintaining $100\%$ safety. In hardware experiments, our method improves empirical safety by $40\%$ over baselines and reduces obstacle misdetection by $93.3\%$. The safety gap widens to $46.7\%$ when navigation speed increases, highlighting our approach's robustness under more demanding conditions.
comment: Videos and code can be found at https://perceive-with-confidence.github.io
♻ ☆ Scalable Multi-Robot Motion Planning Using Guidance-Informed Hypergraphs
In this work, we propose a method for multiple mobile robot motion planning that efficiently plans for robot teams up to an order of magnitude larger than existing state-of-the-art methods in congested settings with narrow passages in the environment. We achieve this improvement in scalability by adapting the state-of-the-art Decomposable State Space Hypergraph (DaSH) planning framework to expand the set of problems it can support to include those without a highly structured planning space and those with kinodynamic constraints. We accomplish this by exploiting guidance about a problem's structure to limit exploration of the planning space and through modifying DaSH's conflict resolution scheme. This guidance captures when coordination between robots is necessary, allowing us to decompose the intractably large multi-robot search space while limiting risk of inter-robot conflicts by composing relevant robot groups together while planning.
comment: This work has been submitted for review
♻ ☆ Learn2Decompose: Learning Problem Decomposition for Efficient Sequential Multi-object Manipulation Planning
We present a Reactive Task and Motion Planning (TAMP) approach for efficient sequential multi-object manipulation in dynamic environments. Conventional TAMP solvers experience an exponential increase in planning time as the planning horizon and number of objects grow, limiting their applicability in real-world scenarios. To address this, we propose learning problem decomposition from demonstrations to accelerate TAMP solvers. Our approach consists of three key components: goal decomposition learning, temporal distance learning, and object reduction. Goal decomposition identifies the necessary sequences of states that the system must pass through before reaching the final goal, treating them as subgoal sequences. Temporal distance learning predicts the temporal distance between two states, enabling the system to identify the closest subgoal from a disturbed state. Object reduction minimizes the set of active objects considered during replanning, further improving efficiency. We evaluate our approach on three benchmarks, demonstrating its effectiveness in improving replanning efficiency for sequential multi-object manipulation tasks in dynamic environments.
♻ ☆ Minimum-Violation Temporal Logic Planning for Heterogeneous Robots under Robot Skill Failures
In this paper, we consider teams of robots with heterogeneous skills (e.g., sensing and manipulation) tasked with collaborative missions described by Linear Temporal Logic (LTL) formulas. These LTL-encoded tasks require robots to apply their skills to specific regions and objects in a temporal and logical order. While existing temporal logic planning algorithms can synthesize correct-by-construction plans, they typically lack reactivity to unexpected failures of robot skills, which can compromise mission performance. This paper addresses this challenge by proposing a reactive LTL planning algorithm that adapts to unexpected failures during deployment. Specifically, the proposed algorithm reassigns sub-tasks to robots based on their functioning skills and locally revises team plans to accommodate these new assignments and ensure mission completion. The main novelty of the proposed algorithm is its ability to handle cases where mission completion becomes impossible due to limited functioning robots. Instead of reporting mission failure, the algorithm strategically prioritizes the most crucial sub-tasks and locally revises the team's plans, as per user-specified priorities, to minimize mission violations. We provide theoretical conditions under which the proposed framework computes the minimum-violation task reassignments and team plans. We provide numerical and hardware experiments to demonstrate the efficiency of the proposed method.
♻ ☆ CAP-Net: A Unified Network for 6D Pose and Size Estimation of Categorical Articulated Parts from a Single RGB-D Image CVPR 2025
This paper tackles category-level pose estimation of articulated objects in robotic manipulation tasks and introduces a new benchmark dataset. While recent methods estimate part poses and sizes at the category level, they often rely on geometric cues and complex multi-stage pipelines that first segment parts from the point cloud, followed by Normalized Part Coordinate Space (NPCS) estimation for 6D poses. These approaches overlook dense semantic cues from RGB images, leading to suboptimal accuracy, particularly for objects with small parts. To address these limitations, we propose a single-stage Network, CAP-Net, for estimating the 6D poses and sizes of Categorical Articulated Parts. This method combines RGB-D features to generate instance segmentation and NPCS representations for each part in an end-to-end manner. CAP-Net uses a unified network to simultaneously predict point-wise class labels, centroid offsets, and NPCS maps. A clustering algorithm then groups points of the same predicted class based on their estimated centroid distances to isolate each part. Finally, the NPCS region of each part is aligned with the point cloud to recover its final pose and size. To bridge the sim-to-real domain gap, we introduce the RGBD-Art dataset, the largest RGB-D articulated dataset to date, featuring photorealistic RGB images and depth noise simulated from real sensors. Experimental evaluations on the RGBD-Art dataset demonstrate that our method significantly outperforms the state-of-the-art approach. Real-world deployments of our model in robotic tasks underscore its robustness and exceptional sim-to-real transfer capabilities, confirming its substantial practical utility. Our dataset, code and pre-trained models are available on the project page.
comment: To appear in CVPR 2025 (Highlight)
♻ ☆ Robotic Optimization of Powdered Beverages Leveraging Computer Vision and Bayesian Optimization
The growing demand for innovative research in the food industry is driving the adoption of robots in large-scale experimentation, as it offers increased precision, replicability, and efficiency in product manufacturing and evaluation. To this end, we introduce a robotic system designed to optimize food product quality, focusing on powdered cappuccino preparation as a case study. By leveraging optimization algorithms and computer vision, the robot explores the parameter space to identify the ideal conditions for producing a cappuccino with the best foam quality. The system also incorporates computer vision-driven feedback in a closed-loop control to further improve the beverage. Our findings demonstrate the effectiveness of robotic automation in achieving high repeatability and extensive parameter exploration, paving the way for more advanced and reliable food product development.
♻ ☆ Curriculum-based Sample Efficient Reinforcement Learning for Robust Stabilization of a Quadrotor
This article introduces a curriculum learning approach to develop a reinforcement learning-based robust stabilizing controller for a Quadrotor that meets predefined performance criteria. The learning objective is to achieve desired positions from random initial conditions while adhering to both transient and steady-state performance specifications. This objective is challenging for conventional one-stage end-to-end reinforcement learning, due to the strong coupling between position and orientation dynamics, the complexity in designing and tuning the reward function, and poor sample efficiency, which necessitates substantial computational resources and leads to extended convergence times. To address these challenges, this work decomposes the learning objective into a three-stage curriculum that incrementally increases task complexity. The curriculum begins with learning to achieve stable hovering from a fixed initial condition, followed by progressively introducing randomization in initial positions, orientations and velocities. A novel additive reward function is proposed, to incorporate transient and steady-state performance specifications. The results demonstrate that the Proximal Policy Optimization (PPO)-based curriculum learning approach, coupled with the proposed reward structure, achieves superior performance compared to a single-stage PPO-trained policy with the same reward function, while significantly reducing computational resource requirements and convergence time. The curriculum-trained policy's performance and robustness are thoroughly validated under random initial conditions and in the presence of disturbances.
comment: 8 pages, 7 figures
♻ ☆ Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions CVPR
Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges. This survey provides a wide-ranging examination of the anti-UAV domain, centering on three core objectives-classification, detection, and tracking-while detailing emerging methodologies such as diffusion-based data synthesis, multi-modal fusion, vision-language modeling, self-supervised learning, and reinforcement learning. We systematically evaluate state-of-the-art solutions across both single-modality and multi-sensor pipelines (spanning RGB, infrared, audio, radar, and RF) and discuss large-scale as well as adversarially oriented benchmarks. Our analysis reveals persistent gaps in real-time performance, stealth detection, and swarm-based scenarios, underscoring pressing needs for robust, adaptive anti-UAV systems. By highlighting open research directions, we aim to foster innovation and guide the development of next-generation defense strategies in an era marked by the extensive use of UAVs.
comment: Accepted at CVPR Workshop Anti-UAV 2025. 15 pages
♻ ☆ Causality-enhanced Decision-Making for Autonomous Mobile Robots in Dynamic Environments
The growing integration of robots in shared environments -- such as warehouses, shopping centres, and hospitals -- demands a deep understanding of the underlying dynamics and human behaviours, including how, when, and where individuals engage in various activities and interactions. This knowledge goes beyond simple correlation studies and requires a more comprehensive causal analysis. By leveraging causal inference to model cause-and-effect relationships, we can better anticipate critical environmental factors and enable autonomous robots to plan and execute tasks more effectively. To this end, we propose a novel causality-based decision-making framework that reasons over a learned causal model to predict battery usage and human obstructions, understanding how these factors could influence robot task execution. Such reasoning framework assists the robot in deciding when and how to complete a given task. To achieve this, we developed also PeopleFlow, a new Gazebo-based simulator designed to model context-sensitive human-robot spatial interactions in shared workspaces. PeopleFlow features realistic human and robot trajectories influenced by contextual factors such as time, environment layout, and robot state, and can simulate a large number of agents. While the simulator is general-purpose, in this paper we focus on a warehouse-like environment as a case study, where we conduct an extensive evaluation benchmarking our causal approach against a non-causal baseline. Our findings demonstrate the efficacy of the proposed solutions, highlighting how causal reasoning enables autonomous robots to operate more efficiently and safely in dynamic environments shared with humans.
comment: Causal Discovery and Inference - Robot Autonomy - Human-Robot Spatial Interaction - Decision-Making
♻ ☆ ExploRLLM: Guiding Exploration in Reinforcement Learning with Large Language Models ICRA
In robot manipulation, Reinforcement Learning (RL) often suffers from low sample efficiency and uncertain convergence, especially in large observation and action spaces. Foundation Models (FMs) offer an alternative, demonstrating promise in zero-shot and few-shot settings. However, they can be unreliable due to limited physical and spatial understanding. We introduce ExploRLLM, a method that combines the strengths of both paradigms. In our approach, FMs improve RL convergence by generating policy code and efficient representations, while a residual RL agent compensates for the FMs' limited physical understanding. We show that ExploRLLM outperforms both policies derived from FMs and RL baselines in table-top manipulation tasks. Additionally, real-world experiments show that the policies exhibit promising zero-shot sim-to-real transfer. Supplementary material is available at https://explorllm.github.io.
comment: 6 pages, 6 figures, IEEE International Conference on Robotics and Automation (ICRA) 2025
♻ ☆ Listen to Your Map: An Online Representation for Spatial Sonification
Robotic perception is becoming a key technology for navigation aids, especially helping individuals with visual impairments through spatial sonification. This paper introduces a mapping representation that accurately captures scene geometry for sonification, turning physical spaces into auditory experiences. Using depth sensors, we encode an incrementally built 3D scene into a compact 360-degree representation with angular and distance information, aligning this way with human auditory spatial perception. The proposed framework performs localisation and mapping via VDB-Gaussian Process Distance Fields for efficient online scene reconstruction. The key aspect is a sensor-centric structure that maintains either a 2D-circular or 3D-cylindrical raster-based projection. This spatial representation is then converted into binaural auditory signals using simple pre-recorded responses from a representative room. Quantitative and qualitative evaluations show improvements in accuracy, coverage, timing and suitability for sonification compared to other approaches, with effective handling of dynamic objects as well. An accompanying video demonstrates spatial sonification in room-like environments. https://tinyurl.com/ListenToYourMap
♻ ☆ Embedding high-resolution touch across robotic hands enables adaptive human-like grasping
Developing robotic hands that adapt to real-world dynamics remains a fundamental challenge in robotics and machine intelligence. Despite significant advances in replicating human hand kinematics and control algorithms, robotic systems still struggle to match human capabilities in dynamic environments, primarily due to inadequate tactile feedback. To bridge this gap, we present F-TAC Hand, a biomimetic hand featuring high-resolution tactile sensing (0.1mm spatial resolution) across 70% of its surface area. Through optimized hand design, we overcome traditional challenges in integrating high-resolution tactile sensors while preserving the full range of motion. The hand, powered by our generative algorithm that synthesizes human-like hand configurations, demonstrates robust grasping capabilities in dynamic real-world conditions. Extensive evaluation across 600 real-world trials demonstrates that this tactile-embodied system significantly outperforms non-tactile-informed alternatives in complex manipulation tasks (p<0.0001). These results provide empirical evidence for the critical role of rich tactile embodiment in developing advanced robotic intelligence, offering new perspectives on the relationship between physical sensing capabilities and intelligent behavior.
♻ ☆ Planning for quasi-static manipulation tasks via an intrinsic haptic metric: a book insertion case study
Contact-rich manipulation often requires strategic interactions with objects, such as pushing to accomplish specific tasks. We propose a novel scenario where a robot inserts a book into a crowded shelf by pushing aside neighboring books to create space before slotting the new book into place. Classical planning algorithms fail in this context due to limited space and their tendency to avoid contact. Additionally, they do not handle indirectly manipulable objects or consider force interactions. Our key contributions are: i) reframing quasi-static manipulation as a planning problem on an implicit manifold derived from equilibrium conditions; ii) utilizing an intrinsic haptic metric instead of ad-hoc cost functions; and iii) proposing an adaptive algorithm that simultaneously updates robot states, object positions, contact points, and haptic distances. We evaluate our method on a crowded bookshelf insertion task, and it can be generally applied to rigid body manipulation tasks. We propose proxies to capture contact points and forces, with superellipses to represent objects. This simplified model guarantees differentiability. Our framework autonomously discovers strategic wedging-in policies while our simplified contact model achieves behavior similar to real world scenarios. We also vary the stiffness and initial positions to analyze our framework comprehensively. The video can be found at https://youtu.be/eab8umZ3AQ0.
♻ ☆ Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework
Multimodal foundation models offer a promising framework for robotic perception and planning by processing sensory inputs to generate actionable plans. However, addressing uncertainty in both perception (sensory interpretation) and decision-making (plan generation) remains a critical challenge for ensuring task reliability. We present a comprehensive framework to disentangle, quantify, and mitigate these two forms of uncertainty. We first introduce a framework for uncertainty disentanglement, isolating perception uncertainty arising from limitations in visual understanding and decision uncertainty relating to the robustness of generated plans. To quantify each type of uncertainty, we propose methods tailored to the unique properties of perception and decision-making: we use conformal prediction to calibrate perception uncertainty and introduce Formal-Methods-Driven Prediction (FMDP) to quantify decision uncertainty, leveraging formal verification techniques for theoretical guarantees. Building on this quantification, we implement two targeted intervention mechanisms: an active sensing process that dynamically re-observes high-uncertainty scenes to enhance visual input quality and an automated refinement procedure that fine-tunes the model on high-certainty data, improving its capability to meet task specifications. Empirical validation in real-world and simulated robotic tasks demonstrates that our uncertainty disentanglement framework reduces variability by up to 40% and enhances task success rates by 5% compared to baselines. These improvements are attributed to the combined effect of both interventions and highlight the importance of uncertainty disentanglement, which facilitates targeted interventions that enhance the robustness and reliability of autonomous systems. Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/.
comment: Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/
♻ ☆ A Centralized Planning and Distributed Execution Method for Shape Filling with Homogeneous Mobile Robots
The pattern formation task is commonly seen in a multi-robot system. In this paper, we study the problem of forming complex shapes with functionally limited mobile robots, which have to rely on other robots to precisely locate themselves. The goal is to decide whether a given shape can be filled by a given set of robots; in case the answer is yes, to complete a shape formation process as fast as possible with a minimum amount of communication. Traditional approaches either require global coordinates for each robot or are prone to failure when attempting to form complex shapes beyond the capability of given approaches - the latter calls for a decision procedure that can tell whether a target shape can be formed before the actual shape-forming process starts. In this paper, we develop a method that does not require global coordinate information during the execution process and can effectively decide whether it is feasible to form the desired shape. The latter is achieved via a planning procedure that is capable of handling a variety of complex shapes, in particular, those with holes, and assigning a simple piece of scheduling information to each robot, facilitating subsequent distributed execution, which does not rely on the coordinates of all robots but only those of neighboring ones. The effectiveness of our shape-forming approach is vividly illustrated in several simulation case studies.
♻ ☆ Safety with Agency: Human-Centered Safety Filter with Application to AI-Assisted Motorsports
We propose a human-centered safety filter (HCSF) for shared autonomy that significantly enhances system safety without compromising human agency. Our HCSF is built on a neural safety value function, which we first learn scalably through black-box interactions and then use at deployment to enforce a novel state-action control barrier function (Q-CBF) safety constraint. Since this Q-CBF safety filter does not require any knowledge of the system dynamics for both synthesis and runtime safety monitoring and intervention, our method applies readily to complex, black-box shared autonomy systems. Notably, our HCSF's CBF-based interventions modify the human's actions minimally and smoothly, avoiding the abrupt, last-moment corrections delivered by many conventional safety filters. We validate our approach in a comprehensive in-person user study using Assetto Corsa-a high-fidelity car racing simulator with black-box dynamics-to assess robustness in "driving on the edge" scenarios. We compare both trajectory data and drivers' perceptions of our HCSF assistance against unassisted driving and a conventional safety filter. Experimental results show that 1) compared to having no assistance, our HCSF improves both safety and user satisfaction without compromising human agency or comfort, and 2) relative to a conventional safety filter, our proposed HCSF boosts human agency, comfort, and satisfaction while maintaining robustness.
comment: Accepted to Robotics: Science and Systems (R:SS) 2025, 22 pages, 16 figures, 7 tables
Artificial Intelligence 151
☆ PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding
Vision-language models are integral to computer vision research, yet many high-performing models remain closed-source, obscuring their data, design and training recipe. The research community has responded by using distillation from black-box models to label training data, achieving strong benchmark results, at the cost of measurable scientific progress. However, without knowing the details of the teacher model and its data sources, scientific progress remains difficult to measure. In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training pipelines without distillation from proprietary models and explore large-scale synthetic data to identify critical data gaps, particularly in detailed video understanding. To bridge these gaps, we release 2.8M human-labeled instances of fine-grained video question-answer pairs and spatio-temporally grounded video captions. Additionally, we introduce PLM-VideoBench, a suite for evaluating challenging video understanding tasks focusing on the ability to reason about "what", "where", "when", and "how" of a video. We make our work fully reproducible by providing data, training recipes, code & models.
comment: Technical report
☆ It's All Connected: A Journey Through Test-Time Memorization, Attentional Bias, Retention, and Online Optimization
Designing efficient and effective architectural backbones has been in the core of research efforts to enhance the capability of foundation models. Inspired by the human cognitive phenomenon of attentional bias-the natural tendency to prioritize certain events or stimuli-we reconceptualize neural architectures, including Transformers, Titans, and modern linear recurrent neural networks as associative memory modules that learn a mapping of keys and values using an internal objective, referred to as attentional bias. Surprisingly, we observed that most existing sequence models leverage either (1) dot-product similarity, or (2) L2 regression objectives as their attentional bias. Going beyond these objectives, we present a set of alternative attentional bias configurations along with their effective approximations to stabilize their training procedure. We then reinterpret forgetting mechanisms in modern deep learning architectures as a form of retention regularization, providing a novel set of forget gates for sequence models. Building upon these insights, we present Miras, a general framework to design deep learning architectures based on four choices of: (i) associative memory architecture, (ii) attentional bias objective, (iii) retention gate, and (iv) memory learning algorithm. We present three novel sequence models-Moneta, Yaad, and Memora-that go beyond the power of existing linear RNNs while maintaining a fast parallelizable training process. Our experiments show different design choices in Miras yield models with varying strengths. For example, certain instances of Miras achieve exceptional performance in special tasks such as language modeling, commonsense reasoning, and recall intensive tasks, even outperforming Transformers and other modern linear recurrent models.
☆ Sleep-time Compute: Beyond Inference Scaling at Test-time
Scaling test-time compute has emerged as a key ingredient for enabling large language models (LLMs) to solve difficult problems, but comes with high latency and inference cost. We introduce sleep-time compute, which allows models to "think" offline about contexts before queries are presented: by anticipating what queries users might ask and pre-computing useful quantities, we can significantly reduce the compute requirements at test-time. To demonstrate the efficacy of our method, we create modified versions of two reasoning tasks - Stateful GSM-Symbolic and Stateful AIME. We find that sleep-time compute can reduce the amount of test-time compute needed to achieve the same accuracy by ~ 5x on Stateful GSM-Symbolic and Stateful AIME and that by scaling sleep-time compute we can further increase accuracy by up to 13% on Stateful GSM-Symbolic and 18% on Stateful AIME. Furthermore, we introduce Multi-Query GSM-Symbolic, which extends GSM-Symbolic by including multiple related queries per context. By amortizing sleep-time compute across related queries about the same context using Multi-Query GSM-Symbolic, we can decrease the average cost per query by 2.5x. We then conduct additional analysis to understand when sleep-time compute is most effective, finding the predictability of the user query to be well correlated with the efficacy of sleep-time compute. Finally, we conduct a case-study of applying sleep-time compute to a realistic agentic SWE task.
comment: Code and data released at: https://github.com/letta-ai/sleep-time-compute
☆ RUKA: Rethinking the Design of Humanoid Hands with Learning
Dexterous manipulation is a fundamental capability for robotic systems, yet progress has been limited by hardware trade-offs between precision, compactness, strength, and affordability. Existing control methods impose compromises on hand designs and applications. However, learning-based approaches present opportunities to rethink these trade-offs, particularly to address challenges with tendon-driven actuation and low-cost materials. This work presents RUKA, a tendon-driven humanoid hand that is compact, affordable, and capable. Made from 3D-printed parts and off-the-shelf components, RUKA has 5 fingers with 15 underactuated degrees of freedom enabling diverse human-like grasps. Its tendon-driven actuation allows powerful grasping in a compact, human-sized form factor. To address control challenges, we learn joint-to-actuator and fingertip-to-actuator models from motion-capture data collected by the MANUS glove, leveraging the hand's morphological accuracy. Extensive evaluations demonstrate RUKA's superior reachability, durability, and strength compared to other robotic hands. Teleoperation tasks further showcase RUKA's dexterous movements. The open-source design and assembly instructions of RUKA, code, and data are available at https://ruka-hand.github.io/.
comment: Website at https://ruka-hand.github.io/
☆ MIB: A Mechanistic Interpretability Benchmark
How can we know whether new mechanistic interpretability methods achieve real improvements? In pursuit of meaningful and lasting evaluation standards, we propose MIB, a benchmark with two tracks spanning four tasks and five models. MIB favors methods that precisely and concisely recover relevant causal pathways or specific causal variables in neural language models. The circuit localization track compares methods that locate the model components - and connections between them - most important for performing a task (e.g., attribution patching or information flow routes). The causal variable localization track compares methods that featurize a hidden vector, e.g., sparse autoencoders (SAEs) or distributed alignment search (DAS), and locate model features for a causal variable relevant to the task. Using MIB, we find that attribution and mask optimization methods perform best on circuit localization. For causal variable localization, we find that the supervised DAS method performs best, while SAE features are not better than neurons, i.e., standard dimensions of hidden vectors. These findings illustrate that MIB enables meaningful comparisons of methods, and increases our confidence that there has been real progress in the field.
☆ Readable Twins of Unreadable Models
Creating responsible artificial intelligence (AI) systems is an important issue in contemporary research and development of works on AI. One of the characteristics of responsible AI systems is their explainability. In the paper, we are interested in explainable deep learning (XDL) systems. On the basis of the creation of digital twins of physical objects, we introduce the idea of creating readable twins (in the form of imprecise information flow models) for unreadable deep learning models. The complete procedure for switching from the deep learning model (DLM) to the imprecise information flow model (IIFM) is presented. The proposed approach is illustrated with an example of a deep learning classification model for image recognition of handwritten digits from the MNIST data set.
comment: Based on the abstract accepted for ISFS 2025
☆ Antidistillation Sampling
Frontier models that generate extended reasoning traces inadvertently produce rich token sequences that can facilitate model distillation. Recognizing this vulnerability, model owners may seek sampling strategies that limit the effectiveness of distillation without compromising model performance. \emph{Antidistillation sampling} provides exactly this capability. By strategically modifying a model's next-token probability distribution, antidistillation sampling poisons reasoning traces, rendering them significantly less effective for distillation while preserving the model's practical utility. For further details, see https://antidistillation.com.
☆ Exploring Expert Failures Improves LLM Agent Tuning
Large Language Models (LLMs) have shown tremendous potential as agents, excelling at tasks that require multiple rounds of reasoning and interactions. Rejection Sampling Fine-Tuning (RFT) has emerged as an effective method for finetuning LLMs as agents: it first imitates expert-generated successful trajectories and further improves agentic skills through iterative fine-tuning on successful, self-generated trajectories. However, since the expert (e.g., GPT-4) succeeds primarily on simpler subtasks and RFT inherently favors simpler scenarios, many complex subtasks remain unsolved and persistently out-of-distribution (OOD). Upon investigating these challenging subtasks, we discovered that previously failed expert trajectories can often provide valuable guidance, e.g., plans and key actions, that can significantly improve agent exploration efficiency and acquisition of critical skills. Motivated by these observations, we propose Exploring Expert Failures (EEF), which identifies beneficial actions from failed expert trajectories and integrates them into the training dataset. Potentially harmful actions are meticulously excluded to prevent contamination of the model learning process. By leveraging the beneficial actions in expert failures, EEF successfully solves some previously unsolvable subtasks and improves agent tuning performance. Remarkably, our approach achieved a 62\% win rate in WebShop, outperforming RFT (53. 6\%) and GPT-4 (35. 6\%), and to the best of our knowledge, setting a new state-of-the-art as the first method to surpass a score of 0.81 in WebShop and exceed 81 in SciWorld.
☆ $\texttt{Complex-Edit}$: CoT-Like Instruction Generation for Complexity-Controllable Image Editing Benchmark SC
We introduce $\texttt{Complex-Edit}$, a comprehensive benchmark designed to systematically evaluate instruction-based image editing models across instructions of varying complexity. To develop this benchmark, we harness GPT-4o to automatically collect a diverse set of editing instructions at scale. Our approach follows a well-structured ``Chain-of-Edit'' pipeline: we first generate individual atomic editing tasks independently and then integrate them to form cohesive, complex instructions. Additionally, we introduce a suite of metrics to assess various aspects of editing performance, along with a VLM-based auto-evaluation pipeline that supports large-scale assessments. Our benchmark yields several notable insights: 1) Open-source models significantly underperform relative to proprietary, closed-source models, with the performance gap widening as instruction complexity increases; 2) Increased instructional complexity primarily impairs the models' ability to retain key elements from the input images and to preserve the overall aesthetic quality; 3) Decomposing a complex instruction into a sequence of atomic steps, executed in a step-by-step manner, substantially degrades performance across multiple metrics; 4) A straightforward Best-of-N selection strategy improves results for both direct editing and the step-by-step sequential approach; and 5) We observe a ``curse of synthetic data'': when synthetic data is involved in model training, the edited images from such models tend to appear increasingly synthetic as the complexity of the editing instructions rises -- a phenomenon that intriguingly also manifests in the latest GPT-4o outputs.
comment: Project Page: https://ucsc-vlaa.github.io/Complex-Edit/, Dataset: https://huggingface.co/datasets/UCSC-VLAA/Complex-Edit
☆ Syntactic and Semantic Control of Large Language Models via Sequential Monte Carlo
A wide range of LM applications require generating text that conforms to syntactic or semantic constraints. Imposing such constraints can be naturally framed as probabilistic conditioning, but exact generation from the resulting distribution -- which can differ substantially from the LM's base distribution -- is generally intractable. In this work, we develop an architecture for controlled LM generation based on sequential Monte Carlo (SMC). Our SMC framework allows us to flexibly incorporate domain- and problem-specific constraints at inference time, and efficiently reallocate computational resources in light of new information during the course of generation. By comparing to a number of alternatives and ablations on four challenging domains -- Python code generation for data science, text-to-SQL, goal inference, and molecule synthesis -- we demonstrate that, with little overhead, our approach allows small open-source language models to outperform models over 8x larger, as well as closed-source, fine-tuned ones. In support of the probabilistic perspective, we show that these performance improvements are driven by better approximation to the posterior distribution. Our system builds on the framework of Lew et al. (2023) and integrates with its language model probabilistic programming language, giving users a simple, programmable way to apply SMC to a broad variety of controlled generation problems.
comment: 34 pages, 4 figures
☆ NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement: Methods and Results CVPR
This paper presents a review for the NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement. The challenge comprises two tracks: (i) Efficient Video Quality Assessment (KVQ), and (ii) Diffusion-based Image Super-Resolution (KwaiSR). Track 1 aims to advance the development of lightweight and efficient video quality assessment (VQA) models, with an emphasis on eliminating reliance on model ensembles, redundant weights, and other computationally expensive components in the previous IQA/VQA competitions. Track 2 introduces a new short-form UGC dataset tailored for single image super-resolution, i.e., the KwaiSR dataset. It consists of 1,800 synthetically generated S-UGC image pairs and 1,900 real-world S-UGC images, which are split into training, validation, and test sets using a ratio of 8:1:1. The primary objective of the challenge is to drive research that benefits the user experience of short-form UGC platforms such as Kwai and TikTok. This challenge attracted 266 participants and received 18 valid final submissions with corresponding fact sheets, significantly contributing to the progress of short-form UGC VQA and image superresolution. The project is publicly available at https://github.com/lixinustc/KVQE- ChallengeCVPR-NTIRE2025.
comment: Challenge Report of NTIRE 2025; Methods from 18 Teams; Accepted by CVPR Workshop; 21 pages
☆ Science-T2I: Addressing Scientific Illusions in Image Synthesis CVPR 2025
We present a novel approach to integrating scientific knowledge into generative models, enhancing their realism and consistency in image synthesis. First, we introduce Science-T2I, an expert-annotated adversarial dataset comprising adversarial 20k image pairs with 9k prompts, covering wide distinct scientific knowledge categories. Leveraging Science-T2I, we present SciScore, an end-to-end reward model that refines the assessment of generated images based on scientific knowledge, which is achieved by augmenting both the scientific comprehension and visual capabilities of pre-trained CLIP model. Additionally, based on SciScore, we propose a two-stage training framework, comprising a supervised fine-tuning phase and a masked online fine-tuning phase, to incorporate scientific knowledge into existing generative models. Through comprehensive experiments, we demonstrate the effectiveness of our framework in establishing new standards for evaluating the scientific realism of generated content. Specifically, SciScore attains performance comparable to human-level, demonstrating a 5% improvement similar to evaluations conducted by experienced human evaluators. Furthermore, by applying our proposed fine-tuning method to FLUX, we achieve a performance enhancement exceeding 50% on SciScore.
comment: Accepted to CVPR 2025. Code, docs, weight, benchmark and training data are all avaliable at https://jialuo-li.github.io/Science-T2I-Web
☆ FreshStack: Building Realistic Benchmarks for Evaluating Retrieval on Technical Documents
We introduce FreshStack, a reusable framework for automatically building information retrieval (IR) evaluation benchmarks from community-asked questions and answers. FreshStack conducts the following steps: (1) automatic corpus collection from code and technical documentation, (2) nugget generation from community-asked questions and answers, and (3) nugget-level support, retrieving documents using a fusion of retrieval techniques and hybrid architectures. We use FreshStack to build five datasets on fast-growing, recent, and niche topics to ensure the tasks are sufficiently challenging. On FreshStack, existing retrieval models, when applied out-of-the-box, significantly underperform oracle approaches on all five topics, denoting plenty of headroom to improve IR quality. In addition, we identify cases where rerankers do not clearly improve first-stage retrieval accuracy (two out of five topics). We hope that FreshStack will facilitate future work toward constructing realistic, scalable, and uncontaminated IR and RAG evaluation benchmarks. FreshStack datasets are available at: https://fresh-stack.github.io.
☆ LLMs Meet Finance: Fine-Tuning Foundation Models for the Open FinLLM Leaderboard
This paper investigates the application of large language models (LLMs) to financial tasks. We fine-tuned foundation models using the Open FinLLM Leaderboard as a benchmark. Building on Qwen2.5 and Deepseek-R1, we employed techniques including supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement learning (RL) to enhance their financial capabilities. The fine-tuned models demonstrated substantial performance gains across a wide range of financial tasks. Moreover, we measured the data scaling law in the financial domain. Our work demonstrates the potential of large language models (LLMs) in financial applications.
☆ Low-hallucination Synthetic Captions for Large-Scale Vision-Language Model Pre-training
In recent years, the field of vision-language model pre-training has experienced rapid advancements, driven primarily by the continuous enhancement of textual capabilities in large language models. However, existing training paradigms for multimodal large language models heavily rely on high-quality image-text pairs. As models and data scales grow exponentially, the availability of such meticulously curated data has become increasingly scarce and saturated, thereby severely limiting further advancements in this domain. This study investigates scalable caption generation techniques for vision-language model pre-training and demonstrates that large-scale low-hallucination synthetic captions can serve dual purposes: 1) acting as a viable alternative to real-world data for pre-training paradigms and 2) achieving superior performance enhancement when integrated into vision-language models through empirical validation. This paper presents three key contributions: 1) a novel pipeline for generating high-quality, low-hallucination, and knowledge-rich synthetic captions. Our continuous DPO methodology yields remarkable results in reducing hallucinations. Specifically, the non-hallucination caption rate on a held-out test set increases from 48.2% to 77.9% for a 7B-size model. 2) Comprehensive empirical validation reveals that our synthetic captions confer superior pre-training advantages over their counterparts. Across 35 vision language tasks, the model trained with our data achieves a significant performance gain of at least 6.2% compared to alt-text pairs and other previous work. Meanwhile, it also offers considerable support in the text-to-image domain. With our dataset, the FID score is reduced by 17.1 on a real-world validation benchmark and 13.3 on the MSCOCO validation benchmark. 3) We will release Hunyuan-Recap100M, a low-hallucination and knowledge-intensive synthetic caption dataset.
☆ Probing and Inducing Combinational Creativity in Vision-Language Models
The ability to combine existing concepts into novel ideas stands as a fundamental hallmark of human intelligence. Recent advances in Vision-Language Models (VLMs) like GPT-4V and DALLE-3 have sparked debate about whether their outputs reflect combinational creativity--defined by M. A. Boden (1998) as synthesizing novel ideas through combining existing concepts--or sophisticated pattern matching of training data. Drawing inspiration from cognitive science, we investigate the combinational creativity of VLMs from the lens of concept blending. We propose the Identification-Explanation-Implication (IEI) framework, which decomposes creative processes into three levels: identifying input spaces, extracting shared attributes, and deriving novel semantic implications. To validate this framework, we curate CreativeMashup, a high-quality dataset of 666 artist-generated visual mashups annotated according to the IEI framework. Through extensive experiments, we demonstrate that in comprehension tasks, best VLMs have surpassed average human performance while falling short of expert-level understanding; in generation tasks, incorporating our IEI framework into the generation pipeline significantly enhances the creative quality of VLMs outputs. Our findings establish both a theoretical foundation for evaluating artificial creativity and practical guidelines for improving creative generation in VLMs.
comment: Project page: https://ppyyqq.github.io/aicc/ The first two authors contribute equally
☆ A Multi-task Learning Balanced Attention Convolutional Neural Network Model for Few-shot Underwater Acoustic Target Recognition
Underwater acoustic target recognition (UATR) is of great significance for the protection of marine diversity and national defense security. The development of deep learning provides new opportunities for UATR, but faces challenges brought by the scarcity of reference samples and complex environmental interference. To address these issues, we proposes a multi-task balanced channel attention convolutional neural network (MT-BCA-CNN). The method integrates a channel attention mechanism with a multi-task learning strategy, constructing a shared feature extractor and multi-task classifiers to jointly optimize target classification and feature reconstruction tasks. The channel attention mechanism dynamically enhances discriminative acoustic features such as harmonic structures while suppressing noise. Experiments on the Watkins Marine Life Dataset demonstrate that MT-BCA-CNN achieves 97\% classification accuracy and 95\% $F1$-score in 27-class few-shot scenarios, significantly outperforming traditional CNN and ACNN models, as well as popular state-of-the-art UATR methods. Ablation studies confirm the synergistic benefits of multi-task learning and attention mechanisms, while a dynamic weighting adjustment strategy effectively balances task contributions. This work provides an efficient solution for few-shot underwater acoustic recognition, advancing research in marine bioacoustics and sonar signal processing.
☆ An Empirically Grounded Identifiability Theory Will Accelerate Self-Supervised Learning Research
Self-Supervised Learning (SSL) powers many current AI systems. As research interest and investment grow, the SSL design space continues to expand. The Platonic view of SSL, following the Platonic Representation Hypothesis (PRH), suggests that despite different methods and engineering approaches, all representations converge to the same Platonic ideal. However, this phenomenon lacks precise theoretical explanation. By synthesizing evidence from Identifiability Theory (IT), we show that the PRH can emerge in SSL. However, current IT cannot explain SSL's empirical success. To bridge the gap between theory and practice, we propose expanding IT into what we term Singular Identifiability Theory (SITh), a broader theoretical framework encompassing the entire SSL pipeline. SITh would allow deeper insights into the implicit data assumptions in SSL and advance the field towards learning more interpretable and generalizable representations. We highlight three critical directions for future research: 1) training dynamics and convergence properties of SSL; 2) the impact of finite samples, batch size, and data diversity; and 3) the role of inductive biases in architecture, augmentations, initialization schemes, and optimizers.
☆ Retrieval-Augmented Generation with Conflicting Evidence
Large language model (LLM) agents are increasingly employing retrieval-augmented generation (RAG) to improve the factuality of their responses. However, in practice, these systems often need to handle ambiguous user queries and potentially conflicting information from multiple sources while also suppressing inaccurate information from noisy or irrelevant documents. Prior work has generally studied and addressed these challenges in isolation, considering only one aspect at a time, such as handling ambiguity or robustness to noise and misinformation. We instead consider multiple factors simultaneously, proposing (i) RAMDocs (Retrieval with Ambiguity and Misinformation in Documents), a new dataset that simulates complex and realistic scenarios for conflicting evidence for a user query, including ambiguity, misinformation, and noise; and (ii) MADAM-RAG, a multi-agent approach in which LLM agents debate over the merits of an answer over multiple rounds, allowing an aggregator to collate responses corresponding to disambiguated entities while discarding misinformation and noise, thereby handling diverse sources of conflict jointly. We demonstrate the effectiveness of MADAM-RAG using both closed and open-source models on AmbigDocs -- which requires presenting all valid answers for ambiguous queries -- improving over strong RAG baselines by up to 11.40% and on FaithEval -- which requires suppressing misinformation -- where we improve by up to 15.80% (absolute) with Llama3.3-70B-Instruct. Furthermore, we find that RAMDocs poses a challenge for existing RAG baselines (Llama3.3-70B-Instruct only obtains 32.60 exact match score). While MADAM-RAG begins to address these conflicting factors, our analysis indicates that a substantial gap remains especially when increasing the level of imbalance in supporting evidence and misinformation.
comment: Our data and code is available at: https://github.com/HanNight/RAMDocs
☆ Enhancing Person-to-Person Virtual Try-On with Multi-Garment Virtual Try-Off
Computer vision is transforming fashion through Virtual Try-On (VTON) and Virtual Try-Off (VTOFF). VTON generates images of a person in a specified garment using a target photo and a standardized garment image, while a more challenging variant, Person-to-Person Virtual Try-On (p2p-VTON), uses a photo of another person wearing the garment. VTOFF, on the other hand, extracts standardized garment images from clothed individuals. We introduce TryOffDiff, a diffusion-based VTOFF model. Built on a latent diffusion framework with SigLIP image conditioning, it effectively captures garment properties like texture, shape, and patterns. TryOffDiff achieves state-of-the-art results on VITON-HD and strong performance on DressCode dataset, covering upper-body, lower-body, and dresses. Enhanced with class-specific embeddings, it pioneers multi-garment VTOFF, the first of its kind. When paired with VTON models, it improves p2p-VTON by minimizing unwanted attribute transfer, such as skin color. Code is available at: https://rizavelioglu.github.io/tryoffdiff/
☆ Accuracy is Not Agreement: Expert-Aligned Evaluation of Crash Narrative Classification Models
This study explores the relationship between deep learning (DL) model accuracy and expert agreement in the classification of crash narratives. We evaluate five DL models -- including BERT variants, the Universal Sentence Encoder (USE), and a zero-shot classifier -- against expert-labeled data and narrative text. The analysis is further extended to four large language models (LLMs): GPT-4, LLaMA 3, Qwen, and Claude. Our results reveal a counterintuitive trend: models with higher technical accuracy often exhibit lower agreement with domain experts, whereas LLMs demonstrate greater expert alignment despite relatively lower accuracy scores. To quantify and interpret model-expert agreement, we employ Cohen's Kappa, Principal Component Analysis (PCA), and SHAP-based explainability techniques. Findings indicate that expert-aligned models tend to rely more on contextual and temporal language cues, rather than location-specific keywords. These results underscore that accuracy alone is insufficient for evaluating models in safety-critical NLP applications. We advocate for incorporating expert agreement as a complementary metric in model evaluation frameworks and highlight the promise of LLMs as interpretable, scalable tools for crash analysis pipelines.
☆ RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins CVPR 2025
In the rapidly advancing field of robotics, dual-arm coordination and complex object manipulation are essential capabilities for developing advanced autonomous systems. However, the scarcity of diverse, high-quality demonstration data and real-world-aligned evaluation benchmarks severely limits such development. To address this, we introduce RoboTwin, a generative digital twin framework that uses 3D generative foundation models and large language models to produce diverse expert datasets and provide a real-world-aligned evaluation platform for dual-arm robotic tasks. Specifically, RoboTwin creates varied digital twins of objects from single 2D images, generating realistic and interactive scenarios. It also introduces a spatial relation-aware code generation framework that combines object annotations with large language models to break down tasks, determine spatial constraints, and generate precise robotic movement code. Our framework offers a comprehensive benchmark with both simulated and real-world data, enabling standardized evaluation and better alignment between simulated training and real-world performance. We validated our approach using the open-source COBOT Magic Robot platform. Policies pre-trained on RoboTwin-generated data and fine-tuned with limited real-world samples demonstrate significant potential for enhancing dual-arm robotic manipulation systems by improving success rates by over 70% for single-arm tasks and over 40% for dual-arm tasks compared to models trained solely on real-world data.
comment: CVPR 2025 Highlight. 22 pages. Project page: https://robotwin-benchmark.github.io/
☆ Aspect-Based Summarization with Self-Aspect Retrieval Enhanced Generation
Aspect-based summarization aims to generate summaries tailored to specific aspects, addressing the resource constraints and limited generalizability of traditional summarization approaches. Recently, large language models have shown promise in this task without the need for training. However, they rely excessively on prompt engineering and face token limits and hallucination challenges, especially with in-context learning. To address these challenges, in this paper, we propose a novel framework for aspect-based summarization: Self-Aspect Retrieval Enhanced Summary Generation. Rather than relying solely on in-context learning, given an aspect, we employ an embedding-driven retrieval mechanism to identify its relevant text segments. This approach extracts the pertinent content while avoiding unnecessary details, thereby mitigating the challenge of token limits. Moreover, our framework optimizes token usage by deleting unrelated parts of the text and ensuring that the model generates output strictly based on the given aspect. With extensive experiments on benchmark datasets, we demonstrate that our framework not only achieves superior performance but also effectively mitigates the token limitation problem.
☆ Design Topological Materials by Reinforcement Fine-Tuned Generative Model
Topological insulators (TIs) and topological crystalline insulators (TCIs) are materials with unconventional electronic properties, making their discovery highly valuable for practical applications. However, such materials, particularly those with a full band gap, remain scarce. Given the limitations of traditional approaches that scan known materials for candidates, we focus on the generation of new topological materials through a generative model. Specifically, we apply reinforcement fine-tuning (ReFT) to a pre-trained generative model, thereby aligning the model's objectives with our material design goals. We demonstrate that ReFT is effective in enhancing the model's ability to generate TIs and TCIs, with minimal compromise on the stability of the generated materials. Using the fine-tuned model, we successfully identify a large number of new topological materials, with Ge$_2$Bi$_2$O$_6$ serving as a representative example--a TI with a full band gap of 0.26 eV, ranking among the largest known in this category.
☆ Event-Enhanced Blurry Video Super-Resolution AAAI 2025
In this paper, we tackle the task of blurry video super-resolution (BVSR), aiming to generate high-resolution (HR) videos from low-resolution (LR) and blurry inputs. Current BVSR methods often fail to restore sharp details at high resolutions, resulting in noticeable artifacts and jitter due to insufficient motion information for deconvolution and the lack of high-frequency details in LR frames. To address these challenges, we introduce event signals into BVSR and propose a novel event-enhanced network, Ev-DeblurVSR. To effectively fuse information from frames and events for feature deblurring, we introduce a reciprocal feature deblurring module that leverages motion information from intra-frame events to deblur frame features while reciprocally using global scene context from the frames to enhance event features. Furthermore, to enhance temporal consistency, we propose a hybrid deformable alignment module that fully exploits the complementary motion information from inter-frame events and optical flow to improve motion estimation in the deformable alignment process. Extensive evaluations demonstrate that Ev-DeblurVSR establishes a new state-of-the-art performance on both synthetic and real-world datasets. Notably, on real data, our method is +2.59 dB more accurate and 7.28$\times$ faster than the recent best BVSR baseline FMA-Net. Code: https://github.com/DachunKai/Ev-DeblurVSR.
comment: AAAI 2025. Project page: https://dachunkai.github.io/evtexture.github.io/
☆ Towards Cardiac MRI Foundation Models: Comprehensive Visual-Tabular Representations for Whole-Heart Assessment and Beyond
Cardiac magnetic resonance imaging is the gold standard for non-invasive cardiac assessment, offering rich spatio-temporal views of the cardiac anatomy and physiology. Patient-level health factors, such as demographics, metabolic, and lifestyle, are known to substantially influence cardiovascular health and disease risk, yet remain uncaptured by CMR alone. To holistically understand cardiac health and to enable the best possible interpretation of an individual's disease risk, CMR and patient-level factors must be jointly exploited within an integrated framework. Recent multi-modal approaches have begun to bridge this gap, yet they often rely on limited spatio-temporal data and focus on isolated clinical tasks, thereby hindering the development of a comprehensive representation for cardiac health evaluation. To overcome these limitations, we introduce ViTa, a step toward foundation models that delivers a comprehensive representation of the heart and a precise interpretation of individual disease risk. Leveraging data from 42,000 UK Biobank participants, ViTa integrates 3D+T cine stacks from short-axis and long-axis views, enabling a complete capture of the cardiac cycle. These imaging data are then fused with detailed tabular patient-level factors, enabling context-aware insights. This multi-modal paradigm supports a wide spectrum of downstream tasks, including cardiac phenotype and physiological feature prediction, segmentation, and classification of cardiac and metabolic diseases within a single unified framework. By learning a shared latent representation that bridges rich imaging features and patient context, ViTa moves beyond traditional, task-specific models toward a universal, patient-specific understanding of cardiac health, highlighting its potential to advance clinical utility and scalability in cardiac analysis.
☆ Prototypes are Balanced Units for Efficient and Effective Partially Relevant Video Retrieval
In a retrieval system, simultaneously achieving search accuracy and efficiency is inherently challenging. This challenge is particularly pronounced in partially relevant video retrieval (PRVR), where incorporating more diverse context representations at varying temporal scales for each video enhances accuracy but increases computational and memory costs. To address this dichotomy, we propose a prototypical PRVR framework that encodes diverse contexts within a video into a fixed number of prototypes. We then introduce several strategies to enhance text association and video understanding within the prototypes, along with an orthogonal objective to ensure that the prototypes capture a diverse range of content. To keep the prototypes searchable via text queries while accurately encoding video contexts, we implement cross- and uni-modal reconstruction tasks. The cross-modal reconstruction task aligns the prototypes with textual features within a shared space, while the uni-modal reconstruction task preserves all video contexts during encoding. Additionally, we employ a video mixing technique to provide weak guidance to further align prototypes and associated textual representations. Extensive evaluations on TVR, ActivityNet-Captions, and QVHighlights validate the effectiveness of our approach without sacrificing efficiency.
☆ InstructRAG: Leveraging Retrieval-Augmented Generation on Instruction Graphs for LLM-Based Task Planning SIGIR 2025
Recent advancements in large language models (LLMs) have enabled their use as agents for planning complex tasks. Existing methods typically rely on a thought-action-observation (TAO) process to enhance LLM performance, but these approaches are often constrained by the LLMs' limited knowledge of complex tasks. Retrieval-augmented generation (RAG) offers new opportunities by leveraging external databases to ground generation in retrieved information. In this paper, we identify two key challenges (enlargability and transferability) in applying RAG to task planning. We propose InstructRAG, a novel solution within a multi-agent meta-reinforcement learning framework, to address these challenges. InstructRAG includes a graph to organize past instruction paths (sequences of correct actions), an RL-Agent with Reinforcement Learning to expand graph coverage for enlargability, and an ML-Agent with Meta-Learning to improve task generalization for transferability. The two agents are trained end-to-end to optimize overall planning performance. Our experiments on four widely used task planning datasets demonstrate that InstructRAG significantly enhances performance and adapts efficiently to new tasks, achieving up to a 19.2% improvement over the best existing approach.
comment: This paper has been accepted by SIGIR 2025
☆ Pose and Facial Expression Transfer by using StyleGAN
We propose a method to transfer pose and expression between face images. Given a source and target face portrait, the model produces an output image in which the pose and expression of the source face image are transferred onto the target identity. The architecture consists of two encoders and a mapping network that projects the two inputs into the latent space of StyleGAN2, which finally generates the output. The training is self-supervised from video sequences of many individuals. Manual labeling is not required. Our model enables the synthesis of random identities with controllable pose and expression. Close-to-real-time performance is achieved.
comment: Accepted at CVWW 2024. Presented in Terme Olimia, Slovenia
☆ SHA256 at SemEval-2025 Task 4: Selective Amnesia -- Constrained Unlearning for Large Language Models via Knowledge Isolation SemEval
Large language models (LLMs) frequently memorize sensitive information during training, posing risks when deploying publicly accessible models. Current machine unlearning methods struggle to selectively remove specific data associations without degrading overall model capabilities. This paper presents our solution to SemEval-2025 Task 4 on targeted unlearning, which introduces a two-stage methodology that combines causal mediation analysis with layer-specific optimization. Through systematic causal tracing experiments on OLMo architectures (1B and 7B parameters), we identify the critical role of the first few transformer layers (layers 0-5) in storing subject-attribute associations within MLP modules. Building on this insight, we develop a constrained optimization approach that freezes upper layers while applying a novel joint loss function to lower layers-simultaneously maximizing forget set loss via output token cross-entropy penalties and minimizing retain set deviation through adaptive regularization. Our method achieves 2nd place in the 1B model track, demonstrating strong task performance while maintaining 88% of baseline MMLU accuracy. These results establish causal-informed layer optimization as a promising paradigm for efficient, precise unlearning in LLMs, offering a significant step forward in addressing data privacy concerns in AI systems.
comment: 8 pages, In Proceedings of The 19th International Workshop on Semantic Evaluation (SemEval), 2025
☆ A Virtual Machine for Arbitrary Low-Precision GPGPU Computation in LLM Serving
Serving Large Language Models (LLMs) is critical for AI-powered applications but demands substantial computational resources, particularly in memory bandwidth and computational throughput. Low-precision computation has emerged as a key technique to improve efficiency while reducing resource consumption. Existing approaches for generating low-precision kernels are limited to weight bit widths that are powers of two and suffer from suboptimal performance due to high-level GPU programming abstractions. These abstractions restrict critical optimizations, such as fine-grained register management and optimized memory access patterns, which are essential for efficient low-precision computations. In this paper, we introduce a virtual machine (VM) designed for General-Purpose GPU (GPGPU) computing, enabling support for low-precision data types with arbitrary bit widths while maintaining GPU programmability. The proposed VM features a thread-block-level programming model, a hierarchical memory space, a novel algebraic layout system, and extensive support for diverse low-precision data types. VM programs are compiled into highly efficient GPU programs with automatic vectorization and instruction selection. Extensive experiments demonstrate that our VM efficiently supports a full spectrum of low-precision data types, and outperforms state-of-the-art low-precision kernels on their supported types. Compared to existing compilers like Triton and Ladder, as well as hand-optimized kernels such as QuantLLM and Marlin, our VM achieves performance improvements of 1.75x, 2.61x, 1.29x and 1.03x, respectively.
comment: 18 pages, 15 figures
☆ Accommodate Knowledge Conflicts in Retrieval-augmented LLMs: Towards Reliable Response Generation in the Wild
The proliferation of large language models (LLMs) has significantly advanced information retrieval systems, particularly in response generation (RG). Unfortunately, LLMs often face knowledge conflicts between internal memory and retrievaled external information, arising from misinformation, biases, or outdated knowledge. These conflicts undermine response reliability and introduce uncertainty in decision-making. In this work, we analyze how LLMs navigate knowledge conflicts from an information-theoretic perspective and reveal that when conflicting and supplementary information exhibit significant differences, LLMs confidently resolve their preferences. However, when the distinction is ambiguous, LLMs experience heightened uncertainty. Based on this insight, we propose Swin-VIB, a novel framework that integrates a pipeline of variational information bottleneck models into adaptive augmentation of retrieved information and guiding LLM preference in response generation. Extensive experiments on single-choice, open-ended question-answering (QA), and retrieval augmented generation (RAG) validate our theoretical findings and demonstrate the efficacy of Swin-VIB. Notably, our method improves single-choice task accuracy by at least 7.54\% over competitive baselines.
☆ A Phenomenological Approach to Analyzing User Queries in IT Systems Using Heidegger's Fundamental Ontology
This paper presents a novel research analytical IT system grounded in Martin Heidegger's Fundamental Ontology, distinguishing between beings (das Seiende) and Being (das Sein). The system employs two modally distinct, descriptively complete languages: a categorical language of beings for processing user inputs and an existential language of Being for internal analysis. These languages are bridged via a phenomenological reduction module, enabling the system to analyze user queries (including questions, answers, and dialogues among IT specialists), identify recursive and self-referential structures, and provide actionable insights in categorical terms. Unlike contemporary systems limited to categorical analysis, this approach leverages Heidegger's phenomenological existential analysis to uncover deeper ontological patterns in query processing, aiding in resolving logical traps in complex interactions, such as metaphor usage in IT contexts. The path to full realization involves formalizing the language of Being by a research team based on Heidegger's Fundamental Ontology; given the existing completeness of the language of beings, this reduces the system's computability to completeness, paving the way for a universal query analysis tool. The paper presents the system's architecture, operational principles, technical implementation, use cases--including a case based on real IT specialist dialogues--comparative evaluation with existing tools, and its advantages and limitations.
comment: 12 pages, no figures
☆ Transferrable Surrogates in Expressive Neural Architecture Search Spaces
Neural architecture search (NAS) faces a challenge in balancing the exploration of expressive, broad search spaces that enable architectural innovation with the need for efficient evaluation of architectures to effectively search such spaces. We investigate surrogate model training for improving search in highly expressive NAS search spaces based on context-free grammars. We show that i) surrogate models trained either using zero-cost-proxy metrics and neural graph features (GRAF) or by fine-tuning an off-the-shelf LM have high predictive power for the performance of architectures both within and across datasets, ii) these surrogates can be used to filter out bad architectures when searching on novel datasets, thereby significantly speeding up search and achieving better final performances, and iii) the surrogates can be further used directly as the search objective for huge speed-ups.
comment: Project page at: https://shiwenqin.github.io/TransferrableSurrogate/
☆ QLLM: Do We Really Need a Mixing Network for Credit Assignment in Multi-Agent Reinforcement Learning?
Credit assignment has remained a fundamental challenge in multi-agent reinforcement learning (MARL). Previous studies have primarily addressed this issue through value decomposition methods under the centralized training with decentralized execution paradigm, where neural networks are utilized to approximate the nonlinear relationship between individual Q-values and the global Q-value. Although these approaches have achieved considerable success in various benchmark tasks, they still suffer from several limitations, including imprecise attribution of contributions, limited interpretability, and poor scalability in high-dimensional state spaces. To address these challenges, we propose a novel algorithm, \textbf{QLLM}, which facilitates the automatic construction of credit assignment functions using large language models (LLMs). Specifically, the concept of \textbf{TFCAF} is introduced, wherein the credit allocation process is represented as a direct and expressive nonlinear functional formulation. A custom-designed \textit{coder-evaluator} framework is further employed to guide the generation, verification, and refinement of executable code by LLMs, significantly mitigating issues such as hallucination and shallow reasoning during inference. Extensive experiments conducted on several standard MARL benchmarks demonstrate that the proposed method consistently outperforms existing state-of-the-art baselines. Moreover, QLLM exhibits strong generalization capability and maintains compatibility with a wide range of MARL algorithms that utilize mixing networks, positioning it as a promising and versatile solution for complex multi-agent scenarios.
comment: 9 pages, 7 figures
☆ Are Retrials All You Need? Enhancing Large Language Model Reasoning Without Verbalized Feedback
Recent advancements in large language models (LLMs) have catalyzed the development of general-purpose autonomous agents, demonstrating remarkable performance in complex reasoning tasks across various domains. This surge has spurred the evolution of a plethora of prompt-based reasoning frameworks. A recent focus has been on iterative reasoning strategies that refine outputs through self-evaluation and verbalized feedback. However, these strategies require additional computational complexity to enable models to recognize and correct their mistakes, leading to a significant increase in their cost. In this work, we introduce the concept of ``retrials without feedback'', an embarrassingly simple yet powerful mechanism for enhancing reasoning frameworks by allowing LLMs to retry problem-solving attempts upon identifying incorrect answers. Unlike conventional iterative refinement methods, our method does not require explicit self-reflection or verbalized feedback, simplifying the refinement process. Our findings indicate that simpler retrial-based approaches often outperform more sophisticated reasoning frameworks, suggesting that the benefits of complex methods may not always justify their computational costs. By challenging the prevailing assumption that more intricate reasoning strategies inherently lead to better performance, our work offers new insights into how simpler, more efficient approaches can achieve optimal results. So, are retrials all you need?
comment: 8 pages, 16 figures, 1 table. arXiv admin note: text overlap with arXiv:2405.06691
☆ Benchmarking Multi-National Value Alignment for Large Language Models
Do Large Language Models (LLMs) hold positions that conflict with your country's values? Occasionally they do! However, existing works primarily focus on ethical reviews, failing to capture the diversity of national values, which encompass broader policy, legal, and moral considerations. Furthermore, current benchmarks that rely on spectrum tests using manually designed questionnaires are not easily scalable. To address these limitations, we introduce NaVAB, a comprehensive benchmark to evaluate the alignment of LLMs with the values of five major nations: China, the United States, the United Kingdom, France, and Germany. NaVAB implements a national value extraction pipeline to efficiently construct value assessment datasets. Specifically, we propose a modeling procedure with instruction tagging to process raw data sources, a screening process to filter value-related topics and a generation process with a Conflict Reduction mechanism to filter non-conflicting values.We conduct extensive experiments on various LLMs across countries, and the results provide insights into assisting in the identification of misaligned scenarios. Moreover, we demonstrate that NaVAB can be combined with alignment techniques to effectively reduce value concerns by aligning LLMs' values with the target country.
☆ Information Gain-Guided Causal Intervention for Autonomous Debiasing Large Language Models
Despite significant progress, recent studies indicate that current large language models (LLMs) may still capture dataset biases and utilize them during inference, leading to the poor generalizability of LLMs. However, due to the diversity of dataset biases and the insufficient nature of bias suppression based on in-context learning, the effectiveness of previous prior knowledge-based debiasing methods and in-context learning based automatic debiasing methods is limited. To address these challenges, we explore the combination of causal mechanisms with information theory and propose an information gain-guided causal intervention debiasing (IGCIDB) framework. This framework first utilizes an information gain-guided causal intervention method to automatically and autonomously balance the distribution of instruction-tuning dataset. Subsequently, it employs a standard supervised fine-tuning process to train LLMs on the debiased dataset. Experimental results show that IGCIDB can effectively debias LLM to improve its generalizability across different tasks.
☆ Are AI agents the new machine translation frontier? Challenges and opportunities of single- and multi-agent systems for multilingual digital communication
The rapid evolution of artificial intelligence (AI) has introduced AI agents as a disruptive paradigm across various industries, yet their application in machine translation (MT) remains underexplored. This paper describes and analyses the potential of single- and multi-agent systems for MT, reflecting on how they could enhance multilingual digital communication. While single-agent systems are well-suited for simpler translation tasks, multi-agent systems, which involve multiple specialized AI agents collaborating in a structured manner, may offer a promising solution for complex scenarios requiring high accuracy, domain-specific knowledge, and contextual awareness. To demonstrate the feasibility of multi-agent workflows in MT, we are conducting a pilot study in legal MT. The study employs a multi-agent system involving four specialized AI agents for (i) translation, (ii) adequacy review, (iii) fluency review, and (iv) final editing. Our findings suggest that multi-agent systems may have the potential to significantly improve domain-adaptability and contextual awareness, with superior translation quality to traditional MT or single-agent systems. This paper also sets the stage for future research into multi-agent applications in MT, integration into professional translation workflows, and shares a demo of the system analyzed in the paper.
☆ EmoVoice: LLM-based Emotional Text-To-Speech Model with Freestyle Text Prompting
Human speech goes beyond the mere transfer of information; it is a profound exchange of emotions and a connection between individuals. While Text-to-Speech (TTS) models have made huge progress, they still face challenges in controlling the emotional expression in the generated speech. In this work, we propose EmoVoice, a novel emotion-controllable TTS model that exploits large language models (LLMs) to enable fine-grained freestyle natural language emotion control, and a phoneme boost variant design that makes the model output phoneme tokens and audio tokens in parallel to enhance content consistency, inspired by chain-of-thought (CoT) and modality-of-thought (CoM) techniques. Besides, we introduce EmoVoice-DB, a high-quality 40-hour English emotion dataset featuring expressive speech and fine-grained emotion labels with natural language descriptions. EmoVoice achieves state-of-the-art performance on the English EmoVoice-DB test set using only synthetic training data, and on the Chinese Secap test set using our in-house data. We further investigate the reliability of existing emotion evaluation metrics and their alignment with human perceptual preferences, and explore using SOTA multimodal LLMs GPT-4o-audio and Gemini to assess emotional speech. Demo samples are available at https://anonymous.4open.science/r/EmoVoice-DF55. Dataset, code, and checkpoints will be released.
☆ 3D-PNAS: 3D Industrial Surface Anomaly Synthesis with Perlin Noise
Large pretrained vision foundation models have shown significant potential in various vision tasks. However, for industrial anomaly detection, the scarcity of real defect samples poses a critical challenge in leveraging these models. While 2D anomaly generation has significantly advanced with established generative models, the adoption of 3D sensors in industrial manufacturing has made leveraging 3D data for surface quality inspection an emerging trend. In contrast to 2D techniques, 3D anomaly generation remains largely unexplored, limiting the potential of 3D data in industrial quality inspection. To address this gap, we propose a novel yet simple 3D anomaly generation method, 3D-PNAS, based on Perlin noise and surface parameterization. Our method generates realistic 3D surface anomalies by projecting the point cloud onto a 2D plane, sampling multi-scale noise values from a Perlin noise field, and perturbing the point cloud along its normal direction. Through comprehensive visualization experiments, we demonstrate how key parameters - including noise scale, perturbation strength, and octaves, provide fine-grained control over the generated anomalies, enabling the creation of diverse defect patterns from pronounced deformations to subtle surface variations. Additionally, our cross-category experiments show that the method produces consistent yet geometrically plausible anomalies across different object types, adapting to their specific surface characteristics. We also provide a comprehensive codebase and visualization toolkit to facilitate future research.
☆ ALT: A Python Package for Lightweight Feature Representation in Time Series Classification
We introduce ALT, an open-source Python package created for efficient and accurate time series classification (TSC). The package implements the adaptive law-based transformation (ALT) algorithm, which transforms raw time series data into a linearly separable feature space using variable-length shifted time windows. This adaptive approach enhances its predecessor, the linear law-based transformation (LLT), by effectively capturing patterns of varying temporal scales. The software is implemented for scalability, interpretability, and ease of use, achieving state-of-the-art performance with minimal computational overhead. Extensive benchmarking on real-world datasets demonstrates the utility of ALT for diverse TSC tasks in physics and related domains.
comment: 16 pages, 4 figures
☆ Image-Editing Specialists: An RLAIF Approach for Diffusion Models
We present a novel approach to training specialized instruction-based image-editing diffusion models, addressing key challenges in structural preservation with input images and semantic alignment with user prompts. We introduce an online reinforcement learning framework that aligns the diffusion model with human preferences without relying on extensive human annotations or curating a large dataset. Our method significantly improves the realism and alignment with instructions in two ways. First, the proposed models achieve precise and structurally coherent modifications in complex scenes while maintaining high fidelity in instruction-irrelevant areas. Second, they capture fine nuances in the desired edit by leveraging a visual prompt, enabling detailed control over visual edits without lengthy textual prompts. This approach simplifies users' efforts to achieve highly specific edits, requiring only 5 reference images depicting a certain concept for training. Experimental results demonstrate that our models can perform intricate edits in complex scenes, after just 10 training steps. Finally, we showcase the versatility of our method by applying it to robotics, where enhancing the visual realism of simulated environments through targeted sim-to-real image edits improves their utility as proxies for real-world settings.
☆ Explainable Scene Understanding with Qualitative Representations and Graph Neural Networks
This paper investigates the integration of graph neural networks (GNNs) with Qualitative Explainable Graphs (QXGs) for scene understanding in automated driving. Scene understanding is the basis for any further reactive or proactive decision-making. Scene understanding and related reasoning is inherently an explanation task: why is another traffic participant doing something, what or who caused their actions? While previous work demonstrated QXGs' effectiveness using shallow machine learning models, these approaches were limited to analysing single relation chains between object pairs, disregarding the broader scene context. We propose a novel GNN architecture that processes entire graph structures to identify relevant objects in traffic scenes. We evaluate our method on the nuScenes dataset enriched with DriveLM's human-annotated relevance labels. Experimental results show that our GNN-based approach achieves superior performance compared to baseline methods. The model effectively handles the inherent class imbalance in relevant object identification tasks while considering the complete spatial-temporal relationships between all objects in the scene. Our work demonstrates the potential of combining qualitative representations with deep learning approaches for explainable scene understanding in autonomous driving systems.
comment: Workshop "Advancing Automated Driving in Highly Interactive Scenarios through Behavior Prediction, Trustworthy AI, and Remote Operations" @ 36th IEEE Intelligent Vehicles Symposium (IV)
☆ Hybrid Dense-UNet201 Optimization for Pap Smear Image Segmentation Using Spider Monkey Optimization
Pap smear image segmentation is crucial for cervical cancer diagnosis. However, traditional segmentation models often struggle with complex cellular structures and variations in pap smear images. This study proposes a hybrid Dense-UNet201 optimization approach that integrates a pretrained DenseNet201 as the encoder for the U-Net architecture and optimizes it using the spider monkey optimization (SMO) algorithm. The Dense-UNet201 model excelled at feature extraction. The SMO was modified to handle categorical and discrete parameters. The SIPaKMeD dataset was used in this study and evaluated using key performance metrics, including loss, accuracy, Intersection over Union (IoU), and Dice coefficient. The experimental results showed that Dense-UNet201 outperformed U-Net, Res-UNet50, and Efficient-UNetB0. SMO Dense-UNet201 achieved a segmentation accuracy of 96.16%, an IoU of 91.63%, and a Dice coefficient score of 95.63%. These findings underscore the effectiveness of image preprocessing, pretrained models, and metaheuristic optimization in improving medical image analysis and provide new insights into cervical cell segmentation methods.
☆ A Numerical Gradient Inversion Attack in Variational Quantum Neural-Networks
The loss landscape of Variational Quantum Neural Networks (VQNNs) is characterized by local minima that grow exponentially with increasing qubits. Because of this, it is more challenging to recover information from model gradients during training compared to classical Neural Networks (NNs). In this paper we present a numerical scheme that successfully reconstructs input training, real-world, practical data from trainable VQNNs' gradients. Our scheme is based on gradient inversion that works by combining gradients estimation with the finite difference method and adaptive low-pass filtering. The scheme is further optimized with Kalman filter to obtain efficient convergence. Our experiments show that our algorithm can invert even batch-trained data, given the VQNN model is sufficiently over-parameterized.
comment: 9 pages, 17 figures
☆ Enhancing Explainability and Reliable Decision-Making in Particle Swarm Optimization through Communication Topologies
Swarm intelligence effectively optimizes complex systems across fields like engineering and healthcare, yet algorithm solutions often suffer from low reliability due to unclear configurations and hyperparameters. This study analyzes Particle Swarm Optimization (PSO), focusing on how different communication topologies Ring, Star, and Von Neumann affect convergence and search behaviors. Using an adapted IOHxplainer , an explainable benchmarking tool, we investigate how these topologies influence information flow, diversity, and convergence speed, clarifying the balance between exploration and exploitation. Through visualization and statistical analysis, the research enhances interpretability of PSO's decisions and provides practical guidelines for choosing suitable topologies for specific optimization tasks. Ultimately, this contributes to making swarm based optimization more transparent, robust, and trustworthy.
☆ Set You Straight: Auto-Steering Denoising Trajectories to Sidestep Unwanted Concepts
Ensuring the ethical deployment of text-to-image models requires effective techniques to prevent the generation of harmful or inappropriate content. While concept erasure methods offer a promising solution, existing finetuning-based approaches suffer from notable limitations. Anchor-free methods risk disrupting sampling trajectories, leading to visual artifacts, while anchor-based methods rely on the heuristic selection of anchor concepts. To overcome these shortcomings, we introduce a finetuning framework, dubbed ANT, which Automatically guides deNoising Trajectories to avoid unwanted concepts. ANT is built on a key insight: reversing the condition direction of classifier-free guidance during mid-to-late denoising stages enables precise content modification without sacrificing early-stage structural integrity. This inspires a trajectory-aware objective that preserves the integrity of the early-stage score function field, which steers samples toward the natural image manifold, without relying on heuristic anchor concept selection. For single-concept erasure, we propose an augmentation-enhanced weight saliency map to precisely identify the critical parameters that most significantly contribute to the unwanted concept, enabling more thorough and efficient erasure. For multi-concept erasure, our objective function offers a versatile plug-and-play solution that significantly boosts performance. Extensive experiments demonstrate that ANT achieves state-of-the-art results in both single and multi-concept erasure, delivering high-quality, safe outputs without compromising the generative fidelity. Code is available at https://github.com/lileyang1210/ANT
comment: Preprint
☆ Towards Lossless Token Pruning in Late-Interaction Retrieval Models SIGIR 2025
Late interaction neural IR models like ColBERT offer a competitive effectiveness-efficiency trade-off across many benchmarks. However, they require a huge memory space to store the contextual representation for all the document tokens. Some works have proposed using either heuristics or statistical-based techniques to prune tokens from each document. This however doesn't guarantee that the removed tokens have no impact on the retrieval score. Our work uses a principled approach to define how to prune tokens without impacting the score between a document and a query. We introduce three regularization losses, that induce a solution with high pruning ratios, as well as two pruning strategies. We study them experimentally (in and out-domain), showing that we can preserve ColBERT's performance while using only 30\% of the tokens.
comment: Accepted at SIGIR 2025 Full Paper Track
☆ Multi-Agent Reinforcement Learning Simulation for Environmental Policy Synthesis AAMAS'25
Climate policy development faces significant challenges due to deep uncertainty, complex system dynamics, and competing stakeholder interests. Climate simulation methods, such as Earth System Models, have become valuable tools for policy exploration. However, their typical use is for evaluating potential polices, rather than directly synthesizing them. The problem can be inverted to optimize for policy pathways, but the traditional optimization approaches often struggle with non-linear dynamics, heterogeneous agents, and comprehensive uncertainty quantification. We propose a framework for augmenting climate simulations with Multi-Agent Reinforcement Learning (MARL) to address these limitations. We identify key challenges at the interface between climate simulations and the application of MARL in the context of policy synthesis, including reward definition, scalability with increasing agents and state spaces, uncertainty propagation across linked systems, and solution validation. Additionally, we discuss challenges in making MARL-derived solutions interpretable and useful for policy-makers. Our framework provides a foundation for more sophisticated climate policy exploration while acknowledging important limitations and areas for future research.
comment: Published in AAMAS'25 Blue Sky Ideas Track
☆ Enhancing the Geometric Problem-Solving Ability of Multimodal LLMs via Symbolic-Neural Integration
Recent advances in Multimodal Large Language Models (MLLMs) have achieved remarkable progress in general domains and demonstrated promise in multimodal mathematical reasoning. However, applying MLLMs to geometry problem solving (GPS) remains challenging due to lack of accurate step-by-step solution data and severe hallucinations during reasoning. In this paper, we propose GeoGen, a pipeline that can automatically generates step-wise reasoning paths for geometry diagrams. By leveraging the precise symbolic reasoning, \textbf{GeoGen} produces large-scale, high-quality question-answer pairs. To further enhance the logical reasoning ability of MLLMs, we train \textbf{GeoLogic}, a Large Language Model (LLM) using synthetic data generated by GeoGen. Serving as a bridge between natural language and symbolic systems, GeoLogic enables symbolic tools to help verifying MLLM outputs, making the reasoning process more rigorous and alleviating hallucinations. Experimental results show that our approach consistently improves the performance of MLLMs, achieving remarkable results on benchmarks for geometric reasoning tasks. This improvement stems from our integration of the strengths of LLMs and symbolic systems, which enables a more reliable and interpretable approach for the GPS task. Codes are available at https://github.com/ycpNotFound/GeoGen.
comment: 10 pages, 5 figures
☆ MCP Guardian: A Security-First Layer for Safeguarding MCP-Based AI System
As Agentic AI gain mainstream adoption, the industry invests heavily in model capabilities, achieving rapid leaps in reasoning and quality. However, these systems remain largely confined to data silos, and each new integration requires custom logic that is difficult to scale. The Model Context Protocol (MCP) addresses this challenge by defining a universal, open standard for securely connecting AI-based applications (MCP clients) to data sources (MCP servers). However, the flexibility of the MCP introduces new risks, including malicious tool servers and compromised data integrity. We present MCP Guardian, a framework that strengthens MCP-based communication with authentication, rate-limiting, logging, tracing, and Web Application Firewall (WAF) scanning. Through real-world scenarios and empirical testing, we demonstrate how MCP Guardian effectively mitigates attacks and ensures robust oversight with minimal overheads. Our approach fosters secure, scalable data access for AI assistants, underscoring the importance of a defense-in-depth approach that enables safer and more transparent innovation in AI-driven environments.
☆ Trajectory Adaptation using Large Language Models
Adapting robot trajectories based on human instructions as per new situations is essential for achieving more intuitive and scalable human-robot interactions. This work proposes a flexible language-based framework to adapt generic robotic trajectories produced by off-the-shelf motion planners like RRT, A-star, etc, or learned from human demonstrations. We utilize pre-trained LLMs to adapt trajectory waypoints by generating code as a policy for dense robot manipulation, enabling more complex and flexible instructions than current methods. This approach allows us to incorporate a broader range of commands, including numerical inputs. Compared to state-of-the-art feature-based sequence-to-sequence models which require training, our method does not require task-specific training and offers greater interpretability and more effective feedback mechanisms. We validate our approach through simulation experiments on the robotic manipulator, aerial vehicle, and ground robot in the Pybullet and Gazebo simulation environments, demonstrating that LLMs can successfully adapt trajectories to complex human instructions.
comment: Accepted to CoRL LangRob workshop 2024
☆ GPMFS: Global Foundation and Personalized Optimization for Multi-Label Feature Selection
As artificial intelligence methods are increasingly applied to complex task scenarios, high dimensional multi-label learning has emerged as a prominent research focus. At present, the curse of dimensionality remains one of the major bottlenecks in high-dimensional multi-label learning, which can be effectively addressed through multi-label feature selection methods. However, existing multi-label feature selection methods mostly focus on identifying global features shared across all labels, which overlooks personalized characteristics and specific requirements of individual labels. This global-only perspective may limit the ability to capture label-specific discriminative information, thereby affecting overall performance. In this paper, we propose a novel method called GPMFS (Global Foundation and Personalized Optimization for Multi-Label Feature Selection). GPMFS firstly identifies global features by exploiting label correlations, then adaptively supplements each label with a personalized subset of discriminative features using a threshold-controlled strategy. Experiments on multiple real-world datasets demonstrate that GPMFS achieves superior performance while maintaining strong interpretability and robustness. Furthermore, GPMFS provides insights into the label-specific strength across different multi-label datasets, thereby demonstrating the necessity and potential applicability of personalized feature selection approaches.
☆ The Athenian Academy: A Seven-Layer Architecture Model for Multi-Agent Systems
This paper proposes the "Academy of Athens" multi-agent seven-layer framework, aimed at systematically addressing challenges in multi-agent systems (MAS) within artificial intelligence (AI) art creation, such as collaboration efficiency, role allocation, environmental adaptation, and task parallelism. The framework divides MAS into seven layers: multi-agent collaboration, single-agent multi-role playing, single-agent multi-scene traversal, single-agent multi-capability incarnation, different single agents using the same large model to achieve the same target agent, single-agent using different large models to achieve the same target agent, and multi-agent synthesis of the same target agent. Through experimental validation in art creation, the framework demonstrates its unique advantages in task collaboration, cross-scene adaptation, and model fusion. This paper further discusses current challenges such as collaboration mechanism optimization, model stability, and system security, proposing future exploration through technologies like meta-learning and federated learning. The framework provides a structured methodology for multi-agent collaboration in AI art creation and promotes innovative applications in the art field.
☆ Pandora: A Code-Driven Large Language Model Agent for Unified Reasoning Across Diverse Structured Knowledge
Unified Structured Knowledge Reasoning (USKR) aims to answer natural language questions (NLQs) by using structured sources such as tables, databases, and knowledge graphs in a unified way. Existing USKR methods either rely on employing task-specific strategies or custom-defined representations, which struggle to leverage the knowledge transfer between different SKR tasks or align with the prior of LLMs, thereby limiting their performance. This paper proposes a novel USKR framework named \textsc{Pandora}, which takes advantage of \textsc{Python}'s \textsc{Pandas} API to construct a unified knowledge representation for alignment with LLM pre-training. It employs an LLM to generate textual reasoning steps and executable Python code for each question. Demonstrations are drawn from a memory of training examples that cover various SKR tasks, facilitating knowledge transfer. Extensive experiments on four benchmarks involving three SKR tasks demonstrate that \textsc{Pandora} outperforms existing unified frameworks and competes effectively with task-specific methods.
☆ SimUSER: Simulating User Behavior with Large Language Models for Recommender System Evaluation
Recommender systems play a central role in numerous real-life applications, yet evaluating their performance remains a significant challenge due to the gap between offline metrics and online behaviors. Given the scarcity and limits (e.g., privacy issues) of real user data, we introduce SimUSER, an agent framework that serves as believable and cost-effective human proxies. SimUSER first identifies self-consistent personas from historical data, enriching user profiles with unique backgrounds and personalities. Then, central to this evaluation are users equipped with persona, memory, perception, and brain modules, engaging in interactions with the recommender system. SimUSER exhibits closer alignment with genuine humans than prior work, both at micro and macro levels. Additionally, we conduct insightful experiments to explore the effects of thumbnails on click rates, the exposure effect, and the impact of reviews on user engagement. Finally, we refine recommender system parameters based on offline A/B test results, resulting in improved user engagement in the real world.
☆ TimeCapsule: Solving the Jigsaw Puzzle of Long-Term Time Series Forecasting with Compressed Predictive Representations
Recent deep learning models for Long-term Time Series Forecasting (LTSF) often emphasize complex, handcrafted designs, while simpler architectures like linear models or MLPs have often outperformed these intricate solutions. In this paper, we revisit and organize the core ideas behind several key techniques, such as redundancy reduction and multi-scale modeling, which are frequently employed in advanced LTSF models. Our goal is to streamline these ideas for more efficient deep learning utilization. To this end, we introduce TimeCapsule, a model built around the principle of high-dimensional information compression that unifies these techniques in a generalized yet simplified framework. Specifically, we model time series as a 3D tensor, incorporating temporal, variate, and level dimensions, and leverage mode production to capture multi-mode dependencies while achieving dimensionality compression. We propose an internal forecast within the compressed representation domain, supported by the Joint-Embedding Predictive Architecture (JEPA), to monitor the learning of predictive representations. Extensive experiments on challenging benchmarks demonstrate the versatility of our method, showing that TimeCapsule can achieve state-of-the-art performance.
☆ TUMLS: Trustful Fully Unsupervised Multi-Level Segmentation for Whole Slide Images of Histology
Digital pathology, augmented by artificial intelligence (AI), holds significant promise for improving the workflow of pathologists. However, challenges such as the labor-intensive annotation of whole slide images (WSIs), high computational demands, and trust concerns arising from the absence of uncertainty estimation in predictions hinder the practical application of current AI methodologies in histopathology. To address these issues, we present a novel trustful fully unsupervised multi-level segmentation methodology (TUMLS) for WSIs. TUMLS adopts an autoencoder (AE) as a feature extractor to identify the different tissue types within low-resolution training data. It selects representative patches from each identified group based on an uncertainty measure and then does unsupervised nuclei segmentation in their respective higher-resolution space without using any ML algorithms. Crucially, this solution integrates seamlessly into clinicians workflows, transforming the examination of a whole WSI into a review of concise, interpretable cross-level insights. This integration significantly enhances and accelerates the workflow while ensuring transparency. We evaluated our approach using the UPENN-GBM dataset, where the AE achieved a mean squared error (MSE) of 0.0016. Additionally, nucleus segmentation is assessed on the MoNuSeg dataset, outperforming all unsupervised approaches with an F1 score of 77.46% and a Jaccard score of 63.35%. These results demonstrate the efficacy of TUMLS in advancing the field of digital pathology.
comment: 32 pages, 15 figures, 3 tables, 42 references
☆ Post-pre-training for Modality Alignment in Vision-Language Foundation Models CVPR 2025
Contrastive language image pre-training (CLIP) is an essential component of building modern vision-language foundation models. While CLIP demonstrates remarkable zero-shot performance on downstream tasks, the multi-modal feature spaces still suffer from a modality gap, which is a gap between image and text feature clusters and limits downstream task performance. Although existing works attempt to address the modality gap by modifying pre-training or fine-tuning, they struggle with heavy training costs with large datasets or degradations of zero-shot performance. This paper presents CLIP-Refine, a post-pre-training method for CLIP models at a phase between pre-training and fine-tuning. CLIP-Refine aims to align the feature space with 1 epoch training on small image-text datasets without zero-shot performance degradations. To this end, we introduce two techniques: random feature alignment (RaFA) and hybrid contrastive-distillation (HyCD). RaFA aligns the image and text features to follow a shared prior distribution by minimizing the distance to random reference vectors sampled from the prior. HyCD updates the model with hybrid soft labels generated by combining ground-truth image-text pair labels and outputs from the pre-trained CLIP model. This contributes to achieving both maintaining the past knowledge and learning new knowledge to align features. Our extensive experiments with multiple classification and retrieval tasks show that CLIP-Refine succeeds in mitigating the modality gap and improving the zero-shot performance.
comment: Accepted to CVPR 2025; Code: https://github.com/yshinya6/clip-refine
☆ Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination ICML 2025
Zero-shot coordination (ZSC), the ability to adapt to a new partner in a cooperative task, is a critical component of human-compatible AI. While prior work has focused on training agents to cooperate on a single task, these specialized models do not generalize to new tasks, even if they are highly similar. Here, we study how reinforcement learning on a distribution of environments with a single partner enables learning general cooperative skills that support ZSC with many new partners on many new problems. We introduce two Jax-based, procedural generators that create billions of solvable coordination challenges. We develop a new paradigm called Cross-Environment Cooperation (CEC), and show that it outperforms competitive baselines quantitatively and qualitatively when collaborating with real people. Our findings suggest that learning to collaborate across many unique scenarios encourages agents to develop general norms, which prove effective for collaboration with different partners. Together, our results suggest a new route toward designing generalist cooperative agents capable of interacting with humans without requiring human data.
comment: Accepted to CogSci 2025, In-review for ICML 2025
☆ NTIRE 2025 Challenge on Day and Night Raindrop Removal for Dual-Focused Images: Methods and Results CVPR
This paper reviews the NTIRE 2025 Challenge on Day and Night Raindrop Removal for Dual-Focused Images. This challenge received a wide range of impressive solutions, which are developed and evaluated using our collected real-world Raindrop Clarity dataset. Unlike existing deraining datasets, our Raindrop Clarity dataset is more diverse and challenging in degradation types and contents, which includes day raindrop-focused, day background-focused, night raindrop-focused, and night background-focused degradations. This dataset is divided into three subsets for competition: 14,139 images for training, 240 images for validation, and 731 images for testing. The primary objective of this challenge is to establish a new and powerful benchmark for the task of removing raindrops under varying lighting and focus conditions. There are a total of 361 participants in the competition, and 32 teams submitting valid solutions and fact sheets for the final testing phase. These submissions achieved state-of-the-art (SOTA) performance on the Raindrop Clarity dataset. The project can be found at https://lixinustc.github.io/CVPR-NTIRE2025-RainDrop-Competition.github.io/.
comment: Challenge Report of CVPR NTIRE 2025; 26 pages; Methods from 32 teams
☆ WebLists: Extracting Structured Information From Complex Interactive Websites Using Executable LLM Agents
Most recent web agent research has focused on navigation and transaction tasks, with little emphasis on extracting structured data at scale. We present WebLists, a benchmark of 200 data-extraction tasks across four common business and enterprise use-cases. Each task requires an agent to navigate to a webpage, configure it appropriately, and extract complete datasets with well-defined schemas. We show that both LLMs with search capabilities and SOTA web agents struggle with these tasks, with a recall of 3% and 31%, respectively, despite higher performance on question-answering tasks. To address this challenge, we propose BardeenAgent, a novel framework that enables web agents to convert their execution into repeatable programs, and replay them at scale across pages with similar structure. BardeenAgent is also the first LLM agent to take advantage of the regular structure of HTML. In particular BardeenAgent constructs a generalizable CSS selector to capture all relevant items on the page, then fits the operations to extract the data. On the WebLists benchmark, BardeenAgent achieves 66% recall overall, more than doubling the performance of SOTA web agents, and reducing cost per output row by 3x.
☆ GRAIL: Gradient-Based Adaptive Unlearning for Privacy and Copyright in LLMs IJCNN 2025
Large Language Models (LLMs) trained on extensive datasets often learn sensitive information, which raises significant social and legal concerns under principles such as the "Right to be forgotten." Retraining entire models from scratch to remove undesired information is both costly and impractical. Furthermore, existing single-domain unlearning methods fail to address multi-domain scenarios, where knowledge is interwoven across domains such as privacy and copyright, creating overlapping representations that lead to excessive knowledge removal or degraded performance. To tackle these issues, we propose GRAIL (GRadient-based AdaptIve unLearning), a novel multi-domain unlearning framework. GRAIL leverages gradient information from multiple domains to precisely distinguish the unlearning scope from the retention scope, and applies an adaptive parameter-wise localization strategy to selectively remove targeted knowledge while preserving critical parameters for each domain. Experimental results on unlearning benchmarks show that GRAIL achieves unlearning success on par with the existing approaches, while also demonstrating up to 17% stronger knowledge retention success compared to the previous state-of-art method. Our findings establish a new paradigm for effectively managing and regulating sensitive information in large-scale pre-trained language models.
comment: Accepted by IJCNN 2025
☆ Embodied-R: Collaborative Framework for Activating Embodied Spatial Reasoning in Foundation Models via Reinforcement Learning
Humans can perceive and reason about spatial relationships from sequential visual observations, such as egocentric video streams. However, how pretrained models acquire such abilities, especially high-level reasoning, remains unclear. This paper introduces Embodied-R, a collaborative framework combining large-scale Vision-Language Models (VLMs) for perception and small-scale Language Models (LMs) for reasoning. Using Reinforcement Learning (RL) with a novel reward system considering think-answer logical consistency, the model achieves slow-thinking capabilities with limited computational resources. After training on only 5k embodied video samples, Embodied-R with a 3B LM matches state-of-the-art multimodal reasoning models (OpenAI-o1, Gemini-2.5-pro) on both in-distribution and out-of-distribution embodied spatial reasoning tasks. Embodied-R also exhibits emergent thinking patterns such as systematic analysis and contextual integration. We further explore research questions including response length, training on VLM, strategies for reward design, and differences in model generalization after SFT (Supervised Fine-Tuning) and RL training.
comment: 12 pages, 5 figures
☆ ACoRN: Noise-Robust Abstractive Compression in Retrieval-Augmented Language Models
Abstractive compression utilizes smaller langauge models to condense query-relevant context, reducing computational costs in retrieval-augmented generation (RAG). However,retrieved documents often include information that is either irrelevant to answering the query or misleading due to factual incorrect content, despite having high relevance scores. This behavior indicates that abstractive compressors are more likely to omit important information essential for the correct answer, especially in long contexts where attention dispersion occurs. To address this issue, we categorize retrieved documents in a more fine-grained manner and propose Abstractive Compression Robust against Noise (ACoRN), which introduces two novel training steps. First, we use offline data augmentation on the training dataset to enhance compressor robustness against two distinct types of retrieval noise. Second, since the language modelbased compressor cannot fully utilize information from multiple retrieved documents and exhibits positional bias, we perform finetuning to generate summaries centered around key information that directly supports the correct answer. Our experiments demonstrate that T5-large, trained with ACoRN as a compressor, improves EM and F1 scores while preserving the answer string, which could serve as direct evidence. ACoRN excels on datasets with many accuracy-reducing documents, making it highly useful in real-world scenarios.
☆ Post-processing improves accuracy of Artificial Intelligence weather forecasts
Artificial Intelligence (AI) weather models are now reaching operational-grade performance for some variables, but like traditional Numerical Weather Prediction (NWP) models, they exhibit systematic biases and reliability issues. We test the application of the Bureau of Meteorology's existing statistical post-processing system, IMPROVER, to ECMWF's deterministic Artificial Intelligence Forecasting System (AIFS), and compare results against post-processed outputs from the ECMWF HRES and ENS models. Without any modification to configuration or processing workflows, post-processing yields comparable accuracy improvements for AIFS as for traditional NWP forecasts, in both expected value and probabilistic outputs. We show that blending AIFS with NWP models improves overall forecast skill, even when AIFS alone is not the most accurate component. These findings show that statistical post-processing methods developed for NWP are directly applicable to AI models, enabling national meteorological centres to incorporate AI forecasts into existing workflows in a low-risk, incremental fashion.
☆ Persona-judge: Personalized Alignment of Large Language Models via Token-level Self-judgment
Aligning language models with human preferences presents significant challenges, particularly in achieving personalization without incurring excessive computational costs. Existing methods rely on reward signals and additional annotated data, limiting their scalability and adaptability to diverse human values. To address these challenges, we introduce Persona-judge, a novel discriminative paradigm that enables training-free personalized alignment with unseen preferences. Instead of optimizing policy parameters through external reward feedback, Persona-judge leverages the intrinsic preference judgment capabilities of the model. Specifically, a draft model generates candidate tokens conditioned on a given preference, while a judge model, embodying another preference, cross-validates the predicted tokens whether to be accepted. Experimental results demonstrate that Persona-judge, using the inherent preference evaluation mechanisms of the model, offers a scalable and computationally efficient solution to personalized alignment, paving the way for more adaptive customized alignment.
☆ Quantum Computing Supported Adversarial Attack-Resilient Autonomous Vehicle Perception Module for Traffic Sign Classification
Deep learning (DL)-based image classification models are essential for autonomous vehicle (AV) perception modules since incorrect categorization might have severe repercussions. Adversarial attacks are widely studied cyberattacks that can lead DL models to predict inaccurate output, such as incorrectly classified traffic signs by the perception module of an autonomous vehicle. In this study, we create and compare hybrid classical-quantum deep learning (HCQ-DL) models with classical deep learning (C-DL) models to demonstrate robustness against adversarial attacks for perception modules. Before feeding them into the quantum system, we used transfer learning models, alexnet and vgg-16, as feature extractors. We tested over 1000 quantum circuits in our HCQ-DL models for projected gradient descent (PGD), fast gradient sign attack (FGSA), and gradient attack (GA), which are three well-known untargeted adversarial approaches. We evaluated the performance of all models during adversarial attacks and no-attack scenarios. Our HCQ-DL models maintain accuracy above 95\% during a no-attack scenario and above 91\% for GA and FGSA attacks, which is higher than C-DL models. During the PGD attack, our alexnet-based HCQ-DL model maintained an accuracy of 85\% compared to C-DL models that achieved accuracies below 21\%. Our results highlight that the HCQ-DL models provide improved accuracy for traffic sign classification under adversarial settings compared to their classical counterparts.
☆ Scaling Instruction-Tuned LLMs to Million-Token Contexts via Hierarchical Synthetic Data Generation
Large Language Models (LLMs) struggle with long-context reasoning, not only due to the quadratic scaling of computational complexity with sequence length but also because of the scarcity and expense of annotating long-context data. There has been barely any open-source work that systematically ablates long-context data, nor is there any openly available instruction tuning dataset with contexts surpassing 100K tokens. To bridge this gap, we introduce a novel post-training synthetic data generation strategy designed to efficiently extend the context window of LLMs while preserving their general task performance. Our approach scalably extends to arbitrarily long context lengths, unconstrained by the length of available real-world data, which effectively addresses the scarcity of raw long-context data. Through a step-by-step rotary position embedding (RoPE) scaling training strategy, we demonstrate that our model, with a context length of up to 1M tokens, performs well on the RULER benchmark and InfiniteBench and maintains robust performance on general language tasks.
comment: 26 pages, 5 figures
☆ The Chronicles of Foundation AI for Forensics of Multi-Agent Provenance
Provenance is the chronology of things, resonating with the fundamental pursuit to uncover origins, trace connections, and situate entities within the flow of space and time. As artificial intelligence advances towards autonomous agents capable of interactive collaboration on complex tasks, the provenance of generated content becomes entangled in the interplay of collective creation, where contributions are continuously revised, extended or overwritten. In a multi-agent generative chain, content undergoes successive transformations, often leaving little, if any, trace of prior contributions. In this study, we investigates the problem of tracking multi-agent provenance across the temporal dimension of generation. We propose a chronological system for post hoc attribution of generative history from content alone, without reliance on internal memory states or external meta-information. At its core lies the notion of symbolic chronicles, representing signed and time-stamped records, in a form analogous to the chain of custody in forensic science. The system operates through a feedback loop, whereby each generative timestep updates the chronicle of prior interactions and synchronises it with the synthetic content in the very act of generation. This research seeks to develop an accountable form of collaborative artificial intelligence within evolving cyber ecosystems.
☆ Crossing the Human-Robot Embodiment Gap with Sim-to-Real RL using One Human Demonstration
Teaching robots dexterous manipulation skills often requires collecting hundreds of demonstrations using wearables or teleoperation, a process that is challenging to scale. Videos of human-object interactions are easier to collect and scale, but leveraging them directly for robot learning is difficult due to the lack of explicit action labels from videos and morphological differences between robot and human hands. We propose Human2Sim2Robot, a novel real-to-sim-to-real framework for training dexterous manipulation policies using only one RGB-D video of a human demonstrating a task. Our method utilizes reinforcement learning (RL) in simulation to cross the human-robot embodiment gap without relying on wearables, teleoperation, or large-scale data collection typically necessary for imitation learning methods. From the demonstration, we extract two task-specific components: (1) the object pose trajectory to define an object-centric, embodiment-agnostic reward function, and (2) the pre-manipulation hand pose to initialize and guide exploration during RL training. We found that these two components are highly effective for learning the desired task, eliminating the need for task-specific reward shaping and tuning. We demonstrate that Human2Sim2Robot outperforms object-aware open-loop trajectory replay by 55% and imitation learning with data augmentation by 68% across grasping, non-prehensile manipulation, and multi-step tasks. Project Site: https://human2sim2robot.github.io
comment: 15 pages, 13 figures
☆ Code Copycat Conundrum: Demystifying Repetition in LLM-based Code Generation
Despite recent advances in Large Language Models (LLMs) for code generation, the quality of LLM-generated code still faces significant challenges. One significant issue is code repetition, which refers to the model's tendency to generate structurally redundant code, resulting in inefficiencies and reduced readability. To address this, we conduct the first empirical study to investigate the prevalence and nature of repetition across 19 state-of-the-art code LLMs using three widely-used benchmarks. Our study includes both quantitative and qualitative analyses, revealing that repetition is pervasive and manifests at various granularities and extents, including character, statement, and block levels. We further summarize a taxonomy of 20 repetition patterns. Building on our findings, we propose DeRep, a rule-based technique designed to detect and mitigate repetition in generated code. We evaluate DeRep using both open-source benchmarks and in an industrial setting. Our results demonstrate that DeRep significantly outperforms baselines in reducing repetition (with an average improvements of 91.3%, 93.5%, and 79.9% in rep-3, rep-line, and sim-line metrics) and enhancing code quality (with a Pass@1 increase of 208.3% over greedy search). Furthermore, integrating DeRep improves the performance of existing repetition mitigation methods, with Pass@1 improvements ranging from 53.7% to 215.7%.
☆ Robo-SGG: Exploiting Layout-Oriented Normalization and Restitution for Robust Scene Graph Generation
In this paper, we introduce a novel method named Robo-SGG, i.e., Layout-Oriented Normalization and Restitution for Robust Scene Graph Generation. Compared to the existing SGG setting, the robust scene graph generation aims to perform inference on a diverse range of corrupted images, with the core challenge being the domain shift between the clean and corrupted images. Existing SGG methods suffer from degraded performance due to compromised visual features e.g., corruption interference or occlusions. To obtain robust visual features, we exploit the layout information, which is domain-invariant, to enhance the efficacy of existing SGG methods on corrupted images. Specifically, we employ Instance Normalization(IN) to filter out the domain-specific feature and recover the unchangeable structural features, i.e., the positional and semantic relationships among objects by the proposed Layout-Oriented Restitution. Additionally, we propose a Layout-Embedded Encoder (LEE) that augments the existing object and predicate encoders within the SGG framework, enriching the robust positional and semantic features of objects and predicates. Note that our proposed Robo-SGG module is designed as a plug-and-play component, which can be easily integrated into any baseline SGG model. Extensive experiments demonstrate that by integrating the state-of-the-art method into our proposed Robo-SGG, we achieve relative improvements of 5.6%, 8.0%, and 6.5% in mR@50 for PredCls, SGCls, and SGDet tasks on the VG-C dataset, respectively, and achieve new state-of-the-art performance in corruption scene graph generation benchmark (VG-C and GQA-C). We will release our source code and model.
☆ Identifying and Mitigating the Influence of the Prior Distribution in Large Language Models
Large language models (LLMs) sometimes fail to respond appropriately to deterministic tasks -- such as counting or forming acronyms -- because the implicit prior distribution they have learned over sequences of tokens influences their responses. In this work, we show that, in at least some cases, LLMs actually compute the information needed to perform these tasks correctly, and we identify some interventions that can allow them to access this information to improve their performance. First, we show that simply prompting the language model to not rely on its prior knowledge leads to dramatic improvements in prior-dominated tasks. We then use mechanistic interpretability techniques to localize the prior within the LLM and manipulate the extent to which that prior influences its responses. Specifically, we show that it is possible to identify layers of the underlying neural network that correlate with the prior probability of a response and that lightweight finetuning of these layers with basic prompts on prior-dominated tasks achieves high performance on held-out answers. These results suggest that the information required to produce a correct response is contained within the representations of the problems formed by the models. Furthermore, we show that this finetuning is significantly more effective for prior-dominated tasks, and that the error after finetuning is no longer correlated with the prior. Our results suggest that it may be possible to define effective methods for manipulating the extent to which LLMs rely upon their priors in solving problems, potentially increasing their performance in settings where LLMs hallucinate for reasons related to the prior probability of token sequences.
comment: 16 pages, 5 figures
☆ Local Data Quantity-Aware Weighted Averaging for Federated Learning with Dishonest Clients ICME 2025
Federated learning (FL) enables collaborative training of deep learning models without requiring data to leave local clients, thereby preserving client privacy. The aggregation process on the server plays a critical role in the performance of the resulting FL model. The most commonly used aggregation method is weighted averaging based on the amount of data from each client, which is thought to reflect each client's contribution. However, this method is prone to model bias, as dishonest clients might report inaccurate training data volumes to the server, which is hard to verify. To address this issue, we propose a novel secure \underline{Fed}erated \underline{D}ata q\underline{u}antity-\underline{a}ware weighted averaging method (FedDua). It enables FL servers to accurately predict the amount of training data from each client based on their local model gradients uploaded. Furthermore, it can be seamlessly integrated into any FL algorithms that involve server-side model aggregation. Extensive experiments on three benchmarking datasets demonstrate that FedDua improves the global model performance by an average of 3.17% compared to four popular FL aggregation methods in the presence of inaccurate client data volume declarations.
comment: The paper has been accepted by ICME 2025
☆ CM3AE: A Unified RGB Frame and Event-Voxel/-Frame Pre-training Framework
Event cameras have attracted increasing attention in recent years due to their advantages in high dynamic range, high temporal resolution, low power consumption, and low latency. Some researchers have begun exploring pre-training directly on event data. Nevertheless, these efforts often fail to establish strong connections with RGB frames, limiting their applicability in multi-modal fusion scenarios. To address these issues, we propose a novel CM3AE pre-training framework for the RGB-Event perception. This framework accepts multi-modalities/views of data as input, including RGB images, event images, and event voxels, providing robust support for both event-based and RGB-event fusion based downstream tasks. Specifically, we design a multi-modal fusion reconstruction module that reconstructs the original image from fused multi-modal features, explicitly enhancing the model's ability to aggregate cross-modal complementary information. Additionally, we employ a multi-modal contrastive learning strategy to align cross-modal feature representations in a shared latent space, which effectively enhances the model's capability for multi-modal understanding and capturing global dependencies. We construct a large-scale dataset containing 2,535,759 RGB-Event data pairs for the pre-training. Extensive experiments on five downstream tasks fully demonstrated the effectiveness of CM3AE. Source code and pre-trained models will be released on https://github.com/Event-AHU/CM3AE.
☆ MetaSynth: Meta-Prompting-Driven Agentic Scaffolds for Diverse Synthetic Data Generation
Recent smaller language models such Phi-3.5 and Phi-4 rely on synthetic data generated using larger Language models. Questions remain about leveraging synthetic data for other use cases, such as adapting LLMs to specific domains. A key limitation of synthetic data is low diversity, which negatively impacts its downstream applicability for improving other models. To address this, we propose MetaSynth, a method for generating synthetic data that enhances diversity through meta-prompting, where a language model orchestrates multiple "expert" LLM agents to collaboratively generate data. Using only 25 million tokens of synthetic data generated with MetaSynth, we successfully adapt a well-trained LLM (Mistral-7B-v0.3) to two specialized domains-Finance and Biomedicine-without compromising the capabilities of the resulting model in general tasks. In addition, we evaluate the diversity of our synthetic data using seven automated metrics, and find that it approaches the diversity of LLM pre-training corpora. Continually pre-training Mistral-7B-v0.3 with MetaSynth notably outperforms the base LLM, showing improvements of up to 4.08% in Finance and 13.75% in Biomedicine. The same model shows degraded performance when trained on data generated using a template prompt, even when the template includes prior generations and varying In-Context exemplars of real data. Our findings suggest that a few million tokens of diverse synthetic data without mixing any real data, is sufficient for effective domain adaptation when using MetaSynth.
comment: 33 pages, 17 figures. Preprint
☆ ZeroSumEval: Scaling LLM Evaluation with Inter-Model Competition
Evaluating the capabilities of Large Language Models (LLMs) has traditionally relied on static benchmark datasets, human assessments, or model-based evaluations - methods that often suffer from overfitting, high costs, and biases. ZeroSumEval is a novel competition-based evaluation protocol that leverages zero-sum games to assess LLMs with dynamic benchmarks that resist saturation. ZeroSumEval encompasses a diverse suite of games, including security challenges (PyJail), classic games (Chess, Liar's Dice, Poker), knowledge tests (MathQuiz), and persuasion challenges (Gandalf, Debate). These games are designed to evaluate a range of AI capabilities such as strategic reasoning, planning, knowledge application, and creativity. Building upon recent studies that highlight the effectiveness of game-based evaluations for LLMs, ZeroSumEval enhances these approaches by providing a standardized and extensible framework. To demonstrate this, we conduct extensive experiments with >7000 simulations across 7 games and 13 models. Our results show that while frontier models from the GPT and Claude families can play common games and answer questions, they struggle to play games that require creating novel and challenging questions. We also observe that models cannot reliably jailbreak each other and fail generally at tasks requiring creativity. We release our code at https://github.com/facebookresearch/ZeroSumEval.
☆ TraCeS: Trajectory Based Credit Assignment From Sparse Safety Feedback
In safe reinforcement learning (RL), auxiliary safety costs are used to align the agent to safe decision making. In practice, safety constraints, including cost functions and budgets, are unknown or hard to specify, as it requires anticipation of all possible unsafe behaviors. We therefore address a general setting where the true safety definition is unknown, and has to be learned from sparsely labeled data. Our key contributions are: first, we design a safety model that performs credit assignment to estimate each decision step's impact on the overall safety using a dataset of diverse trajectories and their corresponding binary safety labels (i.e., whether the corresponding trajectory is safe/unsafe). Second, we illustrate the architecture of our safety model to demonstrate its ability to learn a separate safety score for each timestep. Third, we reformulate the safe RL problem using the proposed safety model and derive an effective algorithm to optimize a safe yet rewarding policy. Finally, our empirical results corroborate our findings and show that this approach is effective in satisfying unknown safety definition, and scalable to various continuous control tasks.
☆ Privacy-Preserving Operating Room Workflow Analysis using Digital Twins
Purpose: The operating room (OR) is a complex environment where optimizing workflows is critical to reduce costs and improve patient outcomes. The use of computer vision approaches for the automatic recognition of perioperative events enables identification of bottlenecks for OR optimization. However, privacy concerns limit the use of computer vision for automated event detection from OR videos, which makes privacy-preserving approaches needed for OR workflow analysis. Methods: We propose a two-stage pipeline for privacy-preserving OR video analysis and event detection. In the first stage, we leverage vision foundation models for depth estimation and semantic segmentation to generate de-identified Digital Twins (DT) of the OR from conventional RGB videos. In the second stage, we employ the SafeOR model, a fused two-stream approach that processes segmentation masks and depth maps for OR event detection. We evaluate this method on an internal dataset of 38 simulated surgical trials with five event classes. Results: Our results indicate that this DT-based approach to the OR event detection model achieves performance on par and sometimes even better than raw RGB video-based models on detecting OR events. Conclusion: DTs enable privacy-preserving OR workflow analysis, facilitating the sharing of de-identified data across institutions and they can potentially enhance model generalizability by mitigating domain-specific appearance differences.
☆ Memorization: A Close Look at Books
To what extent can entire books be extracted from LLMs? Using the Llama 3 70B family of models, and the "prefix-prompting" extraction technique, we were able to auto-regressively reconstruct, with a very high level of similarity, one entire book (Alice's Adventures in Wonderland) from just the first 500 tokens. We were also able to obtain high extraction rates on several other books, piece-wise. However, these successes do not extend uniformly to all books. We show that extraction rates of books correlate with book popularity and thus, likely duplication in the training data. We also confirm the undoing of mitigations in the instruction-tuned Llama 3.1, following recent work (Nasr et al., 2025). We further find that this undoing comes from changes to only a tiny fraction of weights concentrated primarily in the lower transformer blocks. Our results provide evidence of the limits of current regurgitation mitigation strategies and introduce a framework for studying how fine-tuning affects the retrieval of verbatim memorization in aligned LLMs.
♻ ☆ A general language model for peptide identification
Advances in peptide identification are revolutionizing our ability to decipher protein functions and accelerate therapeutic discovery. We present PDeepPP, a deep learning framework that integrates pretrained protein language models with parallel transformer-CNN architectures, achieving state-of-the-art performance in peptide characterization tasks. The model's hybrid architecture demonstrates unique capabilities in capturing both local sequence motifs and global structural features, as evidenced by 29% improved cluster separation in UMAP visualizations compared to conventional approaches. Evaluated across 33 biological recognition tasks - including post-translational modification site prediction and bioactive peptide identification - PDeepPP outperformed existing methods in 25 tasks with average AUC improvements of 4.2%. Notably, it achieved 0.9726 accuracy with PR AUC 0.9977 in antimicrobial peptide detection while reducing false negatives by 37.5% in antimalarial recognition scenarios. This framework enables accurate large-scale peptide analysis, achieving 218* acceleration over sequence-alignment-based methods while maintaining 99.5% specificity in critical glycosylation site detection.PDeepPP establishes a new paradigm for computational peptide analysis through its synergistic architecture design, enabling rapid yet precise functional annotation that bridges molecular pattern recognition with translational biomedical applications.We have made our implementation, including code, data, and pretrained models, publicly available via GitHub (https://github.com/fondress/PDeepPP) and Hugging Face (https://huggingface.co/fondress/PDeppPP).
comment: 21 pages, 9 figures, 4 tables, submitted to arXiv
♻ ☆ SIFT-50M: A Large-Scale Multilingual Dataset for Speech Instruction Fine-Tuning
We introduce SIFT (Speech Instruction Fine-Tuning), a 50M-example dataset designed for instruction fine-tuning and pre-training of speech-text large language models (LLMs). SIFT-50M is built from publicly available speech corpora, which collectively contain 14K hours of speech, and leverages LLMs along with off-the-shelf expert models. The dataset spans five languages, encompassing a diverse range of speech understanding as well as controllable speech generation instructions. Using SIFT-50M, we train SIFT-LLM, which outperforms existing speech-text LLMs on instruction-following benchmarks while achieving competitive performance on foundational speech tasks. To support further research, we also introduce EvalSIFT, a benchmark dataset specifically designed to evaluate the instruction-following capabilities of speech-text LLMs.
♻ ☆ Citation-Enhanced Generation for LLM-based Chatbots
Large language models (LLMs) exhibit powerful general intelligence across diverse scenarios, including their integration into chatbots. However, a vital challenge of LLM-based chatbots is that they may produce hallucinated content in responses, which significantly limits their applicability. Various efforts have been made to alleviate hallucination, such as retrieval augmented generation and reinforcement learning with human feedback, but most of them require additional training and data annotation. In this paper, we propose a novel post-hoc Citation-Enhanced Generation (CEG) approach combined with retrieval argumentation. Unlike previous studies that focus on preventing hallucinations during generation, our method addresses this issue in a post-hoc way. It incorporates a retrieval module to search for supporting documents relevant to the generated content, and employs a natural language inference-based citation generation module. Once the statements in the generated content lack of reference, our model can regenerate responses until all statements are supported by citations. Note that our method is a training-free plug-and-play plugin that is capable of various LLMs. Experiments on various hallucination-related datasets show our framework outperforms state-of-the-art methods in both hallucination detection and response regeneration on three benchmarks. Our codes and dataset will be publicly available.
♻ ☆ Multimodal LLMs Can Reason about Aesthetics in Zero-Shot
The rapid progress of generative art has democratized the creation of visually pleasing imagery. However, achieving genuine artistic impact - the kind that resonates with viewers on a deeper, more meaningful level - requires a sophisticated aesthetic sensibility. This sensibility involves a multi-faceted reasoning process extending beyond mere visual appeal, which is often overlooked by current computational models. This paper pioneers an approach to capture this complex process by investigating how the reasoning capabilities of Multimodal LLMs (MLLMs) can be effectively elicited for aesthetic judgment. Our analysis reveals a critical challenge: MLLMs exhibit a tendency towards hallucinations during aesthetic reasoning, characterized by subjective opinions and unsubstantiated artistic interpretations. We further demonstrate that these limitations can be overcome by employing an evidence-based, objective reasoning process, as substantiated by our proposed baseline, ArtCoT. MLLMs prompted by this principle produce multi-faceted and in-depth aesthetic reasoning that aligns significantly better with human judgment. These findings have direct applications in areas such as AI art tutoring and as reward models for generative art. Ultimately, our work paves the way for AI systems that can truly understand, appreciate, and generate artworks that align with the sensible human aesthetic standard.
comment: WIP, Homepage https://github.com/songrise/MLLM4Art
♻ ☆ Understanding LLM Behaviors via Compression: Data Generation, Knowledge Acquisition and Scaling Laws
Large Language Models (LLMs) have demonstrated remarkable capabilities across numerous tasks, yet principled explanations for their underlying mechanisms and several phenomena, such as scaling laws, hallucinations, and related behaviors, remain elusive. In this work, we revisit the classical relationship between compression and prediction, grounded in Kolmogorov complexity and Shannon information theory, to provide deeper insights into LLM behaviors. By leveraging the Kolmogorov Structure Function and interpreting LLM compression as a two-part coding process, we offer a detailed view of how LLMs acquire and store information across increasing model and data scales -- from pervasive syntactic patterns to progressively rarer knowledge elements. Motivated by this theoretical perspective and natural assumptions inspired by Heap's and Zipf's laws, we introduce a simplified yet representative hierarchical data-generation framework called the Syntax-Knowledge model. Under the Bayesian setting, we show that prediction and compression within this model naturally lead to diverse learning and scaling behaviors of LLMs. In particular, our theoretical analysis offers intuitive and principled explanations for both data and model scaling laws, the dynamics of knowledge acquisition during training and fine-tuning, factual knowledge hallucinations in LLMs. The experimental results validate our theoretical predictions.
♻ ☆ ReTool: Reinforcement Learning for Strategic Tool Use in LLMs
While reasoning models (e.g., DeepSeek R1) trained with reinforcement learning (RL), excel in textual reasoning, they struggle in scenarios requiring structured problem-solving, such as geometric reasoning, concise computation, or complex equation solving-areas where computational tools like code interpreters (CI) demonstrate distinct advantages. To bridge this gap, we propose ReTool, which enhances long-form reasoning with tool-integrated learning, including two key features: (1) dynamic interleaving of real-time code execution within natural language reasoning processes, and (2) an automated RL paradigm that allows policy rollouts with multi-turn real-time code execution and teaches the model in learning when and how to invoke tools based on outcome feedback. ReTool employs a systematic training framework, beginning with synthetic cold-start data generation to produce code-augmented long-form reasoning traces for fine-tuning base models. Subsequent RL training leverages task outcomes as rewards to iteratively refine the model's tool use strategy, enabling autonomous discovery of optimal tool invocation patterns without human priors. Experiments on the challenging MATH Olympiad benchmark AIME demonstrate ReTool's superiority: Our 32B model achieves 67% accuracy with 400 training steps, outperforming text-based RL baseline (40% accuracy, 1080 steps) in efficiency and performance. Remarkably, ReTool-32B attains 72.5% accuracy in extended settings, surpassing OpenAI's o1-preview by 27.9%. Further analysis reveals emergent behaviors such as code self-correction, signaling an ''aha moment'' in which the model autonomously masters adaptive tool use. These findings highlight the promise of outcome-driven tool integration for advancing complex mathematical reasoning and offer new insights into hybrid neuro-symbolic systems.
comment: fix typos
♻ ☆ Benchmarking the Spatial Robustness of DNNs via Natural and Adversarial Localized Corruptions
The robustness of DNNs is a crucial factor in safety-critical applications, particularly in complex and dynamic environments where localized corruptions can arise. While previous studies have evaluated the robustness of semantic segmentation (SS) models under whole-image natural or adversarial corruptions, a comprehensive investigation into the spatial robustness of dense vision models under localized corruptions remained underexplored. This paper fills this gap by introducing specialized metrics for benchmarking the spatial robustness of segmentation models, alongside with an evaluation framework to assess the impact of localized corruptions. Furthermore, we uncover the inherent complexity of characterizing worst-case robustness using a single localized adversarial perturbation. To address this, we propose region-aware multi-attack adversarial analysis, a method that enables a deeper understanding of model robustness against adversarial perturbations applied to specific regions. The proposed metrics and analysis were exploited to evaluate 14 segmentation models in driving scenarios, uncovering key insights into the effects of localized corruption in both natural and adversarial forms. The results reveal that models respond to these two types of threats differently; for instance, transformer-based segmentation models demonstrate notable robustness to localized natural corruptions but are highly vulnerable to adversarial ones and vice-versa for CNN-based models. Consequently, we also address the challenge of balancing robustness to both natural and adversarial localized corruptions by means of ensemble models, thereby achieving a broader threat coverage and improved reliability for dense vision tasks.
comment: Under review
♻ ☆ REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites
We introduce REAL, a benchmark and framework for multi-turn agent evaluations on deterministic simulations of real-world websites. REAL comprises high-fidelity, deterministic replicas of 11 widely-used websites across domains such as e-commerce, travel, communication, and professional networking. We also release a benchmark consisting of 112 practical tasks that mirror everyday complex user interactions requiring both accurate information retrieval and state-changing actions. All interactions occur within this fully controlled setting, eliminating safety risks and enabling robust, reproducible evaluation of agent capability and reliability. Our novel evaluation framework combines programmatic checks of website state for action-based tasks with rubric-guided LLM-based judgments for information retrieval. The framework supports both open-source and proprietary agent systems through a flexible evaluation harness that accommodates black-box commands within browser environments, allowing research labs to test agentic systems without modification. Our empirical results show that frontier language models achieve at most a 41% success rate on REAL, highlighting critical gaps in autonomous web navigation and task completion capabilities. Our framework supports easy integration of new tasks, reproducible evaluation, and scalable post-training data generation, marking a significant step forward in evaluating and advancing agent capabilities.
comment: The websites, framework, and leaderboard are available at https://realevals.xyz and https://github.com/agi-inc/REAL
♻ ☆ Relevance for Human Robot Collaboration
Inspired by the human ability to selectively focus on relevant information, this paper introduces relevance, a novel dimensionality reduction process for human-robot collaboration (HRC). Our approach incorporates a continuously operating perception module, evaluates cue sufficiency within the scene, and applies a flexible formulation and computation framework. To accurately and efficiently quantify relevance, we developed an event-based framework that maintains a continuous perception of the scene and selectively triggers relevance determination. Within this framework, we developed a probabilistic methodology, which considers various factors and is built on a novel structured scene representation. Simulation results demonstrate that the relevance framework and methodology accurately predict the relevance of a general HRC setup, achieving a precision of 0.99, a recall of 0.94, an F1 score of 0.96, and an object ratio of 0.94. Relevance can be broadly applied to several areas in HRC to accurately improve task planning time by 79.56% compared with pure planning for a cereal task, reduce perception latency by up to 26.53% for an object detector, improve HRC safety by up to 13.50% and reduce the number of inquiries for HRC by 80.84%. A real-world demonstration showcases the relevance framework's ability to intelligently and seamlessly assist humans in everyday tasks.
comment: under review
♻ ☆ SparseDM: Toward Sparse Efficient Diffusion Models ICME 2025
Diffusion models represent a powerful family of generative models widely used for image and video generation. However, the time-consuming deployment, long inference time, and requirements on large memory hinder their applications on resource constrained devices. In this paper, we propose a method based on the improved Straight-Through Estimator to improve the deployment efficiency of diffusion models. Specifically, we add sparse masks to the Convolution and Linear layers in a pre-trained diffusion model, then transfer learn the sparse model during the fine-tuning stage and turn on the sparse masks during inference. Experimental results on a Transformer and UNet-based diffusion models demonstrate that our method reduces MACs by 50% while maintaining FID. Sparse models are accelerated by approximately 1.2x on the GPU. Under other MACs conditions, the FID is also lower than 1 compared to other methods.
comment: This paper has been accepted by ICME 2025
♻ ☆ A Coding-Theoretic Analysis of Hyperspherical Prototypical Learning Geometry
Hyperspherical Prototypical Learning (HPL) is a supervised approach to representation learning that designs class prototypes on the unit hypersphere. The prototypes bias the representations to class separation in a scale invariant and known geometry. Previous approaches to HPL have either of the following shortcomings: (i) they follow an unprincipled optimisation procedure; or (ii) they are theoretically sound, but are constrained to only one possible latent dimension. In this paper, we address both shortcomings. To address (i), we present a principled optimisation procedure whose solution we show is optimal. To address (ii), we construct well-separated prototypes in a wide range of dimensions using linear block codes. Additionally, we give a full characterisation of the optimal prototype placement in terms of achievable and converse bounds, showing that our proposed methods are near-optimal.
comment: Changes in version 2: Minor formatting changes. Published in the Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM), PMLR 251. Available at: https://proceedings.mlr.press/v251/lindstrom24a.html 14 pages: 9 of the main paper, 2 of references, and 3 of appendices.. Code is available at: https://github.com/martinlindstrom/coding_theoretic_hpl
♻ ☆ Impact of Data Duplication on Deep Neural Network-Based Image Classifiers: Robust vs. Standard Models
The accuracy and robustness of machine learning models against adversarial attacks are significantly influenced by factors such as training data quality, model architecture, the training process, and the deployment environment. In recent years, duplicated data in training sets, especially in language models, has attracted considerable attention. It has been shown that deduplication enhances both training performance and model accuracy in language models. While the importance of data quality in training image classifier Deep Neural Networks (DNNs) is widely recognized, the impact of duplicated images in the training set on model generalization and performance has received little attention. In this paper, we address this gap and provide a comprehensive study on the effect of duplicates in image classification. Our analysis indicates that the presence of duplicated images in the training set not only negatively affects the efficiency of model training but also may result in lower accuracy of the image classifier. This negative impact of duplication on accuracy is particularly evident when duplicated data is non-uniform across classes or when duplication, whether uniform or non-uniform, occurs in the training set of an adversarially trained model. Even when duplicated samples are selected in a uniform way, increasing the amount of duplication does not lead to a significant improvement in accuracy.
♻ ☆ GPG: A Simple and Strong Reinforcement Learning Baseline for Model Reasoning
Reinforcement Learning (RL) can directly enhance the reasoning capabilities of large language models without extensive reliance on Supervised Fine-Tuning (SFT). In this work, we revisit the traditional Policy Gradient (PG) mechanism and propose a minimalist RL approach termed Group Policy Gradient (GPG). Unlike conventional methods, GPG directly optimize the original RL objective, thus obviating the need for surrogate loss functions. By eliminating the critic and reference models, avoiding KL divergence constraints, and addressing the advantage and gradient estimation bias, our approach significantly simplifies the training process compared to Group Relative Policy Optimization (GRPO). Our approach achieves superior performance without relying on auxiliary techniques or adjustments. As illustrated in Figure 1, extensive experiments demonstrate that our method not only reduces computational costs but also consistently outperforms GRPO across various unimodal and multimodal tasks. Our code is available at https://github.com/AMAP-ML/GPG.
♻ ☆ A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations AAAI 2025
Prototype-based classification learning methods are known to be inherently interpretable. However, this paradigm suffers from major limitations compared to deep models, such as lower performance. This led to the development of the so-called deep Prototype-Based Networks (PBNs), also known as prototypical parts models. In this work, we analyze these models with respect to different properties, including interpretability. In particular, we focus on the Classification-by-Components (CBC) approach, which uses a probabilistic model to ensure interpretability and can be used as a shallow or deep architecture. We show that this model has several shortcomings, like creating contradicting explanations. Based on these findings, we propose an extension of CBC that solves these issues. Moreover, we prove that this extension has robustness guarantees and derive a loss that optimizes robustness. Additionally, our analysis shows that most (deep) PBNs are related to (deep) RBF classifiers, which implies that our robustness guarantees generalize to shallow RBF classifiers. The empirical evaluation demonstrates that our deep PBN yields state-of-the-art classification accuracy on different benchmarks while resolving the interpretability shortcomings of other approaches. Further, our shallow PBN variant outperforms other shallow PBNs while being inherently interpretable and exhibiting provable robustness guarantees.
comment: To appear at AAAI 2025. Includes the Appendix of the AAAI submission. In v2, the font size has been increased in some figures. In v3, an incorrect hyperparameter specification (Table 6; $\lambda$) has been corrected
♻ ☆ The Impact of Environment Configurations on the Stability of AI-Enabled Systems
Nowadays, software systems tend to include Artificial Intelligence (AI) components. Changes in the operational environment have been known to negatively impact the stability of AI-enabled software systems by causing unintended changes in behavior. However, how an environment configuration impacts the behavior of such systems has yet to be explored. Understanding and quantifying the degree of instability caused by different environment settings can help practitioners decide the best environment configuration for the most stable AI systems. To achieve this goal, we performed experiments with eight different combinations of three key environment variables (operating system, Python version, and CPU architecture) on $30$ open-source AI-enabled systems using the Travis CI platform. We determine the existence and the degree of instability introduced by each configuration using three metrics: the output of an AI component of the system (model performance), the time required to build and run the system (processing time), and the cost associated with building and running the system (expense). Our results indicate that changes in environment configurations lead to instability across all three metrics; however, it is observed more frequently with respect to processing time and expense rather than model performance. For example, between Linux and MacOS, instability is observed in 23\%, 96.67\%, and 100\% of the studied projects in model performance, processing time, and expense, respectively. Our findings underscore the importance of identifying the optimal combination of configuration settings to mitigate drops in model performance and reduce the processing time and expense before deploying an AI-enabled system.
comment: Accepted for publication at the International Conference on Evaluation and Assessment in Software Engineering (EASE 2025)
♻ ☆ What Are the Odds? Improving the foundations of Statistical Model Checking
Markov decision processes (MDPs) are a fundamental model for decision making under uncertainty. They exhibit non-deterministic choice as well as probabilistic uncertainty. Traditionally, verification algorithms assume exact knowledge of the probabilities that govern the behaviour of an MDP. As this assumption is often unrealistic in practice, statistical model checking (SMC) was developed in the past two decades. It allows to analyse MDPs with unknown transition probabilities and provide probably approximately correct (PAC) guarantees on the result. Model-based SMC algorithms sample the MDP and build a model of it by estimating all transition probabilities, essentially for every transition answering the question: ``What are the odds?'' However, so far the statistical methods employed by the state of the art SMC algorithms are quite naive. Our contribution are several fundamental improvements to those methods: On the one hand, we survey statistics literature for better concentration inequalities; on the other hand, we propose specialised approaches that exploit our knowledge of the MDP. Our improvements are generally applicable to many kinds of problem statements because they are largely independent of the setting. Moreover, our experimental evaluation shows that they lead to significant gains, reducing the number of samples that the SMC algorithm has to collect by up to two orders of magnitude.
♻ ☆ SemML: Enhancing Automata-Theoretic LTL Synthesis with Machine Learning
Synthesizing a reactive system from specifications given in linear temporal logic (LTL) is a classical problem, finding its applications in safety-critical systems design. We present our tool SemML, which won this year's LTL realizability tracks of SYNTCOMP, after years of domination by Strix. While both tools are based on the automata-theoretic approach, ours relies heavily on (i) Semantic labelling, additional information of logical nature, coming from recent LTL-to-automata translations and decorating the resulting parity game, and (ii) Machine Learning approaches turning this information into a guidance oracle for on-the-fly exploration of the parity game (whence the name SemML). Our tool fills the missing gaps of previous suggestions to use such an oracle and provides an efficeint implementation with additional algorithmic improvements. We evaluate SemML both on the entire set of SYNTCOMP as well as a synthetic data set, compare it to Strix, and analyze the advantages and limitations. As SemML solves more instances on SYNTCOMP and does so significantly faster on larger instances, this demonstrates for the first time that machine-learning-aided approaches can out-perform state-of-the-art tools in real LTL synthesis.
♻ ☆ Which Optimizer Works Best for Physics-Informed Neural Networks and Kolmogorov-Arnold Networks?
Physics-Informed Neural Networks (PINNs) have revolutionized the computation of PDE solutions by integrating partial differential equations (PDEs) into the neural network's training process as soft constraints, becoming an important component of the scientific machine learning (SciML) ecosystem. More recently, physics-informed Kolmogorv-Arnold networks (PIKANs) have also shown to be effective and comparable in accuracy with PINNs. In their current implementation, both PINNs and PIKANs are mainly optimized using first-order methods like Adam, as well as quasi-Newton methods such as BFGS and its low-memory variant, L-BFGS. However, these optimizers often struggle with highly non-linear and non-convex loss landscapes, leading to challenges such as slow convergence, local minima entrapment, and (non)degenerate saddle points. In this study, we investigate the performance of Self-Scaled BFGS (SSBFGS), Self-Scaled Broyden (SSBroyden) methods and other advanced quasi-Newton schemes, including BFGS and L-BFGS with different line search strategies approaches. These methods dynamically rescale updates based on historical gradient information, thus enhancing training efficiency and accuracy. We systematically compare these optimizers -- using both PINNs and PIKANs -- on key challenging linear, stiff, multi-scale and non-linear PDEs, including the Burgers, Allen-Cahn, Kuramoto-Sivashinsky, and Ginzburg-Landau equations. Our findings provide state-of-the-art results with orders-of-magnitude accuracy improvements without the use of adaptive weights or any other enhancements typically employed in PINNs. More broadly, our results reveal insights into the effectiveness of second-order optimization strategies in significantly improving the convergence and accurate generalization of PINNs and PIKANs.
comment: 36 pages, 27 figures
♻ ☆ Uncertainty Quantification via Hölder Divergence for Multi-View Representation Learning
Evidence-based deep learning represents a burgeoning paradigm for uncertainty estimation, offering reliable predictions with negligible extra computational overheads. Existing methods usually adopt Kullback-Leibler divergence to estimate the uncertainty of network predictions, ignoring domain gaps among various modalities. To tackle this issue, this paper introduces a novel algorithm based on H\"older Divergence (HD) to enhance the reliability of multi-view learning by addressing inherent uncertainty challenges from incomplete or noisy data. Generally, our method extracts the representations of multiple modalities through parallel network branches, and then employs HD to estimate the prediction uncertainties. Through the Dempster-Shafer theory, integration of uncertainty from different modalities, thereby generating a comprehensive result that considers all available representations. Mathematically, HD proves to better measure the ``distance'' between real data distribution and predictive distribution of the model and improve the performances of multi-class recognition tasks. Specifically, our method surpass the existing state-of-the-art counterparts on all evaluating benchmarks. We further conduct extensive experiments on different backbones to verify our superior robustness. It is demonstrated that our method successfully pushes the corresponding performance boundaries. Finally, we perform experiments on more challenging scenarios, \textit{i.e.}, learning with incomplete or noisy data, revealing that our method exhibits a high tolerance to such corrupted data.
comment: NA
♻ ☆ Frozen Layers: Memory-efficient Many-fidelity Hyperparameter Optimization
As model sizes grow, finding efficient and cost-effective hyperparameter optimization (HPO) methods becomes increasingly crucial for deep learning pipelines. While multi-fidelity HPO (MF-HPO) trades off computational resources required for DL training with lower fidelity estimations, existing fidelity sources often fail under lower compute and memory constraints. We propose a novel fidelity source: the number of layers that are trained or frozen during training. For deep networks, this approach offers significant compute and memory savings while preserving rank correlations between hyperparameters at low fidelities compared to full model training. We demonstrate this in our empirical evaluation across ResNets and Transformers and additionally analyze the utility of frozen layers as a fidelity in using GPU resources as a fidelity in HPO, and for a combined MF-HPO with other fidelity sources. This contribution opens new applications for MF-HPO with hardware resources as a fidelity and creates opportunities for improved algorithms navigating joint fidelity spaces.
♻ ☆ ArtCrafter: Text-Image Aligning Style Transfer via Embedding Reframing
Recent years have witnessed significant advancements in text-guided style transfer, primarily attributed to innovations in diffusion models. These models excel in conditional guidance, utilizing text or images to direct the sampling process. However, despite their capabilities, direct conditional guidance approaches often face challenges in balancing the expressiveness of textual semantics with the diversity of output results while capturing stylistic features. To address these challenges, we introduce ArtCrafter, a novel framework for text-to-image style transfer. Specifically, we introduce an attention-based style extraction module, meticulously engineered to capture the subtle stylistic elements within an image. This module features a multi-layer architecture that leverages the capabilities of perceiver attention mechanisms to integrate fine-grained information. Additionally, we present a novel text-image aligning augmentation component that adeptly balances control over both modalities, enabling the model to efficiently map image and text embeddings into a shared feature space. We achieve this through attention operations that enable smooth information flow between modalities. Lastly, we incorporate an explicit modulation that seamlessly blends multimodal enhanced embeddings with original embeddings through an embedding reframing design, empowering the model to generate diverse outputs. Extensive experiments demonstrate that ArtCrafter yields impressive results in visual stylization, exhibiting exceptional levels of stylistic intensity, controllability, and diversity.
comment: 13 pages, 17 figures, submitted to a journal
♻ ☆ Test-time Alignment of Diffusion Models without Reward Over-optimization ICLR 2025
Diffusion models excel in generative tasks, but aligning them with specific objectives while maintaining their versatility remains challenging. Existing fine-tuning methods often suffer from reward over-optimization, while approximate guidance approaches fail to optimize target rewards effectively. Addressing these limitations, we propose a training-free, test-time method based on Sequential Monte Carlo (SMC) to sample from the reward-aligned target distribution. Our approach, tailored for diffusion sampling and incorporating tempering techniques, achieves comparable or superior target rewards to fine-tuning methods while preserving diversity and cross-reward generalization. We demonstrate its effectiveness in single-reward optimization, multi-objective scenarios, and online black-box optimization. This work offers a robust solution for aligning diffusion models with diverse downstream objectives without compromising their general capabilities. Code is available at https://github.com/krafton-ai/DAS.
comment: ICLR 2025 (Spotlight). The Thirteenth International Conference on Learning Representations. 2025
♻ ☆ Unveiling Molecular Moieties through Hierarchical Grad-CAM Graph Explainability
Background: Virtual Screening (VS) has become an essential tool in drug discovery, enabling the rapid and cost-effective identification of potential bioactive molecules. Among recent advancements, Graph Neural Networks (GNNs) have gained prominence for their ability to model complex molecular structures using graph-based representations. However, the integration of explainable methods to elucidate the specific contributions of molecular substructures to biological activity remains a significant challenge. This limitation hampers both the interpretability of predictive models and the rational design of novel therapeutics.\\ Results: We trained 20 GNN models on a dataset of small molecules with the goal of predicting their activity on 20 distinct protein targets from the Kinase family. These classifiers achieved state-of-the-art performance in virtual screening tasks, demonstrating high accuracy and robustness on different targets. Building upon these models, we implemented the Hierarchical Grad-CAM graph Explainer (HGE) framework, enabling an in-depth analysis of the molecular moieties driving protein-ligand binding stabilization. HGE exploits Grad-CAM explanations at the atom, ring, and whole-molecule levels, leveraging the message-passing mechanism to highlight the most relevant chemical moieties. Validation against experimental data from the literature confirmed the ability of the explainer to recognize a molecular pattern of drugs and correctly annotate them to the known target. Conclusion: Our approach may represent a valid support to shorten both the screening and the hit discovery process. Detailed knowledge of the molecular substructures that play a role in the binding process can help the computational chemist to gain insights into the structure optimization, as well as in drug repurposing tasks.
♻ ☆ Comprehending Knowledge Graphs with Large Language Models for Recommender Systems SIGIR'25
In recent years, the introduction of knowledge graphs (KGs) has significantly advanced recommender systems by facilitating the discovery of potential associations between items. However, existing methods still face several limitations. First, most KGs suffer from missing facts or limited scopes. Second, existing methods convert textual information in KGs into IDs, resulting in the loss of natural semantic connections between different items. Third, existing methods struggle to capture high-order connections in the global KG. To address these limitations, we propose a novel method called CoLaKG, which leverages large language models (LLMs) to improve KG-based recommendations. The extensive knowledge and remarkable reasoning capabilities of LLMs enable our method to supplement missing facts in KGs, and their powerful text understanding abilities allow for better utilization of semantic information. Specifically, CoLaKG extracts useful information from KGs at both local and global levels. By employing the item-centered subgraph extraction and prompt engineering, it can accurately understand the local information. In addition, through the semantic-based retrieval module, each item is enriched by related items from the entire knowledge graph, effectively harnessing global information. Furthermore, the local and global information are effectively integrated into the recommendation model through a representation fusion module and a retrieval-augmented representation learning module, respectively. Extensive experiments on four real-world datasets demonstrate the superiority of our method.
comment: Accepted as a full paper by SIGIR'25
♻ ☆ Large Language Models as Attribution Regularizers for Efficient Model Training
Large Language Models (LLMs) have demonstrated remarkable performance across diverse domains. However, effectively leveraging their vast knowledge for training smaller downstream models remains an open challenge, especially in domains like tabular data learning, where simpler models are often preferred due to interpretability and efficiency. In this paper, we introduce a novel yet straightforward method for incorporating LLM-generated global task feature attributions into the training process of smaller networks. Specifically, we propose an attribution-matching regularization term that aligns the training dynamics of the smaller model with the insights provided by the LLM. By doing so, our approach yields superior performance in few-shot learning scenarios. Notably, our method requires only black-box API access to the LLM, making it easy to integrate into existing training pipelines with minimal computational overhead. Furthermore, we demonstrate how this method can be used to address common issues in real-world datasets, such as skewness and bias. By integrating high-level knowledge from LLMs, our approach improves generalization, even when training data is limited or imbalanced. We validate its effectiveness through extensive experiments across multiple tasks, demonstrating improved learning efficiency and model robustness.
♻ ☆ Multi-Stakeholder Disaster Insights from Social Media Using Large Language Models
In recent years, social media has emerged as a primary channel for users to promptly share feedback and issues during disasters and emergencies, playing a key role in crisis management. While significant progress has been made in collecting and analyzing social media content, there remains a pressing need to enhance the automation, aggregation, and customization of this data to deliver actionable insights tailored to diverse stakeholders, including the press, police, EMS, and firefighters. This effort is essential for improving the coordination of activities such as relief efforts, resource distribution, and media communication. This paper presents a methodology that leverages the capabilities of LLMs to enhance disaster response and management. Our approach combines classification techniques with generative AI to bridge the gap between raw user feedback and stakeholder-specific reports. Social media posts shared during catastrophic events are analyzed with a focus on user-reported issues, service interruptions, and encountered challenges. We employ full-spectrum LLMs, using analytical models like BERT for precise, multi-dimensional classification of content type, sentiment, emotion, geolocation, and topic. Generative models such as ChatGPT are then used to produce human-readable, informative reports tailored to distinct audiences, synthesizing insights derived from detailed classifications. We compare standard approaches, which analyze posts directly using prompts in ChatGPT, to our advanced method, which incorporates multi-dimensional classification, sub-event selection, and tailored report generation. Our methodology demonstrates superior performance in both quantitative metrics, such as text coherence scores and latent representations, and qualitative assessments by automated tools and field experts, delivering precise insights for diverse disaster response stakeholders.
♻ ☆ Curriculum-based Sample Efficient Reinforcement Learning for Robust Stabilization of a Quadrotor
This article introduces a curriculum learning approach to develop a reinforcement learning-based robust stabilizing controller for a Quadrotor that meets predefined performance criteria. The learning objective is to achieve desired positions from random initial conditions while adhering to both transient and steady-state performance specifications. This objective is challenging for conventional one-stage end-to-end reinforcement learning, due to the strong coupling between position and orientation dynamics, the complexity in designing and tuning the reward function, and poor sample efficiency, which necessitates substantial computational resources and leads to extended convergence times. To address these challenges, this work decomposes the learning objective into a three-stage curriculum that incrementally increases task complexity. The curriculum begins with learning to achieve stable hovering from a fixed initial condition, followed by progressively introducing randomization in initial positions, orientations and velocities. A novel additive reward function is proposed, to incorporate transient and steady-state performance specifications. The results demonstrate that the Proximal Policy Optimization (PPO)-based curriculum learning approach, coupled with the proposed reward structure, achieves superior performance compared to a single-stage PPO-trained policy with the same reward function, while significantly reducing computational resource requirements and convergence time. The curriculum-trained policy's performance and robustness are thoroughly validated under random initial conditions and in the presence of disturbances.
comment: 8 pages, 7 figures
♻ ☆ Abstract Meaning Representation-Based Logic-Driven Data Augmentation for Logical Reasoning ACL 2024
Combining large language models with logical reasoning enhances their capacity to address problems in a robust and reliable manner. Nevertheless, the intricate nature of logical reasoning poses challenges when gathering reliable data from the web to build comprehensive training datasets, subsequently affecting performance on downstream tasks. To address this, we introduce a novel logic-driven data augmentation approach, AMR-LDA. AMR-LDA converts the original text into an Abstract Meaning Representation (AMR) graph, a structured semantic representation that encapsulates the logical structure of the sentence, upon which operations are performed to generate logically modified AMR graphs. The modified AMR graphs are subsequently converted back into text to create augmented data. Notably, our methodology is architecture-agnostic and enhances both generative large language models, such as GPT-3.5 and GPT-4, through prompt augmentation, and discriminative large language models through contrastive learning with logic-driven data augmentation. Empirical evidence underscores the efficacy of our proposed method with improvement in performance across seven downstream tasks, such as reading comprehension requiring logical reasoning, textual entailment, and natural language inference. Furthermore, our method leads on the ReClor leaderboard at https://eval.ai/web/challenges/challenge-page/503/leaderboard/1347. The source code and data are publicly available at https://github.com/Strong-AI-Lab/Logical-Equivalence-driven-AMR-Data-Augmentation-for-Representation-Learning.
comment: 21 pages, 8 figures, the Findings of ACL 2024
♻ ☆ Multi-Step Deductive Reasoning Over Natural Language: An Empirical Study on Out-of-Distribution Generalisation
Combining deep learning with symbolic logic reasoning aims to capitalize on the success of both fields and is drawing increasing attention. Inspired by DeepLogic, an end-to-end model trained to perform inference on logic programs, we introduce IMA-GloVe-GA, an iterative neural inference network for multi-step reasoning expressed in natural language. In our model, reasoning is performed using an iterative memory neural network based on RNN with a gated attention mechanism. We evaluate IMA-GloVe-GA on three datasets: PARARULES, CONCEPTRULES V1 and CONCEPTRULES V2. Experimental results show DeepLogic with gated attention can achieve higher test accuracy than DeepLogic and other RNN baseline models. Our model achieves better out-of-distribution generalisation than RoBERTa-Large when the rules have been shuffled. Furthermore, to address the issue of unbalanced distribution of reasoning depths in the current multi-step reasoning datasets, we develop PARARULE-Plus, a large dataset with more examples that require deeper reasoning steps. Experimental results show that the addition of PARARULE-Plus can increase the model's performance on examples requiring deeper reasoning depths. The source code and data are available at https://github.com/Strong-AI-Lab/Multi-Step-Deductive-Reasoning-Over-Natural-Language.
comment: 10 pages, 3 figures, The 2nd International Joint Conference on Learning & Reasoning and 16th International Workshop on Neural-Symbolic Learning and Reasoning (IJCLR-NeSy 2022)
♻ ☆ Unified Domain Adaptive Semantic Segmentation
Unsupervised Domain Adaptive Semantic Segmentation (UDA-SS) aims to transfer the supervision from a labeled source domain to an unlabeled target domain. The majority of existing UDA-SS works typically consider images whilst recent attempts have extended further to tackle videos by modeling the temporal dimension. Although the two lines of research share the major challenges -- overcoming the underlying domain distribution shift, their studies are largely independent, resulting in fragmented insights, a lack of holistic understanding, and missed opportunities for cross-pollination of ideas. This fragmentation prevents the unification of methods, leading to redundant efforts and suboptimal knowledge transfer across image and video domains. Under this observation, we advocate unifying the study of UDA-SS across video and image scenarios, enabling a more comprehensive understanding, synergistic advancements, and efficient knowledge sharing. To that end, we explore the unified UDA-SS from a general data augmentation perspective, serving as a unifying conceptual framework, enabling improved generalization, and potential for cross-pollination of ideas, ultimately contributing to the overall progress and practical impact of this field of research. Specifically, we propose a Quad-directional Mixup (QuadMix) method, characterized by tackling distinct point attributes and feature inconsistencies through four-directional paths for intra- and inter-domain mixing in a feature space. To deal with temporal shifts with videos, we incorporate optical flow-guided feature aggregation across spatial and temporal dimensions for fine-grained domain alignment. Extensive experiments show that our method outperforms the state-of-the-art works by large margins on four challenging UDA-SS benchmarks. Our source code and models will be released at https://github.com/ZHE-SAPI/UDASS.
comment: 17 pages,11 figures, 11 tables. Accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2025
♻ ☆ Predicting and Publishing Accurate Imbalance Prices Using Monte Carlo Tree Search
The growing reliance on renewable energy sources, particularly solar and wind, has introduced challenges due to their uncontrollable production. This complicates maintaining the electrical grid balance, prompting some transmission system operators in Western Europe to implement imbalance tariffs that penalize unsustainable power deviations. These tariffs create an implicit demand response framework to mitigate grid instability. Yet, several challenges limit active participation. In Belgium, for example, imbalance prices are only calculated at the end of each 15-minute settlement period, creating high risk due to price uncertainty. This risk is further amplified by the inherent volatility of imbalance prices, discouraging participation. Although transmission system operators provide minute-based price predictions, the system imbalance volatility makes accurate price predictions challenging to obtain and requires sophisticated techniques. Moreover, publishing price estimates can prompt participants to adjust their schedules, potentially affecting the system balance and the final price, adding further complexity. To address these challenges, we propose a Monte Carlo Tree Search method that publishes accurate imbalance prices while accounting for potential response actions. Our approach models the system dynamics using a neural network forecaster and a cluster of virtual batteries controlled by reinforcement learning agents. Compared to Belgium's current publication method, our technique improves price accuracy by 20.4% under ideal conditions and by 12.8% in more realistic scenarios. This research addresses an unexplored, yet crucial problem, positioning this paper as a pioneering work in analyzing the potential of more advanced imbalance price publishing techniques.
♻ ☆ LSEAttention is All You Need for Time Series Forecasting
Transformer-based architectures have achieved remarkable success in natural language processing and computer vision. However, their performance in multivariate long-term forecasting often falls short compared to simpler linear baselines. Previous research has identified the traditional attention mechanism as a key factor limiting their effectiveness in this domain. To bridge this gap, we introduce LATST, a novel approach designed to mitigate entropy collapse and training instability common challenges in Transformer-based time series forecasting. We rigorously evaluate LATST across multiple real-world multivariate time series datasets, demonstrating its ability to outperform existing state-of-the-art Transformer models. Notably, LATST manages to achieve competitive performance with fewer parameters than some linear models on certain datasets, highlighting its efficiency and effectiveness.
comment: 8 pages with referencing, 1 figure, 5 tables
♻ ☆ Near, far: Patch-ordering enhances vision foundation models' scene understanding ICLR25
We introduce NeCo: Patch Neighbor Consistency, a novel self-supervised training loss that enforces patch-level nearest neighbor consistency across a student and teacher model. Compared to contrastive approaches that only yield binary learning signals, i.e., 'attract' and 'repel', this approach benefits from the more fine-grained learning signal of sorting spatially dense features relative to reference patches. Our method leverages differentiable sorting applied on top of pretrained representations, such as DINOv2-registers to bootstrap the learning signal and further improve upon them. This dense post-pretraining leads to superior performance across various models and datasets, despite requiring only 19 hours on a single GPU. This method generates high-quality dense feature encoders and establishes several new state-of-the-art results such as +5.5% and +6% for non-parametric in-context semantic segmentation on ADE20k and Pascal VOC, +7.2% and +5.7% for linear segmentation evaluations on COCO-Things and -Stuff and improvements in the 3D understanding of multi-view consistency on SPair-71k, by more than 1.5%.
comment: Accepted at ICLR25. The webpage is accessible at: https://vpariza.github.io/NeCo/
♻ ☆ Why We Feel: Breaking Boundaries in Emotional Reasoning with Multimodal Large Language Models CVPR
Most existing emotion analysis emphasizes which emotion arises (e.g., happy, sad, angry) but neglects the deeper why. We propose Emotion Interpretation (EI), focusing on causal factors-whether explicit (e.g., observable objects, interpersonal interactions) or implicit (e.g., cultural context, off-screen events)-that drive emotional responses. Unlike traditional emotion recognition, EI tasks require reasoning about triggers instead of mere labeling. To facilitate EI research, we present EIBench, a large-scale benchmark encompassing 1,615 basic EI samples and 50 complex EI samples featuring multifaceted emotions. Each instance demands rationale-based explanations rather than straightforward categorization. We further propose a Coarse-to-Fine Self-Ask (CFSA) annotation pipeline, which guides Vision-Language Models (VLLMs) through iterative question-answer rounds to yield high-quality labels at scale. Extensive evaluations on open-source and proprietary large language models under four experimental settings reveal consistent performance gaps-especially for more intricate scenarios-underscoring EI's potential to enrich empathetic, context-aware AI applications. Our benchmark and methods are publicly available at: https://github.com/Lum1104/EIBench, offering a foundation for advanced multimodal causal analysis and next-generation affective computing.
comment: Accepted at CVPR Workshop NEXD 2025. 21 pages, Project: https://github.com/Lum1104/EIBench
♻ ☆ Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions CVPR
Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges. This survey provides a wide-ranging examination of the anti-UAV domain, centering on three core objectives-classification, detection, and tracking-while detailing emerging methodologies such as diffusion-based data synthesis, multi-modal fusion, vision-language modeling, self-supervised learning, and reinforcement learning. We systematically evaluate state-of-the-art solutions across both single-modality and multi-sensor pipelines (spanning RGB, infrared, audio, radar, and RF) and discuss large-scale as well as adversarially oriented benchmarks. Our analysis reveals persistent gaps in real-time performance, stealth detection, and swarm-based scenarios, underscoring pressing needs for robust, adaptive anti-UAV systems. By highlighting open research directions, we aim to foster innovation and guide the development of next-generation defense strategies in an era marked by the extensive use of UAVs.
comment: Accepted at CVPR Workshop Anti-UAV 2025. 15 pages
♻ ☆ Decentralized Federated Domain Generalization with Style Sharing: A Formal Modeling and Convergence Analysis
Much of the federated learning (FL) literature focuses on settings where local dataset statistics remain the same between training and testing time. Recent advances in domain generalization (DG) aim to use data from source (training) domains to train a model that generalizes well to data from unseen target (testing) domains. In this paper, we are motivated by two major gaps in existing work on FL and DG: (1) the lack of formal mathematical analysis of DG objectives and training processes; and (2) DG research in FL being limited to the conventional star-topology architecture. Addressing the second gap, we develop $\textit{Decentralized Federated Domain Generalization with Style Sharing}$ ($\texttt{StyleDDG}$), a fully decentralized DG algorithm designed to allow devices in a peer-to-peer network to achieve DG based on sharing style information inferred from their datasets. Additionally, we fill the first gap by providing the first systematic approach to mathematically analyzing style-based DG training optimization. We cast existing centralized DG algorithms within our framework, and employ their formalisms to model $\texttt{StyleDDG}$. Based on this, we obtain analytical conditions under which a sub-linear convergence rate of $\texttt{StyleDDG}$ can be obtained. Through experiments on two popular DG datasets, we demonstrate that $\texttt{StyleDDG}$ can obtain significant improvements in accuracy across target domains with minimal added communication overhead compared to decentralized gradient methods that do not employ style sharing.
♻ ☆ Causality-enhanced Decision-Making for Autonomous Mobile Robots in Dynamic Environments
The growing integration of robots in shared environments -- such as warehouses, shopping centres, and hospitals -- demands a deep understanding of the underlying dynamics and human behaviours, including how, when, and where individuals engage in various activities and interactions. This knowledge goes beyond simple correlation studies and requires a more comprehensive causal analysis. By leveraging causal inference to model cause-and-effect relationships, we can better anticipate critical environmental factors and enable autonomous robots to plan and execute tasks more effectively. To this end, we propose a novel causality-based decision-making framework that reasons over a learned causal model to predict battery usage and human obstructions, understanding how these factors could influence robot task execution. Such reasoning framework assists the robot in deciding when and how to complete a given task. To achieve this, we developed also PeopleFlow, a new Gazebo-based simulator designed to model context-sensitive human-robot spatial interactions in shared workspaces. PeopleFlow features realistic human and robot trajectories influenced by contextual factors such as time, environment layout, and robot state, and can simulate a large number of agents. While the simulator is general-purpose, in this paper we focus on a warehouse-like environment as a case study, where we conduct an extensive evaluation benchmarking our causal approach against a non-causal baseline. Our findings demonstrate the efficacy of the proposed solutions, highlighting how causal reasoning enables autonomous robots to operate more efficiently and safely in dynamic environments shared with humans.
comment: Causal Discovery and Inference - Robot Autonomy - Human-Robot Spatial Interaction - Decision-Making
♻ ☆ Evaluating Semantic Variation in Text-to-Image Synthesis: A Causal Perspective ICLR 2025
Accurate interpretation and visualization of human instructions are crucial for text-to-image (T2I) synthesis. However, current models struggle to capture semantic variations from word order changes, and existing evaluations, relying on indirect metrics like text-image similarity, fail to reliably assess these challenges. This often obscures poor performance on complex or uncommon linguistic patterns by the focus on frequent word combinations. To address these deficiencies, we propose a novel metric called SemVarEffect and a benchmark named SemVarBench, designed to evaluate the causality between semantic variations in inputs and outputs in T2I synthesis. Semantic variations are achieved through two types of linguistic permutations, while avoiding easily predictable literal variations. Experiments reveal that the CogView-3-Plus and Ideogram 2 performed the best, achieving a score of 0.2/1. Semantic variations in object relations are less understood than attributes, scoring 0.07/1 compared to 0.17-0.19/1. We found that cross-modal alignment in UNet or Transformers plays a crucial role in handling semantic variations, a factor previously overlooked by a focus on textual encoders. Our work establishes an effective evaluation framework that advances the T2I synthesis community's exploration of human instruction understanding. Our benchmark and code are available at https://github.com/zhuxiangru/SemVarBench .
comment: Accepted by ICLR 2025
♻ ☆ De Novo Generation of Hit-like Molecules from Gene Expression Profiles via Deep Learning
De novo generation of hit-like molecules is a challenging task in the drug discovery process. Most methods in previous studies learn the semantics and syntax of molecular structures by analyzing molecular graphs or simplified molecular input line entry system (SMILES) strings; however, they do not take into account the drug responses of the biological systems consisting of genes and proteins. In this study we propose a hybrid neural network, HNN2Mol, which utilizes gene expression profiles to generate molecular structures with desirable phenotypes for arbitrary target proteins. In the algorithm, a variational autoencoder is employed as a feature extractor to learn the latent feature distribution of the gene expression profiles. Then, a long short-term memory is leveraged as the chemical generator to produce syntactically valid SMILES strings that satisfy the feature conditions of the gene expression profile extracted by the feature extractor. Experimental results and case studies demonstrate that the proposed HNN2Mol model can produce new molecules with potential bioactivities and drug-like properties.
♻ ☆ Arithmetic Transformers Can Length-Generalize in Both Operand Length and Count ICLR 2025
Transformers often struggle with length generalization, meaning they fail to generalize to sequences longer than those encountered during training. While arithmetic tasks are commonly used to study length generalization, certain tasks are considered notoriously difficult, e.g., multi-operand addition (requiring generalization over both the number of operands and their lengths) and multiplication (requiring generalization over both operand lengths). In this work, we achieve approximately 2-3x length generalization on both tasks, which is the first such achievement in arithmetic Transformers. We design task-specific scratchpads enabling the model to focus on a fixed number of tokens per each next-token prediction step, and apply multi-level versions of \Position Coupling (Cho et al., 2024; McLeish et al., 2024) to let Transformers know the right position to attend to. On the theory side, we prove that a 1-layer Transformer using our method can solve multi-operand addition, up to operand length and operand count that are exponential in embedding dimension.
comment: 44 pages, 20 figures, 26 tables, accepted to ICLR 2025
♻ ☆ Look Before You Decide: Prompting Active Deduction of MLLMs for Assumptive Reasoning
Recently, Multimodal Large Language Models (MLLMs) have achieved significant success across multiple disciplines due to their exceptional instruction-following capabilities and extensive world knowledge. However, whether these MLLMs possess human-like compositional reasoning abilities remains an open problem. To unveil their reasoning behaviors, we first curate a \textbf{M}ultimodal \textbf{A}ssumptive \textbf{R}ea\textbf{s}oning Benchmark (MARS-Bench) in this paper. Interestingly, we find that most prevalent MLLMs can be easily fooled by the introduction of a presupposition into the question, whereas such presuppositions appear naive to human reasoning. Besides, we also propose a simple yet effective method, Active Deduction (AD), a novel reinforcement learning paradigm to encourage the model to actively perform composite deduction before reaching a final decision. Equipped with the proposed AD method, a MLLM demonstrates significant improvements in assumptive reasoning abilities without compromising its general-purpose question-answering performance. We also provide extensive evaluations of both open-source and private MLLMs on MARS-Bench, along with experimental analyses of the AD method.
♻ ☆ Systems with Switching Causal Relations: A Meta-Causal Perspective ICLR 2025
Most work on causality in machine learning assumes that causal relationships are driven by a constant underlying process. However, the flexibility of agents' actions or tipping points in the environmental process can change the qualitative dynamics of the system. As a result, new causal relationships may emerge, while existing ones change or disappear, resulting in an altered causal graph. To analyze these qualitative changes on the causal graph, we propose the concept of meta-causal states, which groups classical causal models into clusters based on equivalent qualitative behavior and consolidates specific mechanism parameterizations. We demonstrate how meta-causal states can be inferred from observed agent behavior, and discuss potential methods for disentangling these states from unlabeled data. Finally, we direct our analysis towards the application of a dynamical system, showing that meta-causal states can also emerge from inherent system dynamics, and thus constitute more than a context-dependent framework in which mechanisms emerge only as a result of external factors.
comment: 21 pages, 3 figures, 4 tables, ICLR 2025 Camera Ready Version
♻ ☆ Applications of Statistical Field Theory in Deep Learning
Deep learning algorithms have made incredible strides in the past decade, yet due to their complexity, the science of deep learning remains in its early stages. Being an experimentally driven field, it is natural to seek a theory of deep learning within the physics paradigm. As deep learning is largely about learning functions and distributions over functions, statistical field theory, a rich and versatile toolbox for tackling complex distributions over functions (fields) is an obvious choice of formalism. Research efforts carried out in the past few years have demonstrated the ability of field theory to provide useful insights on generalization, implicit bias, and feature learning effects. Here we provide a pedagogical review of this emerging line of research.
♻ ☆ Towards Scientific Intelligence: A Survey of LLM-based Scientific Agents
As scientific research becomes increasingly complex, innovative tools are needed to manage vast data, facilitate interdisciplinary collaboration, and accelerate discovery. Large language models (LLMs) are now evolving into LLM-based scientific agents that automate critical tasks, ranging from hypothesis generation and experiment design to data analysis and simulation. Unlike general-purpose LLMs, these specialized agents integrate domain-specific knowledge, advanced tool sets, and robust validation mechanisms, enabling them to handle complex data types, ensure reproducibility, and drive scientific breakthroughs. This survey provides a focused review of the architectures, design, benchmarks, applications, and ethical considerations surrounding LLM-based scientific agents. We highlight why they differ from general agents and the ways in which they advance research across various scientific fields. By examining their development and challenges, this survey offers a comprehensive roadmap for researchers and practitioners to harness these agents for more efficient, reliable, and ethically sound scientific discovery.
comment: 34 pages, 10 figures
♻ ☆ A Shapley Value Estimation Speedup for Efficient Explainable Quantum AI
This work focuses on developing efficient post-hoc explanations for quantum AI algorithms. In classical contexts, the cooperative game theory concept of the Shapley value adapts naturally to post-hoc explanations, where it can be used to identify which factors are important in an AI's decision-making process. An interesting question is how to translate Shapley values to the quantum setting and whether quantum effects could be used to accelerate their calculation. We propose quantum algorithms that can extract Shapley values within some confidence interval. Our method is capable of quadratically outperforming classical Monte Carlo approaches to approximating Shapley values up to polylogarithmic factors in various circumstances. We demonstrate the validity of our approach empirically with specific voting games and provide rigorous proofs of performance for general cooperative games.
comment: 34 pages, 4 figures, 4 tables, 45 citations
♻ ☆ A Systematic Review on Sleep Stage Classification and Sleep Disorder Detection Using Artificial Intelligence
Sleep is vital for people's physical and mental health, and sound sleep can help them focus on daily activities. Therefore, a sleep study that includes sleep patterns and sleep disorders is crucial to enhancing our knowledge about individuals' health status. This study aims to provide a comprehensive, systematic review of the recent literature to analyze the different approaches and their outcomes in sleep studies, which includes works on "sleep stages classification" and "sleep disorder detection" using AI. In this review, 183 articles were initially selected from different journals, among which 80 records were enlisted for explicit review, ranging from 2016 to 2023. Brain waves were the most commonly employed body parameters for sleep staging and disorder studies (almost 29% of the research used brain activity signals exclusively, and 77% combined with the other signals). The convolutional neural network (CNN), the most widely used of the 34 distinct artificial intelligence models, comprised 27%. The other models included the long short-term memory (LSTM), support vector machine (SVM), random forest (RF), and recurrent neural network (RNN), which consisted of 11%, 6%, 6%, and 5% sequentially. For performance metrics, accuracy was widely used for a maximum of 83.75% of the cases, the F1 score of 45%, Kappa of 36.25%, Sensitivity of 31.25%, and Specificity of 30% of cases, along with the other metrics. This article would help physicians and researchers get the gist of AI's contribution to sleep studies and the feasibility of their intended work.
comment: 39 pages, 11 Figures, 8 Tables
♻ ☆ Selective Attention Federated Learning: Improving Privacy and Efficiency for Clinical Text Classification
Federated Learning (FL) faces major challenges regarding communication overhead and model privacy when training large language models (LLMs), especially in healthcare applications. To address these, we introduce Selective Attention Federated Learning (SAFL), a novel approach that dynamically fine-tunes only those transformer layers identified as attention-critical. By employing attention patterns to determine layer importance, SAFL significantly reduces communication bandwidth and enhances differential privacy resilience. Evaluations on clinical NLP benchmarks (i2b2 Clinical Concept Extraction and MIMIC-III discharge summaries) demonstrate that SAFL achieves competitive performance with centralized models while substantially improving communication efficiency and privacy preservation.
♻ ☆ Uncertainty Calibration for Counterfactual Propensity Estimation in Recommendation
Post-click conversion rate (CVR) is a reliable indicator of online customers' preferences, making it crucial for developing recommender systems. A major challenge in predicting CVR is severe selection bias, arising from users' inherent self-selection behavior and the system's item selection process. To mitigate this issue, the inverse propensity score (IPS) is employed to weight the prediction error of each observed instance. However, current propensity score estimations are unreliable due to the lack of a quality measure. To address this, we evaluate the quality of propensity scores from the perspective of uncertainty calibration, proposing the use of Expected Calibration Error (ECE) as a measure of propensity-score quality, which quantifies the extent to which predicted probabilities are overconfident by assessing the difference between predicted probabilities and actual observed frequencies. Miscalibrated propensity scores can lead to distorted IPS weights, thereby compromising the debiasing process in CVR prediction. In this paper, we introduce a model-agnostic calibration framework for propensity-based debiasing of CVR predictions. Theoretical analysis on bias and generalization bounds demonstrates the superiority of calibrated propensity estimates over uncalibrated ones. Experiments conducted on the Coat, Yahoo and KuaiRand datasets show improved uncertainty calibration, as evidenced by lower ECE values, leading to enhanced CVR prediction outcomes.
comment: This is the accepted manuscript version of the IEEE TKDE paper. The final published version will be available at: https://doi.ieeecomputersociety.org/10.1109/TKDE.2025.3552658
♻ ☆ Transfer Learning for Temporal Link Prediction
Link prediction on graphs has applications spanning from recommender systems to drug discovery. Temporal link prediction (TLP) refers to predicting future links in a temporally evolving graph and adds additional complexity related to the dynamic nature of graphs. State-of-the-art TLP models incorporate memory modules alongside graph neural networks to learn both the temporal mechanisms of incoming nodes and the evolving graph topology. However, memory modules only store information about nodes seen at train time, and hence such models cannot be directly transferred to entirely new graphs at test time and deployment. In this work, we study a new transfer learning task for temporal link prediction, and develop transfer-effective methods for memory-laden models. Specifically, motivated by work showing the informativeness of structural signals for the TLP task, we augment a structural mapping module to the existing TLP model architectures, which learns a mapping from graph structural (topological) features to memory embeddings. Our work paves the way for a memory-free foundation model for TLP.
comment: 14 pages, 7 figures
♻ ☆ DeepResearcher: Scaling Deep Research via Reinforcement Learning in Real-world Environments
Large Language Models (LLMs) equipped with web search capabilities have demonstrated impressive potential for deep research tasks. However, current approaches predominantly rely on either manually engineered prompts (prompt engineering-based) with brittle performance or reinforcement learning within controlled Retrieval-Augmented Generation (RAG) environments (RAG-based) that fail to capture the complexities of real-world interaction. In this paper, we introduce DeepResearcher, the first comprehensive framework for end-to-end training of LLM-based deep research agents through scaling reinforcement learning (RL) in real-world environments with authentic web search interactions. Unlike RAG-based approaches that assume all necessary information exists within a fixed corpus, our method trains agents to navigate the noisy, unstructured, and dynamic nature of the open web. We implement a specialized multi-agent architecture where browsing agents extract relevant information from various webpage structures and overcoming significant technical challenges. Extensive experiments on open-domain research tasks demonstrate that DeepResearcher achieves substantial improvements of up to 28.9 points over prompt engineering-based baselines and up to 7.2 points over RAG-based RL agents. Our qualitative analysis reveals emergent cognitive behaviors from end-to-end RL training, including the ability to formulate plans, cross-validate information from multiple sources, engage in self-reflection to redirect research, and maintain honesty when unable to find definitive answers. Our results highlight that end-to-end training in real-world web environments is not merely an implementation detail but a fundamental requirement for developing robust research capabilities aligned with real-world applications. We release DeepResearcher at https://github.com/GAIR-NLP/DeepResearcher.
♻ ☆ UMSPU: Universal Multi-Size Phase Unwrapping via Mutual Self-Distillation and Adaptive Boosting Ensemble Segmenters
Spatial phase unwrapping is a key technique for extracting phase information to obtain 3D morphology and other features. Modern industrial measurement scenarios demand high precision, large image sizes, and high speed. However, conventional methods struggle with noise resistance and processing speed. Current deep learning methods are limited by the receptive field size and sparse semantic information, making them ineffective for large size images. To address this issue, we propose a mutual self-distillation (MSD) mechanism and adaptive boosting ensemble segmenters to construct a universal multi-size phase unwrapping network (UMSPU). MSD performs hierarchical attention refinement and achieves cross-layer collaborative learning through bidirectional distillation, ensuring fine-grained semantic representation across image sizes. The adaptive boosting ensemble segmenters combine weak segmenters with different receptive fields into a strong one, ensuring stable segmentation across spatial frequencies. Experimental results show that UMSPU overcomes image size limitations, achieving high precision across image sizes ranging from 256*256 to 2048*2048 (an 8 times increase). It also outperforms existing methods in speed, robustness, and generalization. Its practicality is further validated in structured light imaging and InSAR. We believe that UMSPU offers a universal solution for phase unwrapping, with broad potential for industrial applications.
♻ ☆ Exploring the Role of Knowledge Graph-Based RAG in Japanese Medical Question Answering with Small-Scale LLMs
Large language models (LLMs) perform well in medical QA, but their effectiveness in Japanese contexts is limited due to privacy constraints that prevent the use of commercial models like GPT-4 in clinical settings. As a result, recent efforts focus on instruction-tuning open-source LLMs, though the potential of combining them with retrieval-augmented generation (RAG) remains underexplored. To bridge this gap, we are the first to explore a knowledge graph-based (KG) RAG framework for Japanese medical QA small-scale open-source LLMs. Experimental results show that KG-based RAG has only a limited impact on Japanese medical QA using small-scale open-source LLMs. Further case studies reveal that the effectiveness of the RAG is sensitive to the quality and relevance of the external retrieved content. These findings offer valuable insights into the challenges and potential of applying RAG in Japanese medical QA, while also serving as a reference for other low-resource languages.
comment: 10 pages
♻ ☆ Engineering Artificial Intelligence: Framework, Challenges, and Future Direction
Over the past ten years, the application of artificial intelligence (AI) and machine learning (ML) in engineering domains has gained significant popularity, showcasing their potential in data-driven contexts. However, the complexity and diversity of engineering problems often require the development of domain-specific AI approaches, which are frequently hindered by a lack of systematic methodologies, scalability, and robustness during the development process. To address this gap, this paper introduces the "ABCDE" as the key elements of Engineering AI and proposes a unified, systematic engineering AI ecosystem framework, including eight essential layers, along with attributes, goals, and applications, to guide the development and deployment of AI solutions for specific engineering needs. Additionally, key challenges are examined, and eight future research directions are highlighted. By providing a comprehensive perspective, this paper aims to advance the strategic implementation of AI, fostering the development of next-generation engineering AI solutions.
comment: The paper submitted to the Journal Machine Learning: Engineering has been accepted
♻ ☆ Learning to Help in Multi-Class Settings ICLR 2025
Deploying complex machine learning models on resource-constrained devices is challenging due to limited computational power, memory, and model retrainability. To address these limitations, a hybrid system can be established by augmenting the local model with a server-side model, where samples are selectively deferred by a rejector and then sent to the server for processing. The hybrid system enables efficient use of computational resources while minimizing the overhead associated with server usage. The recently proposed Learning to Help (L2H) model trains a server model given a fixed local (client) model, differing from the Learning to Defer (L2D) framework, which trains the client for a fixed (expert) server. In both L2D and L2H, the training includes learning a rejector at the client to determine when to query the server. In this work, we extend the L2H model from binary to multi-class classification problems and demonstrate its applicability in a number of different scenarios of practical interest in which access to the server may be limited by cost, availability, or policy. We derive a stage-switching surrogate loss function that is differentiable, convex, and consistent with the Bayes rule corresponding to the 0-1 loss for the L2H model. Experiments show that our proposed methods offer an efficient and practical solution for multi-class classification in resource-constrained environments.
comment: 30 pages, 7 figures, conference, ICLR 2025
♻ ☆ Rethinking industrial artificial intelligence: a unified foundation framework
Recent advancements in industrial artificial intelligence (AI) are reshaping the industry by driving smarter manufacturing, predictive maintenance, and intelligent decision-making. However, existing approaches often focus primarily on algorithms and models while overlooking the importance of systematically integrating domain knowledge, data, and models to develop more comprehensive and effective AI solutions. Therefore, the effective development and deployment of industrial AI require a more comprehensive and systematic approach. To address this gap, this paper reviews previous research, rethinks the role of industrial AI, and proposes a unified industrial AI foundation framework comprising three core modules: the knowledge module, data module, and model module. These modules help to extend and enhance the industrial AI methodology platform, supporting various industrial applications. In addition, a case study on rotating machinery diagnosis is presented to demonstrate the effectiveness of the proposed framework, and several future directions are highlighted for the development of the industrial AI foundation framework.
comment: The paper submitted to IJAMD, the International Journal of AI for Materials and Design, has been accepted
♻ ☆ Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework
Multimodal foundation models offer a promising framework for robotic perception and planning by processing sensory inputs to generate actionable plans. However, addressing uncertainty in both perception (sensory interpretation) and decision-making (plan generation) remains a critical challenge for ensuring task reliability. We present a comprehensive framework to disentangle, quantify, and mitigate these two forms of uncertainty. We first introduce a framework for uncertainty disentanglement, isolating perception uncertainty arising from limitations in visual understanding and decision uncertainty relating to the robustness of generated plans. To quantify each type of uncertainty, we propose methods tailored to the unique properties of perception and decision-making: we use conformal prediction to calibrate perception uncertainty and introduce Formal-Methods-Driven Prediction (FMDP) to quantify decision uncertainty, leveraging formal verification techniques for theoretical guarantees. Building on this quantification, we implement two targeted intervention mechanisms: an active sensing process that dynamically re-observes high-uncertainty scenes to enhance visual input quality and an automated refinement procedure that fine-tunes the model on high-certainty data, improving its capability to meet task specifications. Empirical validation in real-world and simulated robotic tasks demonstrates that our uncertainty disentanglement framework reduces variability by up to 40% and enhances task success rates by 5% compared to baselines. These improvements are attributed to the combined effect of both interventions and highlight the importance of uncertainty disentanglement, which facilitates targeted interventions that enhance the robustness and reliability of autonomous systems. Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/.
comment: Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/
♻ ☆ Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge
LLM-as-a-Judge has emerged as a promising alternative to human evaluators across various tasks, yet inherent biases - particularly position bias, the tendency to favor solutions based on their position within the prompt - compromise its reliability. This exploratory study evaluates position bias in LLM judges across pairwise and list-wise comparison settings, introducing three metrics: repetition stability, position consistency, and preference fairness. Our experiments, involving 15 LLM judges across MTBench and DevBench with 22 tasks and approximately 40 solution-generating models, result in over 150,000 evaluation instances. We identify Judge-Level, Candidate-Level, and Task-Level factors contributing to bias. The findings confirm that position bias is not due to random chance and varies significantly across judges and tasks. While position bias is weakly influenced by the length of prompt components, it is strongly affected by the quality gap between solutions. Our agreement and disagreement analysis among judges further provides insights into the distribution of judging difficulty across the dataset, and highlights the potential for dataset modifications.
♻ ☆ The Hitchhiker's Guide to Program Analysis, Part II: Deep Thoughts by LLMs
Static analysis is a cornerstone for software vulnerability detection, yet it often struggles with the classic precision-scalability trade-off. In practice, such tools often produce high false positive rates, particularly in large codebases like the Linux kernel. This imprecision can arise from simplified vulnerability modeling and over-approximation of path and data constraints. While large language models (LLMs) show promise in code understanding, their naive application to program analysis yields unreliable results due to inherent reasoning limitations. We introduce BugLens, a post-refinement framework that significantly improves static analysis precision. BugLens guides an LLM to follow traditional analysis steps by assessing buggy code patterns for security impact and validating the constraints associated with static warnings. Evaluated on real-world Linux kernel bugs, BugLens raises precision from 0.10 (raw) and 0.50 (semi-automated refinement) to 0.72, substantially reducing false positives and revealing four previously unreported vulnerabilities. Our results suggest that a structured LLM-based workflow can meaningfully enhance the effectiveness of static analysis tools.
♻ ☆ ZeroSumEval: An Extensible Framework For Scaling LLM Evaluation with Inter-Model Competition
We introduce ZeroSumEval, a dynamic, competition-based, and evolving evaluation framework for Large Language Models (LLMs) that leverages competitive games. ZeroSumEval encompasses a diverse suite of games, including security challenges (Capture the Flag), classic board games (chess), and knowledge tests (MathQuiz). These games are designed to evaluate a range of capabilities such as strategic reasoning, planning, knowledge application, safety, and adaptability. Building upon recent studies that highlight the effectiveness of game-based evaluations for LLMs, ZeroSumEval enhances these approaches by providing a standardized and extensible framework for easily implementing games and leverages DSPy to provide a better abstraction for LLM player strategies.
♻ ☆ Fleet of Agents: Coordinated Problem Solving with Large Language Models
While numerous frameworks have been developed to enhance the reasoning abilities of large language models (LLMs), there is a scarcity of methods that effectively balance the trade-off between cost and quality. In this paper, we introduce Fleet of Agents (FoA), a novel and intuitive yet principled framework utilizing LLMs as agents to navigate through dynamic tree searches, employing a genetic-type particle filtering approach. FoA spawns a multitude of agents, each exploring the search space autonomously, followed by a selection phase where resampling based on a heuristic value function optimizes the balance between exploration and exploitation. This mechanism enables dynamic branching, adapting the exploration strategy based on discovered solutions. We conduct extensive experiments on three benchmark tasks, ``Game of 24'', ``Mini-Crosswords'', and ``WebShop'', utilizing four different LLMs, ``GPT-3.5'', ``GPT-4'', ``LLaMA3.2-11B'', and ``LLaMA3.2-90B''. On average across all tasks and LLMs, FoA obtains a quality improvement of ~5% while requiring only ~40% of the cost of previous SOTA methods. Notably, our analyses reveal that (1) FoA achieves the best cost-quality trade-off among all benchmarked methods and (2) FoA + LLaMA3.2-11B surpasses the Llama3.2-90B model. FoA is publicly available at https://github.com/au-clan/FoA.
comment: 28 pages, 68 figures, 8 tables
♻ ☆ PR-Attack: Coordinated Prompt-RAG Attacks on Retrieval-Augmented Generation in Large Language Models via Bilevel Optimization SIGIR 2025
Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of applications, e.g., medical question-answering, mathematical sciences, and code generation. However, they also exhibit inherent limitations, such as outdated knowledge and susceptibility to hallucinations. Retrieval-Augmented Generation (RAG) has emerged as a promising paradigm to address these issues, but it also introduces new vulnerabilities. Recent efforts have focused on the security of RAG-based LLMs, yet existing attack methods face three critical challenges: (1) their effectiveness declines sharply when only a limited number of poisoned texts can be injected into the knowledge database, (2) they lack sufficient stealth, as the attacks are often detectable by anomaly detection systems, which compromises their effectiveness, and (3) they rely on heuristic approaches to generate poisoned texts, lacking formal optimization frameworks and theoretic guarantees, which limits their effectiveness and applicability. To address these issues, we propose coordinated Prompt-RAG attack (PR-attack), a novel optimization-driven attack that introduces a small number of poisoned texts into the knowledge database while embedding a backdoor trigger within the prompt. When activated, the trigger causes the LLM to generate pre-designed responses to targeted queries, while maintaining normal behavior in other contexts. This ensures both high effectiveness and stealth. We formulate the attack generation process as a bilevel optimization problem leveraging a principled optimization framework to develop optimal poisoned texts and triggers. Extensive experiments across diverse LLMs and datasets demonstrate the effectiveness of PR-Attack, achieving a high attack success rate even with a limited number of poisoned texts and significantly improved stealth compared to existing methods.
comment: Accepted at SIGIR 2025
♻ ☆ EmoSphere++: Emotion-Controllable Zero-Shot Text-to-Speech via Emotion-Adaptive Spherical Vector
Emotional text-to-speech (TTS) technology has achieved significant progress in recent years; however, challenges remain owing to the inherent complexity of emotions and limitations of the available emotional speech datasets and models. Previous studies typically relied on limited emotional speech datasets or required extensive manual annotations, restricting their ability to generalize across different speakers and emotional styles. In this paper, we present EmoSphere++, an emotion-controllable zero-shot TTS model that can control emotional style and intensity to resemble natural human speech. We introduce a novel emotion-adaptive spherical vector that models emotional style and intensity without human annotation. Moreover, we propose a multi-level style encoder that can ensure effective generalization for both seen and unseen speakers. We also introduce additional loss functions to enhance the emotion transfer performance for zero-shot scenarios. We employ a conditional flow matching-based decoder to achieve high-quality and expressive emotional TTS in a few sampling steps. Experimental results demonstrate the effectiveness of the proposed framework.
♻ ☆ CoDiff: Conditional Diffusion Model for Collaborative 3D Object Detection
Collaborative 3D object detection holds significant importance in the field of autonomous driving, as it greatly enhances the perception capabilities of each individual agent by facilitating information exchange among multiple agents. However, in practice, due to pose estimation errors and time delays, the fusion of information across agents often results in feature representations with spatial and temporal noise, leading to detection errors. Diffusion models naturally have the ability to denoise noisy samples to the ideal data, which motivates us to explore the use of diffusion models to address the noise problem between multi-agent systems. In this work, we propose CoDiff, a novel robust collaborative perception framework that leverages the potential of diffusion models to generate more comprehensive and clearer feature representations. To the best of our knowledge, this is the first work to apply diffusion models to multi-agent collaborative perception. Specifically, we project high-dimensional feature map into the latent space of a powerful pre-trained autoencoder. Within this space, individual agent information serves as a condition to guide the diffusion model's sampling. This process denoises coarse feature maps and progressively refines the fused features. Experimental study on both simulated and real-world datasets demonstrates that the proposed framework CoDiff consistently outperforms existing relevant methods in terms of the collaborative object detection performance, and exhibits highly desired robustness when the pose and delay information of agents is with high-level noise. The code is released at https://github.com/HuangZhe885/CoDiff
♻ ☆ HDC: Hierarchical Distillation for Multi-level Noisy Consistency in Semi-Supervised Fetal Ultrasound Segmentation
Transvaginal ultrasound is a critical imaging modality for evaluating cervical anatomy and detecting physiological changes. However, accurate segmentation of cervical structures remains challenging due to low contrast, shadow artifacts, and indistinct boundaries. While convolutional neural networks (CNNs) have demonstrated efficacy in medical image segmentation, their reliance on large-scale annotated datasets presents a significant limitation in clinical ultrasound imaging. Semi-supervised learning (SSL) offers a potential solution by utilizing unlabeled data, yet existing teacher-student frameworks often encounter confirmation bias and high computational costs. In this paper, a novel semi-supervised segmentation framework, called HDC, is proposed incorporating adaptive consistency learning with a single-teacher architecture. The framework introduces a hierarchical distillation mechanism with two objectives: Correlation Guidance Loss for aligning feature representations and Mutual Information Loss for stabilizing noisy student learning. The proposed approach reduces model complexity while enhancing generalization. Experiments on fetal ultrasound datasets, FUGC and PSFH, demonstrate competitive performance with reduced computational overhead compared to multi-teacher models.
♻ ☆ Exploring the Trade-Offs: Quantization Methods, Task Difficulty, and Model Size in Large Language Models From Edge to Giant
Quantization has gained attention as a promising solution for the cost-effective deployment of large and small language models. However, most prior work has been limited to perplexity or basic knowledge tasks and lacks a comprehensive evaluation of recent models like Llama-3.3. In this paper, we conduct a comprehensive evaluation of instruction-tuned models spanning 1B to 405B parameters, applying four quantization methods across 13 datasets. Our findings reveal that (1) quantized models generally surpass smaller FP16 baselines, yet they often struggle with instruction-following and hallucination detection; (2) FP8 consistently emerges as the most robust option across tasks, and AWQ tends to outperform GPTQ in weight-only quantization; (3) smaller models can suffer severe accuracy drops at 4-bit quantization, while 70B-scale models maintain stable performance; (4) notably, \textit{hard} tasks do not always experience the largest accuracy losses, indicating that quantization magnifies a model's inherent weaknesses rather than simply correlating with task difficulty; and (5) an LLM-based judge (MT-Bench) highlights significant performance declines in coding and STEM tasks, though reasoning may sometimes improve.
comment: 21 pages, 2 figure
♻ ☆ DeepSeek-Inspired Exploration of RL-based LLMs and Synergy with Wireless Networks: A Survey
Reinforcement learning (RL)-based large language models (LLMs), such as ChatGPT, DeepSeek, and Grok-3, have gained significant attention for their exceptional capabilities in natural language processing and multimodal data understanding. Meanwhile, the rapid expansion of information services has driven the growing need for intelligence, efficient, and adaptable wireless networks. Wireless networks require the empowerment of RL-based LLMs while these models also benefit from wireless networks to broaden their application scenarios. Specifically, RL-based LLMs can enhance wireless communication systems through intelligent resource allocation, adaptive network optimization, and real-time decision-making. Conversely, wireless networks provide a vital infrastructure for the efficient training, deployment, and distributed inference of RL-based LLMs, especially in decentralized and edge computing environments. This mutual empowerment highlights the need for a deeper exploration of the interplay between these two domains. We first review recent advancements in wireless communications, highlighting the associated challenges and potential solutions. We then discuss the progress of RL-based LLMs, focusing on key technologies for LLM training, challenges, and potential solutions. Subsequently, we explore the mutual empowerment between these two fields, highlighting key motivations, open challenges, and potential solutions. Finally, we provide insights into future directions, applications, and their societal impact to further explore this intersection, paving the way for next-generation intelligent communication systems. Overall, this survey provides a comprehensive overview of the relationship between RL-based LLMs and wireless networks, offering a vision where these domains empower each other to drive innovations.
comment: 45 pages, 12 figures
♻ ☆ Deep Reinforcement Learning Algorithms for Option Hedging
Dynamic hedging is a financial strategy that consists in periodically transacting one or multiple financial assets to offset the risk associated with a correlated liability. Deep Reinforcement Learning (DRL) algorithms have been used to find optimal solutions to dynamic hedging problems by framing them as sequential decision-making problems. However, most previous work assesses the performance of only one or two DRL algorithms, making an objective comparison across algorithms difficult. In this paper, we compare the performance of eight DRL algorithms in the context of dynamic hedging; Monte Carlo Policy Gradient (MCPG), Proximal Policy Optimization (PPO), along with four variants of Deep Q-Learning (DQL) and two variants of Deep Deterministic Policy Gradient (DDPG). Two of these variants represent a novel application to the task of dynamic hedging. In our experiments, we use the Black-Scholes delta hedge as a baseline and simulate the dataset using a GJR-GARCH(1,1) model. Results show that MCPG, followed by PPO, obtain the best performance in terms of the root semi-quadratic penalty. Moreover, MCPG is the only algorithm to outperform the Black-Scholes delta hedge baseline with the allotted computational budget, possibly due to the sparsity of rewards in our environment.
♻ ☆ AI-Assisted Transport of Radioactive Ion Beams
Beams of radioactive heavy ions allow researchers to study rare and unstable atomic nuclei, shedding light into the internal structure of exotic nuclei and on how chemical elements are formed in stars. However, the extraction and transport of radioactive beams rely on time-consuming expert-driven tuning methods, where hundreds of parameters are manually optimized. Here, we introduce a system that employs Artificial Intelligence (AI), specifically utilizing Bayesian Optimization, to assist in the transport process of radioactive beams. We apply our methodology to real-life scenarios showing advantages when compared with standard tuning methods. This AI-assisted approach can be extended to other radioactive beam facilities around the world to improve operational efficiency and enhance scientific output.
comment: 6 pages, 6 figures; Section headings added for clarity. Implementation and Results sections expanded. Minor revisions to Abstract and to Summary and Conclusion
♻ ☆ The Disagreement Problem in Explainable Machine Learning: A Practitioner's Perspective
As various post hoc explanation methods are increasingly being leveraged to explain complex models in high-stakes settings, it becomes critical to develop a deeper understanding of whether and when the explanations output by these methods disagree with each other, and how such disagreements are resolved in practice. However, there is little to no research that provides answers to these critical questions. In this work, we formalize and study the disagreement problem in explainable machine learning. More specifically, we define the notion of disagreement between explanations, analyze how often such disagreements occur in practice, and how practitioners resolve these disagreements. We first conduct interviews with data scientists to understand what constitutes disagreement between explanations generated by different methods for the same model prediction, and introduce a novel quantitative framework to formalize this understanding. We then leverage this framework to carry out a rigorous empirical analysis with four real-world datasets, six state-of-the-art post hoc explanation methods, and six different predictive models, to measure the extent of disagreement between the explanations generated by various popular explanation methods. In addition, we carry out an online user study with data scientists to understand how they resolve the aforementioned disagreements. Our results indicate that (1) state-of-the-art explanation methods often disagree in terms of the explanations they output, and (2) machine learning practitioners often employ ad hoc heuristics when resolving such disagreements. These findings suggest that practitioners may be relying on misleading explanations when making consequential decisions. They also underscore the importance of developing principled frameworks for effectively evaluating and comparing explanations output by various explanation techniques.
comment: Published in Transactions on Machine Learning Research (TMLR). Updated Related Work and Acknowledgment Section
Computer Vision and Pattern Recognition 150
☆ Perception Encoder: The best visual embeddings are not at the output of the network
We introduce Perception Encoder (PE), a state-of-the-art encoder for image and video understanding trained via simple vision-language learning. Traditionally, vision encoders have relied on a variety of pretraining objectives, each tailored to specific downstream tasks such as classification, captioning, or localization. Surprisingly, after scaling our carefully tuned image pretraining recipe and refining with our robust video data engine, we find that contrastive vision-language training alone can produce strong, general embeddings for all of these downstream tasks. There is only one caveat: these embeddings are hidden within the intermediate layers of the network. To draw them out, we introduce two alignment methods, language alignment for multimodal language modeling, and spatial alignment for dense prediction. Together with the core contrastive checkpoint, our PE family of models achieves state-of-the-art performance on a wide variety of tasks, including zero-shot image and video classification and retrieval; document, image, and video Q&A; and spatial tasks such as detection, depth estimation, and tracking. To foster further research, we are releasing our models, code, and a novel dataset of synthetically and human-annotated videos.
comment: Initial Submission
☆ ViTa-Zero: Zero-shot Visuotactile Object 6D Pose Estimation ICRA 2025
Object 6D pose estimation is a critical challenge in robotics, particularly for manipulation tasks. While prior research combining visual and tactile (visuotactile) information has shown promise, these approaches often struggle with generalization due to the limited availability of visuotactile data. In this paper, we introduce ViTa-Zero, a zero-shot visuotactile pose estimation framework. Our key innovation lies in leveraging a visual model as its backbone and performing feasibility checking and test-time optimization based on physical constraints derived from tactile and proprioceptive observations. Specifically, we model the gripper-object interaction as a spring-mass system, where tactile sensors induce attractive forces, and proprioception generates repulsive forces. We validate our framework through experiments on a real-world robot setup, demonstrating its effectiveness across representative visual backbones and manipulation scenarios, including grasping, object picking, and bimanual handover. Compared to the visual models, our approach overcomes some drastic failure modes while tracking the in-hand object pose. In our experiments, our approach shows an average increase of 55% in AUC of ADD-S and 60% in ADD, along with an 80% lower position error compared to FoundationPose.
comment: Accepted by ICRA 2025
☆ PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding
Vision-language models are integral to computer vision research, yet many high-performing models remain closed-source, obscuring their data, design and training recipe. The research community has responded by using distillation from black-box models to label training data, achieving strong benchmark results, at the cost of measurable scientific progress. However, without knowing the details of the teacher model and its data sources, scientific progress remains difficult to measure. In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training pipelines without distillation from proprietary models and explore large-scale synthetic data to identify critical data gaps, particularly in detailed video understanding. To bridge these gaps, we release 2.8M human-labeled instances of fine-grained video question-answer pairs and spatio-temporally grounded video captions. Additionally, we introduce PLM-VideoBench, a suite for evaluating challenging video understanding tasks focusing on the ability to reason about "what", "where", "when", and "how" of a video. We make our work fully reproducible by providing data, training recipes, code & models.
comment: Technical report
☆ Single-Shot Shape and Reflectance with Spatial Polarization Multiplexing
We propose spatial polarization multiplexing (SPM) for reconstructing object shape and reflectance from a single polarimetric image and demonstrate its application to dynamic surface recovery. Although single-pattern structured light enables single-shot shape reconstruction, the reflectance is challenging to recover due to the lack of angular sampling of incident light and the entanglement of the projected pattern and the surface color texture. We design a spatially multiplexed pattern of polarization that can be robustly and uniquely decoded for shape reconstruction by quantizing the AoLP values. At the same time, our spatial-multiplexing enables single-shot ellipsometry of linear polarization by projecting differently polarized light within a local region, which separates the specular and diffuse reflections for BRDF estimation. We achieve this spatial polarization multiplexing with a constrained de Bruijn sequence. Unlike single-pattern structured light with intensity and color, our polarization pattern is invisible to the naked eye and retains the natural surface appearance which is essential for accurate appearance modeling and also interaction with people. We experimentally validate our method on real data. The results show that our method can recover the shape, the Mueller matrix, and the BRDF from a single-shot polarimetric image. We also demonstrate the application of our method to dynamic surfaces.
☆ IMAGGarment-1: Fine-Grained Garment Generation for Controllable Fashion Design
This paper presents IMAGGarment-1, a fine-grained garment generation (FGG) framework that enables high-fidelity garment synthesis with precise control over silhouette, color, and logo placement. Unlike existing methods that are limited to single-condition inputs, IMAGGarment-1 addresses the challenges of multi-conditional controllability in personalized fashion design and digital apparel applications. Specifically, IMAGGarment-1 employs a two-stage training strategy to separately model global appearance and local details, while enabling unified and controllable generation through end-to-end inference. In the first stage, we propose a global appearance model that jointly encodes silhouette and color using a mixed attention module and a color adapter. In the second stage, we present a local enhancement model with an adaptive appearance-aware module to inject user-defined logos and spatial constraints, enabling accurate placement and visual consistency. To support this task, we release GarmentBench, a large-scale dataset comprising over 180K garment samples paired with multi-level design conditions, including sketches, color references, logo placements, and textual prompts. Extensive experiments demonstrate that our method outperforms existing baselines, achieving superior structural stability, color fidelity, and local controllability performance. The code and model are available at https://github.com/muzishen/IMAGGarment-1.
☆ Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling
Vision-Language Models (VLMs) excel at visual understanding but often suffer from visual hallucinations, where they generate descriptions of nonexistent objects, actions, or concepts, posing significant risks in safety-critical applications. Existing hallucination mitigation methods typically follow one of two paradigms: generation adjustment, which modifies decoding behavior to align text with visual inputs, and post-hoc verification, where external models assess and correct outputs. While effective, generation adjustment methods often rely on heuristics and lack correction mechanisms, while post-hoc verification is complicated, typically requiring multiple models and tending to reject outputs rather than refine them. In this work, we introduce REVERSE, a unified framework that integrates hallucination-aware training with on-the-fly self-verification. By leveraging a new hallucination-verification dataset containing over 1.3M semi-synthetic samples, along with a novel inference-time retrospective resampling technique, our approach enables VLMs to both detect hallucinations during generation and dynamically revise those hallucinations. Our evaluations show that REVERSE achieves state-of-the-art hallucination reduction, outperforming the best existing methods by up to 12% on CHAIR-MSCOCO and 28% on HaloQuest. Our dataset, model, and code are available at: https://reverse-vlm.github.io.
comment: Preprint. Project Page: https://reverse-vlm.github.io
☆ ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos CVPR 2025
Creating a photorealistic scene and human reconstruction from a single monocular in-the-wild video figures prominently in the perception of a human-centric 3D world. Recent neural rendering advances have enabled holistic human-scene reconstruction but require pre-calibrated camera and human poses, and days of training time. In this work, we introduce a novel unified framework that simultaneously performs camera tracking, human pose estimation and human-scene reconstruction in an online fashion. 3D Gaussian Splatting is utilized to learn Gaussian primitives for humans and scenes efficiently, and reconstruction-based camera tracking and human pose estimation modules are designed to enable holistic understanding and effective disentanglement of pose and appearance. Specifically, we design a human deformation module to reconstruct the details and enhance generalizability to out-of-distribution poses faithfully. Aiming to learn the spatial correlation between human and scene accurately, we introduce occlusion-aware human silhouette rendering and monocular geometric priors, which further improve reconstruction quality. Experiments on the EMDB and NeuMan datasets demonstrate superior or on-par performance with existing methods in camera tracking, human pose estimation, novel view synthesis and runtime. Our project page is at https://eth-ait.github.io/ODHSR.
comment: Accepted at CVPR 2025
☆ Personalized Text-to-Image Generation with Auto-Regressive Models
Personalized image synthesis has emerged as a pivotal application in text-to-image generation, enabling the creation of images featuring specific subjects in diverse contexts. While diffusion models have dominated this domain, auto-regressive models, with their unified architecture for text and image modeling, remain underexplored for personalized image generation. This paper investigates the potential of optimizing auto-regressive models for personalized image synthesis, leveraging their inherent multimodal capabilities to perform this task. We propose a two-stage training strategy that combines optimization of text embeddings and fine-tuning of transformer layers. Our experiments on the auto-regressive model demonstrate that this method achieves comparable subject fidelity and prompt following to the leading diffusion-based personalization methods. The results highlight the effectiveness of auto-regressive models in personalized image generation, offering a new direction for future research in this area.
comment: Project page: https://github.com/KaiyueSun98/T2I-Personalization-with-AR
☆ Digital Twin Generation from Visual Data: A Survey
This survey explores recent developments in generating digital twins from videos. Such digital twins can be used for robotics application, media content creation, or design and construction works. We analyze various approaches, including 3D Gaussian Splatting, generative in-painting, semantic segmentation, and foundation models highlighting their advantages and limitations. Additionally, we discuss challenges such as occlusions, lighting variations, and scalability, as well as potential future research directions. This survey aims to provide a comprehensive overview of state-of-the-art methodologies and their implications for real-world applications. Awesome list: https://github.com/ndrwmlnk/awesome-digital-twins
☆ AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesis CVPR 2025
We explore the task of geometric reconstruction of images captured from a mixture of ground and aerial views. Current state-of-the-art learning-based approaches fail to handle the extreme viewpoint variation between aerial-ground image pairs. Our hypothesis is that the lack of high-quality, co-registered aerial-ground datasets for training is a key reason for this failure. Such data is difficult to assemble precisely because it is difficult to reconstruct in a scalable way. To overcome this challenge, we propose a scalable framework combining pseudo-synthetic renderings from 3D city-wide meshes (e.g., Google Earth) with real, ground-level crowd-sourced images (e.g., MegaDepth). The pseudo-synthetic data simulates a wide range of aerial viewpoints, while the real, crowd-sourced images help improve visual fidelity for ground-level images where mesh-based renderings lack sufficient detail, effectively bridging the domain gap between real images and pseudo-synthetic renderings. Using this hybrid dataset, we fine-tune several state-of-the-art algorithms and achieve significant improvements on real-world, zero-shot aerial-ground tasks. For example, we observe that baseline DUSt3R localizes fewer than 5% of aerial-ground pairs within 5 degrees of camera rotation error, while fine-tuning with our data raises accuracy to nearly 56%, addressing a major failure point in handling large viewpoint changes. Beyond camera estimation and scene reconstruction, our dataset also improves performance on downstream tasks like novel-view synthesis in challenging aerial-ground scenarios, demonstrating the practical value of our approach in real-world applications.
comment: Appearing in CVPR 2025. Project page: https://aerial-megadepth.github.io
☆ Training-Free Hierarchical Scene Understanding for Gaussian Splatting with Superpoint Graphs
Bridging natural language and 3D geometry is a crucial step toward flexible, language-driven scene understanding. While recent advances in 3D Gaussian Splatting (3DGS) have enabled fast and high-quality scene reconstruction, research has also explored incorporating open-vocabulary understanding into 3DGS. However, most existing methods require iterative optimization over per-view 2D semantic feature maps, which not only results in inefficiencies but also leads to inconsistent 3D semantics across views. To address these limitations, we introduce a training-free framework that constructs a superpoint graph directly from Gaussian primitives. The superpoint graph partitions the scene into spatially compact and semantically coherent regions, forming view-consistent 3D entities and providing a structured foundation for open-vocabulary understanding. Based on the graph structure, we design an efficient reprojection strategy that lifts 2D semantic features onto the superpoints, avoiding costly multi-view iterative training. The resulting representation ensures strong 3D semantic coherence and naturally supports hierarchical understanding, enabling both coarse- and fine-grained open-vocabulary perception within a unified semantic field. Extensive experiments demonstrate that our method achieves state-of-the-art open-vocabulary segmentation performance, with semantic field reconstruction completed over $30\times$ faster. Our code will be available at https://github.com/Atrovast/THGS.
☆ St4RTrack: Simultaneous 4D Reconstruction and Tracking in the World
Dynamic 3D reconstruction and point tracking in videos are typically treated as separate tasks, despite their deep connection. We propose St4RTrack, a feed-forward framework that simultaneously reconstructs and tracks dynamic video content in a world coordinate frame from RGB inputs. This is achieved by predicting two appropriately defined pointmaps for a pair of frames captured at different moments. Specifically, we predict both pointmaps at the same moment, in the same world, capturing both static and dynamic scene geometry while maintaining 3D correspondences. Chaining these predictions through the video sequence with respect to a reference frame naturally computes long-range correspondences, effectively combining 3D reconstruction with 3D tracking. Unlike prior methods that rely heavily on 4D ground truth supervision, we employ a novel adaptation scheme based on a reprojection loss. We establish a new extensive benchmark for world-frame reconstruction and tracking, demonstrating the effectiveness and efficiency of our unified, data-driven framework. Our code, model, and benchmark will be released.
comment: Project page: https://St4RTrack.github.io/
☆ Readable Twins of Unreadable Models
Creating responsible artificial intelligence (AI) systems is an important issue in contemporary research and development of works on AI. One of the characteristics of responsible AI systems is their explainability. In the paper, we are interested in explainable deep learning (XDL) systems. On the basis of the creation of digital twins of physical objects, we introduce the idea of creating readable twins (in the form of imprecise information flow models) for unreadable deep learning models. The complete procedure for switching from the deep learning model (DLM) to the imprecise information flow model (IIFM) is presented. The proposed approach is illustrated with an example of a deep learning classification model for image recognition of handwritten digits from the MNIST data set.
comment: Based on the abstract accepted for ISFS 2025
☆ $\texttt{Complex-Edit}$: CoT-Like Instruction Generation for Complexity-Controllable Image Editing Benchmark SC
We introduce $\texttt{Complex-Edit}$, a comprehensive benchmark designed to systematically evaluate instruction-based image editing models across instructions of varying complexity. To develop this benchmark, we harness GPT-4o to automatically collect a diverse set of editing instructions at scale. Our approach follows a well-structured ``Chain-of-Edit'' pipeline: we first generate individual atomic editing tasks independently and then integrate them to form cohesive, complex instructions. Additionally, we introduce a suite of metrics to assess various aspects of editing performance, along with a VLM-based auto-evaluation pipeline that supports large-scale assessments. Our benchmark yields several notable insights: 1) Open-source models significantly underperform relative to proprietary, closed-source models, with the performance gap widening as instruction complexity increases; 2) Increased instructional complexity primarily impairs the models' ability to retain key elements from the input images and to preserve the overall aesthetic quality; 3) Decomposing a complex instruction into a sequence of atomic steps, executed in a step-by-step manner, substantially degrades performance across multiple metrics; 4) A straightforward Best-of-N selection strategy improves results for both direct editing and the step-by-step sequential approach; and 5) We observe a ``curse of synthetic data'': when synthetic data is involved in model training, the edited images from such models tend to appear increasingly synthetic as the complexity of the editing instructions rises -- a phenomenon that intriguingly also manifests in the latest GPT-4o outputs.
comment: Project Page: https://ucsc-vlaa.github.io/Complex-Edit/, Dataset: https://huggingface.co/datasets/UCSC-VLAA/Complex-Edit
☆ PCBEAR: Pose Concept Bottleneck for Explainable Action Recognition CVPR
Human action recognition (HAR) has achieved impressive results with deep learning models, but their decision-making process remains opaque due to their black-box nature. Ensuring interpretability is crucial, especially for real-world applications requiring transparency and accountability. Existing video XAI methods primarily rely on feature attribution or static textual concepts, both of which struggle to capture motion dynamics and temporal dependencies essential for action understanding. To address these challenges, we propose Pose Concept Bottleneck for Explainable Action Recognition (PCBEAR), a novel concept bottleneck framework that introduces human pose sequences as motion-aware, structured concepts for video action recognition. Unlike methods based on pixel-level features or static textual descriptions, PCBEAR leverages human skeleton poses, which focus solely on body movements, providing robust and interpretable explanations of motion dynamics. We define two types of pose-based concepts: static pose concepts for spatial configurations at individual frames, and dynamic pose concepts for motion patterns across multiple frames. To construct these concepts, PCBEAR applies clustering to video pose sequences, allowing for automatic discovery of meaningful concepts without manual annotation. We validate PCBEAR on KTH, Penn-Action, and HAA500, showing that it achieves high classification performance while offering interpretable, motion-driven explanations. Our method provides both strong predictive performance and human-understandable insights into the model's reasoning process, enabling test-time interventions for debugging and improving model behavior.
comment: This paper is accepted by CVPRW 2025
☆ NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement: Methods and Results CVPR
This paper presents a review for the NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement. The challenge comprises two tracks: (i) Efficient Video Quality Assessment (KVQ), and (ii) Diffusion-based Image Super-Resolution (KwaiSR). Track 1 aims to advance the development of lightweight and efficient video quality assessment (VQA) models, with an emphasis on eliminating reliance on model ensembles, redundant weights, and other computationally expensive components in the previous IQA/VQA competitions. Track 2 introduces a new short-form UGC dataset tailored for single image super-resolution, i.e., the KwaiSR dataset. It consists of 1,800 synthetically generated S-UGC image pairs and 1,900 real-world S-UGC images, which are split into training, validation, and test sets using a ratio of 8:1:1. The primary objective of the challenge is to drive research that benefits the user experience of short-form UGC platforms such as Kwai and TikTok. This challenge attracted 266 participants and received 18 valid final submissions with corresponding fact sheets, significantly contributing to the progress of short-form UGC VQA and image superresolution. The project is publicly available at https://github.com/lixinustc/KVQE- ChallengeCVPR-NTIRE2025.
comment: Challenge Report of NTIRE 2025; Methods from 18 Teams; Accepted by CVPR Workshop; 21 pages
☆ Science-T2I: Addressing Scientific Illusions in Image Synthesis CVPR 2025
We present a novel approach to integrating scientific knowledge into generative models, enhancing their realism and consistency in image synthesis. First, we introduce Science-T2I, an expert-annotated adversarial dataset comprising adversarial 20k image pairs with 9k prompts, covering wide distinct scientific knowledge categories. Leveraging Science-T2I, we present SciScore, an end-to-end reward model that refines the assessment of generated images based on scientific knowledge, which is achieved by augmenting both the scientific comprehension and visual capabilities of pre-trained CLIP model. Additionally, based on SciScore, we propose a two-stage training framework, comprising a supervised fine-tuning phase and a masked online fine-tuning phase, to incorporate scientific knowledge into existing generative models. Through comprehensive experiments, we demonstrate the effectiveness of our framework in establishing new standards for evaluating the scientific realism of generated content. Specifically, SciScore attains performance comparable to human-level, demonstrating a 5% improvement similar to evaluations conducted by experienced human evaluators. Furthermore, by applying our proposed fine-tuning method to FLUX, we achieve a performance enhancement exceeding 50% on SciScore.
comment: Accepted to CVPR 2025. Code, docs, weight, benchmark and training data are all avaliable at https://jialuo-li.github.io/Science-T2I-Web
☆ Low-hallucination Synthetic Captions for Large-Scale Vision-Language Model Pre-training
In recent years, the field of vision-language model pre-training has experienced rapid advancements, driven primarily by the continuous enhancement of textual capabilities in large language models. However, existing training paradigms for multimodal large language models heavily rely on high-quality image-text pairs. As models and data scales grow exponentially, the availability of such meticulously curated data has become increasingly scarce and saturated, thereby severely limiting further advancements in this domain. This study investigates scalable caption generation techniques for vision-language model pre-training and demonstrates that large-scale low-hallucination synthetic captions can serve dual purposes: 1) acting as a viable alternative to real-world data for pre-training paradigms and 2) achieving superior performance enhancement when integrated into vision-language models through empirical validation. This paper presents three key contributions: 1) a novel pipeline for generating high-quality, low-hallucination, and knowledge-rich synthetic captions. Our continuous DPO methodology yields remarkable results in reducing hallucinations. Specifically, the non-hallucination caption rate on a held-out test set increases from 48.2% to 77.9% for a 7B-size model. 2) Comprehensive empirical validation reveals that our synthetic captions confer superior pre-training advantages over their counterparts. Across 35 vision language tasks, the model trained with our data achieves a significant performance gain of at least 6.2% compared to alt-text pairs and other previous work. Meanwhile, it also offers considerable support in the text-to-image domain. With our dataset, the FID score is reduced by 17.1 on a real-world validation benchmark and 13.3 on the MSCOCO validation benchmark. 3) We will release Hunyuan-Recap100M, a low-hallucination and knowledge-intensive synthetic caption dataset.
☆ VistaDPO: Video Hierarchical Spatial-Temporal Direct Preference Optimization for Large Video Models
Large Video Models (LVMs) built upon Large Language Models (LLMs) have shown promise in video understanding but often suffer from misalignment with human intuition and video hallucination issues. To address these challenges, we introduce VistaDPO, a novel framework for Video Hierarchical Spatial-Temporal Direct Preference Optimization. VistaDPO enhances text-video preference alignment across three hierarchical levels: i) Instance Level, aligning overall video content with responses; ii) Temporal Level, aligning video temporal semantics with event descriptions; and iii) Perceptive Level, aligning spatial objects with language tokens. Given the lack of datasets for fine-grained video-language preference alignment, we construct VistaDPO-7k, a dataset of 7.2K QA pairs annotated with chosen and rejected responses, along with spatial-temporal grounding information such as timestamps, keyframes, and bounding boxes. Extensive experiments on benchmarks such as Video Hallucination, Video QA, and Captioning performance tasks demonstrate that VistaDPO significantly improves the performance of existing LVMs, effectively mitigating video-language misalignment and hallucination. The code and data are available at https://github.com/HaroldChen19/VistaDPO.
comment: Code and Data: https://github.com/HaroldChen19/VistaDPO
☆ Probing and Inducing Combinational Creativity in Vision-Language Models
The ability to combine existing concepts into novel ideas stands as a fundamental hallmark of human intelligence. Recent advances in Vision-Language Models (VLMs) like GPT-4V and DALLE-3 have sparked debate about whether their outputs reflect combinational creativity--defined by M. A. Boden (1998) as synthesizing novel ideas through combining existing concepts--or sophisticated pattern matching of training data. Drawing inspiration from cognitive science, we investigate the combinational creativity of VLMs from the lens of concept blending. We propose the Identification-Explanation-Implication (IEI) framework, which decomposes creative processes into three levels: identifying input spaces, extracting shared attributes, and deriving novel semantic implications. To validate this framework, we curate CreativeMashup, a high-quality dataset of 666 artist-generated visual mashups annotated according to the IEI framework. Through extensive experiments, we demonstrate that in comprehension tasks, best VLMs have surpassed average human performance while falling short of expert-level understanding; in generation tasks, incorporating our IEI framework into the generation pipeline significantly enhances the creative quality of VLMs outputs. Our findings establish both a theoretical foundation for evaluating artificial creativity and practical guidelines for improving creative generation in VLMs.
comment: Project page: https://ppyyqq.github.io/aicc/ The first two authors contribute equally
☆ UniEdit-Flow: Unleashing Inversion and Editing in the Era of Flow Models
Flow matching models have emerged as a strong alternative to diffusion models, but existing inversion and editing methods designed for diffusion are often ineffective or inapplicable to them. The straight-line, non-crossing trajectories of flow models pose challenges for diffusion-based approaches but also open avenues for novel solutions. In this paper, we introduce a predictor-corrector-based framework for inversion and editing in flow models. First, we propose Uni-Inv, an effective inversion method designed for accurate reconstruction. Building on this, we extend the concept of delayed injection to flow models and introduce Uni-Edit, a region-aware, robust image editing approach. Our methodology is tuning-free, model-agnostic, efficient, and effective, enabling diverse edits while ensuring strong preservation of edit-irrelevant regions. Extensive experiments across various generative models demonstrate the superiority and generalizability of Uni-Inv and Uni-Edit, even under low-cost settings. Project page: https://uniedit-flow.github.io/
comment: Project page: https://uniedit-flow.github.io/
☆ RF-DETR Object Detection vs YOLOv12 : A Study of Transformer-based and CNN-based Architectures for Single-Class and Multi-Class Greenfruit Detection in Complex Orchard Environments Under Label Ambiguity
This study conducts a detailed comparison of RF-DETR object detection base model and YOLOv12 object detection model configurations for detecting greenfruits in a complex orchard environment marked by label ambiguity, occlusions, and background blending. A custom dataset was developed featuring both single-class (greenfruit) and multi-class (occluded and non-occluded greenfruits) annotations to assess model performance under dynamic real-world conditions. RF-DETR object detection model, utilizing a DINOv2 backbone and deformable attention, excelled in global context modeling, effectively identifying partially occluded or ambiguous greenfruits. In contrast, YOLOv12 leveraged CNN-based attention for enhanced local feature extraction, optimizing it for computational efficiency and edge deployment. RF-DETR achieved the highest mean Average Precision (mAP50) of 0.9464 in single-class detection, proving its superior ability to localize greenfruits in cluttered scenes. Although YOLOv12N recorded the highest mAP@50:95 of 0.7620, RF-DETR consistently outperformed in complex spatial scenarios. For multi-class detection, RF-DETR led with an mAP@50 of 0.8298, showing its capability to differentiate between occluded and non-occluded fruits, while YOLOv12L scored highest in mAP@50:95 with 0.6622, indicating better classification in detailed occlusion contexts. Training dynamics analysis highlighted RF-DETR's swift convergence, particularly in single-class settings where it plateaued within 10 epochs, demonstrating the efficiency of transformer-based architectures in adapting to dynamic visual data. These findings validate RF-DETR's effectiveness for precision agricultural applications, with YOLOv12 suited for fast-response scenarios. >Index Terms: RF-DETR object detection, YOLOv12, YOLOv13, YOLOv14, YOLOv15, YOLOE, YOLO World, YOLO, You Only Look Once, Roboflow, Detection Transformers, CNNs
☆ EventVAD: Training-Free Event-Aware Video Anomaly Detection
Video Anomaly Detection~(VAD) focuses on identifying anomalies within videos. Supervised methods require an amount of in-domain training data and often struggle to generalize to unseen anomalies. In contrast, training-free methods leverage the intrinsic world knowledge of large language models (LLMs) to detect anomalies but face challenges in localizing fine-grained visual transitions and diverse events. Therefore, we propose EventVAD, an event-aware video anomaly detection framework that combines tailored dynamic graph architectures and multimodal LLMs through temporal-event reasoning. Specifically, EventVAD first employs dynamic spatiotemporal graph modeling with time-decay constraints to capture event-aware video features. Then, it performs adaptive noise filtering and uses signal ratio thresholding to detect event boundaries via unsupervised statistical features. The statistical boundary detection module reduces the complexity of processing long videos for MLLMs and improves their temporal reasoning through event consistency. Finally, it utilizes a hierarchical prompting strategy to guide MLLMs in performing reasoning before determining final decisions. We conducted extensive experiments on the UCF-Crime and XD-Violence datasets. The results demonstrate that EventVAD with a 7B MLLM achieves state-of-the-art (SOTA) in training-free settings, outperforming strong baselines that use 7B or larger MLLMs.
☆ Enhancing Person-to-Person Virtual Try-On with Multi-Garment Virtual Try-Off
Computer vision is transforming fashion through Virtual Try-On (VTON) and Virtual Try-Off (VTOFF). VTON generates images of a person in a specified garment using a target photo and a standardized garment image, while a more challenging variant, Person-to-Person Virtual Try-On (p2p-VTON), uses a photo of another person wearing the garment. VTOFF, on the other hand, extracts standardized garment images from clothed individuals. We introduce TryOffDiff, a diffusion-based VTOFF model. Built on a latent diffusion framework with SigLIP image conditioning, it effectively captures garment properties like texture, shape, and patterns. TryOffDiff achieves state-of-the-art results on VITON-HD and strong performance on DressCode dataset, covering upper-body, lower-body, and dresses. Enhanced with class-specific embeddings, it pioneers multi-garment VTOFF, the first of its kind. When paired with VTON models, it improves p2p-VTON by minimizing unwanted attribute transfer, such as skin color. Code is available at: https://rizavelioglu.github.io/tryoffdiff/
☆ Effective Dual-Region Augmentation for Reduced Reliance on Large Amounts of Labeled Data SP
This paper introduces a novel dual-region augmentation approach designed to reduce reliance on large-scale labeled datasets while improving model robustness and adaptability across diverse computer vision tasks, including source-free domain adaptation (SFDA) and person re-identification (ReID). Our method performs targeted data transformations by applying random noise perturbations to foreground objects and spatially shuffling background patches. This effectively increases the diversity of the training data, improving model robustness and generalization. Evaluations on the PACS dataset for SFDA demonstrate that our augmentation strategy consistently outperforms existing methods, achieving significant accuracy improvements in both single-target and multi-target adaptation settings. By augmenting training data through structured transformations, our method enables model generalization across domains, providing a scalable solution for reducing reliance on manually annotated datasets. Furthermore, experiments on Market-1501 and DukeMTMC-reID datasets validate the effectiveness of our approach for person ReID, surpassing traditional augmentation techniques.
comment: 9 pages, 2 figures, Accepted to SPIE DSC 2025 Conference: Synthetic Data for Artificial Intelligence and Machine Learning: Tools, Techniques, and Applications III
☆ SkyReels-V2: Infinite-length Film Generative Model
Recent advances in video generation have been driven by diffusion models and autoregressive frameworks, yet critical challenges persist in harmonizing prompt adherence, visual quality, motion dynamics, and duration: compromises in motion dynamics to enhance temporal visual quality, constrained video duration (5-10 seconds) to prioritize resolution, and inadequate shot-aware generation stemming from general-purpose MLLMs' inability to interpret cinematic grammar, such as shot composition, actor expressions, and camera motions. These intertwined limitations hinder realistic long-form synthesis and professional film-style generation. To address these limitations, we propose SkyReels-V2, an Infinite-length Film Generative Model, that synergizes Multi-modal Large Language Model (MLLM), Multi-stage Pretraining, Reinforcement Learning, and Diffusion Forcing Framework. Firstly, we design a comprehensive structural representation of video that combines the general descriptions by the Multi-modal LLM and the detailed shot language by sub-expert models. Aided with human annotation, we then train a unified Video Captioner, named SkyCaptioner-V1, to efficiently label the video data. Secondly, we establish progressive-resolution pretraining for the fundamental video generation, followed by a four-stage post-training enhancement: Initial concept-balanced Supervised Fine-Tuning (SFT) improves baseline quality; Motion-specific Reinforcement Learning (RL) training with human-annotated and synthetic distortion data addresses dynamic artifacts; Our diffusion forcing framework with non-decreasing noise schedules enables long-video synthesis in an efficient search space; Final high-quality SFT refines visual fidelity. All the code and models are available at https://github.com/SkyworkAI/SkyReels-V2.
comment: 31 pages,10 figures
☆ HiScene: Creating Hierarchical 3D Scenes with Isometric View Generation
Scene-level 3D generation represents a critical frontier in multimedia and computer graphics, yet existing approaches either suffer from limited object categories or lack editing flexibility for interactive applications. In this paper, we present HiScene, a novel hierarchical framework that bridges the gap between 2D image generation and 3D object generation and delivers high-fidelity scenes with compositional identities and aesthetic scene content. Our key insight is treating scenes as hierarchical "objects" under isometric views, where a room functions as a complex object that can be further decomposed into manipulatable items. This hierarchical approach enables us to generate 3D content that aligns with 2D representations while maintaining compositional structure. To ensure completeness and spatial alignment of each decomposed instance, we develop a video-diffusion-based amodal completion technique that effectively handles occlusions and shadows between objects, and introduce shape prior injection to ensure spatial coherence within the scene. Experimental results demonstrate that our method produces more natural object arrangements and complete object instances suitable for interactive applications, while maintaining physical plausibility and alignment with user inputs.
comment: Project webpage: https://zju3dv.github.io/hiscene/
☆ EchoWorld: Learning Motion-Aware World Models for Echocardiography Probe Guidance CVPR 2025
Echocardiography is crucial for cardiovascular disease detection but relies heavily on experienced sonographers. Echocardiography probe guidance systems, which provide real-time movement instructions for acquiring standard plane images, offer a promising solution for AI-assisted or fully autonomous scanning. However, developing effective machine learning models for this task remains challenging, as they must grasp heart anatomy and the intricate interplay between probe motion and visual signals. To address this, we present EchoWorld, a motion-aware world modeling framework for probe guidance that encodes anatomical knowledge and motion-induced visual dynamics, while effectively leveraging past visual-motion sequences to enhance guidance precision. EchoWorld employs a pre-training strategy inspired by world modeling principles, where the model predicts masked anatomical regions and simulates the visual outcomes of probe adjustments. Built upon this pre-trained model, we introduce a motion-aware attention mechanism in the fine-tuning stage that effectively integrates historical visual-motion data, enabling precise and adaptive probe guidance. Trained on more than one million ultrasound images from over 200 routine scans, EchoWorld effectively captures key echocardiographic knowledge, as validated by qualitative analysis. Moreover, our method significantly reduces guidance errors compared to existing visual backbones and guidance frameworks, excelling in both single-frame and sequential evaluation protocols. Code is available at https://github.com/LeapLabTHU/EchoWorld.
comment: Accepted by CVPR 2025
☆ ArtistAuditor: Auditing Artist Style Pirate in Text-to-Image Generation Models
Text-to-image models based on diffusion processes, such as DALL-E, Stable Diffusion, and Midjourney, are capable of transforming texts into detailed images and have widespread applications in art and design. As such, amateur users can easily imitate professional-level paintings by collecting an artist's work and fine-tuning the model, leading to concerns about artworks' copyright infringement. To tackle these issues, previous studies either add visually imperceptible perturbation to the artwork to change its underlying styles (perturbation-based methods) or embed post-training detectable watermarks in the artwork (watermark-based methods). However, when the artwork or the model has been published online, i.e., modification to the original artwork or model retraining is not feasible, these strategies might not be viable. To this end, we propose a novel method for data-use auditing in the text-to-image generation model. The general idea of ArtistAuditor is to identify if a suspicious model has been finetuned using the artworks of specific artists by analyzing the features related to the style. Concretely, ArtistAuditor employs a style extractor to obtain the multi-granularity style representations and treats artworks as samplings of an artist's style. Then, ArtistAuditor queries a trained discriminator to gain the auditing decisions. The experimental results on six combinations of models and datasets show that ArtistAuditor can achieve high AUC values (> 0.937). By studying ArtistAuditor's transferability and core modules, we provide valuable insights into the practical implementation. Finally, we demonstrate the effectiveness of ArtistAuditor in real-world cases by an online platform Scenario. ArtistAuditor is open-sourced at https://github.com/Jozenn/ArtistAuditor.
comment: To appear in the ACM Web Conference 2025, Sydney, Australia
☆ Imaging for All-Day Wearable Smart Glasses
In recent years smart glasses technology has rapidly advanced, opening up entirely new areas for mobile computing. We expect future smart glasses will need to be all-day wearable, adopting a small form factor to meet the requirements of volume, weight, fashionability and social acceptability, which puts significant constraints on the space of possible solutions. Additional challenges arise due to the fact that smart glasses are worn in arbitrary environments while their wearer moves and performs everyday activities. In this paper, we systematically analyze the space of imaging from smart glasses and derive several fundamental limits that govern this imaging domain. We discuss the impact of these limits on achievable image quality and camera module size -- comparing in particular to related devices such as mobile phones. We then propose a novel distributed imaging approach that allows to minimize the size of the individual camera modules when compared to a standard monolithic camera design. Finally, we demonstrate the properties of this novel approach in a series of experiments using synthetic data as well as images captured with two different prototype implementations.
☆ NoisyRollout: Reinforcing Visual Reasoning with Data Augmentation
Recent advances in reinforcement learning (RL) have strengthened the reasoning capabilities of vision-language models (VLMs). However, enhancing policy exploration to more effectively scale test-time compute remains underexplored in VLMs. In addition, VLMs continue to struggle with imperfect visual perception, which in turn affects the subsequent reasoning process. To this end, we propose NoisyRollout, a simple yet effective RL approach that mixes trajectories from both clean and moderately distorted images to introduce targeted diversity in visual perception and the resulting reasoning patterns. Without additional training cost, NoisyRollout enhances the exploration capabilities of VLMs by incorporating a vision-oriented inductive bias. Furthermore, NoisyRollout employs a noise annealing schedule that gradually reduces distortion strength over training, ensuring benefit from noisy signals early while maintaining training stability and scalability in later stages. With just 2.1K training samples, NoisyRollout achieves state-of-the-art performance among open-source RL-tuned models on 5 out-of-domain benchmarks spanning both reasoning and perception tasks, while preserving comparable or even better in-domain performance.
comment: Technical Report
☆ Expert Kernel Generation Network Driven by Contextual Mapping for Hyperspectral Image Classification
Deep neural networks face several challenges in hyperspectral image classification, including high-dimensional data, sparse distribution of ground objects, and spectral redundancy, which often lead to classification overfitting and limited generalization capability. To more efficiently adapt to ground object distributions while extracting image features without introducing excessive parameters and skipping redundant information, this paper proposes EKGNet based on an improved 3D-DenseNet model, consisting of a context-aware mapping network and a dynamic kernel generation module. The context-aware mapping module translates global contextual information of hyperspectral inputs into instructions for combining base convolutional kernels, while the dynamic kernels are composed of K groups of base convolutions, analogous to K different types of experts specializing in fundamental patterns across various dimensions. The mapping module and dynamic kernel generation mechanism form a tightly coupled system - the former generates meaningful combination weights based on inputs, while the latter constructs an adaptive expert convolution system using these weights. This dynamic approach enables the model to focus more flexibly on key spatial structures when processing different regions, rather than relying on the fixed receptive field of a single static convolutional kernel. EKGNet enhances model representation capability through a 3D dynamic expert convolution system without increasing network depth or width. The proposed method demonstrates superior performance on IN, UP, and KSC datasets, outperforming mainstream hyperspectral image classification approaches.
comment: arXiv admin note: substantial text overlap with arXiv:2503.23472
☆ Event-Enhanced Blurry Video Super-Resolution AAAI 2025
In this paper, we tackle the task of blurry video super-resolution (BVSR), aiming to generate high-resolution (HR) videos from low-resolution (LR) and blurry inputs. Current BVSR methods often fail to restore sharp details at high resolutions, resulting in noticeable artifacts and jitter due to insufficient motion information for deconvolution and the lack of high-frequency details in LR frames. To address these challenges, we introduce event signals into BVSR and propose a novel event-enhanced network, Ev-DeblurVSR. To effectively fuse information from frames and events for feature deblurring, we introduce a reciprocal feature deblurring module that leverages motion information from intra-frame events to deblur frame features while reciprocally using global scene context from the frames to enhance event features. Furthermore, to enhance temporal consistency, we propose a hybrid deformable alignment module that fully exploits the complementary motion information from inter-frame events and optical flow to improve motion estimation in the deformable alignment process. Extensive evaluations demonstrate that Ev-DeblurVSR establishes a new state-of-the-art performance on both synthetic and real-world datasets. Notably, on real data, our method is +2.59 dB more accurate and 7.28$\times$ faster than the recent best BVSR baseline FMA-Net. Code: https://github.com/DachunKai/Ev-DeblurVSR.
comment: AAAI 2025. Project page: https://dachunkai.github.io/evtexture.github.io/
☆ Towards Cardiac MRI Foundation Models: Comprehensive Visual-Tabular Representations for Whole-Heart Assessment and Beyond
Cardiac magnetic resonance imaging is the gold standard for non-invasive cardiac assessment, offering rich spatio-temporal views of the cardiac anatomy and physiology. Patient-level health factors, such as demographics, metabolic, and lifestyle, are known to substantially influence cardiovascular health and disease risk, yet remain uncaptured by CMR alone. To holistically understand cardiac health and to enable the best possible interpretation of an individual's disease risk, CMR and patient-level factors must be jointly exploited within an integrated framework. Recent multi-modal approaches have begun to bridge this gap, yet they often rely on limited spatio-temporal data and focus on isolated clinical tasks, thereby hindering the development of a comprehensive representation for cardiac health evaluation. To overcome these limitations, we introduce ViTa, a step toward foundation models that delivers a comprehensive representation of the heart and a precise interpretation of individual disease risk. Leveraging data from 42,000 UK Biobank participants, ViTa integrates 3D+T cine stacks from short-axis and long-axis views, enabling a complete capture of the cardiac cycle. These imaging data are then fused with detailed tabular patient-level factors, enabling context-aware insights. This multi-modal paradigm supports a wide spectrum of downstream tasks, including cardiac phenotype and physiological feature prediction, segmentation, and classification of cardiac and metabolic diseases within a single unified framework. By learning a shared latent representation that bridges rich imaging features and patient context, ViTa moves beyond traditional, task-specific models toward a universal, patient-specific understanding of cardiac health, highlighting its potential to advance clinical utility and scalability in cardiac analysis.
☆ Prototypes are Balanced Units for Efficient and Effective Partially Relevant Video Retrieval
In a retrieval system, simultaneously achieving search accuracy and efficiency is inherently challenging. This challenge is particularly pronounced in partially relevant video retrieval (PRVR), where incorporating more diverse context representations at varying temporal scales for each video enhances accuracy but increases computational and memory costs. To address this dichotomy, we propose a prototypical PRVR framework that encodes diverse contexts within a video into a fixed number of prototypes. We then introduce several strategies to enhance text association and video understanding within the prototypes, along with an orthogonal objective to ensure that the prototypes capture a diverse range of content. To keep the prototypes searchable via text queries while accurately encoding video contexts, we implement cross- and uni-modal reconstruction tasks. The cross-modal reconstruction task aligns the prototypes with textual features within a shared space, while the uni-modal reconstruction task preserves all video contexts during encoding. Additionally, we employ a video mixing technique to provide weak guidance to further align prototypes and associated textual representations. Extensive evaluations on TVR, ActivityNet-Captions, and QVHighlights validate the effectiveness of our approach without sacrificing efficiency.
☆ TTRD3: Texture Transfer Residual Denoising Dual Diffusion Model for Remote Sensing Image Super-Resolution
Remote Sensing Image Super-Resolution (RSISR) reconstructs high-resolution (HR) remote sensing images from low-resolution inputs to support fine-grained ground object interpretation. Existing methods face three key challenges: (1) Difficulty in extracting multi-scale features from spatially heterogeneous RS scenes, (2) Limited prior information causing semantic inconsistency in reconstructions, and (3) Trade-off imbalance between geometric accuracy and visual quality. To address these issues, we propose the Texture Transfer Residual Denoising Dual Diffusion Model (TTRD3) with three innovations: First, a Multi-scale Feature Aggregation Block (MFAB) employing parallel heterogeneous convolutional kernels for multi-scale feature extraction. Second, a Sparse Texture Transfer Guidance (STTG) module that transfers HR texture priors from reference images of similar scenes. Third, a Residual Denoising Dual Diffusion Model (RDDM) framework combining residual diffusion for deterministic reconstruction and noise diffusion for diverse generation. Experiments on multi-source RS datasets demonstrate TTRD3's superiority over state-of-the-art methods, achieving 1.43% LPIPS improvement and 3.67% FID enhancement compared to best-performing baselines. Code/model: https://github.com/LED-666/TTRD3.
☆ Riemannian Patch Assignment Gradient Flows
This paper introduces patch assignment flows for metric data labeling on graphs. Labelings are determined by regularizing initial local labelings through the dynamic interaction of both labels and label assignments across the graph, entirely encoded by a dictionary of competing labeled patches and mediated by patch assignment variables. Maximal consistency of patch assignments is achieved by geometric numerical integration of a Riemannian ascent flow, as critical point of a Lagrangian action functional. Experiments illustrate properties of the approach, including uncertainty quantification of label assignments.
☆ ChatEXAONEPath: An Expert-level Multimodal Large Language Model for Histopathology Using Whole Slide Images
Recent studies have made significant progress in developing large language models (LLMs) in the medical domain, which can answer expert-level questions and demonstrate the potential to assist clinicians in real-world clinical scenarios. Studies have also witnessed the importance of integrating various modalities with the existing LLMs for a better understanding of complex clinical contexts, which are innately multi-faceted by nature. Although studies have demonstrated the ability of multimodal LLMs in histopathology to answer questions from given images, they lack in understanding of thorough clinical context due to the patch-level data with limited information from public datasets. Thus, developing WSI-level MLLMs is significant in terms of the scalability and applicability of MLLMs in histopathology. In this study, we introduce an expert-level MLLM for histopathology using WSIs, dubbed as ChatEXAONEPath. We present a retrieval-based data generation pipeline using 10,094 pairs of WSIs and histopathology reports from The Cancer Genome Atlas (TCGA). We also showcase an AI-based evaluation protocol for a comprehensive understanding of the medical context from given multimodal information and evaluate generated answers compared to the original histopathology reports. We demonstrate the ability of diagnosing the given histopathology images using ChatEXAONEPath with the acceptance rate of 62.9% from 1,134 pairs of WSIs and reports. Our proposed model can understand pan-cancer WSIs and clinical context from various cancer types. We argue that our proposed model has the potential to assist clinicians by comprehensively understanding complex morphology of WSIs for cancer diagnosis through the integration of multiple modalities.
☆ CompGS++: Compressed Gaussian Splatting for Static and Dynamic Scene Representation
Gaussian splatting demonstrates proficiency for 3D scene modeling but suffers from substantial data volume due to inherent primitive redundancy. To enable future photorealistic 3D immersive visual communication applications, significant compression is essential for transmission over the existing Internet infrastructure. Hence, we propose Compressed Gaussian Splatting (CompGS++), a novel framework that leverages compact Gaussian primitives to achieve accurate 3D modeling with substantial size reduction for both static and dynamic scenes. Our design is based on the principle of eliminating redundancy both between and within primitives. Specifically, we develop a comprehensive prediction paradigm to address inter-primitive redundancy through spatial and temporal primitive prediction modules. The spatial primitive prediction module establishes predictive relationships for scene primitives and enables most primitives to be encoded as compact residuals, substantially reducing the spatial redundancy. We further devise a temporal primitive prediction module to handle dynamic scenes, which exploits primitive correlations across timestamps to effectively reduce temporal redundancy. Moreover, we devise a rate-constrained optimization module that jointly minimizes reconstruction error and rate consumption. This module effectively eliminates parameter redundancy within primitives and enhances the overall compactness of scene representations. Comprehensive evaluations across multiple benchmark datasets demonstrate that CompGS++ significantly outperforms existing methods, achieving superior compression performance while preserving accurate scene modeling. Our implementation will be made publicly available on GitHub to facilitate further research.
comment: Submitted to a journal
☆ Pose and Facial Expression Transfer by using StyleGAN
We propose a method to transfer pose and expression between face images. Given a source and target face portrait, the model produces an output image in which the pose and expression of the source face image are transferred onto the target identity. The architecture consists of two encoders and a mapping network that projects the two inputs into the latent space of StyleGAN2, which finally generates the output. The training is self-supervised from video sequences of many individuals. Manual labeling is not required. Our model enables the synthesis of random identities with controllable pose and expression. Close-to-real-time performance is achieved.
comment: Accepted at CVWW 2024. Presented in Terme Olimia, Slovenia
☆ Hierarchical Feature Learning for Medical Point Clouds via State Space Model
Deep learning-based point cloud modeling has been widely investigated as an indispensable component of general shape analysis. Recently, transformer and state space model (SSM) have shown promising capacities in point cloud learning. However, limited research has been conducted on medical point clouds, which have great potential in disease diagnosis and treatment. This paper presents an SSM-based hierarchical feature learning framework for medical point cloud understanding. Specifically, we down-sample input into multiple levels through the farthest point sampling. At each level, we perform a series of k-nearest neighbor (KNN) queries to aggregate multi-scale structural information. To assist SSM in processing point clouds, we introduce coordinate-order and inside-out scanning strategies for efficient serialization of irregular points. Point features are calculated progressively from short neighbor sequences and long point sequences through vanilla and group Point SSM blocks, to capture both local patterns and long-range dependencies. To evaluate the proposed method, we build a large-scale medical point cloud dataset named MedPointS for anatomy classification, completion, and segmentation. Extensive experiments conducted on MedPointS demonstrate that our method achieves superior performance across all tasks. The dataset is available at https://flemme-docs.readthedocs.io/en/latest/medpoints.html. Code is merged to a public medical imaging platform: https://github.com/wlsdzyzl/flemme.
comment: 10 pages, 3 figures
☆ GSAC: Leveraging Gaussian Splatting for Photorealistic Avatar Creation with Unity Integration
Photorealistic avatars have become essential for immersive applications in virtual reality (VR) and augmented reality (AR), enabling lifelike interactions in areas such as training simulations, telemedicine, and virtual collaboration. These avatars bridge the gap between the physical and digital worlds, improving the user experience through realistic human representation. However, existing avatar creation techniques face significant challenges, including high costs, long creation times, and limited utility in virtual applications. Manual methods, such as MetaHuman, require extensive time and expertise, while automatic approaches, such as NeRF-based pipelines often lack efficiency, detailed facial expression fidelity, and are unable to be rendered at a speed sufficent for real-time applications. By involving several cutting-edge modern techniques, we introduce an end-to-end 3D Gaussian Splatting (3DGS) avatar creation pipeline that leverages monocular video input to create a scalable and efficient photorealistic avatar directly compatible with the Unity game engine. Our pipeline incorporates a novel Gaussian splatting technique with customized preprocessing that enables the user of "in the wild" monocular video capture, detailed facial expression reconstruction and embedding within a fully rigged avatar model. Additionally, we present a Unity-integrated Gaussian Splatting Avatar Editor, offering a user-friendly environment for VR/AR application development. Experimental results validate the effectiveness of our preprocessing pipeline in standardizing custom data for 3DGS training and demonstrate the versatility of Gaussian avatars in Unity, highlighting the scalability and practicality of our approach.
☆ All-in-One Transferring Image Compression from Human Perception to Multi-Machine Perception
Efficiently transferring Learned Image Compression (LIC) model from human perception to machine perception is an emerging challenge in vision-centric representation learning. Existing approaches typically adapt LIC to downstream tasks in a single-task manner, which is inefficient, lacks task interaction, and results in multiple task-specific bitstreams. To address these limitations, we propose an asymmetric adaptor framework that supports multi-task adaptation within a single model. Our method introduces a shared adaptor to learn general semantic features and task-specific adaptors to preserve task-level distinctions. With only lightweight plug-in modules and a frozen base codec, our method achieves strong performance across multiple tasks while maintaining compression efficiency. Experiments on the PASCAL-Context benchmark demonstrate that our method outperforms both Fully Fine-Tuned and other Parameter Efficient Fine-Tuned (PEFT) baselines, and validating the effectiveness of multi-vision transferring.
comment: 8 pages, 5 figures
☆ Enhancing Cocoa Pod Disease Classification via Transfer Learning and Ensemble Methods: Toward Robust Predictive Modeling
This study presents an ensemble-based approach for cocoa pod disease classification by integrating transfer learning with three ensemble learning strategies: Bagging, Boosting, and Stacking. Pre-trained convolutional neural networks, including VGG16, VGG19, ResNet50, ResNet101, InceptionV3, and Xception, were fine-tuned and employed as base learners to detect three disease categories: Black Pod Rot, Pod Borer, and Healthy. A balanced dataset of 6,000 cocoa pod images was curated and augmented to ensure robustness against variations in lighting, orientation, and disease severity. The performance of each ensemble method was evaluated using accuracy, precision, recall, and F1-score. Experimental results show that Bagging consistently achieved superior classification performance with a test accuracy of 100%, outperforming Boosting (97%) and Stacking (92%). The findings confirm that combining transfer learning with ensemble techniques improves model generalization and reliability, making it a promising direction for precision agriculture and automated crop disease management.
☆ MathPhys-Guided Coarse-to-Fine Anomaly Synthesis with SQE-Driven Bi-Level Optimization for Anomaly Detection
Anomaly detection is a crucial task in computer vision, yet collecting real-world defect images is inherently difficult due to the rarity and unpredictability of anomalies. Consequently, researchers have turned to synthetic methods for training data augmentation. However, existing synthetic strategies (e.g., naive cut-and-paste or inpainting) overlook the underlying physical causes of defects, leading to inconsistent, low-fidelity anomalies that hamper model generalization to real-world complexities. In this thesis, we introduced a novel pipeline that generates synthetic anomalies through Math-Physics model guidance, refines them via a Coarse-to-Fine approach and employs a bi-level optimization strategy with a Synthesis Quality Estimator(SQE). By incorporating physical modeling of cracks, corrosion, and deformation, our method produces realistic defect masks, which are subsequently enhanced in two phases. The first stage (npcF) enforces a PDE-based consistency to achieve a globally coherent anomaly structure, while the second stage (npcF++) further improves local fidelity using wavelet transforms and boundary synergy blocks. Additionally, we leverage SQE-driven weighting, ensuring that high-quality synthetic samples receive greater emphasis during training. To validate our approach, we conducted comprehensive experiments on three widely adopted industrial anomaly detection benchmarks: MVTec AD, VisA, and BTAD. Across these datasets, the proposed pipeline achieves state-of-the-art (SOTA) results in both image-AUROC and pixel-AUROC, confirming the effectiveness of our MaPhC2F and BiSQAD.
☆ Vision and Language Integration for Domain Generalization
Domain generalization aims at training on source domains to uncover a domain-invariant feature space, allowing the model to perform robust generalization ability on unknown target domains. However, due to domain gaps, it is hard to find reliable common image feature space, and the reason for that is the lack of suitable basic units for images. Different from image in vision space, language has comprehensive expression elements that can effectively convey semantics. Inspired by the semantic completeness of language and intuitiveness of image, we propose VLCA, which combine language space and vision space, and connect the multiple image domains by using semantic space as the bridge domain. Specifically, in language space, by taking advantage of the completeness of language basic units, we tend to capture the semantic representation of the relations between categories through word vector distance. Then, in vision space, by taking advantage of the intuitiveness of image features, the common pattern of sample features with the same class is explored through low-rank approximation. In the end, the language representation is aligned with the vision representation through the multimodal space of text and image. Experiments demonstrate the effectiveness of the proposed method.
☆ Rethinking Temporal Fusion with a Unified Gradient Descent View for 3D Semantic Occupancy Prediction CVPR 2025
We present GDFusion, a temporal fusion method for vision-based 3D semantic occupancy prediction (VisionOcc). GDFusion opens up the underexplored aspects of temporal fusion within the VisionOcc framework, focusing on both temporal cues and fusion strategies. It systematically examines the entire VisionOcc pipeline, identifying three fundamental yet previously overlooked temporal cues: scene-level consistency, motion calibration, and geometric complementation. These cues capture diverse facets of temporal evolution and make distinct contributions across various modules in the VisionOcc framework. To effectively fuse temporal signals across heterogeneous representations, we propose a novel fusion strategy by reinterpreting the formulation of vanilla RNNs. This reinterpretation leverages gradient descent on features to unify the integration of diverse temporal information, seamlessly embedding the proposed temporal cues into the network. Extensive experiments on nuScenes demonstrate that GDFusion significantly outperforms established baselines. Notably, on Occ3D benchmark, it achieves 1.4\%-4.8\% mIoU improvements and reduces memory consumption by 27\%-72\%.
comment: CVPR 2025
☆ Disentangling Polysemantic Channels in Convolutional Neural Networks CVPR 2025
Mechanistic interpretability is concerned with analyzing individual components in a (convolutional) neural network (CNN) and how they form larger circuits representing decision mechanisms. These investigations are challenging since CNNs frequently learn polysemantic channels that encode distinct concepts, making them hard to interpret. To address this, we propose an algorithm to disentangle a specific kind of polysemantic channel into multiple channels, each responding to a single concept. Our approach restructures weights in a CNN, utilizing that different concepts within the same channel exhibit distinct activation patterns in the previous layer. By disentangling these polysemantic features, we enhance the interpretability of CNNs, ultimately improving explanatory techniques such as feature visualizations.
comment: Accepted at CVPR 2025 Workshop on Mechanistic Interpretability for Vision (MIV). Code: https://github.com/visinf/disentangle-channels
☆ Efficient Masked Image Compression with Position-Indexed Self-Attention
In recent years, image compression for high-level vision tasks has attracted considerable attention from researchers. Given that object information in images plays a far more crucial role in downstream tasks than background information, some studies have proposed semantically structuring the bitstream to selectively transmit and reconstruct only the information required by these tasks. However, such methods structure the bitstream after encoding, meaning that the coding process still relies on the entire image, even though much of the encoded information will not be transmitted. This leads to redundant computations. Traditional image compression methods require a two-dimensional image as input, and even if the unimportant regions of the image are set to zero by applying a semantic mask, these regions still participate in subsequent computations as part of the image. To address such limitations, we propose an image compression method based on a position-indexed self-attention mechanism that encodes and decodes only the visible parts of the masked image. Compared to existing semantic-structured compression methods, our approach can significantly reduce computational costs.
☆ Real-time High-fidelity Gaussian Human Avatars with Position-based Interpolation of Spatially Distributed MLPs CVPR 2025
Many works have succeeded in reconstructing Gaussian human avatars from multi-view videos. However, they either struggle to capture pose-dependent appearance details with a single MLP, or rely on a computationally intensive neural network to reconstruct high-fidelity appearance but with rendering performance degraded to non-real-time. We propose a novel Gaussian human avatar representation that can reconstruct high-fidelity pose-dependence appearance with details and meanwhile can be rendered in real time. Our Gaussian avatar is empowered by spatially distributed MLPs which are explicitly located on different positions on human body. The parameters stored in each Gaussian are obtained by interpolating from the outputs of its nearby MLPs based on their distances. To avoid undesired smooth Gaussian property changing during interpolation, for each Gaussian we define a set of Gaussian offset basis, and a linear combination of basis represents the Gaussian property offsets relative to the neutral properties. Then we propose to let the MLPs output a set of coefficients corresponding to the basis. In this way, although Gaussian coefficients are derived from interpolation and change smoothly, the Gaussian offset basis is learned freely without constraints. The smoothly varying coefficients combined with freely learned basis can still produce distinctly different Gaussian property offsets, allowing the ability to learn high-frequency spatial signals. We further use control points to constrain the Gaussians distributed on a surface layer rather than allowing them to be irregularly distributed inside the body, to help the human avatar generalize better when animated under novel poses. Compared to the state-of-the-art method, our method achieves better appearance quality with finer details while the rendering speed is significantly faster under novel views and novel poses.
comment: CVPR 2025
☆ Taccel: Scaling Up Vision-based Tactile Robotics via High-performance GPU Simulation
Tactile sensing is crucial for achieving human-level robotic capabilities in manipulation tasks. VBTSs have emerged as a promising solution, offering high spatial resolution and cost-effectiveness by sensing contact through camera-captured deformation patterns of elastic gel pads. However, these sensors' complex physical characteristics and visual signal processing requirements present unique challenges for robotic applications. The lack of efficient and accurate simulation tools for VBTS has significantly limited the scale and scope of tactile robotics research. Here we present Taccel, a high-performance simulation platform that integrates IPC and ABD to model robots, tactile sensors, and objects with both accuracy and unprecedented speed, achieving an 18-fold acceleration over real-time across thousands of parallel environments. Unlike previous simulators that operate at sub-real-time speeds with limited parallelization, Taccel provides precise physics simulation and realistic tactile signals while supporting flexible robot-sensor configurations through user-friendly APIs. Through extensive validation in object recognition, robotic grasping, and articulated object manipulation, we demonstrate precise simulation and successful sim-to-real transfer. These capabilities position Taccel as a powerful tool for scaling up tactile robotics research and development. By enabling large-scale simulation and experimentation with tactile sensing, Taccel accelerates the development of more capable robotic systems, potentially transforming how robots interact with and understand their physical environment.
comment: 17 pages, 7 figures
☆ Second-order Optimization of Gaussian Splats with Importance Sampling
3D Gaussian Splatting (3DGS) is widely used for novel view synthesis due to its high rendering quality and fast inference time. However, 3DGS predominantly relies on first-order optimizers such as Adam, which leads to long training times. To address this limitation, we propose a novel second-order optimization strategy based on Levenberg-Marquardt (LM) and Conjugate Gradient (CG), which we specifically tailor towards Gaussian Splatting. Our key insight is that the Jacobian in 3DGS exhibits significant sparsity since each Gaussian affects only a limited number of pixels. We exploit this sparsity by proposing a matrix-free and GPU-parallelized LM optimization. To further improve its efficiency, we propose sampling strategies for both the camera views and loss function and, consequently, the normal equation, significantly reducing the computational complexity. In addition, we increase the convergence rate of the second-order approximation by introducing an effective heuristic to determine the learning rate that avoids the expensive computation cost of line search methods. As a result, our method achieves a $3\times$ speedup over standard LM and outperforms Adam by $~6\times$ when the Gaussian count is low while remaining competitive for moderate counts. Project Page: https://vcai.mpi-inf.mpg.de/projects/LM-IS
☆ Tree-NeRV: A Tree-Structured Neural Representation for Efficient Non-Uniform Video Encoding
Implicit Neural Representations for Videos (NeRV) have emerged as a powerful paradigm for video representation, enabling direct mappings from frame indices to video frames. However, existing NeRV-based methods do not fully exploit temporal redundancy, as they rely on uniform sampling along the temporal axis, leading to suboptimal rate-distortion (RD) performance. To address this limitation, we propose Tree-NeRV, a novel tree-structured feature representation for efficient and adaptive video encoding. Unlike conventional approaches, Tree-NeRV organizes feature representations within a Binary Search Tree (BST), enabling non-uniform sampling along the temporal axis. Additionally, we introduce an optimization-driven sampling strategy, dynamically allocating higher sampling density to regions with greater temporal variation. Extensive experiments demonstrate that Tree-NeRV achieves superior compression efficiency and reconstruction quality, outperforming prior uniform sampling-based methods. Code will be released.
comment: 16 pages, 14 figures
☆ SC3EF: A Joint Self-Correlation and Cross-Correspondence Estimation Framework for Visible and Thermal Image Registration
Multispectral imaging plays a critical role in a range of intelligent transportation applications, including advanced driver assistance systems (ADAS), traffic monitoring, and night vision. However, accurate visible and thermal (RGB-T) image registration poses a significant challenge due to the considerable modality differences. In this paper, we present a novel joint Self-Correlation and Cross-Correspondence Estimation Framework (SC3EF), leveraging both local representative features and global contextual cues to effectively generate RGB-T correspondences. For this purpose, we design a convolution-transformer-based pipeline to extract local representative features and encode global correlations of intra-modality for inter-modality correspondence estimation between unaligned visible and thermal images. After merging the local and global correspondence estimation results, we further employ a hierarchical optical flow estimation decoder to progressively refine the estimated dense correspondence maps. Extensive experiments demonstrate the effectiveness of our proposed method, outperforming the current state-of-the-art (SOTA) methods on representative RGB-T datasets. Furthermore, it also shows competitive generalization capabilities across challenging scenarios, including large parallax, severe occlusions, adverse weather, and other cross-modal datasets (e.g., RGB-N and RGB-D).
☆ Computer-Aided Design of Personalized Occlusal Positioning Splints Using Multimodal 3D Data
Contemporary digital technology has a pivotal role in the design of customized medical appliances, including occlusal splints used in the treatment of stomatognathic system dysfunctions. We present an approach to computer-aided design and precision assessment of positioning occlusal splints, bridging clinical concepts with current digital dental practice. In our model, a 3D splint is generated based on a transformation matrix that represents the therapeutic change in mandibular position, defined by a specialist using a virtual patient model reconstructed from intraoral scans, CBCT, 3D facial scans and plaster model digitisation. The paper introduces a novel method for generating splints that accurately reproduce occlusal conditions in the therapeutic position, including a mechanism for resolving surface conflicts through virtual embossing. We demonstrate how transformation matrices can be acquired through clinical tools and intraoral devices, and evaluate the accuracy of the designed and printed splints using profile and surface deviation analysis. The proposed method enables reproducible, patient-specific splint fabrication and opens new possibilities in diagnostics, multimodal image registration and quantification of occlusal discrepancies.
☆ 3D-PNAS: 3D Industrial Surface Anomaly Synthesis with Perlin Noise
Large pretrained vision foundation models have shown significant potential in various vision tasks. However, for industrial anomaly detection, the scarcity of real defect samples poses a critical challenge in leveraging these models. While 2D anomaly generation has significantly advanced with established generative models, the adoption of 3D sensors in industrial manufacturing has made leveraging 3D data for surface quality inspection an emerging trend. In contrast to 2D techniques, 3D anomaly generation remains largely unexplored, limiting the potential of 3D data in industrial quality inspection. To address this gap, we propose a novel yet simple 3D anomaly generation method, 3D-PNAS, based on Perlin noise and surface parameterization. Our method generates realistic 3D surface anomalies by projecting the point cloud onto a 2D plane, sampling multi-scale noise values from a Perlin noise field, and perturbing the point cloud along its normal direction. Through comprehensive visualization experiments, we demonstrate how key parameters - including noise scale, perturbation strength, and octaves, provide fine-grained control over the generated anomalies, enabling the creation of diverse defect patterns from pronounced deformations to subtle surface variations. Additionally, our cross-category experiments show that the method produces consistent yet geometrically plausible anomalies across different object types, adapting to their specific surface characteristics. We also provide a comprehensive codebase and visualization toolkit to facilitate future research.
☆ High-Fidelity Image Inpainting with Multimodal Guided GAN Inversion
Generative Adversarial Network (GAN) inversion have demonstrated excellent performance in image inpainting that aims to restore lost or damaged image texture using its unmasked content. Previous GAN inversion-based methods usually utilize well-trained GAN models as effective priors to generate the realistic regions for missing holes. Despite excellence, they ignore a hard constraint that the unmasked regions in the input and the output should be the same, resulting in a gap between GAN inversion and image inpainting and thus degrading the performance. Besides, existing GAN inversion approaches often consider a single modality of the input image, neglecting other auxiliary cues in images for improvements. Addressing these problems, we propose a novel GAN inversion approach, dubbed MMInvertFill, for image inpainting. MMInvertFill contains primarily a multimodal guided encoder with a pre-modulation and a GAN generator with F&W+ latent space. Specifically, the multimodal encoder aims to enhance the multi-scale structures with additional semantic segmentation edge texture modalities through a gated mask-aware attention module. Afterwards, a pre-modulation is presented to encode these structures into style vectors. To mitigate issues of conspicuous color discrepancy and semantic inconsistency, we introduce the F&W+ latent space to bridge the gap between GAN inversion and image inpainting. Furthermore, in order to reconstruct faithful and photorealistic images, we devise a simple yet effective Soft-update Mean Latent module to capture more diversified in-domain patterns for generating high-fidelity textures for massive corruptions. In our extensive experiments on six challenging datasets, we show that our MMInvertFill qualitatively and quantitatively outperforms other state-of-the-arts and it supports the completion of out-of-domain images effectively.
comment: Accepted to IJCV. arXiv admin note: text overlap with arXiv:2208.11850
☆ ALT: A Python Package for Lightweight Feature Representation in Time Series Classification
We introduce ALT, an open-source Python package created for efficient and accurate time series classification (TSC). The package implements the adaptive law-based transformation (ALT) algorithm, which transforms raw time series data into a linearly separable feature space using variable-length shifted time windows. This adaptive approach enhances its predecessor, the linear law-based transformation (LLT), by effectively capturing patterns of varying temporal scales. The software is implemented for scalability, interpretability, and ease of use, achieving state-of-the-art performance with minimal computational overhead. Extensive benchmarking on real-world datasets demonstrates the utility of ALT for diverse TSC tasks in physics and related domains.
comment: 16 pages, 4 figures
☆ Image-Editing Specialists: An RLAIF Approach for Diffusion Models
We present a novel approach to training specialized instruction-based image-editing diffusion models, addressing key challenges in structural preservation with input images and semantic alignment with user prompts. We introduce an online reinforcement learning framework that aligns the diffusion model with human preferences without relying on extensive human annotations or curating a large dataset. Our method significantly improves the realism and alignment with instructions in two ways. First, the proposed models achieve precise and structurally coherent modifications in complex scenes while maintaining high fidelity in instruction-irrelevant areas. Second, they capture fine nuances in the desired edit by leveraging a visual prompt, enabling detailed control over visual edits without lengthy textual prompts. This approach simplifies users' efforts to achieve highly specific edits, requiring only 5 reference images depicting a certain concept for training. Experimental results demonstrate that our models can perform intricate edits in complex scenes, after just 10 training steps. Finally, we showcase the versatility of our method by applying it to robotics, where enhancing the visual realism of simulated environments through targeted sim-to-real image edits improves their utility as proxies for real-world settings.
☆ UncAD: Towards Safe End-to-end Autonomous Driving via Online Map Uncertainty
End-to-end autonomous driving aims to produce planning trajectories from raw sensors directly. Currently, most approaches integrate perception, prediction, and planning modules into a fully differentiable network, promising great scalability. However, these methods typically rely on deterministic modeling of online maps in the perception module for guiding or constraining vehicle planning, which may incorporate erroneous perception information and further compromise planning safety. To address this issue, we delve into the importance of online map uncertainty for enhancing autonomous driving safety and propose a novel paradigm named UncAD. Specifically, UncAD first estimates the uncertainty of the online map in the perception module. It then leverages the uncertainty to guide motion prediction and planning modules to produce multi-modal trajectories. Finally, to achieve safer autonomous driving, UncAD proposes an uncertainty-collision-aware planning selection strategy according to the online map uncertainty to evaluate and select the best trajectory. In this study, we incorporate UncAD into various state-of-the-art (SOTA) end-to-end methods. Experiments on the nuScenes dataset show that integrating UncAD, with only a 1.9% increase in parameters, can reduce collision rates by up to 26% and drivable area conflict rate by up to 42%. Codes, pre-trained models, and demo videos can be accessed at https://github.com/pengxuanyang/UncAD.
☆ TwoSquared: 4D Generation from 2D Image Pairs
Despite the astonishing progress in generative AI, 4D dynamic object generation remains an open challenge. With limited high-quality training data and heavy computing requirements, the combination of hallucinating unseen geometry together with unseen movement poses great challenges to generative models. In this work, we propose TwoSquared as a method to obtain a 4D physically plausible sequence starting from only two 2D RGB images corresponding to the beginning and end of the action. Instead of directly solving the 4D generation problem, TwoSquared decomposes the problem into two steps: 1) an image-to-3D module generation based on the existing generative model trained on high-quality 3D assets, and 2) a physically inspired deformation module to predict intermediate movements. To this end, our method does not require templates or object-class-specific prior knowledge and can take in-the-wild images as input. In our experiments, we demonstrate that TwoSquared is capable of producing texture-consistent and geometry-consistent 4D sequences only given 2D images.
☆ Explainable Scene Understanding with Qualitative Representations and Graph Neural Networks
This paper investigates the integration of graph neural networks (GNNs) with Qualitative Explainable Graphs (QXGs) for scene understanding in automated driving. Scene understanding is the basis for any further reactive or proactive decision-making. Scene understanding and related reasoning is inherently an explanation task: why is another traffic participant doing something, what or who caused their actions? While previous work demonstrated QXGs' effectiveness using shallow machine learning models, these approaches were limited to analysing single relation chains between object pairs, disregarding the broader scene context. We propose a novel GNN architecture that processes entire graph structures to identify relevant objects in traffic scenes. We evaluate our method on the nuScenes dataset enriched with DriveLM's human-annotated relevance labels. Experimental results show that our GNN-based approach achieves superior performance compared to baseline methods. The model effectively handles the inherent class imbalance in relevant object identification tasks while considering the complete spatial-temporal relationships between all objects in the scene. Our work demonstrates the potential of combining qualitative representations with deep learning approaches for explainable scene understanding in autonomous driving systems.
comment: Workshop "Advancing Automated Driving in Highly Interactive Scenarios through Behavior Prediction, Trustworthy AI, and Remote Operations" @ 36th IEEE Intelligent Vehicles Symposium (IV)
☆ AAA-Gaussians: Anti-Aliased and Artifact-Free 3D Gaussian Rendering
Although 3D Gaussian Splatting (3DGS) has revolutionized 3D reconstruction, it still faces challenges such as aliasing, projection artifacts, and view inconsistencies, primarily due to the simplification of treating splats as 2D entities. We argue that incorporating full 3D evaluation of Gaussians throughout the 3DGS pipeline can effectively address these issues while preserving rasterization efficiency. Specifically, we introduce an adaptive 3D smoothing filter to mitigate aliasing and present a stable view-space bounding method that eliminates popping artifacts when Gaussians extend beyond the view frustum. Furthermore, we promote tile-based culling to 3D with screen-space planes, accelerating rendering and reducing sorting costs for hierarchical rasterization. Our method achieves state-of-the-art quality on in-distribution evaluation sets and significantly outperforms other approaches for out-of-distribution views. Our qualitative evaluations further demonstrate the effective removal of aliasing, distortions, and popping artifacts, ensuring real-time, artifact-free rendering.
☆ Saliency-Aware Diffusion Reconstruction for Effective Invisible Watermark Removal
As digital content becomes increasingly ubiquitous, the need for robust watermark removal techniques has grown due to the inadequacy of existing embedding techniques, which lack robustness. This paper introduces a novel Saliency-Aware Diffusion Reconstruction (SADRE) framework for watermark elimination on the web, combining adaptive noise injection, region-specific perturbations, and advanced diffusion-based reconstruction. SADRE disrupts embedded watermarks by injecting targeted noise into latent representations guided by saliency masks although preserving essential image features. A reverse diffusion process ensures high-fidelity image restoration, leveraging adaptive noise levels determined by watermark strength. Our framework is theoretically grounded with stability guarantees and achieves robust watermark removal across diverse scenarios. Empirical evaluations on state-of-the-art (SOTA) watermarking techniques demonstrate SADRE's superiority in balancing watermark disruption and image quality. SADRE sets a new benchmark for watermark elimination, offering a flexible and reliable solution for real-world web content. Code is available on~\href{https://github.com/inzamamulDU/SADRE}{\textbf{https://github.com/inzamamulDU/SADRE}}.
comment: Accepted at The Web Conference 2025
☆ Hybrid Dense-UNet201 Optimization for Pap Smear Image Segmentation Using Spider Monkey Optimization
Pap smear image segmentation is crucial for cervical cancer diagnosis. However, traditional segmentation models often struggle with complex cellular structures and variations in pap smear images. This study proposes a hybrid Dense-UNet201 optimization approach that integrates a pretrained DenseNet201 as the encoder for the U-Net architecture and optimizes it using the spider monkey optimization (SMO) algorithm. The Dense-UNet201 model excelled at feature extraction. The SMO was modified to handle categorical and discrete parameters. The SIPaKMeD dataset was used in this study and evaluated using key performance metrics, including loss, accuracy, Intersection over Union (IoU), and Dice coefficient. The experimental results showed that Dense-UNet201 outperformed U-Net, Res-UNet50, and Efficient-UNetB0. SMO Dense-UNet201 achieved a segmentation accuracy of 96.16%, an IoU of 91.63%, and a Dice coefficient score of 95.63%. These findings underscore the effectiveness of image preprocessing, pretrained models, and metaheuristic optimization in improving medical image analysis and provide new insights into cervical cell segmentation methods.
☆ Sign-In to the Lottery: Reparameterizing Sparse Training From Scratch
The performance gap between training sparse neural networks from scratch (PaI) and dense-to-sparse training presents a major roadblock for efficient deep learning. According to the Lottery Ticket Hypothesis, PaI hinges on finding a problem specific parameter initialization. As we show, to this end, determining correct parameter signs is sufficient. Yet, they remain elusive to PaI. To address this issue, we propose Sign-In, which employs a dynamic reparameterization that provably induces sign flips. Such sign flips are complementary to the ones that dense-to-sparse training can accomplish, rendering Sign-In as an orthogonal method. While our experiments and theory suggest performance improvements of PaI, they also carve out the main open challenge to close the gap between PaI and dense-to-sparse training.
comment: 21 pages, 9 figures
☆ CAGE-GS: High-fidelity Cage Based 3D Gaussian Splatting Deformation
As 3D Gaussian Splatting (3DGS) gains popularity as a 3D representation of real scenes, enabling user-friendly deformation to create novel scenes while preserving fine details from the original 3DGS has attracted significant research attention. We introduce CAGE-GS, a cage-based 3DGS deformation method that seamlessly aligns a source 3DGS scene with a user-defined target shape. Our approach learns a deformation cage from the target, which guides the geometric transformation of the source scene. While the cages effectively control structural alignment, preserving the textural appearance of 3DGS remains challenging due to the complexity of covariance parameters. To address this, we employ a Jacobian matrix-based strategy to update the covariance parameters of each Gaussian, ensuring texture fidelity post-deformation. Our method is highly flexible, accommodating various target shape representations, including texts, images, point clouds, meshes and 3DGS models. Extensive experiments and ablation studies on both public datasets and newly proposed scenes demonstrate that our method significantly outperforms existing techniques in both efficiency and deformation quality.
☆ TSGS: Improving Gaussian Splatting for Transparent Surface Reconstruction via Normal and De-lighting Priors
Reconstructing transparent surfaces is essential for tasks such as robotic manipulation in labs, yet it poses a significant challenge for 3D reconstruction techniques like 3D Gaussian Splatting (3DGS). These methods often encounter a transparency-depth dilemma, where the pursuit of photorealistic rendering through standard $\alpha$-blending undermines geometric precision, resulting in considerable depth estimation errors for transparent materials. To address this issue, we introduce Transparent Surface Gaussian Splatting (TSGS), a new framework that separates geometry learning from appearance refinement. In the geometry learning stage, TSGS focuses on geometry by using specular-suppressed inputs to accurately represent surfaces. In the second stage, TSGS improves visual fidelity through anisotropic specular modeling, crucially maintaining the established opacity to ensure geometric accuracy. To enhance depth inference, TSGS employs a first-surface depth extraction method. This technique uses a sliding window over $\alpha$-blending weights to pinpoint the most likely surface location and calculates a robust weighted average depth. To evaluate the transparent surface reconstruction task under realistic conditions, we collect a TransLab dataset that includes complex transparent laboratory glassware. Extensive experiments on TransLab show that TSGS achieves accurate geometric reconstruction and realistic rendering of transparent objects simultaneously within the efficient 3DGS framework. Specifically, TSGS significantly surpasses current leading methods, achieving a 37.3% reduction in chamfer distance and an 8.0% improvement in F1 score compared to the top baseline. The code and dataset will be released at https://longxiang-ai.github.io/TSGS/.
comment: Project page: https://longxiang-ai.github.io/TSGS/
☆ EarthGPT-X: Enabling MLLMs to Flexibly and Comprehensively Understand Multi-Source Remote Sensing Imagery
Recent advances in the visual-language area have developed natural multi-modal large language models (MLLMs) for spatial reasoning through visual prompting. However, due to remote sensing (RS) imagery containing abundant geospatial information that differs from natural images, it is challenging to effectively adapt natural spatial models to the RS domain. Moreover, current RS MLLMs are limited in overly narrow interpretation levels and interaction manner, hindering their applicability in real-world scenarios. To address those challenges, a spatial MLLM named EarthGPT-X is proposed, enabling a comprehensive understanding of multi-source RS imagery, such as optical, synthetic aperture radar (SAR), and infrared. EarthGPT-X offers zoom-in and zoom-out insight, and possesses flexible multi-grained interactive abilities. Moreover, EarthGPT-X unifies two types of critical spatial tasks (i.e., referring and grounding) into a visual prompting framework. To achieve these versatile capabilities, several key strategies are developed. The first is the multi-modal content integration method, which enhances the interplay between images, visual prompts, and text instructions. Subsequently, a cross-domain one-stage fusion training strategy is proposed, utilizing the large language model (LLM) as a unified interface for multi-source multi-task learning. Furthermore, by incorporating a pixel perception module, the referring and grounding tasks are seamlessly unified within a single framework. In addition, the experiments conducted demonstrate the superiority of the proposed EarthGPT-X in multi-grained tasks and its impressive flexibility in multi-modal interaction, revealing significant advancements of MLLM in the RS field.
☆ ARAP-GS: Drag-driven As-Rigid-As-Possible 3D Gaussian Splatting Editing with Diffusion Prior
Drag-driven editing has become popular among designers for its ability to modify complex geometric structures through simple and intuitive manipulation, allowing users to adjust and reshape content with minimal technical skill. This drag operation has been incorporated into numerous methods to facilitate the editing of 2D images and 3D meshes in design. However, few studies have explored drag-driven editing for the widely-used 3D Gaussian Splatting (3DGS) representation, as deforming 3DGS while preserving shape coherence and visual continuity remains challenging. In this paper, we introduce ARAP-GS, a drag-driven 3DGS editing framework based on As-Rigid-As-Possible (ARAP) deformation. Unlike previous 3DGS editing methods, we are the first to apply ARAP deformation directly to 3D Gaussians, enabling flexible, drag-driven geometric transformations. To preserve scene appearance after deformation, we incorporate an advanced diffusion prior for image super-resolution within our iterative optimization process. This approach enhances visual quality while maintaining multi-view consistency in the edited results. Experiments show that ARAP-GS outperforms current methods across diverse 3D scenes, demonstrating its effectiveness and superiority for drag-driven 3DGS editing. Additionally, our method is highly efficient, requiring only 10 to 20 minutes to edit a scene on a single RTX 3090 GPU.
☆ Set You Straight: Auto-Steering Denoising Trajectories to Sidestep Unwanted Concepts
Ensuring the ethical deployment of text-to-image models requires effective techniques to prevent the generation of harmful or inappropriate content. While concept erasure methods offer a promising solution, existing finetuning-based approaches suffer from notable limitations. Anchor-free methods risk disrupting sampling trajectories, leading to visual artifacts, while anchor-based methods rely on the heuristic selection of anchor concepts. To overcome these shortcomings, we introduce a finetuning framework, dubbed ANT, which Automatically guides deNoising Trajectories to avoid unwanted concepts. ANT is built on a key insight: reversing the condition direction of classifier-free guidance during mid-to-late denoising stages enables precise content modification without sacrificing early-stage structural integrity. This inspires a trajectory-aware objective that preserves the integrity of the early-stage score function field, which steers samples toward the natural image manifold, without relying on heuristic anchor concept selection. For single-concept erasure, we propose an augmentation-enhanced weight saliency map to precisely identify the critical parameters that most significantly contribute to the unwanted concept, enabling more thorough and efficient erasure. For multi-concept erasure, our objective function offers a versatile plug-and-play solution that significantly boosts performance. Extensive experiments demonstrate that ANT achieves state-of-the-art results in both single and multi-concept erasure, delivering high-quality, safe outputs without compromising the generative fidelity. Code is available at https://github.com/lileyang1210/ANT
comment: Preprint
☆ Stronger, Steadier & Superior: Geometric Consistency in Depth VFM Forges Domain Generalized Semantic Segmentation
Vision Foundation Models (VFMs) have delivered remarkable performance in Domain Generalized Semantic Segmentation (DGSS). However, recent methods often overlook the fact that visual cues are susceptible, whereas the underlying geometry remains stable, rendering depth information more robust. In this paper, we investigate the potential of integrating depth information with features from VFMs, to improve the geometric consistency within an image and boost the generalization performance of VFMs. We propose a novel fine-tuning DGSS framework, named DepthForge, which integrates the visual cues from frozen DINOv2 or EVA02 and depth cues from frozen Depth Anything V2. In each layer of the VFMs, we incorporate depth-aware learnable tokens to continuously decouple domain-invariant visual and spatial information, thereby enhancing depth awareness and attention of the VFMs. Finally, we develop a depth refinement decoder and integrate it into the model architecture to adaptively refine multi-layer VFM features and depth-aware learnable tokens. Extensive experiments are conducted based on various DGSS settings and five different datsets as unseen target domains. The qualitative and quantitative results demonstrate that our method significantly outperforms alternative approaches with stronger performance, steadier visual-spatial attention, and superior generalization ability. In particular, DepthForge exhibits outstanding performance under extreme conditions (e.g., night and snow). Code is available at https://github.com/anonymouse-xzrptkvyqc/DepthForge.
☆ LAD-Reasoner: Tiny Multimodal Models are Good Reasoners for Logical Anomaly Detection
Recent advances in industrial anomaly detection have highlighted the need for deeper logical anomaly analysis, where unexpected relationships among objects, counts, and spatial configurations must be identified and explained. Existing approaches often rely on large-scale external reasoning modules or elaborate pipeline designs, hindering practical deployment and interpretability. To address these limitations, we introduce a new task, Reasoning Logical Anomaly Detection (RLAD), which extends traditional anomaly detection by incorporating logical reasoning. We propose a new framework, LAD-Reasoner, a customized tiny multimodal language model built on Qwen2.5-VL 3B. Our approach leverages a two-stage training paradigm that first employs Supervised Fine-Tuning (SFT) for fine-grained visual understanding, followed by Group Relative Policy Optimization (GRPO) to refine logical anomaly detection and enforce coherent, human-readable reasoning. Crucially, reward signals are derived from both the detection accuracy and the structural quality of the outputs, obviating the need for building chain of thought (CoT) reasoning data. Experiments on the MVTec LOCO AD dataset show that LAD-Reasoner, though significantly smaller, matches the performance of Qwen2.5-VL-72B in accuracy and F1 score, and further excels in producing concise and interpretable rationales. This unified design reduces reliance on large models and complex pipelines, while offering transparent and interpretable insights into logical anomaly detection. Code and data will be released.
☆ Privacy Protection Against Personalized Text-to-Image Synthesis via Cross-image Consistency Constraints
The rapid advancement of diffusion models and personalization techniques has made it possible to recreate individual portraits from just a few publicly available images. While such capabilities empower various creative applications, they also introduce serious privacy concerns, as adversaries can exploit them to generate highly realistic impersonations. To counter these threats, anti-personalization methods have been proposed, which add adversarial perturbations to published images to disrupt the training of personalization models. However, existing approaches largely overlook the intrinsic multi-image nature of personalization and instead adopt a naive strategy of applying perturbations independently, as commonly done in single-image settings. This neglects the opportunity to leverage inter-image relationships for stronger privacy protection. Therefore, we advocate for a group-level perspective on privacy protection against personalization. Specifically, we introduce Cross-image Anti-Personalization (CAP), a novel framework that enhances resistance to personalization by enforcing style consistency across perturbed images. Furthermore, we develop a dynamic ratio adjustment strategy that adaptively balances the impact of the consistency loss throughout the attack iterations. Extensive experiments on the classical CelebHQ and VGGFace2 benchmarks show that CAP substantially improves existing methods.
☆ Mask Image Watermarking
We present MaskMark, a simple, efficient and flexible framework for image watermarking. MaskMark has two variants: MaskMark-D, which supports global watermark embedding, watermark localization, and local watermark extraction for applications such as tamper detection, and MaskMark-ED, which focuses on local watermark embedding and extraction with enhanced robustness in small regions, enabling localized image protection. Built upon the classical Encoder- Distortion-Decoder training paradigm, MaskMark-D introduces a simple masking mechanism during the decoding stage to support both global and local watermark extraction. A mask is applied to the watermarked image before extraction, allowing the decoder to focus on selected regions and learn local extraction. A localization module is also integrated into the decoder to identify watermark regions during inference, reducing interference from irrelevant content and improving accuracy. MaskMark-ED extends this design by incorporating the mask into the encoding stage as well, guiding the encoder to embed the watermark in designated local regions for enhanced robustness. Comprehensive experiments show that MaskMark achieves state-of-the-art performance in global watermark extraction, local watermark extraction, watermark localization, and multi-watermark embedding. It outperforms all existing baselines, including the recent leading model WAM for local watermarking, while preserving high visual quality of the watermarked images. MaskMark is also flexible, by adjusting the distortion layer, it can adapt to different robustness requirements with just a few steps of fine-tuning. Moreover, our approach is efficient and easy to optimize, requiring only 20 hours on a single A6000 GPU with just 1/15 the computational cost of WAM.
comment: 23 pages, 18 figures
☆ TUMLS: Trustful Fully Unsupervised Multi-Level Segmentation for Whole Slide Images of Histology
Digital pathology, augmented by artificial intelligence (AI), holds significant promise for improving the workflow of pathologists. However, challenges such as the labor-intensive annotation of whole slide images (WSIs), high computational demands, and trust concerns arising from the absence of uncertainty estimation in predictions hinder the practical application of current AI methodologies in histopathology. To address these issues, we present a novel trustful fully unsupervised multi-level segmentation methodology (TUMLS) for WSIs. TUMLS adopts an autoencoder (AE) as a feature extractor to identify the different tissue types within low-resolution training data. It selects representative patches from each identified group based on an uncertainty measure and then does unsupervised nuclei segmentation in their respective higher-resolution space without using any ML algorithms. Crucially, this solution integrates seamlessly into clinicians workflows, transforming the examination of a whole WSI into a review of concise, interpretable cross-level insights. This integration significantly enhances and accelerates the workflow while ensuring transparency. We evaluated our approach using the UPENN-GBM dataset, where the AE achieved a mean squared error (MSE) of 0.0016. Additionally, nucleus segmentation is assessed on the MoNuSeg dataset, outperforming all unsupervised approaches with an F1 score of 77.46% and a Jaccard score of 63.35%. These results demonstrate the efficacy of TUMLS in advancing the field of digital pathology.
comment: 32 pages, 15 figures, 3 tables, 42 references
☆ Post-pre-training for Modality Alignment in Vision-Language Foundation Models CVPR 2025
Contrastive language image pre-training (CLIP) is an essential component of building modern vision-language foundation models. While CLIP demonstrates remarkable zero-shot performance on downstream tasks, the multi-modal feature spaces still suffer from a modality gap, which is a gap between image and text feature clusters and limits downstream task performance. Although existing works attempt to address the modality gap by modifying pre-training or fine-tuning, they struggle with heavy training costs with large datasets or degradations of zero-shot performance. This paper presents CLIP-Refine, a post-pre-training method for CLIP models at a phase between pre-training and fine-tuning. CLIP-Refine aims to align the feature space with 1 epoch training on small image-text datasets without zero-shot performance degradations. To this end, we introduce two techniques: random feature alignment (RaFA) and hybrid contrastive-distillation (HyCD). RaFA aligns the image and text features to follow a shared prior distribution by minimizing the distance to random reference vectors sampled from the prior. HyCD updates the model with hybrid soft labels generated by combining ground-truth image-text pair labels and outputs from the pre-trained CLIP model. This contributes to achieving both maintaining the past knowledge and learning new knowledge to align features. Our extensive experiments with multiple classification and retrieval tasks show that CLIP-Refine succeeds in mitigating the modality gap and improving the zero-shot performance.
comment: Accepted to CVPR 2025; Code: https://github.com/yshinya6/clip-refine
☆ NTIRE 2025 Challenge on Day and Night Raindrop Removal for Dual-Focused Images: Methods and Results CVPR
This paper reviews the NTIRE 2025 Challenge on Day and Night Raindrop Removal for Dual-Focused Images. This challenge received a wide range of impressive solutions, which are developed and evaluated using our collected real-world Raindrop Clarity dataset. Unlike existing deraining datasets, our Raindrop Clarity dataset is more diverse and challenging in degradation types and contents, which includes day raindrop-focused, day background-focused, night raindrop-focused, and night background-focused degradations. This dataset is divided into three subsets for competition: 14,139 images for training, 240 images for validation, and 731 images for testing. The primary objective of this challenge is to establish a new and powerful benchmark for the task of removing raindrops under varying lighting and focus conditions. There are a total of 361 participants in the competition, and 32 teams submitting valid solutions and fact sheets for the final testing phase. These submissions achieved state-of-the-art (SOTA) performance on the Raindrop Clarity dataset. The project can be found at https://lixinustc.github.io/CVPR-NTIRE2025-RainDrop-Competition.github.io/.
comment: Challenge Report of CVPR NTIRE 2025; 26 pages; Methods from 32 teams
Self-Supervised Pre-training with Combined Datasets for 3D Perception in Autonomous Driving
The significant achievements of pre-trained models leveraging large volumes of data in the field of NLP and 2D vision inspire us to explore the potential of extensive data pre-training for 3D perception in autonomous driving. Toward this goal, this paper proposes to utilize massive unlabeled data from heterogeneous datasets to pre-train 3D perception models. We introduce a self-supervised pre-training framework that learns effective 3D representations from scratch on unlabeled data, combined with a prompt adapter based domain adaptation strategy to reduce dataset bias. The approach significantly improves model performance on downstream tasks such as 3D object detection, BEV segmentation, 3D object tracking, and occupancy prediction, and shows steady performance increase as the training data volume scales up, demonstrating the potential of continually benefit 3D perception models for autonomous driving. We will release the source code to inspire further investigations in the community.
☆ SmartFreeEdit: Mask-Free Spatial-Aware Image Editing with Complex Instruction Understanding
Recent advancements in image editing have utilized large-scale multimodal models to enable intuitive, natural instruction-driven interactions. However, conventional methods still face significant challenges, particularly in spatial reasoning, precise region segmentation, and maintaining semantic consistency, especially in complex scenes. To overcome these challenges, we introduce SmartFreeEdit, a novel end-to-end framework that integrates a multimodal large language model (MLLM) with a hypergraph-enhanced inpainting architecture, enabling precise, mask-free image editing guided exclusively by natural language instructions. The key innovations of SmartFreeEdit include:(1)the introduction of region aware tokens and a mask embedding paradigm that enhance the spatial understanding of complex scenes;(2) a reasoning segmentation pipeline designed to optimize the generation of editing masks based on natural language instructions;and (3) a hypergraph-augmented inpainting module that ensures the preservation of both structural integrity and semantic coherence during complex edits, overcoming the limitations of local-based image generation. Extensive experiments on the Reason-Edit benchmark demonstrate that SmartFreeEdit surpasses current state-of-the-art methods across multiple evaluation metrics, including segmentation accuracy, instruction adherence, and visual quality preservation, while addressing the issue of local information focus and improving global consistency in the edited image. Our project will be available at https://github.com/smileformylove/SmartFreeEdit.
☆ Unsupervised Cross-Domain 3D Human Pose Estimation via Pseudo-Label-Guided Global Transforms
Existing 3D human pose estimation methods often suffer in performance, when applied to cross-scenario inference, due to domain shifts in characteristics such as camera viewpoint, position, posture, and body size. Among these factors, camera viewpoints and locations {have been shown} to contribute significantly to the domain gap by influencing the global positions of human poses. To address this, we propose a novel framework that explicitly conducts global transformations between pose positions in the camera coordinate systems of source and target domains. We start with a Pseudo-Label Generation Module that is applied to the 2D poses of the target dataset to generate pseudo-3D poses. Then, a Global Transformation Module leverages a human-centered coordinate system as a novel bridging mechanism to seamlessly align the positional orientations of poses across disparate domains, ensuring consistent spatial referencing. To further enhance generalization, a Pose Augmentor is incorporated to address variations in human posture and body size. This process is iterative, allowing refined pseudo-labels to progressively improve guidance for domain adaptation. Our method is evaluated on various cross-dataset benchmarks, including Human3.6M, MPI-INF-3DHP, and 3DPW. The proposed method outperforms state-of-the-art approaches and even outperforms the target-trained model.
comment: 11 pages, 6 figures, including appendix. This work has been submitted to the IEEE for possible publication
☆ Collaborative Perception Datasets for Autonomous Driving: A Review
Collaborative perception has attracted growing interest from academia and industry due to its potential to enhance perception accuracy, safety, and robustness in autonomous driving through multi-agent information fusion. With the advancement of Vehicle-to-Everything (V2X) communication, numerous collaborative perception datasets have emerged, varying in cooperation paradigms, sensor configurations, data sources, and application scenarios. However, the absence of systematic summarization and comparative analysis hinders effective resource utilization and standardization of model evaluation. As the first comprehensive review focused on collaborative perception datasets, this work reviews and compares existing resources from a multi-dimensional perspective. We categorize datasets based on cooperation paradigms, examine their data sources and scenarios, and analyze sensor modalities and supported tasks. A detailed comparative analysis is conducted across multiple dimensions. We also outline key challenges and future directions, including dataset scalability, diversity, domain adaptation, standardization, privacy, and the integration of large language models. To support ongoing research, we provide a continuously updated online repository of collaborative perception datasets and related literature: https://github.com/frankwnb/Collaborative-Perception-Datasets-for-Autonomous-Driving.
comment: 18pages, 7figures, journal
☆ HSS-IAD: A Heterogeneous Same-Sort Industrial Anomaly Detection Dataset ICME 2025
Multi-class Unsupervised Anomaly Detection algorithms (MUAD) are receiving increasing attention due to their relatively low deployment costs and improved training efficiency. However, the real-world effectiveness of MUAD methods is questioned due to limitations in current Industrial Anomaly Detection (IAD) datasets. These datasets contain numerous classes that are unlikely to be produced by the same factory and fail to cover multiple structures or appearances. Additionally, the defects do not reflect real-world characteristics. Therefore, we introduce the Heterogeneous Same-Sort Industrial Anomaly Detection (HSS-IAD) dataset, which contains 8,580 images of metallic-like industrial parts and precise anomaly annotations. These parts exhibit variations in structure and appearance, with subtle defects that closely resemble the base materials. We also provide foreground images for synthetic anomaly generation. Finally, we evaluate popular IAD methods on this dataset under multi-class and class-separated settings, demonstrating its potential to bridge the gap between existing datasets and real factory conditions. The dataset is available at https://github.com/Qiqigeww/HSS-IAD-Dataset.
comment: Accepted to IEEE ICME 2025
☆ SOPHY: Generating Simulation-Ready Objects with Physical Materials
We present SOPHY, a generative model for 3D physics-aware shape synthesis. Unlike existing 3D generative models that focus solely on static geometry or 4D models that produce physics-agnostic animations, our approach jointly synthesizes shape, texture, and material properties related to physics-grounded dynamics, making the generated objects ready for simulations and interactive, dynamic environments. To train our model, we introduce a dataset of 3D objects annotated with detailed physical material attributes, along with an annotation pipeline for efficient material annotation. Our method enables applications such as text-driven generation of interactive, physics-aware 3D objects and single-image reconstruction of physically plausible shapes. Furthermore, our experiments demonstrate that jointly modeling shape and material properties enhances the realism and fidelity of generated shapes, improving performance on generative geometry evaluation metrics.
comment: Project page: https://xjay18.github.io/SOPHY
☆ Embodied-R: Collaborative Framework for Activating Embodied Spatial Reasoning in Foundation Models via Reinforcement Learning
Humans can perceive and reason about spatial relationships from sequential visual observations, such as egocentric video streams. However, how pretrained models acquire such abilities, especially high-level reasoning, remains unclear. This paper introduces Embodied-R, a collaborative framework combining large-scale Vision-Language Models (VLMs) for perception and small-scale Language Models (LMs) for reasoning. Using Reinforcement Learning (RL) with a novel reward system considering think-answer logical consistency, the model achieves slow-thinking capabilities with limited computational resources. After training on only 5k embodied video samples, Embodied-R with a 3B LM matches state-of-the-art multimodal reasoning models (OpenAI-o1, Gemini-2.5-pro) on both in-distribution and out-of-distribution embodied spatial reasoning tasks. Embodied-R also exhibits emergent thinking patterns such as systematic analysis and contextual integration. We further explore research questions including response length, training on VLM, strategies for reward design, and differences in model generalization after SFT (Supervised Fine-Tuning) and RL training.
comment: 12 pages, 5 figures
☆ TongUI: Building Generalized GUI Agents by Learning from Multimodal Web Tutorials
Building Graphical User Interface (GUI) agents is a promising research direction, which simulates human interaction with computers or mobile phones to perform diverse GUI tasks. However, a major challenge in developing generalized GUI agents is the lack of sufficient trajectory data across various operating systems and applications, mainly due to the high cost of manual annotations. In this paper, we propose the TongUI framework that builds generalized GUI agents by learning from rich multimodal web tutorials. Concretely, we crawl and process online GUI tutorials (such as videos and articles) into GUI agent trajectory data, through which we produce the GUI-Net dataset containing 143K trajectory data across five operating systems and more than 200 applications. We develop the TongUI agent by fine-tuning Qwen2.5-VL-3B/7B models on GUI-Net, which show remarkable performance improvements on commonly used grounding and navigation benchmarks, outperforming baseline agents about 10\% on multiple benchmarks, showing the effectiveness of the GUI-Net dataset and underscoring the significance of our TongUI framework. We will fully open-source the code, the GUI-Net dataset, and the trained models soon.
☆ Accurate Tracking of Arabidopsis Root Cortex Cell Nuclei in 3D Time-Lapse Microscopy Images Based on Genetic Algorithm
Arabidopsis is a widely used model plant to gain basic knowledge on plant physiology and development. Live imaging is an important technique to visualize and quantify elemental processes in plant development. To uncover novel theories underlying plant growth and cell division, accurate cell tracking on live imaging is of utmost importance. The commonly used cell tracking software, TrackMate, adopts tracking-by-detection fashion, which applies Laplacian of Gaussian (LoG) for blob detection, and Linear Assignment Problem (LAP) tracker for tracking. However, they do not perform sufficiently when cells are densely arranged. To alleviate the problems mentioned above, we propose an accurate tracking method based on Genetic algorithm (GA) using knowledge of Arabidopsis root cellular patterns and spatial relationship among volumes. Our method can be described as a coarse-to-fine method, in which we first conducted relatively easy line-level tracking of cell nuclei, then performed complicated nuclear tracking based on known linear arrangement of cell files and their spatial relationship between nuclei. Our method has been evaluated on a long-time live imaging dataset of Arabidopsis root tips, and with minor manual rectification, it accurately tracks nuclei. To the best of our knowledge, this research represents the first successful attempt to address a long-standing problem in the field of time-lapse microscopy in the root meristem by proposing an accurate tracking method for Arabidopsis root nuclei.
☆ Two Tasks, One Goal: Uniting Motion and Planning for Excellent End To End Autonomous Driving Performance
End-to-end autonomous driving has made impressive progress in recent years. Former end-to-end autonomous driving approaches often decouple planning and motion tasks, treating them as separate modules. This separation overlooks the potential benefits that planning can gain from learning out-of-distribution data encountered in motion tasks. However, unifying these tasks poses significant challenges, such as constructing shared contextual representations and handling the unobservability of other vehicles' states. To address these challenges, we propose TTOG, a novel two-stage trajectory generation framework. In the first stage, a diverse set of trajectory candidates is generated, while the second stage focuses on refining these candidates through vehicle state information. To mitigate the issue of unavailable surrounding vehicle states, TTOG employs a self-vehicle data-trained state estimator, subsequently extended to other vehicles. Furthermore, we introduce ECSA (equivariant context-sharing scene adapter) to enhance the generalization of scene representations across different agents. Experimental results demonstrate that TTOG achieves state-of-the-art performance across both planning and motion tasks. Notably, on the challenging open-loop nuScenes dataset, TTOG reduces the L2 distance by 36.06\%. Furthermore, on the closed-loop Bench2Drive dataset, our approach achieves a 22\% improvement in the driving score (DS), significantly outperforming existing baselines.
☆ VLMGuard-R1: Proactive Safety Alignment for VLMs via Reasoning-Driven Prompt Optimization
Aligning Vision-Language Models (VLMs) with safety standards is essential to mitigate risks arising from their multimodal complexity, where integrating vision and language unveils subtle threats beyond the reach of conventional safeguards. Inspired by the insight that reasoning across modalities is key to preempting intricate vulnerabilities, we propose a novel direction for VLM safety: multimodal reasoning-driven prompt rewriting. To this end, we introduce VLMGuard-R1, a proactive framework that refines user inputs through a reasoning-guided rewriter, dynamically interpreting text-image interactions to deliver refined prompts that bolster safety across diverse VLM architectures without altering their core parameters. To achieve this, we devise a three-stage reasoning pipeline to synthesize a dataset that trains the rewriter to infer subtle threats, enabling tailored, actionable responses over generic refusals. Extensive experiments across three benchmarks with five VLMs reveal that VLMGuard-R1 outperforms four baselines. In particular, VLMGuard-R1 achieves a remarkable 43.59\% increase in average safety across five models on the SIUO benchmark.
☆ AdaptoVision: A Multi-Resolution Image Recognition Model for Robust and Scalable Classification
This paper introduces AdaptoVision, a novel convolutional neural network (CNN) architecture designed to efficiently balance computational complexity and classification accuracy. By leveraging enhanced residual units, depth-wise separable convolutions, and hierarchical skip connections, AdaptoVision significantly reduces parameter count and computational requirements while preserving competitive performance across various benchmark and medical image datasets. Extensive experimentation demonstrates that AdaptoVision achieves state-of-the-art on BreakHis dataset and comparable accuracy levels, notably 95.3\% on CIFAR-10 and 85.77\% on CIFAR-100, without relying on any pretrained weights. The model's streamlined architecture and strategic simplifications promote effective feature extraction and robust generalization, making it particularly suitable for deployment in real-time and resource-constrained environments.
☆ Quantum Computing Supported Adversarial Attack-Resilient Autonomous Vehicle Perception Module for Traffic Sign Classification
Deep learning (DL)-based image classification models are essential for autonomous vehicle (AV) perception modules since incorrect categorization might have severe repercussions. Adversarial attacks are widely studied cyberattacks that can lead DL models to predict inaccurate output, such as incorrectly classified traffic signs by the perception module of an autonomous vehicle. In this study, we create and compare hybrid classical-quantum deep learning (HCQ-DL) models with classical deep learning (C-DL) models to demonstrate robustness against adversarial attacks for perception modules. Before feeding them into the quantum system, we used transfer learning models, alexnet and vgg-16, as feature extractors. We tested over 1000 quantum circuits in our HCQ-DL models for projected gradient descent (PGD), fast gradient sign attack (FGSA), and gradient attack (GA), which are three well-known untargeted adversarial approaches. We evaluated the performance of all models during adversarial attacks and no-attack scenarios. Our HCQ-DL models maintain accuracy above 95\% during a no-attack scenario and above 91\% for GA and FGSA attacks, which is higher than C-DL models. During the PGD attack, our alexnet-based HCQ-DL model maintained an accuracy of 85\% compared to C-DL models that achieved accuracies below 21\%. Our results highlight that the HCQ-DL models provide improved accuracy for traffic sign classification under adversarial settings compared to their classical counterparts.
☆ RoPETR: Improving Temporal Camera-Only 3D Detection by Integrating Enhanced Rotary Position Embedding
This technical report introduces a targeted improvement to the StreamPETR framework, specifically aimed at enhancing velocity estimation, a critical factor influencing the overall NuScenes Detection Score. While StreamPETR exhibits strong 3D bounding box detection performance as reflected by its high mean Average Precision our analysis identified velocity estimation as a substantial bottleneck when evaluated on the NuScenes dataset. To overcome this limitation, we propose a customized positional embedding strategy tailored to enhance temporal modeling capabilities. Experimental evaluations conducted on the NuScenes test set demonstrate that our improved approach achieves a state-of-the-art NDS of 70.86% using the ViT-L backbone, setting a new benchmark for camera-only 3D object detection.
☆ Packing Input Frame Context in Next-Frame Prediction Models for Video Generation
We present a neural network structure, FramePack, to train next-frame (or next-frame-section) prediction models for video generation. The FramePack compresses input frames to make the transformer context length a fixed number regardless of the video length. As a result, we are able to process a large number of frames using video diffusion with computation bottleneck similar to image diffusion. This also makes the training video batch sizes significantly higher (batch sizes become comparable to image diffusion training). We also propose an anti-drifting sampling method that generates frames in inverted temporal order with early-established endpoints to avoid exposure bias (error accumulation over iterations). Finally, we show that existing video diffusion models can be finetuned with FramePack, and their visual quality may be improved because the next-frame prediction supports more balanced diffusion schedulers with less extreme flow shift timesteps.
comment: https://github.com/lllyasviel/FramePack
☆ SAM-Based Building Change Detection with Distribution-Aware Fourier Adaptation and Edge-Constrained Warping
Building change detection remains challenging for urban development, disaster assessment, and military reconnaissance. While foundation models like Segment Anything Model (SAM) show strong segmentation capabilities, SAM is limited in the task of building change detection due to domain gap issues. Existing adapter-based fine-tuning approaches face challenges with imbalanced building distribution, resulting in poor detection of subtle changes and inaccurate edge extraction. Additionally, bi-temporal misalignment in change detection, typically addressed by optical flow, remains vulnerable to background noises. This affects the detection of building changes and compromises both detection accuracy and edge recognition. To tackle these challenges, we propose a new SAM-Based Network with Distribution-Aware Fourier Adaptation and Edge-Constrained Warping (FAEWNet) for building change detection. FAEWNet utilizes the SAM encoder to extract rich visual features from remote sensing images. To guide SAM in focusing on specific ground objects in remote sensing scenes, we propose a Distribution-Aware Fourier Aggregated Adapter to aggregate task-oriented changed information. This adapter not only effectively addresses the domain gap issue, but also pays attention to the distribution of changed buildings. Furthermore, to mitigate noise interference and misalignment in height offset estimation, we design a novel flow module that refines building edge extraction and enhances the perception of changed buildings. Our state-of-the-art results on the LEVIR-CD, S2Looking and WHU-CD datasets highlight the effectiveness of FAEWNet. The code is available at https://github.com/SUPERMAN123000/FAEWNet.
☆ Robo-SGG: Exploiting Layout-Oriented Normalization and Restitution for Robust Scene Graph Generation
In this paper, we introduce a novel method named Robo-SGG, i.e., Layout-Oriented Normalization and Restitution for Robust Scene Graph Generation. Compared to the existing SGG setting, the robust scene graph generation aims to perform inference on a diverse range of corrupted images, with the core challenge being the domain shift between the clean and corrupted images. Existing SGG methods suffer from degraded performance due to compromised visual features e.g., corruption interference or occlusions. To obtain robust visual features, we exploit the layout information, which is domain-invariant, to enhance the efficacy of existing SGG methods on corrupted images. Specifically, we employ Instance Normalization(IN) to filter out the domain-specific feature and recover the unchangeable structural features, i.e., the positional and semantic relationships among objects by the proposed Layout-Oriented Restitution. Additionally, we propose a Layout-Embedded Encoder (LEE) that augments the existing object and predicate encoders within the SGG framework, enriching the robust positional and semantic features of objects and predicates. Note that our proposed Robo-SGG module is designed as a plug-and-play component, which can be easily integrated into any baseline SGG model. Extensive experiments demonstrate that by integrating the state-of-the-art method into our proposed Robo-SGG, we achieve relative improvements of 5.6%, 8.0%, and 6.5% in mR@50 for PredCls, SGCls, and SGDet tasks on the VG-C dataset, respectively, and achieve new state-of-the-art performance in corruption scene graph generation benchmark (VG-C and GQA-C). We will release our source code and model.
☆ AdaQual-Diff: Diffusion-Based Image Restoration via Adaptive Quality Prompting
Restoring images afflicted by complex real-world degradations remains challenging, as conventional methods often fail to adapt to the unique mixture and severity of artifacts present. This stems from a reliance on indirect cues which poorly capture the true perceptual quality deficit. To address this fundamental limitation, we introduce AdaQual-Diff, a diffusion-based framework that integrates perceptual quality assessment directly into the generative restoration process. Our approach establishes a mathematical relationship between regional quality scores from DeQAScore and optimal guidance complexity, implemented through an Adaptive Quality Prompting mechanism. This mechanism systematically modulates prompt structure according to measured degradation severity: regions with lower perceptual quality receive computationally intensive, structurally complex prompts with precise restoration directives, while higher quality regions receive minimal prompts focused on preservation rather than intervention. The technical core of our method lies in the dynamic allocation of computational resources proportional to degradation severity, creating a spatially-varying guidance field that directs the diffusion process with mathematical precision. By combining this quality-guided approach with content-specific conditioning, our framework achieves fine-grained control over regional restoration intensity without requiring additional parameters or inference iterations. Experimental results demonstrate that AdaQual-Diff achieves visually superior restorations across diverse synthetic and real-world datasets.
☆ 3DResT: A Strong Baseline for Semi-Supervised 3D Referring Expression Segmentation
3D Referring Expression Segmentation (3D-RES) typically requires extensive instance-level annotations, which are time-consuming and costly. Semi-supervised learning (SSL) mitigates this by using limited labeled data alongside abundant unlabeled data, improving performance while reducing annotation costs. SSL uses a teacher-student paradigm where teacher generates high-confidence-filtered pseudo-labels to guide student. However, in the context of 3D-RES, where each label corresponds to a single mask and labeled data is scarce, existing SSL methods treat high-quality pseudo-labels merely as auxiliary supervision, which limits the model's learning potential. The reliance on high-confidence thresholds for filtering often results in potentially valuable pseudo-labels being discarded, restricting the model's ability to leverage the abundant unlabeled data. Therefore, we identify two critical challenges in semi-supervised 3D-RES, namely, inefficient utilization of high-quality pseudo-labels and wastage of useful information from low-quality pseudo-labels. In this paper, we introduce the first semi-supervised learning framework for 3D-RES, presenting a robust baseline method named 3DResT. To address these challenges, we propose two novel designs called Teacher-Student Consistency-Based Sampling (TSCS) and Quality-Driven Dynamic Weighting (QDW). TSCS aids in the selection of high-quality pseudo-labels, integrating them into the labeled dataset to strengthen the labeled supervision signals. QDW preserves low-quality pseudo-labels by dynamically assigning them lower weights, allowing for the effective extraction of useful information rather than discarding them. Extensive experiments conducted on the widely used benchmark demonstrate the effectiveness of our method. Notably, with only 1% labeled data, 3DResT achieves an mIoU improvement of 8.34 points compared to the fully supervised method.
☆ CM3AE: A Unified RGB Frame and Event-Voxel/-Frame Pre-training Framework
Event cameras have attracted increasing attention in recent years due to their advantages in high dynamic range, high temporal resolution, low power consumption, and low latency. Some researchers have begun exploring pre-training directly on event data. Nevertheless, these efforts often fail to establish strong connections with RGB frames, limiting their applicability in multi-modal fusion scenarios. To address these issues, we propose a novel CM3AE pre-training framework for the RGB-Event perception. This framework accepts multi-modalities/views of data as input, including RGB images, event images, and event voxels, providing robust support for both event-based and RGB-event fusion based downstream tasks. Specifically, we design a multi-modal fusion reconstruction module that reconstructs the original image from fused multi-modal features, explicitly enhancing the model's ability to aggregate cross-modal complementary information. Additionally, we employ a multi-modal contrastive learning strategy to align cross-modal feature representations in a shared latent space, which effectively enhances the model's capability for multi-modal understanding and capturing global dependencies. We construct a large-scale dataset containing 2,535,759 RGB-Event data pairs for the pre-training. Extensive experiments on five downstream tasks fully demonstrated the effectiveness of CM3AE. Source code and pre-trained models will be released on https://github.com/Event-AHU/CM3AE.
♻ ☆ Do Vision-Language Models Represent Space and How? Evaluating Spatial Frame of Reference Under Ambiguities ICLR 2025
Spatial expressions in situated communication can be ambiguous, as their meanings vary depending on the frames of reference (FoR) adopted by speakers and listeners. While spatial language understanding and reasoning by vision-language models (VLMs) have gained increasing attention, potential ambiguities in these models are still under-explored. To address this issue, we present the COnsistent Multilingual Frame Of Reference Test (COMFORT), an evaluation protocol to systematically assess the spatial reasoning capabilities of VLMs. We evaluate nine state-of-the-art VLMs using COMFORT. Despite showing some alignment with English conventions in resolving ambiguities, our experiments reveal significant shortcomings of VLMs: notably, the models (1) exhibit poor robustness and consistency, (2) lack the flexibility to accommodate multiple FoRs, and (3) fail to adhere to language-specific or culture-specific conventions in cross-lingual tests, as English tends to dominate other languages. With a growing effort to align vision-language models with human cognitive intuitions, we call for more attention to the ambiguous nature and cross-cultural diversity of spatial reasoning.
comment: Accepted to ICLR 2025 (Oral) | Project page: https://spatial-comfort.github.io/
♻ ☆ Multimodal LLMs Can Reason about Aesthetics in Zero-Shot
The rapid progress of generative art has democratized the creation of visually pleasing imagery. However, achieving genuine artistic impact - the kind that resonates with viewers on a deeper, more meaningful level - requires a sophisticated aesthetic sensibility. This sensibility involves a multi-faceted reasoning process extending beyond mere visual appeal, which is often overlooked by current computational models. This paper pioneers an approach to capture this complex process by investigating how the reasoning capabilities of Multimodal LLMs (MLLMs) can be effectively elicited for aesthetic judgment. Our analysis reveals a critical challenge: MLLMs exhibit a tendency towards hallucinations during aesthetic reasoning, characterized by subjective opinions and unsubstantiated artistic interpretations. We further demonstrate that these limitations can be overcome by employing an evidence-based, objective reasoning process, as substantiated by our proposed baseline, ArtCoT. MLLMs prompted by this principle produce multi-faceted and in-depth aesthetic reasoning that aligns significantly better with human judgment. These findings have direct applications in areas such as AI art tutoring and as reward models for generative art. Ultimately, our work paves the way for AI systems that can truly understand, appreciate, and generate artworks that align with the sensible human aesthetic standard.
comment: WIP, Homepage https://github.com/songrise/MLLM4Art
♻ ☆ Benchmarking the Spatial Robustness of DNNs via Natural and Adversarial Localized Corruptions
The robustness of DNNs is a crucial factor in safety-critical applications, particularly in complex and dynamic environments where localized corruptions can arise. While previous studies have evaluated the robustness of semantic segmentation (SS) models under whole-image natural or adversarial corruptions, a comprehensive investigation into the spatial robustness of dense vision models under localized corruptions remained underexplored. This paper fills this gap by introducing specialized metrics for benchmarking the spatial robustness of segmentation models, alongside with an evaluation framework to assess the impact of localized corruptions. Furthermore, we uncover the inherent complexity of characterizing worst-case robustness using a single localized adversarial perturbation. To address this, we propose region-aware multi-attack adversarial analysis, a method that enables a deeper understanding of model robustness against adversarial perturbations applied to specific regions. The proposed metrics and analysis were exploited to evaluate 14 segmentation models in driving scenarios, uncovering key insights into the effects of localized corruption in both natural and adversarial forms. The results reveal that models respond to these two types of threats differently; for instance, transformer-based segmentation models demonstrate notable robustness to localized natural corruptions but are highly vulnerable to adversarial ones and vice-versa for CNN-based models. Consequently, we also address the challenge of balancing robustness to both natural and adversarial localized corruptions by means of ensemble models, thereby achieving a broader threat coverage and improved reliability for dense vision tasks.
comment: Under review
♻ ☆ CDXLSTM: Boosting Remote Sensing Change Detection with Extended Long Short-Term Memory
In complex scenes and varied conditions, effectively integrating spatial-temporal context is crucial for accurately identifying changes. However, current RS-CD methods lack a balanced consideration of performance and efficiency. CNNs lack global context, Transformers are computationally expensive, and Mambas face CUDA dependence and local correlation loss. In this paper, we propose CDXLSTM, with a core component that is a powerful XLSTM-based feature enhancement layer, integrating the advantages of linear computational complexity, global context perception, and strong interpret-ability. Specifically, we introduce a scale-specific Feature Enhancer layer, incorporating a Cross-Temporal Global Perceptron customized for semantic-accurate deep features, and a Cross-Temporal Spatial Refiner customized for detail-rich shallow features. Additionally, we propose a Cross-Scale Interactive Fusion module to progressively interact global change representations with spatial responses. Extensive experimental results demonstrate that CDXLSTM achieves state-of-the-art performance across three benchmark datasets, offering a compelling balance between efficiency and accuracy. Code is available at https://github.com/xwmaxwma/rschange.
♻ ☆ A Coding-Theoretic Analysis of Hyperspherical Prototypical Learning Geometry
Hyperspherical Prototypical Learning (HPL) is a supervised approach to representation learning that designs class prototypes on the unit hypersphere. The prototypes bias the representations to class separation in a scale invariant and known geometry. Previous approaches to HPL have either of the following shortcomings: (i) they follow an unprincipled optimisation procedure; or (ii) they are theoretically sound, but are constrained to only one possible latent dimension. In this paper, we address both shortcomings. To address (i), we present a principled optimisation procedure whose solution we show is optimal. To address (ii), we construct well-separated prototypes in a wide range of dimensions using linear block codes. Additionally, we give a full characterisation of the optimal prototype placement in terms of achievable and converse bounds, showing that our proposed methods are near-optimal.
comment: Changes in version 2: Minor formatting changes. Published in the Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM), PMLR 251. Available at: https://proceedings.mlr.press/v251/lindstrom24a.html 14 pages: 9 of the main paper, 2 of references, and 3 of appendices.. Code is available at: https://github.com/martinlindstrom/coding_theoretic_hpl
♻ ☆ A Survey and Evaluation of Adversarial Attacks for Object Detection
Deep learning models achieve remarkable accuracy in computer vision tasks, yet remain vulnerable to adversarial examples--carefully crafted perturbations to input images that can deceive these models into making confident but incorrect predictions. This vulnerability pose significant risks in high-stakes applications such as autonomous vehicles, security surveillance, and safety-critical inspection systems. While the existing literature extensively covers adversarial attacks in image classification, comprehensive analyses of such attacks on object detection systems remain limited. This paper presents a novel taxonomic framework for categorizing adversarial attacks specific to object detection architectures, synthesizes existing robustness metrics, and provides a comprehensive empirical evaluation of state-of-the-art attack methodologies on popular object detection models, including both traditional detectors and modern detectors with vision-language pretraining. Through rigorous analysis of open-source attack implementations and their effectiveness across diverse detection architectures, we derive key insights into attack characteristics. Furthermore, we delineate critical research gaps and emerging challenges to guide future investigations in securing object detection systems against adversarial threats. Our findings establish a foundation for developing more robust detection models while highlighting the urgent need for standardized evaluation protocols in this rapidly evolving domain.
comment: Accepted for publication in the IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
♻ ☆ DC-SAM: In-Context Segment Anything in Images and Videos via Dual Consistency
Given a single labeled example, in-context segmentation aims to segment corresponding objects. This setting, known as one-shot segmentation in few-shot learning, explores the segmentation model's generalization ability and has been applied to various vision tasks, including scene understanding and image/video editing. While recent Segment Anything Models have achieved state-of-the-art results in interactive segmentation, these approaches are not directly applicable to in-context segmentation. In this work, we propose the Dual Consistency SAM (DC-SAM) method based on prompt-tuning to adapt SAM and SAM2 for in-context segmentation of both images and videos. Our key insights are to enhance the features of the SAM's prompt encoder in segmentation by providing high-quality visual prompts. When generating a mask prior, we fuse the SAM features to better align the prompt encoder. Then, we design a cycle-consistent cross-attention on fused features and initial visual prompts. Next, a dual-branch design is provided by using the discriminative positive and negative prompts in the prompt encoder. Furthermore, we design a simple mask-tube training strategy to adopt our proposed dual consistency method into the mask tube. Although the proposed DC-SAM is primarily designed for images, it can be seamlessly extended to the video domain with the support of SAM2. Given the absence of in-context segmentation in the video domain, we manually curate and construct the first benchmark from existing video segmentation datasets, named In-Context Video Object Segmentation (IC-VOS), to better assess the in-context capability of the model. Extensive experiments demonstrate that our method achieves 55.5 (+1.4) mIoU on COCO-20i, 73.0 (+1.1) mIoU on PASCAL-5i, and a J&F score of 71.52 on the proposed IC-VOS benchmark. Our source code and benchmark are available at https://github.com/zaplm/DC-SAM.
comment: V1 has been withdrawn due to a template issue, because of the arXiv policy, we can't delete it. Please refer to the newest version v2
♻ ☆ Rethinking Few-Shot Image Fusion: Granular Ball Priors Enable General-Purpose Deep Fusion
In image fusion tasks, the absence of real fused images as priors presents a fundamental challenge. Most deep learning-based fusion methods rely on large-scale paired datasets to extract global weighting features from raw images, thereby generating fused outputs that approximate real fused images. In contrast to previous studies, this paper explores few-shot training of neural networks under the condition of having prior knowledge. We propose a novel fusion framework named GBFF, and a Granular Ball Significant Extraction algorithm specifically designed for the few-shot prior setting. All pixel pairs involved in the fusion process are initially modeled as a Coarse-Grained Granular Ball. At the local level, Fine-Grained Granular Balls are used to slide through the brightness space to extract Non-Salient Pixel Pairs, and perform splitting operations to obtain Salient Pixel Pairs. Pixel-wise weights are then computed to generate a pseudo-supervised image. At the global level, pixel pairs with significant contributions to the fusion process are categorized into the Positive Region, while those whose contributions cannot be accurately determined are assigned to the Boundary Region. The Granular Ball performs modality-aware adaptation based on the proportion of the positive region, thereby adjusting the neural network's loss function and enabling it to complement the information of the boundary region. Extensive experiments demonstrate the effectiveness of both the proposed algorithm and the underlying theory. Compared with state-of-the-art (SOTA) methods, our approach shows strong competitiveness in terms of both fusion time and image expressiveness. Our code is publicly available at:
♻ ☆ Long-Context Autoregressive Video Modeling with Next-Frame Prediction
Long-context autoregressive modeling has significantly advanced language generation, but video generation still struggles to fully utilize extended temporal contexts. To investigate long-context video modeling, we introduce Frame AutoRegressive (FAR), a strong baseline for video autoregressive modeling. Just as language models learn causal dependencies between tokens (i.e., Token AR), FAR models temporal causal dependencies between continuous frames, achieving better convergence than Token AR and video diffusion transformers. Building on FAR, we observe that long-context video modeling faces challenges due to visual redundancy. Training on long videos is computationally expensive, as vision tokens grow much faster than language tokens. To tackle this issue, we propose balancing locality and long-range dependency through long short-term context modeling. A high-resolution short-term context window ensures fine-grained temporal consistency, while an unlimited long-term context window encodes long-range information using fewer tokens. With this approach, we can train on long video sequences with a manageable token context length, thereby significantly reducing training time and memory usage. Furthermore, we propose a multi-level KV cache designed to support the long short-term context modeling, which accelerating inference on long video sequences. We demonstrate that FAR achieves state-of-the-art performance in both short- and long-video generation, providing a simple yet effective baseline for video autoregressive modeling. The code is released at https://github.com/showlab/FAR.
comment: Project page at https://farlongctx.github.io/
♻ ☆ Multimodal Fake News Video Explanation: Dataset, Analysis and Evaluation
Multimodal fake news videos are difficult to interpret because they require comprehensive consideration of the correlation and consistency between multiple modes. Existing methods deal with fake news videos as a classification problem, but it's not clear why news videos are identified as fake. Without proper explanation, the end user may not understand the underlying meaning of the falsehood. Therefore, we propose a new problem - Fake news video Explanation (FNVE) - given a multimodal news post containing a video and title, our goal is to generate natural language explanations to reveal the falsity of the news video. To that end, we developed FakeVE, a new dataset of 2,672 fake news video posts that can definitively explain four real-life fake news video aspects. In order to understand the characteristics of fake news video explanation, we conducted an exploratory analysis of FakeVE from different perspectives. In addition, we propose a Multimodal Relation Graph Transformer (MRGT) based on the architecture of multimodal Transformer to benchmark FakeVE. The empirical results show that the results of the various benchmarks (adopted by FakeVE) are convincing and provide a detailed analysis of the differences in explanation generation of the benchmark models.
♻ ☆ A Robust Prototype-Based Network with Interpretable RBF Classifier Foundations AAAI 2025
Prototype-based classification learning methods are known to be inherently interpretable. However, this paradigm suffers from major limitations compared to deep models, such as lower performance. This led to the development of the so-called deep Prototype-Based Networks (PBNs), also known as prototypical parts models. In this work, we analyze these models with respect to different properties, including interpretability. In particular, we focus on the Classification-by-Components (CBC) approach, which uses a probabilistic model to ensure interpretability and can be used as a shallow or deep architecture. We show that this model has several shortcomings, like creating contradicting explanations. Based on these findings, we propose an extension of CBC that solves these issues. Moreover, we prove that this extension has robustness guarantees and derive a loss that optimizes robustness. Additionally, our analysis shows that most (deep) PBNs are related to (deep) RBF classifiers, which implies that our robustness guarantees generalize to shallow RBF classifiers. The empirical evaluation demonstrates that our deep PBN yields state-of-the-art classification accuracy on different benchmarks while resolving the interpretability shortcomings of other approaches. Further, our shallow PBN variant outperforms other shallow PBNs while being inherently interpretable and exhibiting provable robustness guarantees.
comment: To appear at AAAI 2025. Includes the Appendix of the AAAI submission. In v2, the font size has been increased in some figures. In v3, an incorrect hyperparameter specification (Table 6; $\lambda$) has been corrected
♻ ☆ Beyond the Frame: Generating 360° Panoramic Videos from Perspective Videos
360{\deg} videos have emerged as a promising medium to represent our dynamic visual world. Compared to the "tunnel vision" of standard cameras, their borderless field of view offers a more complete perspective of our surroundings. While existing video models excel at producing standard videos, their ability to generate full panoramic videos remains elusive. In this paper, we investigate the task of video-to-360{\deg} generation: given a perspective video as input, our goal is to generate a full panoramic video that is consistent with the original video. Unlike conventional video generation tasks, the output's field of view is significantly larger, and the model is required to have a deep understanding of both the spatial layout of the scene and the dynamics of objects to maintain spatio-temporal consistency. To address these challenges, we first leverage the abundant 360{\deg} videos available online and develop a high-quality data filtering pipeline to curate pairwise training data. We then carefully design a series of geometry- and motion-aware operations to facilitate the learning process and improve the quality of 360{\deg} video generation. Experimental results demonstrate that our model can generate realistic and coherent 360{\deg} videos from in-the-wild perspective video. In addition, we showcase its potential applications, including video stabilization, camera viewpoint control, and interactive visual question answering.
comment: Project page: https://red-fairy.github.io/argus/
♻ ☆ CAP-Net: A Unified Network for 6D Pose and Size Estimation of Categorical Articulated Parts from a Single RGB-D Image CVPR 2025
This paper tackles category-level pose estimation of articulated objects in robotic manipulation tasks and introduces a new benchmark dataset. While recent methods estimate part poses and sizes at the category level, they often rely on geometric cues and complex multi-stage pipelines that first segment parts from the point cloud, followed by Normalized Part Coordinate Space (NPCS) estimation for 6D poses. These approaches overlook dense semantic cues from RGB images, leading to suboptimal accuracy, particularly for objects with small parts. To address these limitations, we propose a single-stage Network, CAP-Net, for estimating the 6D poses and sizes of Categorical Articulated Parts. This method combines RGB-D features to generate instance segmentation and NPCS representations for each part in an end-to-end manner. CAP-Net uses a unified network to simultaneously predict point-wise class labels, centroid offsets, and NPCS maps. A clustering algorithm then groups points of the same predicted class based on their estimated centroid distances to isolate each part. Finally, the NPCS region of each part is aligned with the point cloud to recover its final pose and size. To bridge the sim-to-real domain gap, we introduce the RGBD-Art dataset, the largest RGB-D articulated dataset to date, featuring photorealistic RGB images and depth noise simulated from real sensors. Experimental evaluations on the RGBD-Art dataset demonstrate that our method significantly outperforms the state-of-the-art approach. Real-world deployments of our model in robotic tasks underscore its robustness and exceptional sim-to-real transfer capabilities, confirming its substantial practical utility. Our dataset, code and pre-trained models are available on the project page.
comment: To appear in CVPR 2025 (Highlight)
♻ ☆ Uncertainty Quantification via Hölder Divergence for Multi-View Representation Learning
Evidence-based deep learning represents a burgeoning paradigm for uncertainty estimation, offering reliable predictions with negligible extra computational overheads. Existing methods usually adopt Kullback-Leibler divergence to estimate the uncertainty of network predictions, ignoring domain gaps among various modalities. To tackle this issue, this paper introduces a novel algorithm based on H\"older Divergence (HD) to enhance the reliability of multi-view learning by addressing inherent uncertainty challenges from incomplete or noisy data. Generally, our method extracts the representations of multiple modalities through parallel network branches, and then employs HD to estimate the prediction uncertainties. Through the Dempster-Shafer theory, integration of uncertainty from different modalities, thereby generating a comprehensive result that considers all available representations. Mathematically, HD proves to better measure the ``distance'' between real data distribution and predictive distribution of the model and improve the performances of multi-class recognition tasks. Specifically, our method surpass the existing state-of-the-art counterparts on all evaluating benchmarks. We further conduct extensive experiments on different backbones to verify our superior robustness. It is demonstrated that our method successfully pushes the corresponding performance boundaries. Finally, we perform experiments on more challenging scenarios, \textit{i.e.}, learning with incomplete or noisy data, revealing that our method exhibits a high tolerance to such corrupted data.
comment: NA
♻ ☆ RGB-Phase Speckle: Cross-Scene Stereo 3D Reconstruction via Wrapped Pre-Normalization ICCV 2025
3D reconstruction garners increasing attention alongside the advancement of high-level image applications, where dense stereo matching (DSM) serves as a pivotal technique. Previous studies often rely on publicly available datasets for training, focusing on modifying network architectures or incorporating specialized modules to extract domain-invariant features and thus improve model robustness. In contrast, inspired by single-frame structured-light phase-shifting encoding, this study introduces RGB-Speckle, a cross-scene 3D reconstruction framework based on an active stereo camera system, designed to enhance robustness. Specifically, we propose a novel phase pre-normalization encoding-decoding method: first, we randomly perturb phase-shift maps and embed them into the three RGB channels to generate color speckle patterns; subsequently, the camera captures phase-encoded images modulated by objects as input to a stereo matching network. This technique effectively mitigates external interference and ensures consistent input data for RGB-Speckle, thereby bolstering cross-domain 3D reconstruction stability. To validate the proposed method, we conduct complex experiments: (1) construct a color speckle dataset for complex scenarios based on the proposed encoding scheme; (2) evaluate the impact of the phase pre-normalization encoding-decoding technique on 3D reconstruction accuracy; and (3) further investigate its robustness across diverse conditions. Experimental results demonstrate that the proposed RGB-Speckle model offers significant advantages in cross-domain and cross-scene 3D reconstruction tasks, enhancing model generalization and reinforcing robustness in challenging environments, thus providing a novel solution for robust 3D reconstruction research.
comment: Submitted to ICCV 2025
♻ ☆ ArtCrafter: Text-Image Aligning Style Transfer via Embedding Reframing
Recent years have witnessed significant advancements in text-guided style transfer, primarily attributed to innovations in diffusion models. These models excel in conditional guidance, utilizing text or images to direct the sampling process. However, despite their capabilities, direct conditional guidance approaches often face challenges in balancing the expressiveness of textual semantics with the diversity of output results while capturing stylistic features. To address these challenges, we introduce ArtCrafter, a novel framework for text-to-image style transfer. Specifically, we introduce an attention-based style extraction module, meticulously engineered to capture the subtle stylistic elements within an image. This module features a multi-layer architecture that leverages the capabilities of perceiver attention mechanisms to integrate fine-grained information. Additionally, we present a novel text-image aligning augmentation component that adeptly balances control over both modalities, enabling the model to efficiently map image and text embeddings into a shared feature space. We achieve this through attention operations that enable smooth information flow between modalities. Lastly, we incorporate an explicit modulation that seamlessly blends multimodal enhanced embeddings with original embeddings through an embedding reframing design, empowering the model to generate diverse outputs. Extensive experiments demonstrate that ArtCrafter yields impressive results in visual stylization, exhibiting exceptional levels of stylistic intensity, controllability, and diversity.
comment: 13 pages, 17 figures, submitted to a journal
♻ ☆ CameraBench: Benchmarking Visual Reasoning in MLLMs via Photography
Large language models (LLMs) and multimodal large language models (MLLMs) have significantly advanced artificial intelligence. However, visual reasoning, reasoning involving both visual and textual inputs, remains underexplored. Recent advancements, including the reasoning models like OpenAI o1 and Gemini 2.0 Flash Thinking, which incorporate image inputs, have opened this capability. In this ongoing work, we focus specifically on photography-related tasks because a photo is a visual snapshot of the physical world where the underlying physics (i.e., illumination, blur extent, etc.) interplay with the camera parameters. Successfully reasoning from the visual information of a photo to identify these numerical camera settings requires the MLLMs to have a deeper understanding of the underlying physics for precise visual comprehension, representing a challenging and intelligent capability essential for practical applications like photography assistant agents. We aim to evaluate MLLMs on their ability to distinguish visual differences related to numerical camera settings, extending a methodology previously proposed for vision-language models (VLMs). Our preliminary results demonstrate the importance of visual reasoning in photography-related tasks. Moreover, these results show that no single MLLM consistently dominates across all evaluation tasks, demonstrating ongoing challenges and opportunities in developing MLLMs with better visual reasoning.
♻ ☆ Feature Calibration enhanced Parameter Synthesis for CLIP-based Class-incremental Learning
Class-Incremental Learning (CIL) enables models to continuously learn new class knowledge while retaining previous classes, facilitating adaptation and evolution in dynamic, real-world environments. Traditional CIL methods primarily rely on visual features, which limits their effectiveness in complex, multimodal scenarios. In contrast, VLMs show promising potential for enhancing CIL by leveraging pre-trained knowledge and integrating multi-modal semantic cues such as text and vision. However, existing approaches struggle to mitigate catastrophic forgetting while preserving the generalization strengths of VLMs across diverse modalities. To address these challenges, we propose a Feature Calibration Enhanced Parameter Synthesis (FCPS) framework. Specifically, FCPS introduces a dynamic parameter adjustment mechanism that iteratively calibrates the contribution of original visual features to the final class decision, thus preserving the model's intrinsic generalization capability across modalities. Simultaneously, parameter integration enables effective knowledge transfer, maintaining a balance between acquiring new class representations and preserving old knowledge. Experimental results on popular benchmarks (e.g., CIFAR100 and ImageNet100) validate the superiority of the proposed method.
♻ ☆ Test-time Alignment of Diffusion Models without Reward Over-optimization ICLR 2025
Diffusion models excel in generative tasks, but aligning them with specific objectives while maintaining their versatility remains challenging. Existing fine-tuning methods often suffer from reward over-optimization, while approximate guidance approaches fail to optimize target rewards effectively. Addressing these limitations, we propose a training-free, test-time method based on Sequential Monte Carlo (SMC) to sample from the reward-aligned target distribution. Our approach, tailored for diffusion sampling and incorporating tempering techniques, achieves comparable or superior target rewards to fine-tuning methods while preserving diversity and cross-reward generalization. We demonstrate its effectiveness in single-reward optimization, multi-objective scenarios, and online black-box optimization. This work offers a robust solution for aligning diffusion models with diverse downstream objectives without compromising their general capabilities. Code is available at https://github.com/krafton-ai/DAS.
comment: ICLR 2025 (Spotlight). The Thirteenth International Conference on Learning Representations. 2025
♻ ☆ VariFace: Fair and Diverse Synthetic Dataset Generation for Face Recognition
The use of large-scale, web-scraped datasets to train face recognition models has raised significant privacy and bias concerns. Synthetic methods mitigate these concerns and provide scalable and controllable face generation to enable fair and accurate face recognition. However, existing synthetic datasets display limited intraclass and interclass diversity and do not match the face recognition performance obtained using real datasets. Here, we propose VariFace, a two-stage diffusion-based pipeline to create fair and diverse synthetic face datasets to train face recognition models. Specifically, we introduce three methods: Face Recognition Consistency to refine demographic labels, Face Vendi Score Guidance to improve interclass diversity, and Divergence Score Conditioning to balance the identity preservation-intraclass diversity trade-off. When constrained to the same dataset size, VariFace considerably outperforms previous synthetic datasets (0.9200 $\rightarrow$ 0.9405) and achieves comparable performance to face recognition models trained with real data (Real Gap = -0.0065). In an unconstrained setting, VariFace not only consistently achieves better performance compared to previous synthetic methods across dataset sizes but also, for the first time, outperforms the real dataset (CASIA-WebFace) across six evaluation datasets. This sets a new state-of-the-art performance with an average face verification accuracy of 0.9567 (Real Gap = +0.0097) across LFW, CFP-FP, CPLFW, AgeDB, and CALFW datasets and 0.9366 (Real Gap = +0.0380) on the RFW dataset.
♻ ☆ QMix: Quality-aware Learning with Mixed Noise for Robust Retinal Disease Diagnosis
Due to the complexity of medical image acquisition and the difficulty of annotation, medical image datasets inevitably contain noise. Noisy data with wrong labels affects the robustness and generalization ability of deep neural networks. Previous noise learning methods mainly considered noise arising from images being mislabeled, i.e. label noise, assuming that all mislabeled images are of high image quality. However, medical images are prone to suffering extreme quality issues, i.e. data noise, where discriminative visual features are missing for disease diagnosis. In this paper, we propose a noise learning framework, termed as QMix, that learns a robust disease diagnosis model under mixed noise. QMix alternates between sample separation and quality-aware semisupervised training in each training epoch. In the sample separation phase, we design a joint uncertainty-loss criterion to effectively separate (1) correctly labeled images; (2) mislabeled images with high quality and (3) mislabeled images with low quality. In the semi-supervised training phase, we train a disease diagnosis model to learn robust feature representation from the separated samples. Specifically, we devise a sample-reweighing loss to mitigate the effect of mislabeled images with low quality during training. Meanwhile, a contrastive enhancement loss is proposed to further distinguish mislabeled images with low quality from correctly labeled images. QMix achieved state-of-the-art disease diagnosis performance on five public retinal image datasets and exhibited substantial improvement on robustness against mixed noise.
comment: Accepted to IEEE Transactions on Medical Imaging
♻ ☆ Unified Domain Adaptive Semantic Segmentation
Unsupervised Domain Adaptive Semantic Segmentation (UDA-SS) aims to transfer the supervision from a labeled source domain to an unlabeled target domain. The majority of existing UDA-SS works typically consider images whilst recent attempts have extended further to tackle videos by modeling the temporal dimension. Although the two lines of research share the major challenges -- overcoming the underlying domain distribution shift, their studies are largely independent, resulting in fragmented insights, a lack of holistic understanding, and missed opportunities for cross-pollination of ideas. This fragmentation prevents the unification of methods, leading to redundant efforts and suboptimal knowledge transfer across image and video domains. Under this observation, we advocate unifying the study of UDA-SS across video and image scenarios, enabling a more comprehensive understanding, synergistic advancements, and efficient knowledge sharing. To that end, we explore the unified UDA-SS from a general data augmentation perspective, serving as a unifying conceptual framework, enabling improved generalization, and potential for cross-pollination of ideas, ultimately contributing to the overall progress and practical impact of this field of research. Specifically, we propose a Quad-directional Mixup (QuadMix) method, characterized by tackling distinct point attributes and feature inconsistencies through four-directional paths for intra- and inter-domain mixing in a feature space. To deal with temporal shifts with videos, we incorporate optical flow-guided feature aggregation across spatial and temporal dimensions for fine-grained domain alignment. Extensive experiments show that our method outperforms the state-of-the-art works by large margins on four challenging UDA-SS benchmarks. Our source code and models will be released at https://github.com/ZHE-SAPI/UDASS.
comment: 17 pages,11 figures, 11 tables. Accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2025
♻ ☆ Online Video Understanding: OVBench and VideoChat-Online CVPR 2025
Multimodal Large Language Models (MLLMs) have significantly progressed in offline video understanding. However, applying these models to real-world scenarios, such as autonomous driving and human-computer interaction, presents unique challenges due to the need for real-time processing of continuous online video streams. To this end, this paper presents systematic efforts from three perspectives: evaluation benchmark, model architecture, and training strategy. First, we introduce OVBench, a comprehensive question-answering benchmark designed to evaluate models' ability to perceive, memorize, and reason within online video contexts. It features 6 core task types across three temporal contexts-past, current, and future-forming 16 subtasks from diverse datasets. Second, we propose a new Pyramid Memory Bank (PMB) that effectively retains key spatiotemporal information in video streams. Third, we proposed an offline-to-online learning paradigm, designing an interleaved dialogue format for online video data and constructing an instruction-tuning dataset tailored for online video training. This framework led to the development of VideoChat-Online, a robust and efficient model for online video understanding. Despite the lower computational cost and higher efficiency, VideoChat-Online outperforms existing state-of-the-art offline and online models across popular offline video benchmarks and OVBench, demonstrating the effectiveness of our model architecture and training strategy. % Our approach surpasses existing state-of-the-art offline models Qwen2-VL 7B and online models Flash-VStream, by 4.19% and 23.7% on OVBench, respectively.
comment: CVPR 2025 Camera Ready Version. Project Page: https://videochat-online.github.io
♻ ☆ FANeRV: Frequency Separation and Augmentation based Neural Representation for Video
Neural representations for video (NeRV) have gained considerable attention for their strong performance across various video tasks. However, existing NeRV methods often struggle to capture fine spatial details, resulting in vague reconstructions. In this paper, we present a Frequency Separation and Augmentation based Neural Representation for video (FANeRV), which addresses these limitations with its core Wavelet Frequency Upgrade Block. This block explicitly separates input frames into high and low-frequency components using discrete wavelet transform, followed by targeted enhancement using specialized modules. Finally, a specially designed gated network effectively fuses these frequency components for optimal reconstruction. Additionally, convolutional residual enhancement blocks are integrated into the later stages of the network to balance parameter distribution and improve the restoration of high-frequency details. Experimental results demonstrate that FANeRV significantly improves reconstruction performance and excels in multiple tasks, including video compression, inpainting, and interpolation, outperforming existing NeRV methods.
♻ ☆ Near, far: Patch-ordering enhances vision foundation models' scene understanding ICLR25
We introduce NeCo: Patch Neighbor Consistency, a novel self-supervised training loss that enforces patch-level nearest neighbor consistency across a student and teacher model. Compared to contrastive approaches that only yield binary learning signals, i.e., 'attract' and 'repel', this approach benefits from the more fine-grained learning signal of sorting spatially dense features relative to reference patches. Our method leverages differentiable sorting applied on top of pretrained representations, such as DINOv2-registers to bootstrap the learning signal and further improve upon them. This dense post-pretraining leads to superior performance across various models and datasets, despite requiring only 19 hours on a single GPU. This method generates high-quality dense feature encoders and establishes several new state-of-the-art results such as +5.5% and +6% for non-parametric in-context semantic segmentation on ADE20k and Pascal VOC, +7.2% and +5.7% for linear segmentation evaluations on COCO-Things and -Stuff and improvements in the 3D understanding of multi-view consistency on SPair-71k, by more than 1.5%.
comment: Accepted at ICLR25. The webpage is accessible at: https://vpariza.github.io/NeCo/
♻ ☆ Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions CVPR
Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges. This survey provides a wide-ranging examination of the anti-UAV domain, centering on three core objectives-classification, detection, and tracking-while detailing emerging methodologies such as diffusion-based data synthesis, multi-modal fusion, vision-language modeling, self-supervised learning, and reinforcement learning. We systematically evaluate state-of-the-art solutions across both single-modality and multi-sensor pipelines (spanning RGB, infrared, audio, radar, and RF) and discuss large-scale as well as adversarially oriented benchmarks. Our analysis reveals persistent gaps in real-time performance, stealth detection, and swarm-based scenarios, underscoring pressing needs for robust, adaptive anti-UAV systems. By highlighting open research directions, we aim to foster innovation and guide the development of next-generation defense strategies in an era marked by the extensive use of UAVs.
comment: Accepted at CVPR Workshop Anti-UAV 2025. 15 pages
♻ ☆ Evaluating Semantic Variation in Text-to-Image Synthesis: A Causal Perspective ICLR 2025
Accurate interpretation and visualization of human instructions are crucial for text-to-image (T2I) synthesis. However, current models struggle to capture semantic variations from word order changes, and existing evaluations, relying on indirect metrics like text-image similarity, fail to reliably assess these challenges. This often obscures poor performance on complex or uncommon linguistic patterns by the focus on frequent word combinations. To address these deficiencies, we propose a novel metric called SemVarEffect and a benchmark named SemVarBench, designed to evaluate the causality between semantic variations in inputs and outputs in T2I synthesis. Semantic variations are achieved through two types of linguistic permutations, while avoiding easily predictable literal variations. Experiments reveal that the CogView-3-Plus and Ideogram 2 performed the best, achieving a score of 0.2/1. Semantic variations in object relations are less understood than attributes, scoring 0.07/1 compared to 0.17-0.19/1. We found that cross-modal alignment in UNet or Transformers plays a crucial role in handling semantic variations, a factor previously overlooked by a focus on textual encoders. Our work establishes an effective evaluation framework that advances the T2I synthesis community's exploration of human instruction understanding. Our benchmark and code are available at https://github.com/zhuxiangru/SemVarBench .
comment: Accepted by ICLR 2025
♻ ☆ unPIC: A Geometric Multiview Prior for Image to 3D Synthesis
We introduce a hierarchical probabilistic approach to go from a 2D image to multiview 3D: a diffusion "prior" predicts the unseen 3D geometry, which then conditions a diffusion "decoder" to generate novel views of the subject. We use a pointmap-based geometric representation to coordinate the generation of multiple target views simultaneously. We construct a predictable distribution of geometric features per target view to enable learnability across examples, and generalization to arbitrary inputs images. Our modular, geometry-driven approach to novel-view synthesis (called "unPIC") beats competing baselines such as CAT3D, EscherNet, Free3D, and One-2-3-45 on held-out objects from ObjaverseXL, as well as unseen real-world objects from Google Scanned Objects, Amazon Berkeley Objects, and the Digital Twin Catalog.
♻ ☆ Look Before You Decide: Prompting Active Deduction of MLLMs for Assumptive Reasoning
Recently, Multimodal Large Language Models (MLLMs) have achieved significant success across multiple disciplines due to their exceptional instruction-following capabilities and extensive world knowledge. However, whether these MLLMs possess human-like compositional reasoning abilities remains an open problem. To unveil their reasoning behaviors, we first curate a \textbf{M}ultimodal \textbf{A}ssumptive \textbf{R}ea\textbf{s}oning Benchmark (MARS-Bench) in this paper. Interestingly, we find that most prevalent MLLMs can be easily fooled by the introduction of a presupposition into the question, whereas such presuppositions appear naive to human reasoning. Besides, we also propose a simple yet effective method, Active Deduction (AD), a novel reinforcement learning paradigm to encourage the model to actively perform composite deduction before reaching a final decision. Equipped with the proposed AD method, a MLLM demonstrates significant improvements in assumptive reasoning abilities without compromising its general-purpose question-answering performance. We also provide extensive evaluations of both open-source and private MLLMs on MARS-Bench, along with experimental analyses of the AD method.
♻ ☆ Towards Training-Free Open-World Classification with 3D Generative Models
3D open-world classification is a challenging yet essential task in dynamic and unstructured real-world scenarios, requiring both open-category and open-pose recognition. To address these challenges, recent wisdom often takes sophisticated 2D pre-trained models to provide enriched and stable representations. However, these methods largely rely on how 3D objects can be projected into 2D space, which is unfortunately not well solved, and thus significantly limits their performance. Unlike these present efforts, in this paper we make a pioneering exploration of 3D generative models for 3D open-world classification. Drawing on abundant prior knowledge from 3D generative models, we additionally craft a rotation-invariant feature extractor. This innovative synergy endows our pipeline with the advantages of being training-free, open-category, and pose-invariant, thus well suited to 3D open-world classification. Extensive experiments on benchmark datasets demonstrate the potential of generative models in 3D open-world classification, achieving state-of-the-art performance on ModelNet10 and McGill with 32.0% and 8.7% overall accuracy improvement, respectively.
♻ ☆ Scene-Text Grounding for Text-Based Video Question Answering
Existing efforts in text-based video question answering (TextVideoQA) are criticized for their opaque decisionmaking and heavy reliance on scene-text recognition. In this paper, we propose to study Grounded TextVideoQA by forcing models to answer questions and spatio-temporally localize the relevant scene-text regions, thus decoupling QA from scenetext recognition and promoting research towards interpretable QA. The task has three-fold significance. First, it encourages scene-text evidence versus other short-cuts for answer predictions. Second, it directly accepts scene-text regions as visual answers, thus circumventing the problem of ineffective answer evaluation by stringent string matching. Third, it isolates the challenges inherited in VideoQA and scene-text recognition. This enables the diagnosis of the root causes for failure predictions, e.g., wrong QA or wrong scene-text recognition? To achieve Grounded TextVideoQA, we propose the T2S-QA model that highlights a disentangled temporal-to-spatial contrastive learning strategy for weakly-supervised scene-text grounding and grounded TextVideoQA. To facilitate evaluation, we construct a new dataset ViTXT-GQA which features 52K scene-text bounding boxes within 2.2K temporal segments related to 2K questions and 729 videos. With ViTXT-GQA, we perform extensive experiments and demonstrate the severe limitations of existing techniques in Grounded TextVideoQA. While T2S-QA achieves superior results, the large performance gap with human leaves ample space for improvement. Our further analysis of oracle scene-text inputs posits that the major challenge is scene-text recognition. To advance the research of Grounded TextVideoQA, our dataset and code are at https://github.com/zhousheng97/ViTXT-GQA.git
♻ ☆ Human-Activity AGV Quality Assessment: A Benchmark Dataset and an Objective Evaluation Metric
AI-driven video generation techniques have made significant progress in recent years. However, AI-generated videos (AGVs) involving human activities often exhibit substantial visual and semantic distortions, hindering the practical application of video generation technologies in real-world scenarios. To address this challenge, we conduct a pioneering study on human activity AGV quality assessment, focusing on visual quality evaluation and the identification of semantic distortions. First, we construct the AI-Generated Human activity Video Quality Assessment (Human-AGVQA) dataset, consisting of 6,000 AGVs derived from 15 popular text-to-video (T2V) models using 400 text prompts that describe diverse human activities. We conduct a subjective study to evaluate the human appearance quality, action continuity quality, and overall video quality of AGVs, and identify semantic issues of human body parts. Based on Human-AGVQA, we benchmark the performance of T2V models and analyze their strengths and weaknesses in generating different categories of human activities. Second, we develop an objective evaluation metric, named AI-Generated Human activity Video Quality metric (GHVQ), to automatically analyze the quality of human activity AGVs. GHVQ systematically extracts human-focused quality features, AI-generated content-aware quality features, and temporal continuity features, making it a comprehensive and explainable quality metric for human activity AGVs. The extensive experimental results show that GHVQ outperforms existing quality metrics on the Human-AGVQA dataset by a large margin, demonstrating its efficacy in assessing the quality of human activity AGVs. The Human-AGVQA dataset and GHVQ metric will be released publicly.
♻ ☆ UI-E2I-Synth: Advancing GUI Grounding with Large-Scale Instruction Synthesis
Recent advancements in Large Vision-Language Models are accelerating the development of Graphical User Interface (GUI) agents that utilize human-like vision perception capabilities to enhance productivity on digital devices. Compared to approaches predicated on GUI metadata, which are platform-dependent and vulnerable to implementation variations, vision-based approaches offer broader applicability. In this vision-based paradigm, the GUI instruction grounding, which maps user instruction to the location of corresponding element on the given screenshot, remains a critical challenge, particularly due to limited public training dataset and resource-intensive manual instruction data annotation. In this paper, we delve into unexplored challenges in this task including element-to-screen ratio, unbalanced element type, and implicit instruction. To address these challenges, we introduce a large-scale data synthesis pipeline UI-E2I-Synth for generating varying complex instruction datasets using GPT-4o instead of human annotators. Furthermore, we propose a new GUI instruction grounding benchmark UI-I2E-Bench, which is designed to address the limitations of existing benchmarks by incorporating diverse annotation aspects. Our model, trained on the synthesized data, achieves superior performance in GUI instruction grounding, demonstrating the advancements of proposed data synthesis pipeline. The proposed benchmark, accompanied by extensive analyses, provides practical insights for future research in GUI grounding. We will release corresponding artifacts at https://colmon46.github.io/i2e-bench-leaderboard/ .
♻ ☆ RegMixMatch: Optimizing Mixup Utilization in Semi-Supervised Learning AAAI
Consistency regularization and pseudo-labeling have significantly advanced semi-supervised learning (SSL). Prior works have effectively employed Mixup for consistency regularization in SSL. However, our findings indicate that applying Mixup for consistency regularization may degrade SSL performance by compromising the purity of artificial labels. Moreover, most pseudo-labeling based methods utilize thresholding strategy to exclude low-confidence data, aiming to mitigate confirmation bias; however, this approach limits the utility of unlabeled samples. To address these challenges, we propose RegMixMatch, a novel framework that optimizes the use of Mixup with both high- and low-confidence samples in SSL. First, we introduce semi-supervised RegMixup, which effectively addresses reduced artificial labels purity by using both mixed samples and clean samples for training. Second, we develop a class-aware Mixup technique that integrates information from the top-2 predicted classes into low-confidence samples and their artificial labels, reducing the confirmation bias associated with these samples and enhancing their effective utilization. Experimental results demonstrate that RegMixMatch achieves state-of-the-art performance across various SSL benchmarks.
comment: Accepted in AAAI Conference on Artificial Intelligence (AAAI-25)
♻ ☆ Efficient Fourier Filtering Network with Contrastive Learning for UAV-based Unaligned Bi-modal Salient Object Detection
Unmanned aerial vehicle (UAV)-based bi-modal salient object detection (BSOD) aims to segment salient objects in a scene utilizing complementary cues in unaligned RGB and thermal image pairs. However, the high computational expense of existing UAV-based BSOD models limits their applicability to real-world UAV devices. To address this problem, we propose an efficient Fourier filter network with contrastive learning that achieves both real-time and accurate performance. Specifically, we first design a semantic contrastive alignment loss to align the two modalities at the semantic level, which facilitates mutual refinement in a parameter-free way. Second, inspired by the fast Fourier transform that obtains global relevance in linear complexity, we propose synchronized alignment fusion, which aligns and fuses bi-modal features in the channel and spatial dimensions by a hierarchical filtering mechanism. Our proposed model, AlignSal, reduces the number of parameters by 70.0%, decreases the floating point operations by 49.4%, and increases the inference speed by 152.5% compared to the cutting-edge BSOD model (i.e., MROS). Extensive experiments on the UAV RGB-T 2400 and seven bi-modal dense prediction datasets demonstrate that AlignSal achieves both real-time inference speed and better performance and generalizability compared to nineteen state-of-the-art models across most evaluation metrics. In addition, our ablation studies further verify AlignSal's potential in boosting the performance of existing aligned BSOD models on UAV-based unaligned data. The code is available at: https://github.com/JoshuaLPF/AlignSal.
comment: Accepted by TGRS 2025
♻ ☆ MCANet: Medical Image Segmentation with Multi-Scale Cross-Axis Attention
Efficiently capturing multi-scale information and building long-range dependencies among pixels are essential for medical image segmentation because of the various sizes and shapes of the lesion regions or organs. In this paper, we present Multi-scale Cross-axis Attention (MCA) to solve the above challenging issues based on the efficient axial attention. Instead of simply connecting axial attention along the horizontal and vertical directions sequentially, we propose to calculate dual cross attentions between two parallel axial attentions to capture global information better. To process the significant variations of lesion regions or organs in individual sizes and shapes, we also use multiple convolutions of strip-shape kernels with different kernel sizes in each axial attention path to improve the efficiency of the proposed MCA in encoding spatial information. We build the proposed MCA upon the MSCAN backbone, yielding our network, termed MCANet. Our MCANet with only 4M+ parameters performs even better than most previous works with heavy backbones (e.g., Swin Transformer) on four challenging tasks, including skin lesion segmentation, nuclei segmentation, abdominal multi-organ segmentation, and polyp segmentation. Code is available at https://github.com/haoshao-nku/medical_seg.
comment: accept to Machine Intelligence Research.DOI: 10.1007/s11633-025-1552-6
♻ ☆ ViTOC: Vision Transformer and Object-aware Captioner
This paper presents ViTOC (Vision Transformer and Object-aware Captioner), a novel vision-language model for image captioning that addresses the challenges of accuracy and diversity in generated descriptions. Unlike conventional approaches, ViTOC employs a dual-path architecture based on Vision Transformer and object detector, effectively fusing global visual features and local object information through learnable vectors. The model introduces an innovative object-aware prompting strategy that significantly enhances its capability in handling long-tail data. Experiments on the standard COCO dataset demonstrate that ViTOC outperforms baseline models across all evaluation metrics. Additionally, we propose a reference-free evaluation method based on CLIP to further validate the model's effectiveness. By utilizing pretrained visual model parameters, ViTOC achieves efficient end-to-end training.
comment: The core idea is too close to what has been published in other journals
♻ ☆ PATFinger: Prompt-Adapted Transferable Fingerprinting against Unauthorized Multimodal Dataset Usage
The multimodal datasets can be leveraged to pre-train large-scale vision-language models by providing cross-modal semantics. Current endeavors for determining the usage of datasets mainly focus on single-modal dataset ownership verification through intrusive methods and non-intrusive techniques, while cross-modal approaches remain under-explored. Intrusive methods can adapt to multimodal datasets but degrade model accuracy, while non-intrusive methods rely on label-driven decision boundaries that fail to guarantee stable behaviors for verification. To address these issues, we propose a novel prompt-adapted transferable fingerprinting scheme from a training-free perspective, called PATFinger, which incorporates the global optimal perturbation (GOP) and the adaptive prompts to capture dataset-specific distribution characteristics. Our scheme utilizes inherent dataset attributes as fingerprints instead of compelling the model to learn triggers. The GOP is derived from the sample distribution to maximize embedding drifts between different modalities. Subsequently, our PATFinger re-aligns the adaptive prompt with GOP samples to capture the cross-modal interactions on the carefully crafted surrogate model. This allows the dataset owner to check the usage of datasets by observing specific prediction behaviors linked to the PATFinger during retrieval queries. Extensive experiments demonstrate the effectiveness of our scheme against unauthorized multimodal dataset usage on various cross-modal retrieval architectures by 30% over state-of-the-art baselines.
♻ ☆ C2GM: Cascading conditional generative cartography framework for multi-scale tile map generation with geographic feature constraints
Multi-scale maps are essential representations of surveying and cartographic results, serving as fundamental components of geographic services. Current image generation networks can quickly produce map tiles from remote-sensing images. However, generative models designed for natural images often focus on texture features, neglecting the unique characteristics of remote-sensing features and the scale attributes of tile maps. This limitation in generative models impairs the accurate representation of geographic information, and the quality of tile map generation still needs improvement. Diffusion models have demonstrated remarkable success in various image generation tasks, highlighting their potential to address this challenge. This paper presents C2GM, a novel framework for generating multi-scale tile maps through conditional guided diffusion and multi-scale cascade generation. Specifically, we implement a conditional feature fusion encoder to extract object priors from remote sensing images and cascade reference double branch input, ensuring an accurate representation of complex features. Low-level generated tiles act as constraints for high-level map generation, enhancing visual continuity. Moreover, we incorporate map scale modality information using CLIP to simulate the relationship between map scale and cartographic generalization in tile maps. Extensive experimental evaluations demonstrate that C2GM consistently achieves the state-of-the-art (SOTA) performance on all metrics, facilitating the rapid and effective generation of multi-scale large-format maps for emergency response and remote mapping applications.
♻ ☆ Adaptive Decision Boundary for Few-Shot Class-Incremental Learning
Few-Shot Class-Incremental Learning (FSCIL) aims to continuously learn new classes from a limited set of training samples without forgetting knowledge of previously learned classes. Conventional FSCIL methods typically build a robust feature extractor during the base training session with abundant training samples and subsequently freeze this extractor, only fine-tuning the classifier in subsequent incremental phases. However, current strategies primarily focus on preventing catastrophic forgetting, considering only the relationship between novel and base classes, without paying attention to the specific decision spaces of each class. To address this challenge, we propose a plug-and-play Adaptive Decision Boundary Strategy (ADBS), which is compatible with most FSCIL methods. Specifically, we assign a specific decision boundary to each class and adaptively adjust these boundaries during training to optimally refine the decision spaces for the classes in each session. Furthermore, to amplify the distinctiveness between classes, we employ a novel inter-class constraint loss that optimizes the decision boundaries and prototypes for each class. Extensive experiments on three benchmarks, namely CIFAR100, miniImageNet, and CUB200, demonstrate that incorporating our ADBS method with existing FSCIL techniques significantly improves performance, achieving overall state-of-the-art results.
♻ ☆ DVLTA-VQA: Decoupled Vision-Language Modeling with Text-Guided Adaptation for Blind Video Quality Assessment
Inspired by the dual-stream theory of the human visual system (HVS) - where the ventral stream is responsible for object recognition and detail analysis, while the dorsal stream focuses on spatial relationships and motion perception - an increasing number of video quality assessment (VQA) works built upon this framework are proposed. Recent advancements in large multi-modal models, notably Contrastive Language-Image Pretraining (CLIP), have motivated researchers to incorporate CLIP into dual-stream-based VQA methods. This integration aims to harness the model's superior semantic understanding capabilities to replicate the object recognition and detail analysis in ventral stream, as well as spatial relationship analysis in dorsal stream. However, CLIP is originally designed for images and lacks the ability to capture temporal and motion information inherent in videos.To address the limitation, this paper propose a Decoupled Vision-Language Modeling with Text-Guided Adaptation for Blind Video Quality Assessment (DVLTA-VQA), which decouples CLIP's visual and textual components, and integrates them into different stages of the NR-VQA pipeline. Specifically, a Video-Based Temporal CLIP module is proposed to explicitly model temporal dynamics and enhance motion perception, aligning with the dorsal stream. Additionally, a Temporal Context Module is developed to refine inter-frame dependencies, further improving motion modeling. On the ventral stream side, a Basic Visual Feature Extraction Module is employed to strengthen detail analysis. Finally, a text-guided adaptive fusion strategy is proposed to enable dynamic weighting of features, facilitating more effective integration of spatial and temporal information.
♻ ☆ Distribution Backtracking Builds A Faster Convergence Trajectory for Diffusion Distillation
Accelerating the sampling speed of diffusion models remains a significant challenge. Recent score distillation methods distill a heavy teacher model into a student generator to achieve one-step generation, which is optimized by calculating the difference between the two score functions on the samples generated by the student model. However, there is a score mismatch issue in the early stage of the distillation process, because existing methods mainly focus on using the endpoint of pre-trained diffusion models as teacher models, overlooking the importance of the convergence trajectory between the student generator and the teacher model. To address this issue, we extend the score distillation process by introducing the entire convergence trajectory of teacher models and propose Distribution Backtracking Distillation (DisBack). DisBask is composed of two stages: Degradation Recording and Distribution Backtracking. Degradation Recording is designed to obtain the convergence trajectory of the teacher model, which records the degradation path from the trained teacher model to the untrained initial student generator. The degradation path implicitly represents the teacher model's intermediate distributions, and its reverse can be viewed as the convergence trajectory from the student generator to the teacher model. Then Distribution Backtracking trains a student generator to backtrack the intermediate distributions along the path to approximate the convergence trajectory of teacher models. Extensive experiments show that DisBack achieves faster and better convergence than the existing distillation method and accomplishes comparable generation performance, with FID score of 1.38 on ImageNet 64x64 dataset. Notably, DisBack is easy to implement and can be generalized to existing distillation methods to boost performance. Our code is publicly available on https://github.com/SYZhang0805/DisBack.
comment: Our code is publicly available on https://github.com/SYZhang0805/DisBack
♻ ☆ Transferable Foundation Models for Geometric Tasks on Point Cloud Representations: Geometric Neural Operators
We introduce methods for obtaining pretrained Geometric Neural Operators (GNPs) that can serve as basal foundation models for use in obtaining geometric features. These can be used within data processing pipelines for machine learning tasks and numerical methods. We show how our GNPs can be trained to learn robust latent representations for the differential geometry of point-clouds to provide estimates of metric, curvature, and other shape-related features. We demonstrate how our pre-trained GNPs can be used (i) to estimate the geometric properties of surfaces of arbitrary shape and topologies with robustness in the presence of noise, (ii) to approximate solutions of geometric partial differential equations (PDEs) on manifolds, and (iii) to solve equations for shape deformations such as curvature driven flows. We release codes and weights for using GNPs in the package geo_neural_op. This allows for incorporating our pre-trained GNPs as components for reuse within existing and new data processing pipelines. The GNPs also can be used as part of numerical solvers involving geometry or as part of methods for performing inference and other geometric tasks.
♻ ☆ UMSPU: Universal Multi-Size Phase Unwrapping via Mutual Self-Distillation and Adaptive Boosting Ensemble Segmenters
Spatial phase unwrapping is a key technique for extracting phase information to obtain 3D morphology and other features. Modern industrial measurement scenarios demand high precision, large image sizes, and high speed. However, conventional methods struggle with noise resistance and processing speed. Current deep learning methods are limited by the receptive field size and sparse semantic information, making them ineffective for large size images. To address this issue, we propose a mutual self-distillation (MSD) mechanism and adaptive boosting ensemble segmenters to construct a universal multi-size phase unwrapping network (UMSPU). MSD performs hierarchical attention refinement and achieves cross-layer collaborative learning through bidirectional distillation, ensuring fine-grained semantic representation across image sizes. The adaptive boosting ensemble segmenters combine weak segmenters with different receptive fields into a strong one, ensuring stable segmentation across spatial frequencies. Experimental results show that UMSPU overcomes image size limitations, achieving high precision across image sizes ranging from 256*256 to 2048*2048 (an 8 times increase). It also outperforms existing methods in speed, robustness, and generalization. Its practicality is further validated in structured light imaging and InSAR. We believe that UMSPU offers a universal solution for phase unwrapping, with broad potential for industrial applications.
♻ ☆ StrokeFusion: Vector Sketch Generation via Joint Stroke-UDF Encoding and Latent Sequence Diffusion
In the field of sketch generation, raster-format trained models often produce non-stroke artifacts, while vector-format trained models typically lack a holistic understanding of sketches, leading to compromised recognizability. Moreover, existing methods struggle to extract common features from similar elements (e.g., eyes of animals) appearing at varying positions across sketches. To address these challenges, we propose StrokeFusion, a two-stage framework for vector sketch generation. It contains a dual-modal sketch feature learning network that maps strokes into a high-quality latent space. This network decomposes sketches into normalized strokes and jointly encodes stroke sequences with Unsigned Distance Function (UDF) maps, representing sketches as sets of stroke feature vectors. Building upon this representation, our framework exploits a stroke-level latent diffusion model that simultaneously adjusts stroke position, scale, and trajectory during generation. This enables high-fidelity sketch generation while supporting stroke interpolation editing. Extensive experiments on the QuickDraw dataset demonstrate that our framework outperforms state-of-the-art techniques, validating its effectiveness in preserving structural integrity and semantic features. Code and models will be made publicly available upon publication.
♻ ☆ Structured 3D Latents for Scalable and Versatile 3D Generation
We introduce a novel 3D generation method for versatile and high-quality 3D asset creation. The cornerstone is a unified Structured LATent (SLAT) representation which allows decoding to different output formats, such as Radiance Fields, 3D Gaussians, and meshes. This is achieved by integrating a sparsely-populated 3D grid with dense multiview visual features extracted from a powerful vision foundation model, comprehensively capturing both structural (geometry) and textural (appearance) information while maintaining flexibility during decoding. We employ rectified flow transformers tailored for SLAT as our 3D generation models and train models with up to 2 billion parameters on a large 3D asset dataset of 500K diverse objects. Our model generates high-quality results with text or image conditions, significantly surpassing existing methods, including recent ones at similar scales. We showcase flexible output format selection and local 3D editing capabilities which were not offered by previous models. Code, model, and data will be released.
comment: Project Page: https://github.com/Microsoft/TRELLIS
♻ ☆ ChemVLM: Exploring the Power of Multimodal Large Language Models in Chemistry Area
Large Language Models (LLMs) have achieved remarkable success and have been applied across various scientific fields, including chemistry. However, many chemical tasks require the processing of visual information, which cannot be successfully handled by existing chemical LLMs. This brings a growing need for models capable of integrating multimodal information in the chemical domain. In this paper, we introduce \textbf{ChemVLM}, an open-source chemical multimodal large language model specifically designed for chemical applications. ChemVLM is trained on a carefully curated bilingual multimodal dataset that enhances its ability to understand both textual and visual chemical information, including molecular structures, reactions, and chemistry examination questions. We develop three datasets for comprehensive evaluation, tailored to Chemical Optical Character Recognition (OCR), Multimodal Chemical Reasoning (MMCR), and Multimodal Molecule Understanding tasks. We benchmark ChemVLM against a range of open-source and proprietary multimodal large language models on various tasks. Experimental results demonstrate that ChemVLM achieves competitive performance across all evaluated tasks. Our model can be found at https://huggingface.co/AI4Chem/ChemVLM-26B.
comment: 11 pages, updated version
♻ ☆ StructRe: Rewriting for Structured Shape Modeling
Man-made 3D shapes are naturally organized in parts and hierarchies; such structures provide important constraints for shape reconstruction and generation. Modeling shape structures is difficult, because there can be multiple hierarchies for a given shape, causing ambiguity, and across different categories the shape structures are correlated with semantics, limiting generalization. We present StructRe, a structure rewriting system, as a novel approach to structured shape modeling. Given a 3D object represented by points and components, StructRe can rewrite it upward into more concise structures, or downward into more detailed structures; by iterating the rewriting process, hierarchies are obtained. Such a localized rewriting process enables probabilistic modeling of ambiguous structures and robust generalization across object categories. We train StructRe on PartNet data and show its generalization to cross-category and multiple object hierarchies, and test its extension to ShapeNet. We also demonstrate the benefits of probabilistic and generalizable structure modeling for shape reconstruction, generation and editing tasks.
comment: Our project page: https://jiepengwang.github.io/StructRe/
♻ ☆ Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework
Multimodal foundation models offer a promising framework for robotic perception and planning by processing sensory inputs to generate actionable plans. However, addressing uncertainty in both perception (sensory interpretation) and decision-making (plan generation) remains a critical challenge for ensuring task reliability. We present a comprehensive framework to disentangle, quantify, and mitigate these two forms of uncertainty. We first introduce a framework for uncertainty disentanglement, isolating perception uncertainty arising from limitations in visual understanding and decision uncertainty relating to the robustness of generated plans. To quantify each type of uncertainty, we propose methods tailored to the unique properties of perception and decision-making: we use conformal prediction to calibrate perception uncertainty and introduce Formal-Methods-Driven Prediction (FMDP) to quantify decision uncertainty, leveraging formal verification techniques for theoretical guarantees. Building on this quantification, we implement two targeted intervention mechanisms: an active sensing process that dynamically re-observes high-uncertainty scenes to enhance visual input quality and an automated refinement procedure that fine-tunes the model on high-certainty data, improving its capability to meet task specifications. Empirical validation in real-world and simulated robotic tasks demonstrates that our uncertainty disentanglement framework reduces variability by up to 40% and enhances task success rates by 5% compared to baselines. These improvements are attributed to the combined effect of both interventions and highlight the importance of uncertainty disentanglement, which facilitates targeted interventions that enhance the robustness and reliability of autonomous systems. Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/.
comment: Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/
♻ ☆ SEAL: Semantic Attention Learning for Long Video Representation
Long video understanding presents challenges due to the inherent high computational complexity and redundant temporal information. An effective representation for long videos must efficiently process such redundancy while preserving essential contents for downstream tasks. This paper introduces SEmantic Attention Learning (SEAL), a novel unified representation for long videos. To reduce computational complexity, long videos are decomposed into three distinct types of semantic entities: scenes, objects, and actions, allowing models to operate on a compact set of entities rather than a large number of frames or pixels. To further address redundancy, we propose an attention learning module that balances token relevance with diversity, formulated as a subset selection optimization problem. Our representation is versatile and applicable across various long video understanding tasks. Extensive experiments demonstrate that SEAL significantly outperforms state-of-the-art methods in video question answering and temporal grounding tasks across diverse benchmarks, including LVBench, MovieChat-1K, and Ego4D.
♻ ☆ Understanding Attention Mechanism in Video Diffusion Models
Text-to-video (T2V) synthesis models, such as OpenAI's Sora, have garnered significant attention due to their ability to generate high-quality videos from a text prompt. In diffusion-based T2V models, the attention mechanism is a critical component. However, it remains unclear what intermediate features are learned and how attention blocks in T2V models affect various aspects of video synthesis, such as image quality and temporal consistency. In this paper, we conduct an in-depth perturbation analysis of the spatial and temporal attention blocks of T2V models using an information-theoretic approach. Our results indicate that temporal and spatial attention maps affect not only the timing and layout of the videos but also the complexity of spatiotemporal elements and the aesthetic quality of the synthesized videos. Notably, high-entropy attention maps are often key elements linked to superior video quality, whereas low-entropy attention maps are associated with the video's intra-frame structure. Based on our findings, we propose two novel methods to enhance video quality and enable text-guided video editing. These methods rely entirely on lightweight manipulation of the attention matrices in T2V models. The efficacy and effectiveness of our methods are further validated through experimental evaluation across multiple datasets.
♻ ☆ CoDiff: Conditional Diffusion Model for Collaborative 3D Object Detection
Collaborative 3D object detection holds significant importance in the field of autonomous driving, as it greatly enhances the perception capabilities of each individual agent by facilitating information exchange among multiple agents. However, in practice, due to pose estimation errors and time delays, the fusion of information across agents often results in feature representations with spatial and temporal noise, leading to detection errors. Diffusion models naturally have the ability to denoise noisy samples to the ideal data, which motivates us to explore the use of diffusion models to address the noise problem between multi-agent systems. In this work, we propose CoDiff, a novel robust collaborative perception framework that leverages the potential of diffusion models to generate more comprehensive and clearer feature representations. To the best of our knowledge, this is the first work to apply diffusion models to multi-agent collaborative perception. Specifically, we project high-dimensional feature map into the latent space of a powerful pre-trained autoencoder. Within this space, individual agent information serves as a condition to guide the diffusion model's sampling. This process denoises coarse feature maps and progressively refines the fused features. Experimental study on both simulated and real-world datasets demonstrates that the proposed framework CoDiff consistently outperforms existing relevant methods in terms of the collaborative object detection performance, and exhibits highly desired robustness when the pose and delay information of agents is with high-level noise. The code is released at https://github.com/HuangZhe885/CoDiff
Information Retrieval 16
☆ SemCORE: A Semantic-Enhanced Generative Cross-Modal Retrieval Framework with MLLMs
Cross-modal retrieval (CMR) is a fundamental task in multimedia research, focused on retrieving semantically relevant targets across different modalities. While traditional CMR methods match text and image via embedding-based similarity calculations, recent advancements in pre-trained generative models have established generative retrieval as a promising alternative. This paradigm assigns each target a unique identifier and leverages a generative model to directly predict identifiers corresponding to input queries without explicit indexing. Despite its great potential, current generative CMR approaches still face semantic information insufficiency in both identifier construction and generation processes. To address these limitations, we propose a novel unified Semantic-enhanced generative Cross-mOdal REtrieval framework (SemCORE), designed to unleash the semantic understanding capabilities in generative cross-modal retrieval task. Specifically, we first construct a Structured natural language IDentifier (SID) that effectively aligns target identifiers with generative models optimized for natural language comprehension and generation. Furthermore, we introduce a Generative Semantic Verification (GSV) strategy enabling fine-grained target discrimination. Additionally, to the best of our knowledge, SemCORE is the first framework to simultaneously consider both text-to-image and image-to-text retrieval tasks within generative cross-modal retrieval. Extensive experiments demonstrate that our framework outperforms state-of-the-art generative cross-modal retrieval methods. Notably, SemCORE achieves substantial improvements across benchmark datasets, with an average increase of 8.65 points in Recall@1 for text-to-image retrieval.
☆ FreshStack: Building Realistic Benchmarks for Evaluating Retrieval on Technical Documents
We introduce FreshStack, a reusable framework for automatically building information retrieval (IR) evaluation benchmarks from community-asked questions and answers. FreshStack conducts the following steps: (1) automatic corpus collection from code and technical documentation, (2) nugget generation from community-asked questions and answers, and (3) nugget-level support, retrieving documents using a fusion of retrieval techniques and hybrid architectures. We use FreshStack to build five datasets on fast-growing, recent, and niche topics to ensure the tasks are sufficiently challenging. On FreshStack, existing retrieval models, when applied out-of-the-box, significantly underperform oracle approaches on all five topics, denoting plenty of headroom to improve IR quality. In addition, we identify cases where rerankers do not clearly improve first-stage retrieval accuracy (two out of five topics). We hope that FreshStack will facilitate future work toward constructing realistic, scalable, and uncontaminated IR and RAG evaluation benchmarks. FreshStack datasets are available at: https://fresh-stack.github.io.
☆ Should We Tailor the Talk? Understanding the Impact of Conversational Styles on Preference Elicitation in Conversational Recommender Systems
Conversational recommender systems (CRSs) provide users with an interactive means to express preferences and receive real-time personalized recommendations. The success of these systems is heavily influenced by the preference elicitation process. While existing research mainly focuses on what questions to ask during preference elicitation, there is a notable gap in understanding what role broader interaction patterns including tone, pacing, and level of proactiveness play in supporting users in completing a given task. This study investigates the impact of different conversational styles on preference elicitation, task performance, and user satisfaction with CRSs. We conducted a controlled experiment in the context of scientific literature recommendation, contrasting two distinct conversational styles, high involvement (fast paced, direct, and proactive with frequent prompts) and high considerateness (polite and accommodating, prioritizing clarity and user comfort) alongside a flexible experimental condition where users could switch between the two. Our results indicate that adapting conversational strategies based on user expertise and allowing flexibility between styles can enhance both user satisfaction and the effectiveness of recommendations in CRSs. Overall, our findings hold important implications for the design of future CRSs.
comment: To appear in: Proceedings of the 33rd ACM Conference on User Modeling, Adaptation and Personalization (UMAP '25), June 16--19, 2025, New York City, NY, USA
☆ InstructRAG: Leveraging Retrieval-Augmented Generation on Instruction Graphs for LLM-Based Task Planning SIGIR 2025
Recent advancements in large language models (LLMs) have enabled their use as agents for planning complex tasks. Existing methods typically rely on a thought-action-observation (TAO) process to enhance LLM performance, but these approaches are often constrained by the LLMs' limited knowledge of complex tasks. Retrieval-augmented generation (RAG) offers new opportunities by leveraging external databases to ground generation in retrieved information. In this paper, we identify two key challenges (enlargability and transferability) in applying RAG to task planning. We propose InstructRAG, a novel solution within a multi-agent meta-reinforcement learning framework, to address these challenges. InstructRAG includes a graph to organize past instruction paths (sequences of correct actions), an RL-Agent with Reinforcement Learning to expand graph coverage for enlargability, and an ML-Agent with Meta-Learning to improve task generalization for transferability. The two agents are trained end-to-end to optimize overall planning performance. Our experiments on four widely used task planning datasets demonstrate that InstructRAG significantly enhances performance and adapts efficiently to new tasks, achieving up to a 19.2% improvement over the best existing approach.
comment: This paper has been accepted by SIGIR 2025
☆ CSMF: Cascaded Selective Mask Fine-Tuning for Multi-Objective Embedding-Based Retrieval SIGIR
Multi-objective embedding-based retrieval (EBR) has become increasingly critical due to the growing complexity of user behaviors and commercial objectives. While traditional approaches often suffer from data sparsity and limited information sharing between objectives, recent methods utilizing a shared network alongside dedicated sub-networks for each objective partially address these limitations. However, such methods significantly increase the model parameters, leading to an increased retrieval latency and a limited ability to model causal relationships between objectives. To address these challenges, we propose the Cascaded Selective Mask Fine-Tuning (CSMF), a novel method that enhances both retrieval efficiency and serving performance for multi-objective EBR. The CSMF framework selectively masks model parameters to free up independent learning space for each objective, leveraging the cascading relationships between objectives during the sequential fine-tuning. Without increasing network parameters or online retrieval overhead, CSMF computes a linearly weighted fusion score for multiple objective probabilities while supporting flexible adjustment of each objective's weight across various recommendation scenarios. Experimental results on real-world datasets demonstrate the superior performance of CSMF, and online experiments validate its significant practical value.
comment: 10 pages, 8 figures, Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '25), July 13--18, 2025, Padua, Italy
☆ ConExion: Concept Extraction with Large Language Models
In this paper, an approach for concept extraction from documents using pre-trained large language models (LLMs) is presented. Compared with conventional methods that extract keyphrases summarizing the important information discussed in a document, our approach tackles a more challenging task of extracting all present concepts related to the specific domain, not just the important ones. Through comprehensive evaluations of two widely used benchmark datasets, we demonstrate that our method improves the F1 score compared to state-of-the-art techniques. Additionally, we explore the potential of using prompts within these models for unsupervised concept extraction. The extracted concepts are intended to support domain coverage evaluation of ontologies and facilitate ontology learning, highlighting the effectiveness of LLMs in concept extraction tasks. Our source code and datasets are publicly available at https://github.com/ISE-FIZKarlsruhe/concept_extraction.
☆ FashionDPO:Fine-tune Fashion Outfit Generation Model using Direct Preference Optimization SIGIR'25
Personalized outfit generation aims to construct a set of compatible and personalized fashion items as an outfit. Recently, generative AI models have received widespread attention, as they can generate fashion items for users to complete an incomplete outfit or create a complete outfit. However, they have limitations in terms of lacking diversity and relying on the supervised learning paradigm. Recognizing this gap, we propose a novel framework FashionDPO, which fine-tunes the fashion outfit generation model using direct preference optimization. This framework aims to provide a general fine-tuning approach to fashion generative models, refining a pre-trained fashion outfit generation model using automatically generated feedback, without the need to design a task-specific reward function. To make sure that the feedback is comprehensive and objective, we design a multi-expert feedback generation module which covers three evaluation perspectives, \ie quality, compatibility and personalization. Experiments on two established datasets, \ie iFashion and Polyvore-U, demonstrate the effectiveness of our framework in enhancing the model's ability to align with users' personalized preferences while adhering to fashion compatibility principles. Our code and model checkpoints are available at https://github.com/Yzcreator/FashionDPO.
comment: Accepted by SIGIR'25
☆ Building Russian Benchmark for Evaluation of Information Retrieval Models
We introduce RusBEIR, a comprehensive benchmark designed for zero-shot evaluation of information retrieval (IR) models in the Russian language. Comprising 17 datasets from various domains, it integrates adapted, translated, and newly created datasets, enabling systematic comparison of lexical and neural models. Our study highlights the importance of preprocessing for lexical models in morphologically rich languages and confirms BM25 as a strong baseline for full-document retrieval. Neural models, such as mE5-large and BGE-M3, demonstrate superior performance on most datasets, but face challenges with long-document retrieval due to input size constraints. RusBEIR offers a unified, open-source framework that promotes research in Russian-language information retrieval.
☆ Towards Lossless Token Pruning in Late-Interaction Retrieval Models SIGIR 2025
Late interaction neural IR models like ColBERT offer a competitive effectiveness-efficiency trade-off across many benchmarks. However, they require a huge memory space to store the contextual representation for all the document tokens. Some works have proposed using either heuristics or statistical-based techniques to prune tokens from each document. This however doesn't guarantee that the removed tokens have no impact on the retrieval score. Our work uses a principled approach to define how to prune tokens without impacting the score between a document and a query. We introduce three regularization losses, that induce a solution with high pruning ratios, as well as two pruning strategies. We study them experimentally (in and out-domain), showing that we can preserve ColBERT's performance while using only 30\% of the tokens.
comment: Accepted at SIGIR 2025 Full Paper Track
☆ Validating LLM-Generated Relevance Labels for Educational Resource Search WSDM '25
Manual relevance judgements in Information Retrieval are costly and require expertise, driving interest in using Large Language Models (LLMs) for automatic assessment. While LLMs have shown promise in general web search scenarios, their effectiveness for evaluating domain-specific search results, such as educational resources, remains unexplored. To investigate different ways of including domain-specific criteria in LLM prompts for relevance judgement, we collected and released a dataset of 401 human relevance judgements from a user study involving teaching professionals performing search tasks related to lesson planning. We compared three approaches to structuring these prompts: a simple two-aspect evaluation baseline from prior work on using LLMs as relevance judges, a comprehensive 12-dimensional rubric derived from educational literature, and criteria directly informed by the study participants. Using domain-specific frameworks, LLMs achieved strong agreement with human judgements (Cohen's $\kappa$ up to 0.650), significantly outperforming the baseline approach. The participant-derived framework proved particularly robust, with GPT-3.5 achieving $\kappa$ scores of 0.639 and 0.613 for 10-dimension and 5-dimension versions respectively. System-level evaluation showed that LLM judgements reliably identified top-performing retrieval approaches (RBO scores 0.71-0.76) while maintaining reasonable discrimination between systems (RBO 0.52-0.56). These findings suggest that LLMs can effectively evaluate educational resources when prompted with domain-specific criteria, though performance varies with framework complexity and input structure.
comment: Presented in the LLM4Eval Workshop Co-located with WSDM '25 in Hannover, Germany
☆ SimUSER: Simulating User Behavior with Large Language Models for Recommender System Evaluation
Recommender systems play a central role in numerous real-life applications, yet evaluating their performance remains a significant challenge due to the gap between offline metrics and online behaviors. Given the scarcity and limits (e.g., privacy issues) of real user data, we introduce SimUSER, an agent framework that serves as believable and cost-effective human proxies. SimUSER first identifies self-consistent personas from historical data, enriching user profiles with unique backgrounds and personalities. Then, central to this evaluation are users equipped with persona, memory, perception, and brain modules, engaging in interactions with the recommender system. SimUSER exhibits closer alignment with genuine humans than prior work, both at micro and macro levels. Additionally, we conduct insightful experiments to explore the effects of thumbnails on click rates, the exposure effect, and the impact of reviews on user engagement. Finally, we refine recommender system parameters based on offline A/B test results, resulting in improved user engagement in the real world.
☆ Benchmarking LLM-based Relevance Judgment Methods
Large Language Models (LLMs) are increasingly deployed in both academic and industry settings to automate the evaluation of information seeking systems, particularly by generating graded relevance judgments. Previous work on LLM-based relevance assessment has primarily focused on replicating graded human relevance judgments through various prompting strategies. However, there has been limited exploration of alternative assessment methods or comprehensive comparative studies. In this paper, we systematically compare multiple LLM-based relevance assessment methods, including binary relevance judgments, graded relevance assessments, pairwise preference-based methods, and two nugget-based evaluation methods~--~document-agnostic and document-dependent. In addition to a traditional comparison based on system rankings using Kendall correlations, we also examine how well LLM judgments align with human preferences, as inferred from relevance grades. We conduct extensive experiments on datasets from three TREC Deep Learning tracks 2019, 2020 and 2021 as well as the ANTIQUE dataset, which focuses on non-factoid open-domain question answering. As part of our data release, we include relevance judgments generated by both an open-source (Llama3.2b) and a commercial (gpt-4o) model. Our goal is to \textit{reproduce} various LLM-based relevance judgment methods to provide a comprehensive comparison. All code, data, and resources are publicly available in our GitHub Repository at https://github.com/Narabzad/llm-relevance-judgement-comparison.
♻ ☆ dsld: A Socially Relevant Tool for Teaching Statistics
The growing power of data science can play a crucial role in addressing social discrimination, necessitating nuanced understanding and effective mitigation strategies for biases. "Data Science Looks At Discrimination" (DSLD) is an R and Python package designed to provide users with a comprehensive toolkit of statistical and graphical methods for assessing possible discrimination related to protected groups such as race, gender, and age. The package addresses critical issues by identifying and mitigating confounders and reducing bias against protected groups in prediction algorithms. In educational settings, DSLD offers instructors powerful tools to teach statistical principles through motivating real world examples of discrimination analysis. The inclusion of an 80 page Quarto book further supports users from statistics educators to legal professionals in effectively applying these analytical tools to real world scenarios.
comment: To be submitted to journal
♻ ☆ Comprehending Knowledge Graphs with Large Language Models for Recommender Systems SIGIR'25
In recent years, the introduction of knowledge graphs (KGs) has significantly advanced recommender systems by facilitating the discovery of potential associations between items. However, existing methods still face several limitations. First, most KGs suffer from missing facts or limited scopes. Second, existing methods convert textual information in KGs into IDs, resulting in the loss of natural semantic connections between different items. Third, existing methods struggle to capture high-order connections in the global KG. To address these limitations, we propose a novel method called CoLaKG, which leverages large language models (LLMs) to improve KG-based recommendations. The extensive knowledge and remarkable reasoning capabilities of LLMs enable our method to supplement missing facts in KGs, and their powerful text understanding abilities allow for better utilization of semantic information. Specifically, CoLaKG extracts useful information from KGs at both local and global levels. By employing the item-centered subgraph extraction and prompt engineering, it can accurately understand the local information. In addition, through the semantic-based retrieval module, each item is enriched by related items from the entire knowledge graph, effectively harnessing global information. Furthermore, the local and global information are effectively integrated into the recommendation model through a representation fusion module and a retrieval-augmented representation learning module, respectively. Extensive experiments on four real-world datasets demonstrate the superiority of our method.
comment: Accepted as a full paper by SIGIR'25
♻ ☆ PATFinger: Prompt-Adapted Transferable Fingerprinting against Unauthorized Multimodal Dataset Usage
The multimodal datasets can be leveraged to pre-train large-scale vision-language models by providing cross-modal semantics. Current endeavors for determining the usage of datasets mainly focus on single-modal dataset ownership verification through intrusive methods and non-intrusive techniques, while cross-modal approaches remain under-explored. Intrusive methods can adapt to multimodal datasets but degrade model accuracy, while non-intrusive methods rely on label-driven decision boundaries that fail to guarantee stable behaviors for verification. To address these issues, we propose a novel prompt-adapted transferable fingerprinting scheme from a training-free perspective, called PATFinger, which incorporates the global optimal perturbation (GOP) and the adaptive prompts to capture dataset-specific distribution characteristics. Our scheme utilizes inherent dataset attributes as fingerprints instead of compelling the model to learn triggers. The GOP is derived from the sample distribution to maximize embedding drifts between different modalities. Subsequently, our PATFinger re-aligns the adaptive prompt with GOP samples to capture the cross-modal interactions on the carefully crafted surrogate model. This allows the dataset owner to check the usage of datasets by observing specific prediction behaviors linked to the PATFinger during retrieval queries. Extensive experiments demonstrate the effectiveness of our scheme against unauthorized multimodal dataset usage on various cross-modal retrieval architectures by 30% over state-of-the-art baselines.
♻ ☆ Uncertainty Calibration for Counterfactual Propensity Estimation in Recommendation
Post-click conversion rate (CVR) is a reliable indicator of online customers' preferences, making it crucial for developing recommender systems. A major challenge in predicting CVR is severe selection bias, arising from users' inherent self-selection behavior and the system's item selection process. To mitigate this issue, the inverse propensity score (IPS) is employed to weight the prediction error of each observed instance. However, current propensity score estimations are unreliable due to the lack of a quality measure. To address this, we evaluate the quality of propensity scores from the perspective of uncertainty calibration, proposing the use of Expected Calibration Error (ECE) as a measure of propensity-score quality, which quantifies the extent to which predicted probabilities are overconfident by assessing the difference between predicted probabilities and actual observed frequencies. Miscalibrated propensity scores can lead to distorted IPS weights, thereby compromising the debiasing process in CVR prediction. In this paper, we introduce a model-agnostic calibration framework for propensity-based debiasing of CVR predictions. Theoretical analysis on bias and generalization bounds demonstrates the superiority of calibrated propensity estimates over uncalibrated ones. Experiments conducted on the Coat, Yahoo and KuaiRand datasets show improved uncertainty calibration, as evidenced by lower ECE values, leading to enhanced CVR prediction outcomes.
comment: This is the accepted manuscript version of the IEEE TKDE paper. The final published version will be available at: https://doi.ieeecomputersociety.org/10.1109/TKDE.2025.3552658
Robotics 40
☆ EmoACT: a Framework to Embed Emotions into Artificial Agents Based on Affect Control Theory
As robots and artificial agents become increasingly integrated into daily life, enhancing their ability to interact with humans is essential. Emotions, which play a crucial role in human interactions, can improve the naturalness and transparency of human-robot interactions (HRI) when embodied in artificial agents. This study aims to employ Affect Control Theory (ACT), a psychological model of emotions deeply rooted in interaction, for the generation of synthetic emotions. A platform-agnostic framework inspired by ACT was developed and implemented in a humanoid robot to assess its impact on human perception. Results show that the frequency of emotional displays impacts how users perceive the robot. Moreover, appropriate emotional expressions seem to enhance the robot's perceived emotional and cognitive agency. The findings suggest that ACT can be successfully employed to embed synthetic emotions into robots, resulting in effective human-robot interactions, where the robot is perceived more as a social agent than merely a machine.
☆ GripMap: An Efficient, Spatially Resolved Constraint Framework for Offline and Online Trajectory Planning in Autonomous Racing
Conventional trajectory planning approaches for autonomous vehicles often assume a fixed vehicle model that remains constant regardless of the vehicle's location. This overlooks the critical fact that the tires and the surface are the two force-transmitting partners in vehicle dynamics; while the tires stay with the vehicle, surface conditions vary with location. Recognizing these challenges, this paper presents a novel framework for spatially resolving dynamic constraints in both offline and online planning algorithms applied to autonomous racing. We introduce the GripMap concept, which provides a spatial resolution of vehicle dynamic constraints in the Frenet frame, allowing adaptation to locally varying grip conditions. This enables compensation for location-specific effects, more efficient vehicle behavior, and increased safety, unattainable with spatially invariant vehicle models. The focus is on low storage demand and quick access through perfect hashing. This framework proved advantageous in real-world applications in the presented form. Experiments inspired by autonomous racing demonstrate its effectiveness. In future work, this framework can serve as a foundational layer for developing future interpretable learning algorithms that adjust to varying grip conditions in real-time.
☆ An Extended Generalized Prandtl-Ishlinskii Hysteresis Model for I2RIS Robot
Retinal surgery requires extreme precision due to constrained anatomical spaces in the human retina. To assist surgeons achieve this level of accuracy, the Improved Integrated Robotic Intraocular Snake (I2RIS) with dexterous capability has been developed. However, such flexible tendon-driven robots often suffer from hysteresis problems, which significantly challenges precise control and positioning. In particular, we observed multi-stage hysteresis phenomena in the small-scale I2RIS. In this paper, we propose an Extended Generalized Prandtl-Ishlinskii (EGPI) model to increase the fitting accuracy of the hysteresis. The model incorporates a novel switching mechanism that enables it to describe multi-stage hysteresis in the regions of monotonic input. Experimental validation on I2RIS data demonstrate that the EGPI model outperforms the conventional Generalized Prandtl-Ishlinskii (GPI) model in terms of RMSE, NRMSE, and MAE across multiple motor input directions. The EGPI model in our study highlights the potential in modeling multi-stage hysteresis in minimally invasive flexible robots.
comment: Submitted to the 5th Modeling, Estimation and Control Conference (MECC 2025)
Self-Supervised Traversability Learning with Online Prototype Adaptation for Off-Road Autonomous Driving
Achieving reliable and safe autonomous driving in off-road environments requires accurate and efficient terrain traversability analysis. However, this task faces several challenges, including the scarcity of large-scale datasets tailored for off-road scenarios, the high cost and potential errors of manual annotation, the stringent real-time requirements of motion planning, and the limited computational power of onboard units. To address these challenges, this paper proposes a novel traversability learning method that leverages self-supervised learning, eliminating the need for manual annotation. For the first time, a Birds-Eye View (BEV) representation is used as input, reducing computational burden and improving adaptability to downstream motion planning. During vehicle operation, the proposed method conducts online analysis of traversed regions and dynamically updates prototypes to adaptively assess the traversability of the current environment, effectively handling dynamic scene changes. We evaluate our approach against state-of-the-art benchmarks on both public datasets and our own dataset, covering diverse seasons and geographical locations. Experimental results demonstrate that our method significantly outperforms recent approaches. Additionally, real-world vehicle experiments show that our method operates at 10 Hz, meeting real-time requirements, while a 5.5 km autonomous driving experiment further validates the generated traversability cost maps compatibility with downstream motion planning.
☆ Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions CVPR
Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges. This survey provides a wide-ranging examination of the anti-UAV domain, centering on three core objectives-classification, detection, and tracking-while detailing emerging methodologies such as diffusion-based data synthesis, multi-modal fusion, vision-language modeling, self-supervised learning, and reinforcement learning. We systematically evaluate state-of-the-art solutions across both single-modality and multi-sensor pipelines (spanning RGB, infrared, audio, radar, and RF) and discuss large-scale as well as adversarially oriented benchmarks. Our analysis reveals persistent gaps in real-time performance, stealth detection, and swarm-based scenarios, underscoring pressing needs for robust, adaptive anti-UAV systems. By highlighting open research directions, we aim to foster innovation and guide the development of next-generation defense strategies in an era marked by the extensive use of UAVs.
comment: Accepted at CVPR Workshop Anti-UAV 2025. 15 pages
☆ Exploring Video-Based Driver Activity Recognition under Noisy Labels
As an open research topic in the field of deep learning, learning with noisy labels has attracted much attention and grown rapidly over the past ten years. Learning with label noise is crucial for driver distraction behavior recognition, as real-world video data often contains mislabeled samples, impacting model reliability and performance. However, label noise learning is barely explored in the driver activity recognition field. In this paper, we propose the first label noise learning approach for the driver activity recognition task. Based on the cluster assumption, we initially enable the model to learn clustering-friendly low-dimensional representations from given videos and assign the resultant embeddings into clusters. We subsequently perform co-refinement within each cluster to smooth the classifier outputs. Furthermore, we propose a flexible sample selection strategy that combines two selection criteria without relying on any hyperparameters to filter clean samples from the training dataset. We also incorporate a self-adaptive parameter into the sample selection process to enforce balancing across classes. A comprehensive variety of experiments on the public Drive&Act dataset for all granularity levels demonstrates the superior performance of our method in comparison with other label-denoising methods derived from the image classification field. The source code is available at https://github.com/ilonafan/DAR-noisy-labels.
comment: The source code is available at https://github.com/ilonafan/DAR-noisy-labels
☆ A Graph-Based Reinforcement Learning Approach with Frontier Potential Based Reward for Safe Cluttered Environment Exploration IROS 2025
Autonomous exploration of cluttered environments requires efficient exploration strategies that guarantee safety against potential collisions with unknown random obstacles. This paper presents a novel approach combining a graph neural network-based exploration greedy policy with a safety shield to ensure safe navigation goal selection. The network is trained using reinforcement learning and the proximal policy optimization algorithm to maximize exploration efficiency while reducing the safety shield interventions. However, if the policy selects an infeasible action, the safety shield intervenes to choose the best feasible alternative, ensuring system consistency. Moreover, this paper proposes a reward function that includes a potential field based on the agent's proximity to unexplored regions and the expected information gain from reaching them. Overall, the approach investigated in this paper merges the benefits of the adaptability of reinforcement learning-driven exploration policies and the guarantee ensured by explicit safety mechanisms. Extensive evaluations in simulated environments demonstrate that the approach enables efficient and safe exploration in cluttered environments.
comment: 6 pages, 4 figures, submitted to the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025)
☆ Causality-enhanced Decision-Making for Autonomous Mobile Robots in Dynamic Environments
The growing integration of robots in shared environments -- such as warehouses, shopping centres, and hospitals -- demands a deep understanding of the underlying dynamics and human behaviours, including how, when, and where individuals engage in various activities and interactions. This knowledge goes beyond simple correlation studies and requires a more comprehensive causal analysis. By leveraging causal inference to model cause-and-effect relationships, we can better anticipate critical environmental factors and enable autonomous robots to plan and execute tasks more effectively. To this end, we propose a novel causality-based decision-making framework that reasons over a learned causal model to predict battery usage and human obstructions, understanding how these factors could influence robot task execution. Such reasoning framework assists the robot in deciding when and how to complete a given task. To achieve this, we developed also PeopleFlow, a new Gazebo-based simulator designed to model context-sensitive human-robot spatial interactions in shared workspaces. PeopleFlow features realistic human and robot trajectories influenced by contextual factors such as time, environment layout, and robot state, and can simulate a large number of agents. While the simulator is general-purpose, in this paper we focus on a warehouse-like environment as a case study, where we conduct an extensive evaluation benchmarking our causal approach against a non-causal baseline. Our findings demonstrate the efficacy of the proposed solutions, highlighting how causal reasoning enables autonomous robots to operate more efficiently and safely in dynamic environments shared with humans.
comment: Under review at The International Journal of Robotics Research (IJRR)
☆ Real-Time Shape Estimation of Tensegrity Structures Using Strut Inclination Angles
Tensegrity structures are becoming widely used in robotics, such as continuously bending soft manipulators and mobile robots to explore unknown and uneven environments dynamically. Estimating their shape, which is the foundation of their state, is essential for establishing control. However, on-board sensor-based shape estimation remains difficult despite its importance, because tensegrity structures lack well-defined joints, which makes it challenging to use conventional angle sensors such as potentiometers or encoders for shape estimation. To our knowledge, no existing work has successfully achieved shape estimation using only onboard sensors such as Inertial Measurement Units (IMUs). This study addresses this issue by proposing a novel approach that uses energy minimization to estimate the shape. We validated our method through experiments on a simple Class 1 tensegrity structure, and the results show that the proposed algorithm can estimate the real-time shape of the structure using onboard sensors, even in the presence of external disturbances.
☆ Towards Forceful Robotic Foundation Models: a Literature Survey
This article reviews contemporary methods for integrating force, including both proprioception and tactile sensing, in robot manipulation policy learning. We conduct a comparative analysis on various approaches for sensing force, data collection, behavior cloning, tactile representation learning, and low-level robot control. From our analysis, we articulate when and why forces are needed, and highlight opportunities to improve learning of contact-rich, generalist robot policies on the path toward highly capable touch-based robot foundation models. We generally find that while there are few tasks such as pouring, peg-in-hole insertion, and handling delicate objects, the performance of imitation learning models is not at a level of dynamics where force truly matters. Also, force and touch are abstract quantities that can be inferred through a wide range of modalities and are often measured and controlled implicitly. We hope that juxtaposing the different approaches currently in use will help the reader to gain a systemic understanding and help inspire the next generation of robot foundation models.
comment: 20 pages, 9 figures
☆ Multi-goal Rapidly Exploring Random Tree with Safety and Dynamic Constraints for UAV Cooperative Path Planning
Cooperative path planning is gaining its importance due to the increasing demand on using multiple unmanned aerial vehicles (UAVs) for complex missions. This work addresses the problem by introducing a new algorithm named MultiRRT that extends the rapidly exploring random tree (RRT) to generate paths for a group of UAVs to reach multiple goal locations at the same time. We first derive the dynamics constraint of the UAV and include it in the problem formulation. MultiRRT is then developed, taking into account the cooperative requirements and safe constraints during its path-searching process. The algorithm features two new mechanisms, node reduction and Bezier interpolation, to ensure the feasibility and optimality of the paths generated. Importantly, the interpolated paths are proven to meet the safety and dynamics constraints imposed by obstacles and the UAVs. A number of simulations, comparisons, and experiments have been conducted to evaluate the performance of the proposed approach. The results show that MultiRRT can generate collision-free paths for multiple UAVs to reach their goals with better scores in path length and smoothness metrics than state-of-the-art RRT variants including Theta-RRT, FN-RRT, RRT*, and RRT*-Smart. The generated paths are also tested in practical flights with real UAVs to evaluate their validity for cooperative tasks. The source code of the algorithm is available at https://github.com/duynamrcv/multi-target_RRT
☆ GrabS: Generative Embodied Agent for 3D Object Segmentation without Scene Supervision ICLR 2025
We study the hard problem of 3D object segmentation in complex point clouds without requiring human labels of 3D scenes for supervision. By relying on the similarity of pretrained 2D features or external signals such as motion to group 3D points as objects, existing unsupervised methods are usually limited to identifying simple objects like cars or their segmented objects are often inferior due to the lack of objectness in pretrained features. In this paper, we propose a new two-stage pipeline called GrabS. The core concept of our method is to learn generative and discriminative object-centric priors as a foundation from object datasets in the first stage, and then design an embodied agent to learn to discover multiple objects by querying against the pretrained generative priors in the second stage. We extensively evaluate our method on two real-world datasets and a newly created synthetic dataset, demonstrating remarkable segmentation performance, clearly surpassing all existing unsupervised methods.
comment: ICLR 2025 Spotlight. Code and data are available at: https://github.com/vLAR-group/GrabS
☆ Steerable rolling of a 1-DoF robot using an internal pendulum ICRA 2025
We present ROCK (Rolling One-motor Controlled rocK), a 1 degree-of-freedom robot consisting of a round shell and an internal pendulum. An uneven shell surface enables steering by using only the movement of the pendulum, allowing for mechanically simple designs that may be feasible to scale to large quantities or small sizes. We train a control policy using reinforcement learning in simulation and deploy it onto the robot to complete a rectangular trajectory.
comment: 2 pages, submitted to 2nd Unconventional Robots Workshop: Rethinking Robotic Systems Beyond Convention at IEEE ICRA 2025
☆ Inversion of biological strategies in engineering technology: in case underwater soft robot
This paper proposes a biomimetic design framework based on biological strategy inversion, aiming to systematically map solutions evolved in nature to the engineering field. By constructing a "Function-Behavior-Feature-Environment" (F-B-Cs in E) knowledge model, combined with natural language processing (NLP) and multi-criteria decision-making methods, it achieves efficient conversion from biological strategies to engineering solutions. Using underwater soft robot design as a case study, the effectiveness of the framework in optimizing drive mechanisms, power distribution, and motion pattern design is verified. This research provides scalable methodological support for interdisciplinary biomimetic innovation.
comment: 26 pages, 8 figures
☆ Safety with Agency: Human-Centered Safety Filter with Application to AI-Assisted Motorsports
We propose a human-centered safety filter (HCSF) for shared autonomy that significantly enhances system safety without compromising human agency. Our HCSF is built on a neural safety value function, which we first learn scalably through black-box interactions and then use at deployment to enforce a novel quality control barrier function (Q-CBF) safety constraint. Since this Q-CBF safety filter does not require any knowledge of the system dynamics for both synthesis and runtime safety monitoring and intervention, our method applies readily to complex, black-box shared autonomy systems. Notably, our HCSF's CBF-based interventions modify the human's actions minimally and smoothly, avoiding the abrupt, last-moment corrections delivered by many conventional safety filters. We validate our approach in a comprehensive in-person user study using Assetto Corsa-a high-fidelity car racing simulator with black-box dynamics-to assess robustness in "driving on the edge" scenarios. We compare both trajectory data and drivers' perceptions of our HCSF assistance against unassisted driving and a conventional safety filter. Experimental results show that 1) compared to having no assistance, our HCSF improves both safety and user satisfaction without compromising human agency or comfort, and 2) relative to a conventional safety filter, our proposed HCSF boosts human agency, comfort, and satisfaction while maintaining robustness.
comment: Accepted to Robotics: Science and Systems (RSS) 2025, 22 pages, 16 figures, 7 tables
☆ An Online Adaptation Method for Robust Depth Estimation and Visual Odometry in the Open World
Recently, learning-based robotic navigation systems have gained extensive research attention and made significant progress. However, the diversity of open-world scenarios poses a major challenge for the generalization of such systems to practical scenarios. Specifically, learned systems for scene measurement and state estimation tend to degrade when the application scenarios deviate from the training data, resulting to unreliable depth and pose estimation. Toward addressing this problem, this work aims to develop a visual odometry system that can fast adapt to diverse novel environments in an online manner. To this end, we construct a self-supervised online adaptation framework for monocular visual odometry aided by an online-updated depth estimation module. Firstly, we design a monocular depth estimation network with lightweight refiner modules, which enables efficient online adaptation. Then, we construct an objective for self-supervised learning of the depth estimation module based on the output of the visual odometry system and the contextual semantic information of the scene. Specifically, a sparse depth densification module and a dynamic consistency enhancement module are proposed to leverage camera poses and contextual semantics to generate pseudo-depths and valid masks for the online adaptation. Finally, we demonstrate the robustness and generalization capability of the proposed method in comparison with state-of-the-art learning-based approaches on urban, in-house datasets and a robot platform. Code is publicly available at: https://github.com/jixingwu/SOL-SLAM.
comment: 11 pages, 14 figures
☆ DM-OSVP++: One-Shot View Planning Using 3D Diffusion Models for Active RGB-Based Object Reconstruction
Active object reconstruction is crucial for many robotic applications. A key aspect in these scenarios is generating object-specific view configurations to obtain informative measurements for reconstruction. One-shot view planning enables efficient data collection by predicting all views at once, eliminating the need for time-consuming online replanning. Our primary insight is to leverage the generative power of 3D diffusion models as valuable prior information. By conditioning on initial multi-view images, we exploit the priors from the 3D diffusion model to generate an approximate object model, serving as the foundation for our view planning. Our novel approach integrates the geometric and textural distributions of the object model into the view planning process, generating views that focus on the complex parts of the object to be reconstructed. We validate the proposed active object reconstruction system through both simulation and real-world experiments, demonstrating the effectiveness of using 3D diffusion priors for one-shot view planning.
☆ Practical Insights on Grasp Strategies for Mobile Manipulation in the Wild IROS 2025
Mobile manipulation robots are continuously advancing, with their grasping capabilities rapidly progressing. However, there are still significant gaps preventing state-of-the-art mobile manipulators from widespread real-world deployments, including their ability to reliably grasp items in unstructured environments. To help bridge this gap, we developed SHOPPER, a mobile manipulation robot platform designed to push the boundaries of reliable and generalizable grasp strategies. We develop these grasp strategies and deploy them in a real-world grocery store -- an exceptionally challenging setting chosen for its vast diversity of manipulable items, fixtures, and layouts. In this work, we present our detailed approach to designing general grasp strategies towards picking any item in a real grocery store. Additionally, we provide an in-depth analysis of our latest real-world field test, discussing key findings related to fundamental failure modes over hundreds of distinct pick attempts. Through our detailed analysis, we aim to offer valuable practical insights and identify key grasping challenges, which can guide the robotics community towards pressing open problems in the field.
comment: 8 pages, 8 figures, submitted to IROS 2025
☆ Robust Visual Servoing under Human Supervision for Assembly Tasks
We propose a framework enabling mobile manipulators to reliably complete pick-and-place tasks for assembling structures from construction blocks. The picking uses an eye-in-hand visual servoing controller for object tracking with Control Barrier Functions (CBFs) to ensure fiducial markers in the blocks remain visible. An additional robot with an eye-to-hand setup ensures precise placement, critical for structural stability. We integrate human-in-the-loop capabilities for flexibility and fault correction and analyze robustness to camera pose errors, proposing adapted barrier functions to handle them. Lastly, experiments validate the framework on 6-DoF mobile arms.
comment: This work has been submitted to the IEEE for possible publication
☆ Learning Transferable Friction Models and LuGre Identification via Physics Informed Neural Networks
Accurately modeling friction in robotics remains a core challenge, as robotics simulators like Mujoco and PyBullet use simplified friction models or heuristics to balance computational efficiency with accuracy, where these simplifications and approximations can lead to substantial differences between simulated and physical performance. In this paper, we present a physics-informed friction estimation framework that enables the integration of well-established friction models with learnable components-requiring only minimal, generic measurement data. Our approach enforces physical consistency yet retains the flexibility to adapt to real-world complexities. We demonstrate, on an underactuated and nonlinear system, that the learned friction models, trained solely on small and noisy datasets, accurately simulate dynamic friction properties and reduce the sim-to-real gap. Crucially, we show that our approach enables the learned models to be transferable to systems they are not trained on. This ability to generalize across multiple systems streamlines friction modeling for complex, underactuated tasks, offering a scalable and interpretable path toward bridging the sim-to-real gap in robotics and control.
comment: 7 pages, 8 figures, Submitted to 2025 64th IEEE Conference on Decision and Control (CDC)
☆ Learning-based Delay Compensation for Enhanced Control of Assistive Soft Robots
Soft robots are increasingly used in healthcare, especially for assistive care, due to their inherent safety and adaptability. Controlling soft robots is challenging due to their nonlinear dynamics and the presence of time delays, especially in applications like a soft robotic arm for patient care. This paper presents a learning-based approach to approximate the nonlinear state predictor (Smith Predictor), aiming to improve tracking performance in a two-module soft robot arm with a short inherent input delay. The method uses Kernel Recursive Least Squares Tracker (KRLST) for online learning of the system dynamics and a Legendre Delay Network (LDN) to compress past input history for efficient delay compensation. Experimental results demonstrate significant improvement in tracking performance compared to a baseline model-based non-linear controller. Statistical analysis confirms the significance of the improvements. The method is computationally efficient and adaptable online, making it suitable for real-world scenarios and highlighting its potential for enabling safer and more accurate control of soft robots in assistive care applications.
☆ Diffusion Based Robust LiDAR Place Recognition ICRA 2025
Mobile robots on construction sites require accurate pose estimation to perform autonomous surveying and inspection missions. Localization in construction sites is a particularly challenging problem due to the presence of repetitive features such as flat plastered walls and perceptual aliasing due to apartments with similar layouts inter and intra floors. In this paper, we focus on the global re-positioning of a robot with respect to an accurate scanned mesh of the building solely using LiDAR data. In our approach, a neural network is trained on synthetic LiDAR point clouds generated by simulating a LiDAR in an accurate real-life large-scale mesh. We train a diffusion model with a PointNet++ backbone, which allows us to model multiple position candidates from a single LiDAR point cloud. The resulting model can successfully predict the global position of LiDAR in confined and complex sites despite the adverse effects of perceptual aliasing. The learned distribution of potential global positions can provide multi-modal position distribution. We evaluate our approach across five real-world datasets and show the place recognition accuracy of 77% +/-2m on average while outperforming baselines at a factor of 2 in mean error.
comment: accepted for ICRA 2025
♻ ☆ RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins (early version)
In the rapidly advancing field of robotics, dual-arm coordination and complex object manipulation are essential capabilities for developing advanced autonomous systems. However, the scarcity of diverse, high-quality demonstration data and real-world-aligned evaluation benchmarks severely limits such development. To address this, we introduce RoboTwin, a generative digital twin framework that uses 3D generative foundation models and large language models to produce diverse expert datasets and provide a real-world-aligned evaluation platform for dual-arm robotic tasks. Specifically, RoboTwin creates varied digital twins of objects from single 2D images, generating realistic and interactive scenarios. It also introduces a spatial relation-aware code generation framework that combines object annotations with large language models to break down tasks, determine spatial constraints, and generate precise robotic movement code. Our framework offers a comprehensive benchmark with both simulated and real-world data, enabling standardized evaluation and better alignment between simulated training and real-world performance. We validated our approach using the open-source COBOT Magic Robot platform. Policies pre-trained on RoboTwin-generated data and fine-tuned with limited real-world samples improve the success rate of over 70% for single-arm tasks and over 40% for dual-arm tasks compared to models trained solely on real-world data. This significant improvement demonstrates RoboTwin's potential to enhance the development and evaluation of dual-arm robotic manipulation systems. Project Page: https://robotwin-benchmark.github.io/early-version/.
comment: Project page: https://robotwin-benchmark.github.io/early-version/
♻ ☆ Decentralized Vehicle Coordination: The Berkeley DeepDrive Drone Dataset and Consensus-Based Models
A significant portion of roads, particularly in densely populated developing countries, lacks explicitly defined right-of-way rules. These understructured roads pose substantial challenges for autonomous vehicle motion planning, where efficient and safe navigation relies on understanding decentralized human coordination for collision avoidance. This coordination, often termed "social driving etiquette," remains underexplored due to limited open-source empirical data and suitable modeling frameworks. In this paper, we present a novel dataset and modeling framework designed to study motion planning in these understructured environments. The dataset includes 20 aerial videos of representative scenarios, an image dataset for training vehicle detection models, and a development kit for vehicle trajectory estimation. We demonstrate that a consensus-based modeling approach can effectively explain the emergence of priority orders observed in our dataset, and is therefore a viable framework for decentralized collision avoidance planning.
comment: 7 pages, 7 figures, 1 table
♻ ☆ SpiritSight Agent: Advanced GUI Agent with One Look CVPR 2025
Graphical User Interface (GUI) agents show amazing abilities in assisting human-computer interaction, automating human user's navigation on digital devices. An ideal GUI agent is expected to achieve high accuracy, low latency, and compatibility for different GUI platforms. Recent vision-based approaches have shown promise by leveraging advanced Vision Language Models (VLMs). While they generally meet the requirements of compatibility and low latency, these vision-based GUI agents tend to have low accuracy due to their limitations in element grounding. To address this issue, we propose $\textbf{SpiritSight}$, a vision-based, end-to-end GUI agent that excels in GUI navigation tasks across various GUI platforms. First, we create a multi-level, large-scale, high-quality GUI dataset called $\textbf{GUI-Lasagne}$ using scalable methods, empowering SpiritSight with robust GUI understanding and grounding capabilities. Second, we introduce the $\textbf{Universal Block Parsing (UBP)}$ method to resolve the ambiguity problem in dynamic high-resolution of visual inputs, further enhancing SpiritSight's ability to ground GUI objects. Through these efforts, SpiritSight agent outperforms other advanced methods on diverse GUI benchmarks, demonstrating its superior capability and compatibility in GUI navigation tasks. Models and datasets are available at https://hzhiyuan.github.io/SpiritSight-Agent.
comment: Paper accepted to CVPR 2025
♻ ☆ Exploring Self-supervised Skeleton-based Action Recognition in Occluded Environments IJCNN 2025
To integrate action recognition into autonomous robotic systems, it is essential to address challenges such as person occlusions-a common yet often overlooked scenario in existing self-supervised skeleton-based action recognition methods. In this work, we propose IosPSTL, a simple and effective self-supervised learning framework designed to handle occlusions. IosPSTL combines a cluster-agnostic KNN imputer with an Occluded Partial Spatio-Temporal Learning (OPSTL) strategy. First, we pre-train the model on occluded skeleton sequences. Then, we introduce a cluster-agnostic KNN imputer that performs semantic grouping using k-means clustering on sequence embeddings. It imputes missing skeleton data by applying K-Nearest Neighbors in the latent space, leveraging nearby sample representations to restore occluded joints. This imputation generates more complete skeleton sequences, which significantly benefits downstream self-supervised models. To further enhance learning, the OPSTL module incorporates Adaptive Spatial Masking (ASM) to make better use of intact, high-quality skeleton sequences during training. Our method achieves state-of-the-art performance on the occluded versions of the NTU-60 and NTU-120 datasets, demonstrating its robustness and effectiveness under challenging conditions. Code is available at https://github.com/cyfml/OPSTL.
comment: Accepted to IJCNN 2025. Code is available at https://github.com/cyfml/OPSTL
♻ ☆ Stochastic Trajectory Optimization for Robotic Skill Acquisition From a Suboptimal Demonstration
Learning from Demonstration (LfD) has emerged as a crucial method for robots to acquire new skills. However, when given suboptimal task trajectory demonstrations with shape characteristics reflecting human preferences but subpar dynamic attributes such as slow motion, robots not only need to mimic the behaviors but also optimize the dynamic performance. In this work, we leverage optimization-based methods to search for a superior-performing trajectory whose shape is similar to that of the demonstrated trajectory. Specifically, we use Dynamic Time Warping (DTW) to quantify the difference between two trajectories and combine it with additional performance metrics, such as collision cost, to construct the cost function. Moreover, we develop a multi-policy version of the Stochastic Trajectory Optimization for Motion Planning (STOMP), called MSTOMP, which is more stable and robust to parameter changes. To deal with the jitter in the demonstrated trajectory, we further utilize the gain-controlling method in the frequency domain to denoise the demonstration and propose a computationally more efficient metric, called Mean Square Error in the Spectrum (MSES), that measures the trajectories' differences in the frequency domain. We also theoretically highlight the connections between the time domain and the frequency domain methods. Finally, we verify our method in both simulation experiments and real-world experiments, showcasing its improved optimization performance and stability compared to existing methods. The source code can be found at https://ming-bot.github.io/MSTOMP.github.io.
♻ ☆ WMNav: Integrating Vision-Language Models into World Models for Object Goal Navigation
Object Goal Navigation-requiring an agent to locate a specific object in an unseen environment-remains a core challenge in embodied AI. Although recent progress in Vision-Language Model (VLM)-based agents has demonstrated promising perception and decision-making abilities through prompting, none has yet established a fully modular world model design that reduces risky and costly interactions with the environment by predicting the future state of the world. We introduce WMNav, a novel World Model-based Navigation framework powered by Vision-Language Models (VLMs). It predicts possible outcomes of decisions and builds memories to provide feedback to the policy module. To retain the predicted state of the environment, WMNav proposes the online maintained Curiosity Value Map as part of the world model memory to provide dynamic configuration for navigation policy. By decomposing according to a human-like thinking process, WMNav effectively alleviates the impact of model hallucination by making decisions based on the feedback difference between the world model plan and observation. To further boost efficiency, we implement a two-stage action proposer strategy: broad exploration followed by precise localization. Extensive evaluation on HM3D and MP3D validates WMNav surpasses existing zero-shot benchmarks in both success rate and exploration efficiency (absolute improvement: +3.2% SR and +3.2% SPL on HM3D, +13.5% SR and +1.1% SPL on MP3D). Project page: https://b0b8k1ng.github.io/WMNav/.
comment: 8 pages, 5 figures
♻ ☆ ADAPT: An Autonomous Forklift for Construction Site Operation
Efficient material logistics play a critical role in controlling costs and schedules in the construction industry. However, manual material handling remains prone to inefficiencies, delays, and safety risks. Autonomous forklifts offer a promising solution to streamline on-site logistics, reducing reliance on human operators and mitigating labor shortages. This paper presents the development and evaluation of ADAPT (Autonomous Dynamic All-terrain Pallet Transporter), a fully autonomous off-road forklift designed for construction environments. Unlike structured warehouse settings, construction sites pose significant challenges, including dynamic obstacles, unstructured terrain, and varying weather conditions. To address these challenges, our system integrates AI-driven perception techniques with traditional approaches for decision making, planning, and control, enabling reliable operation in complex environments. We validate the system through extensive real-world testing, comparing its continuous performance against an experienced human operator across various weather conditions. Our findings demonstrate that autonomous outdoor forklifts can operate near human-level performance, offering a viable path toward safer and more efficient construction logistics.
♻ ☆ A Real-time Anomaly Detection Method for Robots based on a Flexible and Sparse Latent Space
The growing demand for robots to operate effectively in diverse environments necessitates the need for robust real-time anomaly detection techniques during robotic operations. However, deep learning-based models in robotics face significant challenges due to limited training data and highly noisy signal features. In this paper, we present Sparse Masked Autoregressive Flow-based Adversarial AutoEncoders model to address these problems. This approach integrates Masked Autoregressive Flow model into Adversarial AutoEncoders to construct a flexible latent space and utilize Sparse autoencoder to efficiently focus on important features, even in scenarios with limited feature space. Our experiments demonstrate that the proposed model achieves a 4.96% to 9.75% higher area under the receiver operating characteristic curve for pick-and-place robotic operations with randomly placed cans, compared to existing state-of-the-art methods. Notably, it showed up to 19.67% better performance in scenarios involving collisions with lightweight objects. Additionally, unlike the existing state-of-the-art model, our model performs inferences within 1 millisecond, ensuring real-time anomaly detection. These capabilities make our model highly applicable to machine learning-based robotic safety systems in dynamic environments. The code will be made publicly available after acceptance.
comment: 20 pages, 11 figures
♻ ☆ $π$-MPPI: A Projection-based Model Predictive Path Integral Scheme for Smooth Optimal Control of Fixed-Wing Aerial Vehicles
Model Predictive Path Integral (MPPI) is a popular sampling-based Model Predictive Control (MPC) algorithm for nonlinear systems. It optimizes trajectories by sampling control sequences and averaging them. However, a key issue with MPPI is the non-smoothness of the optimal control sequence, leading to oscillations in systems like fixed-wing aerial vehicles (FWVs). Existing solutions use post-hoc smoothing, which fails to bound control derivatives. This paper introduces a new approach: we add a projection filter $\pi$ to minimally correct control samples, ensuring bounds on control magnitude and higher-order derivatives. The filtered samples are then averaged using MPPI, leading to our $\pi$-MPPI approach. We minimize computational overhead by using a neural accelerated custom optimizer for the projection filter. $\pi$-MPPI offers a simple way to achieve arbitrary smoothness in control sequences. While we focus on FWVs, this projection filter can be integrated into any MPPI pipeline. Applied to FWVs, $\pi$-MPPI is easier to tune than the baseline, resulting in smoother, more robust performance.
comment: 8 pages, 4 figures, submitted to IEEE RA-L
♻ ☆ Geometry-aware RL for Manipulation of Varying Shapes and Deformable Objects ICLR 2025
Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hanging require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that comprises smaller sub-graphs, such as actuators and objects, accompanied by different edge types describing their interactions. This graph representation serves as a unified structure for both rigid and deformable objects tasks, and can be extended further to tasks comprising multiple actuators. To evaluate this setup, we present a novel and challenging reinforcement learning benchmark, including rigid insertion of diverse objects, as well as rope and cloth manipulation with multiple end-effectors. These tasks present a large search space, as both the initial and target configurations are uniformly sampled in 3D space. To address this issue, we propose a novel graph-based policy model, dubbed Heterogeneous Equivariant Policy (HEPi), utilizing $SE(3)$ equivariant message passing networks as the main backbone to exploit the geometric symmetry. In addition, by modeling explicit heterogeneity, HEPi can outperform Transformer-based and non-heterogeneous equivariant policies in terms of average returns, sample efficiency, and generalization to unseen objects. Our project page is available at https://thobotics.github.io/hepi.
comment: Accepted at ICLR 2025 (Oral)
♻ ☆ SceneFactory: A Workflow-centric and Unified Framework for Incremental Scene Modeling
We present SceneFactory, a workflow-centric and unified framework for incremental scene modeling, that conveniently supports a wide range of applications, such as (unposed and/or uncalibrated) multi-view depth estimation, LiDAR completion, (dense) RGB-D/RGB-L/Mono/Depth-only reconstruction and SLAM. The workflow-centric design uses multiple blocks as the basis for constructing different production lines. The supported applications, i.e., productions avoid redundancy in their designs. Thus, the focus is placed on each block itself for independent expansion. To support all input combinations, our implementation consists of four building blocks that form SceneFactory: (1) tracking, (2) flexion, (3) depth estimation, and (4) scene reconstruction. The tracking block is based on Mono SLAM and is extended to support RGB-D and RGB-LiDAR (RGB-L) inputs. Flexion is used to convert the depth image (untrackable) into a trackable image. For general-purpose depth estimation, we propose an unposed \& uncalibrated multi-view depth estimation model (U$^2$-MVD) to estimate dense geometry. U$^2$-MVD exploits dense bundle adjustment to solve for poses, intrinsics, and inverse depth. A semantic-aware ScaleCov step is then introduced to complete the multi-view depth. Relying on U$^2$-MVD, SceneFactory both supports user-friendly 3D creation (with just images) and bridges the applications of Dense RGB-D and Dense Mono. For high-quality surface and color reconstruction, we propose Dual-purpose Multi-resolutional Neural Points (DM-NPs) for the first surface accessible Surface Color Field design, where we introduce Improved Point Rasterization (IPR) for point cloud based surface query. ...
comment: Accepted to IEEE Transactions on Robotics (T-RO). For project page and code, please find https://jarrome.github.io/SceneFactory/
♻ ☆ H2O+: An Improved Framework for Hybrid Offline-and-Online RL with Dynamics Gaps ICRA 2025
Solving real-world complex tasks using reinforcement learning (RL) without high-fidelity simulation environments or large amounts of offline data can be quite challenging. Online RL agents trained in imperfect simulation environments can suffer from severe sim-to-real issues. Offline RL approaches although bypass the need for simulators, often pose demanding requirements on the size and quality of the offline datasets. The recently emerged hybrid offline-and-online RL provides an attractive framework that enables joint use of limited offline data and imperfect simulator for transferable policy learning. In this paper, we develop a new algorithm, called H2O+, which offers great flexibility to bridge various choices of offline and online learning methods, while also accounting for dynamics gaps between the real and simulation environment. Through extensive simulation and real-world robotics experiments, we demonstrate superior performance and flexibility over advanced cross-domain online and offline RL algorithms.
comment: ICRA 2025
♻ ☆ RoboMorph: Evolving Robot Morphology using Large Language Models
We introduce RoboMorph, an automated approach for generating and optimizing modular robot designs using large language models (LLMs) and evolutionary algorithms. In this framework, we represent each robot design as a grammar and leverage the capabilities of LLMs to navigate the extensive robot design space, which is traditionally time-consuming and computationally demanding. By introducing a best-shot prompting technique and a reinforcement learning-based control algorithm, RoboMorph iteratively improves robot designs through feedback loops. Experimental results demonstrate that RoboMorph successfully generates nontrivial robots optimized for different terrains while showcasing improvements in robot morphology over successive evolutions. Our approach highlights the potential of using LLMs for data-driven, modular robot design, providing a promising methodology that can be extended to other domains with similar design frameworks.
♻ ☆ Inferring Short-Sightedness in Dynamic Noncooperative Games
Dynamic game theory is an increasingly popular tool for modeling multi-agent, e.g. human-robot, interactions. Game-theoretic models presume that each agent wishes to minimize a private cost function that depends on others' actions. These games typically evolve over a fixed time horizon, specifying how far into the future each agent plans. In practical settings, however, decision-makers may vary in foresightedness. We conjecture that quantifying and estimating each agent's foresightedness from online data will enable safer and more efficient interactions with other agents. To this end, we frame this inference problem as an \emph{inverse} dynamic game. We consider a specific parametrization of each agent's objective function that smoothly interpolates myopic and farsighted planning. Games of this form are readily transformed into parametric mixed complementarity problems; we exploit the directional differentiability of solutions to these problems with respect to their hidden parameters to solve for agents' foresightedness. We conduct two types of experiments: one with synthetically generated pedestrian motion at a crosswalk and the other with real-world intersection data involving people walking, biking, and driving vehicles. The results of these experiments demonstrate that explicitly inferring agents' foresightedness enables game-theoretic models to more accurately model agents' behavior. Specifically, our results show 33% more accurate prediction of foresighted behavior on average compared to the baseline in real-world scenarios.
♻ ☆ Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets
Imitation learning has emerged as a promising approach towards building generalist robots. However, scaling imitation learning for large robot foundation models remains challenging due to its reliance on high-quality expert demonstrations. Meanwhile, large amounts of video data depicting a wide range of environments and diverse behaviors are readily available. This data provides a rich source of information about real-world dynamics and agent-environment interactions. Leveraging this data directly for imitation learning, however, has proven difficult due to the lack of action annotation required for most contemporary methods. In this work, we present Unified World Models (UWM), a framework that allows for leveraging both video and action data for policy learning. Specifically, a UWM integrates an action diffusion process and a video diffusion process within a unified transformer architecture, where independent diffusion timesteps govern each modality. By simply controlling each diffusion timestep, UWM can flexibly represent a policy, a forward dynamics, an inverse dynamics, and a video generator. Through simulated and real-world experiments, we show that: (1) UWM enables effective pretraining on large-scale multitask robot datasets with both dynamics and action predictions, resulting in more generalizable and robust policies than imitation learning, (2) UWM naturally facilitates learning from action-free video data through independent control of modality-specific diffusion timesteps, further improving the performance of finetuned policies. Our results suggest that UWM offers a promising step toward harnessing large, heterogeneous datasets for scalable robot learning, and provides a simple unification between the often disparate paradigms of imitation learning and world modeling. Videos and code are available at https://weirdlabuw.github.io/uwm/.
♻ ☆ Analysis of Deep Learning-Based Colorization and Super-Resolution Techniques for Lidar Imagery
Modern lidar systems can produce not only dense point clouds but also 360 degrees low-resolution images. This advancement facilitates the application of deep learning (DL) techniques initially developed for conventional RGB cameras and simplifies fusion of point cloud data and images without complex processes like lidar-camera calibration. Compared to RGB images from traditional cameras, lidar-generated images show greater robustness under low-light and harsh conditions, such as foggy weather. However, these images typically have lower resolution and often appear overly dark. While various studies have explored DL-based computer vision tasks such as object detection, segmentation, and keypoint detection on lidar imagery, other potentially valuable techniques remain underexplored. This paper provides a comprehensive review and qualitative analysis of DL-based colorization and super-resolution methods applied to lidar imagery. Additionally, we assess the computational performance of these approaches, offering insights into their suitability for downstream robotic and autonomous system applications like odometry and 3D reconstruction.
comment: 6 pages
♻ ☆ Submodular Optimization for Keyframe Selection & Usage in SLAM ICRA
Keyframes are LiDAR scans saved for future reference in Simultaneous Localization And Mapping (SLAM), but despite their central importance most algorithms leave choices of which scans to save and how to use them to wasteful heuristics. This work proposes two novel keyframe selection strategies for localization and map summarization, as well as a novel approach to submap generation which selects keyframes that best constrain localization. Our results show that online keyframe selection and submap generation reduce the number of saved keyframes and improve per scan computation time without compromising localization performance. We also present a map summarization feature for quickly capturing environments under strict map size constraints.
comment: Accepted to the International Conference on Robotics and Automation (ICRA) 2025
♻ ☆ GroundSLAM: A Robust Visual SLAM System for Warehouse Robots Using Ground Textures
A robust visual localization and mapping system is essential for warehouse robot navigation, as cameras offer a more cost-effective alternative to LiDAR sensors. However, existing forward-facing camera systems often encounter challenges in dynamic environments and open spaces, leading to significant performance degradation during deployment. To address these limitations, a localization system utilizing a single downward-facing camera to capture ground textures presents a promising solution. Nevertheless, existing feature-based ground-texture localization methods face difficulties when operating on surfaces with sparse features or repetitive patterns. To address this limitation, we propose GroundSLAM, a novel feature-free and ground-texture-based simultaneous localization and mapping (SLAM) system. GroundSLAM consists of three components: feature-free visual odometry, ground-texture-based loop detection and map optimization, and map reuse. Specifically, we introduce a kernel cross-correlator (KCC) for image-level pose tracking, loop detection, and map reuse to improve localization accuracy and robustness, and incorporate adaptive pruning strategies to enhance efficiency. Due to these specific designs, GroundSLAM is able to deliver efficient and stable localization across various ground surfaces such as those with sparse features and repetitive patterns. To advance research in this area, we introduce the first ground-texture dataset with precise ground-truth poses, consisting of 131k images collected from 10 kinds of indoor and outdoor ground surfaces. Extensive experimental results show that GroundSLAM outperforms state-of-the-art methods for both indoor and outdoor localization. We release our code and dataset at https://github.com/sair-lab/GroundSLAM.
Artificial Intelligence 138
☆ Adapting a World Model for Trajectory Following in a 3D Game
Imitation learning is a powerful tool for training agents by leveraging expert knowledge, and being able to replicate a given trajectory is an integral part of it. In complex environments, like modern 3D video games, distribution shift and stochasticity necessitate robust approaches beyond simple action replay. In this study, we apply Inverse Dynamics Models (IDM) with different encoders and policy heads to trajectory following in a modern 3D video game -- Bleeding Edge. Additionally, we investigate several future alignment strategies that address the distribution shift caused by the aleatoric uncertainty and imperfections of the agent. We measure both the trajectory deviation distance and the first significant deviation point between the reference and the agent's trajectory and show that the optimal configuration depends on the chosen setting. Our results show that in a diverse data setting, a GPT-style policy head with an encoder trained from scratch performs the best, DINOv2 encoder with the GPT-style policy head gives the best results in the low data regime, and both GPT-style and MLP-style policy heads had comparable results when pre-trained on a diverse setting and fine-tuned for a specific behaviour setting.
☆ SHeaP: Self-Supervised Head Geometry Predictor Learned via 2D Gaussians
Accurate, real-time 3D reconstruction of human heads from monocular images and videos underlies numerous visual applications. As 3D ground truth data is hard to come by at scale, previous methods have sought to learn from abundant 2D videos in a self-supervised manner. Typically, this involves the use of differentiable mesh rendering, which is effective but faces limitations. To improve on this, we propose SHeaP (Self-supervised Head Geometry Predictor Learned via 2D Gaussians). Given a source image, we predict a 3DMM mesh and a set of Gaussians that are rigged to this mesh. We then reanimate this rigged head avatar to match a target frame, and backpropagate photometric losses to both the 3DMM and Gaussian prediction networks. We find that using Gaussians for rendering substantially improves the effectiveness of this self-supervised approach. Training solely on 2D data, our method surpasses existing self-supervised approaches in geometric evaluations on the NoW benchmark for neutral faces and a new benchmark for non-neutral expressions. Our method also produces highly expressive meshes, outperforming state-of-the-art in emotion classification.
comment: For video demonstrations and additional materials please see https://nlml.github.io/sheap/
☆ How Do I Do That? Synthesizing 3D Hand Motion and Contacts for Everyday Interactions CVPR 2025
We tackle the novel problem of predicting 3D hand motion and contact maps (or Interaction Trajectories) given a single RGB view, action text, and a 3D contact point on the object as input. Our approach consists of (1) Interaction Codebook: a VQVAE model to learn a latent codebook of hand poses and contact points, effectively tokenizing interaction trajectories, (2) Interaction Predictor: a transformer-decoder module to predict the interaction trajectory from test time inputs by using an indexer module to retrieve a latent affordance from the learned codebook. To train our model, we develop a data engine that extracts 3D hand poses and contact trajectories from the diverse HoloAssist dataset. We evaluate our model on a benchmark that is 2.5-10X larger than existing works, in terms of diversity of objects and interactions observed, and test for generalization of the model across object categories, action categories, tasks, and scenes. Experimental results show the effectiveness of our approach over transformer & diffusion baselines across all settings.
comment: CVPR 2025, Project page: https://ap229997.github.io/projects/latentact
☆ HLS-Eval: A Benchmark and Framework for Evaluating LLMs on High-Level Synthesis Design Tasks
The rapid scaling of large language model (LLM) training and inference has driven their adoption in semiconductor design across academia and industry. While most prior work evaluates LLMs on hardware description language (HDL) tasks, particularly Verilog, designers are increasingly using high-level synthesis (HLS) to build domain-specific accelerators and complex hardware systems. However, benchmarks and tooling to comprehensively evaluate LLMs for HLS design tasks remain scarce. To address this, we introduce HLS-Eval, the first complete benchmark and evaluation framework for LLM-driven HLS design. HLS-Eval targets two core tasks: (1) generating HLS code from natural language descriptions, and (2) performing HLS-specific code edits to optimize performance and hardware efficiency. The benchmark includes 94 unique designs drawn from standard HLS benchmarks and novel sources. Each case is prepared via a semi-automated flow that produces a natural language description and a paired testbench for C-simulation and synthesis validation, ensuring each task is "LLM-ready." Beyond the benchmark, HLS-Eval offers a modular Python framework for automated, parallel evaluation of both local and hosted LLMs. It includes a parallel evaluation engine, direct HLS tool integration, and abstractions for to support different LLM interaction paradigms, enabling rapid prototyping of new benchmarks, tasks, and LLM methods. We demonstrate HLS-Eval through baseline evaluations of open-source LLMs on Vitis HLS, measuring outputs across four key metrics - parseability, compilability, runnability, and synthesizability - reflecting the iterative HLS design cycle. We also report pass@k metrics, establishing clear baselines and reusable infrastructure for the broader LLM-for-hardware community. All benchmarks, framework code, and results are open-sourced at https://github.com/stefanpie/hls-eval.
☆ SCENT: Robust Spatiotemporal Learning for Continuous Scientific Data via Scalable Conditioned Neural Fields
Spatiotemporal learning is challenging due to the intricate interplay between spatial and temporal dependencies, the high dimensionality of the data, and scalability constraints. These challenges are further amplified in scientific domains, where data is often irregularly distributed (e.g., missing values from sensor failures) and high-volume (e.g., high-fidelity simulations), posing additional computational and modeling difficulties. In this paper, we present SCENT, a novel framework for scalable and continuity-informed spatiotemporal representation learning. SCENT unifies interpolation, reconstruction, and forecasting within a single architecture. Built on a transformer-based encoder-processor-decoder backbone, SCENT introduces learnable queries to enhance generalization and a query-wise cross-attention mechanism to effectively capture multi-scale dependencies. To ensure scalability in both data size and model complexity, we incorporate a sparse attention mechanism, enabling flexible output representations and efficient evaluation at arbitrary resolutions. We validate SCENT through extensive simulations and real-world experiments, demonstrating state-of-the-art performance across multiple challenging tasks while achieving superior scalability.
comment: 25 pages, 5 main figures, 3 tables, under review
☆ FLIP Reasoning Challenge ICLR 2025
Over the past years, advances in artificial intelligence (AI) have demonstrated how AI can solve many perception and generation tasks, such as image classification and text writing, yet reasoning remains a challenge. This paper introduces the FLIP dataset, a benchmark for evaluating AI reasoning capabilities based on human verification tasks on the Idena blockchain. FLIP challenges present users with two orderings of 4 images, requiring them to identify the logically coherent one. By emphasizing sequential reasoning, visual storytelling, and common sense, FLIP provides a unique testbed for multimodal AI systems. Our experiments evaluate state-of-the-art models, leveraging both vision-language models (VLMs) and large language models (LLMs). Results reveal that even the best open-sourced and closed-sourced models achieve maximum accuracies of 75.5% and 77.9%, respectively, in zero-shot settings, compared to human performance of 95.3%. Captioning models aid reasoning models by providing text descriptions of images, yielding better results than when using the raw images directly, 69.6% vs. 75.2% for Gemini 1.5 Pro. Combining the predictions from 15 models in an ensemble increases the accuracy to 85.2%. These findings highlight the limitations of existing reasoning models and the need for robust multimodal benchmarks like FLIP. The full codebase and dataset will be available at https://github.com/aplesner/FLIP-Reasoning-Challenge.
comment: Published at First Workshop on Open Science for Foundation Models at ICLR 2025
☆ Advancing Arabic Speech Recognition Through Large-Scale Weakly Supervised Learning
Automatic speech recognition (ASR) is crucial for human-machine interaction in diverse applications like conversational agents, industrial robotics, call center automation, and automated subtitling. However, developing high-performance ASR models remains challenging, particularly for low-resource languages like Arabic, due to the scarcity of large, labeled speech datasets, which are costly and labor-intensive to produce. In this work, we employ weakly supervised learning to train an Arabic ASR model using the Conformer architecture. Our model is trained from scratch on 15,000 hours of weakly annotated speech data covering both Modern Standard Arabic (MSA) and Dialectal Arabic (DA), eliminating the need for costly manual transcriptions. Despite the absence of human-verified labels, our approach attains state-of-the-art (SOTA) performance, exceeding all previous efforts in the field of Arabic ASR on the standard benchmarks. By demonstrating the effectiveness of weak supervision as a scalable, cost-efficient alternative to traditional supervised approaches, paving the way for improved ASR systems in low resource settings.
☆ Uncertainty-Guided Coarse-to-Fine Tumor Segmentation with Anatomy-Aware Post-Processing SC
Reliable tumor segmentation in thoracic computed tomography (CT) remains challenging due to boundary ambiguity, class imbalance, and anatomical variability. We propose an uncertainty-guided, coarse-to-fine segmentation framework that combines full-volume tumor localization with refined region-of-interest (ROI) segmentation, enhanced by anatomically aware post-processing. The first-stage model generates a coarse prediction, followed by anatomically informed filtering based on lung overlap, proximity to lung surfaces, and component size. The resulting ROIs are segmented by a second-stage model trained with uncertainty-aware loss functions to improve accuracy and boundary calibration in ambiguous regions. Experiments on private and public datasets demonstrate improvements in Dice and Hausdorff scores, with fewer false positives and enhanced spatial interpretability. These results highlight the value of combining uncertainty modeling and anatomical priors in cascaded segmentation pipelines for robust and clinically meaningful tumor delineation. On the Orlando dataset, our framework improved Swin UNETR Dice from 0.4690 to 0.6447. Reduction in spurious components was strongly correlated with segmentation gains, underscoring the value of anatomically informed post-processing.
comment: 6 pages, 2 figures, to appear in IEEE ADSCA 2025
☆ Communication Optimization for Decentralized Learning atop Bandwidth-limited Edge Networks
Decentralized federated learning (DFL) is a promising machine learning paradigm for bringing artificial intelligence (AI) capabilities to the network edge. Running DFL on top of edge networks, however, faces severe performance challenges due to the extensive parameter exchanges between agents. Most existing solutions for these challenges were based on simplistic communication models, which cannot capture the case of learning over a multi-hop bandwidth-limited network. In this work, we address this problem by jointly designing the communication scheme for the overlay network formed by the agents and the mixing matrix that controls the communication demands between the agents. By carefully analyzing the properties of our problem, we cast each design problem into a tractable optimization and develop an efficient algorithm with guaranteed performance. Our evaluations based on real topology and data show that the proposed algorithm can reduce the total training time by over $80\%$ compared to the baseline without sacrificing accuracy, while significantly improving the computational efficiency over the state of the art.
comment: arXiv admin note: text overlap with arXiv:2408.04705
☆ From Requirements to Architecture: Semi-Automatically Generating Software Architectures
To support junior and senior architects, I propose developing a new architecture creation method that leverages LLMs' evolving capabilities to support the architect. This method involves the architect's close collaboration with LLM-fueled tooling over the whole process. The architect is guided through Domain Model creation, Use Case specification, architectural decisions, and architecture evaluation. While the architect can take complete control of the process and the results, and use the tooling as a building set, they can follow the intended process for maximum tooling support. The preliminary results suggest the feasibility of this process and indicate major time savings for the architect.
comment: to be published in EMISA 2025
☆ What Do Large Language Models Know? Tacit Knowledge as a Potential Causal-Explanatory Structure
It is sometimes assumed that Large Language Models (LLMs) know language, or for example that they know that Paris is the capital of France. But what -- if anything -- do LLMs actually know? In this paper, I argue that LLMs can acquire tacit knowledge as defined by Martin Davies (1990). Whereas Davies himself denies that neural networks can acquire tacit knowledge, I demonstrate that certain architectural features of LLMs satisfy the constraints of semantic description, syntactic structure, and causal systematicity. Thus, tacit knowledge may serve as a conceptual framework for describing, explaining, and intervening on LLMs and their behavior.
comment: Accepted for publication in Philosophy of Science
☆ SALAD: Improving Robustness and Generalization through Contrastive Learning with Structure-Aware and LLM-Driven Augmented Data NAACL 2025
In various natural language processing (NLP) tasks, fine-tuning Pre-trained Language Models (PLMs) often leads to the issue of spurious correlations, which negatively impacts performance, particularly when dealing with out-of-distribution data. To address this problem, we propose SALAD}(Structure Aware and LLM-driven Augmented Data), a novel approach designed to enhance model robustness and generalization by generating structure-aware and counterfactually augmented data for contrastive learning. Our method leverages a tagging-based approach to generate structure-aware positive samples and utilizes large language models (LLMs) to generate counterfactual negative samples with diverse sentence patterns. By applying contrastive learning, SALAD enables the model to focus on learning the structural relationships between key sentence components while minimizing reliance on spurious correlations. We validate our approach through experiments on three tasks: Sentiment Classification, Sexism Detection, and Natural Language Inference. The results demonstrate that SALAD not only improves model robustness and performance across different environments but also enhances generalization to out-of-distribution datasets and cross-domain scenarios.
comment: Accepted to NAACL 2025 main. 15 pages, 4 figures
☆ Trusting CHATGPT: how minor tweaks in the prompts lead to major differences in sentiment classification
One fundamental question for the social sciences today is: how much can we trust highly complex predictive models like ChatGPT? This study tests the hypothesis that subtle changes in the structure of prompts do not produce significant variations in the classification results of sentiment polarity analysis generated by the Large Language Model GPT-4o mini. Using a dataset of 100.000 comments in Spanish on four Latin American presidents, the model classified the comments as positive, negative, or neutral on 10 occasions, varying the prompts slightly each time. The experimental methodology included exploratory and confirmatory analyses to identify significant discrepancies among classifications. The results reveal that even minor modifications to prompts such as lexical, syntactic, or modal changes, or even their lack of structure impact the classifications. In certain cases, the model produced inconsistent responses, such as mixing categories, providing unsolicited explanations, or using languages other than Spanish. Statistical analysis using Chi-square tests confirmed significant differences in most comparisons between prompts, except in one case where linguistic structures were highly similar. These findings challenge the robustness and trust of Large Language Models for classification tasks, highlighting their vulnerability to variations in instructions. Moreover, it was evident that the lack of structured grammar in prompts increases the frequency of hallucinations. The discussion underscores that trust in Large Language Models is based not only on technical performance but also on the social and institutional relationships underpinning their use.
comment: in Spanish language
☆ Mapping Controversies Using Artificial Intelligence: An Analysis of the Hamas-Israel Conflict on YouTube
This article analyzes the Hamas-Israel controversy through 253,925 Spanish-language YouTube comments posted between October 2023 and January 2024, following the October 7 attack that escalated the conflict. Adopting an interdisciplinary approach, the study combines the analysis of controversies from Science and Technology Studies (STS) with advanced computational methodologies, specifically Natural Language Processing (NLP) using the BERT (Bidirectional Encoder Representations from Transformers) model. Using this approach, the comments were automatically classified into seven categories, reflecting pro-Palestinian, pro-Israeli, anti- Palestinian, anti-Israeli positions, among others. The results show a predominance of pro- Palestinian comments, although pro-Israeli and anti-Palestinian comments received more "likes." This study also applies the agenda-setting theory to demonstrate how media coverage significantly influences public perception, observing a notable shift in public opinion, transitioning from a pro- Palestinian stance to a more critical position towards Israel. This work highlights the importance of combining social science perspectives with technological tools in the analysis of controversies, presenting a methodological innovation by integrating computational analysis with critical social theories to address complex public opinion phenomena and media narratives.
comment: in Spanish language
☆ Poem Meter Classification of Recited Arabic Poetry: Integrating High-Resource Systems for a Low-Resource Task
Arabic poetry is an essential and integral part of Arabic language and culture. It has been used by the Arabs to spot lights on their major events such as depicting brutal battles and conflicts. They also used it, as in many other languages, for various purposes such as romance, pride, lamentation, etc. Arabic poetry has received major attention from linguistics over the decades. One of the main characteristics of Arabic poetry is its special rhythmic structure as opposed to prose. This structure is referred to as a meter. Meters, along with other poetic characteristics, are intensively studied in an Arabic linguistic field called "\textit{Aroud}". Identifying these meters for a verse is a lengthy and complicated process. It also requires technical knowledge in \textit{Aruod}. For recited poetry, it adds an extra layer of processing. Developing systems for automatic identification of poem meters for recited poems need large amounts of labelled data. In this study, we propose a state-of-the-art framework to identify the poem meters of recited Arabic poetry, where we integrate two separate high-resource systems to perform the low-resource task. To ensure generalization of our proposed architecture, we publish a benchmark for this task for future research.
☆ Towards Explainable Fusion and Balanced Learning in Multimodal Sentiment Analysis
Multimodal Sentiment Analysis (MSA) faces two critical challenges: the lack of interpretability in the decision logic of multimodal fusion and modality imbalance caused by disparities in inter-modal information density. To address these issues, we propose KAN-MCP, a novel framework that integrates the interpretability of Kolmogorov-Arnold Networks (KAN) with the robustness of the Multimodal Clean Pareto (MCPareto) framework. First, KAN leverages its univariate function decomposition to achieve transparent analysis of cross-modal interactions. This structural design allows direct inspection of feature transformations without relying on external interpretation tools, thereby ensuring both high expressiveness and interpretability. Second, the proposed MCPareto enhances robustness by addressing modality imbalance and noise interference. Specifically, we introduce the Dimensionality Reduction and Denoising Modal Information Bottleneck (DRD-MIB) method, which jointly denoises and reduces feature dimensionality. This approach provides KAN with discriminative low-dimensional inputs to reduce the modeling complexity of KAN while preserving critical sentiment-related information. Furthermore, MCPareto dynamically balances gradient contributions across modalities using the purified features output by DRD-MIB, ensuring lossless transmission of auxiliary signals and effectively alleviating modality imbalance. This synergy of interpretability and robustness not only achieves superior performance on benchmark datasets such as CMU-MOSI, CMU-MOSEI, and CH-SIMS v2 but also offers an intuitive visualization interface through KAN's interpretable architecture.
☆ ARCeR: an Agentic RAG for the Automated Definition of Cyber Ranges
The growing and evolving landscape of cybersecurity threats necessitates the development of supporting tools and platforms that allow for the creation of realistic IT environments operating within virtual, controlled settings as Cyber Ranges (CRs). CRs can be exploited for analyzing vulnerabilities and experimenting with the effectiveness of devised countermeasures, as well as serving as training environments for building cyber security skills and abilities for IT operators. This paper proposes ARCeR as an innovative solution for the automatic generation and deployment of CRs, starting from user-provided descriptions in a natural language. ARCeR relies on the Agentic RAG paradigm, which allows it to fully exploit state-of-art AI technologies. Experimental results show that ARCeR is able to successfully process prompts even in cases that LLMs or basic RAG systems are not able to cope with. Furthermore, ARCeR is able to target any CR framework provided that specific knowledge is made available to it.
☆ Efficient Contrastive Decoding with Probabilistic Hallucination Detection - Mitigating Hallucinations in Large Vision Language Models -
Despite recent advances in Large Vision Language Models (LVLMs), these models still suffer from generating hallucinatory responses that do not align with the visual input provided. To mitigate such hallucinations, we introduce Efficient Contrastive Decoding (ECD), a simple method that leverages probabilistic hallucination detection to shift the output distribution towards contextually accurate answers at inference time. By contrasting token probabilities and hallucination scores, ECD subtracts hallucinated concepts from the original distribution, effectively suppressing hallucinations. Notably, our proposed method can be applied to any open-source LVLM and does not require additional LVLM training. We evaluate our method on several benchmark datasets and across different LVLMs. Our experiments show that ECD effectively mitigates hallucinations, outperforming state-of-the-art methods with respect to performance on LVLM benchmarks and computation time.
☆ Towards LLM Agents for Earth Observation
Earth Observation (EO) provides critical planetary data for environmental monitoring, disaster management, climate science, and other scientific domains. Here we ask: Are AI systems ready for reliable Earth Observation? We introduce \datasetnamenospace, a benchmark of 140 yes/no questions from NASA Earth Observatory articles across 13 topics and 17 satellite sensors. Using Google Earth Engine API as a tool, LLM agents can only achieve an accuracy of 33% because the code fails to run over 58% of the time. We improve the failure rate for open models by fine-tuning synthetic data, allowing much smaller models (Llama-3.1-8B) to achieve comparable accuracy to much larger ones (e.g., DeepSeek-R1). Taken together, our findings identify significant challenges to be solved before AI agents can automate earth observation, and suggest paths forward. The project page is available at https://iandrover.github.io/UnivEarth.
comment: 36 pages
☆ Reasoning-Based AI for Startup Evaluation (R.A.I.S.E.): A Memory-Augmented, Multi-Step Decision Framework
We present a novel framework that bridges the gap between the interpretability of decision trees and the advanced reasoning capabilities of large language models (LLMs) to predict startup success. Our approach leverages chain-of-thought prompting to generate detailed reasoning logs, which are subsequently distilled into structured, human-understandable logical rules. The pipeline integrates multiple enhancements - efficient data ingestion, a two-step refinement process, ensemble candidate sampling, simulated reinforcement learning scoring, and persistent memory - to ensure both stable decision-making and transparent output. Experimental evaluations on curated startup datasets demonstrate that our combined pipeline improves precision by 54% from 0.225 to 0.346 and accuracy by 50% from 0.46 to 0.70 compared to a standalone OpenAI o3 model. Notably, our model achieves over 2x the precision of a random classifier (16%). By combining state-of-the-art AI reasoning with explicit rule-based explanations, our method not only augments traditional decision-making processes but also facilitates expert intervention and continuous policy refinement. This work lays the foundation for the implementation of interpretable LLM-powered decision frameworks in high-stakes investment environments and other domains that require transparent and data-driven insights.
☆ AttentionDrop: A Novel Regularization Method for Transformer Models
Transformer-based architectures achieve state-of-the-art performance across a wide range of tasks in natural language processing, computer vision, and speech. However, their immense capacity often leads to overfitting, especially when training data is limited or noisy. We propose AttentionDrop, a unified family of stochastic regularization techniques that operate directly on the self-attention distributions. We introduces three variants: 1. Hard Attention Masking: randomly zeroes out top-k attention logits per query to encourage diverse context utilization. 2. Blurred Attention Smoothing: applies a dynamic Gaussian convolution over attention logits to diffuse overly peaked distributions. 3. Consistency-Regularized AttentionDrop: enforces output stability under multiple independent AttentionDrop perturbations via a KL-based consistency loss.
comment: 26 pages
☆ Selective Demonstration Retrieval for Improved Implicit Hate Speech Detection
Hate speech detection is a crucial area of research in natural language processing, essential for ensuring online community safety. However, detecting implicit hate speech, where harmful intent is conveyed in subtle or indirect ways, remains a major challenge. Unlike explicit hate speech, implicit expressions often depend on context, cultural subtleties, and hidden biases, making them more challenging to identify consistently. Additionally, the interpretation of such speech is influenced by external knowledge and demographic biases, resulting in varied detection results across different language models. Furthermore, Large Language Models often show heightened sensitivity to toxic language and references to vulnerable groups, which can lead to misclassifications. This over-sensitivity results in false positives (incorrectly identifying harmless statements as hateful) and false negatives (failing to detect genuinely harmful content). Addressing these issues requires methods that not only improve detection precision but also reduce model biases and enhance robustness. To address these challenges, we propose a novel method, which utilizes in-context learning without requiring model fine-tuning. By adaptively retrieving demonstrations that focus on similar groups or those with the highest similarity scores, our approach enhances contextual comprehension. Experimental results show that our method outperforms current state-of-the-art techniques. Implementation details and code are available at TBD.
☆ Optimizing Compound Retrieval Systems SIGIR 2025
Modern retrieval systems do not rely on a single ranking model to construct their rankings. Instead, they generally take a cascading approach where a sequence of ranking models are applied in multiple re-ranking stages. Thereby, they balance the quality of the top-K ranking with computational costs by limiting the number of documents each model re-ranks. However, the cascading approach is not the only way models can interact to form a retrieval system. We propose the concept of compound retrieval systems as a broader class of retrieval systems that apply multiple prediction models. This encapsulates cascading models but also allows other types of interactions than top-K re-ranking. In particular, we enable interactions with large language models (LLMs) which can provide relative relevance comparisons. We focus on the optimization of compound retrieval system design which uniquely involves learning where to apply the component models and how to aggregate their predictions into a final ranking. This work shows how our compound approach can combine the classic BM25 retrieval model with state-of-the-art (pairwise) LLM relevance predictions, while optimizing a given ranking metric and efficiency target. Our experimental results show optimized compound retrieval systems provide better trade-offs between effectiveness and efficiency than cascading approaches, even when applied in a self-supervised manner. With the introduction of compound retrieval systems, we hope to inspire the information retrieval field to more out-of-the-box thinking on how prediction models can interact to form rankings.
comment: SIGIR 2025
☆ RadMamba: Efficient Human Activity Recognition through Radar-based Micro-Doppler-Oriented Mamba State-Space Model
Radar-based HAR has emerged as a promising alternative to conventional monitoring approaches, such as wearable devices and camera-based systems, due to its unique privacy preservation and robustness advantages. However, existing solutions based on convolutional and recurrent neural networks, although effective, are computationally demanding during deployment. This limits their applicability in scenarios with constrained resources or those requiring multiple sensors. Advanced architectures, such as ViT and SSM architectures, offer improved modeling capabilities and have made efforts toward lightweight designs. However, their computational complexity remains relatively high. To leverage the strengths of transformer architectures while simultaneously enhancing accuracy and reducing computational complexity, this paper introduces RadMamba, a parameter-efficient, radar micro-Doppler-oriented Mamba SSM specifically tailored for radar-based HAR. Across three diverse datasets, RadMamba matches the top-performing previous model's 99.8% classification accuracy on Dataset DIAT with only 1/400 of its parameters and equals the leading models' 92.0% accuracy on Dataset CI4R with merely 1/10 of their parameters. In scenarios with continuous sequences of actions evaluated on Dataset UoG2020, RadMamba surpasses other models with significantly higher parameter counts by at least 3%, achieving this with only 6.7k parameters. Our code is available at: https://github.com/lab-emi/AIRHAR.
comment: Under Review
☆ Proof-Carrying Neuro-Symbolic Code
This invited paper introduces the concept of "proof-carrying neuro-symbolic code" and explains its meaning and value, from both the "neural" and the "symbolic" perspectives. The talk outlines the first successes and challenges that this new area of research faces.
comment: Invited paper at CiE 2025. arXiv admin note: text overlap with arXiv:2501.05867
☆ Purposefully Induced Psychosis (PIP): Embracing Hallucination as Imagination in Large Language Models
Hallucinations in Large Language Models (LLMs) are widely regarded as errors - outputs that deviate from factual accuracy. However, in creative or exploratory contexts, these "mistakes" may represent unexpected avenues for innovation. We introduce Purposefully Induced Psychosis (PIP), a novel approach that amplifies LLM hallucinations for imaginative tasks such as speculative fiction, interactive storytelling, and mixed-reality simulations. Drawing on Herman Melville's Moby-Dick, where Pip's "madness" reveals profound insight, we reframe hallucinations as a source of computational imagination rather than a flaw. Our method fine-tunes LLMs to encourage speculative, metaphorical, and surreal outputs - hallucinations that are useful when factual accuracy is not the chief objective. Inspired by the consensual illusions of theater and stage magic, PIP situates these creative missteps in contexts where users willingly suspend disbelief, thereby transforming "errors" into catalysts for new ways of thinking. We discuss potential applications, design principles for ensuring user consent, preliminary observations, and implications for broader AI ethics and human-AI collaboration.
comment: 5 pages, 3 figures
☆ Balancing Graph Embedding Smoothness in Self-Supervised Learning via Information-Theoretic Decomposition WWW
Self-supervised learning (SSL) in graphs has garnered significant attention, particularly in employing Graph Neural Networks (GNNs) with pretext tasks initially designed for other domains, such as contrastive learning and feature reconstruction. However, it remains uncertain whether these methods effectively reflect essential graph properties, precisely representation similarity with its neighbors. We observe that existing methods position opposite ends of a spectrum driven by the graph embedding smoothness, with each end corresponding to outperformance on specific downstream tasks. Decomposing the SSL objective into three terms via an information-theoretic framework with a neighbor representation variable reveals that this polarization stems from an imbalance among the terms, which existing methods may not effectively maintain. Further insights suggest that balancing between the extremes can lead to improved performance across a wider range of downstream tasks. A framework, BSG (Balancing Smoothness in Graph SSL), introduces novel loss functions designed to supplement the representation quality in graph-based SSL by balancing the derived three terms: neighbor loss, minimal loss, and divergence loss. We present a theoretical analysis of the effects of these loss functions, highlighting their significance from both the SSL and graph smoothness perspectives. Extensive experiments on multiple real-world datasets across node classification and link prediction consistently demonstrate that BSG achieves state-of-the-art performance, outperforming existing methods. Our implementation code is available at https://github.com/steve30572/BSG.
comment: Accepted to the Web Conference (WWW) 2025
☆ Generative Recommendation with Continuous-Token Diffusion
In recent years, there has been a significant trend toward using large language model (LLM)-based recommender systems (RecSys). Current research primarily focuses on representing complex user-item interactions within a discrete space to align with the inherent discrete nature of language models. However, this approach faces limitations due to its discrete nature: (i) information is often compressed during discretization; (ii) the tokenization and generation for the vast number of users and items in real-world scenarios are constrained by a limited vocabulary. Embracing continuous data presents a promising alternative to enhance expressive capabilities, though this approach is still in its early stages. To address this gap, we propose a novel framework, DeftRec, which incorporates \textbf{de}noising di\textbf{f}fusion models to enable LLM-based RecSys to seamlessly support continuous \textbf{t}oken as input and target. First, we introduce a robust tokenizer with a masking operation and an additive K-way architecture to index users and items, capturing their complex collaborative relationships into continuous tokens. Crucially, we develop a denoising diffusion model to process user preferences within continuous domains by conditioning on reasoning content from pre-trained large language model. During the denoising process, we reformulate the objective to include negative interactions, building a comprehensive understanding of user preferences for effective and accurate recommendation generation. Finally, given a continuous token as output, recommendations can be easily generated through score-based retrieval. Extensive experiments demonstrate the effectiveness of the proposed methods, showing that DeftRec surpasses competitive benchmarks, including both traditional and emerging LLM-based RecSys.
☆ A Computationally Efficient Algorithm for Infinite-Horizon Average-Reward Linear MDPs
We study reinforcement learning in infinite-horizon average-reward settings with linear MDPs. Previous work addresses this problem by approximating the average-reward setting by discounted setting and employing a value iteration-based algorithm that uses clipping to constrain the span of the value function for improved statistical efficiency. However, the clipping procedure requires computing the minimum of the value function over the entire state space, which is prohibitive since the state space in linear MDP setting can be large or even infinite. In this paper, we introduce a value iteration method with efficient clipping operation that only requires computing the minimum of value functions over the set of states visited by the algorithm. Our algorithm enjoys the same regret bound as the previous work while being computationally efficient, with computational complexity that is independent of the size of the state space.
☆ Language Models as Quasi-Crystalline Thought: Structure, Constraint, and Emergence in Generative Systems
This essay proposes an analogy between large language models (LLMs) and quasicrystals: systems that exhibit global coherence without periodic repetition and that are generated through local constraints. While LLMs are often evaluated in terms of predictive accuracy, factuality, or alignment, this structural perspective suggests that their most characteristic behavior is the production of internally resonant linguistic patterns. Just as quasicrystals forced a redefinition of order in physical systems, viewing LLMs as generators of quasi-structured language opens new paths for evaluation and design: privileging propagation of constraint over token-level accuracy, and coherence of form over fixed meaning. LLM outputs should be read not only for what they say, but for the patterns of constraint and coherence that organize them. This shift reframes generative language as a space of emergent patterning: LLMs are neither fully random nor strictly rule-based, but defined by a logic of constraint, resonance, and structural depth.
☆ Leveraging Machine Learning Models to Predict the Outcome of Digital Medical Triage Interviews
Many existing digital triage systems are questionnaire-based, guiding patients to appropriate care levels based on information (e.g., symptoms, medical history, and urgency) provided by the patients answering questionnaires. Such a system often uses a deterministic model with predefined rules to determine care levels. It faces challenges with incomplete triage interviews since it can only assist patients who finish the process. In this study, we explore the use of machine learning (ML) to predict outcomes of unfinished interviews, aiming to enhance patient care and service quality. Predicting triage outcomes from incomplete data is crucial for patient safety and healthcare efficiency. Our findings show that decision-tree models, particularly LGBMClassifier and CatBoostClassifier, achieve over 80\% accuracy in predicting outcomes from complete interviews while having a linear correlation between the prediction accuracy and interview completeness degree. For example, LGBMClassifier achieves 88,2\% prediction accuracy for interviews with 100\% completeness, 79,6\% accuracy for interviews with 80\% completeness, 58,9\% accuracy for 60\% completeness, and 45,7\% accuracy for 40\% completeness. The TabTransformer model demonstrated exceptional accuracy of over 80\% for all degrees of completeness but required extensive training time, indicating a need for more powerful computational resources. The study highlights the linear correlation between interview completeness and predictive power of the decision-tree models.
comment: 8 pages, 4 figures, 8 tables
☆ Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions CVPR
Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges. This survey provides a wide-ranging examination of the anti-UAV domain, centering on three core objectives-classification, detection, and tracking-while detailing emerging methodologies such as diffusion-based data synthesis, multi-modal fusion, vision-language modeling, self-supervised learning, and reinforcement learning. We systematically evaluate state-of-the-art solutions across both single-modality and multi-sensor pipelines (spanning RGB, infrared, audio, radar, and RF) and discuss large-scale as well as adversarially oriented benchmarks. Our analysis reveals persistent gaps in real-time performance, stealth detection, and swarm-based scenarios, underscoring pressing needs for robust, adaptive anti-UAV systems. By highlighting open research directions, we aim to foster innovation and guide the development of next-generation defense strategies in an era marked by the extensive use of UAVs.
comment: Accepted at CVPR Workshop Anti-UAV 2025. 15 pages
☆ Robust and Fine-Grained Detection of AI Generated Texts ACL 2025
An ideal detection system for machine generated content is supposed to work well on any generator as many more advanced LLMs come into existence day by day. Existing systems often struggle with accurately identifying AI-generated content over shorter texts. Further, not all texts might be entirely authored by a human or LLM, hence we focused more over partial cases i.e human-LLM co-authored texts. Our paper introduces a set of models built for the task of token classification which are trained on an extensive collection of human-machine co-authored texts, which performed well over texts of unseen domains, unseen generators, texts by non-native speakers and those with adversarial inputs. We also introduce a new dataset of over 2.4M such texts mostly co-authored by several popular proprietary LLMs over 23 languages. We also present findings of our models' performance over each texts of each domain and generator. Additional findings include comparison of performance against each adversarial method, length of input texts and characteristics of generated texts compared to the original human authored texts.
comment: ACL 2025 Feb ARR Submission
☆ VIPO: Value Function Inconsistency Penalized Offline Reinforcement Learning
Offline reinforcement learning (RL) learns effective policies from pre-collected datasets, offering a practical solution for applications where online interactions are risky or costly. Model-based approaches are particularly advantageous for offline RL, owing to their data efficiency and generalizability. However, due to inherent model errors, model-based methods often artificially introduce conservatism guided by heuristic uncertainty estimation, which can be unreliable. In this paper, we introduce VIPO, a novel model-based offline RL algorithm that incorporates self-supervised feedback from value estimation to enhance model training. Specifically, the model is learned by additionally minimizing the inconsistency between the value learned directly from the offline data and the one estimated from the model. We perform comprehensive evaluations from multiple perspectives to show that VIPO can learn a highly accurate model efficiently and consistently outperform existing methods. It offers a general framework that can be readily integrated into existing model-based offline RL algorithms to systematically enhance model accuracy. As a result, VIPO achieves state-of-the-art performance on almost all tasks in both D4RL and NeoRL benchmarks.
☆ ADAT: Time-Series-Aware Adaptive Transformer Architecture for Sign Language Translation
Current sign language machine translation systems rely on recognizing hand movements, facial expressions and body postures, and natural language processing, to convert signs into text. Recent approaches use Transformer architectures to model long-range dependencies via positional encoding. However, they lack accuracy in recognizing fine-grained, short-range temporal dependencies between gestures captured at high frame rates. Moreover, their high computational complexity leads to inefficient training. To mitigate these issues, we propose an Adaptive Transformer (ADAT), which incorporates components for enhanced feature extraction and adaptive feature weighting through a gating mechanism to emphasize contextually relevant features while reducing training overhead and maintaining translation accuracy. To evaluate ADAT, we introduce MedASL, the first public medical American Sign Language dataset. In sign-to-gloss-to-text experiments, ADAT outperforms the encoder-decoder transformer, improving BLEU-4 accuracy by 0.1% while reducing training time by 14.33% on PHOENIX14T and 3.24% on MedASL. In sign-to-text experiments, it improves accuracy by 8.7% and reduces training time by 2.8% on PHOENIX14T and achieves 4.7% higher accuracy and 7.17% faster training on MedASL. Compared to encoder-only and decoder-only baselines in sign-to-text, ADAT is at least 6.8% more accurate despite being up to 12.1% slower due to its dual-stream structure.
☆ Rethinking the Generation of High-Quality CoT Data from the Perspective of LLM-Adaptive Question Difficulty Grading
Recently, DeepSeek-R1 (671B) (DeepSeek-AIet al., 2025) has demonstrated its excellent reasoning ability in complex tasks and has publiclyshared its methodology. This provides potentially high-quality chain-of-thought (CoT) data for stimulating the reasoning abilities of small-sized large language models (LLMs). To generate high-quality CoT data for different LLMs, we seek an efficient method for generating high-quality CoT data with LLM-Adaptive questiondifficulty levels. First, we grade the difficulty of the questions according to the reasoning ability of the LLMs themselves and construct a LLM-Adaptive question database. Second, we sample the problem database based on a distribution of difficulty levels of the questions and then use DeepSeek-R1 (671B) (DeepSeek-AI et al., 2025) to generate the corresponding high-quality CoT data with correct answers. Thanks to the construction of CoT data with LLM-Adaptive difficulty levels, we have significantly reduced the cost of data generation and enhanced the efficiency of model supervised fine-tuning (SFT). Finally, we have validated the effectiveness and generalizability of the proposed method in the fields of complex mathematical competitions and code generation tasks. Notably, with only 2k high-quality mathematical CoT data, our ZMath-32B surpasses DeepSeek-Distill-32B in math reasoning task. Similarly, with only 2k high-quality code CoT data, our ZCode-32B surpasses DeepSeek-Distill-32B in code reasoning tasks.
☆ Causality-enhanced Decision-Making for Autonomous Mobile Robots in Dynamic Environments
The growing integration of robots in shared environments -- such as warehouses, shopping centres, and hospitals -- demands a deep understanding of the underlying dynamics and human behaviours, including how, when, and where individuals engage in various activities and interactions. This knowledge goes beyond simple correlation studies and requires a more comprehensive causal analysis. By leveraging causal inference to model cause-and-effect relationships, we can better anticipate critical environmental factors and enable autonomous robots to plan and execute tasks more effectively. To this end, we propose a novel causality-based decision-making framework that reasons over a learned causal model to predict battery usage and human obstructions, understanding how these factors could influence robot task execution. Such reasoning framework assists the robot in deciding when and how to complete a given task. To achieve this, we developed also PeopleFlow, a new Gazebo-based simulator designed to model context-sensitive human-robot spatial interactions in shared workspaces. PeopleFlow features realistic human and robot trajectories influenced by contextual factors such as time, environment layout, and robot state, and can simulate a large number of agents. While the simulator is general-purpose, in this paper we focus on a warehouse-like environment as a case study, where we conduct an extensive evaluation benchmarking our causal approach against a non-causal baseline. Our findings demonstrate the efficacy of the proposed solutions, highlighting how causal reasoning enables autonomous robots to operate more efficiently and safely in dynamic environments shared with humans.
comment: Under review at The International Journal of Robotics Research (IJRR)
☆ Learning Physics-Informed Color-Aware Transforms for Low-Light Image Enhancement ICME 2025
Image decomposition offers deep insights into the imaging factors of visual data and significantly enhances various advanced computer vision tasks. In this work, we introduce a novel approach to low-light image enhancement based on decomposed physics-informed priors. Existing methods that directly map low-light to normal-light images in the sRGB color space suffer from inconsistent color predictions and high sensitivity to spectral power distribution (SPD) variations, resulting in unstable performance under diverse lighting conditions. To address these challenges, we introduce a Physics-informed Color-aware Transform (PiCat), a learning-based framework that converts low-light images from the sRGB color space into deep illumination-invariant descriptors via our proposed Color-aware Transform (CAT). This transformation enables robust handling of complex lighting and SPD variations. Complementing this, we propose the Content-Noise Decomposition Network (CNDN), which refines the descriptor distributions to better align with well-lit conditions by mitigating noise and other distortions, thereby effectively restoring content representations to low-light images. The CAT and the CNDN collectively act as a physical prior, guiding the transformation process from low-light to normal-light domains. Our proposed PiCat framework demonstrates superior performance compared to state-of-the-art methods across five benchmark datasets.
comment: Accepted by ICME 2025
☆ Seeking and leveraging alternative variable dependency concepts in gray-box-elusive bimodal land-use allocation problems
Solving land-use allocation problems can help us to deal with some of the most urgent global environmental issues. Since these problems are NP-hard, effective optimizers are needed to handle them. The knowledge about variable dependencies allows for proposing such tools. However, in this work, we consider a real-world multi-objective problem for which standard variable dependency discovery techniques are inapplicable. Therefore, using linkage-based variation operators is unreachable. To address this issue, we propose a definition of problem-dedicated variable dependency. On this base, we propose obtaining masks of dependent variables. Using them, we construct three novel crossover operators. The results concerning real-world test cases show that introducing our propositions into two well-known optimizers (NSGA-II, MOEA/D) dedicated to multi-objective optimization significantly improves their effectiveness.
☆ Moving between high-quality optima using multi-satisfiability characteristics in hard-to-solve Max3Sat instances
Gray-box optimization proposes effective and efficient optimizers of general use. To this end, it leverages information about variable dependencies and the subfunction-based problem representation. These approaches were already shown effective by enabling \textit{tunnelling} between local optima even if these moves require the modification of many dependent variables. Tunnelling is useful in solving the maximum satisfiability problem (MaxSat), which can be reformulated to Max3Sat. Since many real-world problems can be brought to solving the MaxSat/Max3Sat instances, it is important to solve them effectively and efficiently. Therefore, we focus on Max3Sat instances for which tunnelling fails to introduce improving moves between locally optimal high-quality solutions and the region of globally optimal solutions. We analyze the features of such instances on the ground of phase transitions. Based on these observations, we propose manipulating clause-satisfiability characteristics that allow connecting high-quality solutions distant in the solution space. We utilize multi-satisfiability characteristics in the optimizer built from typical gray-box mechanisms. The experimental study shows that the proposed optimizer can solve those Max3Sat instances that are out of the grasp of state-of-the-art gray-box optimizers. At the same time, it remains effective for instances that have already been successfully solved by gray-box.
☆ EngramNCA: a Neural Cellular Automaton Model of Memory Transfer
This study introduces EngramNCA, a neural cellular automaton (NCA) that integrates both publicly visible states and private, cell-internal memory channels, drawing inspiration from emerging biological evidence suggesting that memory storage extends beyond synaptic modifications to include intracellular mechanisms. The proposed model comprises two components: GeneCA, an NCA trained to develop distinct morphologies from seed cells containing immutable "gene" encodings, and GenePropCA, an auxiliary NCA that modulates the private "genetic" memory of cells without altering their visible states. This architecture enables the encoding and propagation of complex morphologies through the interaction of visible and private channels, facilitating the growth of diverse structures from a shared "genetic" substrate. EngramNCA supports the emergence of hierarchical and coexisting morphologies, offering insights into decentralized memory storage and transfer in artificial systems. These findings have potential implications for the development of adaptive, self-organizing systems and may contribute to the broader understanding of memory mechanisms in both biological and synthetic contexts.
☆ Evaluating the Goal-Directedness of Large Language Models
To what extent do LLMs use their capabilities towards their given goal? We take this as a measure of their goal-directedness. We evaluate goal-directedness on tasks that require information gathering, cognitive effort, and plan execution, where we use subtasks to infer each model's relevant capabilities. Our evaluations of LLMs from Google DeepMind, OpenAI, and Anthropic show that goal-directedness is relatively consistent across tasks, differs from task performance, and is only moderately sensitive to motivational prompts. Notably, most models are not fully goal-directed. We hope our goal-directedness evaluations will enable better monitoring of LLM progress, and enable more deliberate design choices of agentic properties in LLMs.
☆ FiSMiness: A Finite State Machine Based Paradigm for Emotional Support Conversations
Emotional support conversation (ESC) aims to alleviate the emotional distress of individuals through effective conversations. Although large language models (LLMs) have obtained remarkable progress on ESC, most of these studies might not define the diagram from the state model perspective, therefore providing a suboptimal solution for long-term satisfaction. To address such an issue, we leverage the Finite State Machine (FSM) on LLMs, and propose a framework called FiSMiness. Our framework allows a single LLM to bootstrap the planning during ESC, and self-reason the seeker's emotion, support strategy and the final response upon each conversational turn. Substantial experiments on ESC datasets suggest that FiSMiness outperforms many baselines, including direct inference, self-refine, chain of thought, finetuning, and external-assisted methods, even those with many more parameters.
comment: accepted by CMCL
☆ Déjà Vu: Multilingual LLM Evaluation through the Lens of Machine Translation Evaluation
Generation capabilities and language coverage of multilingual large language models (mLLMs) are advancing rapidly. However, evaluation practices for generative abilities of mLLMs are still lacking comprehensiveness, scientific rigor, and consistent adoption across research labs, which undermines their potential to meaningfully guide mLLM development. We draw parallels with machine translation (MT) evaluation, a field that faced similar challenges and has, over decades, developed transparent reporting standards and reliable evaluations for multilingual generative models. Through targeted experiments across key stages of the generative evaluation pipeline, we demonstrate how best practices from MT evaluation can deepen the understanding of quality differences between models. Additionally, we identify essential components for robust meta-evaluation of mLLMs, ensuring the evaluation methods themselves are rigorously assessed. We distill these insights into a checklist of actionable recommendations for mLLM research and development.
☆ Real-World Depth Recovery via Structure Uncertainty Modeling and Inaccurate GT Depth Fitting
The low-quality structure in raw depth maps is prevalent in real-world RGB-D datasets, which makes real-world depth recovery a critical task in recent years. However, the lack of paired raw-ground truth (raw-GT) data in the real world poses challenges for generalized depth recovery. Existing methods insufficiently consider the diversity of structure misalignment in raw depth maps, which leads to poor generalization in real-world depth recovery. Notably, random structure misalignments are not limited to raw depth data but also affect GT depth in real-world datasets. In the proposed method, we tackle the generalization problem from both input and output perspectives. For input, we enrich the diversity of structure misalignment in raw depth maps by designing a new raw depth generation pipeline, which helps the network avoid overfitting to a specific condition. Furthermore, a structure uncertainty module is designed to explicitly identify the misaligned structure for input raw depth maps to better generalize in unseen scenarios. Notably the well-trained depth foundation model (DFM) can help the structure uncertainty module estimate the structure uncertainty better. For output, a robust feature alignment module is designed to precisely align with the accurate structure of RGB images avoiding the interference of inaccurate GT depth. Extensive experiments on multiple datasets demonstrate the proposed method achieves competitive accuracy and generalization capabilities across various challenging raw depth maps.
☆ Learning Strategies in Particle Swarm Optimizer: A Critical Review and Performance Analysis
Nature has long inspired the development of swarm intelligence (SI), a key branch of artificial intelligence that models collective behaviors observed in biological systems for solving complex optimization problems. Particle swarm optimization (PSO) is widely adopted among SI algorithms due to its simplicity and efficiency. Despite numerous learning strategies proposed to enhance PSO's performance in terms of convergence speed, robustness, and adaptability, no comprehensive and systematic analysis of these strategies exists. We review and classify various learning strategies to address this gap, assessing their impact on optimization performance. Additionally, a comparative experimental evaluation is conducted to examine how these strategies influence PSO's search dynamics. Finally, we discuss open challenges and future directions, emphasizing the need for self-adaptive, intelligent PSO variants capable of addressing increasingly complex real-world problems.
comment: 53 pages, 14 figures
☆ Selective Attention Federated Learning: Improving Privacy and Efficiency for Clinical Text Classification
Federated Learning (FL) faces major challenges regarding communication overhead and model privacy when training large language models (LLMs), especially in healthcare applications. To address these, we introduce Selective Attention Federated Learning (SAFL), a novel approach that dynamically fine-tunes only those transformer layers identified as attention-critical. By employing attention patterns to determine layer importance, SAFL significantly reduces communication bandwidth and enhances differential privacy resilience. Evaluations on clinical NLP benchmarks (i2b2 Clinical Concept Extraction and MIMIC-III discharge summaries) demonstrate that SAFL achieves competitive performance with centralized models while substantially improving communication efficiency and privacy preservation.
☆ Large Language Models for Drug Overdose Prediction from Longitudinal Medical Records
The ability to predict drug overdose risk from a patient's medical records is crucial for timely intervention and prevention. Traditional machine learning models have shown promise in analyzing longitudinal medical records for this task. However, recent advancements in large language models (LLMs) offer an opportunity to enhance prediction performance by leveraging their ability to process long textual data and their inherent prior knowledge across diverse tasks. In this study, we assess the effectiveness of Open AI's GPT-4o LLM in predicting drug overdose events using patients' longitudinal insurance claims records. We evaluate its performance in both fine-tuned and zero-shot settings, comparing them to strong traditional machine learning methods as baselines. Our results show that LLMs not only outperform traditional models in certain settings but can also predict overdose risk in a zero-shot setting without task-specific training. These findings highlight the potential of LLMs in clinical decision support, particularly for drug overdose risk prediction.
☆ Enhancing Web Agents with Explicit Rollback Mechanisms
With recent advancements in large language models, web agents have been greatly improved. However, dealing with complex and dynamic web environments requires more advanced planning and search abilities. Previous studies usually adopt a greedy one-way search strategy, which may struggle to recover from erroneous states. In this work, we enhance web agents with an explicit rollback mechanism, enabling the agent to revert back to a previous state in its navigation trajectory. This mechanism gives the model the flexibility to directly control the search process, leading to an effective and efficient web navigation method. We conduct experiments on two live web navigation benchmarks with zero-shot and fine-tuning settings. The results demonstrate the effectiveness of our proposed approach.
☆ ACMamba: Fast Unsupervised Anomaly Detection via An Asymmetrical Consensus State Space Model
Unsupervised anomaly detection in hyperspectral images (HSI), aiming to detect unknown targets from backgrounds, is challenging for earth surface monitoring. However, current studies are hindered by steep computational costs due to the high-dimensional property of HSI and dense sampling-based training paradigm, constraining their rapid deployment. Our key observation is that, during training, not all samples within the same homogeneous area are indispensable, whereas ingenious sampling can provide a powerful substitute for reducing costs. Motivated by this, we propose an Asymmetrical Consensus State Space Model (ACMamba) to significantly reduce computational costs without compromising accuracy. Specifically, we design an asymmetrical anomaly detection paradigm that utilizes region-level instances as an efficient alternative to dense pixel-level samples. In this paradigm, a low-cost Mamba-based module is introduced to discover global contextual attributes of regions that are essential for HSI reconstruction. Additionally, we develop a consensus learning strategy from the optimization perspective to simultaneously facilitate background reconstruction and anomaly compression, further alleviating the negative impact of anomaly reconstruction. Theoretical analysis and extensive experiments across eight benchmarks verify the superiority of ACMamba, demonstrating a faster speed and stronger performance over the state-of-the-art.
comment: 15 pages, 9 figures
☆ Agile Retrospectives: What went well? What didn't go well? What should we do? SC
In Agile/Scrum software development, the idea of retrospective meetings (retros) is one of the core elements of the project process. In this paper, we present our work in progress focusing on two aspects: analysis of potential usage of generative AI for information interaction within retrospective meetings, and visualisation of retros' information to software development teams. We also present our prototype tool RetroAI++, focusing on retros-related functionalities.
comment: Preprint. Accepted to the 20th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2025). Final version to be published by SCITEPRESS, http://www.scitepress.org
☆ PCDiff: Proactive Control for Ownership Protection in Diffusion Models with Watermark Compatibility
With the growing demand for protecting the intellectual property (IP) of text-to-image diffusion models, we propose PCDiff -- a proactive access control framework that redefines model authorization by regulating generation quality. At its core, PCDIFF integrates a trainable fuser module and hierarchical authentication layers into the decoder architecture, ensuring that only users with valid encrypted credentials can generate high-fidelity images. In the absence of valid keys, the system deliberately degrades output quality, effectively preventing unauthorized exploitation.Importantly, while the primary mechanism enforces active access control through architectural intervention, its decoupled design retains compatibility with existing watermarking techniques. This satisfies the need of model owners to actively control model ownership while preserving the traceability capabilities provided by traditional watermarking approaches.Extensive experimental evaluations confirm a strong dependency between credential verification and image quality across various attack scenarios. Moreover, when combined with typical post-processing operations, PCDIFF demonstrates powerful performance alongside conventional watermarking methods. This work shifts the paradigm from passive detection to proactive enforcement of authorization, laying the groundwork for IP management of diffusion models.
☆ Shared Disk KV Cache Management for Efficient Multi-Instance Inference in RAG-Powered LLMs
Recent large language models (LLMs) face increasing inference latency as input context length and model size continue to grow. In particular, the retrieval-augmented generation (RAG) technique, which enhances LLM responses by incorporating external knowledge, exacerbates this issue by significantly increasing the number of input tokens. This expansion in token length leads to a substantial rise in computational overhead, particularly during the prefill stage, resulting in prolonged time-to-first-token (TTFT). To address this issue, this paper proposes a method to reduce TTFT by leveraging a disk-based key-value (KV) cache to lessen the computational burden during the prefill stage. We also introduce a disk-based shared KV cache management system, called Shared RAG-DCache, for multi-instance LLM RAG service environments. This system, together with an optimal system configuration, improves both throughput and latency under given resource constraints. Shared RAG-DCache exploits the locality of documents related to user queries in RAG, as well as the queueing delay in LLM inference services. It proactively generates and stores disk KV caches for query-related documents and shares them across multiple LLM instances to enhance inference performance. In experiments on a single host equipped with 2 GPUs and 1 CPU, Shared RAG-DCache achieved a 15~71% increase in throughput and up to a 12~65% reduction in latency, depending on the resource configuration.
☆ GrabS: Generative Embodied Agent for 3D Object Segmentation without Scene Supervision ICLR 2025
We study the hard problem of 3D object segmentation in complex point clouds without requiring human labels of 3D scenes for supervision. By relying on the similarity of pretrained 2D features or external signals such as motion to group 3D points as objects, existing unsupervised methods are usually limited to identifying simple objects like cars or their segmented objects are often inferior due to the lack of objectness in pretrained features. In this paper, we propose a new two-stage pipeline called GrabS. The core concept of our method is to learn generative and discriminative object-centric priors as a foundation from object datasets in the first stage, and then design an embodied agent to learn to discover multiple objects by querying against the pretrained generative priors in the second stage. We extensively evaluate our method on two real-world datasets and a newly created synthetic dataset, demonstrating remarkable segmentation performance, clearly surpassing all existing unsupervised methods.
comment: ICLR 2025 Spotlight. Code and data are available at: https://github.com/vLAR-group/GrabS
☆ Characterizing and Optimizing LLM Inference Workloads on CPU-GPU Coupled Architectures SP
Large language model (LLM)-based inference workloads increasingly dominate data center costs and resource utilization. Therefore, understanding the inference workload characteristics on evolving CPU-GPU coupled architectures is crucial for optimization. This paper presents an in-depth analysis of LLM inference behavior on loosely-coupled (PCIe A100/H100) and closely-coupled (GH200) systems. We analyze performance dynamics using fine-grained operator-to-kernel trace analysis, facilitated by our novel profiler SKIP and metrics like Total Kernel Launch and Queuing Time (TKLQT). Results show that closely-coupled (CC) GH200 significantly outperforms loosely-coupled (LC) systems at large batch sizes, achieving 1.9x-2.7x faster prefill latency for Llama 3.2-1B. However, our analysis also reveals that GH200 remains CPU-bound up to 4x larger batch sizes than LC systems. In this extended CPU-bound region, we identify the performance characteristics of the Grace CPU as a key factor contributing to higher inference latency at low batch sizes on GH200. We demonstrate that TKLQT accurately identifies this CPU/GPU-bound transition point. Based on this analysis, we further show that kernel fusion offers significant potential to mitigate GH200's low-batch latency bottleneck by reducing kernel launch overhead. This detailed kernel-level characterization provides critical insights for optimizing diverse CPU-GPU coupling strategies. This work is an initial effort, and we plan to explore other major AI/DL workloads that demand different degrees of CPU-GPU heterogeneous architectures.
comment: Accepted for ISPASS 2025
☆ Climbing the Ladder of Reasoning: What LLMs Can-and Still Can't-Solve after SFT?
Recent supervised fine-tuning (SFT) approaches have significantly improved language models' performance on mathematical reasoning tasks, even when models are trained at a small scale. However, the specific capabilities enhanced through such fine-tuning remain poorly understood. In this paper, we conduct a detailed analysis of model performance on the AIME24 dataset to understand how reasoning capabilities evolve. We discover a ladder-like structure in problem difficulty, categorize questions into four tiers (Easy, Medium, Hard, and Extremely Hard (Exh)), and identify the specific requirements for advancing between tiers. We find that progression from Easy to Medium tier requires adopting an R1 reasoning style with minimal SFT (500-1K instances), while Hard-level questions suffer from frequent model's errors at each step of the reasoning chain, with accuracy plateauing at around 65% despite logarithmic scaling. Exh-level questions present a fundamentally different challenge; they require unconventional problem-solving skills that current models uniformly struggle with. Additional findings reveal that carefully curated small-scale datasets offer limited advantage-scaling dataset size proves far more effective. Our analysis provides a clearer roadmap for advancing language model capabilities in mathematical reasoning.
☆ Saga: Capturing Multi-granularity Semantics from Massive Unlabelled IMU Data for User Perception
Inertial measurement units (IMUs), have been prevalently used in a wide range of mobile perception applications such as activity recognition and user authentication, where a large amount of labelled data are normally required to train a satisfactory model. However, it is difficult to label micro-activities in massive IMU data due to the hardness of understanding raw IMU data and the lack of ground truth. In this paper, we propose a novel fine-grained user perception approach, called Saga, which only needs a small amount of labelled IMU data to achieve stunning user perception accuracy. The core idea of Saga is to first pre-train a backbone feature extraction model, utilizing the rich semantic information of different levels embedded in the massive unlabelled IMU data. Meanwhile, for a specific downstream user perception application, Bayesian Optimization is employed to determine the optimal weights for pre-training tasks involving different semantic levels. We implement Saga on five typical mobile phones and evaluate Saga on three typical tasks on three IMU datasets. Results show that when only using about 100 training samples per class, Saga can achieve over 90% accuracy of the full-fledged model trained on over ten thousands training samples with no additional system overhead.
comment: 2025 IEEE 45th International Conference on Distributed Computing Systems (ICDCS)
☆ Adjoint Sampling: Highly Scalable Diffusion Samplers via Adjoint Matching
We introduce Adjoint Sampling, a highly scalable and efficient algorithm for learning diffusion processes that sample from unnormalized densities, or energy functions. It is the first on-policy approach that allows significantly more gradient updates than the number of energy evaluations and model samples, allowing us to scale to much larger problem settings than previously explored by similar methods. Our framework is theoretically grounded in stochastic optimal control and shares the same theoretical guarantees as Adjoint Matching, being able to train without the need for corrective measures that push samples towards the target distribution. We show how to incorporate key symmetries, as well as periodic boundary conditions, for modeling molecules in both cartesian and torsional coordinates. We demonstrate the effectiveness of our approach through extensive experiments on classical energy functions, and further scale up to neural network-based energy models where we perform amortized conformer generation across many molecular systems. To encourage further research in developing highly scalable sampling methods, we plan to open source these challenging benchmarks, where successful methods can directly impact progress in computational chemistry.
☆ The Hitchhiker's Guide to Program Analysis, Part II: Deep Thoughts by LLMs
Static analysis is a cornerstone for software vulnerability detection, yet it often struggles with the classic precision-scalability trade-off. In practice, such tools often produce high false positive rates, particularly in large codebases like the Linux kernel. This imprecision can arise from simplified vulnerability modeling and over-approximation of path and data constraints. While large language models (LLMs) show promise in code understanding, their naive application to program analysis yields unreliable results due to inherent reasoning limitations. We introduce BugLens, a post-refinement framework that significantly improves static analysis precision. BugLens guides an LLM to follow traditional analysis steps by assessing buggy code patterns for security impact and validating the constraints associated with static warnings. Evaluated on real-world Linux kernel bugs, BugLens raises precision from 0.10 (raw) and 0.50 (semi-automated refinement) to 0.72, substantially reducing false positives and revealing four previously unreported vulnerabilities. Our results suggest that a structured LLM-based workflow can meaningfully enhance the effectiveness of static analysis tools.
☆ Towards Safe Synthetic Image Generation On the Web: A Multimodal Robust NSFW Defense and Million Scale Dataset
In the past years, we have witnessed the remarkable success of Text-to-Image (T2I) models and their widespread use on the web. Extensive research in making T2I models produce hyper-realistic images has led to new concerns, such as generating Not-Safe-For-Work (NSFW) web content and polluting the web society. To help prevent misuse of T2I models and create a safer web environment for users features like NSFW filters and post-hoc security checks are used in these models. However, recent work unveiled how these methods can easily fail to prevent misuse. In particular, adversarial attacks on text and image modalities can easily outplay defensive measures. %Exploiting such leads to the growing concern of preventing adversarial attacks on text and image modalities. Moreover, there is currently no robust multimodal NSFW dataset that includes both prompt and image pairs and adversarial examples. This work proposes a million-scale prompt and image dataset generated using open-source diffusion models. Second, we develop a multimodal defense to distinguish safe and NSFW text and images, which is robust against adversarial attacks and directly alleviates current challenges. Our extensive experiments show that our model performs well against existing SOTA NSFW detection methods in terms of accuracy and recall, drastically reducing the Attack Success Rate (ASR) in multimodal adversarial attack scenarios. Code: https://github.com/shahidmuneer/multimodal-nsfw-defense.
comment: Short Paper The Web Conference
☆ A Library of LLM Intrinsics for Retrieval-Augmented Generation
In the developer community for large language models (LLMs), there is not yet a clean pattern analogous to a software library, to support very large scale collaboration. Even for the commonplace use case of Retrieval-Augmented Generation (RAG), it is not currently possible to write a RAG application against a well-defined set of APIs that are agreed upon by different LLM providers. Inspired by the idea of compiler intrinsics, we propose some elements of such a concept through introducing a library of LLM Intrinsics for RAG. An LLM intrinsic is defined as a capability that can be invoked through a well-defined API that is reasonably stable and independent of how the LLM intrinsic itself is implemented. The intrinsics in our library are released as LoRA adapters on HuggingFace, and through a software interface with clear structured input/output characteristics on top of vLLM as an inference platform, accompanied in both places with documentation and code. This article describes the intended usage, training details, and evaluations for each intrinsic, as well as compositions of multiple intrinsics.
☆ Progent: Programmable Privilege Control for LLM Agents
LLM agents are an emerging form of AI systems where large language models (LLMs) serve as the central component, utilizing a diverse set of tools to complete user-assigned tasks. Despite their great potential, LLM agents pose significant security risks. When interacting with the external world, they may encounter malicious commands from attackers, leading to the execution of dangerous actions. A promising way to address this is by enforcing the principle of least privilege: allowing only essential actions for task completion while blocking unnecessary ones. However, achieving this is challenging, as it requires covering diverse agent scenarios while preserving both security and utility. We introduce Progent, the first privilege control mechanism for LLM agents. At its core is a domain-specific language for flexibly expressing privilege control policies applied during agent execution. These policies provide fine-grained constraints over tool calls, deciding when tool calls are permissible and specifying fallbacks if they are not. This enables agent developers and users to craft suitable policies for their specific use cases and enforce them deterministically to guarantee security. Thanks to its modular design, integrating Progent does not alter agent internals and requires only minimal changes to agent implementation, enhancing its practicality and potential for widespread adoption. To automate policy writing, we leverage LLMs to generate policies based on user queries, which are then updated dynamically for improved security and utility. Our extensive evaluation shows that it enables strong security while preserving high utility across three distinct scenarios or benchmarks: AgentDojo, ASB, and AgentPoison. Furthermore, we perform an in-depth analysis, showcasing the effectiveness of its core components and the resilience of its automated policy generation against adaptive attacks.
☆ Can GPT tell us why these images are synthesized? Empowering Multimodal Large Language Models for Forensics
The rapid development of generative AI facilitates content creation and makes image manipulation easier and more difficult to detect. While multimodal Large Language Models (LLMs) have encoded rich world knowledge, they are not inherently tailored for combating AI-generated Content (AIGC) and struggle to comprehend local forgery details. In this work, we investigate the application of multimodal LLMs in forgery detection. We propose a framework capable of evaluating image authenticity, localizing tampered regions, providing evidence, and tracing generation methods based on semantic tampering clues. Our method demonstrates that the potential of LLMs in forgery analysis can be effectively unlocked through meticulous prompt engineering and the application of few-shot learning techniques. We conduct qualitative and quantitative experiments and show that GPT4V can achieve an accuracy of 92.1% in Autosplice and 86.3% in LaMa, which is competitive with state-of-the-art AIGC detection methods. We further discuss the limitations of multimodal LLMs in such tasks and propose potential improvements.
comment: 12 pages, 11 figures, 13IHMMSec2025
☆ Steering Prosocial AI Agents: Computational Basis of LLM's Decision Making in Social Simulation
Large language models (LLMs) increasingly serve as human-like decision-making agents in social science and applied settings. These LLM-agents are typically assigned human-like characters and placed in real-life contexts. However, how these characters and contexts shape an LLM's behavior remains underexplored. This study proposes and tests methods for probing, quantifying, and modifying an LLM's internal representations in a Dictator Game -- a classic behavioral experiment on fairness and prosocial behavior. We extract ``vectors of variable variations'' (e.g., ``male'' to ``female'') from the LLM's internal state. Manipulating these vectors during the model's inference can substantially alter how those variables relate to the model's decision-making. This approach offers a principled way to study and regulate how social concepts can be encoded and engineered within transformer-based models, with implications for alignment, debiasing, and designing AI agents for social simulations in both academic and commercial applications.
♻ ☆ Understanding and Optimizing Multi-Stage AI Inference Pipelines
The rapid evolution of Large Language Models (LLMs) has driven the need for increasingly sophisticated inference pipelines and hardware platforms. Modern LLM serving extends beyond traditional prefill-decode workflows, incorporating multi-stage processes such as Retrieval Augmented Generation (RAG), key-value (KV) cache retrieval, dynamic model routing, and multi step reasoning. These stages exhibit diverse computational demands, requiring distributed systems that integrate GPUs, ASICs, CPUs, and memory-centric architectures. However, existing simulators lack the fidelity to model these heterogeneous, multi-engine workflows, limiting their ability to inform architectural decisions. To address this gap, we introduce HERMES, a Heterogeneous Multi-stage LLM inference Execution Simulator. HERMES models diverse request stages; including RAG, KV retrieval, reasoning, prefill, and decode across complex hardware hierarchies. HERMES supports heterogeneous clients executing multiple models concurrently unlike prior frameworks while incorporating advanced batching strategies and multi-level memory hierarchies. By integrating real hardware traces with analytical modeling, HERMES captures critical trade-offs such as memory bandwidth contention, inter-cluster communication latency, and batching efficiency in hybrid CPU-accelerator deployments. Through case studies, we explore the impact of reasoning stages on end-to-end latency, optimal batching strategies for hybrid pipelines, and the architectural implications of remote KV cache retrieval. HERMES empowers system designers to navigate the evolving landscape of LLM inference, providing actionable insights into optimizing hardware-software co-design for next-generation AI workloads.
comment: Inference System Design for Multi-Stage AI Inference Pipelines. 13 Pages, 15 Figues, 3 Tables
♻ ☆ RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins (early version)
In the rapidly advancing field of robotics, dual-arm coordination and complex object manipulation are essential capabilities for developing advanced autonomous systems. However, the scarcity of diverse, high-quality demonstration data and real-world-aligned evaluation benchmarks severely limits such development. To address this, we introduce RoboTwin, a generative digital twin framework that uses 3D generative foundation models and large language models to produce diverse expert datasets and provide a real-world-aligned evaluation platform for dual-arm robotic tasks. Specifically, RoboTwin creates varied digital twins of objects from single 2D images, generating realistic and interactive scenarios. It also introduces a spatial relation-aware code generation framework that combines object annotations with large language models to break down tasks, determine spatial constraints, and generate precise robotic movement code. Our framework offers a comprehensive benchmark with both simulated and real-world data, enabling standardized evaluation and better alignment between simulated training and real-world performance. We validated our approach using the open-source COBOT Magic Robot platform. Policies pre-trained on RoboTwin-generated data and fine-tuned with limited real-world samples improve the success rate of over 70% for single-arm tasks and over 40% for dual-arm tasks compared to models trained solely on real-world data. This significant improvement demonstrates RoboTwin's potential to enhance the development and evaluation of dual-arm robotic manipulation systems. Project Page: https://robotwin-benchmark.github.io/early-version/.
comment: Project page: https://robotwin-benchmark.github.io/early-version/
♻ ☆ BioMaze: Benchmarking and Enhancing Large Language Models for Biological Pathway Reasoning
The applications of large language models (LLMs) in various biological domains have been explored recently, but their reasoning ability in complex biological systems, such as pathways, remains underexplored, which is crucial for predicting biological phenomena, formulating hypotheses, and designing experiments. This work explores the potential of LLMs in pathway reasoning. We introduce BioMaze, a dataset with 5.1K complex pathway problems derived from real research, covering various biological contexts including natural dynamic changes, disturbances, additional intervention conditions, and multi-scale research targets. Our evaluation of methods such as CoT and graph-augmented reasoning, shows that LLMs struggle with pathway reasoning, especially in perturbed systems. To address this, we propose PathSeeker, an LLM agent that enhances reasoning through interactive subgraph-based navigation, enabling a more effective approach to handling the complexities of biological systems in a scientifically aligned manner. The dataset and code are available at https://github.com/zhao-ht/BioMaze.
♻ ☆ Multimodal Lego: Model Merging and Fine-Tuning Across Topologies and Modalities in Biomedicine
Learning holistic computational representations in physical, chemical or biological systems requires the ability to process information from different distributions and modalities within the same model. Thus, the demand for multimodal machine learning models has sharply risen for modalities that go beyond vision and language, such as sequences, graphs, time series, or tabular data. While there are many available multimodal fusion and alignment approaches, most of them require end-to-end training, scale quadratically with the number of modalities, cannot handle cases of high modality imbalance in the training set, or are highly topology-specific, making them too restrictive for many biomedical learning tasks. This paper presents Multimodal Lego (MM-Lego), a general-purpose fusion framework to turn any set of encoders into a competitive multimodal model with no or minimal fine-tuning. We achieve this by introducing a wrapper for any unimodal encoder that enforces shape consistency between modality representations. It harmonises these representations by learning features in the frequency domain to enable model merging with little signal interference. We show that MM-Lego 1) can be used as a model merging method which achieves competitive performance with end-to-end fusion models without any fine-tuning, 2) can operate on any unimodal encoder, and 3) is a model fusion method that, with minimal fine-tuning, surpasses all benchmarks in five out of seven datasets.
♻ ☆ BPP-Search: Enhancing Tree of Thought Reasoning for Mathematical Modeling Problem Solving
LLMs exhibit advanced reasoning capabilities, offering the potential to transform natural language questions into mathematical models. However, existing open-source datasets in operations research domain lack detailed annotations of the modeling process, such as variable definitions, focusing solely on objective values, which hinders reinforcement learning applications. To address this, we release the StructuredOR dataset, annotated with comprehensive labels that capture the complete mathematical modeling process. We further propose BPP-Search, an algorithm that integrates reinforcement learning into a tree-of-thought structure using Beam search, a Process reward model, and a pairwise Preference algorithm. This approach enables efficient exploration of tree structures, avoiding exhaustive search while improving accuracy. Extensive experiments on StructuredOR, NL4OPT, and MAMO-ComplexLP datasets show that BPP-Search significantly outperforms state-of-the-art methods. In tree-based reasoning, BPP-Search excels in accuracy and efficiency, enabling faster retrieval of correct solutions. The StructuredOR dataset is available at https://github.com/tengwang0318/StructuredOR.
♻ ☆ Science Out of Its Ivory Tower: Improving Accessibility with Reinforcement Learning
A vast amount of scholarly work is published daily, yet much of it remains inaccessible to the general public due to dense jargon and complex language. To address this challenge in science communication, we introduce a reinforcement learning framework that fine-tunes a language model to rewrite scholarly abstracts into more comprehensible versions. Guided by a carefully balanced combination of word- and sentence-level accessibility rewards, our language model effectively substitutes technical terms with more accessible alternatives, a task which models supervised fine-tuned or guided by conventional readability measures struggle to accomplish. Our best model adjusts the readability level of scholarly abstracts by approximately six U.S. grade levels -- in other words, from a postgraduate to a high school level. This translates to roughly a 90% relative boost over the supervised fine-tuning baseline, all while maintaining factual accuracy and high-quality language. An in-depth analysis of our approach shows that balanced rewards lead to systematic modifications in the base model, likely contributing to smoother optimization and superior performance. We envision this work as a step toward bridging the gap between scholarly research and the general public, particularly younger readers and those without a college degree.
♻ ☆ Care for the Mind Amid Chronic Diseases: An Interpretable AI Approach Using IoT
Health sensing for chronic disease management creates immense benefits for social welfare. Existing health sensing studies primarily focus on the prediction of physical chronic diseases. Depression, a widespread complication of chronic diseases, is however understudied. We draw on the medical literature to support depression detection using motion sensor data. To connect humans in this decision-making, safeguard trust, and ensure algorithm transparency, we develop an interpretable deep learning model: Temporal Prototype Network (TempPNet). TempPNet is built upon the emergent prototype learning models. To accommodate the temporal characteristic of sensor data and the progressive property of depression, TempPNet differs from existing prototype learning models in its capability of capturing temporal progressions of prototypes. Extensive empirical analyses using real-world motion sensor data show that TempPNet outperforms state-of-the-art benchmarks in depression detection. Moreover, TempPNet interprets its decision by visualizing the temporal progression of depression and its corresponding symptoms detected from sensor data. We further employ a user study and a medical expert panel to demonstrate its superiority over the benchmarks in interpretability. This study offers an algorithmic solution for impactful social good -- collaborative care of chronic diseases and depression in health sensing. Methodologically, it contributes to extant literature with a novel interpretable deep learning model for depression detection from sensor data. Patients, doctors, and caregivers can deploy our model on mobile devices to monitor patients' depression risks in real-time. Our model's interpretability also allows human experts to participate in the decision-making by reviewing the interpretation and making informed interventions.
comment: Accepted by Management Science
♻ ☆ Automatic Item Generation for Personality Situational Judgment Tests with Large Language Models
Personality assessment, particularly through situational judgment tests (SJTs), is a vital tool for psychological research, talent selection, and educational evaluation. This study explores the potential of GPT-4, a state-of-the-art large language model (LLM), to automate the generation of personality situational judgment tests (PSJTs) in Chinese. Traditional SJT development is labor-intensive and prone to biases, while GPT-4 offers a scalable, efficient alternative. Two studies were conducted: Study 1 evaluated the impact of prompt design and temperature settings on content validity, finding that optimized prompts with a temperature of 1.0 produced creative and accurate items. Study 2 assessed the psychometric properties of GPT-4-generated PSJTs, revealing that they demonstrated satisfactory reliability and validity, surpassing the performance of manually developed tests in measuring the Big Five personality traits. This research highlights GPT-4's effectiveness in developing high-quality PSJTs, providing a scalable and innovative method for psychometric test development. These findings expand the possibilities of automatic item generation and the application of LLMs in psychology, and offer practical implications for streamlining test development processes in resource-limited settings.
comment: Submitted to Psychological Methods. 56 pages (main text), 12 pages (appendix), and 5 figures
♻ ☆ Score as Action: Fine-Tuning Diffusion Generative Models by Continuous-time Reinforcement Learning
Reinforcement learning from human feedback (RLHF), which aligns a diffusion model with input prompt, has become a crucial step in building reliable generative AI models. Most works in this area use a discrete-time formulation, which is prone to induced errors, and often not applicable to models with higher-order/black-box solvers. The objective of this study is to develop a disciplined approach to fine-tune diffusion models using continuous-time RL, formulated as a stochastic control problem with a reward function that aligns the end result (terminal state) with input prompt. The key idea is to treat score matching as controls or actions, and thereby making connections to policy optimization and regularization in continuous-time RL. To carry out this idea, we lay out a new policy optimization framework for continuous-time RL, and illustrate its potential in enhancing the value networks design space via leveraging the structural property of diffusion models. We validate the advantages of our method by experiments in downstream tasks of fine-tuning large-scale Text2Image models of Stable Diffusion v1.5.
comment: arXiv admin note: text overlap with arXiv:2409.08400
♻ ☆ Bypassing Prompt Injection and Jailbreak Detection in LLM Guardrails
Large Language Models (LLMs) guardrail systems are designed to protect against prompt injection and jailbreak attacks. However, they remain vulnerable to evasion techniques. We demonstrate two approaches for bypassing LLM prompt injection and jailbreak detection systems via traditional character injection methods and algorithmic Adversarial Machine Learning (AML) evasion techniques. Through testing against six prominent protection systems, including Microsoft's Azure Prompt Shield and Meta's Prompt Guard, we show that both methods can be used to evade detection while maintaining adversarial utility achieving in some instances up to 100% evasion success. Furthermore, we demonstrate that adversaries can enhance Attack Success Rates (ASR) against black-box targets by leveraging word importance ranking computed by offline white-box models. Our findings reveal vulnerabilities within current LLM protection mechanisms and highlight the need for more robust guardrail systems.
comment: 12 pages, 5 figures, 6 tables
♻ ☆ Fine-Tuning Diffusion Generative Models via Rich Preference Optimization
We introduce Rich Preference Optimization (RPO), a novel pipeline that leverages rich feedback signals to improve the curation of preference pairs for fine-tuning text-to-image diffusion models. Traditional methods, like Diffusion-DPO, often rely solely on reward model labeling, which can be opaque, offer limited insights into the rationale behind preferences, and are prone to issues such as reward hacking or overfitting. In contrast, our approach begins with generating detailed critiques of synthesized images to extract reliable and actionable image editing instructions. By implementing these instructions, we create refined images, resulting in synthetic, informative preference pairs that serve as enhanced tuning datasets. We demonstrate the effectiveness of our pipeline and the resulting datasets in fine-tuning state-of-the-art diffusion models.
♻ ☆ Future Aware Safe Active Learning of Time Varying Systems using Gaussian Processes
Experimental exploration of high-cost systems with safety constraints, common in engineering applications, is a challenging endeavor. Data-driven models offer a promising solution, but acquiring the requisite data remains expensive and is potentially unsafe. Safe active learning techniques prove essential, enabling the learning of high-quality models with minimal expensive data points and high safety. This paper introduces a safe active learning framework tailored for time-varying systems, addressing drift, seasonal changes, and complexities due to dynamic behavior. The proposed Time-aware Integrated Mean Squared Prediction Error (T-IMSPE) method minimizes posterior variance over current and future states, optimizing information gathering also in the time domain. Empirical results highlight T-IMSPE's advantages in model quality through toy and real-world examples. State of the art Gaussian processes are compatible with T-IMSPE. Our theoretical contributions include a clear delineation which Gaussian process kernels, domains, and weighting measures are suitable for T-IMSPE and even beyond for its non-time aware predecessor IMSPE.
♻ ☆ Document Parsing Unveiled: Techniques, Challenges, and Prospects for Structured Information Extraction
Document parsing is essential for converting unstructured and semi-structured documents such as contracts, academic papers, and invoices into structured, machine-readable data. Document parsing reliable structured data from unstructured inputs, providing huge convenience for numerous applications. Especially with recent achievements in Large Language Models, document parsing plays an indispensable role in both knowledge base construction and training data generation. This survey presents a comprehensive review of the current state of document parsing, covering key methodologies, from modular pipeline systems to end-to-end models driven by large vision-language models. Core components such as layout detection, content extraction (including text, tables, and mathematical expressions), and multi-modal data integration are examined in detail. Additionally, this paper discusses the challenges faced by modular document parsing systems and vision-language models in handling complex layouts, integrating multiple modules, and recognizing high-density text. It outlines future research directions and emphasizes the importance of developing larger and more diverse datasets.
♻ ☆ VeriMind: Agentic LLM for Automated Verilog Generation with a Novel Evaluation Metric
Designing Verilog modules requires meticulous attention to correctness, efficiency, and adherence to design specifications. However, manually writing Verilog code remains a complex and time-consuming task that demands both expert knowledge and iterative refinement. Leveraging recent advancements in large language models (LLMs) and their structured text generation capabilities, we propose VeriMind, an agentic LLM framework for Verilog code generation that significantly automates and optimizes the synthesis process. Unlike traditional LLM-based code generators, VeriMind employs a structured reasoning approach: given a user-provided prompt describing design requirements, the system first formulates a detailed train of thought before the final Verilog code is generated. This multi-step methodology enhances interpretability, accuracy, and adaptability in hardware design. In addition, we introduce a novel evaluation metric-pass@ARC-which combines the conventional pass@k measure with Average Refinement Cycles (ARC) to capture both success rate and the efficiency of iterative refinement. Experimental results on diverse hardware design tasks demonstrated that our approach achieved up to $8.3\%$ improvement on pass@k metric and $8.1\%$ on pass@ARC metric. These findings underscore the transformative potential of agentic LLMs in automated hardware design, RTL development, and digital system synthesis.
♻ ☆ Heimdall: test-time scaling on the generative verification
An AI system can create and maintain knowledge only to the extent that it can verify that knowledge itself. Recent work on long Chain-of-Thought reasoning has demonstrated great potential of LLMs on solving competitive problems, but their verification ability remains to be weak and not sufficiently investigated. In this paper, we propose Heimdall, the long CoT verification LLM that can accurately judge the correctness of solutions. With pure reinforcement learning, we boost the verification accuracy from 62.5% to 94.5% on competitive math problems. By scaling with repeated sampling, the accuracy further increases to 97.5%. Through human evaluation, Heimdall demonstrates impressive generalization capabilities, successfully detecting most issues in challenging math proofs, the type of which is not included during training. Furthermore, we propose Pessimistic Verification to extend the functionality of Heimdall to scaling up the problem solving. It calls Heimdall to judge the solutions from a solver model and based on the pessimistic principle, selects the most likely correct solution with the least uncertainty. Taking DeepSeek-R1-Distill-Qwen-32B as the solver model, Pessimistic Verification improves the solution accuracy on AIME2025 from 54.2% to 70.0% with 16x compute budget and to 83.3% with more compute budget. With the stronger solver Gemini 2.5 Pro, the score reaches 93.0%. Finally, we prototype an automatic knowledge discovery system, a ternary system where one poses questions, another provides solutions, and the third verifies the solutions. Using the data synthesis work NuminaMath for the first two components, Heimdall effectively identifies problematic records within the dataset and reveals that nearly half of the data is flawed, which interestingly aligns with the recent ablation studies from NuminaMath.
♻ ☆ Task Memory Engine (TME): A Structured Memory Framework with Graph-Aware Extensions for Multi-Step LLM Agent Tasks
Large Language Models (LLMs) are increasingly used as autonomous agents for multi-step tasks. However, most existing frameworks fail to maintain a structured understanding of the task state, often relying on linear prompt concatenation or shallow memory buffers. This leads to brittle performance, frequent hallucinations, and poor long-range coherence. In this work, we propose the Task Memory Engine (TME), a lightweight and structured memory module that tracks task execution using a hierarchical Task Memory Tree (TMT). Each node in the tree corresponds to a task step, storing relevant input, output, status, and sub-task relationships. We introduce a prompt synthesis method that dynamically generates LLM prompts based on the active node path, significantly improving execution consistency and contextual grounding. Through case studies and comparative experiments on multi-step agent tasks, we demonstrate that TME leads to better task completion accuracy and more interpretable behavior with minimal implementation overhead. A reference implementation of the core TME components is available at https://github.com/biubiutomato/TME-Agent, including basic examples and structured memory integration. While the current implementation uses a tree-based structure, TME is designed to be graph-aware, supporting reusable substeps, converging task paths, and shared dependencies. This lays the groundwork for future DAG-based memory architectures.
comment: 14 pages, 5 figures. Preprint prepared for future submission. Includes implementation and token-efficiency analysis. Code at https://github.com/biubiutomato/TME-Agent
♻ ☆ LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks
Large language model unlearning has become a critical challenge in ensuring safety and controlled model behavior by removing undesired data-model influences from the pretrained model while preserving general utility. Significant recent efforts have been dedicated to developing LLM unlearning benchmarks such as WMDP (Weapons of Mass Destruction Proxy) and MUSE (Machine Unlearning Six-way Evaluation), facilitating standardized unlearning performance assessment and method comparison. Despite their usefulness, we uncover for the first time a novel coreset effect within these benchmarks. Specifically, we find that LLM unlearning achieved with the original (full) forget set can be effectively maintained using a significantly smaller subset (functioning as a "coreset"), e.g., as little as 5% of the forget set, even when selected at random. This suggests that LLM unlearning in these benchmarks can be performed surprisingly easily, even in an extremely low-data regime. We demonstrate that this coreset effect remains strong, regardless of the LLM unlearning method used, such as NPO (Negative Preference Optimization) and RMU (Representation Misdirection Unlearning), the popular ones in these benchmarks. The surprisingly strong coreset effect is also robust across various data selection methods, ranging from random selection to more sophisticated heuristic approaches. We explain the coreset effect in LLM unlearning through a keyword-based perspective, showing that keywords extracted from the forget set alone contribute significantly to unlearning effectiveness and indicating that current unlearning is driven by a compact set of high-impact tokens rather than the entire dataset. We further justify the faithfulness of coreset-unlearned models along additional dimensions, such as mode connectivity and robustness to jailbreaking attacks. Codes are available at https://github.com/OPTML-Group/MU-Coreset.
♻ ☆ GST: Precise 3D Human Body from a Single Image with Gaussian Splatting Transformers CVPR 2025
Reconstructing posed 3D human models from monocular images has important applications in the sports industry, including performance tracking, injury prevention and virtual training. In this work, we combine 3D human pose and shape estimation with 3D Gaussian Splatting (3DGS), a representation of the scene composed of a mixture of Gaussians. This allows training or fine-tuning a human model predictor on multi-view images alone, without 3D ground truth. Predicting such mixtures for a human from a single input image is challenging due to self-occlusions and dependence on articulations, while also needing to retain enough flexibility to accommodate a variety of clothes and poses. Our key observation is that the vertices of standardized human meshes (such as SMPL) can provide an adequate spatial density and approximate initial position for the Gaussians. We can then train a transformer model to jointly predict comparatively small adjustments to these positions, as well as the other 3DGS attributes and the SMPL parameters. We show empirically that this combination (using only multi-view supervision) can achieve near real-time inference of 3D human models from a single image without expensive diffusion models or 3D points supervision, thus making it ideal for the sport industry at any level. More importantly, rendering is an effective auxiliary objective to refine 3D pose estimation by accounting for clothes and other geometric variations. The code is available at https://github.com/prosperolo/GST.
comment: Camera ready for CVSports workshop at CVPR 2025
♻ ☆ StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text
Text-to-video diffusion models enable the generation of high-quality videos that follow text instructions, making it easy to create diverse and individual content. However, existing approaches mostly focus on high-quality short video generation (typically 16 or 24 frames), ending up with hard-cuts when naively extended to the case of long video synthesis. To overcome these limitations, we introduce StreamingT2V, an autoregressive approach for long video generation of 80, 240, 600, 1200 or more frames with smooth transitions. The key components are:(i) a short-term memory block called conditional attention module (CAM), which conditions the current generation on the features extracted from the previous chunk via an attentional mechanism, leading to consistent chunk transitions, (ii) a long-term memory block called appearance preservation module, which extracts high-level scene and object features from the first video chunk to prevent the model from forgetting the initial scene, and (iii) a randomized blending approach that enables to apply a video enhancer autoregressively for infinitely long videos without inconsistencies between chunks. Experiments show that StreamingT2V generates high motion amount. In contrast, all competing image-to-video methods are prone to video stagnation when applied naively in an autoregressive manner. Thus, we propose with StreamingT2V a high-quality seamless text-to-long video generator that outperforms competitors with consistency and motion. Our code will be available at: https://github.com/Picsart-AI-Research/StreamingT2V
comment: https://github.com/Picsart-AI-Research/StreamingT2V
♻ ☆ Formal Verification of Graph Convolutional Networks with Uncertain Node Features and Uncertain Graph Structure
Graph neural networks are becoming increasingly popular in the field of machine learning due to their unique ability to process data structured in graphs. They have also been applied in safety-critical environments where perturbations inherently occur. However, these perturbations require us to formally verify neural networks before their deployment in safety-critical environments as neural networks are prone to adversarial attacks. While there exists research on the formal verification of neural networks, there is no work verifying the robustness of generic graph convolutional network architectures with uncertainty in the node features and in the graph structure over multiple message-passing steps. This work addresses this research gap by explicitly preserving the non-convex dependencies of all elements in the underlying computations through reachability analysis with (matrix) polynomial zonotopes. We demonstrate our approach on three popular benchmark datasets.
comment: published at Transactions on Machine Learning Research (TMLR) 2025
♻ ☆ BoTTA: Benchmarking on-device Test Time Adaptation
The performance of deep learning models depends heavily on test samples at runtime, and shifts from the training data distribution can significantly reduce accuracy. Test-time adaptation (TTA) addresses this by adapting models during inference without requiring labeled test data or access to the original training set. While research has explored TTA from various perspectives like algorithmic complexity, data and class distribution shifts, model architectures, and offline versus continuous learning, constraints specific to mobile and edge devices remain underexplored. We propose BoTTA, a benchmark designed to evaluate TTA methods under practical constraints on mobile and edge devices. Our evaluation targets four key challenges caused by limited resources and usage conditions: (i) limited test samples, (ii) limited exposure to categories, (iii) diverse distribution shifts, and (iv) overlapping shifts within a sample. We assess state-of-the-art TTA methods under these scenarios using benchmark datasets and report system-level metrics on a real testbed. Furthermore, unlike prior work, we align with on-device requirements by advocating periodic adaptation instead of continuous inference-time adaptation. Experiments reveal key insights: many recent TTA algorithms struggle with small datasets, fail to generalize to unseen categories, and depend on the diversity and complexity of distribution shifts. BoTTA also reports device-specific resource use. For example, while SHOT improves accuracy by $2.25\times$ with $512$ adaptation samples, it uses $1.08\times$ peak memory on Raspberry Pi versus the base model. BoTTA offers actionable guidance for TTA in real-world, resource-constrained deployments.
♻ ☆ RSTeller: Scaling Up Visual Language Modeling in Remote Sensing with Rich Linguistic Semantics from Openly Available Data and Large Language Models SP
Abundant, well-annotated multimodal data in remote sensing are pivotal for aligning complex visual remote sensing (RS) scenes with human language, enabling the development of specialized vision language models across diverse RS interpretation tasks. However, annotating RS images with rich linguistic semantics at scale demands expertise in RS and substantial human labor, making it costly and often impractical. In this study, we propose a workflow that leverages large language models (LLMs) to generate multimodal datasets with semantically rich captions at scale from plain OpenStreetMap (OSM) data for images sourced from the Google Earth Engine (GEE) platform. This approach facilitates the generation of paired remote sensing data and can be readily scaled up using openly available data. Within this framework, we present RSTeller, a multimodal dataset comprising over 1.3 million RS images, each accompanied by two descriptive captions. Extensive experiments demonstrate that RSTeller enhances the performance of multiple existing vision language models for RS scene understanding through continual pre-training. Our methodology significantly reduces the manual effort and expertise needed for annotating remote sensing imagery while democratizing access to high-quality annotated data. This advancement fosters progress in visual language modeling and encourages broader participation in remote sensing research and applications. The RSTeller dataset is available at https://github.com/SlytherinGe/RSTeller.
comment: Submitted to ISPRS
♻ ☆ SoK: Decentralized AI (DeAI)
Centralization enhances the efficiency of Artificial Intelligence (AI), but it also brings critical challenges, such as single points of failure, inherent biases, data privacy concerns, and scalability issues, for AI systems. These problems are especially common in closed-source large language models (LLMs), where user data is collected and used with full transparency. To address these issues, blockchain-based decentralized AI (DeAI) has been introduced. DeAI leverages the strengths of blockchain technologies to enhance the transparency, security, decentralization, as well as trustworthiness of AI systems. Although DeAI has been widely developed in industry, a comprehensive understanding of state-of-the-art practical DeAI solutions is still lacking. In this work, we present a Systematization of Knowledge (SoK) for blockchain-based DeAI solutions. We propose a taxonomy to classify existing DeAI protocols based on the model lifecycle. Based on this taxonomy, we provide a structured way to clarify the landscape of DeAI protocols and identify their similarities and differences. Specifically, we analyze the functionalities of blockchain in DeAI, investigate how blockchain features contribute to enhancing the security, transparency, and trustworthiness of AI processes, and also ensure fair incentives for AI data and model contributors. In addition, we provide key insights and research gaps in developing DeAI protocols for future research.
comment: This is a Systematization of Knowledge (SoK) for the rapidly evolving field of Decentralized AI (DeAI). We welcome valuable comments, suggestions, and collaboration to further refine and enhance this work. We hope our contribution will help accelerate the advancement of DeAI
♻ ☆ TS-ACL: Closed-Form Solution for Time Series-oriented Continual Learning
Time series classification underpins critical applications such as healthcare diagnostics and gesture-driven interactive systems in multimedia scenarios. However, time series class-incremental learning (TSCIL) faces two major challenges: catastrophic forgetting and intra-class variations. Catastrophic forgetting occurs because gradient-based parameter update strategies inevitably erase past knowledge. And unlike images, time series data exhibits subject-specific patterns, also known as intra-class variations, which refer to differences in patterns observed within the same class. While exemplar-based methods fail to cover diverse variation with limited samples, existing exemplar-free methods lack explicit mechanisms to handle intra-class variations. To address these two challenges, we propose TS-ACL, which leverages a gradient-free closed-form solution to avoid the catastrophic forgetting problem inherent in gradient-based optimization methods while simultaneously learning global distributions to resolve intra-class variations. Additionally, it provides privacy protection and efficiency. Extensive experiments on five benchmark datasets covering various sensor modalities and tasks demonstrate that TS-ACL achieves performance close to joint training on four datasets, outperforming existing methods and establishing a new state-of-the-art (SOTA) for TSCIL.
comment: 12 pages, 5 figures, 3 tables
♻ ☆ Attribute Inference Attacks for Federated Regression Tasks
Federated Learning (FL) enables multiple clients, such as mobile phones and IoT devices, to collaboratively train a global machine learning model while keeping their data localized. However, recent studies have revealed that the training phase of FL is vulnerable to reconstruction attacks, such as attribute inference attacks (AIA), where adversaries exploit exchanged messages and auxiliary public information to uncover sensitive attributes of targeted clients. While these attacks have been extensively studied in the context of classification tasks, their impact on regression tasks remains largely unexplored. In this paper, we address this gap by proposing novel model-based AIAs specifically designed for regression tasks in FL environments. Our approach considers scenarios where adversaries can either eavesdrop on exchanged messages or directly interfere with the training process. We benchmark our proposed attacks against state-of-the-art methods using real-world datasets. The results demonstrate a significant increase in reconstruction accuracy, particularly in heterogeneous client datasets, a common scenario in FL. The efficacy of our model-based AIAs makes them better candidates for empirically quantifying privacy leakage for federated regression tasks.
♻ ☆ RLSA-PFL: Robust Lightweight Secure Aggregation with Model Inconsistency Detection in Privacy-Preserving Federated Learning
Federated Learning (FL) allows users to collaboratively train a global machine learning model by sharing local model only, without exposing their private data to a central server. This distributed learning is particularly appealing in scenarios where data privacy is crucial, and it has garnered substantial attention from both industry and academia. However, studies have revealed privacy vulnerabilities in FL, where adversaries can potentially infer sensitive information from the shared model parameters. In this paper, we present an efficient masking-based secure aggregation scheme utilizing lightweight cryptographic primitives to mitigate privacy risks. Our scheme offers several advantages over existing methods. First, it requires only a single setup phase for the entire FL training session, significantly reducing communication overhead. Second, it minimizes user-side overhead by eliminating the need for user-to-user interactions, utilizing an intermediate server layer and a lightweight key negotiation method. Third, the scheme is highly resilient to user dropouts, and the users can join at any FL round. Fourth, it can detect and defend against malicious server activities, including recently discovered model inconsistency attacks. Finally, our scheme ensures security in both semi-honest and malicious settings. We provide security analysis to formally prove the robustness of our approach. Furthermore, we implemented an end-to-end prototype of our scheme. We conducted comprehensive experiments and comparisons, which show that it outperforms existing solutions in terms of communication and computation overhead, functionality, and security.
comment: 16 pages, 10 Figures
♻ ☆ Neural Dueling Bandits: Preference-Based Optimization with Human Feedback ICLR 2025
Contextual dueling bandit is used to model the bandit problems, where a learner's goal is to find the best arm for a given context using observed noisy human preference feedback over the selected arms for the past contexts. However, existing algorithms assume the reward function is linear, which can be complex and non-linear in many real-life applications like online recommendations or ranking web search results. To overcome this challenge, we use a neural network to estimate the reward function using preference feedback for the previously selected arms. We propose upper confidence bound- and Thompson sampling-based algorithms with sub-linear regret guarantees that efficiently select arms in each round. We also extend our theoretical results to contextual bandit problems with binary feedback, which is in itself a non-trivial contribution. Experimental results on the problem instances derived from synthetic datasets corroborate our theoretical results.
comment: Accepted at ICLR 2025. Also, accepted at ICML 2024 Workshop on Foundations of Reinforcement Learning and Control
♻ ☆ Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation
With the advent of large multimodal language models, science is now at a threshold of an AI-based technological transformation. Recently, a plethora of new AI models and tools has been proposed, promising to empower researchers and academics worldwide to conduct their research more effectively and efficiently. This includes all aspects of the research cycle, especially (1) searching for relevant literature; (2) generating research ideas and conducting experimentation; generating (3) text-based and (4) multimodal content (e.g., scientific figures and diagrams); and (5) AI-based automatic peer review. In this survey, we provide an in-depth overview over these exciting recent developments, which promise to fundamentally alter the scientific research process for good. Our survey covers the five aspects outlined above, indicating relevant datasets, methods and results (including evaluation) as well as limitations and scope for future research. Ethical concerns regarding shortcomings of these tools and potential for misuse (fake science, plagiarism, harms to research integrity) take a particularly prominent place in our discussion. We hope that our survey will not only become a reference guide for newcomers to the field but also a catalyst for new AI-based initiatives in the area of "AI4Science".
comment: 44 pages, 7 figures, 8 tables
♻ ☆ Neural Network Parameter-optimization of Gaussian pmDAGs
Finding the parameters of a latent variable causal model is central to causal inference and causal identification. In this article, we show that existing graphical structures that are used in causal inference are not stable under marginalization of Gaussian Bayesian networks, and present a graphical structure that faithfully represent margins of Gaussian Bayesian networks. We present the first duality between parameter optimization of a latent variable model and training a feed-forward neural network in the parameter space of the assumed family of distributions. Based on this observation, we develop an algorithm for parameter optimization of these graphical structures based on a given observational distribution. Then, we provide conditions for causal effect identifiability in the Gaussian setting. We propose an meta-algorithm that checks whether a causal effect is identifiable or not. Moreover, we lay a grounding for generalizing the duality between a neural network and a causal model from the Gaussian to other distributions.
comment: 48 pages
♻ ☆ Independence Is Not an Issue in Neurosymbolic AI
A popular approach to neurosymbolic AI is to take the output of the last layer of a neural network, e.g. a softmax activation, and pass it through a sparse computation graph encoding certain logical constraints one wishes to enforce. This induces a probability distribution over a set of random variables, which happen to be conditionally independent of each other in many commonly used neurosymbolic AI models. Such conditionally independent random variables have been deemed harmful as their presence has been observed to co-occur with a phenomenon dubbed deterministic bias, where systems learn to deterministically prefer one of the valid solutions from the solution space over the others. We provide evidence contesting this conclusion and show that the phenomenon of deterministic bias is an artifact of improperly applying neurosymbolic AI.
♻ ☆ Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation
Sequential Recommendation (SeqRec) aims to predict the next item by capturing sequential patterns from users' historical interactions, playing a crucial role in many real-world recommender systems. However, existing approaches predominantly adopt a direct forward computation paradigm, where the final hidden state of the sequence encoder serves as the user representation. We argue that this inference paradigm, due to its limited computational depth, struggles to model the complex evolving nature of user preferences and lacks a nuanced understanding of long-tail items, leading to suboptimal performance. To address this issue, we propose \textbf{ReaRec}, the first inference-time computing framework for recommender systems, which enhances user representations through implicit multi-step reasoning. Specifically, ReaRec autoregressively feeds the sequence's last hidden state into the sequential recommender while incorporating special reasoning position embeddings to decouple the original item encoding space from the multi-step reasoning space. Moreover, we introduce two lightweight reasoning-based learning methods, Ensemble Reasoning Learning (ERL) and Progressive Reasoning Learning (PRL), to further effectively exploit ReaRec's reasoning potential. Extensive experiments on five public real-world datasets and different SeqRec architectures demonstrate the generality and effectiveness of our proposed ReaRec. Remarkably, post-hoc analyses reveal that ReaRec significantly elevates the performance ceiling of multiple sequential recommendation backbones by approximately 30\%-50\%. Thus, we believe this work can open a new and promising avenue for future research in inference-time computing for sequential recommendation.
♻ ☆ ChatTS: Aligning Time Series with LLMs via Synthetic Data for Enhanced Understanding and Reasoning VLDB' 25
Understanding time series is crucial for its application in real-world scenarios. Recently, large language models (LLMs) have been increasingly applied to time series tasks, leveraging their strong language capabilities to enhance various applications. However, research on multimodal LLMs (MLLMs) for time series understanding and reasoning remains limited, primarily due to the scarcity of high-quality datasets that align time series with textual information. This paper introduces ChatTS, a novel MLLM designed for time series analysis. ChatTS treats time series as a modality, similar to how vision MLLMs process images, enabling it to perform both understanding and reasoning with time series. To address the scarcity of training data, we propose an attribute-based method for generating synthetic time series with detailed attribute descriptions. We further introduce Time Series Evol-Instruct, a novel approach that generates diverse time series Q&As, enhancing the model's reasoning capabilities. To the best of our knowledge, ChatTS is the first TS-MLLM that takes multivariate time series as input for understanding and reasoning, which is fine-tuned exclusively on synthetic datasets. We evaluate its performance using benchmark datasets with real-world data, including six alignment tasks and four reasoning tasks. Our results show that ChatTS significantly outperforms existing vision-based MLLMs (e.g., GPT-4o) and text/agent-based LLMs, achieving a 46.0% improvement in alignment tasks and a 25.8% improvement in reasoning tasks. We have open-sourced the source code, model checkpoint and datasets at https://github.com/NetManAIOps/ChatTS.
comment: accepted by VLDB' 25
♻ ☆ Lotus at SemEval-2025 Task 11: RoBERTa with Llama-3 Generated Explanations for Multi-Label Emotion Classification SemEval 2025
This paper presents a novel approach for multi-label emotion detection, where Llama-3 is used to generate explanatory content that clarifies ambiguous emotional expressions, thereby enhancing RoBERTa's emotion classification performance. By incorporating explanatory context, our method improves F1-scores, particularly for emotions like fear, joy, and sadness, and outperforms text-only models. The addition of explanatory content helps resolve ambiguity, addresses challenges like overlapping emotional cues, and enhances multi-label classification, marking a significant advancement in emotion detection tasks.
comment: 8 pages , submitted to SemEval 2025-Task 11
♻ ☆ Orthus: Autoregressive Interleaved Image-Text Generation with Modality-Specific Heads
We introduce Orthus, an autoregressive (AR) transformer that excels in generating images given textual prompts, answering questions based on visual inputs, and even crafting lengthy image-text interleaved contents. Unlike prior arts on unified multimodal modeling, Orthus simultaneously copes with discrete text tokens and continuous image features under the AR modeling principle. The continuous treatment of visual signals minimizes the information loss for both image understanding and generation while the fully AR formulation renders the characterization of the correlation between modalities straightforward. The key mechanism enabling Orthus to leverage these advantages lies in its modality-specific heads -- one regular language modeling (LM) head predicts discrete text tokens and one diffusion head generates continuous image features conditioning on the output of the backbone. We devise an efficient strategy for building Orthus -- by substituting the Vector Quantization (VQ) operation in the existing unified AR model with a soft alternative, introducing a diffusion head, and tuning the added modules to reconstruct images, we can create an Orthus-base model effortlessly (e.g., within mere 72 A100 GPU hours). Orthus-base can further embrace post-training to better model interleaved images and texts. Empirically, Orthus surpasses competing baselines including Show-o and Chameleon across standard benchmarks, achieving a GenEval score of 0.58 and an MME-P score of 1265.8 using 7B parameters. Orthus also shows exceptional mixed-modality generation capabilities, reflecting the potential for handling intricate practical generation tasks.
♻ ☆ Neural ODE Transformers: Analyzing Internal Dynamics and Adaptive Fine-tuning ICLR 2025
Recent advancements in large language models (LLMs) based on transformer architectures have sparked significant interest in understanding their inner workings. In this paper, we introduce a novel approach to modeling transformer architectures using highly flexible non-autonomous neural ordinary differential equations (ODEs). Our proposed model parameterizes all weights of attention and feed-forward blocks through neural networks, expressing these weights as functions of a continuous layer index. Through spectral analysis of the model's dynamics, we uncover an increase in eigenvalue magnitude that challenges the weight-sharing assumption prevalent in existing theoretical studies. We also leverage the Lyapunov exponent to examine token-level sensitivity, enhancing model interpretability. Our neural ODE transformer demonstrates performance comparable to or better than vanilla transformers across various configurations and datasets, while offering flexible fine-tuning capabilities that can adapt to different architectural constraints.
comment: ICLR 2025
♻ ☆ Efficient Federated Finetuning of Tiny Transformers with Resource-Constrained Devices
In recent years, Large Language Models (LLMs) through Transformer structures have dominated many machine learning tasks, especially text processing. However, these models require massive amounts of data for training and induce high resource requirements, particularly in terms of the large number of Floating Point Operations (FLOPs) and the high amounts of memory needed. To fine-tune such a model in a parameter-efficient way, techniques like Adapter or LoRA have been developed. However, we observe that the application of LoRA, when used in federated learning (FL), while still being parameter-efficient, is memory and FLOP inefficient. Based on that observation, we develop a novel layer finetuning scheme that allows devices in cross-device FL to make use of pretrained neural networks (NNs) while adhering to given resource constraints. We show that our presented scheme outperforms the current state of the art when dealing with homogeneous or heterogeneous computation and memory constraints and is on par with LoRA regarding limited communication, thereby achieving significantly higher accuracies in FL training.
♻ ☆ AI threats to national security can be countered through an incident regime
Recent progress in AI capabilities has heightened concerns that AI systems could pose a threat to national security, for example, by making it easier for malicious actors to perform cyberattacks on critical national infrastructure, or through loss of control of autonomous AI systems. In parallel, federal legislators in the US have proposed nascent 'AI incident regimes' to identify and counter similar threats. In this paper, we consolidate these two trends and present a timely proposal for a legally mandated post-deployment AI incident regime that aims to counter potential national security threats from AI systems. We start the paper by introducing the concept of 'security-critical' to describe sectors that pose extreme risks to national security, before arguing that 'security-critical' describes civilian nuclear power, aviation, life science dual-use research of concern, and frontier AI development. We then present in detail our AI incident regime proposal, justifying each component of the proposal by demonstrating its similarity to US domestic incident regimes in other 'security-critical' sectors. Finally, we sketch a hypothetical scenario where our proposed AI incident regime deals with an AI cyber incident. Our proposed AI incident regime is split into three phases. The first phase revolves around a novel operationalization of what counts as an 'AI incident' and we suggest that AI providers must create a 'national security case' before deploying a frontier AI system. The second and third phases spell out that AI providers should notify a government agency about incidents, and that the government agency should be involved in amending AI providers' security and safety procedures, in order to counter future threats to national security.
♻ ☆ Enhancing Contrastive Learning Inspired by the Philosophy of "The Blind Men and the Elephant" AAAI 2025
Contrastive learning is a prevalent technique in self-supervised vision representation learning, typically generating positive pairs by applying two data augmentations to the same image. Designing effective data augmentation strategies is crucial for the success of contrastive learning. Inspired by the story of the blind men and the elephant, we introduce JointCrop and JointBlur. These methods generate more challenging positive pairs by leveraging the joint distribution of the two augmentation parameters, thereby enabling contrastive learning to acquire more effective feature representations. To the best of our knowledge, this is the first effort to explicitly incorporate the joint distribution of two data augmentation parameters into contrastive learning. As a plug-and-play framework without additional computational overhead, JointCrop and JointBlur enhance the performance of SimCLR, BYOL, MoCo v1, MoCo v2, MoCo v3, SimSiam, and Dino baselines with notable improvements.
comment: Accepted by AAAI 2025
♻ ☆ Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents
Although LLM-based agents, powered by Large Language Models (LLMs), can use external tools and memory mechanisms to solve complex real-world tasks, they may also introduce critical security vulnerabilities. However, the existing literature does not comprehensively evaluate attacks and defenses against LLM-based agents. To address this, we introduce Agent Security Bench (ASB), a comprehensive framework designed to formalize, benchmark, and evaluate the attacks and defenses of LLM-based agents, including 10 scenarios (e.g., e-commerce, autonomous driving, finance), 10 agents targeting the scenarios, over 400 tools, 27 different types of attack/defense methods, and 7 evaluation metrics. Based on ASB, we benchmark 10 prompt injection attacks, a memory poisoning attack, a novel Plan-of-Thought backdoor attack, 4 mixed attacks, and 11 corresponding defenses across 13 LLM backbones. Our benchmark results reveal critical vulnerabilities in different stages of agent operation, including system prompt, user prompt handling, tool usage, and memory retrieval, with the highest average attack success rate of 84.30\%, but limited effectiveness shown in current defenses, unveiling important works to be done in terms of agent security for the community. We also introduce a new metric to evaluate the agents' capability to balance utility and security. Our code can be found at https://github.com/agiresearch/ASB.
♻ ☆ Dynamical errors in machine learning forecasts
In machine learning forecasting, standard error metrics such as mean absolute error (MAE) and mean squared error (MSE) quantify discrepancies between predictions and target values. However, these metrics do not directly evaluate the physical and/or dynamical consistency of forecasts, an increasingly critical concern in scientific and engineering applications. Indeed, a fundamental yet often overlooked question is whether machine learning forecasts preserve the dynamical behavior of the underlying system. Addressing this issue is essential for assessing the fidelity of machine learning models and identifying potential failure modes, particularly in applications where maintaining correct dynamical behavior is crucial. In this work, we investigate the relationship between standard forecasting error metrics, such as MAE and MSE, and the dynamical properties of the underlying system. To achieve this goal, we use two recently developed dynamical indices: the instantaneous dimension ($d$), and the inverse persistence ($\theta$). Our results indicate that larger forecast errors -- e.g., higher MSE -- tend to occur in states with higher $d$ (higher complexity) and higher $\theta$ (lower persistence). To further assess dynamical consistency, we propose error metrics based on the dynamical indices that measure the discrepancy of the forecasted $d$ and $\theta$ versus their correct values. Leveraging these dynamical indices-based metrics, we analyze direct and recursive forecasting strategies for three canonical datasets -- Lorenz, Kuramoto-Sivashinsky equation, and Kolmogorov flow -- as well as a real-world weather forecasting task. Our findings reveal substantial distortions in dynamical properties in ML forecasts, especially for long forecast lead times or long recursive simulations, providing complementary information on ML forecast fidelity that can be used to improve ML models.
♻ ☆ A Comparative Study of Artificial Potential Fields and Reciprocal Control Barrier Function-based Safety Filters
In this paper, we demonstrate that controllers designed by artificial potential fields (APFs) can be derived from reciprocal control barrier function quadratic program (RCBF-QP) safety filters. By integrating APFs within the RCBF-QP framework, we explicitly establish the relationship between these two approaches. Specifically, we first introduce the concepts of tightened control Lyapunov functions (T-CLFs) and tightened reciprocal control barrier functions (T-RCBFs), each of which incorporates a flexible auxiliary function. We then utilize an attractive potential field as a T-CLF to guide the nominal controller design, and a repulsive potential field as a T-RCBF to formulate an RCBF-QP safety filter. With appropriately chosen auxiliary functions, we show that controllers designed by APFs and those derived by RCBF-QP safety filters are equivalent. Based on this insight, we further generalize the APF-based controllers (equivalently, RCBF-QP safety filter-based controllers) to more general scenarios without restricting the choice of auxiliary functions. Finally, we present a collision avoidance example to clearly illustrate the connection and equivalence between the two methods.
♻ ☆ OpenSDI: Spotting Diffusion-Generated Images in the Open World
This paper identifies OpenSDI, a challenge for spotting diffusion-generated images in open-world settings. In response to this challenge, we define a new benchmark, the OpenSDI dataset (OpenSDID), which stands out from existing datasets due to its diverse use of large vision-language models that simulate open-world diffusion-based manipulations. Another outstanding feature of OpenSDID is its inclusion of both detection and localization tasks for images manipulated globally and locally by diffusion models. To address the OpenSDI challenge, we propose a Synergizing Pretrained Models (SPM) scheme to build up a mixture of foundation models. This approach exploits a collaboration mechanism with multiple pretrained foundation models to enhance generalization in the OpenSDI context, moving beyond traditional training by synergizing multiple pretrained models through prompting and attending strategies. Building on this scheme, we introduce MaskCLIP, an SPM-based model that aligns Contrastive Language-Image Pre-Training (CLIP) with Masked Autoencoder (MAE). Extensive evaluations on OpenSDID show that MaskCLIP significantly outperforms current state-of-the-art methods for the OpenSDI challenge, achieving remarkable relative improvements of 14.23% in IoU (14.11% in F1) and 2.05% in accuracy (2.38% in F1) compared to the second-best model in localization and detection tasks, respectively. Our dataset and code are available at https://github.com/iamwangyabin/OpenSDI.
♻ ☆ Towards Personalized Conversational Sales Agents : Contextual User Profiling for Strategic Action
Conversational Recommender Systems (CRSs) aim to engage users in dialogue to provide tailored recommendations. While traditional CRSs focus on eliciting preferences and retrieving items, real-world e-commerce interactions involve more complex decision-making, where users consider multiple factors beyond simple attributes. To bridge this gap, we introduce Conversational Sales (CSales), a novel task that unifies preference elicitation, recommendation, and persuasion to better support user decision-making. For a realistic evaluation of CSales, we present CSUser, an LLM-based user simulator constructed from real-world data, modeling diverse user profiles with needs and personalities. Additionally, we propose CSI, a conversational sales agent that proactively infers contextual profiles through dialogue for personalized action planning. Extensive experiments demonstrate that CSUser effectively replicates real-world users and emphasize the importance of contextual profiling for strategic action selection, ultimately driving successful purchases in e-commerce.
♻ ☆ RAB$^2$-DEF: Dynamic and explainable defense against adversarial attacks in Federated Learning to fair poor clients
At the same time that artificial intelligence is becoming popular, concern and the need for regulation is growing, including among other requirements the data privacy. In this context, Federated Learning is proposed as a solution to data privacy concerns derived from different source data scenarios due to its distributed learning. The defense mechanisms proposed in literature are just focused on defending against adversarial attacks and the performance, leaving aside other important qualities such as explainability, fairness to poor quality clients, dynamism in terms of attacks configuration and generality in terms of being resilient against different kinds of attacks. In this work, we propose RAB$^2$-DEF, a $\textbf{r}$esilient $\textbf{a}$gainst $\textbf{b}\text{yzantine}$ and $\textbf{b}$ackdoor attacks which is $\textbf{d}$ynamic, $\textbf{e}$xplainable and $\textbf{f}$air to poor clients using local linear explanations. We test the performance of RAB$^2$-DEF in image datasets and both byzantine and backdoor attacks considering the state-of-the-art defenses and achieve that RAB$^2$-DEF is a proper defense at the same time that it boosts the other qualities towards trustworthy artificial intelligence.
♻ ☆ VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow CVPR 2025
Scene flow estimation aims to recover per-point motion from two adjacent LiDAR scans. However, in real-world applications such as autonomous driving, points rarely move independently of others, especially for nearby points belonging to the same object, which often share the same motion. Incorporating this locally rigid motion constraint has been a key challenge in self-supervised scene flow estimation, which is often addressed by post-processing or appending extra regularization. While these approaches are able to improve the rigidity of predicted flows, they lack an architectural inductive bias for local rigidity within the model structure, leading to suboptimal learning efficiency and inferior performance. In contrast, we enforce local rigidity with a lightweight add-on module in neural network design, enabling end-to-end learning. We design a discretized voting space that accommodates all possible translations and then identify the one shared by nearby points by differentiable voting. Additionally, to ensure computational efficiency, we operate on pillars rather than points and learn representative features for voting per pillar. We plug the Voting Module into popular model designs and evaluate its benefit on Argoverse 2 and Waymo datasets. We outperform baseline works with only marginal compute overhead. Code is available at https://github.com/tudelft-iv/VoteFlow.
comment: CVPR 2025. Code is available at https://github.com/tudelft-iv/VoteFlow. Yancong Lin and Shiming Wang have equal contributions
♻ ☆ A Computational Framework for Efficient Model Evaluation with Causal Guarantees
In order to reduce the cost of experimental evaluation for models, we introduce a computational theory of evaluation for prediction and decision models: build evaluation model to accelerate the evaluation procedures. We prove upper bounds of generalized error and generalized causal effect error of given evaluation models. We also prove efficiency, and consistency to estimated causal effect from deployed subject to evaluation metric by prediction. To learn evaluation models, we propose a meta-learner to handle heterogeneous evaluation subjects space problem. Comparing with existed evaluation approaches, our (conditional) evaluation model reduced 24.1\%-99.0\% evaluation errors across 12 scenes, including individual medicine, scientific simulation, social experiment, business activity, and quantum trade. The evaluation time is reduced 3-7 order of magnitude comparing with experiments or simulations.
♻ ☆ Natural Language Outlines for Code: Literate Programming in the LLM Era
We propose using natural language outlines as a novel modality and interaction surface for providing AI assistance to developers throughout the software development process. An NL outline for a code function comprises multiple statements written in concise prose, which partition the code and summarize its main ideas in the style of literate programming. Crucially, we find that modern LLMs can generate accurate and high-quality NL outlines in practice. Moreover, NL outlines enable a bidirectional sync between code and NL: a developer can change one and the LLM automatically updates the other. We discuss many use cases for NL outlines: they can accelerate understanding and navigation of code and diffs, simplify code maintenance, augment code search, steer code generation, and more. We then propose and compare multiple LLM prompting techniques for generating outlines and ask professional developers to judge outline quality. Finally, we present two case studies applying NL outlines toward code review and malware detection.
comment: Accepted to FSE'25 Industry Track
♻ ☆ Generalization in medical AI: a perspective on developing scalable models
The scientific community is increasingly recognizing the importance of generalization in medical AI for translating research into practical clinical applications. A three-level scale is introduced to characterize out-of-distribution generalization performance of medical AI models. This scale addresses the diversity of real-world medical scenarios as well as whether target domain data and labels are available for model recalibration. It serves as a tool to help researchers characterize their development settings and determine the best approach to tackling the challenge of out-of-distribution generalization.
♻ ☆ UI-R1: Enhancing Action Prediction of GUI Agents by Reinforcement Learning
The recent DeepSeek-R1 has showcased the emergence of reasoning capabilities in LLMs through reinforcement learning (RL) with rule-based rewards. Despite its success in language models, its application in multi-modal domains, particularly in graphic user interface (GUI) agent tasks, remains under-explored. To address this issue, we propose UI-R1, the first framework to explore how rule-based RL can enhance the reasoning capabilities of multimodal large language models (MLLMs) for GUI action prediction tasks. Specifically, UI-R1 introduces a novel rule-based action reward, enabling model optimization via policy-based algorithms such as Group Relative Policy Optimization (GRPO). For efficient training, we curate a small yet high-quality dataset of 136 challenging tasks, encompassing five common action types on mobile devices. Experimental results demonstrate that our proposed UI-R1-3B achieves significant improvements over the base model (i.e. Qwen2.5-VL-3B) on both in-domain (ID) and out-of-domain (OOD) tasks, with average accuracy gains of 22.1% on ScreenSpot, 6.0% on ScreenSpot-Pro, and 12.7% on ANDROIDCONTROL. Furthermore, UI-R1-3B delivers competitive performance compared to larger models (e.g., OS-Atlas-7B) trained via supervised fine-tuning (SFT) on 76K samples. These results underscore the potential of rule-based reinforcement learning to advance GUI understanding and control, paving the way for future research in this domain. Code website: https://github.com/lll6gg/UI-R1.
♻ ☆ Multi-Mission Tool Bench: Assessing the Robustness of LLM based Agents through Related and Dynamic Missions
Large language models (LLMs) demonstrate strong potential as agents for tool invocation due to their advanced comprehension and planning capabilities. Users increasingly rely on LLM-based agents to solve complex missions through iterative interactions. However, existing benchmarks predominantly access agents in single-mission scenarios, failing to capture real-world complexity. To bridge this gap, we propose the Multi-Mission Tool Bench. In the benchmark, each test case comprises multiple interrelated missions. This design requires agents to dynamically adapt to evolving demands. Moreover, the proposed benchmark explores all possible mission-switching patterns within a fixed mission number. Specifically, we propose a multi-agent data generation framework to construct the benchmark. We also propose a novel method to evaluate the accuracy and efficiency of agent decisions with dynamic decision trees. Experiments on diverse open-source and closed-source LLMs reveal critical factors influencing agent robustness and provide actionable insights to the tool invocation society.
♻ ☆ Direction-Aware Diagonal Autoregressive Image Generation
The raster-ordered image token sequence exhibits a significant Euclidean distance between index-adjacent tokens at line breaks, making it unsuitable for autoregressive generation. To address this issue, this paper proposes Direction-Aware Diagonal Autoregressive Image Generation (DAR) method, which generates image tokens following a diagonal scanning order. The proposed diagonal scanning order ensures that tokens with adjacent indices remain in close proximity while enabling causal attention to gather information from a broader range of directions. Additionally, two direction-aware modules: 4D-RoPE and direction embeddings are introduced, enhancing the model's capability to handle frequent changes in generation direction. To leverage the representational capacity of the image tokenizer, we use its codebook as the image token embeddings. We propose models of varying scales, ranging from 485M to 2.0B. On the 256$\times$256 ImageNet benchmark, our DAR-XL (2.0B) outperforms all previous autoregressive image generators, achieving a state-of-the-art FID score of 1.37.
♻ ☆ H2O+: An Improved Framework for Hybrid Offline-and-Online RL with Dynamics Gaps ICRA 2025
Solving real-world complex tasks using reinforcement learning (RL) without high-fidelity simulation environments or large amounts of offline data can be quite challenging. Online RL agents trained in imperfect simulation environments can suffer from severe sim-to-real issues. Offline RL approaches although bypass the need for simulators, often pose demanding requirements on the size and quality of the offline datasets. The recently emerged hybrid offline-and-online RL provides an attractive framework that enables joint use of limited offline data and imperfect simulator for transferable policy learning. In this paper, we develop a new algorithm, called H2O+, which offers great flexibility to bridge various choices of offline and online learning methods, while also accounting for dynamics gaps between the real and simulation environment. Through extensive simulation and real-world robotics experiments, we demonstrate superior performance and flexibility over advanced cross-domain online and offline RL algorithms.
comment: ICRA 2025
♻ ☆ AI-Driven Sentiment Analytics: Unlocking Business Value in the E-Commerce Landscape_v1
The rapid growth of e-commerce has led to an overwhelming volume of customer feedback, from product reviews to service interactions. Extracting meaningful insights from this data is crucial for businesses aiming to improve customer satisfaction and optimize decision-making. This paper presents an AI-driven sentiment analysis system designed specifically for e-commerce applications, balancing accuracy with interpretability. Our approach integrates traditional machine learning techniques with modern deep learning models, allowing for a more nuanced understanding of customer sentiment while ensuring transparency in decision-making. Experimental results show that our system outperforms standard sentiment analysis methods, achieving an accuracy of 89.7% on diverse, large-scale datasets. Beyond technical performance, real-world implementation across multiple e-commerce platforms demonstrates tangible improvements in customer engagement and operational efficiency. This study highlights both the potential and the challenges of applying AI to sentiment analysis in a commercial setting, offering insights into practical deployment strategies and areas for future refinement.
comment: 7 pages
♻ ☆ Leveraging Social Determinants of Health in Alzheimer's Research Using LLM-Augmented Literature Mining and Knowledge Graphs
Growing evidence suggests that social determinants of health (SDoH), a set of nonmedical factors, affect individuals' risks of developing Alzheimer's disease (AD) and related dementias. Nevertheless, the etiological mechanisms underlying such relationships remain largely unclear, mainly due to difficulties in collecting relevant information. This study presents a novel, automated framework that leverages recent advancements of large language model (LLM) and natural language processing techniques to mine SDoH knowledge from extensive literature and integrate it with AD-related biological entities extracted from the general-purpose knowledge graph PrimeKG. Utilizing graph neural networks, we performed link prediction tasks to evaluate the resultant SDoH-augmented knowledge graph. Our framework shows promise for enhancing knowledge discovery in AD and can be generalized to other SDoH-related research areas, offering a new tool for exploring the impact of social determinants on health outcomes. Our code is available at: https://github.com/hwq0726/SDoHenPKG
comment: Accepted by AMIA-IS'25: AMIA Informatics Summit
♻ ☆ What, How, Where, and How Well? A Survey on Test-Time Scaling in Large Language Models
As enthusiasm for scaling computation (data and parameters) in the pretraining era gradually diminished, test-time scaling (TTS), also referred to as ``test-time computing'' has emerged as a prominent research focus. Recent studies demonstrate that TTS can further elicit the problem-solving capabilities of large language models (LLMs), enabling significant breakthroughs not only in specialized reasoning tasks, such as mathematics and coding, but also in general tasks like open-ended Q&A. However, despite the explosion of recent efforts in this area, there remains an urgent need for a comprehensive survey offering a systemic understanding. To fill this gap, we propose a unified, multidimensional framework structured along four core dimensions of TTS research: what to scale, how to scale, where to scale, and how well to scale. Building upon this taxonomy, we conduct an extensive review of methods, application scenarios, and assessment aspects, and present an organized decomposition that highlights the unique functional roles of individual techniques within the broader TTS landscape. From this analysis, we distill the major developmental trajectories of TTS to date and offer hands-on guidelines for practical deployment. Furthermore, we identify several open challenges and offer insights into promising future directions, including further scaling, clarifying the functional essence of techniques, generalizing to more tasks, and more attributions. Our repository is available on https://github.com/testtimescaling/testtimescaling.github.io/
comment: v2: Creating the GitHub repository, Citing some missed works, Incorporating two new domains (agentic and evaluation) in where to scale, Incorporating one direction (thoughtology research) in challenge and future work
♻ ☆ ChaosEater: Fully Automating Chaos Engineering with Large Language Models
Chaos Engineering (CE) is an engineering technique aimed at improving the resiliency of distributed systems. It involves artificially injecting specific failures into a distributed system and observing its behavior in response. Based on the observation, the system can be proactively improved to handle those failures. Recent CE tools implement the automated execution of predefined CE experiments. However, defining these experiments and improving the system based on the experimental results still remain manual. To reduce the costs of the manual operations, we propose ChaosEater, a system for automating the entire CE operations with Large Language Models (LLMs). It predefines the agentic workflow according to a systematic CE cycle and assigns subdivided operations within the workflow to LLMs. ChaosEater targets CE for Kubernetes systems, which are managed through code (i.e., Infrastructure as Code). Therefore, the LLMs in ChaosEater perform software engineering tasks to complete CE cycles, including requirement definition, code generation, debugging, and testing. We evaluate ChaosEater through case studies on both small and large Kubernetes systems. The results demonstrate that it stably completes reasonable single CE cycles with significantly low time and monetary costs. The CE cycles are also qualitatively validated by human engineers and LLMs.
comment: 114 pages (7 main), 11 figures. Project page: https://ntt-dkiku.github.io/chaos-eater
♻ ☆ Efficient Distributed Retrieval-Augmented Generation for Enhancing Language Model Performance
Small language models (SLMs) support efficient deployments on resource-constrained edge devices, but their limited capacity compromises inference performance. Retrieval-augmented generation (RAG) is a promising solution to enhance model performance by integrating external databases, without requiring intensive on-device model retraining. However, large-scale public databases and user-specific private contextual documents are typically located on the cloud and the device separately, while existing RAG implementations are primarily centralized. To bridge this gap, we propose DRAGON, a distributed RAG framework to enhance on-device SLMs through both general and personal knowledge without the risk of leaking document privacy. Specifically, DRAGON decomposes multi-document RAG into multiple parallel token generation processes performed independently and locally on the cloud and the device, and employs a newly designed Speculative Aggregation, a dual-side speculative algorithm to avoid frequent output synchronization between the cloud and device. A new scheduling algorithm is further introduced to identify the optimal aggregation side based on real-time network conditions. Evaluations on real-world hardware testbed demonstrate a significant performance improvement of DRAGON-up to 1.9x greater gains over standalone SLM compared to the centralized RAG, substantial reduction in per-token latency, and negligible Time to First Token (TTFT) overhead.
♻ ☆ Unlocking the Power of Function Vectors for Characterizing and Mitigating Catastrophic Forgetting in Continual Instruction Tuning
Catastrophic forgetting (CF) poses a significant challenge in machine learning, where a model forgets previously learned information upon learning new tasks. Despite the advanced capabilities of Large Language Models (LLMs), they continue to face challenges with CF during continual learning. The majority of existing research focuses on analyzing forgetting patterns through a singular training sequence, thereby overlooking the intricate effects that diverse tasks have on model behavior. Our study explores CF across various settings, discovering that model forgetting is influenced by both the specific training tasks and the models themselves. To this end, we interpret forgetting by examining the function vector (FV), a compact representation of functions in LLMs, offering a model-dependent indicator for the occurrence of CF. Through theoretical and empirical analyses, we demonstrated that CF in LLMs primarily stems from biases in function activation rather than the overwriting of task processing functions. Leveraging these insights, we propose a novel function vector guided training methodology, incorporating a regularization technique to stabilize the FV and mitigate forgetting. Empirical tests on four benchmarks confirm the effectiveness of our proposed training method, substantiating our theoretical framework concerning CF and model function dynamics. We plan to make our code publicly accessible in the near future.
comment: 10pages
♻ ☆ Chemical Language Model Linker: blending text and molecules with modular adapters
The development of large language models and multi-modal models has enabled the appealing idea of generating novel molecules from text descriptions. Generative modeling would shift the paradigm from relying on large-scale chemical screening to find molecules with desired properties to directly generating those molecules. However, multi-modal models combining text and molecules are often trained from scratch, without leveraging existing high-quality pretrained models. Training from scratch consumes more computational resources and prohibits model scaling. In contrast, we propose a lightweight adapter-based strategy named Chemical Language Model Linker (ChemLML). ChemLML blends the two single domain models and obtains conditional molecular generation from text descriptions while still operating in the specialized embedding spaces of the molecular domain. ChemLML can tailor diverse pretrained text models for molecule generation by training relatively few adapter parameters. We find that the choice of molecular representation used within ChemLML, SMILES versus SELFIES, has a strong influence on conditional molecular generation performance. SMILES is often preferable despite not guaranteeing valid molecules. We raise issues in using the entire PubChem dataset of molecules and their associated descriptions for evaluating molecule generation and provide a filtered version of the dataset as a generation test set. To demonstrate how ChemLML could be used in practice, we generate candidate protein inhibitors and use docking to assess their quality and also generate candidate membrane permeable molecules.
comment: 60 pages, 12 figures
♻ ☆ MomentSeeker: A Comprehensive Benchmark and A Strong Baseline For Moment Retrieval Within Long Videos
Retrieval augmented generation (RAG) holds great promise in addressing challenges associated with long video understanding. These methods retrieve useful moments from long videos for their presented tasks, thereby enabling multimodal large language models (MLLMs) to generate high-quality answers in a cost-effective way. In this work, we present MomentSeeker, a comprehensive benchmark to evaluate retrieval models' performance in handling general long-video moment retrieval (LVMR) tasks. MomentSeeker offers three key advantages. First, it incorporates long videos of over 500 seconds on average, making it the first benchmark specialized for long-video moment retrieval. Second, it covers a wide range of task categories (including Moment Search, Caption Alignment, Image-conditioned Moment Search, and Video-conditioned Moment Search) and diverse application scenarios (e.g., sports, movies, cartoons, and ego), making it a comprehensive tool for assessing retrieval models' general LVMR performance. Additionally, the evaluation tasks are carefully curated through human annotation, ensuring the reliability of assessment. We further fine-tune an MLLM-based LVMR retriever on synthetic data, which demonstrates strong performance on our benchmark. We perform extensive experiments with various popular multimodal retrievers based on our benchmark, whose results highlight the challenges of LVMR and limitations for existing methods. Our created resources will be shared with community to advance future research in this field.
♻ ☆ An Attentive Graph Agent for Topology-Adaptive Cyber Defence
As cyber threats grow increasingly sophisticated, reinforcement learning (RL) is emerging as a promising technique to create intelligent and adaptive cyber defense systems. However, most existing autonomous defensive agents have overlooked the inherent graph structure of computer networks subject to cyber attacks, potentially missing critical information and constraining their adaptability. To overcome these limitations, we developed a custom version of the Cyber Operations Research Gym (CybORG) environment, encoding network state as a directed graph with realistic low-level features. We employ a Graph Attention Network (GAT) architecture to process node, edge, and global features, and adapt its output to be compatible with policy gradient methods in RL. Our GAT-based approach offers key advantages over flattened alternatives: policies that demonstrate resilience to certain types of unexpected dynamic network topology changes, reasonable generalisation to networks of varying sizes within the same structural distribution, and interpretable defensive actions grounded in tangible network properties. We demonstrate that GAT defensive policies can be trained using our low-level directed graph observations, even when unexpected connections arise during simulation. Evaluations across networks of different sizes, but consistent subnetwork structure, show our policies achieve comparable performance to policies trained specifically for each network configuration. Our study contributes to the development of robust cyber defence systems that can better adapt to real-world network security challenges.
comment: Draft requires substantial revision
♻ ☆ Imagery as Inquiry: Exploring A Multimodal Dataset for Conversational Recommendation
We introduce a multimodal dataset where users express preferences through images. These images encompass a broad spectrum of visual expressions ranging from landscapes to artistic depictions. Users request recommendations for books or music that evoke similar feelings to those captured in the images, and recommendations are endorsed by the community through upvotes. This dataset supports two recommendation tasks: title generation and multiple-choice selection. Our experiments with large foundation models reveal their limitations in these tasks. Particularly, vision-language models show no significant advantage over language-only counterparts that use descriptions, which we hypothesize is due to underutilized visual capabilities. To better harness these abilities, we propose the chain-of-imagery prompting, which results in notable improvements. We release our code and datasets.
♻ ☆ Versatile Multimodal Controls for Expressive Talking Human Animation
In filmmaking, directors typically allow actors to perform freely based on the script before providing specific guidance on how to present key actions. AI-generated content faces similar requirements, where users not only need automatic generation of lip synchronization and basic gestures from audio input but also desire semantically accurate and expressive body movement that can be ``directly guided'' through text descriptions. Therefore, we present VersaAnimator, a versatile framework that synthesizes expressive talking human videos from arbitrary portrait images. Specifically, we design a motion generator that produces basic rhythmic movements from audio input and supports text-prompt control for specific actions. The generated whole-body 3D motion tokens can animate portraits of various scales, producing talking heads, half-body gestures and even leg movements for whole-body images. Besides, we introduce a multi-modal controlled video diffusion that generates photorealistic videos, where speech signals govern lip synchronization, facial expressions, and head motions while body movements are guided by the 2D poses. Furthermore, we introduce a token2pose translator to smoothly map 3D motion tokens to 2D pose sequences. This design mitigates the stiffness resulting from direct 3D to 2D conversion and enhances the details of the generated body movements. Extensive experiments shows that VersaAnimator synthesizes lip-synced and identity-preserving videos while generating expressive and semantically meaningful whole-body motions.
♻ ☆ PSBD: Prediction Shift Uncertainty Unlocks Backdoor Detection
Deep neural networks are susceptible to backdoor attacks, where adversaries manipulate model predictions by inserting malicious samples into the training data. Currently, there is still a significant challenge in identifying suspicious training data to unveil potential backdoor samples. In this paper, we propose a novel method, Prediction Shift Backdoor Detection (PSBD), leveraging an uncertainty-based approach requiring minimal unlabeled clean validation data. PSBD is motivated by an intriguing Prediction Shift (PS) phenomenon, where poisoned models' predictions on clean data often shift away from true labels towards certain other labels with dropout applied during inference, while backdoor samples exhibit less PS. We hypothesize PS results from the neuron bias effect, making neurons favor features of certain classes. PSBD identifies backdoor training samples by computing the Prediction Shift Uncertainty (PSU), the variance in probability values when dropout layers are toggled on and off during model inference. Extensive experiments have been conducted to verify the effectiveness and efficiency of PSBD, which achieves state-of-the-art results among mainstream detection methods. The code is available at https://github.com/WL-619/PSBD.
♻ ☆ Elucidating the Design Space of Multimodal Protein Language Models
Multimodal protein language models (PLMs) integrate sequence and token-based structural information, serving as a powerful foundation for protein modeling, generation, and design. However, the reliance on tokenizing 3D structures into discrete tokens causes substantial loss of fidelity about fine-grained structural details and correlations. In this paper, we systematically elucidate the design space of multimodal PLMs to overcome their limitations. We identify tokenization loss and inaccurate structure token predictions by the PLMs as major bottlenecks. To address these, our proposed design space covers improved generative modeling, structure-aware architectures and representation learning, and data exploration. Our advancements approach finer-grained supervision, demonstrating that token-based multimodal PLMs can achieve robust structural modeling. The effective design methods dramatically improve the structure generation diversity, and notably, folding abilities of our 650M model by reducing the RMSD from 5.52 to 2.36 on PDB testset, even outperforming 3B baselines and on par with the specialized folding models.
comment: Project Page: https://bytedance.github.io/dplm/dplm-2.1/
♻ ☆ KPC-cF: Aspect-Based Sentiment Analysis via Implicit-Feature Alignment with Corpus Filtering ICML 2024
Investigations into Aspect-Based Sentiment Analysis (ABSA) for Korean industrial reviews are notably lacking in the existing literature. Our research proposes an intuitive and effective framework for ABSA in low-resource languages such as Korean. It optimizes prediction labels by integrating translated benchmark and unlabeled Korean data. Using a model fine-tuned on translated data, we pseudo-labeled the actual Korean NLI set. Subsequently, we applied LaBSE and \MSP{}-based filtering to this pseudo-NLI set as implicit feature, enhancing Aspect Category Detection and Polarity determination through additional training. Incorporating dual filtering, this model bridged dataset gaps and facilitates feature alignment with minimal resources. By implementing alignment pipelines, our approach aims to leverage high-resource datasets to develop reliable predictive and refined models within corporate or individual communities in low-resource language countries. Compared to English ABSA, our framework showed an approximately 3\% difference in F1 scores and accuracy. We will release our dataset and code for Korean ABSA, at this link.
comment: Work in Progress, DMLR@ICML 2024
♻ ☆ Exploring the Role of Knowledge Graph-Based RAG in Japanese Medical Question Answering with Small-Scale LLMs
Large language models (LLMs) perform well in medical QA, but their effectiveness in Japanese contexts is limited due to privacy constraints that prevent the use of commercial models like GPT-4 in clinical settings. As a result, recent efforts focus on instruction-tuning open-source LLMs, though the potential of combining them with retrieval-augmented generation (RAG) remains underexplored. To bridge this gap, we are the first to explore a knowledge graph-based (KG) RAG framework for Japanese medical QA small-scale open-source LLMs. Experimental results show that KG-based RAG has only a limited impact on Japanese medical QA using small-scale open-source LLMs. Further case studies reveal that the effectiveness of the RAG is sensitive to the quality and relevance of the external retrieved content. These findings offer valuable insights into the challenges and potential of applying RAG in Japanese medical QA, while also serving as a reference for other low-resource languages.
comment: 10 pages
♻ ☆ GATE3D: Generalized Attention-based Task-synergized Estimation in 3D*
The emerging trend in computer vision emphasizes developing universal models capable of simultaneously addressing multiple diverse tasks. Such universality typically requires joint training across multi-domain datasets to ensure effective generalization. However, monocular 3D object detection presents unique challenges in multi-domain training due to the scarcity of datasets annotated with accurate 3D ground-truth labels, especially beyond typical road-based autonomous driving contexts. To address this challenge, we introduce a novel weakly supervised framework leveraging pseudo-labels. Current pretrained models often struggle to accurately detect pedestrians in non-road environments due to inherent dataset biases. Unlike generalized image-based 2D object detection models, achieving similar generalization in monocular 3D detection remains largely unexplored. In this paper, we propose GATE3D, a novel framework designed specifically for generalized monocular 3D object detection via weak supervision. GATE3D effectively bridges domain gaps by employing consistency losses between 2D and 3D predictions. Remarkably, our model achieves competitive performance on the KITTI benchmark as well as on an indoor-office dataset collected by us to evaluate the generalization capabilities of our framework. Our results demonstrate that GATE3D significantly accelerates learning from limited annotated data through effective pre-training strategies, highlighting substantial potential for broader impacts in robotics, augmented reality, and virtual reality applications. Project page: https://ies0411.github.io/GATE3D/
comment: 9pages, 1 supple
♻ ☆ Visual Language Models show widespread visual deficits on neuropsychological tests
Visual Language Models (VLMs) show remarkable performance in visual reasoning tasks, successfully tackling college-level challenges that require high-level understanding of images. However, some recent reports of VLMs struggling to reason about elemental visual concepts like orientation, position, continuity, and occlusion suggest a potential gulf between human and VLM vision. Here we use the toolkit of neuropsychology to systematically assess the capabilities of three state-of-the-art VLMs across visual domains. Using 51 tests drawn from six clinical and experimental batteries, we characterise the visual abilities of leading VLMs relative to normative performance in healthy adults. While the models excel in straightforward object recognition tasks, we find widespread deficits in low- and mid-level visual abilities that would be considered clinically significant in humans. These selective deficits, profiled through validated test batteries, suggest that an artificial system can achieve complex object recognition without developing foundational visual concepts that in humans require no explicit training.
comment: 31 pages, 3 figures, 1 supplementary document with 1 figure and 51 sample images; corrected typo in Fig 1
♻ ☆ Incrementally Learning Multiple Diverse Data Domains via Multi-Source Dynamic Expansion Model
Continual Learning seeks to develop a model capable of incrementally assimilating new information while retaining prior knowledge. However, current research predominantly addresses a straightforward learning context, wherein all data samples originate from a singular data domain. This paper shifts focus to a more complex and realistic learning environment, characterized by data samples sourced from multiple distinct domains. We tackle this intricate learning challenge by introducing a novel methodology, termed the Multi-Source Dynamic Expansion Model (MSDEM), which leverages various pre-trained models as backbones and progressively establishes new experts based on them to adapt to emerging tasks. Additionally, we propose an innovative dynamic expandable attention mechanism designed to selectively harness knowledge from multiple backbones, thereby accelerating the new task learning. Moreover, we introduce a dynamic graph weight router that strategically reuses all previously acquired parameters and representations for new task learning, maximizing the positive knowledge transfer effect, which further improves generalization performance. We conduct a comprehensive series of experiments, and the empirical findings indicate that our proposed approach achieves state-of-the-art performance.
comment: 10 pages, 5 figures
♻ ☆ Self-Controlled Dynamic Expansion Model for Continual Learning
Continual Learning (CL) epitomizes an advanced training paradigm wherein prior data samples remain inaccessible during the acquisition of new tasks. Numerous investigations have delved into leveraging a pre-trained Vision Transformer (ViT) to enhance model efficacy in continual learning. Nonetheless, these approaches typically utilize a singular, static backbone, which inadequately adapts to novel tasks, particularly when engaging with diverse data domains, due to a substantial number of inactive parameters. This paper addresses this limitation by introducing an innovative Self-Controlled Dynamic Expansion Model (SCDEM), which orchestrates multiple distinct trainable pre-trained ViT backbones to furnish diverse and semantically enriched representations. Specifically, by employing the multi-backbone architecture as a shared module, the proposed SCDEM dynamically generates a new expert with minimal parameters to accommodate a new task. A novel Collaborative Optimization Mechanism (COM) is introduced to synergistically optimize multiple backbones by harnessing prediction signals from historical experts, thereby facilitating new task learning without erasing previously acquired knowledge. Additionally, a novel Feature Distribution Consistency (FDC) approach is proposed to align semantic similarity between previously and currently learned representations through an optimal transport distance-based mechanism, effectively mitigating negative knowledge transfer effects. Furthermore, to alleviate over-regularization challenges, this paper presents a novel Dynamic Layer-Wise Feature Attention Mechanism (DLWFAM) to autonomously determine the penalization intensity on each trainable representation layer. An extensive series of experiments have been conducted to evaluate the proposed methodology's efficacy, with empirical results corroborating that the approach attains state-of-the-art performance.
comment: 10 pages, 3 figures, 6 tables, Continual Learning, Cross-Domain Continual Learning, Mixture Model
♻ ☆ InterChat: Enhancing Generative Visual Analytics using Multimodal Interactions
The rise of Large Language Models (LLMs) and generative visual analytics systems has transformed data-driven insights, yet significant challenges persist in accurately interpreting users' analytical and interaction intents. While language inputs offer flexibility, they often lack precision, making the expression of complex intents inefficient, error-prone, and time-intensive. To address these limitations, we investigate the design space of multimodal interactions for generative visual analytics through a literature review and pilot brainstorming sessions. Building on these insights, we introduce a highly extensible workflow that integrates multiple LLM agents for intent inference and visualization generation. We develop InterChat, a generative visual analytics system that combines direct manipulation of visual elements with natural language inputs. This integration enables precise intent communication and supports progressive, visually driven exploratory data analyses. By employing effective prompt engineering, and contextual interaction linking, alongside intuitive visualization and interaction designs, InterChat bridges the gap between user interactions and LLM-driven visualizations, enhancing both interpretability and usability. Extensive evaluations, including two usage scenarios, a user study, and expert feedback, demonstrate the effectiveness of InterChat. Results show significant improvements in the accuracy and efficiency of handling complex visual analytics tasks, highlighting the potential of multimodal interactions to redefine user engagement and analytical depth in generative visual analytics.
comment: This work is accepted by the 27th Eurographics Conference on Visualization (EuroVis 2025). The paper contains 12 pages and 7 figures
♻ ☆ COMBO: Compositional World Models for Embodied Multi-Agent Cooperation ICLR 2025
In this paper, we investigate the problem of embodied multi-agent cooperation, where decentralized agents must cooperate given only egocentric views of the world. To effectively plan in this setting, in contrast to learning world dynamics in a single-agent scenario, we must simulate world dynamics conditioned on an arbitrary number of agents' actions given only partial egocentric visual observations of the world. To address this issue of partial observability, we first train generative models to estimate the overall world state given partial egocentric observations. To enable accurate simulation of multiple sets of actions on this world state, we then propose to learn a compositional world model for multi-agent cooperation by factorizing the naturally composable joint actions of multiple agents and compositionally generating the video conditioned on the world state. By leveraging this compositional world model, in combination with Vision Language Models to infer the actions of other agents, we can use a tree search procedure to integrate these modules and facilitate online cooperative planning. We evaluate our methods on three challenging benchmarks with 2-4 agents. The results show our compositional world model is effective and the framework enables the embodied agents to cooperate efficiently with different agents across various tasks and an arbitrary number of agents, showing the promising future of our proposed methods. More videos can be found at https://umass-embodied-agi.github.io/COMBO/.
comment: Published at ICLR 2025. 24 pages. The first three authors contributed equally
♻ ☆ Large Visual-Language Models Are Also Good Classifiers: A Study of In-Context Multimodal Fake News Detection
Large visual-language models (LVLMs) exhibit exceptional performance in visual-language reasoning across diverse cross-modal benchmarks. Despite these advances, recent research indicates that Large Language Models (LLMs), like GPT-3.5-turbo, underachieve compared to well-trained smaller models, such as BERT, in Fake News Detection (FND), prompting inquiries into LVLMs' efficacy in FND tasks. Although performance could improve through fine-tuning LVLMs, the substantial parameters and requisite pre-trained weights render it a resource-heavy endeavor for FND applications. This paper initially assesses the FND capabilities of two notable LVLMs, CogVLM and GPT4V, in comparison to a smaller yet adeptly trained CLIP model in a zero-shot context. The findings demonstrate that LVLMs can attain performance competitive with that of the smaller model. Next, we integrate standard in-context learning (ICL) with LVLMs, noting improvements in FND performance, though limited in scope and consistency. To address this, we introduce the \textbf{I}n-context \textbf{M}ultimodal \textbf{F}ake \textbf{N}ews \textbf{D}etection (IMFND) framework, enriching in-context examples and test inputs with predictions and corresponding probabilities from a well-trained smaller model. This strategic integration directs the LVLMs' focus towards news segments associated with higher probabilities, thereby improving their analytical accuracy. The experimental results suggest that the IMFND framework significantly boosts the FND efficiency of LVLMs, achieving enhanced accuracy over the standard ICL approach across three publicly available FND datasets.
comment: Withdraw for new experiments
Computer Vision and Pattern Recognition 150
☆ Adapting a World Model for Trajectory Following in a 3D Game
Imitation learning is a powerful tool for training agents by leveraging expert knowledge, and being able to replicate a given trajectory is an integral part of it. In complex environments, like modern 3D video games, distribution shift and stochasticity necessitate robust approaches beyond simple action replay. In this study, we apply Inverse Dynamics Models (IDM) with different encoders and policy heads to trajectory following in a modern 3D video game -- Bleeding Edge. Additionally, we investigate several future alignment strategies that address the distribution shift caused by the aleatoric uncertainty and imperfections of the agent. We measure both the trajectory deviation distance and the first significant deviation point between the reference and the agent's trajectory and show that the optimal configuration depends on the chosen setting. Our results show that in a diverse data setting, a GPT-style policy head with an encoder trained from scratch performs the best, DINOv2 encoder with the GPT-style policy head gives the best results in the low data regime, and both GPT-style and MLP-style policy heads had comparable results when pre-trained on a diverse setting and fine-tuned for a specific behaviour setting.
☆ SHeaP: Self-Supervised Head Geometry Predictor Learned via 2D Gaussians
Accurate, real-time 3D reconstruction of human heads from monocular images and videos underlies numerous visual applications. As 3D ground truth data is hard to come by at scale, previous methods have sought to learn from abundant 2D videos in a self-supervised manner. Typically, this involves the use of differentiable mesh rendering, which is effective but faces limitations. To improve on this, we propose SHeaP (Self-supervised Head Geometry Predictor Learned via 2D Gaussians). Given a source image, we predict a 3DMM mesh and a set of Gaussians that are rigged to this mesh. We then reanimate this rigged head avatar to match a target frame, and backpropagate photometric losses to both the 3DMM and Gaussian prediction networks. We find that using Gaussians for rendering substantially improves the effectiveness of this self-supervised approach. Training solely on 2D data, our method surpasses existing self-supervised approaches in geometric evaluations on the NoW benchmark for neutral faces and a new benchmark for non-neutral expressions. Our method also produces highly expressive meshes, outperforming state-of-the-art in emotion classification.
comment: For video demonstrations and additional materials please see https://nlml.github.io/sheap/
☆ How Do I Do That? Synthesizing 3D Hand Motion and Contacts for Everyday Interactions CVPR 2025
We tackle the novel problem of predicting 3D hand motion and contact maps (or Interaction Trajectories) given a single RGB view, action text, and a 3D contact point on the object as input. Our approach consists of (1) Interaction Codebook: a VQVAE model to learn a latent codebook of hand poses and contact points, effectively tokenizing interaction trajectories, (2) Interaction Predictor: a transformer-decoder module to predict the interaction trajectory from test time inputs by using an indexer module to retrieve a latent affordance from the learned codebook. To train our model, we develop a data engine that extracts 3D hand poses and contact trajectories from the diverse HoloAssist dataset. We evaluate our model on a benchmark that is 2.5-10X larger than existing works, in terms of diversity of objects and interactions observed, and test for generalization of the model across object categories, action categories, tasks, and scenes. Experimental results show the effectiveness of our approach over transformer & diffusion baselines across all settings.
comment: CVPR 2025, Project page: https://ap229997.github.io/projects/latentact
☆ The Tenth NTIRE 2025 Image Denoising Challenge Report
This paper presents an overview of the NTIRE 2025 Image Denoising Challenge ({\sigma} = 50), highlighting the proposed methodologies and corresponding results. The primary objective is to develop a network architecture capable of achieving high-quality denoising performance, quantitatively evaluated using PSNR, without constraints on computational complexity or model size. The task assumes independent additive white Gaussian noise (AWGN) with a fixed noise level of 50. A total of 290 participants registered for the challenge, with 20 teams successfully submitting valid results, providing insights into the current state-of-the-art in image denoising.
☆ Beyond Reconstruction: A Physics Based Neural Deferred Shader for Photo-realistic Rendering
Deep learning based rendering has demonstrated major improvements for photo-realistic image synthesis, applicable to various applications including visual effects in movies and photo-realistic scene building in video games. However, a significant limitation is the difficulty of decomposing the illumination and material parameters, which limits such methods to reconstruct an input scene, without any possibility to control these parameters. This paper introduces a novel physics based neural deferred shading pipeline to decompose the data-driven rendering process, learn a generalizable shading function to produce photo-realistic results for shading and relighting tasks, we also provide a shadow estimator to efficiently mimic shadowing effect. Our model achieves improved performance compared to classical models and a state-of-art neural shading model, and enables generalizable photo-realistic shading from arbitrary illumination input.
☆ Towards Learning to Complete Anything in Lidar
We propose CAL (Complete Anything in Lidar) for Lidar-based shape-completion in-the-wild. This is closely related to Lidar-based semantic/panoptic scene completion. However, contemporary methods can only complete and recognize objects from a closed vocabulary labeled in existing Lidar datasets. Different to that, our zero-shot approach leverages the temporal context from multi-modal sensor sequences to mine object shapes and semantic features of observed objects. These are then distilled into a Lidar-only instance-level completion and recognition model. Although we only mine partial shape completions, we find that our distilled model learns to infer full object shapes from multiple such partial observations across the dataset. We show that our model can be prompted on standard benchmarks for Semantic and Panoptic Scene Completion, localize objects as (amodal) 3D bounding boxes, and recognize objects beyond fixed class vocabularies. Our project page is https://research.nvidia.com/labs/dvl/projects/complete-anything-lidar
☆ VGDFR: Diffusion-based Video Generation with Dynamic Latent Frame Rate
Diffusion Transformer(DiT)-based generation models have achieved remarkable success in video generation. However, their inherent computational demands pose significant efficiency challenges. In this paper, we exploit the inherent temporal non-uniformity of real-world videos and observe that videos exhibit dynamic information density, with high-motion segments demanding greater detail preservation than static scenes. Inspired by this temporal non-uniformity, we propose VGDFR, a training-free approach for Diffusion-based Video Generation with Dynamic Latent Frame Rate. VGDFR adaptively adjusts the number of elements in latent space based on the motion frequency of the latent space content, using fewer tokens for low-frequency segments while preserving detail in high-frequency segments. Specifically, our key contributions are: (1) A dynamic frame rate scheduler for DiT video generation that adaptively assigns frame rates for video segments. (2) A novel latent-space frame merging method to align latent representations with their denoised counterparts before merging those redundant in low-resolution space. (3) A preference analysis of Rotary Positional Embeddings (RoPE) across DiT layers, informing a tailored RoPE strategy optimized for semantic and local information capture. Experiments show that VGDFR can achieve a speedup up to 3x for video generation with minimal quality degradation.
☆ FLIP Reasoning Challenge ICLR 2025
Over the past years, advances in artificial intelligence (AI) have demonstrated how AI can solve many perception and generation tasks, such as image classification and text writing, yet reasoning remains a challenge. This paper introduces the FLIP dataset, a benchmark for evaluating AI reasoning capabilities based on human verification tasks on the Idena blockchain. FLIP challenges present users with two orderings of 4 images, requiring them to identify the logically coherent one. By emphasizing sequential reasoning, visual storytelling, and common sense, FLIP provides a unique testbed for multimodal AI systems. Our experiments evaluate state-of-the-art models, leveraging both vision-language models (VLMs) and large language models (LLMs). Results reveal that even the best open-sourced and closed-sourced models achieve maximum accuracies of 75.5% and 77.9%, respectively, in zero-shot settings, compared to human performance of 95.3%. Captioning models aid reasoning models by providing text descriptions of images, yielding better results than when using the raw images directly, 69.6% vs. 75.2% for Gemini 1.5 Pro. Combining the predictions from 15 models in an ensemble increases the accuracy to 85.2%. These findings highlight the limitations of existing reasoning models and the need for robust multimodal benchmarks like FLIP. The full codebase and dataset will be available at https://github.com/aplesner/FLIP-Reasoning-Challenge.
comment: Published at First Workshop on Open Science for Foundation Models at ICLR 2025
☆ Human Aligned Compression for Robust Models CVPR 2025
Adversarial attacks on image models threaten system robustness by introducing imperceptible perturbations that cause incorrect predictions. We investigate human-aligned learned lossy compression as a defense mechanism, comparing two learned models (HiFiC and ELIC) against traditional JPEG across various quality levels. Our experiments on ImageNet subsets demonstrate that learned compression methods outperform JPEG, particularly for Vision Transformer architectures, by preserving semantically meaningful content while removing adversarial noise. Even in white-box settings where attackers can access the defense, these methods maintain substantial effectiveness. We also show that sequential compression--applying rounds of compression/decompression--significantly enhances defense efficacy while maintaining classification performance. Our findings reveal that human-aligned compression provides an effective, computationally efficient defense that protects the image features most relevant to human and machine understanding. It offers a practical approach to improving model robustness against adversarial threats.
comment: Presented at the Workshop AdvML at CVPR 2025
☆ Comparative Evaluation of Radiomics and Deep Learning Models for Disease Detection in Chest Radiography
The application of artificial intelligence (AI) in medical imaging has revolutionized diagnostic practices, enabling advanced analysis and interpretation of radiological data. This study presents a comprehensive evaluation of radiomics-based and deep learning-based approaches for disease detection in chest radiography, focusing on COVID-19, lung opacity, and viral pneumonia. While deep learning models, particularly convolutional neural networks (CNNs) and vision transformers (ViTs), learn directly from image data, radiomics-based models extract and analyze quantitative features, potentially providing advantages in data-limited scenarios. This study systematically compares the diagnostic accuracy and robustness of various AI models, including Decision Trees, Gradient Boosting, Random Forests, Support Vector Machines (SVM), and Multi-Layer Perceptrons (MLP) for radiomics, against state-of-the-art computer vision deep learning architectures. Performance metrics across varying sample sizes reveal insights into each model's efficacy, highlighting the contexts in which specific AI approaches may offer enhanced diagnostic capabilities. The results aim to inform the integration of AI-driven diagnostic tools in clinical practice, particularly in automated and high-throughput environments where timely, reliable diagnosis is critical. This comparative study addresses an essential gap, establishing guidance for the selection of AI models based on clinical and operational needs.
☆ SIDME: Self-supervised Image Demoiréing via Masked Encoder-Decoder Reconstruction
Moir\'e patterns, resulting from aliasing between object light signals and camera sampling frequencies, often degrade image quality during capture. Traditional demoir\'eing methods have generally treated images as a whole for processing and training, neglecting the unique signal characteristics of different color channels. Moreover, the randomness and variability of moir\'e pattern generation pose challenges to the robustness of existing methods when applied to real-world data. To address these issues, this paper presents SIDME (Self-supervised Image Demoir\'eing via Masked Encoder-Decoder Reconstruction), a novel model designed to generate high-quality visual images by effectively processing moir\'e patterns. SIDME combines a masked encoder-decoder architecture with self-supervised learning, allowing the model to reconstruct images using the inherent properties of camera sampling frequencies. A key innovation is the random masked image reconstructor, which utilizes an encoder-decoder structure to handle the reconstruction task. Furthermore, since the green channel in camera sampling has a higher sampling frequency compared to red and blue channels, a specialized self-supervised loss function is designed to improve the training efficiency and effectiveness. To ensure the generalization ability of the model, a self-supervised moir\'e image generation method has been developed to produce a dataset that closely mimics real-world conditions. Extensive experiments demonstrate that SIDME outperforms existing methods in processing real moir\'e pattern data, showing its superior generalization performance and robustness.
comment: 21 pages, 13 figures
☆ Cobra: Efficient Line Art COlorization with BRoAder References
The comic production industry requires reference-based line art colorization with high accuracy, efficiency, contextual consistency, and flexible control. A comic page often involves diverse characters, objects, and backgrounds, which complicates the coloring process. Despite advancements in diffusion models for image generation, their application in line art colorization remains limited, facing challenges related to handling extensive reference images, time-consuming inference, and flexible control. We investigate the necessity of extensive contextual image guidance on the quality of line art colorization. To address these challenges, we introduce Cobra, an efficient and versatile method that supports color hints and utilizes over 200 reference images while maintaining low latency. Central to Cobra is a Causal Sparse DiT architecture, which leverages specially designed positional encodings, causal sparse attention, and Key-Value Cache to effectively manage long-context references and ensure color identity consistency. Results demonstrate that Cobra achieves accurate line art colorization through extensive contextual reference, significantly enhancing inference speed and interactivity, thereby meeting critical industrial demands. We release our codes and models on our project page: https://zhuang2002.github.io/Cobra/.
comment: Project page with code: https://zhuang2002.github.io/Cobra/
☆ Coding-Prior Guided Diffusion Network for Video Deblurring
While recent video deblurring methods have advanced significantly, they often overlook two valuable prior information: (1) motion vectors (MVs) and coding residuals (CRs) from video codecs, which provide efficient inter-frame alignment cues, and (2) the rich real-world knowledge embedded in pre-trained diffusion generative models. We present CPGDNet, a novel two-stage framework that effectively leverages both coding priors and generative diffusion priors for high-quality deblurring. First, our coding-prior feature propagation (CPFP) module utilizes MVs for efficient frame alignment and CRs to generate attention masks, addressing motion inaccuracies and texture variations. Second, a coding-prior controlled generation (CPC) module network integrates coding priors into a pretrained diffusion model, guiding it to enhance critical regions and synthesize realistic details. Experiments demonstrate our method achieves state-of-the-art perceptual quality with up to 30% improvement in IQA metrics. Both the code and the codingprior-augmented dataset will be open-sourced.
☆ Uncertainty-Guided Coarse-to-Fine Tumor Segmentation with Anatomy-Aware Post-Processing SC
Reliable tumor segmentation in thoracic computed tomography (CT) remains challenging due to boundary ambiguity, class imbalance, and anatomical variability. We propose an uncertainty-guided, coarse-to-fine segmentation framework that combines full-volume tumor localization with refined region-of-interest (ROI) segmentation, enhanced by anatomically aware post-processing. The first-stage model generates a coarse prediction, followed by anatomically informed filtering based on lung overlap, proximity to lung surfaces, and component size. The resulting ROIs are segmented by a second-stage model trained with uncertainty-aware loss functions to improve accuracy and boundary calibration in ambiguous regions. Experiments on private and public datasets demonstrate improvements in Dice and Hausdorff scores, with fewer false positives and enhanced spatial interpretability. These results highlight the value of combining uncertainty modeling and anatomical priors in cascaded segmentation pipelines for robust and clinically meaningful tumor delineation. On the Orlando dataset, our framework improved Swin UNETR Dice from 0.4690 to 0.6447. Reduction in spurious components was strongly correlated with segmentation gains, underscoring the value of anatomically informed post-processing.
comment: 6 pages, 2 figures, to appear in IEEE ADSCA 2025
☆ Towards Realistic Low-Light Image Enhancement via ISP Driven Data Modeling
Deep neural networks (DNNs) have recently become the leading method for low-light image enhancement (LLIE). However, despite significant progress, their outputs may still exhibit issues such as amplified noise, incorrect white balance, or unnatural enhancements when deployed in real world applications. A key challenge is the lack of diverse, large scale training data that captures the complexities of low-light conditions and imaging pipelines. In this paper, we propose a novel image signal processing (ISP) driven data synthesis pipeline that addresses these challenges by generating unlimited paired training data. Specifically, our pipeline begins with easily collected high-quality normal-light images, which are first unprocessed into the RAW format using a reverse ISP. We then synthesize low-light degradations directly in the RAW domain. The resulting data is subsequently processed through a series of ISP stages, including white balance adjustment, color space conversion, tone mapping, and gamma correction, with controlled variations introduced at each stage. This broadens the degradation space and enhances the diversity of the training data, enabling the generated data to capture a wide range of degradations and the complexities inherent in the ISP pipeline. To demonstrate the effectiveness of our synthetic pipeline, we conduct extensive experiments using a vanilla UNet model consisting solely of convolutional layers, group normalization, GeLU activation, and convolutional block attention modules (CBAMs). Extensive testing across multiple datasets reveals that the vanilla UNet model trained with our data synthesis pipeline delivers high fidelity, visually appealing enhancement results, surpassing state-of-the-art (SOTA) methods both quantitatively and qualitatively.
comment: 17 pages, 11 tables, 10 figures
☆ Modality-Independent Explainable Detection of Inaccurate Organ Segmentations Using Denoising Autoencoders
In radiation therapy planning, inaccurate segmentations of organs at risk can result in suboptimal treatment delivery, if left undetected by the clinician. To address this challenge, we developed a denoising autoencoder-based method to detect inaccurate organ segmentations. We applied noise to ground truth organ segmentations, and the autoencoders were tasked to denoise them. Through the application of our method to organ segmentations generated on both MR and CT scans, we demonstrated that the method is independent of imaging modality. By providing reconstructions, our method offers visual information about inaccurate regions of the organ segmentations, leading to more explainable detection of suboptimal segmentations. We compared our method to existing approaches in the literature and demonstrated that it achieved superior performance for the majority of organs.
comment: Short version of this paper was accepted for poster presentation at IEEE ISBI 2025
☆ Beyond Patches: Mining Interpretable Part-Prototypes for Explainable AI
Deep learning has provided considerable advancements for multimedia systems, yet the interpretability of deep models remains a challenge. State-of-the-art post-hoc explainability methods, such as GradCAM, provide visual interpretation based on heatmaps but lack conceptual clarity. Prototype-based approaches, like ProtoPNet and PIPNet, offer a more structured explanation but rely on fixed patches, limiting their robustness and semantic consistency. To address these limitations, a part-prototypical concept mining network (PCMNet) is proposed that dynamically learns interpretable prototypes from meaningful regions. PCMNet clusters prototypes into concept groups, creating semantically grounded explanations without requiring additional annotations. Through a joint process of unsupervised part discovery and concept activation vector extraction, PCMNet effectively captures discriminative concepts and makes interpretable classification decisions. Our extensive experiments comparing PCMNet against state-of-the-art methods on multiple datasets show that it can provide a high level of interpretability, stability, and robustness under clean and occluded scenarios.
☆ CoMotion: Concurrent Multi-person 3D Motion ICLR 2025
We introduce an approach for detecting and tracking detailed 3D poses of multiple people from a single monocular camera stream. Our system maintains temporally coherent predictions in crowded scenes filled with difficult poses and occlusions. Our model performs both strong per-frame detection and a learned pose update to track people from frame to frame. Rather than match detections across time, poses are updated directly from a new input image, which enables online tracking through occlusion. We train on numerous image and video datasets leveraging pseudo-labeled annotations to produce a model that matches state-of-the-art systems in 3D pose estimation accuracy while being faster and more accurate in tracking multiple people through time. Code and weights are provided at https://github.com/apple/ml-comotion
comment: Accepted at ICLR 2025, for code and weights go to https://github.com/apple/ml-comotion
☆ Towards a General-Purpose Zero-Shot Synthetic Low-Light Image and Video Pipeline
Low-light conditions pose significant challenges for both human and machine annotation. This in turn has led to a lack of research into machine understanding for low-light images and (in particular) videos. A common approach is to apply annotations obtained from high quality datasets to synthetically created low light versions. In addition, these approaches are often limited through the use of unrealistic noise models. In this paper, we propose a new Degradation Estimation Network (DEN), which synthetically generates realistic standard RGB (sRGB) noise without the requirement for camera metadata. This is achieved by estimating the parameters of physics-informed noise distributions, trained in a self-supervised manner. This zero-shot approach allows our method to generate synthetic noisy content with a diverse range of realistic noise characteristics, unlike other methods which focus on recreating the noise characteristics of the training data. We evaluate our proposed synthetic pipeline using various methods trained on its synthetic data for typical low-light tasks including synthetic noise replication, video enhancement, and object detection, showing improvements of up to 24\% KLD, 21\% LPIPS, and 62\% AP$_{50-95}$, respectively.
☆ RADLER: Radar Object Detection Leveraging Semantic 3D City Models and Self-Supervised Radar-Image Learning CVPR
Semantic 3D city models are worldwide easy-accessible, providing accurate, object-oriented, and semantic-rich 3D priors. To date, their potential to mitigate the noise impact on radar object detection remains under-explored. In this paper, we first introduce a unique dataset, RadarCity, comprising 54K synchronized radar-image pairs and semantic 3D city models. Moreover, we propose a novel neural network, RADLER, leveraging the effectiveness of contrastive self-supervised learning (SSL) and semantic 3D city models to enhance radar object detection of pedestrians, cyclists, and cars. Specifically, we first obtain the robust radar features via a SSL network in the radar-image pretext task. We then use a simple yet effective feature fusion strategy to incorporate semantic-depth features from semantic 3D city models. Having prior 3D information as guidance, RADLER obtains more fine-grained details to enhance radar object detection. We extensively evaluate RADLER on the collected RadarCity dataset and demonstrate average improvements of 5.46% in mean avarage precision (mAP) and 3.51% in mean avarage recall (mAR) over previous radar object detection methods. We believe this work will foster further research on semantic-guided and map-supported radar object detection. Our project page is publicly available athttps://gpp-communication.github.io/RADLER .
comment: The paper accepted for CVPRW '25 (PBVS 2025 - the Perception Beyond the Visible Spectrum)
☆ CodingHomo: Bootstrapping Deep Homography With Video Coding
Homography estimation is a fundamental task in computer vision with applications in diverse fields. Recent advances in deep learning have improved homography estimation, particularly with unsupervised learning approaches, offering increased robustness and generalizability. However, accurately predicting homography, especially in complex motions, remains a challenge. In response, this work introduces a novel method leveraging video coding, particularly by harnessing inherent motion vectors (MVs) present in videos. We present CodingHomo, an unsupervised framework for homography estimation. Our framework features a Mask-Guided Fusion (MGF) module that identifies and utilizes beneficial features among the MVs, thereby enhancing the accuracy of homography prediction. Additionally, the Mask-Guided Homography Estimation (MGHE) module is presented for eliminating undesired features in the coarse-to-fine homography refinement process. CodingHomo outperforms existing state-of-the-art unsupervised methods, delivering good robustness and generalizability. The code and dataset are available at: \href{github}{https://github.com/liuyike422/CodingHomo
☆ FocusedAD: Character-centric Movie Audio Description
Movie Audio Description (AD) aims to narrate visual content during dialogue-free segments, particularly benefiting blind and visually impaired (BVI) audiences. Compared with general video captioning, AD demands plot-relevant narration with explicit character name references, posing unique challenges in movie understanding.To identify active main characters and focus on storyline-relevant regions, we propose FocusedAD, a novel framework that delivers character-centric movie audio descriptions. It includes: (i) a Character Perception Module(CPM) for tracking character regions and linking them to names; (ii) a Dynamic Prior Module(DPM) that injects contextual cues from prior ADs and subtitles via learnable soft prompts; and (iii) a Focused Caption Module(FCM) that generates narrations enriched with plot-relevant details and named characters. To overcome limitations in character identification, we also introduce an automated pipeline for building character query banks. FocusedAD achieves state-of-the-art performance on multiple benchmarks, including strong zero-shot results on MAD-eval-Named and our newly proposed Cinepile-AD dataset. Code and data will be released at https://github.com/Thorin215/FocusedAD .
comment: Code and Demo link: https://github.com/Thorin215/FocusedAD
☆ Efficient Contrastive Decoding with Probabilistic Hallucination Detection - Mitigating Hallucinations in Large Vision Language Models -
Despite recent advances in Large Vision Language Models (LVLMs), these models still suffer from generating hallucinatory responses that do not align with the visual input provided. To mitigate such hallucinations, we introduce Efficient Contrastive Decoding (ECD), a simple method that leverages probabilistic hallucination detection to shift the output distribution towards contextually accurate answers at inference time. By contrasting token probabilities and hallucination scores, ECD subtracts hallucinated concepts from the original distribution, effectively suppressing hallucinations. Notably, our proposed method can be applied to any open-source LVLM and does not require additional LVLM training. We evaluate our method on several benchmark datasets and across different LVLMs. Our experiments show that ECD effectively mitigates hallucinations, outperforming state-of-the-art methods with respect to performance on LVLM benchmarks and computation time.
☆ Weakly Semi-supervised Whole Slide Image Classification by Two-level Cross Consistency Supervision
Computer-aided Whole Slide Image (WSI) classification has the potential to enhance the accuracy and efficiency of clinical pathological diagnosis. It is commonly formulated as a Multiple Instance Learning (MIL) problem, where each WSI is treated as a bag and the small patches extracted from the WSI are considered instances within that bag. However, obtaining labels for a large number of bags is a costly and time-consuming process, particularly when utilizing existing WSIs for new classification tasks. This limitation renders most existing WSI classification methods ineffective. To address this issue, we propose a novel WSI classification problem setting, more aligned with clinical practice, termed Weakly Semi-supervised Whole slide image Classification (WSWC). In WSWC, a small number of bags are labeled, while a significant number of bags remain unlabeled. The MIL nature of the WSWC problem, coupled with the absence of patch labels, distinguishes it from typical semi-supervised image classification problems, making existing algorithms for natural images unsuitable for directly solving the WSWC problem. In this paper, we present a concise and efficient framework, named CroCo, to tackle the WSWC problem through two-level Cross Consistency supervision. CroCo comprises two heterogeneous classifier branches capable of performing both instance classification and bag classification. The fundamental idea is to establish cross-consistency supervision at both the bag-level and instance-level between the two branches during training. Extensive experiments conducted on four datasets demonstrate that CroCo achieves superior bag classification and instance classification performance compared to other comparative methods when limited WSIs with bag labels are available. To the best of our knowledge, this paper presents for the first time the WSWC problem and gives a successful resolution.
☆ Anti-Aesthetics: Protecting Facial Privacy against Customized Text-to-Image Synthesis
The rise of customized diffusion models has spurred a boom in personalized visual content creation, but also poses risks of malicious misuse, severely threatening personal privacy and copyright protection. Some studies show that the aesthetic properties of images are highly positively correlated with human perception of image quality. Inspired by this, we approach the problem from a novel and intriguing aesthetic perspective to degrade the generation quality of maliciously customized models, thereby achieving better protection of facial identity. Specifically, we propose a Hierarchical Anti-Aesthetic (HAA) framework to fully explore aesthetic cues, which consists of two key branches: 1) Global Anti-Aesthetics: By establishing a global anti-aesthetic reward mechanism and a global anti-aesthetic loss, it can degrade the overall aesthetics of the generated content; 2) Local Anti-Aesthetics: A local anti-aesthetic reward mechanism and a local anti-aesthetic loss are designed to guide adversarial perturbations to disrupt local facial identity. By seamlessly integrating both branches, our HAA effectively achieves the goal of anti-aesthetics from a global to a local level during customized generation. Extensive experiments show that HAA outperforms existing SOTA methods largely in identity removal, providing a powerful tool for protecting facial privacy and copyright.
☆ Remote sensing colour image semantic segmentation of trails created by large herbivorous Mammals
Detection of spatial areas where biodiversity is at risk is of paramount importance for the conservation and monitoring of ecosystems. Large terrestrial mammalian herbivores are keystone species as their activity not only has deep effects on soils, plants, and animals, but also shapes landscapes, as large herbivores act as allogenic ecosystem engineers. One key landscape feature that indicates intense herbivore activity and potentially impacts biodiversity is the formation of grazing trails. Grazing trails are formed by the continuous trampling activity of large herbivores that can produce complex networks of tracks of bare soil. Here, we evaluated different algorithms based on machine learning techniques to identify grazing trails. Our goal is to automatically detect potential areas with intense herbivory activity, which might be beneficial for conservation and management plans. We have applied five semantic segmentation methods combined with fourteen encoders aimed at mapping grazing trails on aerial images. Our results indicate that in most cases the chosen methodology successfully mapped the trails, although there were a few instances where the actual trail structure was underestimated. The UNet architecture with the MambaOut encoder was the best architecture for mapping trails. The proposed approach could be applied to develop tools for mapping and monitoring temporal changes in these landscape structures to support habitat conservation and land management programs. This is the first time, to the best of our knowledge, that competitive image segmentation results are obtained for the detection and delineation of trails of large herbivorous mammals.
comment: 24 pages, 6 figures. Submitted to Computers and Geosciences
☆ A Diffusion-Based Framework for Terrain-Aware Remote Sensing Image Reconstruction
Remote sensing imagery is essential for environmental monitoring, agricultural management, and disaster response. However, data loss due to cloud cover, sensor failures, or incomplete acquisition-especially in high-resolution and high-frequency tasks-severely limits satellite imagery's effectiveness. Traditional interpolation methods struggle with large missing areas and complex structures. Remote sensing imagery consists of multiple bands, each with distinct meanings, and ensuring consistency across bands is critical to avoid anomalies in the combined images. This paper proposes SatelliteMaker, a diffusion-based method that reconstructs missing data across varying levels of data loss while maintaining spatial, spectral, and temporal consistency. We also propose Digital Elevation Model (DEM) as a conditioning input and use tailored prompts to generate realistic images, making diffusion models applicable to quantitative remote sensing tasks. Additionally, we propose a VGG-Adapter module based on Distribution Loss, which reduces distribution discrepancy and ensures style consistency. Extensive experiments show that SatelliteMaker achieves state-of-the-art performance across multiple tasks.
☆ Logits DeConfusion with CLIP for Few-Shot Learning CVPR 2025
With its powerful visual-language alignment capability, CLIP performs well in zero-shot and few-shot learning tasks. However, we found in experiments that CLIP's logits suffer from serious inter-class confusion problems in downstream tasks, and the ambiguity between categories seriously affects the accuracy. To address this challenge, we propose a novel method called Logits DeConfusion, which effectively learns and eliminates inter-class confusion in logits by combining our Multi-level Adapter Fusion (MAF) module with our Inter-Class Deconfusion (ICD) module. Our MAF extracts features from different levels and fuses them uniformly to enhance feature representation. Our ICD learnably eliminates inter-class confusion in logits with a residual structure. Experimental results show that our method can significantly improve the classification performance and alleviate the inter-class confusion problem. The code is available at https://github.com/LiShuo1001/LDC.
comment: CVPR 2025
☆ Metric-Solver: Sliding Anchored Metric Depth Estimation from a Single Image
Accurate and generalizable metric depth estimation is crucial for various computer vision applications but remains challenging due to the diverse depth scales encountered in indoor and outdoor environments. In this paper, we introduce Metric-Solver, a novel sliding anchor-based metric depth estimation method that dynamically adapts to varying scene scales. Our approach leverages an anchor-based representation, where a reference depth serves as an anchor to separate and normalize the scene depth into two components: scaled near-field depth and tapered far-field depth. The anchor acts as a normalization factor, enabling the near-field depth to be normalized within a consistent range while mapping far-field depth smoothly toward zero. Through this approach, any depth from zero to infinity in the scene can be represented within a unified representation, effectively eliminating the need to manually account for scene scale variations. More importantly, for the same scene, the anchor can slide along the depth axis, dynamically adjusting to different depth scales. A smaller anchor provides higher resolution in the near-field, improving depth precision for closer objects while a larger anchor improves depth estimation in far regions. This adaptability enables the model to handle depth predictions at varying distances and ensure strong generalization across datasets. Our design enables a unified and adaptive depth representation across diverse environments. Extensive experiments demonstrate that Metric-Solver outperforms existing methods in both accuracy and cross-dataset generalization.
comment: Our project page: https://tele-ai.github.io/MetricSolver/
☆ Generalized Visual Relation Detection with Diffusion Models
Visual relation detection (VRD) aims to identify relationships (or interactions) between object pairs in an image. Although recent VRD models have achieved impressive performance, they are all restricted to pre-defined relation categories, while failing to consider the semantic ambiguity characteristic of visual relations. Unlike objects, the appearance of visual relations is always subtle and can be described by multiple predicate words from different perspectives, e.g., ``ride'' can be depicted as ``race'' and ``sit on'', from the sports and spatial position views, respectively. To this end, we propose to model visual relations as continuous embeddings, and design diffusion models to achieve generalized VRD in a conditional generative manner, termed Diff-VRD. We model the diffusion process in a latent space and generate all possible relations in the image as an embedding sequence. During the generation, the visual and text embeddings of subject-object pairs serve as conditional signals and are injected via cross-attention. After the generation, we design a subsequent matching stage to assign the relation words to subject-object pairs by considering their semantic similarities. Benefiting from the diffusion-based generative process, our Diff-VRD is able to generate visual relations beyond the pre-defined category labels of datasets. To properly evaluate this generalized VRD task, we introduce two evaluation metrics, i.e., text-to-image retrieval and SPICE PR Curve inspired by image captioning. Extensive experiments in both human-object interaction (HOI) detection and scene graph generation (SGG) benchmarks attest to the superiority and effectiveness of Diff-VRD.
comment: Under review at IEEE TCSVT. The Appendix is provided additionally
☆ AttentionDrop: A Novel Regularization Method for Transformer Models
Transformer-based architectures achieve state-of-the-art performance across a wide range of tasks in natural language processing, computer vision, and speech. However, their immense capacity often leads to overfitting, especially when training data is limited or noisy. We propose AttentionDrop, a unified family of stochastic regularization techniques that operate directly on the self-attention distributions. We introduces three variants: 1. Hard Attention Masking: randomly zeroes out top-k attention logits per query to encourage diverse context utilization. 2. Blurred Attention Smoothing: applies a dynamic Gaussian convolution over attention logits to diffuse overly peaked distributions. 3. Consistency-Regularized AttentionDrop: enforces output stability under multiple independent AttentionDrop perturbations via a KL-based consistency loss.
comment: 26 pages
☆ Self-alignment of Large Video Language Models with Refined Regularized Preference Optimization
Despite recent advances in Large Video Language Models (LVLMs), they still struggle with fine-grained temporal understanding, hallucinate, and often make simple mistakes on even simple video question-answering tasks, all of which pose significant challenges to their safe and reliable deployment in real-world applications. To address these limitations, we propose a self-alignment framework that enables LVLMs to learn from their own errors. Our proposed framework first obtains a training set of preferred and non-preferred response pairs, where non-preferred responses are generated by incorporating common error patterns that often occur due to inadequate spatio-temporal understanding, spurious correlations between co-occurring concepts, and over-reliance on linguistic cues while neglecting the vision modality, among others. To facilitate self-alignment of LVLMs with the constructed preferred and non-preferred response pairs, we introduce Refined Regularized Preference Optimization (RRPO), a novel preference optimization method that utilizes sub-sequence-level refined rewards and token-wise KL regularization to address the limitations of Direct Preference Optimization (DPO). We demonstrate that RRPO achieves more precise alignment and more stable training compared to DPO. Our experiments and analysis validate the effectiveness of our approach across diverse video tasks, including video hallucination, short- and long-video understanding, and fine-grained temporal reasoning.
☆ DC-SAM: In-Context Segment Anything in Images and Videos via Dual Consistency
Given a single labeled example, in-context segmentation aims to segment corresponding objects. This setting, known as one-shot segmentation in few-shot learning, explores the segmentation model's generalization ability and has been applied to various vision tasks, including scene understanding and image/video editing. While recent Segment Anything Models have achieved state-of-the-art results in interactive segmentation, these approaches are not directly applicable to in-context segmentation. In this work, we propose the Dual Consistency SAM (DC-SAM) method based on prompt-tuning to adapt SAM and SAM2 for in-context segmentation of both images and videos. Our key insights are to enhance the features of the SAM's prompt encoder in segmentation by providing high-quality visual prompts. When generating a mask prior, we fuse the SAM features to better align the prompt encoder. Then, we design a cycle-consistent cross-attention on fused features and initial visual prompts. Next, a dual-branch design is provided by using the discriminative positive and negative prompts in the prompt encoder. Furthermore, we design a simple mask-tube training strategy to adopt our proposed dual consistency method into the mask tube. Although the proposed DC-SAM is primarily designed for images, it can be seamlessly extended to the video domain with the support of SAM2. Given the absence of in-context segmentation in the video domain, we manually curate and construct the first benchmark from existing video segmentation datasets, named In-Context Video Object Segmentation (IC-VOS), to better assess the in-context capability of the model. Extensive experiments demonstrate that our method achieves 55.5 (+1.4) mIoU on COCO-20i, 73.0 (+1.1) mIoU on PASCAL-5i, and a J&F score of 71.52 on the proposed IC-VOS benchmark. Our source code and benchmark are available at https://github.com/zaplm/DC-SAM.
☆ Single-shot Star-convex Polygon-based Instance Segmentation for Spatially-correlated Biomedical Objects
Biomedical images often contain objects known to be spatially correlated or nested due to their inherent properties, leading to semantic relations. Examples include cell nuclei being nested within eukaryotic cells and colonies growing exclusively within their culture dishes. While these semantic relations bear key importance, detection tasks are often formulated independently, requiring multi-shot analysis pipelines. Importantly, spatial correlation could constitute a fundamental prior facilitating learning of more meaningful representations for tasks like instance segmentation. This knowledge has, thus far, not been utilised by the biomedical computer vision community. We argue that the instance segmentation of two or more categories of objects can be achieved in parallel. We achieve this via two architectures HydraStarDist (HSD) and the novel (HSD-WBR) based on the widely-used StarDist (SD), to take advantage of the star-convexity of our target objects. HSD and HSD-WBR are constructed to be capable of incorporating their interactions as constraints into account. HSD implicitly incorporates spatial correlation priors based on object interaction through a joint encoder. HSD-WBR further enforces the prior in a regularisation layer with the penalty we proposed named Within Boundary Regularisation Penalty (WBR). Both architectures achieve nested instance segmentation in a single shot. We demonstrate their competitiveness based on $IoU_R$ and AP and superiority in a new, task-relevant criteria, Joint TP rate (JTPR) compared to their baseline SD and Cellpose. Our approach can be further modified to capture partial-inclusion/-exclusion in multi-object interactions in fluorescent or brightfield microscopy or digital imaging. Finally, our strategy suggests gains by making this learning single-shot and computationally efficient.
comment: 12 pages, 8 figures
☆ Modular-Cam: Modular Dynamic Camera-view Video Generation with LLM AAAI 2025
Text-to-Video generation, which utilizes the provided text prompt to generate high-quality videos, has drawn increasing attention and achieved great success due to the development of diffusion models recently. Existing methods mainly rely on a pre-trained text encoder to capture the semantic information and perform cross attention with the encoded text prompt to guide the generation of video. However, when it comes to complex prompts that contain dynamic scenes and multiple camera-view transformations, these methods can not decompose the overall information into separate scenes, as well as fail to smoothly change scenes based on the corresponding camera-views. To solve these problems, we propose a novel method, i.e., Modular-Cam. Specifically, to better understand a given complex prompt, we utilize a large language model to analyze user instructions and decouple them into multiple scenes together with transition actions. To generate a video containing dynamic scenes that match the given camera-views, we incorporate the widely-used temporal transformer into the diffusion model to ensure continuity within a single scene and propose CamOperator, a modular network based module that well controls the camera movements. Moreover, we propose AdaControlNet, which utilizes ControlNet to ensure consistency across scenes and adaptively adjusts the color tone of the generated video. Extensive qualitative and quantitative experiments prove our proposed Modular-Cam's strong capability of generating multi-scene videos together with its ability to achieve fine-grained control of camera movements. Generated results are available at https://modular-cam.github.io.
comment: AAAI 2025 Poster
☆ pix2pockets: Shot Suggestions in 8-Ball Pool from a Single Image in the Wild SC
Computer vision models have seen increased usage in sports, and reinforcement learning (RL) is famous for beating humans in strategic games such as Chess and Go. In this paper, we are interested in building upon these advances and examining the game of classic 8-ball pool. We introduce pix2pockets, a foundation for an RL-assisted pool coach. Given a single image of a pool table, we first aim to detect the table and the balls and then propose the optimal shot suggestion. For the first task, we build a dataset with 195 diverse images where we manually annotate all balls and table dots, leading to 5748 object segmentation masks. For the second task, we build a standardized RL environment that allows easy development and benchmarking of any RL algorithm. Our object detection model yields an AP50 of 91.2 while our ball location pipeline obtains an error of only 0.4 cm. Furthermore, we compare standard RL algorithms to set a baseline for the shot suggestion task and we show that all of them fail to pocket all balls without making a foul move. We also present a simple baseline that achieves a per-shot success rate of 94.7% and clears a full game in a single turn 30% of the time.
comment: 15 pages, 7 figures, to be published in SCIA 2025
☆ RadMamba: Efficient Human Activity Recognition through Radar-based Micro-Doppler-Oriented Mamba State-Space Model
Radar-based HAR has emerged as a promising alternative to conventional monitoring approaches, such as wearable devices and camera-based systems, due to its unique privacy preservation and robustness advantages. However, existing solutions based on convolutional and recurrent neural networks, although effective, are computationally demanding during deployment. This limits their applicability in scenarios with constrained resources or those requiring multiple sensors. Advanced architectures, such as ViT and SSM architectures, offer improved modeling capabilities and have made efforts toward lightweight designs. However, their computational complexity remains relatively high. To leverage the strengths of transformer architectures while simultaneously enhancing accuracy and reducing computational complexity, this paper introduces RadMamba, a parameter-efficient, radar micro-Doppler-oriented Mamba SSM specifically tailored for radar-based HAR. Across three diverse datasets, RadMamba matches the top-performing previous model's 99.8% classification accuracy on Dataset DIAT with only 1/400 of its parameters and equals the leading models' 92.0% accuracy on Dataset CI4R with merely 1/10 of their parameters. In scenarios with continuous sequences of actions evaluated on Dataset UoG2020, RadMamba surpasses other models with significantly higher parameter counts by at least 3%, achieving this with only 6.7k parameters. Our code is available at: https://github.com/lab-emi/AIRHAR.
comment: Under Review
☆ Object Placement for Anything ICME 2025
Object placement aims to determine the appropriate placement (\emph{e.g.}, location and size) of a foreground object when placing it on the background image. Most previous works are limited by small-scale labeled dataset, which hinders the real-world application of object placement. In this work, we devise a semi-supervised framework which can exploit large-scale unlabeled dataset to promote the generalization ability of discriminative object placement models. The discriminative models predict the rationality label for each foreground placement given a foreground-background pair. To better leverage the labeled data, under the semi-supervised framework, we further propose to transfer the knowledge of rationality variation, \emph{i.e.}, whether the change of foreground placement would result in the change of rationality label, from labeled data to unlabeled data. Extensive experiments demonstrate that our framework can effectively enhance the generalization ability of discriminative object placement models.
comment: accepted by ICME 2025
☆ Understanding Attention Mechanism in Video Diffusion Models
Text-to-video (T2V) synthesis models, such as OpenAI's Sora, have garnered significant attention due to their ability to generate high-quality videos from a text prompt. In diffusion-based T2V models, the attention mechanism is a critical component. However, it remains unclear what intermediate features are learned and how attention blocks in T2V models affect various aspects of video synthesis, such as image quality and temporal consistency. In this paper, we conduct an in-depth perturbation analysis of the spatial and temporal attention blocks of T2V models using an information-theoretic approach. Our results indicate that temporal and spatial attention maps affect not only the timing and layout of the videos but also the complexity of spatiotemporal elements and the aesthetic quality of the synthesized videos. Notably, high-entropy attention maps are often key elements linked to superior video quality, whereas low-entropy attention maps are associated with the video's intra-frame structure. Based on our findings, we propose two novel methods to enhance video quality and enable text-guided video editing. These methods rely entirely on lightweight manipulation of the attention matrices in T2V models. The efficacy and effectiveness of our methods are further validated through experimental evaluation across multiple datasets.
☆ Action Anticipation from SoccerNet Football Video Broadcasts CVPR
Artificial intelligence has revolutionized the way we analyze sports videos, whether to understand the actions of games in long untrimmed videos or to anticipate the player's motion in future frames. Despite these efforts, little attention has been given to anticipating game actions before they occur. In this work, we introduce the task of action anticipation for football broadcast videos, which consists in predicting future actions in unobserved future frames, within a five- or ten-second anticipation window. To benchmark this task, we release a new dataset, namely the SoccerNet Ball Action Anticipation dataset, based on SoccerNet Ball Action Spotting. Additionally, we propose a Football Action ANticipation TRAnsformer (FAANTRA), a baseline method that adapts FUTR, a state-of-the-art action anticipation model, to predict ball-related actions. To evaluate action anticipation, we introduce new metrics, including mAP@$\delta$, which evaluates the temporal precision of predicted future actions, as well as mAP@$\infty$, which evaluates their occurrence within the anticipation window. We also conduct extensive ablation studies to examine the impact of various task settings, input configurations, and model architectures. Experimental results highlight both the feasibility and challenges of action anticipation in football videos, providing valuable insights into the design of predictive models for sports analytics. By forecasting actions before they unfold, our work will enable applications in automated broadcasting, tactical analysis, and player decision-making. Our dataset and code are publicly available at https://github.com/MohamadDalal/FAANTRA.
comment: 15 pages, 14 figures. To be published in the CVSports CVPR workshop
☆ MixSignGraph: A Sign Sequence is Worth Mixed Graphs of Nodes
Recent advances in sign language research have benefited from CNN-based backbones, which are primarily transferred from traditional computer vision tasks (\eg object identification, image recognition). However, these CNN-based backbones usually excel at extracting features like contours and texture, but may struggle with capturing sign-related features. In fact, sign language tasks require focusing on sign-related regions, including the collaboration between different regions (\eg left hand region and right hand region) and the effective content in a single region. To capture such region-related features, we introduce MixSignGraph, which represents sign sequences as a group of mixed graphs and designs the following three graph modules for feature extraction, \ie Local Sign Graph (LSG) module, Temporal Sign Graph (TSG) module and Hierarchical Sign Graph (HSG) module. Specifically, the LSG module learns the correlation of intra-frame cross-region features within one frame, \ie focusing on spatial features. The TSG module tracks the interaction of inter-frame cross-region features among adjacent frames, \ie focusing on temporal features. The HSG module aggregates the same-region features from different-granularity feature maps of a frame, \ie focusing on hierarchical features. In addition, to further improve the performance of sign language tasks without gloss annotations, we propose a simple yet counter-intuitive Text-driven CTC Pre-training (TCP) method, which generates pseudo gloss labels from text labels for model pre-training. Extensive experiments conducted on current five public sign language datasets demonstrate the superior performance of the proposed model. Notably, our model surpasses the SOTA models on multiple sign language tasks across several datasets, without relying on any additional cues.
comment: 17 pages, 9 figures, submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI). This is a regular paper submission
☆ Instruction-augmented Multimodal Alignment for Image-Text and Element Matching CVPR 2025
With the rapid advancement of text-to-image (T2I) generation models, assessing the semantic alignment between generated images and text descriptions has become a significant research challenge. Current methods, including those based on Visual Question Answering (VQA), still struggle with fine-grained assessments and precise quantification of image-text alignment. This paper presents an improved evaluation method named Instruction-augmented Multimodal Alignment for Image-Text and Element Matching (iMatch), which evaluates image-text semantic alignment by fine-tuning multimodal large language models. We introduce four innovative augmentation strategies: First, the QAlign strategy creates a precise probabilistic mapping to convert discrete scores from multimodal large language models into continuous matching scores. Second, a validation set augmentation strategy uses pseudo-labels from model predictions to expand training data, boosting the model's generalization performance. Third, an element augmentation strategy integrates element category labels to refine the model's understanding of image-text matching. Fourth, an image augmentation strategy employs techniques like random lighting to increase the model's robustness. Additionally, we propose prompt type augmentation and score perturbation strategies to further enhance the accuracy of element assessments. Our experimental results show that the iMatch method significantly surpasses existing methods, confirming its effectiveness and practical value. Furthermore, our iMatch won first place in the CVPR NTIRE 2025 Text to Image Generation Model Quality Assessment - Track 1 Image-Text Alignment.
comment: Accepted to CVPR 2025 Workshop
☆ A Complex-valued SAR Foundation Model Based on Physically Inspired Representation Learning
Vision foundation models in remote sensing have been extensively studied due to their superior generalization on various downstream tasks. Synthetic Aperture Radar (SAR) offers all-day, all-weather imaging capabilities, providing significant advantages for Earth observation. However, establishing a foundation model for SAR image interpretation inevitably encounters the challenges of insufficient information utilization and poor interpretability. In this paper, we propose a remote sensing foundation model based on complex-valued SAR data, which simulates the polarimetric decomposition process for pre-training, i.e., characterizing pixel scattering intensity as a weighted combination of scattering bases and scattering coefficients, thereby endowing the foundation model with physical interpretability. Specifically, we construct a series of scattering queries, each representing an independent and meaningful scattering basis, which interact with SAR features in the scattering query decoder and output the corresponding scattering coefficient. To guide the pre-training process, polarimetric decomposition loss and power self-supervision loss are constructed. The former aligns the predicted coefficients with Yamaguchi coefficients, while the latter reconstructs power from the predicted coefficients and compares it to the input image's power. The performance of our foundation model is validated on six typical downstream tasks, achieving state-of-the-art results. Notably, the foundation model can extract stable feature representations and exhibits strong generalization, even in data-scarce conditions.
☆ A Review of YOLOv12: Attention-Based Enhancements vs. Previous Versions
The YOLO (You Only Look Once) series has been a leading framework in real-time object detection, consistently improving the balance between speed and accuracy. However, integrating attention mechanisms into YOLO has been challenging due to their high computational overhead. YOLOv12 introduces a novel approach that successfully incorporates attention-based enhancements while preserving real-time performance. This paper provides a comprehensive review of YOLOv12's architectural innovations, including Area Attention for computationally efficient self-attention, Residual Efficient Layer Aggregation Networks for improved feature aggregation, and FlashAttention for optimized memory access. Additionally, we benchmark YOLOv12 against prior YOLO versions and competing object detectors, analyzing its improvements in accuracy, inference speed, and computational efficiency. Through this analysis, we demonstrate how YOLOv12 advances real-time object detection by refining the latency-accuracy trade-off and optimizing computational resources.
☆ Analysis of Pseudo-Labeling for Online Source-Free Universal Domain Adaptation
A domain (distribution) shift between training and test data often hinders the real-world performance of deep neural networks, necessitating unsupervised domain adaptation (UDA) to bridge this gap. Online source-free UDA has emerged as a solution for practical scenarios where access to source data is restricted and target data is received as a continuous stream. However, the open-world nature of many real-world applications additionally introduces category shifts meaning that the source and target label spaces may differ. Online source-free universal domain adaptation (SF-UniDA) addresses this challenge. Existing methods mainly rely on self-training with pseudo-labels, yet the relationship between pseudo-labeling and adaptation outcomes has not been studied yet. To bridge this gap, we conduct a systematic analysis through controlled experiments with simulated pseudo-labeling, offering valuable insights into pseudo-labeling for online SF-UniDA. Our findings reveal a substantial gap between the current state-of-the-art and the upper bound of adaptation achieved with perfect pseudo-labeling. Moreover, we show that a contrastive loss enables effective adaptation even with moderate pseudo-label accuracy, while a cross-entropy loss, though less robust to pseudo-label errors, achieves superior results when pseudo-labeling approaches perfection. Lastly, our findings indicate that pseudo-label accuracy is in general more crucial than quantity, suggesting that prioritizing fewer but high-confidence pseudo-labels is beneficial. Overall, our study highlights the critical role of pseudo-labeling in (online) SF-UniDA and provides actionable insights to drive future advancements in the field. Our code is available at https://github.com/pascalschlachter/PLAnalysis.
comment: Submitted to the 33rd European Signal Processing Conference (EUSIPCO 2025)
☆ Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions CVPR
Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges. This survey provides a wide-ranging examination of the anti-UAV domain, centering on three core objectives-classification, detection, and tracking-while detailing emerging methodologies such as diffusion-based data synthesis, multi-modal fusion, vision-language modeling, self-supervised learning, and reinforcement learning. We systematically evaluate state-of-the-art solutions across both single-modality and multi-sensor pipelines (spanning RGB, infrared, audio, radar, and RF) and discuss large-scale as well as adversarially oriented benchmarks. Our analysis reveals persistent gaps in real-time performance, stealth detection, and swarm-based scenarios, underscoring pressing needs for robust, adaptive anti-UAV systems. By highlighting open research directions, we aim to foster innovation and guide the development of next-generation defense strategies in an era marked by the extensive use of UAVs.
comment: Accepted at CVPR Workshop Anti-UAV 2025. 15 pages
☆ Exploring Video-Based Driver Activity Recognition under Noisy Labels
As an open research topic in the field of deep learning, learning with noisy labels has attracted much attention and grown rapidly over the past ten years. Learning with label noise is crucial for driver distraction behavior recognition, as real-world video data often contains mislabeled samples, impacting model reliability and performance. However, label noise learning is barely explored in the driver activity recognition field. In this paper, we propose the first label noise learning approach for the driver activity recognition task. Based on the cluster assumption, we initially enable the model to learn clustering-friendly low-dimensional representations from given videos and assign the resultant embeddings into clusters. We subsequently perform co-refinement within each cluster to smooth the classifier outputs. Furthermore, we propose a flexible sample selection strategy that combines two selection criteria without relying on any hyperparameters to filter clean samples from the training dataset. We also incorporate a self-adaptive parameter into the sample selection process to enforce balancing across classes. A comprehensive variety of experiments on the public Drive&Act dataset for all granularity levels demonstrates the superior performance of our method in comparison with other label-denoising methods derived from the image classification field. The source code is available at https://github.com/ilonafan/DAR-noisy-labels.
comment: The source code is available at https://github.com/ilonafan/DAR-noisy-labels
☆ Novel-view X-ray Projection Synthesis through Geometry-Integrated Deep Learning
X-ray imaging plays a crucial role in the medical field, providing essential insights into the internal anatomy of patients for diagnostics, image-guided procedures, and clinical decision-making. Traditional techniques often require multiple X-ray projections from various angles to obtain a comprehensive view, leading to increased radiation exposure and more complex clinical processes. This paper explores an innovative approach using the DL-GIPS model, which synthesizes X-ray projections from new viewpoints by leveraging a single existing projection. The model strategically manipulates geometry and texture features extracted from an initial projection to match new viewing angles. It then synthesizes the final projection by merging these modified geometry features with consistent texture information through an advanced image generation process. We demonstrate the effectiveness and broad applicability of the DL-GIPS framework through lung imaging examples, highlighting its potential to revolutionize stereoscopic and volumetric imaging by minimizing the need for extensive data acquisition.
comment: 6 pages, 3 figures, 1 table
☆ Flow Intelligence: Robust Feature Matching via Temporal Signature Correlation
Feature matching across video streams remains a cornerstone challenge in computer vision. Increasingly, robust multimodal matching has garnered interest in robotics, surveillance, remote sensing, and medical imaging. While traditional rely on detecting and matching spatial features, they break down when faced with noisy, misaligned, or cross-modal data. Recent deep learning methods have improved robustness through learned representations, but remain constrained by their dependence on extensive training data and computational demands. We present Flow Intelligence, a paradigm-shifting approach that moves beyond spatial features by focusing on temporal motion patterns exclusively. Instead of detecting traditional keypoints, our method extracts motion signatures from pixel blocks across consecutive frames and extract temporal motion signatures between videos. These motion-based descriptors achieve natural invariance to translation, rotation, and scale variations while remaining robust across different imaging modalities. This novel approach also requires no pretraining data, eliminates the need for spatial feature detection, enables cross-modal matching using only temporal motion, and it outperforms existing methods in challenging scenarios where traditional approaches fail. By leveraging motion rather than appearance, Flow Intelligence enables robust, real-time video feature matching in diverse environments.
☆ R-Meshfusion: Reinforcement Learning Powered Sparse-View Mesh Reconstruction with Diffusion Priors
Mesh reconstruction from multi-view images is a fundamental problem in computer vision, but its performance degrades significantly under sparse-view conditions, especially in unseen regions where no ground-truth observations are available. While recent advances in diffusion models have demonstrated strong capabilities in synthesizing novel views from limited inputs, their outputs often suffer from visual artifacts and lack 3D consistency, posing challenges for reliable mesh optimization. In this paper, we propose a novel framework that leverages diffusion models to enhance sparse-view mesh reconstruction in a principled and reliable manner. To address the instability of diffusion outputs, we propose a Consensus Diffusion Module that filters unreliable generations via interquartile range (IQR) analysis and performs variance-aware image fusion to produce robust pseudo-supervision. Building on this, we design an online reinforcement learning strategy based on the Upper Confidence Bound (UCB) to adaptively select the most informative viewpoints for enhancement, guided by diffusion loss. Finally, the fused images are used to jointly supervise a NeRF-based model alongside sparse-view ground truth, ensuring consistency across both geometry and appearance. Extensive experiments demonstrate that our method achieves significant improvements in both geometric quality and rendering quality.
☆ ADAT: Time-Series-Aware Adaptive Transformer Architecture for Sign Language Translation
Current sign language machine translation systems rely on recognizing hand movements, facial expressions and body postures, and natural language processing, to convert signs into text. Recent approaches use Transformer architectures to model long-range dependencies via positional encoding. However, they lack accuracy in recognizing fine-grained, short-range temporal dependencies between gestures captured at high frame rates. Moreover, their high computational complexity leads to inefficient training. To mitigate these issues, we propose an Adaptive Transformer (ADAT), which incorporates components for enhanced feature extraction and adaptive feature weighting through a gating mechanism to emphasize contextually relevant features while reducing training overhead and maintaining translation accuracy. To evaluate ADAT, we introduce MedASL, the first public medical American Sign Language dataset. In sign-to-gloss-to-text experiments, ADAT outperforms the encoder-decoder transformer, improving BLEU-4 accuracy by 0.1% while reducing training time by 14.33% on PHOENIX14T and 3.24% on MedASL. In sign-to-text experiments, it improves accuracy by 8.7% and reduces training time by 2.8% on PHOENIX14T and achieves 4.7% higher accuracy and 7.17% faster training on MedASL. Compared to encoder-only and decoder-only baselines in sign-to-text, ADAT is at least 6.8% more accurate despite being up to 12.1% slower due to its dual-stream structure.
☆ Beyond Words: Augmenting Discriminative Richness via Diffusions in Unsupervised Prompt Learning
Fine-tuning vision-language models (VLMs) with large amounts of unlabeled data has recently garnered significant interest. However, a key challenge remains the lack of high-quality pseudo-labeled data. Current pseudo-labeling strategies often struggle with mismatches between semantic and visual information, leading to sub-optimal performance of unsupervised prompt learning (UPL) methods. In this paper, we introduce a simple yet effective approach called \textbf{A}ugmenting D\textbf{i}scriminative \textbf{R}ichness via Diffusions (AiR), toward learning a richer discriminating way to represent the class comprehensively and thus facilitate classification. Specifically, our approach includes a pseudo-label generation module that leverages high-fidelity synthetic samples to create an auxiliary classifier, which captures richer visual variation, bridging text-image-pair classification to a more robust image-image-pair classification. Additionally, we exploit the diversity of diffusion-based synthetic samples to enhance prompt learning, providing greater information for semantic-visual alignment. Extensive experiments on five public benchmarks, including RESISC45 and Flowers102, and across three learning paradigms-UL, SSL, and TRZSL-demonstrate that AiR achieves substantial and consistent performance improvements over state-of-the-art unsupervised prompt learning methods.
☆ SemDiff: Generating Natural Unrestricted Adversarial Examples via Semantic Attributes Optimization in Diffusion Models
Unrestricted adversarial examples (UAEs), allow the attacker to create non-constrained adversarial examples without given clean samples, posing a severe threat to the safety of deep learning models. Recent works utilize diffusion models to generate UAEs. However, these UAEs often lack naturalness and imperceptibility due to simply optimizing in intermediate latent noises. In light of this, we propose SemDiff, a novel unrestricted adversarial attack that explores the semantic latent space of diffusion models for meaningful attributes, and devises a multi-attributes optimization approach to ensure attack success while maintaining the naturalness and imperceptibility of generated UAEs. We perform extensive experiments on four tasks on three high-resolution datasets, including CelebA-HQ, AFHQ and ImageNet. The results demonstrate that SemDiff outperforms state-of-the-art methods in terms of attack success rate and imperceptibility. The generated UAEs are natural and exhibit semantically meaningful changes, in accord with the attributes' weights. In addition, SemDiff is found capable of evading different defenses, which further validates its effectiveness and threatening.
☆ Zooming In on Fakes: A Novel Dataset for Localized AI-Generated Image Detection with Forgery Amplification Approach
The rise of AI-generated image editing tools has made localized forgeries increasingly realistic, posing challenges for visual content integrity. Although recent efforts have explored localized AIGC detection, existing datasets predominantly focus on object-level forgeries while overlooking broader scene edits in regions such as sky or ground. To address these limitations, we introduce \textbf{BR-Gen}, a large-scale dataset of 150,000 locally forged images with diverse scene-aware annotations, which are based on semantic calibration to ensure high-quality samples. BR-Gen is constructed through a fully automated Perception-Creation-Evaluation pipeline to ensure semantic coherence and visual realism. In addition, we further propose \textbf{NFA-ViT}, a Noise-guided Forgery Amplification Vision Transformer that enhances the detection of localized forgeries by amplifying forgery-related features across the entire image. NFA-ViT mines heterogeneous regions in images, \emph{i.e.}, potential edited areas, by noise fingerprints. Subsequently, attention mechanism is introduced to compel the interaction between normal and abnormal features, thereby propagating the generalization traces throughout the entire image, allowing subtle forgeries to influence a broader context and improving overall detection robustness. Extensive experiments demonstrate that BR-Gen constructs entirely new scenarios that are not covered by existing methods. Take a step further, NFA-ViT outperforms existing methods on BR-Gen and generalizes well across current benchmarks. All data and codes are available at https://github.com/clpbc/BR-Gen.
☆ AnomalyR1: A GRPO-based End-to-end MLLM for Industrial Anomaly Detection
Industrial Anomaly Detection (IAD) poses a formidable challenge due to the scarcity of defective samples, making it imperative to deploy models capable of robust generalization to detect unseen anomalies effectively. Traditional approaches, often constrained by hand-crafted features or domain-specific expert models, struggle to address this limitation, underscoring the need for a paradigm shift. We introduce AnomalyR1, a pioneering framework that leverages VLM-R1, a Multimodal Large Language Model (MLLM) renowned for its exceptional generalization and interpretability, to revolutionize IAD. By integrating MLLM with Group Relative Policy Optimization (GRPO), enhanced by our novel Reasoned Outcome Alignment Metric (ROAM), AnomalyR1 achieves a fully end-to-end solution that autonomously processes inputs of image and domain knowledge, reasons through analysis, and generates precise anomaly localizations and masks. Based on the latest multimodal IAD benchmark, our compact 3-billion-parameter model outperforms existing methods, establishing state-of-the-art results. As MLLM capabilities continue to advance, this study is the first to deliver an end-to-end VLM-based IAD solution that demonstrates the transformative potential of ROAM-enhanced GRPO, positioning our framework as a forward-looking cornerstone for next-generation intelligent anomaly detection systems in industrial applications with limited defective data.
☆ Learning Physics-Informed Color-Aware Transforms for Low-Light Image Enhancement ICME 2025
Image decomposition offers deep insights into the imaging factors of visual data and significantly enhances various advanced computer vision tasks. In this work, we introduce a novel approach to low-light image enhancement based on decomposed physics-informed priors. Existing methods that directly map low-light to normal-light images in the sRGB color space suffer from inconsistent color predictions and high sensitivity to spectral power distribution (SPD) variations, resulting in unstable performance under diverse lighting conditions. To address these challenges, we introduce a Physics-informed Color-aware Transform (PiCat), a learning-based framework that converts low-light images from the sRGB color space into deep illumination-invariant descriptors via our proposed Color-aware Transform (CAT). This transformation enables robust handling of complex lighting and SPD variations. Complementing this, we propose the Content-Noise Decomposition Network (CNDN), which refines the descriptor distributions to better align with well-lit conditions by mitigating noise and other distortions, thereby effectively restoring content representations to low-light images. The CAT and the CNDN collectively act as a physical prior, guiding the transformation process from low-light to normal-light domains. Our proposed PiCat framework demonstrates superior performance compared to state-of-the-art methods across five benchmark datasets.
comment: Accepted by ICME 2025
☆ Search is All You Need for Few-shot Anomaly Detection
Few-shot anomaly detection (FSAD) has emerged as a crucial yet challenging task in industrial inspection, where normal distribution modeling must be accomplished with only a few normal images. While existing approaches typically employ multi-modal foundation models combining language and vision modalities for prompt-guided anomaly detection, these methods often demand sophisticated prompt engineering and extensive manual tuning. In this paper, we demonstrate that a straightforward nearest-neighbor search framework can surpass state-of-the-art performance in both single-class and multi-class FSAD scenarios. Our proposed method, VisionAD, consists of four simple yet essential components: (1) scalable vision foundation models that extract universal and discriminative features; (2) dual augmentation strategies - support augmentation to enhance feature matching adaptability and query augmentation to address the oversights of single-view prediction; (3) multi-layer feature integration that captures both low-frequency global context and high-frequency local details with minimal computational overhead; and (4) a class-aware visual memory bank enabling efficient one-for-all multi-class detection. Extensive evaluations across MVTec-AD, VisA, and Real-IAD benchmarks demonstrate VisionAD's exceptional performance. Using only 1 normal images as support, our method achieves remarkable image-level AUROC scores of 97.4%, 94.8%, and 70.8% respectively, outperforming current state-of-the-art approaches by significant margins (+1.6%, +3.2%, and +1.4%). The training-free nature and superior few-shot capabilities of VisionAD make it particularly appealing for real-world applications where samples are scarce or expensive to obtain. Code is available at https://github.com/Qiqigeww/VisionAD.
☆ CAGS: Open-Vocabulary 3D Scene Understanding with Context-Aware Gaussian Splatting
Open-vocabulary 3D scene understanding is crucial for applications requiring natural language-driven spatial interpretation, such as robotics and augmented reality. While 3D Gaussian Splatting (3DGS) offers a powerful representation for scene reconstruction, integrating it with open-vocabulary frameworks reveals a key challenge: cross-view granularity inconsistency. This issue, stemming from 2D segmentation methods like SAM, results in inconsistent object segmentations across views (e.g., a "coffee set" segmented as a single entity in one view but as "cup + coffee + spoon" in another). Existing 3DGS-based methods often rely on isolated per-Gaussian feature learning, neglecting the spatial context needed for cohesive object reasoning, leading to fragmented representations. We propose Context-Aware Gaussian Splatting (CAGS), a novel framework that addresses this challenge by incorporating spatial context into 3DGS. CAGS constructs local graphs to propagate contextual features across Gaussians, reducing noise from inconsistent granularity, employs mask-centric contrastive learning to smooth SAM-derived features across views, and leverages a precomputation strategy to reduce computational cost by precomputing neighborhood relationships, enabling efficient training in large-scale scenes. By integrating spatial context, CAGS significantly improves 3D instance segmentation and reduces fragmentation errors on datasets like LERF-OVS and ScanNet, enabling robust language-guided 3D scene understanding.
☆ Learning Compatible Multi-Prize Subnetworks for Asymmetric Retrieval CVPR 2025
Asymmetric retrieval is a typical scenario in real-world retrieval systems, where compatible models of varying capacities are deployed on platforms with different resource configurations. Existing methods generally train pre-defined networks or subnetworks with capacities specifically designed for pre-determined platforms, using compatible learning. Nevertheless, these methods suffer from limited flexibility for multi-platform deployment. For example, when introducing a new platform into the retrieval systems, developers have to train an additional model at an appropriate capacity that is compatible with existing models via backward-compatible learning. In this paper, we propose a Prunable Network with self-compatibility, which allows developers to generate compatible subnetworks at any desired capacity through post-training pruning. Thus it allows the creation of a sparse subnetwork matching the resources of the new platform without additional training. Specifically, we optimize both the architecture and weight of subnetworks at different capacities within a dense network in compatible learning. We also design a conflict-aware gradient integration scheme to handle the gradient conflicts between the dense network and subnetworks during compatible learning. Extensive experiments on diverse benchmarks and visual backbones demonstrate the effectiveness of our method. Our code and model are available at https://github.com/Bunny-Black/PrunNet.
comment: Accepted to CVPR 2025
☆ A Category-Fragment Segmentation Framework for Pelvic Fracture Segmentation in X-ray Images
Pelvic fractures, often caused by high-impact trauma, frequently require surgical intervention. Imaging techniques such as CT and 2D X-ray imaging are used to transfer the surgical plan to the operating room through image registration, enabling quick intraoperative adjustments. Specifically, segmenting pelvic fractures from 2D X-ray imaging can assist in accurately positioning bone fragments and guiding the placement of screws or metal plates. In this study, we propose a novel deep learning-based category and fragment segmentation (CFS) framework for the automatic segmentation of pelvic bone fragments in 2D X-ray images. The framework consists of three consecutive steps: category segmentation, fragment segmentation, and post-processing. Our best model achieves an IoU of 0.91 for anatomical structures and 0.78 for fracture segmentation. Results demonstrate that the CFS framework is effective and accurate.
comment: 5 pages, 2 figures, 1 table
☆ Synthetic Data for Blood Vessel Network Extraction ICLR 2025
Blood vessel networks in the brain play a crucial role in stroke research, where understanding their topology is essential for analyzing blood flow dynamics. However, extracting detailed topological vessel network information from microscopy data remains a significant challenge, mainly due to the scarcity of labeled training data and the need for high topological accuracy. This work combines synthetic data generation with deep learning to automatically extract vessel networks as graphs from volumetric microscopy data. To combat data scarcity, we introduce a comprehensive pipeline for generating large-scale synthetic datasets that mirror the characteristics of real vessel networks. Our three-stage approach progresses from abstract graph generation through vessel mask creation to realistic medical image synthesis, incorporating biological constraints and imaging artifacts at each stage. Using this synthetic data, we develop a two-stage deep learning pipeline of 3D U-Net-based models for node detection and edge prediction. Fine-tuning on real microscopy data shows promising adaptation, improving edge prediction F1 scores from 0.496 to 0.626 by training on merely 5 manually labeled samples. These results suggest that automated vessel network extraction is becoming practically feasible, opening new possibilities for large-scale vascular analysis in stroke research.
comment: Presented at SynthData Workshop at ICLR 2025
☆ Cross-Frequency Collaborative Training Network and Dataset for Semi-supervised First Molar Root Canal Segmentation
Root canal (RC) treatment is a highly delicate and technically complex procedure in clinical practice, heavily influenced by the clinicians' experience and subjective judgment. Deep learning has made significant advancements in the field of computer-aided diagnosis (CAD) because it can provide more objective and accurate diagnostic results. However, its application in RC treatment is still relatively rare, mainly due to the lack of public datasets in this field. To address this issue, in this paper, we established a First Molar Root Canal segmentation dataset called FMRC-2025. Additionally, to alleviate the workload of manual annotation for dentists and fully leverage the unlabeled data, we designed a Cross-Frequency Collaborative training semi-supervised learning (SSL) Network called CFC-Net. It consists of two components: (1) Cross-Frequency Collaborative Mean Teacher (CFC-MT), which introduces two specialized students (SS) and one comprehensive teacher (CT) for collaborative multi-frequency training. The CT and SS are trained on different frequency components while fully integrating multi-frequency knowledge through cross and full frequency consistency supervisions. (2) Uncertainty-guided Cross-Frequency Mix (UCF-Mix) mechanism enables the network to generate high-confidence pseudo-labels while learning to integrate multi-frequency information and maintaining the structural integrity of the targets. Extensive experiments on FMRC-2025 and three public dental datasets demonstrate that CFC-MT is effective for RC segmentation and can also exhibit strong generalizability on other dental segmentation tasks, outperforming state-of-the-art SSL medical image segmentation methods. Codes and dataset will be released.
comment: 12 pages, Initial submission time 25 December 2024, Now Under Review
☆ ACE: Attentional Concept Erasure in Diffusion Models
Large text-to-image diffusion models have demonstrated remarkable image synthesis capabilities, but their indiscriminate training on Internet-scale data has led to learned concepts that enable harmful, copyrighted, or otherwise undesirable content generation. We address the task of concept erasure in diffusion models, i.e., removing a specified concept from a pre-trained model such that prompting the concept (or related synonyms) no longer yields its depiction, while preserving the model's ability to generate other content. We propose a novel method, Attentional Concept Erasure (ACE), that integrates a closed-form attention manipulation with lightweight fine-tuning. Theoretically, we formulate concept erasure as aligning the model's conditional distribution on the target concept with a neutral distribution. Our approach identifies and nullifies concept-specific latent directions in the cross-attention modules via a gated low-rank adaptation, followed by adversarially augmented fine-tuning to ensure thorough erasure of the concept and its synonyms. Empirically, we demonstrate on multiple benchmarks, including object classes, celebrity faces, explicit content, and artistic styles, that ACE achieves state-of-the-art concept removal efficacy and robustness. Compared to prior methods, ACE better balances generality (erasing concept and related terms) and specificity (preserving unrelated content), scales to dozens of concepts, and is efficient, requiring only a few seconds of adaptation per concept. We will release our code to facilitate safer deployment of diffusion models.
comment: Under Review
☆ Boosting Multi-View Stereo with Depth Foundation Model in the Absence of Real-World Labels
Learning-based Multi-View Stereo (MVS) methods have made remarkable progress in recent years. However, how to effectively train the network without using real-world labels remains a challenging problem. In this paper, driven by the recent advancements of vision foundation models, a novel method termed DFM-MVS, is proposed to leverage the depth foundation model to generate the effective depth prior, so as to boost MVS in the absence of real-world labels. Specifically, a depth prior-based pseudo-supervised training mechanism is developed to simulate realistic stereo correspondences using the generated depth prior, thereby constructing effective supervision for the MVS network. Besides, a depth prior-guided error correction strategy is presented to leverage the depth prior as guidance to mitigate the error propagation problem inherent in the widely-used coarse-to-fine network structure. Experimental results on DTU and Tanks & Temples datasets demonstrate that the proposed DFM-MVS significantly outperforms existing MVS methods without using real-world labels.
☆ A Visual RAG Pipeline for Few-Shot Fine-Grained Product Classification
Despite the rapid evolution of learning and computer vision algorithms, Fine-Grained Classification (FGC) still poses an open problem in many practically relevant applications. In the retail domain, for example, the identification of fast changing and visually highly similar products and their properties are key to automated price-monitoring and product recommendation. This paper presents a novel Visual RAG pipeline that combines the Retrieval Augmented Generation (RAG) approach and Vision Language Models (VLMs) for few-shot FGC. This Visual RAG pipeline extracts product and promotion data in advertisement leaflets from various retailers and simultaneously predicts fine-grained product ids along with price and discount information. Compared to previous approaches, the key characteristic of the Visual RAG pipeline is that it allows the prediction of novel products without re-training, simply by adding a few class samples to the RAG database. Comparing several VLM back-ends like GPT-4o [23], GPT-4o-mini [24], and Gemini 2.0 Flash [10], our approach achieves 86.8% accuracy on a diverse dataset.
☆ TextDiffSeg: Text-guided Latent Diffusion Model for 3d Medical Images Segmentation
Diffusion Probabilistic Models (DPMs) have demonstrated significant potential in 3D medical image segmentation tasks. However, their high computational cost and inability to fully capture global 3D contextual information limit their practical applications. To address these challenges, we propose a novel text-guided diffusion model framework, TextDiffSeg. This method leverages a conditional diffusion framework that integrates 3D volumetric data with natural language descriptions, enabling cross-modal embedding and establishing a shared semantic space between visual and textual modalities. By enhancing the model's ability to recognize complex anatomical structures, TextDiffSeg incorporates innovative label embedding techniques and cross-modal attention mechanisms, effectively reducing computational complexity while preserving global 3D contextual integrity. Experimental results demonstrate that TextDiffSeg consistently outperforms existing methods in segmentation tasks involving kidney and pancreas tumors, as well as multi-organ segmentation scenarios. Ablation studies further validate the effectiveness of key components, highlighting the synergistic interaction between text fusion, image feature extractor, and label encoder. TextDiffSeg provides an efficient and accurate solution for 3D medical image segmentation, showcasing its broad applicability in clinical diagnosis and treatment planning.
☆ Real-World Depth Recovery via Structure Uncertainty Modeling and Inaccurate GT Depth Fitting
The low-quality structure in raw depth maps is prevalent in real-world RGB-D datasets, which makes real-world depth recovery a critical task in recent years. However, the lack of paired raw-ground truth (raw-GT) data in the real world poses challenges for generalized depth recovery. Existing methods insufficiently consider the diversity of structure misalignment in raw depth maps, which leads to poor generalization in real-world depth recovery. Notably, random structure misalignments are not limited to raw depth data but also affect GT depth in real-world datasets. In the proposed method, we tackle the generalization problem from both input and output perspectives. For input, we enrich the diversity of structure misalignment in raw depth maps by designing a new raw depth generation pipeline, which helps the network avoid overfitting to a specific condition. Furthermore, a structure uncertainty module is designed to explicitly identify the misaligned structure for input raw depth maps to better generalize in unseen scenarios. Notably the well-trained depth foundation model (DFM) can help the structure uncertainty module estimate the structure uncertainty better. For output, a robust feature alignment module is designed to precisely align with the accurate structure of RGB images avoiding the interference of inaccurate GT depth. Extensive experiments on multiple datasets demonstrate the proposed method achieves competitive accuracy and generalization capabilities across various challenging raw depth maps.
☆ Neighbor-Based Feature and Index Enhancement for Person Re-Identification CVPR
Person re-identification (Re-ID) aims to match the same pedestrian in a large gallery with different cameras and views. Enhancing the robustness of the extracted feature representations is a main challenge in Re-ID. Existing methods usually improve feature representation by improving model architecture, but most methods ignore the potential contextual information, which limits the effectiveness of feature representation and retrieval performance. Neighborhood information, especially the potential information of multi-order neighborhoods, can effectively enrich feature expression and improve retrieval accuracy, but this has not been fully explored in existing research. Therefore, we propose a novel model DMON-ARO that leverages latent neighborhood information to enhance both feature representation and index performance. Our approach is built on two complementary modules: Dynamic Multi-Order Neighbor Modeling (DMON) and Asymmetric Relationship Optimization (ARO). The DMON module dynamically aggregates multi-order neighbor relationships, allowing it to capture richer contextual information and enhance feature representation through adaptive neighborhood modeling. Meanwhile, ARO refines the distance matrix by optimizing query-to-gallery relationships, improving the index accuracy. Extensive experiments on three benchmark datasets demonstrate that our approach achieves performance improvements against baseline models, which illustrate the effectiveness of our model. Specifically, our model demonstrates improvements in Rank-1 accuracy and mAP. Moreover, this method can also be directly extended to other re-identification tasks.
comment: Comment: This paper has been accepted for publication in the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)
☆ DART: Disease-aware Image-Text Alignment and Self-correcting Re-alignment for Trustworthy Radiology Report Generation CVPR
The automatic generation of radiology reports has emerged as a promising solution to reduce a time-consuming task and accurately capture critical disease-relevant findings in X-ray images. Previous approaches for radiology report generation have shown impressive performance. However, there remains significant potential to improve accuracy by ensuring that retrieved reports contain disease-relevant findings similar to those in the X-ray images and by refining generated reports. In this study, we propose a Disease-aware image-text Alignment and self-correcting Re-alignment for Trustworthy radiology report generation (DART) framework. In the first stage, we generate initial reports based on image-to-text retrieval with disease-matching, embedding both images and texts in a shared embedding space through contrastive learning. This approach ensures the retrieval of reports with similar disease-relevant findings that closely align with the input X-ray images. In the second stage, we further enhance the initial reports by introducing a self-correction module that re-aligns them with the X-ray images. Our proposed framework achieves state-of-the-art results on two widely used benchmarks, surpassing previous approaches in both report generation and clinical efficacy metrics, thereby enhancing the trustworthiness of radiology reports.
comment: The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2025
☆ ACMamba: Fast Unsupervised Anomaly Detection via An Asymmetrical Consensus State Space Model
Unsupervised anomaly detection in hyperspectral images (HSI), aiming to detect unknown targets from backgrounds, is challenging for earth surface monitoring. However, current studies are hindered by steep computational costs due to the high-dimensional property of HSI and dense sampling-based training paradigm, constraining their rapid deployment. Our key observation is that, during training, not all samples within the same homogeneous area are indispensable, whereas ingenious sampling can provide a powerful substitute for reducing costs. Motivated by this, we propose an Asymmetrical Consensus State Space Model (ACMamba) to significantly reduce computational costs without compromising accuracy. Specifically, we design an asymmetrical anomaly detection paradigm that utilizes region-level instances as an efficient alternative to dense pixel-level samples. In this paradigm, a low-cost Mamba-based module is introduced to discover global contextual attributes of regions that are essential for HSI reconstruction. Additionally, we develop a consensus learning strategy from the optimization perspective to simultaneously facilitate background reconstruction and anomaly compression, further alleviating the negative impact of anomaly reconstruction. Theoretical analysis and extensive experiments across eight benchmarks verify the superiority of ACMamba, demonstrating a faster speed and stronger performance over the state-of-the-art.
comment: 15 pages, 9 figures
☆ Multimodal Spatio-temporal Graph Learning for Alignment-free RGBT Video Object Detection
RGB-Thermal Video Object Detection (RGBT VOD) can address the limitation of traditional RGB-based VOD in challenging lighting conditions, making it more practical and effective in many applications. However, similar to most RGBT fusion tasks, it still mainly relies on manually aligned multimodal image pairs. In this paper, we propose a novel Multimodal Spatio-temporal Graph learning Network (MSGNet) for alignment-free RGBT VOD problem by leveraging the robust graph representation learning model. Specifically, we first design an Adaptive Partitioning Layer (APL) to estimate the corresponding regions of the Thermal image within the RGB image (high-resolution), achieving a preliminary inexact alignment. Then, we introduce the Spatial Sparse Graph Learning Module (S-SGLM) which employs a sparse information passing mechanism on the estimated inexact alignment to achieve reliable information interaction between different modalities. Moreover, to fully exploit the temporal cues for RGBT VOD problem, we introduce Hybrid Structured Temporal Modeling (HSTM), which involves a Temporal Sparse Graph Learning Module (T-SGLM) and Temporal Star Block (TSB). T-SGLM aims to filter out some redundant information between adjacent frames by employing the sparse aggregation mechanism on the temporal graph. Meanwhile, TSB is dedicated to achieving the complementary learning of local spatial relationships. Extensive comparative experiments conducted on both the aligned dataset VT-VOD50 and the unaligned dataset UVT-VOD2024 demonstrate the effectiveness and superiority of our proposed method. Our project will be made available on our website for free public access.
☆ Bridging the Semantic Gaps: Improving Medical VQA Consistency with LLM-Augmented Question Sets
Medical Visual Question Answering (MVQA) systems can interpret medical images in response to natural language queries. However, linguistic variability in question phrasing often undermines the consistency of these systems. To address this challenge, we propose a Semantically Equivalent Question Augmentation (SEQA) framework, which leverages large language models (LLMs) to generate diverse yet semantically equivalent rephrasings of questions. Specifically, this approach enriches linguistic diversity while preserving semantic meaning. We further introduce an evaluation metric, Total Agreement Rate with Semantically Equivalent Input and Correct Answer (TAR-SC), which assesses a model's capability to generate consistent and correct responses to semantically equivalent linguistic variations. In addition, we also propose three other diversity metrics - average number of QA items per image (ANQI), average number of questions per image with the same answer (ANQA), and average number of open-ended questions per image with the same semantics (ANQS). Using the SEQA framework, we augmented the benchmarked MVQA public datasets of SLAKE, VQA-RAD, and PathVQA. As a result, all three datasets achieved significant improvements by incorporating more semantically equivalent questions: ANQI increased by an average of 86.1, ANQA by 85.1, and ANQS by 46. Subsequent experiments evaluate three MVQA models (M2I2, MUMC, and BiomedGPT) under both zero-shot and fine-tuning settings on the enhanced datasets. Experimental results in MVQA datasets show that fine-tuned models achieve an average accuracy improvement of 19.35%, while our proposed TAR-SC metric shows an average improvement of 11. 61%, indicating a substantial enhancement in model consistency.
comment: The first two listed authors contributed equally to this work
☆ TacoDepth: Towards Efficient Radar-Camera Depth Estimation with One-stage Fusion CVPR 2025
Radar-Camera depth estimation aims to predict dense and accurate metric depth by fusing input images and Radar data. Model efficiency is crucial for this task in pursuit of real-time processing on autonomous vehicles and robotic platforms. However, due to the sparsity of Radar returns, the prevailing methods adopt multi-stage frameworks with intermediate quasi-dense depth, which are time-consuming and not robust. To address these challenges, we propose TacoDepth, an efficient and accurate Radar-Camera depth estimation model with one-stage fusion. Specifically, the graph-based Radar structure extractor and the pyramid-based Radar fusion module are designed to capture and integrate the graph structures of Radar point clouds, delivering superior model efficiency and robustness without relying on the intermediate depth results. Moreover, TacoDepth can be flexible for different inference modes, providing a better balance of speed and accuracy. Extensive experiments are conducted to demonstrate the efficacy of our method. Compared with the previous state-of-the-art approach, TacoDepth improves depth accuracy and processing speed by 12.8% and 91.8%. Our work provides a new perspective on efficient Radar-Camera depth estimation.
comment: Accepted by CVPR 2025 (Oral Presentation)
☆ Extended Short- and Long-Range Mesh Learning for Fast and Generalized Garment Simulation
3D garment simulation is a critical component for producing cloth-based graphics. Recent advancements in graph neural networks (GNNs) offer a promising approach for efficient garment simulation. However, GNNs require extensive message-passing to propagate information such as physical forces and maintain contact awareness across the entire garment mesh, which becomes computationally inefficient at higher resolutions. To address this, we devise a novel GNN-based mesh learning framework with two key components to extend the message-passing range with minimal overhead, namely the Laplacian-Smoothed Dual Message-Passing (LSDMP) and the Geodesic Self-Attention (GSA) modules. LSDMP enhances message-passing with a Laplacian features smoothing process, which efficiently propagates the impact of each vertex to nearby vertices. Concurrently, GSA introduces geodesic distance embeddings to represent the spatial relationship between vertices and utilises attention mechanisms to capture global mesh information. The two modules operate in parallel to ensure both short- and long-range mesh modelling. Extensive experiments demonstrate the state-of-the-art performance of our method, requiring fewer layers and lower inference latency.
☆ GrabS: Generative Embodied Agent for 3D Object Segmentation without Scene Supervision ICLR 2025
We study the hard problem of 3D object segmentation in complex point clouds without requiring human labels of 3D scenes for supervision. By relying on the similarity of pretrained 2D features or external signals such as motion to group 3D points as objects, existing unsupervised methods are usually limited to identifying simple objects like cars or their segmented objects are often inferior due to the lack of objectness in pretrained features. In this paper, we propose a new two-stage pipeline called GrabS. The core concept of our method is to learn generative and discriminative object-centric priors as a foundation from object datasets in the first stage, and then design an embodied agent to learn to discover multiple objects by querying against the pretrained generative priors in the second stage. We extensively evaluate our method on two real-world datasets and a newly created synthetic dataset, demonstrating remarkable segmentation performance, clearly surpassing all existing unsupervised methods.
comment: ICLR 2025 Spotlight. Code and data are available at: https://github.com/vLAR-group/GrabS
☆ SkeletonX: Data-Efficient Skeleton-based Action Recognition via Cross-sample Feature Aggregation
While current skeleton action recognition models demonstrate impressive performance on large-scale datasets, their adaptation to new application scenarios remains challenging. These challenges are particularly pronounced when facing new action categories, diverse performers, and varied skeleton layouts, leading to significant performance degeneration. Additionally, the high cost and difficulty of collecting skeleton data make large-scale data collection impractical. This paper studies one-shot and limited-scale learning settings to enable efficient adaptation with minimal data. Existing approaches often overlook the rich mutual information between labeled samples, resulting in sub-optimal performance in low-data scenarios. To boost the utility of labeled data, we identify the variability among performers and the commonality within each action as two key attributes. We present SkeletonX, a lightweight training pipeline that integrates seamlessly with existing GCN-based skeleton action recognizers, promoting effective training under limited labeled data. First, we propose a tailored sample pair construction strategy on two key attributes to form and aggregate sample pairs. Next, we develop a concise and effective feature aggregation module to process these pairs. Extensive experiments are conducted on NTU RGB+D, NTU RGB+D 120, and PKU-MMD with various GCN backbones, demonstrating that the pipeline effectively improves performance when trained from scratch with limited data. Moreover, it surpasses previous state-of-the-art methods in the one-shot setting, with only 1/10 of the parameters and much fewer FLOPs. The code and data are available at: https://github.com/zzysteve/SkeletonX
comment: Accepted by IEEE Transactions on Multimedia (TMM). 13 pages, 7 figures, 11 tables
☆ The Devil is in the Prompts: Retrieval-Augmented Prompt Optimization for Text-to-Video Generation CVPR2025
The evolution of Text-to-video (T2V) generative models, trained on large-scale datasets, has been marked by significant progress. However, the sensitivity of T2V generative models to input prompts highlights the critical role of prompt design in influencing generative outcomes. Prior research has predominantly relied on Large Language Models (LLMs) to align user-provided prompts with the distribution of training prompts, albeit without tailored guidance encompassing prompt vocabulary and sentence structure nuances. To this end, we introduce \textbf{RAPO}, a novel \textbf{R}etrieval-\textbf{A}ugmented \textbf{P}rompt \textbf{O}ptimization framework. In order to address potential inaccuracies and ambiguous details generated by LLM-generated prompts. RAPO refines the naive prompts through dual optimization branches, selecting the superior prompt for T2V generation. The first branch augments user prompts with diverse modifiers extracted from a learned relational graph, refining them to align with the format of training prompts via a fine-tuned LLM. Conversely, the second branch rewrites the naive prompt using a pre-trained LLM following a well-defined instruction set. Extensive experiments demonstrate that RAPO can effectively enhance both the static and dynamic dimensions of generated videos, demonstrating the significance of prompt optimization for user-provided prompts. Project website: \href{https://whynothaha.github.io/Prompt_optimizer/RAPO.html}{GitHub}.
comment: accepted by CVPR2025
☆ Recent Advance in 3D Object and Scene Generation: A Survey
In recent years, the demand for 3D content has grown exponentially with intelligent upgrading of interactive media, extended reality (XR), and Metaverse industries. In order to overcome the limitation of traditional manual modeling approaches, such as labor-intensive workflows and prolonged production cycles, revolutionary advances have been achieved through the convergence of novel 3D representation paradigms and artificial intelligence generative technologies. In this survey, we conduct a systematically review of the cutting-edge achievements in static 3D object and scene generation, as well as establish a comprehensive technical framework through systematic categorization. Specifically, we initiate our analysis with mainstream 3D object representations, followed by in-depth exploration of two principal technical pathways in object generation: data-driven supervised learning methods and deep generative model-based approaches. Regarding scene generation, we focus on three dominant paradigms: layout-guided compositional synthesis, 2D prior-based scene generation, and rule-driven modeling. Finally, we critically examine persistent challenges in 3D generation and propose potential research directions for future investigation. This survey aims to provide readers with a structured understanding of state-of-the-art 3D generation technologies while inspiring researchers to undertake more exploration in this domain.
comment: 34 pages, 6 figures
☆ DVLTA-VQA: Decoupled Vision-Language Modeling with Text-Guided Adaptation for Blind Video Quality Assessment
Inspired by the dual-stream theory of the human visual system (HVS) - where the ventral stream is responsible for object recognition and detail analysis, while the dorsal stream focuses on spatial relationships and motion perception - an increasing number of video quality assessment (VQA) works built upon this framework are proposed. Recent advancements in large multi-modal models, notably Contrastive Language-Image Pretraining (CLIP), have motivated researchers to incorporate CLIP into dual-stream-based VQA methods. This integration aims to harness the model's superior semantic understanding capabilities to replicate the object recognition and detail analysis in ventral stream, as well as spatial relationship analysis in dorsal stream. However, CLIP is originally designed for images and lacks the ability to capture temporal and motion information inherent in videos. %Furthermore, existing feature fusion strategies in no-reference video quality assessment (NR-VQA) often rely on fixed weighting schemes, which fail to adaptively adjust feature importance. To address the limitation, this paper propose a Decoupled Vision-Language Modeling with Text-Guided Adaptation for Blind Video Quality Assessment (DVLTA-VQA), which decouples CLIP's visual and textual components, and integrates them into different stages of the NR-VQA pipeline.
☆ EgoExo-Gen: Ego-centric Video Prediction by Watching Exo-centric Videos ICLR 2025
Generating videos in the first-person perspective has broad application prospects in the field of augmented reality and embodied intelligence. In this work, we explore the cross-view video prediction task, where given an exo-centric video, the first frame of the corresponding ego-centric video, and textual instructions, the goal is to generate futur frames of the ego-centric video. Inspired by the notion that hand-object interactions (HOI) in ego-centric videos represent the primary intentions and actions of the current actor, we present EgoExo-Gen that explicitly models the hand-object dynamics for cross-view video prediction. EgoExo-Gen consists of two stages. First, we design a cross-view HOI mask prediction model that anticipates the HOI masks in future ego-frames by modeling the spatio-temporal ego-exo correspondence. Next, we employ a video diffusion model to predict future ego-frames using the first ego-frame and textual instructions, while incorporating the HOI masks as structural guidance to enhance prediction quality. To facilitate training, we develop an automated pipeline to generate pseudo HOI masks for both ego- and exo-videos by exploiting vision foundation models. Extensive experiments demonstrate that our proposed EgoExo-Gen achieves better prediction performance compared to previous video prediction models on the Ego-Exo4D and H2O benchmark datasets, with the HOI masks significantly improving the generation of hands and interactive objects in the ego-centric videos.
comment: ICLR 2025
☆ Towards Safe Synthetic Image Generation On the Web: A Multimodal Robust NSFW Defense and Million Scale Dataset
In the past years, we have witnessed the remarkable success of Text-to-Image (T2I) models and their widespread use on the web. Extensive research in making T2I models produce hyper-realistic images has led to new concerns, such as generating Not-Safe-For-Work (NSFW) web content and polluting the web society. To help prevent misuse of T2I models and create a safer web environment for users features like NSFW filters and post-hoc security checks are used in these models. However, recent work unveiled how these methods can easily fail to prevent misuse. In particular, adversarial attacks on text and image modalities can easily outplay defensive measures. %Exploiting such leads to the growing concern of preventing adversarial attacks on text and image modalities. Moreover, there is currently no robust multimodal NSFW dataset that includes both prompt and image pairs and adversarial examples. This work proposes a million-scale prompt and image dataset generated using open-source diffusion models. Second, we develop a multimodal defense to distinguish safe and NSFW text and images, which is robust against adversarial attacks and directly alleviates current challenges. Our extensive experiments show that our model performs well against existing SOTA NSFW detection methods in terms of accuracy and recall, drastically reducing the Attack Success Rate (ASR) in multimodal adversarial attack scenarios. Code: https://github.com/shahidmuneer/multimodal-nsfw-defense.
comment: Short Paper The Web Conference
☆ Learning What NOT to Count
Few/zero-shot object counting methods reduce the need for extensive annotations but often struggle to distinguish between fine-grained categories, especially when multiple similar objects appear in the same scene. To address this limitation, we propose an annotation-free approach that enables the seamless integration of new fine-grained categories into existing few/zero-shot counting models. By leveraging latent generative models, we synthesize high-quality, category-specific crowded scenes, providing a rich training source for adapting to new categories without manual labeling. Our approach introduces an attention prediction network that identifies fine-grained category boundaries trained using only synthetic pseudo-annotated data. At inference, these fine-grained attention estimates refine the output of existing few/zero-shot counting networks. To benchmark our method, we further introduce the FGTC dataset, a taxonomy-specific fine-grained object counting dataset for natural images. Our method substantially enhances pre-trained state-of-the-art models on fine-grained taxon counting tasks, while using only synthetic data. Code and data to be released upon acceptance.
☆ Non-uniform Point Cloud Upsampling via Local Manifold Distribution
Existing learning-based point cloud upsampling methods often overlook the intrinsic data distribution charac?teristics of point clouds, leading to suboptimal results when handling sparse and non-uniform point clouds. We propose a novel approach to point cloud upsampling by imposing constraints from the perspective of manifold distributions. Leveraging the strong fitting capability of Gaussian functions, our method employs a network to iteratively optimize Gaussian components and their weights, accurately representing local manifolds. By utilizing the probabilistic distribution properties of Gaussian functions, we construct a unified statistical manifold to impose distribution constraints on the point cloud. Experimental results on multiple datasets demonstrate that our method generates higher-quality and more uniformly distributed dense point clouds when processing sparse and non-uniform inputs, outperforming state-of-the-art point cloud upsampling techniques.
☆ An Online Adaptation Method for Robust Depth Estimation and Visual Odometry in the Open World
Recently, learning-based robotic navigation systems have gained extensive research attention and made significant progress. However, the diversity of open-world scenarios poses a major challenge for the generalization of such systems to practical scenarios. Specifically, learned systems for scene measurement and state estimation tend to degrade when the application scenarios deviate from the training data, resulting to unreliable depth and pose estimation. Toward addressing this problem, this work aims to develop a visual odometry system that can fast adapt to diverse novel environments in an online manner. To this end, we construct a self-supervised online adaptation framework for monocular visual odometry aided by an online-updated depth estimation module. Firstly, we design a monocular depth estimation network with lightweight refiner modules, which enables efficient online adaptation. Then, we construct an objective for self-supervised learning of the depth estimation module based on the output of the visual odometry system and the contextual semantic information of the scene. Specifically, a sparse depth densification module and a dynamic consistency enhancement module are proposed to leverage camera poses and contextual semantics to generate pseudo-depths and valid masks for the online adaptation. Finally, we demonstrate the robustness and generalization capability of the proposed method in comparison with state-of-the-art learning-based approaches on urban, in-house datasets and a robot platform. Code is publicly available at: https://github.com/jixingwu/SOL-SLAM.
comment: 11 pages, 14 figures
☆ Interpreting the Linear Structure of Vision-language Model Embedding Spaces
Vision-language models encode images and text in a joint space, minimizing the distance between corresponding image and text pairs. How are language and images organized in this joint space, and how do the models encode meaning and modality? To investigate this, we train and release sparse autoencoders (SAEs) on the embedding spaces of four vision-language models (CLIP, SigLIP, SigLIP2, and AIMv2). SAEs approximate model embeddings as sparse linear combinations of learned directions, or "concepts". We find that, compared to other methods of linear feature learning, SAEs are better at reconstructing the real embeddings, while also able to retain the most sparsity. Retraining SAEs with different seeds or different data diet leads to two findings: the rare, specific concepts captured by the SAEs are liable to change drastically, but we also show that the key commonly-activating concepts extracted by SAEs are remarkably stable across runs. Interestingly, while most concepts are strongly unimodal in activation, we find they are not merely encoding modality per se. Many lie close to - but not entirely within - the subspace defining modality, suggesting that they encode cross-modal semantics despite their unimodal usage. To quantify this bridging behavior, we introduce the Bridge Score, a metric that identifies concept pairs which are both co-activated across aligned image-text inputs and geometrically aligned in the shared space. This reveals that even unimodal concepts can collaborate to support cross-modal integration. We release interactive demos of the SAEs for all models, allowing researchers to explore the organization of the concept spaces. Overall, our findings uncover a sparse linear structure within VLM embedding spaces that is shaped by modality, yet stitched together through latent bridges-offering new insight into how multimodal meaning is constructed.
♻ ☆ Taming Data and Transformers for Audio Generation
The scalability of ambient sound generators is hindered by data scarcity, insufficient caption quality, and limited scalability in model architecture. This work addresses these challenges by advancing both data and model scaling. First, we propose an efficient and scalable dataset collection pipeline tailored for ambient audio generation, resulting in AutoReCap-XL, the largest ambient audio-text dataset with over 47 million clips. To provide high-quality textual annotations, we propose AutoCap, a high-quality automatic audio captioning model. By adopting a Q-Former module and leveraging audio metadata, AutoCap substantially enhances caption quality, reaching a CIDEr score of $83.2$, a $3.2\%$ improvement over previous captioning models. Finally, we propose GenAu, a scalable transformer-based audio generation architecture that we scale up to 1.25B parameters. We demonstrate its benefits from data scaling with synthetic captions as well as model size scaling. When compared to baseline audio generators trained at similar size and data scale, GenAu obtains significant improvements of $4.7\%$ in FAD score, $11.1\%$ in IS, and $13.5\%$ in CLAP score. Our code, model checkpoints, and dataset are publicly available.
comment: Project Webpage: https://snap-research.github.io/GenAU/
♻ ☆ Earth-Adapter: Bridge the Geospatial Domain Gaps with Mixture of Frequency Adaptation
Parameter-Efficient Fine-Tuning (PEFT) is a technique that allows us to adapt powerful Foundation Models (FMs) to diverse downstream tasks while preserving and unleashing their inherent capabilities. However, we have observed that existing PEFT methods, which are often designed with natural imagery in mind, struggle when applied to Remote Sensing (RS) scenarios. This is primarily due to their inability to handle artifact influences, a problem particularly severe in RS image features. To tackle this challenge, we introduce Earth-Adapter, the first PEFT method specifically designed for RS artifacts conquering. Earth-Adapter introduces a novel Mixture of Frequency Adaptation process that combines a Mixture of Adapter (MoA) with Discrete Fourier Transformation (DFT). By utilizing DFT, Earth-Adapter can decompose features into different frequency components, precisely separating artifacts from original features. The MoA then dynamically assigns weights to each adapter expert, allowing for the combination of features across various frequency domains. These simple-yet-effective approaches enable Earth-Adapter to more efficiently overcome the disturbances caused by artifacts than previous PEFT methods, significantly enhancing the FMs' performance on RS scenarios. Experiments on Domain Adaptation (DA), and Domain Generalization (DG) semantic segmentation benchmarks showcase the Earth-Adapter's effectiveness. Compared with baseline Rein, Earth-Adapter significantly improves 9.0% mIoU in DA and 3.1% mIoU in DG benchmarks. Our code will be released at https://github.com/VisionXLab/Earth-Adapter.
♻ ☆ KMM: Key Frame Mask Mamba for Extended Motion Generation
Human motion generation is a cut-edge area of research in generative computer vision, with promising applications in video creation, game development, and robotic manipulation. The recent Mamba architecture shows promising results in efficiently modeling long and complex sequences, yet two significant challenges remain: Firstly, directly applying Mamba to extended motion generation is ineffective, as the limited capacity of the implicit memory leads to memory decay. Secondly, Mamba struggles with multimodal fusion compared to Transformers, and lack alignment with textual queries, often confusing directions (left or right) or omitting parts of longer text queries. To address these challenges, our paper presents three key contributions: Firstly, we introduce KMM, a novel architecture featuring Key frame Masking Modeling, designed to enhance Mamba's focus on key actions in motion segments. This approach addresses the memory decay problem and represents a pioneering method in customizing strategic frame-level masking in SSMs. Additionally, we designed a contrastive learning paradigm for addressing the multimodal fusion problem in Mamba and improving the motion-text alignment. Finally, we conducted extensive experiments on the go-to dataset, BABEL, achieving state-of-the-art performance with a reduction of more than 57% in FID and 70% parameters compared to previous state-of-the-art methods. See project website: https://steve-zeyu-zhang.github.io/KMM
♻ ☆ Seedream 3.0 Technical Report
We present Seedream 3.0, a high-performance Chinese-English bilingual image generation foundation model. We develop several technical improvements to address existing challenges in Seedream 2.0, including alignment with complicated prompts, fine-grained typography generation, suboptimal visual aesthetics and fidelity, and limited image resolutions. Specifically, the advancements of Seedream 3.0 stem from improvements across the entire pipeline, from data construction to model deployment. At the data stratum, we double the dataset using a defect-aware training paradigm and a dual-axis collaborative data-sampling framework. Furthermore, we adopt several effective techniques such as mixed-resolution training, cross-modality RoPE, representation alignment loss, and resolution-aware timestep sampling in the pre-training phase. During the post-training stage, we utilize diversified aesthetic captions in SFT, and a VLM-based reward model with scaling, thereby achieving outputs that well align with human preferences. Furthermore, Seedream 3.0 pioneers a novel acceleration paradigm. By employing consistent noise expectation and importance-aware timestep sampling, we achieve a 4 to 8 times speedup while maintaining image quality. Seedream 3.0 demonstrates significant improvements over Seedream 2.0: it enhances overall capabilities, in particular for text-rendering in complicated Chinese characters which is important to professional typography generation. In addition, it provides native high-resolution output (up to 2K), allowing it to generate images with high visual quality.
comment: Seedream 3.0 Technical Report
♻ ☆ Decentralized Vehicle Coordination: The Berkeley DeepDrive Drone Dataset and Consensus-Based Models
A significant portion of roads, particularly in densely populated developing countries, lacks explicitly defined right-of-way rules. These understructured roads pose substantial challenges for autonomous vehicle motion planning, where efficient and safe navigation relies on understanding decentralized human coordination for collision avoidance. This coordination, often termed "social driving etiquette," remains underexplored due to limited open-source empirical data and suitable modeling frameworks. In this paper, we present a novel dataset and modeling framework designed to study motion planning in these understructured environments. The dataset includes 20 aerial videos of representative scenarios, an image dataset for training vehicle detection models, and a development kit for vehicle trajectory estimation. We demonstrate that a consensus-based modeling approach can effectively explain the emergence of priority orders observed in our dataset, and is therefore a viable framework for decentralized collision avoidance planning.
comment: 7 pages, 7 figures, 1 table
♻ ☆ MMCLIP: Cross-modal Attention Masked Modelling for Medical Language-Image Pre-Training
Vision-and-language pretraining (VLP) in the medical field utilizes contrastive learning on image-text pairs to achieve effective transfer across tasks. Yet, current VLP approaches with the masked modeling strategy face two challenges when applied to the medical domain. First, current models struggle to accurately reconstruct key pathological features due to the scarcity of medical data. Second, most methods only adopt either paired image-text or image-only data, failing to exploit the combination of both paired and unpaired data. To this end, this paper proposes the MMCLIP (Masked Medical Contrastive Language-Image Pre-Training) framework to enhance pathological learning and feature learning via unpaired data. First, we introduce the attention-masked image modeling (AttMIM) and entity-driven masked language modeling module (EntMLM), which learns to reconstruct pathological visual and textual tokens via multi-modal feature interaction, thus improving medical-enhanced features. The AttMIM module masks a portion of the image features that are highly responsive to textual features. This allows MMCLIP to improve the reconstruction of highly similar image data in medicine efficiency. Second, our MMCLIP capitalizes unpaired data to enhance multimodal learning by introducing disease-kind prompts. The experimental results show that MMCLIP achieves SOTA for zero-shot and fine-tuning classification performance on five datasets. Our code will be available at https://github.com/AIGeeksGroup/MMCLIP.
♻ ☆ Self-Supervised Enhancement of Forward-Looking Sonar Images: Bridging Cross-Modal Degradation Gaps through Feature Space Transformation and Multi-Frame Fusion
Enhancing forward-looking sonar images is critical for accurate underwater target detection. Current deep learning methods mainly rely on supervised training with simulated data, but the difficulty in obtaining high-quality real-world paired data limits their practical use and generalization. Although self-supervised approaches from remote sensing partially alleviate data shortages, they neglect the cross-modal degradation gap between sonar and remote sensing images. Directly transferring pretrained weights often leads to overly smooth sonar images, detail loss, and insufficient brightness. To address this, we propose a feature-space transformation that maps sonar images from the pixel domain to a robust feature domain, effectively bridging the degradation gap. Additionally, our self-supervised multi-frame fusion strategy leverages complementary inter-frame information to naturally remove speckle noise and enhance target-region brightness. Experiments on three self-collected real-world forward-looking sonar datasets show that our method significantly outperforms existing approaches, effectively suppressing noise, preserving detailed edges, and substantially improving brightness, demonstrating strong potential for underwater target detection applications.
♻ ☆ A Semi-Self-Supervised Approach for Dense-Pattern Video Object Segmentation
Video object segmentation (VOS) -- predicting pixel-level regions for objects within each frame of a video -- is particularly challenging in agricultural scenarios, where videos of crops include hundreds of small, dense, and occluded objects (stems, leaves, flowers, pods) that sway and move unpredictably in the wind. Supervised training is the state-of-the-art for VOS, but it requires large, pixel-accurate, human-annotated videos, which are costly to produce for videos with many densely packed objects in each frame. To address these challenges, we proposed a semi-self-supervised spatiotemporal approach for dense-VOS (DVOS) using a diffusion-based method through multi-task (reconstruction and segmentation) learning. We train the model first with synthetic data that mimics the camera and object motion of real videos and then with pseudo-labeled videos. We evaluate our DVOS method for wheat head segmentation from a diverse set of videos (handheld, drone-captured, different field locations, and different growth stages -- spanning from Boot-stage to Wheat-mature and Harvest-ready). Despite using only a few manually annotated video frames, the proposed approach yielded a high-performing model, achieving a Dice score of 0.79 when tested on a drone-captured external test set. While our method was evaluated on wheat head segmentation, it can be extended to other crops and domains, such as crowd analysis or microscopic image analysis.
♻ ☆ SpiritSight Agent: Advanced GUI Agent with One Look CVPR 2025
Graphical User Interface (GUI) agents show amazing abilities in assisting human-computer interaction, automating human user's navigation on digital devices. An ideal GUI agent is expected to achieve high accuracy, low latency, and compatibility for different GUI platforms. Recent vision-based approaches have shown promise by leveraging advanced Vision Language Models (VLMs). While they generally meet the requirements of compatibility and low latency, these vision-based GUI agents tend to have low accuracy due to their limitations in element grounding. To address this issue, we propose $\textbf{SpiritSight}$, a vision-based, end-to-end GUI agent that excels in GUI navigation tasks across various GUI platforms. First, we create a multi-level, large-scale, high-quality GUI dataset called $\textbf{GUI-Lasagne}$ using scalable methods, empowering SpiritSight with robust GUI understanding and grounding capabilities. Second, we introduce the $\textbf{Universal Block Parsing (UBP)}$ method to resolve the ambiguity problem in dynamic high-resolution of visual inputs, further enhancing SpiritSight's ability to ground GUI objects. Through these efforts, SpiritSight agent outperforms other advanced methods on diverse GUI benchmarks, demonstrating its superior capability and compatibility in GUI navigation tasks. Models and datasets are available at https://hzhiyuan.github.io/SpiritSight-Agent.
comment: Paper accepted to CVPR 2025
♻ ☆ InfoNCE: Identifying the Gap Between Theory and Practice
Prior theory work on Contrastive Learning via the InfoNCE loss showed that, under certain assumptions, the learned representations recover the ground-truth latent factors. We argue that these theories overlook crucial aspects of how CL is deployed in practice. Specifically, they either assume equal variance across all latents or that certain latents are kept invariant. However, in practice, positive pairs are often generated using augmentations such as strong cropping to just a few pixels. Hence, a more realistic assumption is that all latent factors change with a continuum of variability across all factors. We introduce AnInfoNCE, a generalization of InfoNCE that can provably uncover the latent factors in this anisotropic setting, broadly generalizing previous identifiability results in CL. We validate our identifiability results in controlled experiments and show that AnInfoNCE increases the recovery of previously collapsed information in CIFAR10 and ImageNet, albeit at the cost of downstream accuracy. Finally, we discuss the remaining mismatches between theoretical assumptions and practical implementations.
♻ ☆ Deep Anatomical Federated Network (Dafne): An open client-server framework for the continuous, collaborative improvement of deep learning-based medical image segmentation
Purpose: To present and evaluate Dafne (deep anatomical federated network), a freely available decentralized, collaborative deep learning system for the semantic segmentation of radiological images through federated incremental learning. Materials and Methods: Dafne is free software with a client-server architecture. The client side is an advanced user interface that applies the deep learning models stored on the server to the user's data and allows the user to check and refine the prediction. Incremental learning is then performed at the client's side and sent back to the server, where it is integrated into the root model. Dafne was evaluated locally, by assessing the performance gain across model generations on 38 MRI datasets of the lower legs, and through the analysis of real-world usage statistics (n = 639 use-cases). Results: Dafne demonstrated a statistically improvement in the accuracy of semantic segmentation over time (average increase of the Dice Similarity Coefficient by 0.007 points/generation on the local validation set, p < 0.001). Qualitatively, the models showed enhanced performance on various radiologic image types, including those not present in the initial training sets, indicating good model generalizability. Conclusion: Dafne showed improvement in segmentation quality over time, demonstrating potential for learning and generalization.
comment: In this new version: accepted version in Radiology: Artificial Intelligence. Note regarding the license/copyright: This submission is conforming with the RSNA Preprint policy available here: https://pubs.rsna.org/page/ai/author-instructions, which REQUIRES authors to update the version on preprint servers with the accepted version and the copyright notice as indicated in the PDF
♻ ☆ Strategic Client Selection to Address Non-IIDness in HAPS-enabled FL Networks
The deployment of federated learning (FL) in non-terrestrial networks (NTN) that are supported by high-altitude platform stations (HAPS) offers numerous advantages. Due to its large footprint, it facilitates interaction with a large number of line-of-sight (LoS) ground clients, each possessing diverse datasets along with distinct communication and computational capabilities. The presence of many clients enhances the accuracy of the FL model and speeds up convergence. However, the variety of datasets among these clients poses a significant challenge, as it leads to pervasive non-independent and identically distributed (non-IID) data. The data non-IIDness results in markedly reduced training accuracy and slower convergence rates. To address this issue, we propose a novel weighted attribute-based client selection strategy that leverages multiple user-specific attributes, including historical traffic patterns, instantaneous channel conditions, computational capabilities, and previous-round learning performance. By combining these attributes into a composite score for each user at every FL round and selecting users with higher scores as FL clients, the framework ensures more uniform and representative data distributions, effectively mitigating the adverse effects of non-IID data. Simulation results corroborate the effectiveness of the proposed client selection strategy in enhancing FL model accuracy and convergence rate, as well as reducing training loss, by effectively addressing the critical challenge of data non-IIDness in large-scale FL system implementations.
comment: Submitted to IEEE for possible publication
♻ ☆ OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning
The advances in vision-language models (VLMs) have led to a growing interest in autonomous driving to leverage their strong reasoning capabilities. However, extending these capabilities from 2D to full 3D understanding is crucial for real-world applications. To address this challenge, we propose OmniDrive, a holistic vision-language dataset that aligns agent models with 3D driving tasks through counterfactual reasoning. This approach enhances decision-making by evaluating potential scenarios and their outcomes, similar to human drivers considering alternative actions. Our counterfactual-based synthetic data annotation process generates large-scale, high-quality datasets, providing denser supervision signals that bridge planning trajectories and language-based reasoning. Futher, we explore two advanced OmniDrive-Agent frameworks, namely Omni-L and Omni-Q, to assess the importance of vision-language alignment versus 3D perception, revealing critical insights into designing effective LLM-agents. Significant improvements on the DriveLM Q\&A benchmark and nuScenes open-loop planning demonstrate the effectiveness of our dataset and methods.
♻ ☆ Exploring Self-supervised Skeleton-based Action Recognition in Occluded Environments IJCNN 2025
To integrate action recognition into autonomous robotic systems, it is essential to address challenges such as person occlusions-a common yet often overlooked scenario in existing self-supervised skeleton-based action recognition methods. In this work, we propose IosPSTL, a simple and effective self-supervised learning framework designed to handle occlusions. IosPSTL combines a cluster-agnostic KNN imputer with an Occluded Partial Spatio-Temporal Learning (OPSTL) strategy. First, we pre-train the model on occluded skeleton sequences. Then, we introduce a cluster-agnostic KNN imputer that performs semantic grouping using k-means clustering on sequence embeddings. It imputes missing skeleton data by applying K-Nearest Neighbors in the latent space, leveraging nearby sample representations to restore occluded joints. This imputation generates more complete skeleton sequences, which significantly benefits downstream self-supervised models. To further enhance learning, the OPSTL module incorporates Adaptive Spatial Masking (ASM) to make better use of intact, high-quality skeleton sequences during training. Our method achieves state-of-the-art performance on the occluded versions of the NTU-60 and NTU-120 datasets, demonstrating its robustness and effectiveness under challenging conditions. Code is available at https://github.com/cyfml/OPSTL.
comment: Accepted to IJCNN 2025. Code is available at https://github.com/cyfml/OPSTL
♻ ☆ Cross-Modal Mapping: Mitigating the Modality Gap for Few-Shot Image Classification
Few-shot image classification remains a critical challenge in the field of computer vision, particularly in data-scarce environments. Existing methods typically rely on pre-trained visual-language models, such as CLIP. However, due to the modality gap, which is the inconsistent distribution of image and text features in the joint embedding space, directly using these features as class prototypes often leads to suboptimal performance. To address this issue, we propose a novel Cross-Modal Mapping (CMM) method. This method globally aligns image features with the text feature space through linear transformation and optimizes their local spatial relationships using triplet loss, thereby significantly enhancing cross-modal consistency. Experimental results show that compared to other methods, CMM simplifies the training process and demonstrates higher efficiency. Furthermore, CMM improves the average Top-1 accuracy by 1.06% on 11 benchmark datasets compared to methods that partially fine-tune the backbone, and it performs excellently on 4 distribution shift datasets. Notably, CMM effectively mitigates the modality gap in pre-trained models, enabling text features to serve as effective class prototypes for image features, thus providing an efficient and highly generalizable solution for few-shot learning.
♻ ☆ StructRe: Rewriting for Structured Shape Modeling
Man-made 3D shapes are naturally organized in parts and hierarchies; such structures provide important constraints for shape reconstruction and generation. Modeling shape structures is difficult, because there can be multiple hierarchies for a given shape, causing ambiguity, and across different categories the shape structures are correlated with semantics, limiting generalization. We present StructRe, a structure rewriting system, as a novel approach to structured shape modeling. Given a 3D object represented by points and components, StructRe can rewrite it upward into more concise structures, or downward into more detailed structures; by iterating the rewriting process, hierarchies are obtained. Such a localized rewriting process enables probabilistic modeling of ambiguous structures and robust generalization across object categories. We train StructRe on PartNet data and show its generalization to cross-category and multiple object hierarchies, and test its extension to ShapeNet. We also demonstrate the benefits of probabilistic and generalizable structure modeling for shape reconstruction, generation and editing tasks.
comment: Our project page: https://jiepengwang.github.io/StructRe/
♻ ☆ Document Parsing Unveiled: Techniques, Challenges, and Prospects for Structured Information Extraction
Document parsing is essential for converting unstructured and semi-structured documents such as contracts, academic papers, and invoices into structured, machine-readable data. Document parsing reliable structured data from unstructured inputs, providing huge convenience for numerous applications. Especially with recent achievements in Large Language Models, document parsing plays an indispensable role in both knowledge base construction and training data generation. This survey presents a comprehensive review of the current state of document parsing, covering key methodologies, from modular pipeline systems to end-to-end models driven by large vision-language models. Core components such as layout detection, content extraction (including text, tables, and mathematical expressions), and multi-modal data integration are examined in detail. Additionally, this paper discusses the challenges faced by modular document parsing systems and vision-language models in handling complex layouts, integrating multiple modules, and recognizing high-density text. It outlines future research directions and emphasizes the importance of developing larger and more diverse datasets.
♻ ☆ OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning
The advances in vision-language models (VLMs) have led to a growing interest in autonomous driving to leverage their strong reasoning capabilities. However, extending these capabilities from 2D to full 3D understanding is crucial for real-world applications. To address this challenge, we propose OmniDrive, a holistic vision-language dataset that aligns agent models with 3D driving tasks through counterfactual reasoning. This approach enhances decision-making by evaluating potential scenarios and their outcomes, similar to human drivers considering alternative actions. Our counterfactual-based synthetic data annotation process generates large-scale, high-quality datasets, providing denser supervision signals that bridge planning trajectories and language-based reasoning. Futher, we explore two advanced OmniDrive-Agent frameworks, namely Omni-L and Omni-Q, to assess the importance of vision-language alignment versus 3D perception, revealing critical insights into designing effective LLM-agents. Significant improvements on the DriveLM Q\&A benchmark and nuScenes open-loop planning demonstrate the effectiveness of our dataset and methods.
comment: Mistaken resubmission. The original version is at arXiv:2405.01533
♻ ☆ GST: Precise 3D Human Body from a Single Image with Gaussian Splatting Transformers CVPR 2025
Reconstructing posed 3D human models from monocular images has important applications in the sports industry, including performance tracking, injury prevention and virtual training. In this work, we combine 3D human pose and shape estimation with 3D Gaussian Splatting (3DGS), a representation of the scene composed of a mixture of Gaussians. This allows training or fine-tuning a human model predictor on multi-view images alone, without 3D ground truth. Predicting such mixtures for a human from a single input image is challenging due to self-occlusions and dependence on articulations, while also needing to retain enough flexibility to accommodate a variety of clothes and poses. Our key observation is that the vertices of standardized human meshes (such as SMPL) can provide an adequate spatial density and approximate initial position for the Gaussians. We can then train a transformer model to jointly predict comparatively small adjustments to these positions, as well as the other 3DGS attributes and the SMPL parameters. We show empirically that this combination (using only multi-view supervision) can achieve near real-time inference of 3D human models from a single image without expensive diffusion models or 3D points supervision, thus making it ideal for the sport industry at any level. More importantly, rendering is an effective auxiliary objective to refine 3D pose estimation by accounting for clothes and other geometric variations. The code is available at https://github.com/prosperolo/GST.
comment: Camera ready for CVSports workshop at CVPR 2025
♻ ☆ Collaborative Learning for Enhanced Unsupervised Domain Adaptation
Unsupervised Domain Adaptation (UDA) endeavors to bridge the gap between a model trained on a labeled source domain and its deployment in an unlabeled target domain. However, current high-performance models demand significant resources, making deployment costs prohibitive and highlighting the need for compact, yet effective models. For UDA of lightweight models, Knowledge Distillation (KD) leveraging a Teacher-Student framework could be a common approach, but we found that domain shift in UDA leads to a significant increase in non-salient parameters in the teacher model, degrading model's generalization ability and transferring misleading information to the student model. Interestingly, we observed that this phenomenon occurs considerably less in the student model. Driven by this insight, we introduce Collaborative Learning for UDA (CLDA), a method that updates the teacher's non-salient parameters using the student model and at the same time utilizes the updated teacher model to improve UDA performance of the student model. Experiments show consistent performance improvements for both student and teacher models. For example, in semantic segmentation, CLDA achieves an improvement of +0.7% mIoU for the teacher model and +1.4% mIoU for the student model compared to the baseline model in the GTA-to-Cityscapes datasets. In the Synthia-to-Cityscapes dataset, it achieves an improvement of +0.8% mIoU and +2.0% mIoU for the teacher and student models, respectively.
♻ ☆ Trajectory-guided Motion Perception for Facial Expression Quality Assessment in Neurological Disorders
Automated facial expression quality assessment (FEQA) in neurological disorders is critical for enhancing diagnostic accuracy and improving patient care, yet effectively capturing the subtle motions and nuances of facial muscle movements remains a challenge. We propose to analyse facial landmark trajectories, a compact yet informative representation, that encodes these subtle motions from a high-level structural perspective. Hence, we introduce Trajectory-guided Motion Perception Transformer (TraMP-Former), a novel FEQA framework that fuses landmark trajectory features for fine-grained motion capture with visual semantic cues from RGB frames, ultimately regressing the combined features into a quality score. Extensive experiments demonstrate that TraMP-Former achieves new state-of-the-art performance on benchmark datasets with neurological disorders, including PFED5 (up by 6.51%) and an augmented Toronto NeuroFace (up by 7.62%). Our ablation studies further validate the efficiency and effectiveness of landmark trajectories in FEQA. Our code is available at https://github.com/shuchaoduan/TraMP-Former.
comment: Accepted to IEEE FG 2025 (preprint)
♻ ☆ reBEN: Refined BigEarthNet Dataset for Remote Sensing Image Analysis
This paper presents refined BigEarthNet (reBEN) that is a large-scale, multi-modal remote sensing dataset constructed to support deep learning (DL) studies for remote sensing image analysis. The reBEN dataset consists of 549,488 pairs of Sentinel-1 and Sentinel-2 image patches. To construct reBEN, we initially consider the Sentinel-1 and Sentinel-2 tiles used to construct the BigEarthNet dataset and then divide them into patches of size 1200 m x 1200 m. We apply atmospheric correction to the Sentinel-2 patches using the latest version of the sen2cor tool, resulting in higher-quality patches compared to those present in BigEarthNet. Each patch is then associated with a pixel-level reference map and scene-level multi-labels. This makes reBEN suitable for pixel- and scene-based learning tasks. The labels are derived from the most recent CORINE Land Cover (CLC) map of 2018 by utilizing the 19-class nomenclature as in BigEarthNet. The use of the most recent CLC map results in overcoming the label noise present in BigEarthNet. Furthermore, we introduce a new geographical-based split assignment algorithm that significantly reduces the spatial correlation among the train, validation, and test sets with respect to those present in BigEarthNet. This increases the reliability of the evaluation of DL models. To minimize the DL model training time, we introduce software tools that convert the reBEN dataset into a DL-optimized data format. In our experiments, we show the potential of reBEN for multi-modal multi-label image classification problems by considering several state-of-the-art DL models. The pre-trained model weights, associated code, and complete dataset are available at https://bigearth.net.
comment: Accepted at IEEE International Geoscience and Remote Sensing Symposium (IGARSS) 2025. Our code is available at https://github.com/rsim-tu-berlin/bigearthnet-pipeline
♻ ☆ StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text
Text-to-video diffusion models enable the generation of high-quality videos that follow text instructions, making it easy to create diverse and individual content. However, existing approaches mostly focus on high-quality short video generation (typically 16 or 24 frames), ending up with hard-cuts when naively extended to the case of long video synthesis. To overcome these limitations, we introduce StreamingT2V, an autoregressive approach for long video generation of 80, 240, 600, 1200 or more frames with smooth transitions. The key components are:(i) a short-term memory block called conditional attention module (CAM), which conditions the current generation on the features extracted from the previous chunk via an attentional mechanism, leading to consistent chunk transitions, (ii) a long-term memory block called appearance preservation module, which extracts high-level scene and object features from the first video chunk to prevent the model from forgetting the initial scene, and (iii) a randomized blending approach that enables to apply a video enhancer autoregressively for infinitely long videos without inconsistencies between chunks. Experiments show that StreamingT2V generates high motion amount. In contrast, all competing image-to-video methods are prone to video stagnation when applied naively in an autoregressive manner. Thus, we propose with StreamingT2V a high-quality seamless text-to-long video generator that outperforms competitors with consistency and motion. Our code will be available at: https://github.com/Picsart-AI-Research/StreamingT2V
comment: https://github.com/Picsart-AI-Research/StreamingT2V
♻ ☆ WMNav: Integrating Vision-Language Models into World Models for Object Goal Navigation
Object Goal Navigation-requiring an agent to locate a specific object in an unseen environment-remains a core challenge in embodied AI. Although recent progress in Vision-Language Model (VLM)-based agents has demonstrated promising perception and decision-making abilities through prompting, none has yet established a fully modular world model design that reduces risky and costly interactions with the environment by predicting the future state of the world. We introduce WMNav, a novel World Model-based Navigation framework powered by Vision-Language Models (VLMs). It predicts possible outcomes of decisions and builds memories to provide feedback to the policy module. To retain the predicted state of the environment, WMNav proposes the online maintained Curiosity Value Map as part of the world model memory to provide dynamic configuration for navigation policy. By decomposing according to a human-like thinking process, WMNav effectively alleviates the impact of model hallucination by making decisions based on the feedback difference between the world model plan and observation. To further boost efficiency, we implement a two-stage action proposer strategy: broad exploration followed by precise localization. Extensive evaluation on HM3D and MP3D validates WMNav surpasses existing zero-shot benchmarks in both success rate and exploration efficiency (absolute improvement: +3.2% SR and +3.2% SPL on HM3D, +13.5% SR and +1.1% SPL on MP3D). Project page: https://b0b8k1ng.github.io/WMNav/.
comment: 8 pages, 5 figures
♻ ☆ ADAPT: An Autonomous Forklift for Construction Site Operation
Efficient material logistics play a critical role in controlling costs and schedules in the construction industry. However, manual material handling remains prone to inefficiencies, delays, and safety risks. Autonomous forklifts offer a promising solution to streamline on-site logistics, reducing reliance on human operators and mitigating labor shortages. This paper presents the development and evaluation of ADAPT (Autonomous Dynamic All-terrain Pallet Transporter), a fully autonomous off-road forklift designed for construction environments. Unlike structured warehouse settings, construction sites pose significant challenges, including dynamic obstacles, unstructured terrain, and varying weather conditions. To address these challenges, our system integrates AI-driven perception techniques with traditional approaches for decision making, planning, and control, enabling reliable operation in complex environments. We validate the system through extensive real-world testing, comparing its continuous performance against an experienced human operator across various weather conditions. Our findings demonstrate that autonomous outdoor forklifts can operate near human-level performance, offering a viable path toward safer and more efficient construction logistics.
♻ ☆ RSTeller: Scaling Up Visual Language Modeling in Remote Sensing with Rich Linguistic Semantics from Openly Available Data and Large Language Models SP
Abundant, well-annotated multimodal data in remote sensing are pivotal for aligning complex visual remote sensing (RS) scenes with human language, enabling the development of specialized vision language models across diverse RS interpretation tasks. However, annotating RS images with rich linguistic semantics at scale demands expertise in RS and substantial human labor, making it costly and often impractical. In this study, we propose a workflow that leverages large language models (LLMs) to generate multimodal datasets with semantically rich captions at scale from plain OpenStreetMap (OSM) data for images sourced from the Google Earth Engine (GEE) platform. This approach facilitates the generation of paired remote sensing data and can be readily scaled up using openly available data. Within this framework, we present RSTeller, a multimodal dataset comprising over 1.3 million RS images, each accompanied by two descriptive captions. Extensive experiments demonstrate that RSTeller enhances the performance of multiple existing vision language models for RS scene understanding through continual pre-training. Our methodology significantly reduces the manual effort and expertise needed for annotating remote sensing imagery while democratizing access to high-quality annotated data. This advancement fosters progress in visual language modeling and encourages broader participation in remote sensing research and applications. The RSTeller dataset is available at https://github.com/SlytherinGe/RSTeller.
comment: Submitted to ISPRS
♻ ☆ MotionStreamer: Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Space
This paper addresses the challenge of text-conditioned streaming motion generation, which requires us to predict the next-step human pose based on variable-length historical motions and incoming texts. Existing methods struggle to achieve streaming motion generation, e.g., diffusion models are constrained by pre-defined motion lengths, while GPT-based methods suffer from delayed response and error accumulation problem due to discretized non-causal tokenization. To solve these problems, we propose MotionStreamer, a novel framework that incorporates a continuous causal latent space into a probabilistic autoregressive model. The continuous latents mitigate information loss caused by discretization and effectively reduce error accumulation during long-term autoregressive generation. In addition, by establishing temporal causal dependencies between current and historical motion latents, our model fully utilizes the available information to achieve accurate online motion decoding. Experiments show that our method outperforms existing approaches while offering more applications, including multi-round generation, long-term generation, and dynamic motion composition. Project Page: https://zju3dv.github.io/MotionStreamer/
comment: Project Page: https://zju3dv.github.io/MotionStreamer/
♻ ☆ Proxy Denoising for Source-Free Domain Adaptation ICLR 2025
Source-Free Domain Adaptation (SFDA) aims to adapt a pre-trained source model to an unlabeled target domain with no access to the source data. Inspired by the success of large Vision-Language (ViL) models in many applications, the latest research has validated ViL's benefit for SFDA by using their predictions as pseudo supervision. However, we observe that ViL's supervision could be noisy and inaccurate at an unknown rate, introducing additional negative effects during adaption. To address this thus-far ignored challenge, we introduce a novel Proxy Denoising (ProDe) approach. The key idea is to leverage the ViL model as a proxy to facilitate the adaptation process towards the latent domain-invariant space. We design a proxy denoising mechanism to correct ViL's predictions, grounded on a proxy confidence theory that models the dynamic effect of proxy's divergence against the domain-invariant space during adaptation. To capitalize on the corrected proxy, we derive a mutual knowledge distilling regularization. Extensive experiments show that ProDe significantly outperforms current state-of-the-art alternatives under the conventional closed set setting and more challenging open set, partial set, generalized SFDA, multi-target, multi-source, and test-time settings. Our code and data are available at https://github.com/tntek/source-free-domain-adaptation.
comment: This paper is accepted by ICLR 2025 (Oral, Top 1.8%)
♻ ☆ SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis
Text-based generation and editing of 3D scenes hold significant potential for streamlining content creation through intuitive user interactions. While recent advances leverage 3D Gaussian Splatting (3DGS) for high-fidelity and real-time rendering, existing methods are often specialized and task-focused, lacking a unified framework for both generation and editing. In this paper, we introduce SplatFlow, a comprehensive framework that addresses this gap by enabling direct 3DGS generation and editing. SplatFlow comprises two main components: a multi-view rectified flow (RF) model and a Gaussian Splatting Decoder (GSDecoder). The multi-view RF model operates in latent space, generating multi-view images, depths, and camera poses simultaneously, conditioned on text prompts, thus addressing challenges like diverse scene scales and complex camera trajectories in real-world settings. Then, the GSDecoder efficiently translates these latent outputs into 3DGS representations through a feed-forward 3DGS method. Leveraging training-free inversion and inpainting techniques, SplatFlow enables seamless 3DGS editing and supports a broad range of 3D tasks-including object editing, novel view synthesis, and camera pose estimation-within a unified framework without requiring additional complex pipelines. We validate SplatFlow's capabilities on the MVImgNet and DL3DV-7K datasets, demonstrating its versatility and effectiveness in various 3D generation, editing, and inpainting-based tasks.
comment: Project Page: https://gohyojun15.github.io/SplatFlow/
♻ ☆ Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation
With the advent of large multimodal language models, science is now at a threshold of an AI-based technological transformation. Recently, a plethora of new AI models and tools has been proposed, promising to empower researchers and academics worldwide to conduct their research more effectively and efficiently. This includes all aspects of the research cycle, especially (1) searching for relevant literature; (2) generating research ideas and conducting experimentation; generating (3) text-based and (4) multimodal content (e.g., scientific figures and diagrams); and (5) AI-based automatic peer review. In this survey, we provide an in-depth overview over these exciting recent developments, which promise to fundamentally alter the scientific research process for good. Our survey covers the five aspects outlined above, indicating relevant datasets, methods and results (including evaluation) as well as limitations and scope for future research. Ethical concerns regarding shortcomings of these tools and potential for misuse (fake science, plagiarism, harms to research integrity) take a particularly prominent place in our discussion. We hope that our survey will not only become a reference guide for newcomers to the field but also a catalyst for new AI-based initiatives in the area of "AI4Science".
comment: 44 pages, 7 figures, 8 tables
♻ ☆ Two Effects, One Trigger: On the Modality Gap, Object Bias, and Information Imbalance in Contrastive Vision-Language Models ICLR 2025
Contrastive vision-language models (VLMs), like CLIP, have gained popularity for their versatile applicability to various downstream tasks. Despite their successes in some tasks, like zero-shot object recognition, they perform surprisingly poor on other tasks, like attribute recognition. Previous work has attributed these challenges to the modality gap, a separation of image and text in the shared representation space, and to a bias towards objects over other factors, such as attributes. In this analysis paper, we investigate both phenomena thoroughly. We evaluated off-the-shelf VLMs and while the gap's influence on performance is typically overshadowed by other factors, we find indications that closing the gap indeed leads to improvements. Moreover, we find that, contrary to intuition, only few embedding dimensions drive the gap and that the embedding spaces are differently organized. To allow for a clean study of object bias, we introduce a definition and a corresponding measure of it. Equipped with this tool, we find that object bias does not lead to worse performance on other concepts, such as attributes per se. However, why do both phenomena, modality gap and object bias, emerge in the first place? To answer this fundamental question and uncover some of the inner workings of contrastive VLMs, we conducted experiments that allowed us to control the amount of shared information between the modalities. These experiments revealed that the driving factor behind both the modality gap and the object bias, is an information imbalance between images and captions, and unveiled an intriguing connection between the modality gap and entropy of the logits.
comment: ICLR 2025 (Oral)
♻ ☆ SAM2MOT: A Novel Paradigm of Multi-Object Tracking by Segmentation
Segment Anything 2 (SAM2) enables robust single-object tracking using segmentation. To extend this to multi-object tracking (MOT), we propose SAM2MOT, introducing a novel Tracking by Segmentation paradigm. Unlike Tracking by Detection or Tracking by Query, SAM2MOT directly generates tracking boxes from segmentation masks, reducing reliance on detection accuracy. SAM2MOT has two key advantages: zero-shot generalization, allowing it to work across datasets without fine-tuning, and strong object association, inherited from SAM2. To further improve performance, we integrate a trajectory manager system for precise object addition and removal, and a cross-object interaction module to handle occlusions. Experiments on DanceTrack, UAVDT, and BDD100K show state-of-the-art results. Notably, SAM2MOT outperforms existing methods on DanceTrack by +2.1 HOTA and +4.5 IDF1, highlighting its effectiveness in MOT. Code is available at https://github.com/TripleJoy/SAM2MOT.
♻ ☆ Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization AAAI 2025
Deep neural networks (DNNs) are susceptible to universal adversarial perturbations (UAPs). These perturbations are meticulously designed to fool the target model universally across all sample classes. Unlike instance-specific adversarial examples (AEs), generating UAPs is more complex because they must be generalized across a wide range of data samples and models. Our research reveals that existing universal attack methods, which optimize UAPs using DNNs with static model parameter snapshots, do not fully leverage the potential of DNNs to generate more effective UAPs. Rather than optimizing UAPs against static DNN models with a fixed training set, we suggest using dynamic model-data pairs to generate UAPs. In particular, we introduce a dynamic maximin optimization strategy, aiming to optimize the UAP across a variety of optimal model-data pairs. We term this approach DM-UAP. DM-UAP utilizes an iterative max-min-min optimization framework that refines the model-data pairs, coupled with a curriculum UAP learning algorithm to examine the combined space of model parameters and data thoroughly. Comprehensive experiments on the ImageNet dataset demonstrate that the proposed DM-UAP markedly enhances both cross-sample universality and cross-model transferability of UAPs. Using only 500 samples for UAP generation, DM-UAP outperforms the state-of-the-art approach with an average increase in fooling ratio of 12.108%.
comment: Accepted in AAAI 2025
♻ ☆ Orthus: Autoregressive Interleaved Image-Text Generation with Modality-Specific Heads
We introduce Orthus, an autoregressive (AR) transformer that excels in generating images given textual prompts, answering questions based on visual inputs, and even crafting lengthy image-text interleaved contents. Unlike prior arts on unified multimodal modeling, Orthus simultaneously copes with discrete text tokens and continuous image features under the AR modeling principle. The continuous treatment of visual signals minimizes the information loss for both image understanding and generation while the fully AR formulation renders the characterization of the correlation between modalities straightforward. The key mechanism enabling Orthus to leverage these advantages lies in its modality-specific heads -- one regular language modeling (LM) head predicts discrete text tokens and one diffusion head generates continuous image features conditioning on the output of the backbone. We devise an efficient strategy for building Orthus -- by substituting the Vector Quantization (VQ) operation in the existing unified AR model with a soft alternative, introducing a diffusion head, and tuning the added modules to reconstruct images, we can create an Orthus-base model effortlessly (e.g., within mere 72 A100 GPU hours). Orthus-base can further embrace post-training to better model interleaved images and texts. Empirically, Orthus surpasses competing baselines including Show-o and Chameleon across standard benchmarks, achieving a GenEval score of 0.58 and an MME-P score of 1265.8 using 7B parameters. Orthus also shows exceptional mixed-modality generation capabilities, reflecting the potential for handling intricate practical generation tasks.
♻ ☆ Mitigating Long-tail Distribution in Oracle Bone Inscriptions: Dataset, Model, and Benchmark
The oracle bone inscription (OBI) recognition plays a significant role in understanding the history and culture of ancient China. However, the existing OBI datasets suffer from a long-tail distribution problem, leading to biased performance of OBI recognition models across majority and minority classes. With recent advancements in generative models, OBI synthesis-based data augmentation has become a promising avenue to expand the sample size of minority classes. Unfortunately, current OBI datasets lack large-scale structure-aligned image pairs for generative model training. To address these problems, we first present the Oracle-P15K, a structure-aligned OBI dataset for OBI generation and denoising, consisting of 14,542 images infused with domain knowledge from OBI experts. Second, we propose a diffusion model-based pseudo OBI generator, called OBIDiff, to achieve realistic and controllable OBI generation. Given a clean glyph image and a target rubbing-style image, it can effectively transfer the noise style of the original rubbing to the glyph image. Extensive experiments on OBI downstream tasks and user preference studies show the effectiveness of the proposed Oracle-P15K dataset and demonstrate that OBIDiff can accurately preserve inherent glyph structures while transferring authentic rubbing styles effectively.
♻ ☆ Enhancing Contrastive Learning Inspired by the Philosophy of "The Blind Men and the Elephant" AAAI 2025
Contrastive learning is a prevalent technique in self-supervised vision representation learning, typically generating positive pairs by applying two data augmentations to the same image. Designing effective data augmentation strategies is crucial for the success of contrastive learning. Inspired by the story of the blind men and the elephant, we introduce JointCrop and JointBlur. These methods generate more challenging positive pairs by leveraging the joint distribution of the two augmentation parameters, thereby enabling contrastive learning to acquire more effective feature representations. To the best of our knowledge, this is the first effort to explicitly incorporate the joint distribution of two data augmentation parameters into contrastive learning. As a plug-and-play framework without additional computational overhead, JointCrop and JointBlur enhance the performance of SimCLR, BYOL, MoCo v1, MoCo v2, MoCo v3, SimSiam, and Dino baselines with notable improvements.
comment: Accepted by AAAI 2025
♻ ☆ Discrete Distribution Networks ICLR 2025
We introduce a novel generative model, the Discrete Distribution Networks (DDN), that approximates data distribution using hierarchical discrete distributions. We posit that since the features within a network inherently capture distributional information, enabling the network to generate multiple samples simultaneously, rather than a single output, may offer an effective way to represent distributions. Therefore, DDN fits the target distribution, including continuous ones, by generating multiple discrete sample points. To capture finer details of the target data, DDN selects the output that is closest to the Ground Truth (GT) from the coarse results generated in the first layer. This selected output is then fed back into the network as a condition for the second layer, thereby generating new outputs more similar to the GT. As the number of DDN layers increases, the representational space of the outputs expands exponentially, and the generated samples become increasingly similar to the GT. This hierarchical output pattern of discrete distributions endows DDN with unique properties: more general zero-shot conditional generation and 1D latent representation. We demonstrate the efficacy of DDN and its intriguing properties through experiments on CIFAR-10 and FFHQ. The code is available at https://discrete-distribution-networks.github.io/
comment: Published as a conference paper at ICLR 2025
♻ ☆ Multi-modal vision-language model for generalizable annotation-free pathology localization and clinical diagnosis
Defining pathologies automatically from medical images aids the understanding of the emergence and progression of diseases, and such an ability is crucial in clinical diagnostics. However, existing deep learning models heavily rely on expert annotations and lack generalization capabilities in open clinical environments. In this study, we present a generalizable vision-language model for Annotation-Free pathology Localization (AFLoc). The core strength of AFLoc lies in its extensive multi-level semantic structure-based contrastive learning, which comprehensively aligns multi-granularity medical concepts from reports with abundant image features, to adapt to the diverse expressions of pathologies and unseen pathologies without the reliance on image annotations from experts. We conducted primary experiments on a dataset of 220K pairs of image-report chest X-ray images, and performed extensive validation across six external datasets encompassing 20 types of chest pathologies. The results demonstrate that AFLoc outperforms state-of-the-art methods in both annotation-free localization and classification tasks. Additionally, we assessed the generalizability of AFLoc on other modalities, including histopathology and retinal fundus images. Extensive experiments show that AFLoc exhibits robust generalization capabilities, even surpassing human benchmarks in localizing five different types of pathological images. These results highlight the potential of AFLoc in reducing annotation requirements and its applicability in complex clinical environments.
♻ ☆ Spatio-Temporal Fuzzy-oriented Multi-Modal Meta-Learning for Fine-grained Emotion Recognition
Fine-grained emotion recognition (FER) plays a vital role in various fields, such as disease diagnosis, personalized recommendations, and multimedia mining. However, existing FER methods face three key challenges in real-world applications: (i) they rely on large amounts of continuously annotated data to ensure accuracy since emotions are complex and ambiguous in reality, which is costly and time-consuming; (ii) they cannot capture the temporal heterogeneity caused by changing emotion patterns, because they usually assume that the temporal correlation within sampling periods is the same; (iii) they do not consider the spatial heterogeneity of different FER scenarios, that is, the distribution of emotion information in different data may have bias or interference. To address these challenges, we propose a Spatio-Temporal Fuzzy-oriented Multi-modal Meta-learning framework (ST-F2M). Specifically, ST-F2M first divides the multi-modal videos into multiple views, and each view corresponds to one modality of one emotion. Multiple randomly selected views for the same emotion form a meta-training task. Next, ST-F2M uses an integrated module with spatial and temporal convolutions to encode the data of each task, reflecting the spatial and temporal heterogeneity. Then it adds fuzzy semantic information to each task based on generalized fuzzy rules, which helps handle the complexity and ambiguity of emotions. Finally, ST-F2M learns emotion-related general meta-knowledge through meta-recurrent neural networks to achieve fast and robust fine-grained emotion recognition. Extensive experiments show that ST-F2M outperforms various state-of-the-art methods in terms of accuracy and model efficiency. In addition, we construct ablation studies and further analysis to explore why ST-F2M performs well.
comment: This work has been submitted to the IEEE for possible publication
♻ ☆ OpenSDI: Spotting Diffusion-Generated Images in the Open World
This paper identifies OpenSDI, a challenge for spotting diffusion-generated images in open-world settings. In response to this challenge, we define a new benchmark, the OpenSDI dataset (OpenSDID), which stands out from existing datasets due to its diverse use of large vision-language models that simulate open-world diffusion-based manipulations. Another outstanding feature of OpenSDID is its inclusion of both detection and localization tasks for images manipulated globally and locally by diffusion models. To address the OpenSDI challenge, we propose a Synergizing Pretrained Models (SPM) scheme to build up a mixture of foundation models. This approach exploits a collaboration mechanism with multiple pretrained foundation models to enhance generalization in the OpenSDI context, moving beyond traditional training by synergizing multiple pretrained models through prompting and attending strategies. Building on this scheme, we introduce MaskCLIP, an SPM-based model that aligns Contrastive Language-Image Pre-Training (CLIP) with Masked Autoencoder (MAE). Extensive evaluations on OpenSDID show that MaskCLIP significantly outperforms current state-of-the-art methods for the OpenSDI challenge, achieving remarkable relative improvements of 14.23% in IoU (14.11% in F1) and 2.05% in accuracy (2.38% in F1) compared to the second-best model in localization and detection tasks, respectively. Our dataset and code are available at https://github.com/iamwangyabin/OpenSDI.
♻ ☆ SceneFactory: A Workflow-centric and Unified Framework for Incremental Scene Modeling
We present SceneFactory, a workflow-centric and unified framework for incremental scene modeling, that conveniently supports a wide range of applications, such as (unposed and/or uncalibrated) multi-view depth estimation, LiDAR completion, (dense) RGB-D/RGB-L/Mono/Depth-only reconstruction and SLAM. The workflow-centric design uses multiple blocks as the basis for constructing different production lines. The supported applications, i.e., productions avoid redundancy in their designs. Thus, the focus is placed on each block itself for independent expansion. To support all input combinations, our implementation consists of four building blocks that form SceneFactory: (1) tracking, (2) flexion, (3) depth estimation, and (4) scene reconstruction. The tracking block is based on Mono SLAM and is extended to support RGB-D and RGB-LiDAR (RGB-L) inputs. Flexion is used to convert the depth image (untrackable) into a trackable image. For general-purpose depth estimation, we propose an unposed \& uncalibrated multi-view depth estimation model (U$^2$-MVD) to estimate dense geometry. U$^2$-MVD exploits dense bundle adjustment to solve for poses, intrinsics, and inverse depth. A semantic-aware ScaleCov step is then introduced to complete the multi-view depth. Relying on U$^2$-MVD, SceneFactory both supports user-friendly 3D creation (with just images) and bridges the applications of Dense RGB-D and Dense Mono. For high-quality surface and color reconstruction, we propose Dual-purpose Multi-resolutional Neural Points (DM-NPs) for the first surface accessible Surface Color Field design, where we introduce Improved Point Rasterization (IPR) for point cloud based surface query. ...
comment: Accepted to IEEE Transactions on Robotics (T-RO). For project page and code, please find https://jarrome.github.io/SceneFactory/
♻ ☆ GaussVideoDreamer: 3D Scene Generation with Video Diffusion and Inconsistency-Aware Gaussian Splatting
Single-image 3D scene reconstruction presents significant challenges due to its inherently ill-posed nature and limited input constraints. Recent advances have explored two promising directions: multiview generative models that train on 3D consistent datasets but struggle with out-of-distribution generalization, and 3D scene inpainting and completion frameworks that suffer from cross-view inconsistency and suboptimal error handling, as they depend exclusively on depth data or 3D smoothness, which ultimately degrades output quality and computational performance. Building upon these approaches, we present GaussVideoDreamer, which advances generative multimedia approaches by bridging the gap between image, video, and 3D generation, integrating their strengths through two key innovations: (1) A progressive video inpainting strategy that harnesses temporal coherence for improved multiview consistency and faster convergence. (2) A 3D Gaussian Splatting consistency mask to guide the video diffusion with 3D consistent multiview evidence. Our pipeline combines three core components: a geometry-aware initialization protocol, Inconsistency-Aware Gaussian Splatting, and a progressive video inpainting strategy. Experimental results demonstrate that our approach achieves 32% higher LLaVA-IQA scores and at least 2x speedup compared to existing methods while maintaining robust performance across diverse scenes.
♻ ☆ VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow CVPR 2025
Scene flow estimation aims to recover per-point motion from two adjacent LiDAR scans. However, in real-world applications such as autonomous driving, points rarely move independently of others, especially for nearby points belonging to the same object, which often share the same motion. Incorporating this locally rigid motion constraint has been a key challenge in self-supervised scene flow estimation, which is often addressed by post-processing or appending extra regularization. While these approaches are able to improve the rigidity of predicted flows, they lack an architectural inductive bias for local rigidity within the model structure, leading to suboptimal learning efficiency and inferior performance. In contrast, we enforce local rigidity with a lightweight add-on module in neural network design, enabling end-to-end learning. We design a discretized voting space that accommodates all possible translations and then identify the one shared by nearby points by differentiable voting. Additionally, to ensure computational efficiency, we operate on pillars rather than points and learn representative features for voting per pillar. We plug the Voting Module into popular model designs and evaluate its benefit on Argoverse 2 and Waymo datasets. We outperform baseline works with only marginal compute overhead. Code is available at https://github.com/tudelft-iv/VoteFlow.
comment: CVPR 2025. Code is available at https://github.com/tudelft-iv/VoteFlow. Yancong Lin and Shiming Wang have equal contributions
♻ ☆ Masked Autoencoders are Robust Data Augmentors
Deep neural networks are capable of learning powerful representations to tackle complex vision tasks but expose undesirable properties like the over-fitting issue. To this end, regularization techniques like image augmentation are necessary for deep neural networks to generalize well. Nevertheless, most prevalent image augmentation recipes confine themselves to off-the-shelf linear transformations like scale, flip, and colorjitter. Due to their hand-crafted property, these augmentations are insufficient to generate truly hard augmented examples. In this paper, we propose a novel perspective of augmentation to regularize the training process. Inspired by the recent success of applying masked image modeling to self-supervised learning, we adopt the self-supervised masked autoencoder to generate the distorted view of the input images. We show that utilizing such model-based nonlinear transformation as data augmentation can improve high-level recognition tasks. We term the proposed method as \textbf{M}ask-\textbf{R}econstruct \textbf{A}ugmentation (MRA). The extensive experiments on various image classification benchmarks verify the effectiveness of the proposed augmentation. Specifically, MRA consistently enhances the performance on supervised, semi-supervised as well as few-shot classification.
♻ ☆ Direction-Aware Diagonal Autoregressive Image Generation
The raster-ordered image token sequence exhibits a significant Euclidean distance between index-adjacent tokens at line breaks, making it unsuitable for autoregressive generation. To address this issue, this paper proposes Direction-Aware Diagonal Autoregressive Image Generation (DAR) method, which generates image tokens following a diagonal scanning order. The proposed diagonal scanning order ensures that tokens with adjacent indices remain in close proximity while enabling causal attention to gather information from a broader range of directions. Additionally, two direction-aware modules: 4D-RoPE and direction embeddings are introduced, enhancing the model's capability to handle frequent changes in generation direction. To leverage the representational capacity of the image tokenizer, we use its codebook as the image token embeddings. We propose models of varying scales, ranging from 485M to 2.0B. On the 256$\times$256 ImageNet benchmark, our DAR-XL (2.0B) outperforms all previous autoregressive image generators, achieving a state-of-the-art FID score of 1.37.
♻ ☆ DTFSal: Audio-Visual Dynamic Token Fusion for Video Saliency Prediction
Audio-visual saliency prediction aims to mimic human visual attention by identifying salient regions in videos through the integration of both visual and auditory information. Although visual-only approaches have significantly advanced, effectively incorporating auditory cues remains challenging due to complex spatio-temporal interactions and high computational demands. To address these challenges, we propose Dynamic Token Fusion Saliency (DFTSal), a novel audio-visual saliency prediction framework designed to balance accuracy with computational efficiency. Our approach features a multi-scale visual encoder equipped with two novel modules: the Learnable Token Enhancement Block (LTEB), which adaptively weights tokens to emphasize crucial saliency cues, and the Dynamic Learnable Token Fusion Block (DLTFB), which employs a shifting operation to reorganize and merge features, effectively capturing long-range dependencies and detailed spatial information. In parallel, an audio branch processes raw audio signals to extract meaningful auditory features. Both visual and audio features are integrated using our Adaptive Multimodal Fusion Block (AMFB), which employs local, global, and adaptive fusion streams for precise cross-modal fusion. The resulting fused features are processed by a hierarchical multi-decoder structure, producing accurate saliency maps. Extensive evaluations on six audio-visual benchmarks demonstrate that DFTSal achieves SOTA performance while maintaining computational efficiency.
♻ ☆ Kernel-Aware Graph Prompt Learning for Few-Shot Anomaly Detection AAAI 2025
Few-shot anomaly detection (FSAD) aims to detect unseen anomaly regions with the guidance of very few normal support images from the same class. Existing FSAD methods usually find anomalies by directly designing complex text prompts to align them with visual features under the prevailing large vision-language model paradigm. However, these methods, almost always, neglect intrinsic contextual information in visual features, e.g., the interaction relationships between different vision layers, which is an important clue for detecting anomalies comprehensively. To this end, we propose a kernel-aware graph prompt learning framework, termed as KAG-prompt, by reasoning the cross-layer relations among visual features for FSAD. Specifically, a kernel-aware hierarchical graph is built by taking the different layer features focusing on anomalous regions of different sizes as nodes, meanwhile, the relationships between arbitrary pairs of nodes stand for the edges of the graph. By message passing over this graph, KAG-prompt can capture cross-layer contextual information, thus leading to more accurate anomaly prediction. Moreover, to integrate the information of multiple important anomaly signals in the prediction map, we propose a novel image-level scoring method based on multi-level information fusion. Extensive experiments on MVTecAD and VisA datasets show that KAG-prompt achieves state-of-the-art FSAD results for image-level/pixel-level anomaly detection. Code is available at https://github.com/CVL-hub/KAG-prompt.git.
comment: Accepted to AAAI 2025
♻ ☆ Negate or Embrace: On How Misalignment Shapes Multimodal Representation Learning
Multimodal representation learning, exemplified by multimodal contrastive learning (MMCL) using image-text pairs, aims to learn powerful representations by aligning cues across modalities. This approach relies on the core assumption that the exemplar image-text pairs constitute two representations of an identical concept. However, recent research has revealed that real-world datasets often exhibit misalignment. There are two distinct viewpoints on how to address this issue: one suggests mitigating the misalignment, and the other leveraging it. We seek here to reconcile these seemingly opposing perspectives, and to provide a practical guide for practitioners. Using latent variable models we thus formalize misalignment by introducing two specific mechanisms: selection bias, where some semantic variables are missing, and perturbation bias, where semantic variables are distorted -- both affecting latent variables shared across modalities. Our theoretical analysis demonstrates that, under mild assumptions, the representations learned by MMCL capture exactly the information related to the subset of the semantic variables invariant to selection and perturbation biases. This provides a unified perspective for understanding misalignment. Based on this, we further offer actionable insights into how misalignment should inform the design of real-world ML systems. We validate our theoretical findings through extensive empirical studies on both synthetic data and real image-text datasets, shedding light on the nuanced impact of misalignment on multimodal representation learning.
♻ ☆ 3DAffordSplat: Efficient Affordance Reasoning with 3D Gaussians
3D affordance reasoning is essential in associating human instructions with the functional regions of 3D objects, facilitating precise, task-oriented manipulations in embodied AI. However, current methods, which predominantly depend on sparse 3D point clouds, exhibit limited generalizability and robustness due to their sensitivity to coordinate variations and the inherent sparsity of the data. By contrast, 3D Gaussian Splatting (3DGS) delivers high-fidelity, real-time rendering with minimal computational overhead by representing scenes as dense, continuous distributions. This positions 3DGS as a highly effective approach for capturing fine-grained affordance details and improving recognition accuracy. Nevertheless, its full potential remains largely untapped due to the absence of large-scale, 3DGS-specific affordance datasets. To overcome these limitations, we present 3DAffordSplat, the first large-scale, multi-modal dataset tailored for 3DGS-based affordance reasoning. This dataset includes 23,677 Gaussian instances, 8,354 point cloud instances, and 6,631 manually annotated affordance labels, encompassing 21 object categories and 18 affordance types. Building upon this dataset, we introduce AffordSplatNet, a novel model specifically designed for affordance reasoning using 3DGS representations. AffordSplatNet features an innovative cross-modal structure alignment module that exploits structural consistency priors to align 3D point cloud and 3DGS representations, resulting in enhanced affordance recognition accuracy. Extensive experiments demonstrate that the 3DAffordSplat dataset significantly advances affordance learning within the 3DGS domain, while AffordSplatNet consistently outperforms existing methods across both seen and unseen settings, highlighting its robust generalization capabilities.
comment: The first large-scale 3D Gaussians Affordance Reasoning Benchmark
♻ ☆ DyDiT++: Dynamic Diffusion Transformers for Efficient Visual Generation ICLR
Diffusion Transformer (DiT), an emerging diffusion model for visual generation, has demonstrated superior performance but suffers from substantial computational costs. Our investigations reveal that these costs primarily stem from the \emph{static} inference paradigm, which inevitably introduces redundant computation in certain \emph{diffusion timesteps} and \emph{spatial regions}. To overcome this inefficiency, we propose \textbf{Dy}namic \textbf{Di}ffusion \textbf{T}ransformer (DyDiT), an architecture that \emph{dynamically} adjusts its computation along both \emph{timestep} and \emph{spatial} dimensions. Specifically, we introduce a \emph{Timestep-wise Dynamic Width} (TDW) approach that adapts model width conditioned on the generation timesteps. In addition, we design a \emph{Spatial-wise Dynamic Token} (SDT) strategy to avoid redundant computation at unnecessary spatial locations. TDW and SDT can be seamlessly integrated into DiT and significantly accelerates the generation process. Building on these designs, we further enhance DyDiT in three key aspects. First, DyDiT is integrated seamlessly with flow matching-based generation, enhancing its versatility. Furthermore, we enhance DyDiT to tackle more complex visual generation tasks, including video generation and text-to-image generation, thereby broadening its real-world applications. Finally, to address the high cost of full fine-tuning and democratize technology access, we investigate the feasibility of training DyDiT in a parameter-efficient manner and introduce timestep-based dynamic LoRA (TD-LoRA). Extensive experiments on diverse visual generation models, including DiT, SiT, Latte, and FLUX, demonstrate the effectiveness of DyDiT.
comment: Extended journal version for ICLR. arXiv admin note: substantial text overlap with arXiv:2410.03456
♻ ☆ Fine-Grained Rib Fracture Diagnosis with Hyperbolic Embeddings: A Detailed Annotation Framework and Multi-Label Classification Model
Accurate rib fracture identification and classification are essential for treatment planning. However, existing datasets often lack fine-grained annotations, particularly regarding rib fracture characterization, type, and precise anatomical location on individual ribs. To address this, we introduce a novel rib fracture annotation protocol tailored for fracture classification. Further, we enhance fracture classification by leveraging cross-modal embeddings that bridge radiological images and clinical descriptions. Our approach employs hyperbolic embeddings to capture the hierarchical nature of fracture, mapping visual features and textual descriptions into a shared non-Euclidean manifold. This framework enables more nuanced similarity computations between imaging characteristics and clinical descriptions, accounting for the inherent hierarchical relationships in fracture taxonomy. Experimental results demonstrate that our approach outperforms existing methods across multiple classification tasks, with average recall improvements of 6% on the AirRib dataset and 17.5% on the public RibFrac dataset.
♻ ☆ DeepWheel: Generating a 3D Synthetic Wheel Dataset for Design and Performance Evaluation
Data-driven design is emerging as a powerful strategy to accelerate engineering innovation. However, its application to vehicle wheel design remains limited due to the lack of large-scale, high-quality datasets that include 3D geometry and physical performance metrics. To address this gap, this study proposes a synthetic design-performance dataset generation framework using generative AI. The proposed framework first generates 2D rendered images using Stable Diffusion, and then reconstructs the 3D geometry through 2.5D depth estimation. Structural simulations are subsequently performed to extract engineering performance data. To further expand the design and performance space, topology optimization is applied, enabling the generation of a more diverse set of wheel designs. The final dataset, named DeepWheel, consists of over 6,000 photo-realistic images and 900 structurally analyzed 3D models. This multi-modal dataset serves as a valuable resource for surrogate model training, data-driven inverse design, and design space exploration. The proposed methodology is also applicable to other complex design domains. The dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International(CC BY-NC 4.0) and is available on the https://www.smartdesignlab.org/datasets
comment: 28 pages, 18 figures. Not yet submitted to a journal or conference
♻ ☆ Consensus Entropy: Harnessing Multi-VLM Agreement for Self-Verifying and Self-Improving OCR
The Optical Character Recognition (OCR) task is important for evaluating Vision-Language Models (VLMs) and providing high-quality data sources for LLM training data. While state-of-the-art VLMs show improved average OCR accuracy, they still struggle with sample-level quality degradation and lack reliable automatic detection of low-quality outputs. We introduce Consensus Entropy (CE), a training-free post-inference method that quantifies OCR uncertainty by aggregating outputs from multiple VLMs. Our approach exploits a key insight: correct VLM OCR predictions converge in output space while errors diverge. We develop a lightweight multi-model framework that effectively identifies problematic samples, selects the best outputs and combines model strengths. Experiments across multiple OCR benchmarks and VLMs demonstrate that CE outperforms VLM-as-judge approaches and single-model baselines at the same cost and achieves state-of-the-art results across multiple metrics. For instance, our solution demonstrates: achieving 15.2% higher F1 scores than VLM-as-judge methods in quality verification, delivering 6.0% accuracy gains on mathematical calculation tasks, and requiring rephrasing only 7.3% of inputs while maintaining overall performance. Notably, the entire process requires neither training nor supervision while maintaining plug-and-play functionality throughout.
♻ ☆ SignDiff: Diffusion Model for American Sign Language Production
In this paper, we propose a dual-condition diffusion pre-training model named SignDiff that can generate human sign language speakers from a skeleton pose. SignDiff has a novel Frame Reinforcement Network called FR-Net, similar to dense human pose estimation work, which enhances the correspondence between text lexical symbols and sign language dense pose frames, reduces the occurrence of multiple fingers in the diffusion model. In addition, we propose a new method for American Sign Language Production (ASLP), which can generate ASL skeletal pose videos from text input, integrating two new improved modules and a new loss function to improve the accuracy and quality of sign language skeletal posture and enhance the ability of the model to train on large-scale data. We propose the first baseline for ASL production and report the scores of 17.19 and 12.85 on BLEU-4 on the How2Sign dev/test sets. We evaluated our model on the previous mainstream dataset PHOENIX14T, and the experiments achieved the SOTA results. In addition, our image quality far exceeds all previous results by 10 percentage points in terms of SSIM.
comment: Project Page at https://signdiff.github.io
♻ ☆ MomentSeeker: A Comprehensive Benchmark and A Strong Baseline For Moment Retrieval Within Long Videos
Retrieval augmented generation (RAG) holds great promise in addressing challenges associated with long video understanding. These methods retrieve useful moments from long videos for their presented tasks, thereby enabling multimodal large language models (MLLMs) to generate high-quality answers in a cost-effective way. In this work, we present MomentSeeker, a comprehensive benchmark to evaluate retrieval models' performance in handling general long-video moment retrieval (LVMR) tasks. MomentSeeker offers three key advantages. First, it incorporates long videos of over 500 seconds on average, making it the first benchmark specialized for long-video moment retrieval. Second, it covers a wide range of task categories (including Moment Search, Caption Alignment, Image-conditioned Moment Search, and Video-conditioned Moment Search) and diverse application scenarios (e.g., sports, movies, cartoons, and ego), making it a comprehensive tool for assessing retrieval models' general LVMR performance. Additionally, the evaluation tasks are carefully curated through human annotation, ensuring the reliability of assessment. We further fine-tune an MLLM-based LVMR retriever on synthetic data, which demonstrates strong performance on our benchmark. We perform extensive experiments with various popular multimodal retrievers based on our benchmark, whose results highlight the challenges of LVMR and limitations for existing methods. Our created resources will be shared with community to advance future research in this field.
♻ ☆ Partial Label Learning for Emotion Recognition from EEG
Fully supervised learning has recently achieved promising performance in various electroencephalography (EEG) learning tasks by training on large datasets with ground truth labels. However, labeling EEG data for affective experiments is challenging, as it can be difficult for participants to accurately distinguish between similar emotions, resulting in ambiguous labeling (reporting multiple emotions for one EEG instance). This notion could cause model performance degradation, as the ground truth is hidden within multiple candidate labels. To address this issue, Partial Label Learning (PLL) has been proposed to identify the ground truth from candidate labels during the training phase, and has shown good performance in the computer vision domain. However, PLL methods have not yet been adopted for EEG representation learning or implemented for emotion recognition tasks. In this paper, we adapt and re-implement six state-of-the-art PLL approaches for emotion recognition from EEG on two large emotion datasets (SEED-IV and SEED-V). These datasets contain four and five categories of emotions, respectively. We evaluate the performance of all methods in classical, circumplex-based and real-world experiments. The results show that PLL methods can achieve strong results in affective computing from EEG and achieve comparable performance to fully supervised learning. We also investigate the effect of label disambiguation, a key step in many PLL methods. The results show that in most cases, label disambiguation would benefit the model when the candidate labels are generated based on their similarities to the ground truth rather than obeying a uniform distribution. This finding suggests the potential of using label disambiguation-based PLL methods for circumplex-based and real-world affective tasks.
comment: 15 pages, 13 figures
♻ ☆ Imagery as Inquiry: Exploring A Multimodal Dataset for Conversational Recommendation
We introduce a multimodal dataset where users express preferences through images. These images encompass a broad spectrum of visual expressions ranging from landscapes to artistic depictions. Users request recommendations for books or music that evoke similar feelings to those captured in the images, and recommendations are endorsed by the community through upvotes. This dataset supports two recommendation tasks: title generation and multiple-choice selection. Our experiments with large foundation models reveal their limitations in these tasks. Particularly, vision-language models show no significant advantage over language-only counterparts that use descriptions, which we hypothesize is due to underutilized visual capabilities. To better harness these abilities, we propose the chain-of-imagery prompting, which results in notable improvements. We release our code and datasets.
♻ ☆ Versatile Multimodal Controls for Expressive Talking Human Animation
In filmmaking, directors typically allow actors to perform freely based on the script before providing specific guidance on how to present key actions. AI-generated content faces similar requirements, where users not only need automatic generation of lip synchronization and basic gestures from audio input but also desire semantically accurate and expressive body movement that can be ``directly guided'' through text descriptions. Therefore, we present VersaAnimator, a versatile framework that synthesizes expressive talking human videos from arbitrary portrait images. Specifically, we design a motion generator that produces basic rhythmic movements from audio input and supports text-prompt control for specific actions. The generated whole-body 3D motion tokens can animate portraits of various scales, producing talking heads, half-body gestures and even leg movements for whole-body images. Besides, we introduce a multi-modal controlled video diffusion that generates photorealistic videos, where speech signals govern lip synchronization, facial expressions, and head motions while body movements are guided by the 2D poses. Furthermore, we introduce a token2pose translator to smoothly map 3D motion tokens to 2D pose sequences. This design mitigates the stiffness resulting from direct 3D to 2D conversion and enhances the details of the generated body movements. Extensive experiments shows that VersaAnimator synthesizes lip-synced and identity-preserving videos while generating expressive and semantically meaningful whole-body motions.
♻ ☆ UI-E2I-Synth: Advancing GUI Grounding with Large-Scale Instruction Synthesis
Recent advancements in Large Vision-Language Models are accelerating the development of Graphical User Interface (GUI) agents that utilize human-like vision perception capabilities to enhance productivity on digital devices. Compared to approaches predicated on GUI metadata, which are platform-dependent and vulnerable to implementation variations, vision-based approaches offer broader applicability. In this vision-based paradigm, the GUI instruction grounding, which maps user instruction to the location of corresponding element on the given screenshot, remains a critical challenge, particularly due to limited public training dataset and resource-intensive manual instruction data annotation. In this paper, we delve into unexplored challenges in this task including element-to-screen ratio, unbalanced element type, and implicit instruction. To address these challenges, we introduce a large-scale data synthesis pipeline UI-E2I-Synth for generating varying complex instruction datasets using GPT-4o instead of human annotators. Furthermore, we propose a new GUI instruction grounding benchmark UI-I2E-Bench, which is designed to address the limitations of existing benchmarks by incorporating diverse annotation aspects. Our model, trained on the synthesized data, achieves superior performance in GUI instruction grounding, demonstrating the advancements of proposed data synthesis pipeline. The proposed benchmark, accompanied by extensive analyses, provides practical insights for future research in GUI grounding. We will release corresponding artifacts at https://colmon46.github.io/i2e-bench-leaderboard/ .
♻ ☆ OMR-Diffusion:Optimizing Multi-Round Enhanced Training in Diffusion Models for Improved Intent Understanding
Generative AI has significantly advanced text-driven image generation, but it still faces challenges in producing outputs that consistently align with evolving user preferences and intents, particularly in multi-turn dialogue scenarios. In this research, We present a Visual Co-Adaptation (VCA) framework that incorporates human-in-the-loop feedback, utilizing a well-trained reward model specifically designed to closely align with human preferences. Using a diverse multi-turn dialogue dataset, the framework applies multiple reward functions (such as diversity, consistency, and preference feedback) to refine the diffusion model through LoRA, effectively optimizing image generation based on user input. We also constructed multi-round dialogue datasets with prompts and image pairs that well-fit user intent. Experiments show the model achieves 508 wins in human evaluation, outperforming DALL-E 3 (463 wins) and others. It also achieves 3.4 rounds in dialogue efficiency (vs. 13.7 for DALL-E 3) and excels in metrics like LPIPS (0.15) and BLIP (0.59). Various experiments demonstrate the effectiveness of the proposed method over state-of-the-art baselines, with significant improvements in image consistency and alignment with user intent.
♻ ☆ Diffusion Distillation With Direct Preference Optimization For Efficient 3D LiDAR Scene Completion
The application of diffusion models in 3D LiDAR scene completion is limited due to diffusion's slow sampling speed. Score distillation accelerates diffusion sampling but with performance degradation, while post-training with direct policy optimization (DPO) boosts performance using preference data. This paper proposes Distillation-DPO, a novel diffusion distillation framework for LiDAR scene completion with preference aligment. First, the student model generates paired completion scenes with different initial noises. Second, using LiDAR scene evaluation metrics as preference, we construct winning and losing sample pairs. Such construction is reasonable, since most LiDAR scene metrics are informative but non-differentiable to be optimized directly. Third, Distillation-DPO optimizes the student model by exploiting the difference in score functions between the teacher and student models on the paired completion scenes. Such procedure is repeated until convergence. Extensive experiments demonstrate that, compared to state-of-the-art LiDAR scene completion diffusion models, Distillation-DPO achieves higher-quality scene completion while accelerating the completion speed by more than 5-fold. Our method is the first to explore adopting preference learning in distillation to the best of our knowledge and provide insights into preference-aligned distillation. Our code is public available on https://github.com/happyw1nd/DistillationDPO.
comment: Our code is public available on https://github.com/happyw1nd/DistillationDPO
♻ ☆ TDRI: Two-Phase Dialogue Refinement and Co-Adaptation for Interactive Image Generation
Although text-to-image generation technologies have made significant advancements, they still face challenges when dealing with ambiguous prompts and aligning outputs with user intent.Our proposed framework, TDRI (Two-Phase Dialogue Refinement and Co-Adaptation), addresses these issues by enhancing image generation through iterative user interaction. It consists of two phases: the Initial Generation Phase, which creates base images based on user prompts, and the Interactive Refinement Phase, which integrates user feedback through three key modules. The Dialogue-to-Prompt (D2P) module ensures that user feedback is effectively transformed into actionable prompts, which improves the alignment between user intent and model input. By evaluating generated outputs against user expectations, the Feedback-Reflection (FR) module identifies discrepancies and facilitates improvements. In an effort to ensure consistently high-quality results, the Adaptive Optimization (AO) module fine-tunes the generation process by balancing user preferences and maintaining prompt fidelity. Experimental results show that TDRI outperforms existing methods by achieving 33.6% human preference, compared to 6.2% for GPT-4 augmentation, and the highest CLIP and BLIP alignment scores (0.338 and 0.336, respectively). In iterative feedback tasks, user satisfaction increased to 88% after 8 rounds, with diminishing returns beyond 6 rounds. Furthermore, TDRI has been found to reduce the number of iterations and improve personalization in the creation of fashion products. TDRI exhibits a strong potential for a wide range of applications in the creative and industrial domains, as it streamlines the creative process and improves alignment with user preferences
♻ ☆ Learning to Learn Transferable Generative Attack for Person Re-Identification
Deep learning-based person re-identification (re-id) models are widely employed in surveillance systems and inevitably inherit the vulnerability of deep networks to adversarial attacks. Existing attacks merely consider cross-dataset and cross-model transferability, ignoring the cross-test capability to perturb models trained in different domains. To powerfully examine the robustness of real-world re-id models, the Meta Transferable Generative Attack (MTGA) method is proposed, which adopts meta-learning optimization to promote the generative attacker producing highly transferable adversarial examples by learning comprehensively simulated transfer-based cross-model\&dataset\&test black-box meta attack tasks. Specifically, cross-model\&dataset black-box attack tasks are first mimicked by selecting different re-id models and datasets for meta-train and meta-test attack processes. As different models may focus on different feature regions, the Perturbation Random Erasing module is further devised to prevent the attacker from learning to only corrupt model-specific features. To boost the attacker learning to possess cross-test transferability, the Normalization Mix strategy is introduced to imitate diverse feature embedding spaces by mixing multi-domain statistics of target models. Extensive experiments show the superiority of MTGA, especially in cross-model\&dataset and cross-model\&dataset\&test attacks, our MTGA outperforms the SOTA methods by 21.5\% and 11.3\% on mean mAP drop rate, respectively. The code of MTGA will be released after the paper is accepted.
♻ ☆ GATE3D: Generalized Attention-based Task-synergized Estimation in 3D*
The emerging trend in computer vision emphasizes developing universal models capable of simultaneously addressing multiple diverse tasks. Such universality typically requires joint training across multi-domain datasets to ensure effective generalization. However, monocular 3D object detection presents unique challenges in multi-domain training due to the scarcity of datasets annotated with accurate 3D ground-truth labels, especially beyond typical road-based autonomous driving contexts. To address this challenge, we introduce a novel weakly supervised framework leveraging pseudo-labels. Current pretrained models often struggle to accurately detect pedestrians in non-road environments due to inherent dataset biases. Unlike generalized image-based 2D object detection models, achieving similar generalization in monocular 3D detection remains largely unexplored. In this paper, we propose GATE3D, a novel framework designed specifically for generalized monocular 3D object detection via weak supervision. GATE3D effectively bridges domain gaps by employing consistency losses between 2D and 3D predictions. Remarkably, our model achieves competitive performance on the KITTI benchmark as well as on an indoor-office dataset collected by us to evaluate the generalization capabilities of our framework. Our results demonstrate that GATE3D significantly accelerates learning from limited annotated data through effective pre-training strategies, highlighting substantial potential for broader impacts in robotics, augmented reality, and virtual reality applications. Project page: https://ies0411.github.io/GATE3D/
comment: 9pages, 1 supple
♻ ☆ Visual Language Models show widespread visual deficits on neuropsychological tests
Visual Language Models (VLMs) show remarkable performance in visual reasoning tasks, successfully tackling college-level challenges that require high-level understanding of images. However, some recent reports of VLMs struggling to reason about elemental visual concepts like orientation, position, continuity, and occlusion suggest a potential gulf between human and VLM vision. Here we use the toolkit of neuropsychology to systematically assess the capabilities of three state-of-the-art VLMs across visual domains. Using 51 tests drawn from six clinical and experimental batteries, we characterise the visual abilities of leading VLMs relative to normative performance in healthy adults. While the models excel in straightforward object recognition tasks, we find widespread deficits in low- and mid-level visual abilities that would be considered clinically significant in humans. These selective deficits, profiled through validated test batteries, suggest that an artificial system can achieve complex object recognition without developing foundational visual concepts that in humans require no explicit training.
comment: 31 pages, 3 figures, 1 supplementary document with 1 figure and 51 sample images; corrected typo in Fig 1
Information Retrieval 15
☆ Clarifying Ambiguities: on the Role of Ambiguity Types in Prompting Methods for Clarification Generation SIGIR 2025
In information retrieval (IR), providing appropriate clarifications to better understand users' information needs is crucial for building a proactive search-oriented dialogue system. Due to the strong in-context learning ability of large language models (LLMs), recent studies investigate prompting methods to generate clarifications using few-shot or Chain of Thought (CoT) prompts. However, vanilla CoT prompting does not distinguish the characteristics of different information needs, making it difficult to understand how LLMs resolve ambiguities in user queries. In this work, we focus on the concept of ambiguity for clarification, seeking to model and integrate ambiguities in the clarification process. To this end, we comprehensively study the impact of prompting schemes based on reasoning and ambiguity for clarification. The idea is to enhance the reasoning abilities of LLMs by limiting CoT to predict first ambiguity types that can be interpreted as instructions to clarify, then correspondingly generate clarifications. We name this new prompting scheme Ambiguity Type-Chain of Thought (AT-CoT). Experiments are conducted on various datasets containing human-annotated clarifying questions to compare AT-CoT with multiple baselines. We also perform user simulations to implicitly measure the quality of generated clarifications under various IR scenarios.
comment: 11 pages, 3 figures. Accepted at SIGIR 2025
☆ Optimizing Compound Retrieval Systems SIGIR 2025
Modern retrieval systems do not rely on a single ranking model to construct their rankings. Instead, they generally take a cascading approach where a sequence of ranking models are applied in multiple re-ranking stages. Thereby, they balance the quality of the top-K ranking with computational costs by limiting the number of documents each model re-ranks. However, the cascading approach is not the only way models can interact to form a retrieval system. We propose the concept of compound retrieval systems as a broader class of retrieval systems that apply multiple prediction models. This encapsulates cascading models but also allows other types of interactions than top-K re-ranking. In particular, we enable interactions with large language models (LLMs) which can provide relative relevance comparisons. We focus on the optimization of compound retrieval system design which uniquely involves learning where to apply the component models and how to aggregate their predictions into a final ranking. This work shows how our compound approach can combine the classic BM25 retrieval model with state-of-the-art (pairwise) LLM relevance predictions, while optimizing a given ranking metric and efficiency target. Our experimental results show optimized compound retrieval systems provide better trade-offs between effectiveness and efficiency than cascading approaches, even when applied in a self-supervised manner. With the introduction of compound retrieval systems, we hope to inspire the information retrieval field to more out-of-the-box thinking on how prediction models can interact to form rankings.
comment: SIGIR 2025
☆ Generative Recommendation with Continuous-Token Diffusion
In recent years, there has been a significant trend toward using large language model (LLM)-based recommender systems (RecSys). Current research primarily focuses on representing complex user-item interactions within a discrete space to align with the inherent discrete nature of language models. However, this approach faces limitations due to its discrete nature: (i) information is often compressed during discretization; (ii) the tokenization and generation for the vast number of users and items in real-world scenarios are constrained by a limited vocabulary. Embracing continuous data presents a promising alternative to enhance expressive capabilities, though this approach is still in its early stages. To address this gap, we propose a novel framework, DeftRec, which incorporates \textbf{de}noising di\textbf{f}fusion models to enable LLM-based RecSys to seamlessly support continuous \textbf{t}oken as input and target. First, we introduce a robust tokenizer with a masking operation and an additive K-way architecture to index users and items, capturing their complex collaborative relationships into continuous tokens. Crucially, we develop a denoising diffusion model to process user preferences within continuous domains by conditioning on reasoning content from pre-trained large language model. During the denoising process, we reformulate the objective to include negative interactions, building a comprehensive understanding of user preferences for effective and accurate recommendation generation. Finally, given a continuous token as output, recommendations can be easily generated through score-based retrieval. Extensive experiments demonstrate the effectiveness of the proposed methods, showing that DeftRec surpasses competitive benchmarks, including both traditional and emerging LLM-based RecSys.
☆ Rethinking LLM-Based Recommendations: A Query Generation-Based, Training-Free Approach
Existing large language model LLM-based recommendation methods face several challenges, including inefficiency in handling large candidate pools, sensitivity to item order within prompts ("lost in the middle" phenomenon) poor scalability, and unrealistic evaluation due to random negative sampling. To address these issues, we propose a Query-to-Recommendation approach that leverages LLMs to generate personalized queries for retrieving relevant items from the entire candidate pool, eliminating the need for candidate pre-selection. This method can be integrated into an ID-based recommendation system without additional training, enhances recommendation performance and diversity through LLMs' world knowledge, and performs well even for less popular item groups. Experiments on three datasets show up to 57 percent improvement, with an average gain of 31 percent, demonstrating strong zero-shot performance and further gains when ensembled with existing models.
☆ Résumé abstractif à partir d'une transcription audio
Currently, large language models are gaining popularity, their achievements are used in many areas, ranging from text translation to generating answers to queries. However, the main problem with these new machine learning algorithms is that training such models requires large computing resources that only large IT companies have. To avoid this problem, a number of methods (LoRA, quantization) have been proposed so that existing models can be effectively fine-tuned for specific tasks. In this paper, we propose an E2E (end to end) audio summarization model using these techniques. In addition, this paper examines the effectiveness of these approaches to the problem under consideration and draws conclusions about the applicability of these methods.
comment: 35 pages, in French language, 8 tables, 6 figures
☆ A New Paradigm of User-Centric Wireless Communication Driven by Large Language Models
The next generation of wireless communications seeks to deeply integrate artificial intelligence (AI) with user-centric communication networks, with the goal of developing AI-native networks that more accurately address user requirements. The rapid development of large language models (LLMs) offers significant potential in realizing these goals. However, existing efforts that leverage LLMs for wireless communication often overlook the considerable gap between human natural language and the intricacies of real-world communication systems, thus failing to fully exploit the capabilities of LLMs. To address this gap, we propose a novel LLM-driven paradigm for wireless communication that innovatively incorporates the nature language to structured query language (NL2SQL) tool. Specifically, in this paradigm, user personal requirements is the primary focus. Upon receiving a user request, LLMs first analyze the user intent in terms of relevant communication metrics and system parameters. Subsequently, a structured query language (SQL) statement is generated to retrieve the specific parameter values from a high-performance real-time database. We further utilize LLMs to formulate and solve an optimization problem based on the user request and the retrieved parameters. The solution to this optimization problem then drives adjustments in the communication system to fulfill the user's requirements. To validate the feasibility of the proposed paradigm, we present a prototype system. In this prototype, we consider user-request centric semantic communication (URC-SC) system in which a dynamic semantic representation network at the physical layer adapts its encoding depth to meet user requirements. Additionally, two LLMs are employed to analyze user requests and generate SQL statements, respectively. Simulation results demonstrate the effectiveness.
comment: 8 pages, 5 figures
☆ A Human-AI Comparative Analysis of Prompt Sensitivity in LLM-Based Relevance Judgment
Large Language Models (LLMs) are increasingly used to automate relevance judgments for information retrieval (IR) tasks, often demonstrating agreement with human labels that approaches inter-human agreement. To assess the robustness and reliability of LLM-based relevance judgments, we systematically investigate impact of prompt sensitivity on the task. We collected prompts for relevance assessment from 15 human experts and 15 LLMs across three tasks~ -- ~binary, graded, and pairwise~ -- ~yielding 90 prompts in total. After filtering out unusable prompts from three humans and three LLMs, we employed the remaining 72 prompts with three different LLMs as judges to label document/query pairs from two TREC Deep Learning Datasets (2020 and 2021). We compare LLM-generated labels with TREC official human labels using Cohen's $\kappa$ and pairwise agreement measures. In addition to investigating the impact of prompt variations on agreement with human labels, we compare human- and LLM-generated prompts and analyze differences among different LLMs as judges. We also compare human- and LLM-generated prompts with the standard UMBRELA prompt used for relevance assessment by Bing and TREC 2024 Retrieval Augmented Generation (RAG) Track. To support future research in LLM-based evaluation, we release all data and prompts at https://github.com/Narabzad/prompt-sensitivity-relevance-judgements/.
♻ ☆ Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation
Sequential Recommendation (SeqRec) aims to predict the next item by capturing sequential patterns from users' historical interactions, playing a crucial role in many real-world recommender systems. However, existing approaches predominantly adopt a direct forward computation paradigm, where the final hidden state of the sequence encoder serves as the user representation. We argue that this inference paradigm, due to its limited computational depth, struggles to model the complex evolving nature of user preferences and lacks a nuanced understanding of long-tail items, leading to suboptimal performance. To address this issue, we propose \textbf{ReaRec}, the first inference-time computing framework for recommender systems, which enhances user representations through implicit multi-step reasoning. Specifically, ReaRec autoregressively feeds the sequence's last hidden state into the sequential recommender while incorporating special reasoning position embeddings to decouple the original item encoding space from the multi-step reasoning space. Moreover, we introduce two lightweight reasoning-based learning methods, Ensemble Reasoning Learning (ERL) and Progressive Reasoning Learning (PRL), to further effectively exploit ReaRec's reasoning potential. Extensive experiments on five public real-world datasets and different SeqRec architectures demonstrate the generality and effectiveness of our proposed ReaRec. Remarkably, post-hoc analyses reveal that ReaRec significantly elevates the performance ceiling of multiple sequential recommendation backbones by approximately 30\%-50\%. Thus, we believe this work can open a new and promising avenue for future research in inference-time computing for sequential recommendation.
♻ ☆ Towards Personalized Conversational Sales Agents : Contextual User Profiling for Strategic Action
Conversational Recommender Systems (CRSs) aim to engage users in dialogue to provide tailored recommendations. While traditional CRSs focus on eliciting preferences and retrieving items, real-world e-commerce interactions involve more complex decision-making, where users consider multiple factors beyond simple attributes. To bridge this gap, we introduce Conversational Sales (CSales), a novel task that unifies preference elicitation, recommendation, and persuasion to better support user decision-making. For a realistic evaluation of CSales, we present CSUser, an LLM-based user simulator constructed from real-world data, modeling diverse user profiles with needs and personalities. Additionally, we propose CSI, a conversational sales agent that proactively infers contextual profiles through dialogue for personalized action planning. Extensive experiments demonstrate that CSUser effectively replicates real-world users and emphasize the importance of contextual profiling for strategic action selection, ultimately driving successful purchases in e-commerce.
♻ ☆ AI-Driven Sentiment Analytics: Unlocking Business Value in the E-Commerce Landscape_v1
The rapid growth of e-commerce has led to an overwhelming volume of customer feedback, from product reviews to service interactions. Extracting meaningful insights from this data is crucial for businesses aiming to improve customer satisfaction and optimize decision-making. This paper presents an AI-driven sentiment analysis system designed specifically for e-commerce applications, balancing accuracy with interpretability. Our approach integrates traditional machine learning techniques with modern deep learning models, allowing for a more nuanced understanding of customer sentiment while ensuring transparency in decision-making. Experimental results show that our system outperforms standard sentiment analysis methods, achieving an accuracy of 89.7% on diverse, large-scale datasets. Beyond technical performance, real-world implementation across multiple e-commerce platforms demonstrates tangible improvements in customer engagement and operational efficiency. This study highlights both the potential and the challenges of applying AI to sentiment analysis in a commercial setting, offering insights into practical deployment strategies and areas for future refinement.
comment: 7 pages
♻ ☆ Efficient Distributed Retrieval-Augmented Generation for Enhancing Language Model Performance
Small language models (SLMs) support efficient deployments on resource-constrained edge devices, but their limited capacity compromises inference performance. Retrieval-augmented generation (RAG) is a promising solution to enhance model performance by integrating external databases, without requiring intensive on-device model retraining. However, large-scale public databases and user-specific private contextual documents are typically located on the cloud and the device separately, while existing RAG implementations are primarily centralized. To bridge this gap, we propose DRAGON, a distributed RAG framework to enhance on-device SLMs through both general and personal knowledge without the risk of leaking document privacy. Specifically, DRAGON decomposes multi-document RAG into multiple parallel token generation processes performed independently and locally on the cloud and the device, and employs a newly designed Speculative Aggregation, a dual-side speculative algorithm to avoid frequent output synchronization between the cloud and device. A new scheduling algorithm is further introduced to identify the optimal aggregation side based on real-time network conditions. Evaluations on real-world hardware testbed demonstrate a significant performance improvement of DRAGON-up to 1.9x greater gains over standalone SLM compared to the centralized RAG, substantial reduction in per-token latency, and negligible Time to First Token (TTFT) overhead.
♻ ☆ PhishLang: A Real-Time, Fully Client-Side Phishing Detection Framework Using MobileBERT
In this paper, we introduce PhishLang, the first fully client-side anti-phishing framework built on a lightweight ensemble framework that utilizes advanced language models to analyze the contextual features of a website's source code and URL. Unlike traditional heuristic or machine learning approaches that rely on static features and struggle to adapt to evolving threats, or deep learning models that are computationally intensive, our approach utilizes MobileBERT, a fast and memory-efficient variant of the BERT architecture, to capture nuanced features indicative of phishing attacks. To further enhance detection accuracy, PhishLang employs a multi-modal ensemble approach, combining both the URL and Source detection models. This architecture ensures robustness by allowing one model to compensate for scenarios where the other may fail, or if both models provide ambiguous inferences. As a result, PhishLang excels at detecting both regular and evasive phishing threats, including zero-day attacks, outperforming popular anti-phishing tools, while operating without relying on external blocklists and safeguarding user privacy by ensuring that browser history remains entirely local and unshared. We release PhishLang as a Chromium browser extension and also open-source the framework to aid the research community.
♻ ☆ CSPLADE: Learned Sparse Retrieval with Causal Language Models
In recent years, dense retrieval has been the focus of information retrieval (IR) research. While effective, dense retrieval produces uninterpretable dense vectors, and suffers from the drawback of large index size. Learned sparse retrieval (LSR) has emerged as promising alternative, achieving competitive retrieval performance while also being able to leverage the classical inverted index data structure for efficient retrieval. However, limited works have explored scaling LSR beyond BERT scale. In this work, we identify two challenges in training large language models (LLM) for LSR: (1) training instability during the early stage of contrastive training; (2) suboptimal performance due to pre-trained LLM's unidirectional attention. To address these challenges, we propose two corresponding techniques: (1) a lightweight adaptation training phase to eliminate training instability; (2) two model variants to enable bidirectional information. With these techniques, we are able to train LSR models with 8B scale LLM, and achieve competitive retrieval performance with reduced index size. Furthermore, we are among the first to analyze the performance-efficiency tradeoff of LLM-based LSR model through the lens of model quantization. Our findings provide insights into adapting LLMs for efficient retrieval modeling.
♻ ☆ Taxonomy and Analysis of Sensitive User Queries in Generative AI Search NAACL2025
Although there has been a growing interest among industries in integrating generative LLMs into their services, limited experience and scarcity of resources act as a barrier in launching and servicing large-scale LLM-based services. In this paper, we share our experiences in developing and operating generative AI models within a national-scale search engine, with a specific focus on the sensitiveness of user queries. We propose a taxonomy for sensitive search queries, outline our approaches, and present a comprehensive analysis report on sensitive queries from actual users. We believe that our experiences in launching generative AI search systems can contribute to reducing the barrier in building generative LLM-based services.
comment: NAACL2025(Findings), corrected typo in co-corresponding authors
♻ ☆ Multi-Field Adaptive Retrieval ICLR 2025
Document retrieval for tasks such as search and retrieval-augmented generation typically involves datasets that are unstructured: free-form text without explicit internal structure in each document. However, documents can have a structured form, consisting of fields such as an article title, message body, or HTML header. To address this gap, we introduce Multi-Field Adaptive Retrieval (MFAR), a flexible framework that accommodates any number of and any type of document indices on structured data. Our framework consists of two main steps: (1) the decomposition of an existing document into fields, each indexed independently through dense and lexical methods, and (2) learning a model which adaptively predicts the importance of a field by conditioning on the document query, allowing on-the-fly weighting of the most likely field(s). We find that our approach allows for the optimized use of dense versus lexical representations across field types, significantly improves in document ranking over a number of existing retrievers, and achieves state-of-the-art performance for multi-field structured data.
comment: ICLR 2025, Spotlight
Robotics 35
☆ A Clean Slate for Offline Reinforcement Learning
Progress in offline reinforcement learning (RL) has been impeded by ambiguous problem definitions and entangled algorithmic designs, resulting in inconsistent implementations, insufficient ablations, and unfair evaluations. Although offline RL explicitly avoids environment interaction, prior methods frequently employ extensive, undocumented online evaluation for hyperparameter tuning, complicating method comparisons. Moreover, existing reference implementations differ significantly in boilerplate code, obscuring their core algorithmic contributions. We address these challenges by first introducing a rigorous taxonomy and a transparent evaluation protocol that explicitly quantifies online tuning budgets. To resolve opaque algorithmic design, we provide clean, minimalistic, single-file implementations of various model-free and model-based offline RL methods, significantly enhancing clarity and achieving substantial speed-ups. Leveraging these streamlined implementations, we propose Unifloral, a unified algorithm that encapsulates diverse prior approaches within a single, comprehensive hyperparameter space, enabling algorithm development in a shared hyperparameter space. Using Unifloral with our rigorous evaluation protocol, we develop two novel algorithms - TD3-AWR (model-free) and MoBRAC (model-based) - which substantially outperform established baselines. Our implementation is publicly available at https://github.com/EmptyJackson/unifloral.
☆ Improving Swimming Performance in Soft Robotic Fish with Distributed Muscles and Embedded Kinematic Sensing
Bio-inspired underwater vehicles could yield improved efficiency, maneuverability, and environmental compatibility over conventional propeller-driven underwater vehicles. However, to realize the swimming performance of biology, there is a need for soft robotic swimmers with both distributed muscles and kinematic feedback. This study presents the design and swimming performance of a soft robotic fish with independently controllable muscles and embedded kinematic sensing distributed along the body. The soft swimming robot consists of an interior flexible spine, three axially distributed sets of HASEL artificial muscles, embedded strain gauges, a streamlined silicone body, and off-board electronics. In a fixed configuration, the soft robot generates a maximum thrust of 7.9 mN when excited near its first resonant frequency (2 Hz) with synchronized antagonistic actuation of all muscles. When excited near its second resonant frequency (8 Hz), synchronized muscle actuation generates 5.0 mN of thrust. By introducing a sequential phase offset into the muscle actuation, the thrust at the second resonant frequency increases to 7.2 mN, a 44% increase from simple antagonistic activation. The sequential muscle activation improves the thrust by increasing 1) the tail-beat velocity and 2) traveling wave content in the swimming kinematics by four times. Further, the second resonant frequency (8 Hz) generates nearly as much thrust as the first resonance (2 Hz) while requiring only $\approx25$% of the tail displacement, indicating that higher resonant frequencies have benefits for swimming in confined environments where a smaller kinematic envelope is necessary. These results demonstrate the performance benefits of independently controllable muscles and distributed kinematic sensing, and this type of soft robotic swimmer provides a platform to address the open challenge of sensorimotor control.
comment: 7 pages, 5 figures. To be published in the Proceedings of the 8th IEEE - RAS International Conference on Soft Robotics
☆ Next-Future: Sample-Efficient Policy Learning for Robotic-Arm Tasks
Hindsight Experience Replay (HER) is widely regarded as the state-of-the-art algorithm for achieving sample-efficient multi-goal reinforcement learning (RL) in robotic manipulation tasks with binary rewards. HER facilitates learning from failed attempts by replaying trajectories with redefined goals. However, it relies on a heuristic-based replay method that lacks a principled framework. To address this limitation, we introduce a novel replay strategy, "Next-Future", which focuses on rewarding single-step transitions. This approach significantly enhances sample efficiency and accuracy in learning multi-goal Markov decision processes (MDPs), particularly under stringent accuracy requirements -- a critical aspect for performing complex and precise robotic-arm tasks. We demonstrate the efficacy of our method by highlighting how single-step learning enables improved value approximation within the multi-goal RL framework. The performance of the proposed replay strategy is evaluated across eight challenging robotic manipulation tasks, using ten random seeds for training. Our results indicate substantial improvements in sample efficiency for seven out of eight tasks and higher success rates in six tasks. Furthermore, real-world experiments validate the practical feasibility of the learned policies, demonstrating the potential of "Next-Future" in solving complex robotic-arm tasks.
comment: 10 pages, 9 figures, 6 tables
☆ CAP-Net: A Unified Network for 6D Pose and Size Estimation of Categorical Articulated Parts from a Single RGB-D Image CVPR 2025
This paper tackles category-level pose estimation of articulated objects in robotic manipulation tasks and introduces a new benchmark dataset. While recent methods estimate part poses and sizes at the category level, they often rely on geometric cues and complex multi-stage pipelines that first segment parts from the point cloud, followed by Normalized Part Coordinate Space (NPCS) estimation for 6D poses. These approaches overlook dense semantic cues from RGB images, leading to suboptimal accuracy, particularly for objects with small parts. To address these limitations, we propose a single-stage Network, CAP-Net, for estimating the 6D poses and sizes of Categorical Articulated Parts. This method combines RGB-D features to generate instance segmentation and NPCS representations for each part in an end-to-end manner. CAP-Net uses a unified network to simultaneously predict point-wise class labels, centroid offsets, and NPCS maps. A clustering algorithm then groups points of the same predicted class based on their estimated centroid distances to isolate each part. Finally, the NPCS region of each part is aligned with the point cloud to recover its final pose and size. To bridge the sim-to-real domain gap, we introduce the RGBD-Art dataset, the largest RGB-D articulated dataset to date, featuring photorealistic RGB images and depth noise simulated from real sensors. Experimental evaluations on the RGBD-Art dataset demonstrate that our method significantly outperforms the state-of-the-art approach. Real-world deployments of our model in robotic tasks underscore its robustness and exceptional sim-to-real transfer capabilities, confirming its substantial practical utility. Our dataset, code and pre-trained models are available on the project page.
comment: To appear in CVPR 2025 (Highlight)
☆ A Real-time Anomaly Detection Method for Robots based on a Flexible and Sparse Latent Space
The growing demand for robots to operate effectively in diverse environments necessitates the need for robust real-time anomaly detection techniques during robotic operations. However, deep learning-based models in robotics face significant challenges due to limited training data and highly noisy signal features. In this paper, we present Sparse Masked Autoregressive Flow-based Adversarial AutoEncoders model to address these problems. This approach integrates Masked Autoregressive Flow model into Adversarial AutoEncoders to construct a flexible latent space and utilize Sparse autoencoder to efficiently focus on important features, even in scenarios with limited feature space. Our experiments demonstrate that the proposed model achieves a 4.96% to 9.75% higher area under the receiver operating characteristic curve for pick-and-place robotic operations with randomly placed cans, compared to existing state-of-the-art methods. Notably, it showed up to 19.67% better performance in scenarios involving collisions with lightweight objects. Additionally, unlike the existing state-of-the-art model, our model performs inferences within 1 millisecond, ensuring real-time anomaly detection. These capabilities make our model highly applicable to machine learning-based robotic safety systems in dynamic environments. The code will be made publicly available after acceptance.
comment: 20 pages, 11 figures
☆ The Robotability Score: Enabling Harmonious Robot Navigation on Urban Streets
This paper introduces the Robotability Score ($R$), a novel metric that quantifies the suitability of urban environments for autonomous robot navigation. Through expert interviews and surveys, we identify and weigh key features contributing to R for wheeled robots on urban streets. Our findings reveal that pedestrian density, crowd dynamics and pedestrian flow are the most critical factors, collectively accounting for 28% of the total score. Computing robotability across New York City yields significant variation; the area of highest R is 3.0 times more "robotable" than the area of lowest R. Deployments of a physical robot on high and low robotability areas show the adequacy of the score in anticipating the ease of robot navigation. This new framework for evaluating urban landscapes aims to reduce uncertainty in robot deployment while respecting established mobility patterns and urban planning principles, contributing to the discourse on harmonious human-robot environments.
comment: Accepted to CHI '25
☆ FreeDOM: Online Dynamic Object Removal Framework for Static Map Construction Based on Conservative Free Space Estimation
Online map construction is essential for autonomous robots to navigate in unknown environments. However, the presence of dynamic objects may introduce artifacts into the map, which can significantly degrade the performance of localization and path planning. To tackle this problem, a novel online dynamic object removal framework for static map construction based on conservative free space estimation (FreeDOM) is proposed, consisting of a scan-removal front-end and a map-refinement back-end. First, we propose a multi-resolution map structure for fast computation and effective map representation. In the scan-removal front-end, we employ raycast enhancement to improve free space estimation and segment the LiDAR scan based on the estimated free space. In the map-refinement back-end, we further eliminate residual dynamic objects in the map by leveraging incremental free space information. As experimentally verified on SemanticKITTI, HeLiMOS, and indoor datasets with various sensors, our proposed framework overcomes the limitations of visibility-based methods and outperforms state-of-the-art methods with an average F1-score improvement of 9.7%.
☆ A Multi-UAV Formation Obstacle Avoidance Method Combined Improved Simulated Annealing and Adaptive Artificial Potential Field
The traditional Artificial Potential Field (APF) method exhibits limitations in its force distribution: excessive attraction when UAVs are far from the target may cause collisions with obstacles, while insufficient attraction near the goal often results in failure to reach the target. Furthermore, APF is highly susceptible to local minima, compromising motion reliability in complex environments. To address these challenges, this paper presents a novel hybrid obstacle avoidance algorithm-Deflected Simulated Annealing-Adaptive Artificial Potential Field (DSA-AAPF)-which combines an improved simulated annealing mechanism with an enhanced APF model. The proposed approach integrates a Leader-Follower distributed formation strategy with the APF framework, where the resultant force formulation is redefined to smooth UAV trajectories. An adaptive gravitational gain function is introduced to dynamically adjust UAV velocity based on environmental context, and a fast-converging controller ensures accurate and efficient convergence to the target. Moreover, a directional deflection mechanism is embedded within the simulated annealing process, enabling UAVs to escape local minima caused by semi-enclosed obstacles through continuous rotational motion. The simulation results, covering formation reconfiguration, complex obstacle avoidance, and entrapment escape, demonstrate the feasibility, robustness, and superiority of the proposed DSA-AAPF algorithm.
☆ Neural Control Barrier Functions from Physics Informed Neural Networks
As autonomous systems become increasingly prevalent in daily life, ensuring their safety is paramount. Control Barrier Functions (CBFs) have emerged as an effective tool for guaranteeing safety; however, manually designing them for specific applications remains a significant challenge. With the advent of deep learning techniques, recent research has explored synthesizing CBFs using neural networks-commonly referred to as neural CBFs. This paper introduces a novel class of neural CBFs that leverages a physics-inspired neural network framework by incorporating Zubov's Partial Differential Equation (PDE) within the context of safety. This approach provides a scalable methodology for synthesizing neural CBFs applicable to high-dimensional systems. Furthermore, by utilizing reciprocal CBFs instead of zeroing CBFs, the proposed framework allows for the specification of flexible, user-defined safe regions. To validate the effectiveness of the approach, we present case studies on three different systems: an inverted pendulum, autonomous ground navigation, and aerial navigation in obstacle-laden environments.
comment: 8 pages, 5 figures
☆ Acquisition of high-quality images for camera calibration in robotics applications via speech prompts
Accurate intrinsic and extrinsic camera calibration can be an important prerequisite for robotic applications that rely on vision as input. While there is ongoing research on enabling camera calibration using natural images, many systems in practice still rely on using designated calibration targets with e.g. checkerboard patterns or April tag grids. Once calibration images from different perspectives have been acquired and feature descriptors detected, those are typically used in an optimization process to minimize the geometric reprojection error. For this optimization to converge, input images need to be of sufficient quality and particularly sharpness; they should neither contain motion blur nor rolling-shutter artifacts that can arise when the calibration board was not static during image capture. In this work, we present a novel calibration image acquisition technique controlled via voice commands recorded with a clip-on microphone, that can be more robust and user-friendly than e.g. triggering capture with a remote control, or filtering out blurry frames from a video sequence in postprocessing. To achieve this, we use a state-of-the-art speech-to-text transcription model with accurate per-word timestamping to capture trigger words with precise temporal alignment. Our experiments show that the proposed method improves user experience by being fast and efficient, allowing us to successfully calibrate complex multi-camera setups.
comment: 8 pages, 6 figures
☆ $π$-MPPI: A Projection-based Model Predictive Path Integral Scheme for Smooth Optimal Control of Fixed-Wing Aerial Vehicles
Model Predictive Path Integral (MPPI) is a popular sampling-based Model Predictive Control (MPC) algorithm for nonlinear systems. It optimizes trajectories by sampling control sequences and averaging them. However, a key issue with MPPI is the non-smoothness of the optimal control sequence, leading to oscillations in systems like fixed-wing aerial vehicles (FWVs). Existing solutions use post-hoc smoothing, which fails to bound control derivatives. This paper introduces a new approach: we add a projection filter $\pi$ to minimally correct control samples, ensuring bounds on control magnitude and higher-order derivatives. The filtered samples are then averaged using MPPI, leading to our $\pi$-MPPI approach. We minimize computational overhead by using a neural accelerated custom optimizer for the projection filter. $\pi$-MPPI offers a simple way to achieve arbitrary smoothness in control sequences. While we focus on FWVs, this projection filter can be integrated into any MPPI pipeline. Applied to FWVs, $\pi$-MPPI is easier to tune than the baseline, resulting in smoother, more robust performance.
comment: 8 pages, 4 figures, submitted to IEEE RA-L
☆ A Sublinear Algorithm for Path Feasibility Among Rectangular Obstacles
The problem of finding a path between two points while avoiding obstacles is critical in robotic path planning. We focus on the feasibility problem: determining whether such a path exists. We model the robot as a query-specific rectangular object capable of moving parallel to its sides. The obstacles are axis-aligned, rectangular, and may overlap. Most previous works only consider nondisjoint rectangular objects and point-sized or statically sized robots. Our approach introduces a novel technique leveraging generalized Gabriel graphs and constructs a data structure to facilitate online queries regarding path feasibility with varying robot sizes in sublinear time. To efficiently handle feasibility queries, we propose an online algorithm utilizing sweep line to construct a generalized Gabriel graph under the $L_\infty$ norm, capturing key gap constraints between obstacles. We utilize a persistent disjoint-set union data structure to efficiently determine feasibility queries in $\mathcal{O}(\log n)$ time and $\mathcal{O}(n)$ total space.
☆ ZeroGrasp: Zero-Shot Shape Reconstruction Enabled Robotic Grasping CVPR 2025
Robotic grasping is a cornerstone capability of embodied systems. Many methods directly output grasps from partial information without modeling the geometry of the scene, leading to suboptimal motion and even collisions. To address these issues, we introduce ZeroGrasp, a novel framework that simultaneously performs 3D reconstruction and grasp pose prediction in near real-time. A key insight of our method is that occlusion reasoning and modeling the spatial relationships between objects is beneficial for both accurate reconstruction and grasping. We couple our method with a novel large-scale synthetic dataset, which comprises 1M photo-realistic images, high-resolution 3D reconstructions and 11.3B physically-valid grasp pose annotations for 12K objects from the Objaverse-LVIS dataset. We evaluate ZeroGrasp on the GraspNet-1B benchmark as well as through real-world robot experiments. ZeroGrasp achieves state-of-the-art performance and generalizes to novel real-world objects by leveraging synthetic data.
comment: Published at CVPR 2025, Webpage: https://sh8.io/#/zerograsp
☆ Hallucination-Aware Generative Pretrained Transformer for Cooperative Aerial Mobility Control
This paper proposes SafeGPT, a two-tiered framework that integrates generative pretrained transformers (GPTs) with reinforcement learning (RL) for efficient and reliable unmanned aerial vehicle (UAV) last-mile deliveries. In the proposed design, a Global GPT module assigns high-level tasks such as sector allocation, while an On-Device GPT manages real-time local route planning. An RL-based safety filter monitors each GPT decision and overrides unsafe actions that could lead to battery depletion or duplicate visits, effectively mitigating hallucinations. Furthermore, a dual replay buffer mechanism helps both the GPT modules and the RL agent refine their strategies over time. Simulation results demonstrate that SafeGPT achieves higher delivery success rates compared to a GPT-only baseline, while substantially reducing battery consumption and travel distance. These findings validate the efficacy of combining GPT-based semantic reasoning with formal safety guarantees, contributing a viable solution for robust and energy-efficient UAV logistics.
☆ Following Is All You Need: Robot Crowd Navigation Using People As Planners
Navigating in crowded environments requires the robot to be equipped with high-level reasoning and planning techniques. Existing works focus on developing complex and heavyweight planners while ignoring the role of human intelligence. Since humans are highly capable agents who are also widely available in a crowd navigation setting, we propose an alternative scheme where the robot utilises people as planners to benefit from their effective planning decisions and social behaviours. Through a set of rule-based evaluations, we identify suitable human leaders who exhibit the potential to guide the robot towards its goal. Using a simple base planner, the robot follows the selected leader through shorthorizon subgoals that are designed to be straightforward to achieve. We demonstrate through both simulated and real-world experiments that our novel framework generates safe and efficient robot plans compared to existing planners, even without predictive or data-driven modules. Our method also brings human-like robot behaviours without explicitly defining traffic rules and social norms. Code will be available at https://github.com/centiLinda/PeopleAsPlanner.git.
☆ E2E Parking Dataset: An Open Benchmark for End-to-End Autonomous Parking
End-to-end learning has shown great potential in autonomous parking, yet the lack of publicly available datasets limits reproducibility and benchmarking. While prior work introduced a visual-based parking model and a pipeline for data generation, training, and close-loop test, the dataset itself was not released. To bridge this gap, we create and open-source a high-quality dataset for end-to-end autonomous parking. Using the original model, we achieve an overall success rate of 85.16% with lower average position and orientation errors (0.24 meters and 0.34 degrees).
☆ ATLASv2: LLM-Guided Adaptive Landmark Acquisition and Navigation on the Edge
Autonomous systems deployed on edge devices face significant challenges, including resource constraints, real-time processing demands, and adapting to dynamic environments. This work introduces ATLASv2, a novel system that integrates a fine-tuned TinyLLM, real-time object detection, and efficient path planning to enable hierarchical, multi-task navigation and manipulation all on the edge device, Jetson Nano. ATLASv2 dynamically expands its navigable landmarks by detecting and localizing objects in the environment which are saved to its internal knowledge base to be used for future task execution. We evaluate ATLASv2 in real-world environments, including a handcrafted home and office setting constructed with diverse objects and landmarks. Results show that ATLASv2 effectively interprets natural language instructions, decomposes them into low-level actions, and executes tasks with high success rates. By leveraging generative AI in a fully on-board framework, ATLASv2 achieves optimized resource utilization with minimal prompting latency and power consumption, bridging the gap between simulated environments and real-world applications.
☆ Superfast Configuration-Space Convex Set Computation on GPUs for Online Motion Planning
In this work, we leverage GPUs to construct probabilistically collision-free convex sets in robot configuration space on the fly. This extends the use of modern motion planning algorithms that leverage such representations to changing environments. These planners rapidly and reliably optimize high-quality trajectories, without the burden of challenging nonconvex collision-avoidance constraints. We present an algorithm that inflates collision-free piecewise linear paths into sequences of convex sets (SCS) that are probabilistically collision-free using massive parallelism. We then integrate this algorithm into a motion planning pipeline, which leverages dynamic roadmaps to rapidly find one or multiple collision-free paths, and inflates them. We then optimize the trajectory through the probabilistically collision-free sets, simultaneously using the candidate trajectory to detect and remove collisions from the sets. We demonstrate the efficacy of our approach on a simulation benchmark and a KUKA iiwa 7 robot manipulator with perception in the loop. On our benchmark, our approach runs 17.1 times faster and yields a 27.9% increase in reliability over the nonlinear trajectory optimization baseline, while still producing high-quality motion plans.
comment: 14 pages, 7 figures
☆ Linearity, Time Invariance, and Passivity of a Novice Person in Human Teleoperation
Low-cost teleguidance of medical procedures is becoming essential to provide healthcare to remote and underserved communities. Human teleoperation is a promising new method for guiding a novice person with relatively high precision and efficiency through a mixed reality (MR) interface. Prior work has shown that the novice, or "follower", can reliably track the MR input with performance not unlike a telerobotic system. As a consequence, it is of interest to understand and control the follower's dynamics to optimize the system performance and permit stable and transparent bilateral teleoperation. To this end, linearity, time-invariance, inter-axis coupling, and passivity are important in teleoperation and controller design. This paper therefore explores these effects with regard to the follower person in human teleoperation. It is demonstrated through modeling and experiments that the follower can indeed be treated as approximately linear and time invariant, with little coupling and a large excess of passivity at practical frequencies. Furthermore, a stochastic model of the follower dynamics is derived. These results will permit controller design and analysis to improve the performance of human teleoperation.
☆ Doppler-SLAM: Doppler-Aided Radar-Inertial and LiDAR-Inertial Simultaneous Localization and Mapping
Simultaneous localization and mapping (SLAM) is a critical capability for autonomous systems. Traditional SLAM approaches, which often rely on visual or LiDAR sensors, face significant challenges in adverse conditions such as low light or featureless environments. To overcome these limitations, we propose a novel Doppler-aided radar-inertial and LiDAR-inertial SLAM framework that leverages the complementary strengths of 4D radar, FMCW LiDAR, and inertial measurement units. Our system integrates Doppler velocity measurements and spatial data into a tightly-coupled front-end and graph optimization back-end to provide enhanced ego velocity estimation, accurate odometry, and robust mapping. We also introduce a Doppler-based scan-matching technique to improve front-end odometry in dynamic environments. In addition, our framework incorporates an innovative online extrinsic calibration mechanism, utilizing Doppler velocity and loop closure to dynamically maintain sensor alignment. Extensive evaluations on both public and proprietary datasets show that our system significantly outperforms state-of-the-art radar-SLAM and LiDAR-SLAM frameworks in terms of accuracy and robustness. To encourage further research, the code of our Doppler-SLAM and our dataset are available at: https://github.com/Wayne-DWA/Doppler-SLAM.
comment: 8 pages, 7 figures
☆ RESPLE: Recursive Spline Estimation for LiDAR-Based Odometry
We present a novel recursive Bayesian estimation framework for continuous-time six-DoF dynamic motion estimation using B-splines. The state vector consists of a recurrent set of position control points and orientation control point increments, enabling a straightforward modification of the iterated extended Kalman filter without involving the error-state formulation. The resulting recursive spline estimator (RESPLE) provides a versatile, pragmatic and lightweight solution for motion estimation and is further exploited for direct LiDAR-based odometry, supporting integration of one or multiple LiDARs and an IMU. We conduct extensive real-world benchmarking based on public datasets and own experiments, covering aerial, wheeled, legged, and wearable platforms operating in indoor, urban, wild environments with diverse LiDARs. RESPLE-based solutions achieve superior estimation accuracy and robustness over corresponding state-of-the-art systems, while attaining real-time performance. Notably, our LiDAR-only variant outperforms existing LiDAR-inertial systems in scenarios without significant LiDAR degeneracy, and showing further improvements when additional LiDAR and inertial sensors are incorporated for more challenging conditions. We release the source code and own experimental datasets at https://github.com/ASIG-X/RESPLE .
☆ LANGTRAJ: Diffusion Model and Dataset for Language-Conditioned Trajectory Simulation
Evaluating autonomous vehicles with controllability enables scalable testing in counterfactual or structured settings, enhancing both efficiency and safety. We introduce LangTraj, a language-conditioned scene-diffusion model that simulates the joint behavior of all agents in traffic scenarios. By conditioning on natural language inputs, LangTraj provides flexible and intuitive control over interactive behaviors, generating nuanced and realistic scenarios. Unlike prior approaches that depend on domain-specific guidance functions, LangTraj incorporates language conditioning during training, facilitating more intuitive traffic simulation control. We propose a novel closed-loop training strategy for diffusion models, explicitly tailored to enhance stability and realism during closed-loop simulation. To support language-conditioned simulation, we develop Inter-Drive, a large-scale dataset with diverse and interactive labels for training language-conditioned diffusion models. Our dataset is built upon a scalable pipeline for annotating agent-agent interactions and single-agent behaviors, ensuring rich and varied supervision. Validated on the Waymo Motion Dataset, LangTraj demonstrates strong performance in realism, language controllability, and language-conditioned safety-critical simulation, establishing a new paradigm for flexible and scalable autonomous vehicle testing.
comment: Dataset and project website in preparation
☆ Enhancing Autonomous Driving Systems with On-Board Deployed Large Language Models
Neural Networks (NNs) trained through supervised learning struggle with managing edge-case scenarios common in real-world driving due to the intractability of exhaustive datasets covering all edge-cases, making knowledge-driven approaches, akin to how humans intuitively detect unexpected driving behavior, a suitable complement to data-driven methods. This work proposes a hybrid architecture combining low-level Model Predictive Controller (MPC) with locally deployed Large Language Models (LLMs) to enhance decision-making and Human Machine Interaction (HMI). The DecisionxLLM module evaluates robotic state information against natural language instructions to ensure adherence to desired driving behavior. The MPCxLLM module then adjusts MPC parameters based on LLM-generated insights, achieving control adaptability while preserving the safety and constraint guarantees of traditional MPC systems. Further, to enable efficient on-board deployment and to eliminate dependency on cloud connectivity, we shift processing to the on-board computing platform: We propose an approach that exploits Retrieval Augmented Generation (RAG), Low Rank Adaptation (LoRA) fine-tuning, and quantization. Experimental results demonstrate that these enhancements yield significant improvements in reasoning accuracy by up to 10.45%, control adaptability by as much as 52.2%, and up to 10.5x increase in computational efficiency (tokens/s), validating the proposed framework's practicality for real-time deployment even on down-scaled robotic platforms. This work bridges high-level decision-making with low-level control adaptability, offering a synergistic framework for knowledge-driven and adaptive Autonomous Driving Systems (ADS).
☆ Cross-cultural Deployment of Autonomous Vehicles Using Data-light Inverse Reinforcement Learning
More than the adherence to specific traffic regulations, driving culture touches upon a more implicit part - an informal, conventional, collective behavioral pattern followed by drivers - that varies across countries, regions, and even cities. Such cultural divergence has become one of the biggest challenges in deploying autonomous vehicles (AVs) across diverse regions today. The current emergence of data-driven methods has shown a potential solution to enable culture-compatible driving through learning from data, but what if some underdeveloped regions cannot provide sufficient local data to inform driving culture? This issue is particularly significant for a broader global AV market. Here, we propose a cross-cultural deployment scheme for AVs, called data-light inverse reinforcement learning, designed to re-calibrate culture-specific AVs and assimilate them into other cultures. First, we report the divergence in driving cultures through a comprehensive comparative analysis of naturalistic driving datasets on highways from three countries: Germany, China, and the USA. Then, we demonstrate the effectiveness of our scheme by testing the expeditious cross-cultural deployment across these three countries, with cumulative testing mileage of over 56084 km. The performance is particularly advantageous when cross-cultural deployment is carried out without affluent local data. Results show that we can reduce the dependence on local data by a margin of 98.67% at best. This study is expected to bring a broader, fairer AV global market, particularly in those regions that lack enough local data to develop culture-compatible AVs.
♻ ☆ Train Robots in a JIF: Joint Inverse and Forward Dynamics with Human and Robot Demonstrations
Pre-training on large datasets of robot demonstrations is a powerful technique for learning diverse manipulation skills but is often limited by the high cost and complexity of collecting robot-centric data, especially for tasks requiring tactile feedback. This work addresses these challenges by introducing a novel method for pre-training with multi-modal human demonstrations. Our approach jointly learns inverse and forward dynamics to extract latent state representations, towards learning manipulation specific representations. This enables efficient fine-tuning with only a small number of robot demonstrations, significantly improving data efficiency. Furthermore, our method allows for the use of multi-modal data, such as combination of vision and touch for manipulation. By leveraging latent dynamics modeling and tactile sensing, this approach paves the way for scalable robot manipulation learning based on human demonstrations.
comment: 9 pages, 8 figures, submission to RSS 2025
♻ ☆ RoboComm: A DID-based scalable and privacy-preserving Robot-to-Robot interaction over state channels
In a multi robot system establishing trust amongst untrusted robots from different organisations while preserving a robot's privacy is a challenge. Recently decentralized technologies such as smart contract and blockchain are being explored for applications in robotics. However, the limited transaction processing and high maintenance cost hinder the widespread adoption of such approaches. Moreover, blockchain transactions be they on public or private permissioned blockchain are publically readable which further fails to preserve the confidentiality of the robot's data and privacy of the robot. In this work, we propose RoboComm a Decentralized Identity based approach for privacy-preserving interaction between robots. With DID a component of Self-Sovereign Identity; robots can authenticate each other independently without relying on any third-party service. Verifiable Credentials enable private data associated with a robot to be stored within the robot's hardware, unlike existing blockchain based approaches where the data has to be on the blockchain. We improve throughput by allowing message exchange over state channels. Being a blockchain backed solution RoboComm provides a trustworthy system without relying on a single party. Moreover, we implement our proposed approach to demonstrate the feasibility of our solution.
comment: resolved overlapping text from another article
♻ ☆ HAS-RRT: RRT-based Motion Planning using Topological Guidance
We present a hierarchical RRT-based motion planning strategy, Hierarchical Annotated-Skeleton Guided RRT (HAS-RRT), guided by a workspace skeleton, to solve motion planning problems. HAS-RRT provides up to a 91% runtime reduction and builds a tree at least 30% smaller than competitors while still finding competitive-cost paths. This is because our strategy prioritizes paths indicated by the workspace guidance to efficiently find a valid motion plan for the robot. Existing methods either rely too heavily on workspace guidance or have difficulty finding narrow passages. By taking advantage of the assumptions that the workspace skeleton provides, HAS-RRT is able to build a smaller tree and find a path faster than its competitors. Additionally, we show that HAS-RRT is robust to the quality of workspace guidance provided and that, in a worst-case scenario where the workspace skeleton provides no additional insight, our method performs comparably to an unguided method.
comment: 8 pages; Accepted at RA-L, April 2025
♻ ☆ "A Good Bot Always Knows Its Limitations": Assessing Autonomous System Decision-making Competencies through Factorized Machine Self-confidence
How can intelligent machines assess their competency to complete a task? This question has come into focus for autonomous systems that algorithmically make decisions under uncertainty. We argue that machine self-confidence -- a form of meta-reasoning based on self-assessments of system knowledge about the state of the world, itself, and ability to reason about and execute tasks -- leads to many computable and useful competency indicators for such agents. This paper presents our body of work, so far, on this concept in the form of the Factorized Machine Self-confidence (FaMSeC) framework, which holistically considers several major factors driving competency in algorithmic decision-making: outcome assessment, solver quality, model quality, alignment quality, and past experience. In FaMSeC, self-confidence indicators are derived via 'problem-solving statistics' embedded in Markov decision process solvers and related approaches. These statistics come from evaluating probabilistic exceedance margins in relation to certain outcomes and associated competency standards specified by an evaluator. Once designed, and evaluated, the statistics can be easily incorporated into autonomous agents and serve as indicators of competency. We include detailed descriptions and examples for Markov decision process agents, and show how outcome assessment and solver quality factors can be found for a range of tasking contexts through novel use of meta-utility functions, behavior simulations, and surrogate prediction models. Numerical evaluations are performed to demonstrate that FaMSeC indicators perform as desired (references to human subject studies beyond the scope of this paper are provided).
comment: 63 pages, 22 figures, version accepted to ACM THRI
♻ ☆ GarmentTracking: Category-Level Garment Pose Tracking CVPR 2023
Garments are important to humans. A visual system that can estimate and track the complete garment pose can be useful for many downstream tasks and real-world applications. In this work, we present a complete package to address the category-level garment pose tracking task: (1) A recording system VR-Garment, with which users can manipulate virtual garment models in simulation through a VR interface. (2) A large-scale dataset VR-Folding, with complex garment pose configurations in manipulation like flattening and folding. (3) An end-to-end online tracking framework GarmentTracking, which predicts complete garment pose both in canonical space and task space given a point cloud sequence. Extensive experiments demonstrate that the proposed GarmentTracking achieves great performance even when the garment has large non-rigid deformation. It outperforms the baseline approach on both speed and accuracy. We hope our proposed solution can serve as a platform for future research. Codes and datasets are available in https://garment-tracking.robotflow.ai.
comment: CVPR 2023
♻ ☆ MonoRollBot: 3-DOF Spherical Robot with Underactuated Single Compliant Actuator Design
Spherical rolling robots have garnered significant attention in the field of mobile robotics for applications such as inspection and space exploration. Designing underactuated rolling robots poses challenges in achieving multi-directional propulsion with high degrees of freedom while utilizing a limited number of actuators. This paper presents the MonoRollBot, a novel 3-degree-of-freedom (DOF) spherical robot that utilizes an underactuated mechanism driven by only a single spring-motor system. Unlike conventional spherical robots, MonoRollBot employs a minimalist actuation approach, relying on only one motor and a passive spring to control its locomotion. The robot achieves 3-DOF motion through an innovative coupling of spring dynamics and motor control. In this work, we detail the design of the MonoRollBot and evaluate its motion capabilities through design studies. We also do studies on its locomotion behaviours based on changes in rotating mass and stiffness properties.
comment: 6 pages, 11 figures, accepted at IEEE RoboSoft 2025
♻ ☆ PIP-Loco: A Proprioceptive Infinite Horizon Planning Framework for Quadrupedal Robot Locomotion ICRA
A core strength of Model Predictive Control (MPC) for quadrupedal locomotion has been its ability to enforce constraints and provide interpretability of the sequence of commands over the horizon. However, despite being able to plan, MPC struggles to scale with task complexity, often failing to achieve robust behavior on rapidly changing surfaces. On the other hand, model-free Reinforcement Learning (RL) methods have outperformed MPC on multiple terrains, showing emergent motions but inherently lack any ability to handle constraints or perform planning. To address these limitations, we propose a framework that integrates proprioceptive planning with RL, allowing for agile and safe locomotion behaviors through the horizon. Inspired by MPC, we incorporate an internal model that includes a velocity estimator and a Dreamer module. During training, the framework learns an expert policy and an internal model that are co-dependent, facilitating exploration for improved locomotion behaviors. During deployment, the Dreamer module solves an infinite-horizon MPC problem, adapting actions and velocity commands to respect the constraints. We validate the robustness of our training framework through ablation studies on internal model components and demonstrate improved robustness to training noise. Finally, we evaluate our approach across multi-terrain scenarios in both simulation and hardware.
comment: Accepted at IEEE International Conference on Robotics and Automation (ICRA) 2025
♻ ☆ Reasoning in visual navigation of end-to-end trained agents: a dynamical systems approach
Progress in Embodied AI has made it possible for end-to-end-trained agents to navigate in photo-realistic environments with high-level reasoning and zero-shot or language-conditioned behavior, but benchmarks are still dominated by simulation. In this work, we focus on the fine-grained behavior of fast-moving real robots and present a large-scale experimental study involving \numepisodes{} navigation episodes in a real environment with a physical robot, where we analyze the type of reasoning emerging from end-to-end training. In particular, we study the presence of realistic dynamics which the agent learned for open-loop forecasting, and their interplay with sensing. We analyze the way the agent uses latent memory to hold elements of the scene structure and information gathered during exploration. We probe the planning capabilities of the agent, and find in its memory evidence for somewhat precise plans over a limited horizon. Furthermore, we show in a post-hoc analysis that the value function learned by the agent relates to long-term planning. Put together, our experiments paint a new picture on how using tools from computer vision and sequential decision making have led to new capabilities in robotics and control. An interactive tool is available at europe.naverlabs.com/research/publications/reasoning-in-visual-navigation-of-end-to-end-trained-agents.
♻ ☆ LanguageMPC: Large Language Models as Decision Makers for Autonomous Driving
Existing learning-based autonomous driving (AD) systems face challenges in comprehending high-level information, generalizing to rare events, and providing interpretability. To address these problems, this work employs Large Language Models (LLMs) as a decision-making component for complex AD scenarios that require human commonsense understanding. We devise cognitive pathways to enable comprehensive reasoning with LLMs, and develop algorithms for translating LLM decisions into actionable driving commands. Through this approach, LLM decisions are seamlessly integrated with low-level controllers by guided parameter matrix adaptation. Extensive experiments demonstrate that our proposed method not only consistently surpasses baseline approaches in single-vehicle tasks, but also helps handle complex driving behaviors even multi-vehicle coordination, thanks to the commonsense reasoning capabilities of LLMs. This paper presents an initial step toward leveraging LLMs as effective decision-makers for intricate AD scenarios in terms of safety, efficiency, generalizability, and interoperability. We aspire for it to serve as inspiration for future research in this field. Project page: https://sites.google.com/view/llm-mpc
♻ ☆ Constraint-Aware Zero-Shot Vision-Language Navigation in Continuous Environments
We address the task of Vision-Language Navigation in Continuous Environments (VLN-CE) under the zero-shot setting. Zero-shot VLN-CE is particularly challenging due to the absence of expert demonstrations for training and minimal environment structural prior to guide navigation. To confront these challenges, we propose a Constraint-Aware Navigator (CA-Nav), which reframes zero-shot VLN-CE as a sequential, constraint-aware sub-instruction completion process. CA-Nav continuously translates sub-instructions into navigation plans using two core modules: the Constraint-Aware Sub-instruction Manager (CSM) and the Constraint-Aware Value Mapper (CVM). CSM defines the completion criteria for decomposed sub-instructions as constraints and tracks navigation progress by switching sub-instructions in a constraint-aware manner. CVM, guided by CSM's constraints, generates a value map on the fly and refines it using superpixel clustering to improve navigation stability. CA-Nav achieves the state-of-the-art performance on two VLN-CE benchmarks, surpassing the previous best method by 12 percent and 13 percent in Success Rate on the validation unseen splits of R2R-CE and RxR-CE, respectively. Moreover, CA-Nav demonstrates its effectiveness in real-world robot deployments across various indoor scenes and instructions.
♻ ☆ Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework
Multimodal foundation models offer a promising framework for robotic perception and planning by processing sensory inputs to generate actionable plans. However, addressing uncertainty in both perception (sensory interpretation) and decision-making (plan generation) remains a critical challenge for ensuring task reliability. We present a comprehensive framework to disentangle, quantify, and mitigate these two forms of uncertainty. We first introduce a framework for uncertainty disentanglement, isolating perception uncertainty arising from limitations in visual understanding and decision uncertainty relating to the robustness of generated plans. To quantify each type of uncertainty, we propose methods tailored to the unique properties of perception and decision-making: we use conformal prediction to calibrate perception uncertainty and introduce Formal-Methods-Driven Prediction (FMDP) to quantify decision uncertainty, leveraging formal verification techniques for theoretical guarantees. Building on this quantification, we implement two targeted intervention mechanisms: an active sensing process that dynamically re-observes high-uncertainty scenes to enhance visual input quality and an automated refinement procedure that fine-tunes the model on high-certainty data, improving its capability to meet task specifications. Empirical validation in real-world and simulated robotic tasks demonstrates that our uncertainty disentanglement framework reduces variability by up to 40% and enhances task success rates by 5% compared to baselines. These improvements are attributed to the combined effect of both interventions and highlight the importance of uncertainty disentanglement, which facilitates targeted interventions that enhance the robustness and reliability of autonomous systems. Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/.
comment: Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/
Artificial Intelligence 162
☆ DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoning
The capacity for complex mathematical reasoning is a key benchmark for artificial intelligence. While reinforcement learning (RL) applied to LLMs shows promise, progress is significantly hindered by the lack of large-scale training data that is sufficiently challenging, possesses verifiable answer formats suitable for RL, and is free from contamination with evaluation benchmarks. To address these limitations, we introduce DeepMath-103K, a new, large-scale dataset comprising approximately 103K mathematical problems, specifically designed to train advanced reasoning models via RL. DeepMath-103K is curated through a rigorous pipeline involving source analysis, stringent decontamination against numerous benchmarks, and filtering for high difficulty (primarily Levels 5-9), significantly exceeding existing open resources in challenge. Each problem includes a verifiable final answer, enabling rule-based RL, and three distinct R1-generated solutions suitable for diverse training paradigms like supervised fine-tuning or distillation. Spanning a wide range of mathematical topics, DeepMath-103K promotes the development of generalizable reasoning. We demonstrate that models trained on DeepMath-103K achieve significant improvements on challenging mathematical benchmarks, validating its effectiveness. We release DeepMath-103K publicly to facilitate community progress in building more capable AI reasoning systems: https://github.com/zwhe99/DeepMath.
comment: WIP
☆ Elucidating the Design Space of Multimodal Protein Language Models
Multimodal protein language models (PLMs) integrate sequence and token-based structural information, serving as a powerful foundation for protein modeling, generation, and design. However, the reliance on tokenizing 3D structures into discrete tokens causes substantial loss of fidelity about fine-grained structural details and correlations. In this paper, we systematically elucidate the design space of multimodal PLMs to overcome their limitations. We identify tokenization loss and inaccurate structure token predictions by the PLMs as major bottlenecks. To address these, our proposed design space covers improved generative modeling, structure-aware architectures and representation learning, and data exploration. Our advancements approach finer-grained supervision, demonstrating that token-based multimodal PLMs can achieve robust structural modeling. The effective design methods dramatically improve the structure generation diversity, and notably, folding abilities of our 650M model by reducing the RMSD from 5.52 to 2.36 on PDB testset, even outperforming 3B baselines and on par with the specialized folding models.
comment: Project Page: https://bytedance.github.io/dplm/dplm-2.1/
☆ A Clean Slate for Offline Reinforcement Learning
Progress in offline reinforcement learning (RL) has been impeded by ambiguous problem definitions and entangled algorithmic designs, resulting in inconsistent implementations, insufficient ablations, and unfair evaluations. Although offline RL explicitly avoids environment interaction, prior methods frequently employ extensive, undocumented online evaluation for hyperparameter tuning, complicating method comparisons. Moreover, existing reference implementations differ significantly in boilerplate code, obscuring their core algorithmic contributions. We address these challenges by first introducing a rigorous taxonomy and a transparent evaluation protocol that explicitly quantifies online tuning budgets. To resolve opaque algorithmic design, we provide clean, minimalistic, single-file implementations of various model-free and model-based offline RL methods, significantly enhancing clarity and achieving substantial speed-ups. Leveraging these streamlined implementations, we propose Unifloral, a unified algorithm that encapsulates diverse prior approaches within a single, comprehensive hyperparameter space, enabling algorithm development in a shared hyperparameter space. Using Unifloral with our rigorous evaluation protocol, we develop two novel algorithms - TD3-AWR (model-free) and MoBRAC (model-based) - which substantially outperform established baselines. Our implementation is publicly available at https://github.com/EmptyJackson/unifloral.
☆ TextArena
TextArena is an open-source collection of competitive text-based games for training and evaluation of agentic behavior in Large Language Models (LLMs). It spans 57+ unique environments (including single-player, two-player, and multi-player setups) and allows for easy evaluation of model capabilities via an online-play system (against humans and other submitted models) with real-time TrueSkill scores. Traditional benchmarks rarely assess dynamic social skills such as negotiation, theory of mind, and deception, creating a gap that TextArena addresses. Designed with research, community and extensibility in mind, TextArena emphasizes ease of adding new games, adapting the framework, testing models, playing against the models, and training models. Detailed documentation of environments, games, leaderboard, and examples are available on https://github.com/LeonGuertler/TextArena and https://www.textarena.ai/.
comment: work in progress; 5 pages, 3 figures
☆ Greedy Restart Schedules: A Baseline for Dynamic Algorithm Selection on Numerical Black-box Optimization Problems GECCO 2025
In many optimization domains, there are multiple different solvers that contribute to the overall state-of-the-art, each performing better on some, and worse on other types of problem instances. Meta-algorithmic approaches, such as instance-based algorithm selection, configuration and scheduling, aim to close this gap by extracting the most performance possible from a set of (configurable) optimizers. In this context, the best performing individual algorithms are often hand-crafted hybrid heuristics which perform many restarts of fast local optimization approaches. However, data-driven techniques to create optimized restart schedules have not yet been extensively studied. Here, we present a simple scheduling approach that iteratively selects the algorithm performing best on the distribution of unsolved training problems at time of selection, resulting in a problem-independent solver schedule. We demonstrate our approach using well-known optimizers from numerical black-box optimization on the BBOB testbed, bridging much of the gap between single and virtual best solver from the original portfolio across various evaluation protocols. Our greedy restart schedule presents a powerful baseline for more complex dynamic algorithm selection models.
comment: Author version. Accepted as full paper to be presented at the GECCO 2025 conference, July 14-18, M\'alaga, Spain. (DOI 10.1145/3712256.3726408)
☆ Masculine Defaults via Gendered Discourse in Podcasts and Large Language Models
Masculine defaults are widely recognized as a significant type of gender bias, but they are often unseen as they are under-researched. Masculine defaults involve three key parts: (i) the cultural context, (ii) the masculine characteristics or behaviors, and (iii) the reward for, or simply acceptance of, those masculine characteristics or behaviors. In this work, we study discourse-based masculine defaults, and propose a twofold framework for (i) the large-scale discovery and analysis of gendered discourse words in spoken content via our Gendered Discourse Correlation Framework (GDCF); and (ii) the measurement of the gender bias associated with these gendered discourse words in LLMs via our Discourse Word-Embedding Association Test (D-WEAT). We focus our study on podcasts, a popular and growing form of social media, analyzing 15,117 podcast episodes. We analyze correlations between gender and discourse words -- discovered via LDA and BERTopic -- to automatically form gendered discourse word lists. We then study the prevalence of these gendered discourse words in domain-specific contexts, and find that gendered discourse-based masculine defaults exist in the domains of business, technology/politics, and video games. Next, we study the representation of these gendered discourse words from a state-of-the-art LLM embedding model from OpenAI, and find that the masculine discourse words have a more stable and robust representation than the feminine discourse words, which may result in better system performance on downstream tasks for men. Hence, men are rewarded for their discourse patterns with better system performance by one of the state-of-the-art language models -- and this embedding disparity is a representational harm and a masculine default.
comment: To appear in ICWSM 2025
☆ A Dual-Space Framework for General Knowledge Distillation of Large Language Models
Knowledge distillation (KD) is a promising solution to compress large language models (LLMs) by transferring their knowledge to smaller models. During this process, white-box KD methods usually minimize the distance between the output distributions of the teacher model and the student model to transfer more information. However, we reveal that the current white-box KD framework exhibits two limitations: a) bridging probability distributions from different output spaces will limit the similarity between the teacher model and the student model; b) this framework cannot be applied to LLMs with different vocabularies. One of the root causes for these limitations is that the distributions from the teacher and the student for KD are output by different prediction heads, which yield distributions in different output spaces and dimensions. Therefore, in this paper, we propose a dual-space knowledge distillation (DSKD) framework that unifies the prediction heads of the teacher and the student models for KD. Specifically, we first introduce two projectors with ideal initialization to project the teacher/student hidden states into the student/teacher representation spaces. After this, the hidden states from different models can share the same head and unify the output spaces of the distributions. Furthermore, we develop an exact token alignment (ETA) algorithm to align the same tokens in two differently-tokenized sequences. Based on the above, our DSKD framework is a general KD framework that supports both off-policy and on-policy KD, and KD between any two LLMs regardless of their vocabularies. Extensive experiments on instruction-following, mathematical reasoning, and code generation benchmarks show that DSKD significantly outperforms existing methods based on the current white-box KD framework and surpasses other cross-tokenizer KD methods for LLMs with different vocabularies.
comment: 19 pages, 9 figures, 11 tables, under review. Code is available at: https://github.com/songmzhang/DSKDv2. arXiv admin note: text overlap with arXiv:2406.17328
☆ ADT: Tuning Diffusion Models with Adversarial Supervision
Diffusion models have achieved outstanding image generation by reversing a forward noising process to approximate true data distributions. During training, these models predict diffusion scores from noised versions of true samples in a single forward pass, while inference requires iterative denoising starting from white noise. This training-inference divergences hinder the alignment between inference and training data distributions, due to potential prediction biases and cumulative error accumulation. To address this problem, we propose an intuitive but effective fine-tuning framework, called Adversarial Diffusion Tuning (ADT), by stimulating the inference process during optimization and aligning the final outputs with training data by adversarial supervision. Specifically, to achieve robust adversarial training, ADT features a siamese-network discriminator with a fixed pre-trained backbone and lightweight trainable parameters, incorporates an image-to-image sampling strategy to smooth discriminative difficulties, and preserves the original diffusion loss to prevent discriminator hacking. In addition, we carefully constrain the backward-flowing path for back-propagating gradients along the inference path without incurring memory overload or gradient explosion. Finally, extensive experiments on Stable Diffusion models (v1.5, XL, and v3), demonstrate that ADT significantly improves both distribution alignment and image quality.
☆ Embodied World Models Emerge from Navigational Task in Open-Ended Environments
Understanding how artificial systems can develop spatial awareness and reasoning has long been a challenge in AI research. Traditional models often rely on passive observation, but embodied cognition theory suggests that deeper understanding emerges from active interaction with the environment. This study investigates whether neural networks can autonomously internalize spatial concepts through interaction, focusing on planar navigation tasks. Using Gated Recurrent Units (GRUs) combined with Meta-Reinforcement Learning (Meta-RL), we show that agents can learn to encode spatial properties like direction, distance, and obstacle avoidance. We introduce Hybrid Dynamical Systems (HDS) to model the agent-environment interaction as a closed dynamical system, revealing stable limit cycles that correspond to optimal navigation strategies. Ridge Representation allows us to map navigation paths into a fixed-dimensional behavioral space, enabling comparison with neural states. Canonical Correlation Analysis (CCA) confirms strong alignment between these representations, suggesting that the agent's neural states actively encode spatial knowledge. Intervention experiments further show that specific neural dimensions are causally linked to navigation performance. This work provides an approach to bridging the gap between action and perception in AI, offering new insights into building adaptive, interpretable models that can generalize across complex environments. The causal validation of neural representations also opens new avenues for understanding and controlling the internal mechanisms of AI systems, pushing the boundaries of how machines learn and reason in dynamic, real-world scenarios.
comment: Research on explainable meta-reinforcement learning AI
☆ Measures of Variability for Risk-averse Policy Gradient
Risk-averse reinforcement learning (RARL) is critical for decision-making under uncertainty, which is especially valuable in high-stake applications. However, most existing works focus on risk measures, e.g., conditional value-at-risk (CVaR), while measures of variability remain underexplored. In this paper, we comprehensively study nine common measures of variability, namely Variance, Gini Deviation, Mean Deviation, Mean-Median Deviation, Standard Deviation, Inter-Quantile Range, CVaR Deviation, Semi_Variance, and Semi_Standard Deviation. Among them, four metrics have not been previously studied in RARL. We derive policy gradient formulas for these unstudied metrics, improve gradient estimation for Gini Deviation, analyze their gradient properties, and incorporate them with the REINFORCE and PPO frameworks to penalize the dispersion of returns. Our empirical study reveals that variance-based metrics lead to unstable policy updates. In contrast, CVaR Deviation and Gini Deviation show consistent performance across different randomness and evaluation domains, achieving high returns while effectively learning risk-averse policies. Mean Deviation and Semi_Standard Deviation are also competitive across different scenarios. This work provides a comprehensive overview of variability measures in RARL, offering practical insights for risk-aware decision-making and guiding future research on risk metrics and RARL algorithms.
☆ Multi-level Cellular Automata for FLIM networks
The necessity of abundant annotated data and complex network architectures presents a significant challenge in deep-learning Salient Object Detection (deep SOD) and across the broader deep-learning landscape. This challenge is particularly acute in medical applications in developing countries with limited computational resources. Combining modern and classical techniques offers a path to maintaining competitive performance while enabling practical applications. Feature Learning from Image Markers (FLIM) methodology empowers experts to design convolutional encoders through user-drawn markers, with filters learned directly from these annotations. Recent findings demonstrate that coupling a FLIM encoder with an adaptive decoder creates a flyweight network suitable for SOD, requiring significantly fewer parameters than lightweight models and eliminating the need for backpropagation. Cellular Automata (CA) methods have proven successful in data-scarce scenarios but require proper initialization -- typically through user input, priors, or randomness. We propose a practical intersection of these approaches: using FLIM networks to initialize CA states with expert knowledge without requiring user interaction for each image. By decoding features from each level of a FLIM network, we can initialize multiple CAs simultaneously, creating a multi-level framework. Our method leverages the hierarchical knowledge encoded across different network layers, merging multiple saliency maps into a high-quality final output that functions as a CA ensemble. Benchmarks across two challenging medical datasets demonstrate the competitiveness of our multi-level CA approach compared to established models in the deep SOD literature.
☆ VideoPanda: Video Panoramic Diffusion with Multi-view Attention
High resolution panoramic video content is paramount for immersive experiences in Virtual Reality, but is non-trivial to collect as it requires specialized equipment and intricate camera setups. In this work, we introduce VideoPanda, a novel approach for synthesizing 360$^\circ$ videos conditioned on text or single-view video data. VideoPanda leverages multi-view attention layers to augment a video diffusion model, enabling it to generate consistent multi-view videos that can be combined into immersive panoramic content. VideoPanda is trained jointly using two conditions: text-only and single-view video, and supports autoregressive generation of long-videos. To overcome the computational burden of multi-view video generation, we randomly subsample the duration and camera views used during training and show that the model is able to gracefully generalize to generating more frames during inference. Extensive evaluations on both real-world and synthetic video datasets demonstrate that VideoPanda generates more realistic and coherent 360$^\circ$ panoramas across all input conditions compared to existing methods. Visit the project website at https://research-staging.nvidia.com/labs/toronto-ai/VideoPanda/ for results.
comment: Project website at https://research-staging.nvidia.com/labs/toronto-ai/VideoPanda/
☆ Trajectory Encoding Temporal Graph Networks
Temporal Graph Networks (TGNs) have demonstrated significant success in dynamic graph tasks such as link prediction and node classification. Both tasks comprise transductive settings, where the model predicts links among known nodes, and in inductive settings, where it generalises learned patterns to previously unseen nodes. Existing TGN designs face a dilemma under these dual scenarios. Anonymous TGNs, which rely solely on temporal and structural information, offer strong inductive generalisation but struggle to distinguish known nodes. In contrast, non-anonymous TGNs leverage node features to excel in transductive tasks yet fail to adapt to new nodes. To address this challenge, we propose Trajectory Encoding TGN (TETGN). Our approach introduces automatically expandable node identifiers (IDs) as learnable temporal positional features and performs message passing over these IDs to capture each node's historical context. By integrating this trajectory-aware module with a standard TGN using multi-head attention, TETGN effectively balances transductive accuracy with inductive generalisation. Experimental results on three real-world datasets show that TETGN significantly outperforms strong baselines on both link prediction and node classification tasks, demonstrating its ability to unify the advantages of anonymous and non-anonymous models for dynamic graph learning.
☆ A Winner-Takes-All Mechanism for Event Generation
We present a novel framework for central pattern generator design that leverages the intrinsic rebound excitability of neurons in combination with winner-takes-all computation. Our approach unifies decision-making and rhythmic pattern generation within a simple yet powerful network architecture that employs all-to-all inhibitory connections enhanced by designable excitatory interactions. This design offers significant advantages regarding ease of implementation, adaptability, and robustness. We demonstrate its efficacy through a ring oscillator model, which exhibits adaptive phase and frequency modulation, making the framework particularly promising for applications in neuromorphic systems and robotics.
☆ OpenTuringBench: An Open-Model-based Benchmark and Framework for Machine-Generated Text Detection and Attribution
Open Large Language Models (OLLMs) are increasingly leveraged in generative AI applications, posing new challenges for detecting their outputs. We propose OpenTuringBench, a new benchmark based on OLLMs, designed to train and evaluate machine-generated text detectors on the Turing Test and Authorship Attribution problems. OpenTuringBench focuses on a representative set of OLLMs, and features a number of challenging evaluation tasks, including human/machine-manipulated texts, out-of-domain texts, and texts from previously unseen models. We also provide OTBDetector, a contrastive learning framework to detect and attribute OLLM-based machine-generated texts. Results highlight the relevance and varying degrees of difficulty of the OpenTuringBench tasks, with our detector achieving remarkable capabilities across the various tasks and outperforming most existing detectors. Resources are available on the OpenTuringBench Hugging Face repository at https://huggingface.co/datasets/MLNTeam-Unical/OpenTuringBench
comment: Under review with ARR
☆ Teaching Large Language Models to Reason through Learning and Forgetting
Leveraging inference-time search in large language models has proven effective in further enhancing a trained model's capability to solve complex mathematical and reasoning problems. However, this approach significantly increases computational costs and inference time, as the model must generate and evaluate multiple candidate solutions to identify a viable reasoning path. To address this, we propose an effective approach that integrates search capabilities directly into the model by fine-tuning it using both successful (learning) and failed reasoning paths (forgetting) derived from diverse search methods. While fine-tuning the model with these data might seem straightforward, we identify a critical issue: the model's search capability tends to degrade rapidly if fine-tuning is performed naively. We show that this degradation can be substantially mitigated by employing a smaller learning rate. Extensive experiments on the challenging Game-of-24 and Countdown mathematical reasoning benchmarks show that our approach not only outperforms both standard fine-tuning and inference-time search baselines but also significantly reduces inference time by 180$\times$.
☆ DataSentinel: A Game-Theoretic Detection of Prompt Injection Attacks
LLM-integrated applications and agents are vulnerable to prompt injection attacks, where an attacker injects prompts into their inputs to induce attacker-desired outputs. A detection method aims to determine whether a given input is contaminated by an injected prompt. However, existing detection methods have limited effectiveness against state-of-the-art attacks, let alone adaptive ones. In this work, we propose DataSentinel, a game-theoretic method to detect prompt injection attacks. Specifically, DataSentinel fine-tunes an LLM to detect inputs contaminated with injected prompts that are strategically adapted to evade detection. We formulate this as a minimax optimization problem, with the objective of fine-tuning the LLM to detect strong adaptive attacks. Furthermore, we propose a gradient-based method to solve the minimax optimization problem by alternating between the inner max and outer min problems. Our evaluation results on multiple benchmark datasets and LLMs show that DataSentinel effectively detects both existing and adaptive prompt injection attacks.
comment: To appear in IEEE Symposium on Security and Privacy, 2025
☆ Neural Networks for on-chip Model Predictive Control: a Method to Build Optimized Training Datasets and its application to Type-1 Diabetes
Training Neural Networks (NNs) to behave as Model Predictive Control (MPC) algorithms is an effective way to implement them in constrained embedded devices. By collecting large amounts of input-output data, where inputs represent system states and outputs are MPC-generated control actions, NNs can be trained to replicate MPC behavior at a fraction of the computational cost. However, although the composition of the training data critically influences the final NN accuracy, methods for systematically optimizing it remain underexplored. In this paper, we introduce the concept of Optimally-Sampled Datasets (OSDs) as ideal training sets and present an efficient algorithm for generating them. An OSD is a parametrized subset of all the available data that (i) preserves existing MPC information up to a certain numerical resolution, (ii) avoids duplicate or near-duplicate states, and (iii) becomes saturated or complete. We demonstrate the effectiveness of OSDs by training NNs to replicate the University of Virginia's MPC algorithm for automated insulin delivery in Type-1 Diabetes, achieving a four-fold improvement in final accuracy. Notably, two OSD-trained NNs received regulatory clearance for clinical testing as the first NN-based control algorithm for direct human insulin dosing. This methodology opens new pathways for implementing advanced optimizations on resource-constrained embedded platforms, potentially revolutionizing how complex algorithms are deployed.
☆ Kimina-Prover Preview: Towards Large Formal Reasoning Models with Reinforcement Learning
We introduce Kimina-Prover Preview, a large language model that pioneers a novel reasoning-driven exploration paradigm for formal theorem proving, as showcased in this preview release. Trained with a large-scale reinforcement learning pipeline from Qwen2.5-72B, Kimina-Prover demonstrates strong performance in Lean 4 proof generation by employing a structured reasoning pattern we term \textit{formal reasoning pattern}. This approach allows the model to emulate human problem-solving strategies in Lean, iteratively generating and refining proof steps. Kimina-Prover sets a new state-of-the-art on the miniF2F benchmark, reaching 80.7% with pass@8192. Beyond improved benchmark performance, our work yields several key insights: (1) Kimina-Prover exhibits high sample efficiency, delivering strong results even with minimal sampling (pass@1) and scaling effectively with computational budget, stemming from its unique reasoning pattern and RL training; (2) we demonstrate clear performance scaling with model size, a trend previously unobserved for neural theorem provers in formal mathematics; (3) the learned reasoning style, distinct from traditional search algorithms, shows potential to bridge the gap between formal verification and informal mathematical intuition. We open source distilled versions with 1.5B and 7B parameters of Kimina-Prover
comment: 22 pages
☆ Explicit and Implicit Representations in AI-based 3D Reconstruction for Radiology: A systematic literature review
The demand for high-quality medical imaging in clinical practice and assisted diagnosis has made 3D reconstruction in radiological imaging a key research focus. Artificial intelligence (AI) has emerged as a promising approach to enhancing reconstruction accuracy while reducing acquisition and processing time, thereby minimizing patient radiation exposure and discomfort and ultimately benefiting clinical diagnosis. This review explores state-of-the-art AI-based 3D reconstruction algorithms in radiological imaging, categorizing them into explicit and implicit approaches based on their underlying principles. Explicit methods include point-based, volume-based, and Gaussian representations, while implicit methods encompass implicit prior embedding and neural radiance fields. Additionally, we examine commonly used evaluation metrics and benchmark datasets. Finally, we discuss the current state of development, key challenges, and future research directions in this evolving field. Our project available on: https://github.com/Bean-Young/AI4Med.
comment: 43 pages, 5 figures, submit to Medical Image Analysis
☆ Interpretable Hybrid-Rule Temporal Point Processes
Temporal Point Processes (TPPs) are widely used for modeling event sequences in various medical domains, such as disease onset prediction, progression analysis, and clinical decision support. Although TPPs effectively capture temporal dynamics, their lack of interpretability remains a critical challenge. Recent advancements have introduced interpretable TPPs. However, these methods fail to incorporate numerical features, thereby limiting their ability to generate precise predictions. To address this issue, we propose Hybrid-Rule Temporal Point Processes (HRTPP), a novel framework that integrates temporal logic rules with numerical features, improving both interpretability and predictive accuracy in event modeling. HRTPP comprises three key components: basic intensity for intrinsic event likelihood, rule-based intensity for structured temporal dependencies, and numerical feature intensity for dynamic probability modulation. To effectively discover valid rules, we introduce a two-phase rule mining strategy with Bayesian optimization. To evaluate our method, we establish a multi-criteria assessment framework, incorporating rule validity, model fitting, and temporal predictive accuracy. Experimental results on real-world medical datasets demonstrate that HRTPP outperforms state-of-the-art interpretable TPPs in terms of predictive performance and clinical interpretability. In case studies, the rules extracted by HRTPP explain the disease progression, offering valuable contributions to medical diagnosis.
☆ A Minimalist Approach to LLM Reasoning: from Rejection Sampling to Reinforce
Reinforcement learning (RL) has become a prevailing approach for fine-tuning large language models (LLMs) on complex reasoning tasks. Among recent methods, GRPO stands out for its empirical success in training models such as DeepSeek-R1, yet the sources of its effectiveness remain poorly understood. In this work, we revisit GRPO from a reinforce-like algorithm perspective and analyze its core components. Surprisingly, we find that a simple rejection sampling baseline, RAFT, which trains only on positively rewarded samples, yields competitive performance than GRPO and PPO. Our ablation studies reveal that GRPO's main advantage arises from discarding prompts with entirely incorrect responses, rather than from its reward normalization. Motivated by this insight, we propose Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples. Reinforce-Rej improves KL efficiency and stability, serving as a lightweight yet effective alternative to more complex RL algorithms. We advocate RAFT as a robust and interpretable baseline, and suggest that future advances should focus on more principled designs for incorporating negative samples, rather than relying on them indiscriminately. Our findings provide guidance for future work in reward-based LLM post-training.
comment: 12 pages, 4 figures
Transformer-Based Model for Cold Start Mitigation in FaaS Architecture
Serverless architectures, particularly the Function as a Service (FaaS) model, have become a cornerstone of modern cloud computing due to their ability to simplify resource management and enhance application deployment agility. However, a significant challenge remains: the cold start problem. This phenomenon occurs when an idle FaaS function is invoked, requiring a full initialization process, which increases latency and degrades user experience. Existing solutions for cold start mitigation are limited in terms of invocation pattern generalization and implementation complexity. In this study, we propose an innovative approach leveraging Transformer models to mitigate the impact of cold starts in FaaS architectures. Our solution excels in accurately modeling function initialization delays and optimizing serverless system performance. Experimental evaluation using a public dataset provided by Azure demonstrates a significant reduction in cold start times, reaching up to 79\% compared to conventional methods.
☆ Looking beyond the next token
The structure of causal language model training assumes that each token can be accurately predicted from the previous context. This contrasts with humans' natural writing and reasoning process, where goals are typically known before the exact argument or phrasings. While this mismatch has been well studied in the literature, the working assumption has been that architectural changes are needed to address this mismatch. We argue that rearranging and processing the training data sequences can allow models to more accurately imitate the true data-generating process, and does not require any other changes to the architecture or training infrastructure. We demonstrate that this technique, Trelawney, and the inference algorithms derived from it allow us to improve performance on several key benchmarks that span planning, algorithmic reasoning, and story generation tasks. Finally, our method naturally enables the generation of long-term goals at no additional cost. We investigate how using the model's goal-generation capability can further improve planning and reasoning. Additionally, we believe Trelawney could potentially open doors to new capabilities beyond the current language modeling paradigm.
☆ Code Reborn AI-Driven Legacy Systems Modernization from COBOL to Java
This study investigates AI-driven modernization of legacy COBOL code into Java, addressing a critical challenge in aging software systems. Leveraging the Legacy COBOL 2024 Corpus -- 50,000 COBOL files from public and enterprise sources -- Java parses the code, AI suggests upgrades, and React visualizes gains. Achieving 93% accuracy, complexity drops 35% (from 18 to 11.7) and coupling 33% (from 8 to 5.4), surpassing manual efforts (75%) and rule-based tools (82%). The approach offers a scalable path to rejuvenate COBOL systems, vital for industries like banking and insurance.
☆ Optimizing LLM Inference: Fluid-Guided Online Scheduling with Memory Constraints
Large Language Models (LLMs) are indispensable in today's applications, but their inference procedure -- generating responses by processing text in segments and using a memory-heavy Key-Value (KV) cache -- demands significant computational resources, particularly under memory constraints. This paper formulates LLM inference optimization as a multi-stage online scheduling problem where sequential prompt arrivals and KV cache growth render conventional scheduling ineffective. We develop a fluid dynamics approximation to provide a tractable benchmark that guides algorithm design. Building on this, we propose the Waiting for Accumulated Inference Threshold (WAIT) algorithm, which uses multiple thresholds to schedule incoming prompts optimally when output lengths are known, and extend it to Nested WAIT for cases with unknown output lengths. Theoretical analysis shows that both algorithms achieve near-optimal performance against the fluid benchmark in heavy traffic conditions, balancing throughput, latency, and Time to First Token (TTFT). Experiments with the Llama-7B model on an A100 GPU using both synthetic and real-world datasets demonstrate improved throughput and reduced latency relative to established baselines like vLLM and Sarathi. This work bridges operations research and machine learning, offering a rigorous framework for the efficient deployment of LLMs under memory constraints.
comment: 42 pages, 18 figures
☆ CFIS-YOLO: A Lightweight Multi-Scale Fusion Network for Edge-Deployable Wood Defect Detection
Wood defect detection is critical for ensuring quality control in the wood processing industry. However, current industrial applications face two major challenges: traditional methods are costly, subjective, and labor-intensive, while mainstream deep learning models often struggle to balance detection accuracy and computational efficiency for edge deployment. To address these issues, this study proposes CFIS-YOLO, a lightweight object detection model optimized for edge devices. The model introduces an enhanced C2f structure, a dynamic feature recombination module, and a novel loss function that incorporates auxiliary bounding boxes and angular constraints. These innovations improve multi-scale feature fusion and small object localization while significantly reducing computational overhead. Evaluated on a public wood defect dataset, CFIS-YOLO achieves a mean Average Precision (mAP@0.5) of 77.5\%, outperforming the baseline YOLOv10s by 4 percentage points. On SOPHON BM1684X edge devices, CFIS-YOLO delivers 135 FPS, reduces power consumption to 17.3\% of the original implementation, and incurs only a 0.5 percentage point drop in mAP. These results demonstrate that CFIS-YOLO is a practical and effective solution for real-world wood defect detection in resource-constrained environments.
comment: 10 pages, 11 figures
☆ Learning to Be A Doctor: Searching for Effective Medical Agent Architectures
Large Language Model (LLM)-based agents have demonstrated strong capabilities across a wide range of tasks, and their application in the medical domain holds particular promise due to the demand for high generalizability and reliance on interdisciplinary knowledge. However, existing medical agent systems often rely on static, manually crafted workflows that lack the flexibility to accommodate diverse diagnostic requirements and adapt to emerging clinical scenarios. Motivated by the success of automated machine learning (AutoML), this paper introduces a novel framework for the automated design of medical agent architectures. Specifically, we define a hierarchical and expressive agent search space that enables dynamic workflow adaptation through structured modifications at the node, structural, and framework levels. Our framework conceptualizes medical agents as graph-based architectures composed of diverse, functional node types and supports iterative self-improvement guided by diagnostic feedback. Experimental results on skin disease diagnosis tasks demonstrate that the proposed method effectively evolves workflow structures and significantly enhances diagnostic accuracy over time. This work represents the first fully automated framework for medical agent architecture design and offers a scalable, adaptable foundation for deploying intelligent agents in real-world clinical environments.
☆ Bipartite Ranking From Multiple Labels: On Loss Versus Label Aggregation
Bipartite ranking is a fundamental supervised learning problem, with the goal of learning a ranking over instances with maximal area under the ROC curve (AUC) against a single binary target label. However, one may often observe multiple binary target labels, e.g., from distinct human annotators. How can one synthesize such labels into a single coherent ranking? In this work, we formally analyze two approaches to this problem -- loss aggregation and label aggregation -- by characterizing their Bayes-optimal solutions. Based on this, we show that while both methods can yield Pareto-optimal solutions, loss aggregation can exhibit label dictatorship: one can inadvertently (and undesirably) favor one label over others. This suggests that label aggregation can be preferable to loss aggregation, which we empirically verify.
☆ Single-Input Multi-Output Model Merging: Leveraging Foundation Models for Dense Multi-Task Learning
Model merging is a flexible and computationally tractable approach to merge single-task checkpoints into a multi-task model. Prior work has solely focused on constrained multi-task settings where there is a one-to-one mapping between a sample and a task, overlooking the paradigm where multiple tasks may operate on the same sample, e.g., scene understanding. In this paper, we focus on the multi-task setting with single-input-multiple-outputs (SIMO) and show that it qualitatively differs from the single-input-single-output model merging settings studied in the literature due to the existence of task-specific decoders and diverse loss objectives. We identify that existing model merging methods lead to significant performance degradation, primarily due to representation misalignment between the merged encoder and task-specific decoders. We propose two simple and efficient fixes for the SIMO setting to re-align the feature representation after merging. Compared to joint fine-tuning, our approach is computationally effective and flexible, and sheds light into identifying task relationships in an offline manner. Experiments on NYUv2, Cityscapes, and a subset of the Taskonomy dataset demonstrate: (1) task arithmetic suffices to enable multi-task capabilities; however, the representations generated by the merged encoder has to be re-aligned with the task-specific heads; (2) the proposed architecture rivals traditional multi-task learning in performance but requires fewer samples and training steps by leveraging the existence of task-specific models.
comment: 22 pages, 6 figures
☆ DeepSelective: Feature Gating and Representation Matching for Interpretable Clinical Prediction
The rapid accumulation of Electronic Health Records (EHRs) has transformed healthcare by providing valuable data that enhance clinical predictions and diagnoses. While conventional machine learning models have proven effective, they often lack robust representation learning and depend heavily on expert-crafted features. Although deep learning offers powerful solutions, it is often criticized for its lack of interpretability. To address these challenges, we propose DeepSelective, a novel end to end deep learning framework for predicting patient prognosis using EHR data, with a strong emphasis on enhancing model interpretability. DeepSelective combines data compression techniques with an innovative feature selection approach, integrating custom-designed modules that work together to improve both accuracy and interpretability. Our experiments demonstrate that DeepSelective not only enhances predictive accuracy but also significantly improves interpretability, making it a valuable tool for clinical decision-making. The source code is freely available at http://www.healthinformaticslab.org/supp/resources.php .
☆ A Rollout-Based Algorithm and Reward Function for Efficient Resource Allocation in Business Processes
Resource allocation plays a critical role in minimizing cycle time and improving the efficiency of business processes. Recently, Deep Reinforcement Learning (DRL) has emerged as a powerful tool to optimize resource allocation policies in business processes. In the DRL framework, an agent learns a policy through interaction with the environment, guided solely by reward signals that indicate the quality of its decisions. However, existing algorithms are not suitable for dynamic environments such as business processes. Furthermore, existing DRL-based methods rely on engineered reward functions that approximate the desired objective, but a misalignment between reward and objective can lead to undesired decisions or suboptimal policies. To address these issues, we propose a rollout-based DRL algorithm and a reward function to optimize the objective directly. Our algorithm iteratively improves the policy by evaluating execution trajectories following different actions. Our reward function directly decomposes the objective function of minimizing the mean cycle time. Maximizing our reward function guarantees that the objective function is minimized without requiring extensive reward engineering. The results show that our method consistently learns the optimal policy in all six evaluated business processes, outperforming the state-of-the-art algorithm that can only learn the optimal policy in two of the evaluated processes.
comment: Pre-print submitted to the 23rd International Conference on Business Process Management
☆ Respiratory Inhaler Sound Event Classification Using Self-Supervised Learning
Asthma is a chronic respiratory condition that affects millions of people worldwide. While this condition can be managed by administering controller medications through handheld inhalers, clinical studies have shown low adherence to the correct inhaler usage technique. Consequently, many patients may not receive the full benefit of their medication. Automated classification of inhaler sounds has recently been studied to assess medication adherence. However, the existing classification models were typically trained using data from specific inhaler types, and their ability to generalize to sounds from different inhalers remains unexplored. In this study, we adapted the wav2vec 2.0 self-supervised learning model for inhaler sound classification by pre-training and fine-tuning this model on inhaler sounds. The proposed model shows a balanced accuracy of 98% on a dataset collected using a dry powder inhaler and smartwatch device. The results also demonstrate that re-finetuning this model on minimal data from a target inhaler is a promising approach to adapting a generic inhaler sound classification model to a different inhaler device and audio capture hardware. This is the first study in the field to demonstrate the potential of smartwatches as assistive technologies for the personalized monitoring of inhaler adherence using machine learning models.
comment: Accepted at the IEEE EMBC 2025 Conference
☆ Influence Maximization in Temporal Social Networks with a Cold-Start Problem: A Supervised Approach
Influence Maximization (IM) in temporal graphs focuses on identifying influential "seeds" that are pivotal for maximizing network expansion. We advocate defining these seeds through Influence Propagation Paths (IPPs), which is essential for scaling up the network. Our focus lies in efficiently labeling IPPs and accurately predicting these seeds, while addressing the often-overlooked cold-start issue prevalent in temporal networks. Our strategy introduces a motif-based labeling method and a tensorized Temporal Graph Network (TGN) tailored for multi-relational temporal graphs, bolstering prediction accuracy and computational efficiency. Moreover, we augment cold-start nodes with new neighbors from historical data sharing similar IPPs. The recommendation system within an online team-based gaming environment presents subtle impact on the social network, forming multi-relational (i.e., weak and strong) temporal graphs for our empirical IM study. We conduct offline experiments to assess prediction accuracy and model training efficiency, complemented by online A/B testing to validate practical network growth and the effectiveness in addressing the cold-start issue.
comment: Accepted by ICWSM 2025
☆ Towards Automated Safety Requirements Derivation Using Agent-based RAG
We study the automated derivation of safety requirements in a self-driving vehicle use case, leveraging LLMs in combination with agent-based retrieval-augmented generation. Conventional approaches that utilise pre-trained LLMs to assist in safety analyses typically lack domain-specific knowledge. Existing RAG approaches address this issue, yet their performance deteriorates when handling complex queries and it becomes increasingly harder to retrieve the most relevant information. This is particularly relevant for safety-relevant applications. In this paper, we propose the use of agent-based RAG to derive safety requirements and show that the retrieved information is more relevant to the queries. We implement an agent-based approach on a document pool of automotive standards and the Apollo case study, as a representative example of an automated driving perception system. Our solution is tested on a data set of safety requirement questions and answers, extracted from the Apollo data. Evaluating a set of selected RAG metrics, we present and discuss advantages of a agent-based approach compared to default RAG methods.
comment: 9 pages, 3 figures
☆ Nondeterministic Polynomial-time Problem Challenge: An Ever-Scaling Reasoning Benchmark for LLMs
Reasoning is the fundamental capability of large language models (LLMs). Due to the rapid progress of LLMs, there are two main issues of current benchmarks: i) these benchmarks can be crushed in a short time (less than 1 year), and ii) these benchmarks may be easily hacked. To handle these issues, we propose the ever-scalingness for building the benchmarks which are uncrushable, unhackable, auto-verifiable and general. This paper presents Nondeterministic Polynomial-time Problem Challenge (NPPC), an ever-scaling reasoning benchmark for LLMs. Specifically, the NPPC has three main modules: i) npgym, which provides a unified interface of 25 well-known NP-complete problems and can generate any number of instances with any levels of complexities, ii) npsolver: which provides a unified interface to evaluate the problem instances with both online and offline models via APIs and local deployments, respectively, and iii) npeval: which provides the comprehensive and ready-to-use tools to analyze the performances of LLMs over different problems, the number of tokens, the aha moments, the reasoning errors and the solution errors. Extensive experiments over widely-used LLMs demonstrate: i) NPPC can successfully decrease the performances of advanced LLMs' performances to below 10%, demonstrating that NPPC is uncrushable, ii) DeepSeek-R1, Claude-3.7-Sonnet, and o1/o3-mini are the most powerful LLMs, where DeepSeek-R1 outperforms Claude-3.7-Sonnet and o1/o3-mini in most NP-complete problems considered, and iii) the numbers of tokens, aha moments in the advanced LLMs, e.g., Claude-3.7-Sonnet and DeepSeek-R1, are observed first to increase and then decrease when the problem instances become more and more difficult. We believe that NPPC is the first ever-scaling reasoning benchmark, serving as the uncrushable and unhackable testbed for LLMs toward artificial general intelligence (AGI).
comment: Preliminary work, 10 pages for main text
☆ Diversity-Driven Learning: Tackling Spurious Correlations and Data Heterogeneity in Federated Models
Federated Learning (FL) enables decentralized training of machine learning models on distributed data while preserving privacy. However, in real-world FL settings, client data is often non-identically distributed and imbalanced, resulting in statistical data heterogeneity which impacts the generalization capabilities of the server's model across clients, slows convergence and reduces performance. In this paper, we address this challenge by first proposing a characterization of statistical data heterogeneity by means of 6 metrics of global and client attribute imbalance, class imbalance, and spurious correlations. Next, we create and share 7 computer vision datasets for binary and multiclass image classification tasks in Federated Learning that cover a broad range of statistical data heterogeneity and hence simulate real-world situations. Finally, we propose FedDiverse, a novel client selection algorithm in FL which is designed to manage and leverage data heterogeneity across clients by promoting collaboration between clients with complementary data distributions. Experiments on the seven proposed FL datasets demonstrate FedDiverse's effectiveness in enhancing the performance and robustness of a variety of FL methods while having low communication and computational overhead.
☆ Mutual Understanding between People and Systems via Neurosymbolic AI and Knowledge Graphs
This chapter investigates the concept of mutual understanding between humans and systems, positing that Neuro-symbolic Artificial Intelligence (NeSy AI) methods can significantly enhance this mutual understanding by leveraging explicit symbolic knowledge representations with data-driven learning models. We start by introducing three critical dimensions to characterize mutual understanding: sharing knowledge, exchanging knowledge, and governing knowledge. Sharing knowledge involves aligning the conceptual models of different agents to enable a shared understanding of the domain of interest. Exchanging knowledge relates to ensuring the effective and accurate communication between agents. Governing knowledge concerns establishing rules and processes to regulate the interaction between agents. Then, we present several different use case scenarios that demonstrate the application of NeSy AI and Knowledge Graphs to aid meaningful exchanges between human, artificial, and robotic agents. These scenarios highlight both the potential and the challenges of combining top-down symbolic reasoning with bottom-up neural learning, guiding the discussion of the coverage provided by current solutions along the dimensions of sharing, exchanging, and governing knowledge. Concurrently, this analysis facilitates the identification of gaps and less developed aspects in mutual understanding to address in future research.
comment: 26 pages, 13 figures, 1 table; pre-print version of book chapter
☆ Efficient Distributed Retrieval-Augmented Generation for Enhancing Language Model Performance
Small language models (SLMs) support efficient deployments on resource-constrained edge devices, but their limited capacity compromises inference performance. Retrieval-augmented generation (RAG) is a promising solution to enhance model performance by integrating external databases, without requiring intensive on-device model retraining. However, large-scale public databases and user-specific private contextual documents are typically located on the cloud and the device separately, while existing RAG implementations are primarily centralized. To bridge this gap, we propose DRAGON, a distributed RAG framework to enhance on-device SLMs through both general and personal knowledge without the risk of leaking document privacy. Specifically, DRAGON decomposes multi-document RAG into multiple parallel token generation processes performed independently and locally on the cloud and the device, and employs a newly designed Speculative Aggregation, a dual-side speculative algorithm to avoid frequent output synchronization between the cloud and device. A new scheduling algorithm is further introduced to identify the optimal aggregation side based on real-time network conditions. Evaluations on real-world hardware testbed demonstrate a significant performance improvement of DRAGON-up to 1.9x greater gains over standalone SLM compared to the centralized RAG, substantial reduction in per-token latency, and negligible Time to First Token (TTFT) overhead.
☆ Enhancing multimodal analogical reasoning with Logic Augmented Generation
Recent advances in Large Language Models have demonstrated their capabilities across a variety of tasks. However, automatically extracting implicit knowledge from natural language remains a significant challenge, as machines lack active experience with the physical world. Given this scenario, semantic knowledge graphs can serve as conceptual spaces that guide the automated text generation reasoning process to achieve more efficient and explainable results. In this paper, we apply a logic-augmented generation (LAG) framework that leverages the explicit representation of a text through a semantic knowledge graph and applies it in combination with prompt heuristics to elicit implicit analogical connections. This method generates extended knowledge graph triples representing implicit meaning, enabling systems to reason on unlabeled multimodal data regardless of the domain. We validate our work through three metaphor detection and understanding tasks across four datasets, as they require deep analogical reasoning capabilities. The results show that this integrated approach surpasses current baselines, performs better than humans in understanding visual metaphors, and enables more explainable reasoning processes, though still has inherent limitations in metaphor understanding, especially for domain-specific metaphors. Furthermore, we propose a thorough error analysis, discussing issues with metaphorical annotations and current evaluation methods.
☆ Benchmarking Next-Generation Reasoning-Focused Large Language Models in Ophthalmology: A Head-to-Head Evaluation on 5,888 Items
Recent advances in reasoning-focused large language models (LLMs) mark a shift from general LLMs toward models designed for complex decision-making, a crucial aspect in medicine. However, their performance in specialized domains like ophthalmology remains underexplored. This study comprehensively evaluated and compared the accuracy and reasoning capabilities of four newly developed reasoning-focused LLMs, namely DeepSeek-R1, OpenAI o1, o3-mini, and Gemini 2.0 Flash-Thinking. Each model was assessed using 5,888 multiple-choice ophthalmology exam questions from the MedMCQA dataset in zero-shot setting. Quantitative evaluation included accuracy, Macro-F1, and five text-generation metrics (ROUGE-L, METEOR, BERTScore, BARTScore, and AlignScore), computed against ground-truth reasonings. Average inference time was recorded for a subset of 100 randomly selected questions. Additionally, two board-certified ophthalmologists qualitatively assessed clarity, completeness, and reasoning structure of responses to differential diagnosis questions.O1 (0.902) and DeepSeek-R1 (0.888) achieved the highest accuracy, with o1 also leading in Macro-F1 (0.900). The performance of models across the text-generation metrics varied: O3-mini excelled in ROUGE-L (0.151), o1 in METEOR (0.232), DeepSeek-R1 and o3-mini tied for BERTScore (0.673), DeepSeek-R1 (-4.105) and Gemini 2.0 Flash-Thinking (-4.127) performed best in BARTScore, while o3-mini (0.181) and o1 (0.176) led AlignScore. Inference time across the models varied, with DeepSeek-R1 being slowest (40.4 seconds) and Gemini 2.0 Flash-Thinking fastest (6.7 seconds). Qualitative evaluation revealed that DeepSeek-R1 and Gemini 2.0 Flash-Thinking tended to provide detailed and comprehensive intermediate reasoning, whereas o1 and o3-mini displayed concise and summarized justifications.
comment: 83 pages, 6 figures, 3 tables, 9 supplementary figures, 7 supplementary tables
☆ Exploring Backdoor Attack and Defense for LLM-empowered Recommendations
The fusion of Large Language Models (LLMs) with recommender systems (RecSys) has dramatically advanced personalized recommendations and drawn extensive attention. Despite the impressive progress, the safety of LLM-based RecSys against backdoor attacks remains largely under-explored. In this paper, we raise a new problem: Can a backdoor with a specific trigger be injected into LLM-based Recsys, leading to the manipulation of the recommendation responses when the backdoor trigger is appended to an item's title? To investigate the vulnerabilities of LLM-based RecSys under backdoor attacks, we propose a new attack framework termed Backdoor Injection Poisoning for RecSys (BadRec). BadRec perturbs the items' titles with triggers and employs several fake users to interact with these items, effectively poisoning the training set and injecting backdoors into LLM-based RecSys. Comprehensive experiments reveal that poisoning just 1% of the training data with adversarial examples is sufficient to successfully implant backdoors, enabling manipulation of recommendations. To further mitigate such a security threat, we propose a universal defense strategy called Poison Scanner (P-Scanner). Specifically, we introduce an LLM-based poison scanner to detect the poisoned items by leveraging the powerful language understanding and rich knowledge of LLMs. A trigger augmentation agent is employed to generate diverse synthetic triggers to guide the poison scanner in learning domain-specific knowledge of the poisoned item detection task. Extensive experiments on three real-world datasets validate the effectiveness of the proposed P-Scanner.
☆ TerraMind: Large-Scale Generative Multimodality for Earth Observation
We present TerraMind, the first any-to-any generative, multimodal foundation model for Earth observation (EO). Unlike other multimodal models, TerraMind is pretrained on dual-scale representations combining both token-level and pixel-level data across modalities. On a token level, TerraMind encodes high-level contextual information to learn cross-modal relationships, while on a pixel level, TerraMind leverages fine-grained representations to capture critical spatial nuances. We pretrained TerraMind on nine geospatial modalities of a global, large-scale dataset. In this paper, we demonstrate that (i) TerraMind's dual-scale early fusion approach unlocks a range of zero-shot and few-shot applications for Earth observation, (ii) TerraMind introduces "Thinking-in-Modalities" (TiM) -- the capability of generating additional artificial data during finetuning and inference to improve the model output -- and (iii) TerraMind achieves beyond state-of-the-art performance in community-standard benchmarks for EO like PANGAEA. The pretraining dataset, the model weights, and our code is open-sourced under a permissive license.
☆ MuSeD: A Multimodal Spanish Dataset for Sexism Detection in Social Media Videos
Sexism is generally defined as prejudice and discrimination based on sex or gender, affecting every sector of society, from social institutions to relationships and individual behavior. Social media platforms amplify the impact of sexism by conveying discriminatory content not only through text but also across multiple modalities, highlighting the critical need for a multimodal approach to the analysis of sexism online. With the rise of social media platforms where users share short videos, sexism is increasingly spreading through video content. Automatically detecting sexism in videos is a challenging task, as it requires analyzing the combination of verbal, audio, and visual elements to identify sexist content. In this study, (1) we introduce MuSeD, a new Multimodal Spanish dataset for Sexism Detection consisting of $\approx$ 11 hours of videos extracted from TikTok and BitChute; (2) we propose an innovative annotation framework for analyzing the contribution of textual and multimodal labels in the classification of sexist and non-sexist content; and (3) we evaluate a range of large language models (LLMs) and multimodal LLMs on the task of sexism detection. We find that visual information plays a key role in labeling sexist content for both humans and models. Models effectively detect explicit sexism; however, they struggle with implicit cases, such as stereotypes, instances where annotators also show low agreement. This highlights the inherent difficulty of the task, as identifying implicit sexism depends on the social and cultural context.
☆ Bypassing Prompt Injection and Jailbreak Detection in LLM Guardrails
Large Language Models (LLMs) guardrail systems are designed to protect against prompt injection and jailbreak attacks. However, they remain vulnerable to evasion techniques. We demonstrate two approaches for bypassing LLM prompt injection and jailbreak detection systems via traditional character injection methods and algorithmic Adversarial Machine Learning (AML) evasion techniques. Through testing against six prominent protection systems, including Microsoft's Azure Prompt Shield and Meta's Prompt Guard, we show that both methods can be used to evade detection while maintaining adversarial utility achieving in some instances up to 100% evasion success. Furthermore, we demonstrate that adversaries can enhance Attack Success Rates (ASR) against black-box targets by leveraging word importance ranking computed by offline white-box models. Our findings reveal vulnerabilities within current LLM protection mechanisms and highlight the need for more robust guardrail systems.
comment: 12 pages, 5 figures, 6 tables
☆ DMAGaze: Gaze Estimation Based on Feature Disentanglement and Multi-Scale Attention
Gaze estimation, which predicts gaze direction, commonly faces the challenge of interference from complex gaze-irrelevant information in face images. In this work, we propose DMAGaze, a novel gaze estimation framework that exploits information from facial images in three aspects: gaze-relevant global features (disentangled from facial image), local eye features (extracted from cropped eye patch), and head pose estimation features, to improve overall performance. Firstly, we design a new continuous mask-based Disentangler to accurately disentangle gaze-relevant and gaze-irrelevant information in facial images by achieving the dual-branch disentanglement goal through separately reconstructing the eye and non-eye regions. Furthermore, we introduce a new cascaded attention module named Multi-Scale Global Local Attention Module (MS-GLAM). Through a customized cascaded attention structure, it effectively focuses on global and local information at multiple scales, further enhancing the information from the Disentangler. Finally, the global gaze-relevant features disentangled by the upper face branch, combined with head pose and local eye features, are passed through the detection head for high-precision gaze estimation. Our proposed DMAGaze has been extensively validated on two mainstream public datasets, achieving state-of-the-art performance.
☆ C-SHAP for time series: An approach to high-level temporal explanations
Time series are ubiquitous in domains such as energy forecasting, healthcare, and industry. Using AI systems, some tasks within these domains can be efficiently handled. Explainable AI (XAI) aims to increase the reliability of AI solutions by explaining model reasoning. For time series, many XAI methods provide point- or sequence-based attribution maps. These methods explain model reasoning in terms of low-level patterns. However, they do not capture high-level patterns that may also influence model reasoning. We propose a concept-based method to provide explanations in terms of these high-level patterns. In this paper, we present C-SHAP for time series, an approach which determines the contribution of concepts to a model outcome. We provide a general definition of C-SHAP and present an example implementation using time series decomposition. Additionally, we demonstrate the effectiveness of the methodology through a use case from the energy domain.
comment: 10 pages, 6 figures
☆ Divergence of Empirical Neural Tangent Kernel in Classification Problems
This paper demonstrates that in classification problems, fully connected neural networks (FCNs) and residual neural networks (ResNets) cannot be approximated by kernel logistic regression based on the Neural Tangent Kernel (NTK) under overtraining (i.e., when training time approaches infinity). Specifically, when using the cross-entropy loss, regardless of how large the network width is (as long as it is finite), the empirical NTK diverges from the NTK on the training samples as training time increases. To establish this result, we first demonstrate the strictly positive definiteness of the NTKs for multi-layer FCNs and ResNets. Then, we prove that during training, % with the cross-entropy loss, the neural network parameters diverge if the smallest eigenvalue of the empirical NTK matrix (Gram matrix) with respect to training samples is bounded below by a positive constant. This behavior contrasts sharply with the lazy training regime commonly observed in regression problems. Consequently, using a proof by contradiction, we show that the empirical NTK does not uniformly converge to the NTK across all times on the training samples as the network width increases. We validate our theoretical results through experiments on both synthetic data and the MNIST classification task. This finding implies that NTK theory is not applicable in this context, with significant theoretical implications for understanding neural networks in classification problems.
☆ Fine-Tuning Large Language Models on Quantum Optimization Problems for Circuit Generation
Large language models (LLM) have achieved remarkable outcomes in addressing complex problems, including math, coding, and analyzing large amounts of scientific reports. Yet few works have explored the potential of LLM in quantum computing. The most challenging problem is how to leverage LLMs to automatically generate quantum circuits at a large scale. In this paper, we address such a challenge by fine-tuning LLMs and injecting the domain-specific knowledge of quantum computing. In particular, we investigate the mechanisms to generate training data sets and construct the end-to-end pipeline to fine-tune pre-trained LLMs that produce parameterized quantum circuits for optimization problems. We have prepared 14,000 quantum circuits covering a substantial part of the quantum optimization landscape: 12 optimization problem instances and their optimized QAOA, VQE, and adaptive VQE circuits. The fine-tuned LLMs can construct syntactically correct parametrized quantum circuits in the most recent OpenQASM 3.0. We have evaluated the quality of the parameters by comparing them to the optimized expectation values and distributions. Our evaluation shows that the fine-tuned LLM outperforms state-of-the-art models and that the parameters are better than random. The LLM-generated parametrized circuits and initial parameters can be used as a starting point for further optimization, \emph{e.g.,} templates in quantum machine learning and the benchmark for compilers and hardware.
comment: 12 pages, 8 figures, 3 tables
☆ AI-guided Antibiotic Discovery Pipeline from Target Selection to Compound Identification
Antibiotic resistance presents a growing global health crisis, demanding new therapeutic strategies that target novel bacterial mechanisms. Recent advances in protein structure prediction and machine learning-driven molecule generation offer a promising opportunity to accelerate drug discovery. However, practical guidance on selecting and integrating these models into real-world pipelines remains limited. In this study, we develop an end-to-end, artificial intelligence-guided antibiotic discovery pipeline that spans target identification to compound realization. We leverage structure-based clustering across predicted proteomes of multiple pathogens to identify conserved, essential, and non-human-homologous targets. We then systematically evaluate six leading 3D-structure-aware generative models$\unicode{x2014}$spanning diffusion, autoregressive, graph neural network, and language model architectures$\unicode{x2014}$on their usability, chemical validity, and biological relevance. Rigorous post-processing filters and commercial analogue searches reduce over 100 000 generated compounds to a focused, synthesizable set. Our results highlight DeepBlock and TamGen as top performers across diverse criteria, while also revealing critical trade-offs between model complexity, usability, and output quality. This work provides a comparative benchmark and blueprint for deploying artificial intelligence in early-stage antibiotic development.
comment: 12 pages, preprint
☆ QAMA: Quantum annealing multi-head attention operator with classical deep learning framework
As large language models scale up, the conventional attention mechanism faces critical challenges of exponential growth in memory consumption and energy costs. Quantum annealing computing, with its inherent advantages in computational efficiency and low energy consumption, offers an innovative direction for constructing novel deep learning architectures. This study proposes the first Quantum Annealing-based Multi-head Attention (QAMA) mechanism, achieving seamless compatibility with classical attention architectures through quadratic unconstrained binary optimization (QUBO) modeling of forward propagation and energy-based backpropagation. The method innovatively leverages the quantum bit interaction characteristics of Ising models to optimize the conventional $O(n^2)$ spatiotemporal complexity into linear resource consumption. Integrated with the optical computing advantages of coherent Ising machines (CIM), the system maintains millisecond-level real-time responsiveness while significantly reducing energy consumption. Our key contributions include: Theoretical proofs establish QAMA mathematical equivalence to classical attention mechanisms; Dual optimization of multi-head specificity and long-range information capture via QUBO constraints; Explicit gradient proofs for the Ising energy equation are utilized to implement gradient conduction as the only path in the computational graph as a layer; Proposed soft selection mechanism overcoming traditional binary attention limitations to approximate continuous weights. Experiments on QBoson CPQC quantum computer show QAMA achieves comparable accuracy to classical operators while reducing inference time to millisecond level and improving solution quality. This work pioneers architectural-level integration of quantum computing and deep learning, applicable to any attention-based model, driving paradigm innovation in AI foundational computing.
☆ DeepMLF: Multimodal language model with learnable tokens for deep fusion in sentiment analysis
While multimodal fusion has been extensively studied in Multimodal Sentiment Analysis (MSA), the role of fusion depth and multimodal capacity allocation remains underexplored. In this work, we position fusion depth, scalability, and dedicated multimodal capacity as primary factors for effective fusion. We introduce DeepMLF, a novel multimodal language model (LM) with learnable tokens tailored toward deep fusion. DeepMLF leverages an audiovisual encoder and a pretrained decoder LM augmented with multimodal information across its layers. We append learnable tokens to the LM that: 1) capture modality interactions in a controlled fashion and 2) preserve independent information flow for each modality. These fusion tokens gather linguistic information via causal self-attention in LM Blocks and integrate with audiovisual information through cross-attention MM Blocks. Serving as dedicated multimodal capacity, this design enables progressive fusion across multiple layers, providing depth in the fusion process. Our training recipe combines modality-specific losses and language modelling loss, with the decoder LM tasked to predict ground truth polarity. Across three MSA benchmarks with varying dataset characteristics, DeepMLF achieves state-of-the-art performance. Our results confirm that deeper fusion leads to better performance, with optimal fusion depths (5-7) exceeding those of existing approaches. Additionally, our analysis on the number of fusion tokens reveals that small token sets ($\sim$20) achieve optimal performance. We examine the importance of representation learning order (fusion curriculum) through audiovisual encoder initialization experiments. Our ablation studies demonstrate the superiority of the proposed fusion design and gating while providing a holistic examination of DeepMLF's scalability to LLMs, and the impact of each training objective and embedding regularization.
comment: Preprint
☆ Emergence of Goal-Directed Behaviors via Active Inference with Self-Prior
Infants often exhibit goal-directed behaviors, such as reaching for a sensory stimulus, even when no external reward criterion is provided. These intrinsically motivated behaviors facilitate spontaneous exploration and learning of the body and environment during early developmental stages. Although computational modeling can offer insight into the mechanisms underlying such behaviors, many existing studies on intrinsic motivation focus primarily on how exploration contributes to acquiring external rewards. In this paper, we propose a novel density model for an agent's own multimodal sensory experiences, called the "self-prior," and investigate whether it can autonomously induce goal-directed behavior. Integrated within an active inference framework based on the free energy principle, the self-prior generates behavioral references purely from an intrinsic process that minimizes mismatches between average past sensory experiences and current observations. This mechanism is also analogous to the acquisition and utilization of a body schema through continuous interaction with the environment. We examine this approach in a simulated environment and confirm that the agent spontaneously reaches toward a tactile stimulus. Our study implements intrinsically motivated behavior shaped by the agent's own sensory experiences, demonstrating the spontaneous emergence of intentional behavior during early development.
comment: 20 pages, Code is available at https://github.com/kim135797531/self-prior
☆ Dynamical errors in machine learning forecasts
In machine learning forecasting, standard error metrics such as mean absolute error (MAE) and mean squared error (MSE) quantify discrepancies between predictions and target values. However, these metrics do not directly evaluate the physical and/or dynamical consistency of forecasts, an increasingly critical concern in scientific and engineering applications. Indeed, a fundamental yet often overlooked question is whether machine learning forecasts preserve the dynamical behavior of the underlying system. Addressing this issue is essential for assessing the fidelity of machine learning models and identifying potential failure modes, particularly in applications where maintaining correct dynamical behavior is crucial. In this work, we investigate the relationship between standard forecasting error metrics, such as MAE and MSE, and the dynamical properties of the underlying system. To achieve this goal, we use two recently developed dynamical indices: the instantaneous dimension ($d$), and the inverse persistence ($\theta$). Our results indicate that larger forecast errors -- e.g., higher MSE -- tend to occur in states with higher $d$ (higher complexity) and higher $\theta$ (lower persistence). To further assess dynamical consistency, we propose error metrics based on the dynamical indices that measure the discrepancy of the forecasted $d$ and $\theta$ versus their correct values. Leveraging these dynamical indices-based metrics, we analyze direct and recursive forecasting strategies for three canonical datasets -- Lorenz, Kuramoto-Sivashinsky equation, and Kolmogorov flow -- as well as a real-world weather forecasting task. Our findings reveal substantial distortions in dynamical properties in ML forecasts, especially for long forecast lead times or long recursive simulations, providing complementary information on ML forecast fidelity that can be used to improve ML models.
☆ Neural Control Barrier Functions from Physics Informed Neural Networks
As autonomous systems become increasingly prevalent in daily life, ensuring their safety is paramount. Control Barrier Functions (CBFs) have emerged as an effective tool for guaranteeing safety; however, manually designing them for specific applications remains a significant challenge. With the advent of deep learning techniques, recent research has explored synthesizing CBFs using neural networks-commonly referred to as neural CBFs. This paper introduces a novel class of neural CBFs that leverages a physics-inspired neural network framework by incorporating Zubov's Partial Differential Equation (PDE) within the context of safety. This approach provides a scalable methodology for synthesizing neural CBFs applicable to high-dimensional systems. Furthermore, by utilizing reciprocal CBFs instead of zeroing CBFs, the proposed framework allows for the specification of flexible, user-defined safe regions. To validate the effectiveness of the approach, we present case studies on three different systems: an inverted pendulum, autonomous ground navigation, and aerial navigation in obstacle-laden environments.
comment: 8 pages, 5 figures
☆ QAVA: Query-Agnostic Visual Attack to Large Vision-Language Models NAACL 2025
In typical multimodal tasks, such as Visual Question Answering (VQA), adversarial attacks targeting a specific image and question can lead large vision-language models (LVLMs) to provide incorrect answers. However, it is common for a single image to be associated with multiple questions, and LVLMs may still answer other questions correctly even for an adversarial image attacked by a specific question. To address this, we introduce the query-agnostic visual attack (QAVA), which aims to create robust adversarial examples that generate incorrect responses to unspecified and unknown questions. Compared to traditional adversarial attacks focused on specific images and questions, QAVA significantly enhances the effectiveness and efficiency of attacks on images when the question is unknown, achieving performance comparable to attacks on known target questions. Our research broadens the scope of visual adversarial attacks on LVLMs in practical settings, uncovering previously overlooked vulnerabilities, particularly in the context of visual adversarial threats. The code is available at https://github.com/btzyd/qava.
comment: Accepted by NAACL 2025 main
☆ "Even explanations will not help in trusting [this] fundamentally biased system": A Predictive Policing Case-Study
In today's society, where Artificial Intelligence (AI) has gained a vital role, concerns regarding user's trust have garnered significant attention. The use of AI systems in high-risk domains have often led users to either under-trust it, potentially causing inadequate reliance or over-trust it, resulting in over-compliance. Therefore, users must maintain an appropriate level of trust. Past research has indicated that explanations provided by AI systems can enhance user understanding of when to trust or not trust the system. However, the utility of presentation of different explanations forms still remains to be explored especially in high-risk domains. Therefore, this study explores the impact of different explanation types (text, visual, and hybrid) and user expertise (retired police officers and lay users) on establishing appropriate trust in AI-based predictive policing. While we observed that the hybrid form of explanations increased the subjective trust in AI for expert users, it did not led to better decision-making. Furthermore, no form of explanations helped build appropriate trust. The findings of our study emphasize the importance of re-evaluating the use of explanations to build [appropriate] trust in AI based systems especially when the system's use is questionable. Finally, we synthesize potential challenges and policy recommendations based on our results to design for appropriate trust in high-risk based AI-based systems.
comment: 33rd ACM Conference on User Modeling, Adaptation and Personalization (UMAP '25), June 16--19, 2025, New York City, NY, USA
☆ GATE3D: Generalized Attention-based Task-synergized Estimation in 3D*
The emerging trend in computer vision emphasizes developing universal models capable of simultaneously addressing multiple diverse tasks. Such universality typically requires joint training across multi-domain datasets to ensure effective generalization. However, monocular 3D object detection presents unique challenges in multi-domain training due to the scarcity of datasets annotated with accurate 3D ground-truth labels, especially beyond typical road-based autonomous driving contexts. To address this challenge, we introduce a novel weakly supervised framework leveraging pseudo-labels. Current pretrained models often struggle to accurately detect pedestrians in non-road environments due to inherent dataset biases. Unlike generalized image-based 2D object detection models, achieving similar generalization in monocular 3D detection remains largely unexplored. In this paper, we propose GATE3D, a novel framework designed specifically for generalized monocular 3D object detection via weak supervision. GATE3D effectively bridges domain gaps by employing consistency losses between 2D and 3D predictions. Remarkably, our model achieves competitive performance on the KITTI benchmark as well as on an indoor-office dataset collected by us to evaluate the generalization capabilities of our framework. Our results demonstrate that GATE3D significantly accelerates learning from limited annotated data through effective pre-training strategies, highlighting substantial potential for broader impacts in robotics, augmented reality, and virtual reality applications. Project page: https://ies0411.github.io/GATE3D/
comment: 9pages, 1 supple
☆ Document Quality Scoring for Web Crawling
The internet contains large amounts of low-quality content, yet users expect web search engines to deliver high-quality, relevant results. The abundant presence of low-quality pages can negatively impact retrieval and crawling processes by wasting resources on these documents. Therefore, search engines can greatly benefit from techniques that leverage efficient quality estimation methods to mitigate these negative impacts. Quality scoring methods for web pages are useful for many processes typical for web search systems, including static index pruning, index tiering, and crawling. Building on work by Chang et al.~\cite{chang2024neural}, who proposed using neural estimators of semantic quality for static index pruning, we extend their approach and apply their neural quality scorers to assess the semantic quality of web pages in crawling prioritisation tasks. In our experimental analysis, we found that prioritising semantically high-quality pages over low-quality ones can improve downstream search effectiveness. Our software contribution consists of a Docker container that computes an effective quality score for a given web page, allowing the quality scorer to be easily included and used in other components of web search systems.
comment: Presented at WOWS2025
☆ MediSee: Reasoning-based Pixel-level Perception in Medical Images
Despite remarkable advancements in pixel-level medical image perception, existing methods are either limited to specific tasks or heavily rely on accurate bounding boxes or text labels as input prompts. However, the medical knowledge required for input is a huge obstacle for general public, which greatly reduces the universality of these methods. Compared with these domain-specialized auxiliary information, general users tend to rely on oral queries that require logical reasoning. In this paper, we introduce a novel medical vision task: Medical Reasoning Segmentation and Detection (MedSD), which aims to comprehend implicit queries about medical images and generate the corresponding segmentation mask and bounding box for the target object. To accomplish this task, we first introduce a Multi-perspective, Logic-driven Medical Reasoning Segmentation and Detection (MLMR-SD) dataset, which encompasses a substantial collection of medical entity targets along with their corresponding reasoning. Furthermore, we propose MediSee, an effective baseline model designed for medical reasoning segmentation and detection. The experimental results indicate that the proposed method can effectively address MedSD with implicit colloquial queries and outperform traditional medical referring segmentation methods.
comment: 10 pages, 6 figures
☆ Dynamic Compressing Prompts for Efficient Inference of Large Language Models
Large Language Models (LLMs) have shown outstanding performance across a variety of tasks, partly due to advanced prompting techniques. However, these techniques often require lengthy prompts, which increase computational costs and can hinder performance because of the limited context windows of LLMs. While prompt compression is a straightforward solution, existing methods confront the challenges of retaining essential information, adapting to context changes, and remaining effective across different tasks. To tackle these issues, we propose a task-agnostic method called Dynamic Compressing Prompts (LLM-DCP). Our method reduces the number of prompt tokens while aiming to preserve the performance as much as possible. We model prompt compression as a Markov Decision Process (MDP), enabling the DCP-Agent to sequentially remove redundant tokens by adapting to dynamic contexts and retaining crucial content. We develop a reward function for training the DCP-Agent that balances the compression rate, the quality of the LLM output, and the retention of key information. This allows for prompt token reduction without needing an external black-box LLM. Inspired by the progressive difficulty adjustment in curriculum learning, we introduce a Hierarchical Prompt Compression (HPC) training strategy that gradually increases the compression difficulty, enabling the DCP-Agent to learn an effective compression method that maintains information integrity. Experiments demonstrate that our method outperforms state-of-the-art techniques, especially at higher compression rates. The code for our approach will be available at https://github.com/Fhujinwu/DCP.
comment: Under review (submited in 2024.11)
☆ TMCIR: Token Merge Benefits Composed Image Retrieval
Composed Image Retrieval (CIR) retrieves target images using a multi-modal query that combines a reference image with text describing desired modifications. The primary challenge is effectively fusing this visual and textual information. Current cross-modal feature fusion approaches for CIR exhibit an inherent bias in intention interpretation. These methods tend to disproportionately emphasize either the reference image features (visual-dominant fusion) or the textual modification intent (text-dominant fusion through image-to-text conversion). Such an imbalanced representation often fails to accurately capture and reflect the actual search intent of the user in the retrieval results. To address this challenge, we propose TMCIR, a novel framework that advances composed image retrieval through two key innovations: 1) Intent-Aware Cross-Modal Alignment. We first fine-tune CLIP encoders contrastively using intent-reflecting pseudo-target images, synthesized from reference images and textual descriptions via a diffusion model. This step enhances the encoder ability of text to capture nuanced intents in textual descriptions. 2) Adaptive Token Fusion. We further fine-tune all encoders contrastively by comparing adaptive token-fusion features with the target image. This mechanism dynamically balances visual and textual representations within the contrastive learning pipeline, optimizing the composed feature for retrieval. Extensive experiments on Fashion-IQ and CIRR datasets demonstrate that TMCIR significantly outperforms state-of-the-art methods, particularly in capturing nuanced user intent.
comment: arXiv admin note: text overlap with arXiv:2310.05473 by other authors
☆ ProtFlow: Fast Protein Sequence Design via Flow Matching on Compressed Protein Language Model Embeddings
The design of protein sequences with desired functionalities is a fundamental task in protein engineering. Deep generative methods, such as autoregressive models and diffusion models, have greatly accelerated the discovery of novel protein sequences. However, these methods mainly focus on local or shallow residual semantics and suffer from low inference efficiency, large modeling space and high training cost. To address these challenges, we introduce ProtFlow, a fast flow matching-based protein sequence design framework that operates on embeddings derived from semantically meaningful latent space of protein language models. By compressing and smoothing the latent space, ProtFlow enhances performance while training on limited computational resources. Leveraging reflow techniques, ProtFlow enables high-quality single-step sequence generation. Additionally, we develop a joint design pipeline for the design scene of multichain proteins. We evaluate ProtFlow across diverse protein design tasks, including general peptides and long-chain proteins, antimicrobial peptides, and antibodies. Experimental results demonstrate that ProtFlow outperforms task-specific methods in these applications, underscoring its potential and broad applicability in computational protein sequence design and analysis.
☆ Exploring the Role of KG-Based RAG in Japanese Medical Question Answering with Small-Scale LLMs
Large language models (LLMs) perform well in medical QA, but their effectiveness in Japanese contexts is limited due to privacy constraints that prevent the use of commercial models like GPT-4 in clinical settings. As a result, recent efforts focus on instruction-tuning open-source LLMs, though the potential of combining them with retrieval-augmented generation (RAG) remains underexplored. To bridge this gap, we are the first to explore a knowledge graph-based (KG) RAG framework for Japanese medical QA small-scale open-source LLMs. Experimental results show that KG-based RAG has only a limited impact on Japanese medical QA using small-scale open-source LLMs. Further case studies reveal that the effectiveness of the RAG is sensitive to the quality and relevance of the external retrieved content. These findings offer valuable insights into the challenges and potential of applying RAG in Japanese medical QA, while also serving as a reference for other low-resource languages.
comment: 10 pages
☆ Evaluating Trust in AI, Human, and Co-produced Feedback Among Undergraduate Students
As generative AI transforms educational feedback practices, understanding students' perceptions of different feedback providers becomes crucial for effective implementation. This study addresses a critical gap by comparing undergraduate students' trust in AI-generated, human-created, and human-AI co-produced feedback, informing how institutions can adapt feedback practices in this new era. Through a within-subject experiment with 91 participants, we investigated factors predicting students' ability to distinguish between feedback types, perception of feedback quality, and potential biases to AI involvement. Findings revealed that students generally preferred AI and co-produced feedback over human feedback in terms of perceived usefulness and objectivity. Only AI feedback suffered a decline in perceived genuineness when feedback sources were revealed, while co-produced feedback maintained its positive perception. Educational AI experience improved students' ability to identify AI feedback and increased their trust in all feedback types, while general AI experience decreased perceived usefulness and credibility. Male students consistently rated all feedback types as less valuable than their female and non-binary counterparts. These insights inform evidence-based guidelines for integrating AI into higher education feedback systems while addressing trust concerns and fostering AI literacy among students.
comment: 35 pages, 6 figures. Under review at Assessment and Evaluation in Higher Education
☆ BEACON: A Benchmark for Efficient and Accurate Counting of Subgraphs
Subgraph counting the task of determining the number of instances of a query pattern within a large graph lies at the heart of many critical applications, from analyzing financial networks and transportation systems to understanding biological interactions. Despite decades of work yielding efficient algorithmic (AL) solutions and, more recently, machine learning (ML) approaches, a clear comparative understanding is elusive. This gap stems from the absence of a unified evaluation framework, standardized datasets, and accessible ground truths, all of which hinder systematic analysis and fair benchmarking. To overcome these barriers, we introduce BEACON: a comprehensive benchmark designed to rigorously evaluate both AL and ML-based subgraph counting methods. BEACON provides a standardized dataset with verified ground truths, an integrated evaluation environment, and a public leaderboard, enabling reproducible and transparent comparisons across diverse approaches. Our extensive experiments reveal that while AL methods excel in efficiently counting subgraphs on very large graphs, they struggle with complex patterns (e.g., those exceeding six nodes). In contrast, ML methods are capable of handling larger patterns but demand massive graph data inputs and often yield suboptimal accuracy on small, dense graphs. These insights not only highlight the unique strengths and limitations of each approach but also pave the way for future advancements in subgraph counting techniques. Overall, BEACON represents a significant step towards unifying and accelerating research in subgraph counting, encouraging innovative solutions and fostering a deeper understanding of the trade-offs between algorithmic and machine learning paradigms.
☆ Can LLMs Leverage Observational Data? Towards Data-Driven Causal Discovery with LLMs
Causal discovery traditionally relies on statistical methods applied to observational data, often requiring large datasets and assumptions about underlying causal structures. Recent advancements in Large Language Models (LLMs) have introduced new possibilities for causal discovery by providing domain expert knowledge. However, it remains unclear whether LLMs can effectively process observational data for causal discovery. In this work, we explore the potential of LLMs for data-driven causal discovery by integrating observational data for LLM-based reasoning. Specifically, we examine whether LLMs can effectively utilize observational data through two prompting strategies: pairwise prompting and breadth first search (BFS)-based prompting. In both approaches, we incorporate the observational data directly into the prompt to assess LLMs' ability to infer causal relationships from such data. Experiments on benchmark datasets show that incorporating observational data enhances causal discovery, boosting F1 scores by up to 0.11 point using both pairwise and BFS LLM-based prompting, while outperforming traditional statistical causal discovery baseline by up to 0.52 points. Our findings highlight the potential and limitations of LLMs for data-driven causal discovery, demonstrating their ability to move beyond textual metadata and effectively interpret and utilize observational data for more informed causal reasoning. Our studies lays the groundwork for future advancements toward fully LLM-driven causal discovery.
☆ Transfer Learning for Temporal Link Prediction
Link prediction on graphs has applications spanning from recommender systems to drug discovery. Temporal link prediction (TLP) refers to predicting future links in a temporally evolving graph and adds additional complexity related to the dynamic nature of graphs. State-of-the-art TLP models incorporate memory modules alongside graph neural networks to learn both the temporal mechanisms of incoming nodes and the evolving graph topology. However, memory modules only store information about nodes seen at train time, and hence such models cannot be directly transferred to entirely new graphs at test time and deployment. In this work, we study a new transfer learning task for temporal link prediction, and develop transfer-effective methods for memory-laden models. Specifically, motivated by work showing the informativeness of structural signals for the TLP task, we augment a structural mapping module to the existing TLP model architectures, which learns a mapping from graph structural (topological) features to memory embeddings. Our work paves the way for a memory-free foundation model for TLP.
comment: 14 pages, 7 figures
☆ Towards A Universal Graph Structural Encoder
Recent advancements in large-scale pre-training have shown the potential to learn generalizable representations for downstream tasks. In the graph domain, however, capturing and transferring structural information across different graph domains remains challenging, primarily due to the inherent differences in topological patterns across various contexts. Additionally, most existing models struggle to capture the complexity of rich graph structures, leading to inadequate exploration of the embedding space. To address these challenges, we propose GFSE, a universal graph structural encoder designed to capture transferable structural patterns across diverse domains such as molecular graphs, social networks, and citation networks. GFSE is the first cross-domain graph structural encoder pre-trained with multiple self-supervised learning objectives. Built on a Graph Transformer, GFSE incorporates attention mechanisms informed by graph inductive bias, enabling it to encode intricate multi-level and fine-grained topological features. The pre-trained GFSE produces generic and theoretically expressive positional and structural encoding for graphs, which can be seamlessly integrated with various downstream graph feature encoders, including graph neural networks for vectorized features and Large Language Models for text-attributed graphs. Comprehensive experiments on synthetic and real-world datasets demonstrate GFSE's capability to significantly enhance the model's performance while requiring substantially less task-specific fine-tuning. Notably, GFSE achieves state-of-the-art performance in 81.6% evaluated cases, spanning diverse graph models and datasets, highlighting its potential as a powerful and versatile encoder for graph-structured data.
☆ LOKA Protocol: A Decentralized Framework for Trustworthy and Ethical AI Agent Ecosystems
The rise of autonomous AI agents, capable of perceiving, reasoning, and acting independently, signals a profound shift in how digital ecosystems operate, govern, and evolve. As these agents proliferate beyond centralized infrastructures, they expose foundational gaps in identity, accountability, and ethical alignment. Three critical questions emerge: Identity: Who or what is the agent? Accountability: Can its actions be verified, audited, and trusted? Ethical Consensus: Can autonomous systems reliably align with human values and prevent harmful emergent behaviors? We present the novel LOKA Protocol (Layered Orchestration for Knowledgeful Agents), a unified, systems-level architecture for building ethically governed, interoperable AI agent ecosystems. LOKA introduces a proposed Universal Agent Identity Layer (UAIL) for decentralized, verifiable identity; intent-centric communication protocols for semantic coordination across diverse agents; and a Decentralized Ethical Consensus Protocol (DECP) that enables agents to make context-aware decisions grounded in shared ethical baselines. Anchored in emerging standards such as Decentralized Identifiers (DIDs), Verifiable Credentials (VCs), and post-quantum cryptography, LOKA offers a scalable, future-resilient blueprint for multi-agent AI governance. By embedding identity, trust, and ethics into the protocol layer itself, LOKA establishes the foundation for a new era of responsible, transparent, and autonomous AI ecosystems operating across digital and physical domains.
comment: 4 Figures, 1 Table
☆ Efficient Reasoning Models: A Survey
Reasoning models have demonstrated remarkable progress in solving complex and logic-intensive tasks by generating extended Chain-of-Thoughts (CoTs) prior to arriving at a final answer. Yet, the emergence of this "slow-thinking" paradigm, with numerous tokens generated in sequence, inevitably introduces substantial computational overhead. To this end, it highlights an urgent need for effective acceleration. This survey aims to provide a comprehensive overview of recent advances in efficient reasoning. It categorizes existing works into three key directions: (1) shorter - compressing lengthy CoTs into concise yet effective reasoning chains; (2) smaller - developing compact language models with strong reasoning capabilities through techniques such as knowledge distillation, other model compression techniques, and reinforcement learning; and (3) faster - designing efficient decoding strategies to accelerate inference. A curated collection of papers discussed in this survey is available in our GitHub repository.
☆ Bridging Distribution Gaps in Time Series Foundation Model Pretraining with Prototype-Guided Normalization
Foundation models have achieved remarkable success across diverse machine-learning domains through large-scale pretraining on large, diverse datasets. However, pretraining on such datasets introduces significant challenges due to substantial mismatches in data distributions, a problem particularly pronounced with time series data. In this paper, we tackle this issue by proposing a domain-aware adaptive normalization strategy within the Transformer architecture. Specifically, we replace the traditional LayerNorm with a prototype-guided dynamic normalization mechanism (ProtoNorm), where learned prototypes encapsulate distinct data distributions, and sample-to-prototype affinity determines the appropriate normalization layer. This mechanism effectively captures the heterogeneity of time series characteristics, aligning pretrained representations with downstream tasks. Through comprehensive empirical evaluation, we demonstrate that our method significantly outperforms conventional pretraining techniques across both classification and forecasting tasks, while effectively mitigating the adverse effects of distribution shifts during pretraining. Incorporating ProtoNorm is as simple as replacing a single line of code. Extensive experiments on diverse real-world time series benchmarks validate the robustness and generalizability of our approach, advancing the development of more versatile time series foundation models.
☆ Xpose: Bi-directional Engineering for Hidden Query Extraction
Query reverse engineering (QRE) aims to synthesize a SQL query to connect a given database and result instance. A recent variation of QRE is where an additional input, an opaque executable containing a ground-truth query, is provided, and the goal is to non-invasively extract this specific query through only input-output examples. This variant, called Hidden Query Extraction (HQE), has a spectrum of industrial use-cases including query recovery, database security, and vendor migration. The reverse engineering (RE) tools developed for HQE, which are based on database mutation and generation techniques, can only extract flat queries with key-based equi joins and conjunctive arithmetic filter predicates, making them limited wrt both query structure and query operators. In this paper, we present Xpose, a HQE solution that elevates the extraction scope to realistic complex queries, such as those found in the TPCH benchmark. A two-pronged approach is taken: (1) The existing RE scope is substantially extended to incorporate union connectors, algebraic filter predicates, and disjunctions for both values and predicates. (2) The predictive power of LLMs is leveraged to convert business descriptions of the opaque application into extraction guidance, representing ``forward engineering" (FE). The FE module recognizes common constructs, such as nesting of sub-queries, outer joins, and scalar functions. In essence, FE establishes the broad query contours, while RE fleshes out the fine-grained details. We have evaluated Xpose on (a) E-TPCH, a query suite comprising the complete TPCH benchmark extended with queries featuring unions, diverse join types, and sub-queries; and (b) the real-world STACK benchmark. The experimental results demonstrate that its bi-directional engineering approach accurately extracts these complex queries, representing a significant step forward with regard to HQE coverage.
☆ ARise: Towards Knowledge-Augmented Reasoning via Risk-Adaptive Search
Large language models (LLMs) have demonstrated impressive capabilities and are receiving increasing attention to enhance their reasoning through scaling test--time compute. However, their application in open--ended, knowledge--intensive, complex reasoning scenarios is still limited. Reasoning--oriented methods struggle to generalize to open--ended scenarios due to implicit assumptions of complete world knowledge. Meanwhile, knowledge--augmented reasoning (KAR) methods fail to address two core challenges: 1) error propagation, where errors in early steps cascade through the chain, and 2) verification bottleneck, where the explore--exploit tradeoff arises in multi--branch decision processes. To overcome these limitations, we introduce ARise, a novel framework that integrates risk assessment of intermediate reasoning states with dynamic retrieval--augmented generation (RAG) within a Monte Carlo tree search paradigm. This approach enables effective construction and optimization of reasoning plans across multiple maintained hypothesis branches. Experimental results show that ARise significantly outperforms the state--of--the--art KAR methods by up to 23.10%, and the latest RAG-equipped large reasoning models by up to 25.37%.
comment: Project homepage: https://opencausalab.github.io/ARise
☆ CDUPatch: Color-Driven Universal Adversarial Patch Attack for Dual-Modal Visible-Infrared Detectors
Adversarial patches are widely used to evaluate the robustness of object detection systems in real-world scenarios. These patches were initially designed to deceive single-modal detectors (e.g., visible or infrared) and have recently been extended to target visible-infrared dual-modal detectors. However, existing dual-modal adversarial patch attacks have limited attack effectiveness across diverse physical scenarios. To address this, we propose CDUPatch, a universal cross-modal patch attack against visible-infrared object detectors across scales, views, and scenarios. Specifically, we observe that color variations lead to different levels of thermal absorption, resulting in temperature differences in infrared imaging. Leveraging this property, we propose an RGB-to-infrared adapter that maps RGB patches to infrared patches, enabling unified optimization of cross-modal patches. By learning an optimal color distribution on the adversarial patch, we can manipulate its thermal response and generate an adversarial infrared texture. Additionally, we introduce a multi-scale clipping strategy and construct a new visible-infrared dataset, MSDrone, which contains aerial vehicle images in varying scales and perspectives. These data augmentation strategies enhance the robustness of our patch in real-world conditions. Experiments on four benchmark datasets (e.g., DroneVehicle, LLVIP, VisDrone, MSDrone) show that our method outperforms existing patch attacks in the digital domain. Extensive physical tests further confirm strong transferability across scales, views, and scenarios.
☆ Exploring Persona-dependent LLM Alignment for the Moral Machine Experiment ICLR 2025
Deploying large language models (LLMs) with agency in real-world applications raises critical questions about how these models will behave. In particular, how will their decisions align with humans when faced with moral dilemmas? This study examines the alignment between LLM-driven decisions and human judgment in various contexts of the moral machine experiment, including personas reflecting different sociodemographics. We find that the moral decisions of LLMs vary substantially by persona, showing greater shifts in moral decisions for critical tasks than humans. Our data also indicate an interesting partisan sorting phenomenon, where political persona predominates the direction and degree of LLM decisions. We discuss the ethical implications and risks associated with deploying these models in applications that involve moral decisions.
comment: Accepted to ICLR 2025 Workshop - BiAlign (Bidirectional Human-AI Alignment)
☆ PuzzleBench: A Fully Dynamic Evaluation Framework for Large Multimodal Models on Puzzle Solving
Large Multimodal Models (LMMs) have demonstrated impressive capabilities across a wide range of multimodal tasks, achieving ever-increasing performance on various evaluation benchmarks. However, existing benchmarks are typically static and often overlap with pre-training datasets, leading to fixed complexity constraints and substantial data contamination issues. Meanwhile, manually annotated datasets are labor-intensive, time-consuming, and subject to human bias and inconsistency, leading to reliability and reproducibility issues. To address these problems, we propose a fully dynamic multimodal evaluation framework, named Open-ended Visual Puzzle Generation (OVPG), which aims to generate fresh, diverse, and verifiable evaluation data automatically in puzzle-solving tasks. Specifically, the OVPG pipeline consists of a raw material sampling module, a visual content generation module, and a puzzle rule design module, which ensures that each evaluation instance is primitive, highly randomized, and uniquely solvable, enabling continual adaptation to the evolving capabilities of LMMs. Built upon OVPG, we construct PuzzleBench, a dynamic and scalable benchmark comprising 11,840 VQA samples. It features six carefully designed puzzle tasks targeting three core LMM competencies, visual recognition, logical reasoning, and context understanding. PuzzleBench differs from static benchmarks that quickly become outdated. It enables ongoing dataset refreshing through OVPG and a rich set of open-ended puzzle designs, allowing seamless adaptation to the evolving capabilities of LMMs.
☆ Bringing together invertible UNets with invertible attention modules for memory-efficient diffusion models
Diffusion models have recently gained state of the art performance on many image generation tasks. However, most models require significant computational resources to achieve this. This becomes apparent in the application of medical image synthesis due to the 3D nature of medical datasets like CT-scans, MRIs, electron microscope, etc. In this paper we propose a novel architecture for a single GPU memory-efficient training for diffusion models for high dimensional medical datasets. The proposed model is built by using an invertible UNet architecture with invertible attention modules. This leads to the following two contributions: 1. denoising diffusion models and thus enabling memory usage to be independent of the dimensionality of the dataset, and 2. reducing the energy usage during training. While this new model can be applied to a multitude of image generation tasks, we showcase its memory-efficiency on the 3D BraTS2020 dataset leading to up to 15\% decrease in peak memory consumption during training with comparable results to SOTA while maintaining the image quality.
☆ Large Language Model-Informed Feature Discovery Improves Prediction and Interpretation of Credibility Perceptions of Visual Content
In today's visually dominated social media landscape, predicting the perceived credibility of visual content and understanding what drives human judgment are crucial for countering misinformation. However, these tasks are challenging due to the diversity and richness of visual features. We introduce a Large Language Model (LLM)-informed feature discovery framework that leverages multimodal LLMs, such as GPT-4o, to evaluate content credibility and explain its reasoning. We extract and quantify interpretable features using targeted prompts and integrate them into machine learning models to improve credibility predictions. We tested this approach on 4,191 visual social media posts across eight topics in science, health, and politics, using credibility ratings from 5,355 crowdsourced workers. Our method outperformed zero-shot GPT-based predictions by 13 percent in R2, and revealed key features like information concreteness and image format. We discuss the implications for misinformation mitigation, visual credibility, and the role of LLMs in social science.
comment: 26 pages
☆ Can Vision-Language Models Understand and Interpret Dynamic Gestures from Pedestrians? Pilot Datasets and Exploration Towards Instructive Nonverbal Commands for Cooperative Autonomous Vehicles
In autonomous driving, it is crucial to correctly interpret traffic gestures (TGs), such as those of an authority figure providing orders or instructions, or a pedestrian signaling the driver, to ensure a safe and pleasant traffic environment for all road users. This study investigates the capabilities of state-of-the-art vision-language models (VLMs) in zero-shot interpretation, focusing on their ability to caption and classify human gestures in traffic contexts. We create and publicly share two custom datasets with varying formal and informal TGs, such as 'Stop', 'Reverse', 'Hail', etc. The datasets are "Acted TG (ATG)" and "Instructive TG In-The-Wild (ITGI)". They are annotated with natural language, describing the pedestrian's body position and gesture. We evaluate models using three methods utilizing expert-generated captions as baseline and control: (1) caption similarity, (2) gesture classification, and (3) pose sequence reconstruction similarity. Results show that current VLMs struggle with gesture understanding: sentence similarity averages below 0.59, and classification F1 scores reach only 0.14-0.39, well below the expert baseline of 0.70. While pose reconstruction shows potential, it requires more data and refined metrics to be reliable. Our findings reveal that although some SOTA VLMs can interpret zero-shot human traffic gestures, none are accurate and robust enough to be trustworthy, emphasizing the need for further research in this domain.
☆ Understanding the theoretical properties of projected Bellman equation, linear Q-learning, and approximate value iteration
In this paper, we study the theoretical properties of the projected Bellman equation (PBE) and two algorithms to solve this equation: linear Q-learning and approximate value iteration (AVI). We consider two sufficient conditions for the existence of a solution to PBE : strictly negatively row dominating diagonal (SNRDD) assumption and a condition motivated by the convergence of AVI. The SNRDD assumption also ensures the convergence of linear Q-learning, and its relationship with the convergence of AVI is examined. Lastly, several interesting observations on the solution of PBE are provided when using $\epsilon$-greedy policy.
comment: Initial submission
☆ Moving Beyond Next-Token Prediction: Transformers are Context-Sensitive Language Generators
Large Language Models (LLMs), powered by Transformers, have demonstrated human-like intelligence capabilities, yet their underlying mechanisms remain poorly understood. This paper presents a novel framework for interpreting LLMs as probabilistic left context-sensitive languages (CSLs) generators. We hypothesize that Transformers can be effectively decomposed into three fundamental components: context windows, attention mechanisms, and autoregressive generation frameworks. This decomposition allows for the development of more flexible and interpretable computational models, moving beyond the traditional view of attention and autoregression as inseparable processes. We argue that next-token predictions can be understood as probabilistic, dynamic approximations of left CSL production rules, providing an intuitive explanation for how simple token predictions can yield human-like intelligence outputs. Given that all CSLs are left context-sensitive (Penttonen, 1974), we conclude that Transformers stochastically approximate CSLs, which are widely recognized as models of human-like intelligence. This interpretation bridges the gap between Formal Language Theory and the observed generative power of Transformers, laying a foundation for future advancements in generative AI theory and applications. Our novel perspective on Transformer architectures will foster a deeper understanding of LLMs and their future potentials.
comment: 11 pages, 2 figures
☆ Rethinking Theory of Mind Benchmarks for LLMs: Towards A User-Centered Perspective
The last couple of years have witnessed emerging research that appropriates Theory-of-Mind (ToM) tasks designed for humans to benchmark LLM's ToM capabilities as an indication of LLM's social intelligence. However, this approach has a number of limitations. Drawing on existing psychology and AI literature, we summarize the theoretical, methodological, and evaluation limitations by pointing out that certain issues are inherently present in the original ToM tasks used to evaluate human's ToM, which continues to persist and exacerbated when appropriated to benchmark LLM's ToM. Taking a human-computer interaction (HCI) perspective, these limitations prompt us to rethink the definition and criteria of ToM in ToM benchmarks in a more dynamic, interactional approach that accounts for user preferences, needs, and experiences with LLMs in such evaluations. We conclude by outlining potential opportunities and challenges towards this direction.
comment: 7 pages, 1 figure, accepted to the HEAL@CHI 2025 Workshop
☆ Uplink Assisted Joint Channel Estimation and CSI Feedback: An Approach Based on Deep Joint Source-Channel Coding
In frequency division duplex (FDD) multiple-input multiple-output (MIMO) wireless communication systems, the acquisition of downlink channel state information (CSI) is essential for maximizing spatial resource utilization and improving system spectral efficiency. The separate design of modules in AI-based CSI feedback architectures under traditional modular communication frameworks, including channel estimation (CE), CSI compression and feedback, leads to sub-optimal performance. In this paper, we propose an uplink assisted joint CE and and CSI feedback approach via deep learning for downlink CSI acquisition, which mitigates performance degradation caused by distribution bias across separately trained modules in traditional modular communication frameworks. The proposed network adopts a deep joint source-channel coding (DJSCC) architecture to mitigate the cliff effect encountered in the conventional separate source-channel coding. Furthermore, we exploit the uplink CSI as auxiliary information to enhance CSI reconstruction accuracy by leveraging the partial reciprocity between the uplink and downlink channels in FDD systems, without introducing additional overhead. The effectiveness of uplink CSI as assisted information and the necessity of an end-toend multi-module joint training architecture is validated through comprehensive ablation and scalability experiments.
☆ Towards Spatially-Aware and Optimally Faithful Concept-Based Explanations
Post-hoc, unsupervised concept-based explanation methods (U-CBEMs) are a promising tool for generating semantic explanations of the decision-making processes in deep neural networks, having applications in both model improvement and understanding. It is vital that the explanation is accurate, or faithful, to the model, yet we identify several limitations of prior faithfulness metrics that inhibit an accurate evaluation; most notably, prior metrics involve only the set of concepts present, ignoring how they may be spatially distributed. We address these limitations with Surrogate Faithfulness (SF), an evaluation method that introduces a spatially-aware surrogate and two novel faithfulness metrics. Using SF, we produce Optimally Faithful (OF) explanations, where concepts are found that maximize faithfulness. Our experiments show that (1) adding spatial-awareness to prior U-CBEMs increases faithfulness in all cases; (2) OF produces significantly more faithful explanations than prior U-CBEMs (30% or higher improvement in error); (3) OF's learned concepts generalize well to out-of-domain data and are more robust to adversarial examples, where prior U-CBEMs struggle.
☆ Hallucination-Aware Generative Pretrained Transformer for Cooperative Aerial Mobility Control
This paper proposes SafeGPT, a two-tiered framework that integrates generative pretrained transformers (GPTs) with reinforcement learning (RL) for efficient and reliable unmanned aerial vehicle (UAV) last-mile deliveries. In the proposed design, a Global GPT module assigns high-level tasks such as sector allocation, while an On-Device GPT manages real-time local route planning. An RL-based safety filter monitors each GPT decision and overrides unsafe actions that could lead to battery depletion or duplicate visits, effectively mitigating hallucinations. Furthermore, a dual replay buffer mechanism helps both the GPT modules and the RL agent refine their strategies over time. Simulation results demonstrate that SafeGPT achieves higher delivery success rates compared to a GPT-only baseline, while substantially reducing battery consumption and travel distance. These findings validate the efficacy of combining GPT-based semantic reasoning with formal safety guarantees, contributing a viable solution for robust and energy-efficient UAV logistics.
☆ Improving LLM Interpretability and Performance via Guided Embedding Refinement for Sequential Recommendation
The fast development of Large Language Models (LLMs) offers growing opportunities to further improve sequential recommendation systems. Yet for some practitioners, integrating LLMs to their existing base recommendation systems raises questions about model interpretability, transparency and related safety. To partly alleviate challenges from these questions, we propose guided embedding refinement, a method that carries out a guided and interpretable usage of LLM to enhance the embeddings associated with the base recommendation system. Instead of directly using LLMs as the backbone of sequential recommendation systems, we utilize them as auxiliary tools to emulate the sales logic of recommendation and generate guided embeddings that capture domain-relevant semantic information on interpretable attributes. Benefiting from the strong generalization capabilities of the guided embedding, we construct refined embedding by using the guided embedding and reduced-dimension version of the base embedding. We then integrate the refined embedding into the recommendation module for training and inference. A range of numerical experiments demonstrate that guided embedding is adaptable to various given existing base embedding models, and generalizes well across different recommendation tasks. The numerical results show that the refined embedding not only improves recommendation performance, achieving approximately $10\%$ to $50\%$ gains in Mean Reciprocal Rank (MRR), Recall rate, and Normalized Discounted Cumulative Gain (NDCG), but also enhances interpretability, as evidenced by case studies.
☆ Data driven approach towards more efficient Newton-Raphson power flow calculation for distribution grids
Power flow (PF) calculations are fundamental to power system analysis to ensure stable and reliable grid operation. The Newton-Raphson (NR) method is commonly used for PF analysis due to its rapid convergence when initialized properly. However, as power grids operate closer to their capacity limits, ill-conditioned cases and convergence issues pose significant challenges. This work, therefore, addresses these challenges by proposing strategies to improve NR initialization, hence minimizing iterations and avoiding divergence. We explore three approaches: (i) an analytical method that estimates the basin of attraction using mathematical bounds on voltages, (ii) Two data-driven models leveraging supervised learning or physics-informed neural networks (PINNs) to predict optimal initial guesses, and (iii) a reinforcement learning (RL) approach that incrementally adjusts voltages to accelerate convergence. These methods are tested on benchmark systems. This research is particularly relevant for modern power systems, where high penetration of renewables and decentralized generation require robust and scalable PF solutions. In experiments, all three proposed methods demonstrate a strong ability to provide an initial guess for Newton-Raphson method to converge with fewer steps. The findings provide a pathway for more efficient real-time grid operations, which, in turn, support the transition toward smarter and more resilient electricity networks.
comment: 7 pages, 9 figures, 3 tables, 14 equations, 1 lemma, and 2 theorems. ICT for Industry 2025 Alliander usecase workshop paper. Oral presentation of this paper accepted and to be given on 16th April 2025 in ICT.OPEN 2025 conference of Netherlands in the Beatrix Theatre in Utrecht
☆ Achieving Tighter Finite-Time Rates for Heterogeneous Federated Stochastic Approximation under Markovian Sampling
Motivated by collaborative reinforcement learning (RL) and optimization with time-correlated data, we study a generic federated stochastic approximation problem involving $M$ agents, where each agent is characterized by an agent-specific (potentially nonlinear) local operator. The goal is for the agents to communicate intermittently via a server to find the root of the average of the agents' local operators. The generality of our setting stems from allowing for (i) Markovian data at each agent and (ii) heterogeneity in the roots of the agents' local operators. The limited recent work that has accounted for both these features in a federated setting fails to guarantee convergence to the desired point or to show any benefit of collaboration; furthermore, they rely on projection steps in their algorithms to guarantee bounded iterates. Our work overcomes each of these limitations. We develop a novel algorithm titled \texttt{FedHSA}, and prove that it guarantees convergence to the correct point, while enjoying an $M$-fold linear speedup in sample-complexity due to collaboration. To our knowledge, \emph{this is the first finite-time result of its kind}, and establishing it (without relying on a projection step) entails a fairly intricate argument that accounts for the interplay between complex temporal correlations due to Markovian sampling, multiple local steps to save communication, and the drift-effects induced by heterogeneous local operators. Our results have implications for a broad class of heterogeneous federated RL problems (e.g., policy evaluation and control) with function approximation, where the agents' Markov decision processes can differ in their probability transition kernels and reward functions.
☆ Improving Instruct Models for Free: A Study on Partial Adaptation
Instruct models, obtained from various instruction tuning or post-training steps, are commonly deemed superior and more usable than their base counterpart. While the model gains instruction following ability, instruction tuning may lead to forgetting the knowledge from pre-training or it may encourage the model being overly conversational or verbose. This, in turn, can lead to degradation of in-context few-shot learning performance. In this work, we study the performance trajectory between base and instruct models by scaling down the strength of instruction-tuning via the partial adaption method. We show that, across several model families and model sizes, reducing the strength of instruction-tuning results in material improvement on a few-shot in-context learning benchmark covering a variety of classic natural language tasks. This comes at the cost of losing some degree of instruction following ability as measured by AlpacaEval. Our study shines light on the potential trade-off between in-context learning and instruction following abilities that is worth considering in practice.
comment: Author ordering chosen at random
☆ Possibility for Proactive Anomaly Detection ICLR 2025
Time-series anomaly detection, which detects errors and failures in a workflow, is one of the most important topics in real-world applications. The purpose of time-series anomaly detection is to reduce potential damages or losses. However, existing anomaly detection models detect anomalies through the error between the model output and the ground truth (observed) value, which makes them impractical. In this work, we present a \textit{proactive} approach for time-series anomaly detection based on a time-series forecasting model specialized for anomaly detection and a data-driven anomaly detection model. Our proactive approach establishes an anomaly threshold from training data with a data-driven anomaly detection model, and anomalies are subsequently detected by identifying predicted values that exceed the anomaly threshold. In addition, we extensively evaluated the model using four anomaly detection benchmarks and analyzed both predictable and unpredictable anomalies. We attached the source code as supplementary material.
comment: Accepted at ICLR 2025 I Can't Believe It's Not Better: Challenges in Applied Deep Learning Workshop (ICBINB)
♻ ☆ SpoofCeleb: Speech Deepfake Detection and SASV In The Wild SP
This paper introduces SpoofCeleb, a dataset designed for Speech Deepfake Detection (SDD) and Spoofing-robust Automatic Speaker Verification (SASV), utilizing source data from real-world conditions and spoofing attacks generated by Text-To-Speech (TTS) systems also trained on the same real-world data. Robust recognition systems require speech data recorded in varied acoustic environments with different levels of noise to be trained. However, current datasets typically include clean, high-quality recordings (bona fide data) due to the requirements for TTS training; studio-quality or well-recorded read speech is typically necessary to train TTS models. Current SDD datasets also have limited usefulness for training SASV models due to insufficient speaker diversity. SpoofCeleb leverages a fully automated pipeline we developed that processes the VoxCeleb1 dataset, transforming it into a suitable form for TTS training. We subsequently train 23 contemporary TTS systems. SpoofCeleb comprises over 2.5 million utterances from 1,251 unique speakers, collected under natural, real-world conditions. The dataset includes carefully partitioned training, validation, and evaluation sets with well-controlled experimental protocols. We present the baseline results for both SDD and SASV tasks. All data, protocols, and baselines are publicly available at https://jungjee.github.io/spoofceleb.
comment: IEEE OJSP. Official document lives at: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=10839331
♻ ☆ Belief-State Query Policies for User-Aligned POMDPs
Planning in real-world settings often entails addressing partial observability while aligning with users' requirements. We present a novel framework for expressing users' constraints and preferences about agent behavior in a partially observable setting using parameterized belief-state query (BSQ) policies in the setting of goal-oriented partially observable Markov decision processes (gPOMDPs). We present the first formal analysis of such constraints and prove that while the expected cost function of a parameterized BSQ policy w.r.t its parameters is not convex, it is piecewise constant and yields an implicit discrete parameter search space that is finite for finite horizons. This theoretical result leads to novel algorithms that optimize gPOMDP agent behavior with guaranteed user alignment. Analysis proves that our algorithms converge to the optimal user-aligned behavior in the limit. Empirical results show that parameterized BSQ policies provide a computationally feasible approach for user-aligned planning in partially observable settings.
♻ ☆ Weight Ensembling Improves Reasoning in Language Models
We investigate a failure mode that arises during the training of reasoning models, where the diversity of generations begins to collapse, leading to suboptimal test-time scaling. Notably, the Pass@1 rate reliably improves during supervised finetuning (SFT), but Pass@k rapidly deteriorates. Surprisingly, a simple intervention of interpolating the weights of the latest SFT checkpoint with an early checkpoint, otherwise known as WiSE-FT, almost completely recovers Pass@k while also improving Pass@1. The WiSE-FT variant achieves better test-time scaling (Best@k, majority vote) and achieves superior results with less data when tuned further by reinforcement learning. Finally, we find that WiSE-FT provides complementary performance gains that cannot be achieved only through diversity-inducing decoding strategies, like temperature scaling. We formalize a bias-variance tradeoff of Pass@k with respect to the expectation and variance of Pass@1 over the test distribution. We find that WiSE-FT can reduce bias and variance simultaneously, while temperature scaling inherently trades-off between bias and variance.
♻ ☆ OmniXAS: A Universal Deep-Learning Framework for Materials X-ray Absorption Spectra
X-ray absorption spectroscopy (XAS) is a powerful characterization technique for probing the local chemical environment of absorbing atoms. However, analyzing XAS data presents significant challenges, often requiring extensive, computationally intensive simulations, as well as significant domain expertise. These limitations hinder the development of fast, robust XAS analysis pipelines that are essential in high-throughput studies and for autonomous experimentation. We address these challenges with OmniXAS, a framework that contains a suite of transfer learning approaches for XAS prediction, each contributing to improved accuracy and efficiency, as demonstrated on K-edge spectra database covering eight 3d transition metals (Ti-Cu). The OmniXAS framework is built upon three distinct strategies. First, we use M3GNet to derive latent representations of the local chemical environment of absorption sites as input for XAS prediction, achieving up to order-of-magnitude improvements over conventional featurization techniques. Second, we employ a hierarchical transfer learning strategy, training a universal multi-task model across elements before fine-tuning for element-specific predictions. Models based on this cascaded approach after element-wise fine-tuning outperform element-specific models by up to 69%. Third, we implement cross-fidelity transfer learning, adapting a universal model to predict spectra generated by simulation of a different fidelity with a higher computational cost. This approach improves prediction accuracy by up to 11% over models trained on the target fidelity alone. Our approach boosts the throughput of XAS modeling by orders of magnitude versus first-principles simulations and is extendable to XAS prediction for a broader range of elements. This transfer learning framework is generalizable to enhance deep-learning models that target other properties in materials research.
comment: Main manuscript: 23 pages, 14 figures. Supplemental material (13 pages, 6 figures) available as a separate file in arXiv ancillary files (additional downloadable files)
♻ ☆ Breaking the Data Barrier -- Building GUI Agents Through Task Generalization
Graphical User Interface (GUI) agents offer cross-platform solutions for automating complex digital tasks, with significant potential to transform productivity workflows. However, their performance is often constrained by the scarcity of high-quality trajectory data. To address this limitation, we propose training Vision Language Models (VLMs) on data-rich, reasoning-intensive tasks during a dedicated mid-training stage, and then examine how incorporating these tasks facilitates generalization to GUI planning scenarios. Specifically, we explore a range of tasks with readily available instruction-tuning data, including GUI perception, multimodal reasoning, and textual reasoning. Through extensive experiments across 11 mid-training tasks, we demonstrate that: (1) Task generalization proves highly effective, yielding substantial improvements across most settings. For instance, multimodal mathematical reasoning enhances performance on AndroidWorld by an absolute 6.3%. Remarkably, text-only mathematical data significantly boosts GUI web agent performance, achieving a 5.6% improvement on WebArena and 5.4% improvement on AndroidWorld, underscoring notable cross-modal generalization from text-based to visual domains; (2) Contrary to prior assumptions, GUI perception data - previously considered closely aligned with GUI agent tasks and widely utilized for training - has a comparatively limited impact on final performance; (3) Building on these insights, we identify the most effective mid-training tasks and curate optimized mixture datasets, resulting in absolute performance gains of 8.0% on WebArena and 12.2% on AndroidWorld. Our work provides valuable insights into cross-domain knowledge transfer for GUI agents and offers a practical approach to addressing data scarcity challenges in this emerging field. The code, data and models will be available at https://github.com/hkust-nlp/GUIMid.
comment: 24 pages, 11 figures
♻ ☆ AI threats to national security can be countered through an incident regime
Recent progress in AI capabilities has heightened concerns that AI systems could pose a threat to national security, for example, by making it easier for malicious actors to perform cyberattacks on critical national infrastructure, or through loss of control of autonomous AI systems. In parallel, federal legislators in the US have proposed nascent 'AI incident regimes' to identify and counter similar threats. In this paper, we consolidate these two trends and present a timely proposal for a legally mandated post-deployment AI incident regime that aims to counter potential national security threats from AI systems. We start the paper by introducing the concept of 'security-critical' to describe doctors that pose extreme risks to national security, before arguing that 'security-critical' describes civilian nuclear power, aviation, life science dual-use research of concern, and frontier AI development. We then present in detail our AI incident regime proposal, justifying each component of the proposal by demonstrating its similarity to US domestic incident regimes in other 'security-critical' sectors. Finally, we sketch a hypothetical scenario where our proposed AI incident regime deals with an AI cyber incident. Our proposed AI incident regime is split into three phases. The first phase revolves around a novel operationalization of what counts as an 'AI incident' and we suggest that AI providers must create a 'national security case' before deploying a frontier AI system. The second and third phases spell out that AI providers should notify a government agency about incidents, and that the government agency should be involved in amending AI providers' security and safety procedures, in order to counter future threats to national security.
♻ ☆ Deep Learning-Based Longitudinal Prediction of Childhood Myopia Progression Using Fundus Image Sequences and Baseline Refraction Data
Childhood myopia constitutes a significant global health concern. It exhibits an escalating prevalence and has the potential to evolve into severe, irreversible conditions that detrimentally impact familial well-being and create substantial economic costs. Contemporary research underscores the importance of precisely predicting myopia progression to enable timely and effective interventions, thereby averting severe visual impairment in children. Such predictions predominantly rely on subjective clinical assessments, which are inherently biased and resource-intensive, thus hindering their widespread application. In this study, we introduce a novel, high-accuracy method for quantitatively predicting the myopic trajectory and myopia risk in children using only fundus images and baseline refraction data. This approach was validated through a six-year longitudinal study of 3,408 children in Henan, utilizing 16,211 fundus images and corresponding refractive data. Our method based on deep learning demonstrated predictive accuracy with an error margin of 0.311D per year and AUC scores of 0.944 and 0.995 for forecasting the risks of developing myopia and high myopia, respectively. These findings confirm the utility of our model in supporting early intervention strategies and in significantly reducing healthcare costs, particularly by obviating the need for additional metadata and repeated consultations. Furthermore, our method was designed to rely only on fundus images and refractive error data, without the need for meta data or multiple inquiries from doctors, strongly reducing the associated medical costs and facilitating large-scale screening. Our model can even provide good predictions based on only a single time measurement. Consequently, the proposed method is an important means to reduce medical inequities caused by economic disparities.
♻ ☆ "A Good Bot Always Knows Its Limitations": Assessing Autonomous System Decision-making Competencies through Factorized Machine Self-confidence
How can intelligent machines assess their competency to complete a task? This question has come into focus for autonomous systems that algorithmically make decisions under uncertainty. We argue that machine self-confidence -- a form of meta-reasoning based on self-assessments of system knowledge about the state of the world, itself, and ability to reason about and execute tasks -- leads to many computable and useful competency indicators for such agents. This paper presents our body of work, so far, on this concept in the form of the Factorized Machine Self-confidence (FaMSeC) framework, which holistically considers several major factors driving competency in algorithmic decision-making: outcome assessment, solver quality, model quality, alignment quality, and past experience. In FaMSeC, self-confidence indicators are derived via 'problem-solving statistics' embedded in Markov decision process solvers and related approaches. These statistics come from evaluating probabilistic exceedance margins in relation to certain outcomes and associated competency standards specified by an evaluator. Once designed, and evaluated, the statistics can be easily incorporated into autonomous agents and serve as indicators of competency. We include detailed descriptions and examples for Markov decision process agents, and show how outcome assessment and solver quality factors can be found for a range of tasking contexts through novel use of meta-utility functions, behavior simulations, and surrogate prediction models. Numerical evaluations are performed to demonstrate that FaMSeC indicators perform as desired (references to human subject studies beyond the scope of this paper are provided).
comment: 63 pages, 22 figures, version accepted to ACM THRI
♻ ☆ Unmasking Deceptive Visuals: Benchmarking Multimodal Large Language Models on Misleading Chart Question Answering
Misleading chart visualizations, which intentionally manipulate data representations to support specific claims, can distort perceptions and lead to incorrect conclusions. Despite decades of research, misleading visualizations remain a widespread and pressing issue. Recent advances in multimodal large language models (MLLMs) have demonstrated strong chart comprehension capabilities, yet no existing work has systematically evaluated their ability to detect and interpret misleading charts. This paper introduces the Misleading Chart Question Answering (Misleading ChartQA) Benchmark, a large-scale multimodal dataset designed to assess MLLMs in identifying and reasoning about misleading charts. It contains over 3,000 curated examples, covering 21 types of misleaders and 10 chart types. Each example includes standardized chart code, CSV data, and multiple-choice questions with labeled explanations, validated through multi-round MLLM checks and exhausted expert human review. We benchmark 16 state-of-the-art MLLMs on our dataset, revealing their limitations in identifying visually deceptive practices. We also propose a novel pipeline that detects and localizes misleaders, enhancing MLLMs' accuracy in misleading chart interpretation. Our work establishes a foundation for advancing MLLM-driven misleading chart comprehension. We publicly release the sample dataset to support further research in this critical area.
comment: 31 pages in total. Under Review
♻ ☆ Automatic Item Generation for Personality Situational Judgment Tests with Large Language Models
Personality assessment, particularly through situational judgment tests (SJTs), is a vital tool for psychological research, talent selection, and educational evaluation. This study explores the potential of GPT-4, a state-of-the-art large language model (LLM), to automate the generation of personality situational judgment tests (PSJTs) in Chinese. Traditional SJT development is labor-intensive and prone to biases, while GPT-4 offers a scalable, efficient alternative. Two studies were conducted: Study 1 evaluated the impact of prompt design and temperature settings on content validity, finding that optimized prompts with a temperature of 1.0 produced creative and accurate items. Study 2 assessed the psychometric properties of GPT-4-generated PSJTs, revealing that they demonstrated satisfactory reliability and validity, surpassing the performance of manually developed tests in measuring the Big Five personality traits. This research highlights GPT-4's effectiveness in developing high-quality PSJTs, providing a scalable and innovative method for psychometric test development. These findings expand the possibilities of automatic item generation and the application of LLMs in psychology, and offer practical implications for streamlining test development processes in resource-limited settings.
comment: Submitted to Computers in Human Behavior Reports. 54 pages (main text), 12 pages (appendix), and 5 figures
♻ ☆ CMAT: A Multi-Agent Collaboration Tuning Framework for Enhancing Small Language Models
Open large language models (LLMs) have significantly advanced the field of natural language processing, showcasing impressive performance across various tasks.Despite the significant advancements in LLMs, their effective operation still relies heavily on human input to accurately guide the dialogue flow, with agent tuning being a crucial optimization technique that involves human adjustments to the model for better response to such guidance.Addressing this dependency, our work introduces the TinyAgent model, trained on a meticulously curated high-quality dataset. We also present the Collaborative Multi-Agent Tuning (CMAT) framework, an innovative system designed to augment language agent capabilities through adaptive weight updates based on environmental feedback. This framework fosters collaborative learning and real-time adaptation among multiple intelligent agents, enhancing their context-awareness and long-term memory. In this research, we propose a new communication agent framework that integrates multi-agent systems with environmental feedback mechanisms, offering a scalable method to explore cooperative behaviors. Notably, our TinyAgent-7B model exhibits performance on par with GPT-3.5, despite having fewer parameters, signifying a substantial improvement in the efficiency and effectiveness of LLMs.
♻ ☆ Enhancing Commentary Strategies for Imperfect Information Card Games: A Study of Large Language Models in Guandan Commentary
Recent advancements in large language models (LLMs) have unlocked the potential for generating high-quality game commentary. However, producing insightful and engaging commentary for complex games with incomplete information remains a significant challenge. In this paper, we introduce a novel commentary method that combine Reinforcement Learning (RL) and LLMs, tailored specifically for the Chinese card game \textit{Guandan}. Our system leverages RL to generate intricate card-playing scenarios and employs LLMs to generate corresponding commentary text, effectively emulating the strategic analysis and narrative prowess of professional commentators. The framework comprises a state commentary guide, a Theory of Mind (ToM)-based strategy analyzer, and a style retrieval module, which seamlessly collaborate to deliver detailed and context-relevant game commentary in the Chinese language environment. We empower LLMs with ToM capabilities and refine both retrieval and information filtering mechanisms. This facilitates the generation of personalized commentary content. Our experimental results showcase the substantial enhancement in performance achieved by the proposed commentary framework when applied to open-source LLMs, surpassing the performance of GPT-4 across multiple evaluation metrics.
♻ ☆ SCA: Highly Efficient Semantic-Consistent Unrestricted Adversarial Attack
Deep neural network based systems deployed in sensitive environments are vulnerable to adversarial attacks. Unrestricted adversarial attacks typically manipulate the semantic content of an image (e.g., color or texture) to create adversarial examples that are both effective and photorealistic. Recent works have utilized the diffusion inversion process to map images into a latent space, where high-level semantics are manipulated by introducing perturbations. However, they often results in substantial semantic distortions in the denoised output and suffers from low efficiency. In this study, we propose a novel framework called Semantic-Consistent Unrestricted Adversarial Attacks (SCA), which employs an inversion method to extract edit-friendly noise maps and utilizes Multimodal Large Language Model (MLLM) to provide semantic guidance throughout the process. Under the condition of rich semantic information provided by MLLM, we perform the DDPM denoising process of each step using a series of edit-friendly noise maps, and leverage DPM Solver++ to accelerate this process, enabling efficient sampling with semantic consistency. Compared to existing methods, our framework enables the efficient generation of adversarial examples that exhibit minimal discernible semantic changes. Consequently, we for the first time introduce Semantic-Consistent Adversarial Examples (SCAE). Extensive experiments and visualizations have demonstrated the high efficiency of SCA, particularly in being on average 12 times faster than the state-of-the-art attacks. Our research can further draw attention to the security of multimedia information.
♻ ☆ Advanced Architectures Integrated with Agentic AI for Next-Generation Wireless Networks
This paper investigates a range of cutting-edge technologies and architectural innovations aimed at simplifying network operations, reducing operational expenditure (OpEx), and enabling the deployment of new service models. The focus is on (i) Proposing novel, more efficient 6G architectures, with both Control and User planes enabling the seamless expansion of services, while addressing long-term 6G network evolution. (ii) Exploring advanced techniques for constrained artificial intelligence (AI) operations, particularly the design of AI agents for real-time learning, optimizing energy consumption, and the allocation of computational resources. (iii) Identifying technologies and architectures that support the orchestration of backend services using serverless computing models across multiple domains, particularly for vertical industries. (iv) Introducing optically-based, ultra-high-speed, low-latency network architectures, with fast optical switching and real-time control, replacing conventional electronic switching to reduce power consumption by an order of magnitude.
comment: 6 Pages
♻ ☆ Balancing Forecast Accuracy and Switching Costs in Online Optimization of Energy Management Systems
This study investigates the integration of forecasting and optimization in energy management systems, with a focus on the role of switching costs -- penalties incurred from frequent operational adjustments. We develop a theoretical and empirical framework to examine how forecast accuracy and stability interact with switching costs in online decision-making settings. Our analysis spans both deterministic and stochastic optimization approaches, using point and probabilistic forecasts. A novel metric for measuring temporal consistency in probabilistic forecasts is introduced, and the framework is validated in a real-world battery scheduling case based on the CityLearn 2022 challenge. Results show that switching costs significantly alter the trade-off between forecast accuracy and stability, and that more stable forecasts can reduce the performance loss due to switching. Contrary to common practice, the findings suggest that, under non-negligible switching costs, longer commitment periods may lead to better overall outcomes. These insights have practical implications for the design of intelligent, forecast-aware energy management systems.
comment: 34 pages, contains the Appendix with a comment on KPIs, MPC formulation, Theoretical analysis of the MPC performance bounds and extra results on the in-sample performance
♻ ☆ Bridging Technology and Humanities: Evaluating the Impact of Large Language Models on Social Sciences Research with DeepSeek-R1
In recent years, the development of Large Language Models (LLMs) has made significant breakthroughs in the field of natural language processing and has gradually been applied to the field of humanities and social sciences research. LLMs have a wide range of application value in the field of humanities and social sciences because of its strong text understanding, generation and reasoning capabilities. In humanities and social sciences research, LLMs can analyze large-scale text data and make inferences. This article analyzes the large language model DeepSeek-R1 from seven aspects: low-resource language translation, educational question-answering, student writing improvement in higher education, logical reasoning, educational measurement and psychometrics, public health policy analysis, and art education . Then we compare the answers given by DeepSeek-R1 in the seven aspects with the answers given by o1-preview. DeepSeek-R1 performs well in the humanities and social sciences, answering most questions correctly and logically, and can give reasonable analysis processes and explanations. Compared with o1-preview, it can automatically generate reasoning processes and provide more detailed explanations, which is suitable for beginners or people who need to have a detailed understanding of this knowledge, while o1-preview is more suitable for quick reading. Through analysis, it is found that LLM has broad application potential in the field of humanities and social sciences, and shows great advantages in improving text analysis efficiency, language communication and other fields. LLM's powerful language understanding and generation capabilities enable it to deeply explore complex problems in the field of humanities and social sciences, and provide innovative tools for academic research and practical applications.
comment: 52 pages, 19 figures
♻ ☆ Can Learned Optimization Make Reinforcement Learning Less Difficult?
While reinforcement learning (RL) holds great potential for decision making in the real world, it suffers from a number of unique difficulties which often need specific consideration. In particular: it is highly non-stationary; suffers from high degrees of plasticity loss; and requires exploration to prevent premature convergence to local optima and maximize return. In this paper, we consider whether learned optimization can help overcome these problems. Our method, Learned Optimization for Plasticity, Exploration and Non-stationarity (OPEN), meta-learns an update rule whose input features and output structure are informed by previously proposed solutions to these difficulties. We show that our parameterization is flexible enough to enable meta-learning in diverse learning contexts, including the ability to use stochasticity for exploration. Our experiments demonstrate that when meta-trained on single and small sets of environments, OPEN outperforms or equals traditionally used optimizers. Furthermore, OPEN shows strong generalization characteristics across a range of environments and agent architectures.
comment: Added Metadata for Neurips 2024
♻ ☆ Transforming Future Data Center Operations and Management via Physical AI
Data centers (DCs) as mission-critical infrastructures are pivotal in powering the growth of artificial intelligence (AI) and the digital economy. The evolution from Internet DC to AI DC has introduced new challenges in operating and managing data centers for improved business resilience and reduced total cost of ownership. As a result, new paradigms, beyond the traditional approaches based on best practices, must be in order for future data centers. In this research, we propose and develop a novel Physical AI (PhyAI) framework for advancing DC operations and management. Our system leverages the emerging capabilities of state-of-the-art industrial products and our in-house research and development. Specifically, it presents three core modules, namely: 1) an industry-grade in-house simulation engine to simulate DC operations in a highly accurate manner, 2) an AI engine built upon NVIDIA PhysicsNemo for the training and evaluation of physics-informed machine learning (PIML) models, and 3) a digital twin platform built upon NVIDIA Omniverse for our proposed 5-tier digital twin framework. This system presents a scalable and adaptable solution to digitalize, optimize, and automate future data center operations and management, by enabling real-time digital twins for future data centers. To illustrate its effectiveness, we present a compelling case study on building a surrogate model for predicting the thermal and airflow profiles of a large-scale DC in a real-time manner. Our results demonstrate its superior performance over traditional time-consuming Computational Fluid Dynamics/Heat Transfer (CFD/HT) simulation, with a median absolute temperature prediction error of 0.18 {\deg}C. This emerging approach would open doors to several potential research directions for advancing Physical AI in future DC operations.
comment: 9 pages, 5 figures
♻ ☆ Automated Retrosynthesis Planning of Macromolecules Using Large Language Models and Knowledge Graphs
Identifying reliable synthesis pathways in materials chemistry is a complex task, particularly in polymer science, due to the intricate and often non-unique nomenclature of macromolecules. To address this challenge, we propose an agent system that integrates large language models (LLMs) and knowledge graphs. By leveraging LLMs' powerful capabilities for extracting and recognizing chemical substance names, and storing the extracted data in a structured knowledge graph, our system fully automates the retrieval of relevant literatures, extraction of reaction data, database querying, construction of retrosynthetic pathway trees, further expansion through the retrieval of additional literature and recommendation of optimal reaction pathways. By considering the complex interdependencies among chemical reactants, a novel Multi-branched Reaction Pathway Search Algorithm (MBRPS) is proposed to help identify all valid multi-branched reaction pathways, which arise when a single product decomposes into multiple reaction intermediates. In contrast, previous studies were limited to cases where a product decomposes into at most one reaction intermediate. This work represents the first attempt to develop a fully automated retrosynthesis planning agent tailored specially for macromolecules powered by LLMs. Applied to polyimide synthesis, our new approach constructs a retrosynthetic pathway tree with hundreds of pathways and recommends optimized routes, including both known and novel pathways. This demonstrates utilizing LLMs for literature consultation to accomplish specific tasks is possible and crucial for future materials research, given the vast amount of materials-related literature.
comment: The source code of RetroSynthesisAgent is available at https://github.com/QinyuMa316/RetroSynthesisAgent
♻ ☆ Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models
As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.
♻ ☆ An AI-driven multimodal smart home platform for continuous monitoring and intelligent assistance in post-stroke patients
At-home rehabilitation for post-stroke patients presents significant challenges, as continuous, personalized care is often limited outside clinical settings. Additionally, the absence of comprehensive solutions addressing diverse monitoring and assistance needs in home environments complicates recovery efforts. Here, we present a multimodal smart home platform designed for continuous, at-home rehabilitation of post-stroke patients, integrating wearable sensing, ambient monitoring, and adaptive automation. A plantar pressure insole equipped with a machine learning pipeline classifies users into motor recovery stages with up to 94% accuracy, enabling quantitative tracking of walking patterns. A head-mounted eye-tracking module supports cognitive assessments and hands-free control of household devices, while ambient sensors ensure sub-second response times for interaction. These data streams are fused locally via a hierarchical Internet of Things (IoT) architecture, protecting privacy and minimizing latency. An embedded large language model (LLM) agent, Auto-Care, continuously interprets multimodal data to provide real-time interventions-issuing personalized reminders, adjusting environmental conditions, and notifying caregivers. Implemented in a post-stroke context, this integrated smart home platform increases overall user satisfaction by an average of 115% (p<0.01) compared to traditional home environment. Beyond stroke, the system offers a scalable framework for patient-centered, long-term care in broader neurorehabilitation and aging-in-place applications.
comment: 5 figures, 41 references
♻ ☆ Retro-Search: Exploring Untaken Paths for Deeper and Efficient Reasoning
Large reasoning models exhibit remarkable reasoning capabilities via long, elaborate reasoning trajectories. Supervised fine-tuning on such reasoning traces, also known as distillation, can be a cost-effective way to boost reasoning capabilities of student models. However, empirical observations reveal that these reasoning trajectories are often suboptimal, switching excessively between different lines of thought, resulting in under-thinking, over-thinking, and even degenerate responses. We introduce Retro-Search, an MCTS-inspired search algorithm, for distilling higher quality reasoning paths from large reasoning models. Retro-Search retrospectively revises reasoning paths to discover better, yet shorter traces, which can then lead to student models with enhanced reasoning capabilities with shorter, thus faster inference. Our approach can enable two use cases: self-improvement, where models are fine-tuned on their own Retro-Search-ed thought traces, and weak-to-strong improvement, where a weaker model revises stronger model's thought traces via Retro-Search. For self-improving, R1-distill-7B, fine-tuned on its own Retro-Search-ed traces, reduces the average reasoning length by 31.2% while improving performance by 7.7% across seven math benchmarks. For weak-to-strong improvement, we retrospectively revise R1-671B's traces from the OpenThoughts dataset using R1-distill-32B as the Retro-Search-er, a model 20x smaller. Qwen2.5-32B, fine-tuned on this refined data, achieves performance comparable to R1-distill-32B, yielding an 11.3% reduction in reasoning length and a 2.4% performance improvement compared to fine-tuning on the original OpenThoughts data. Our work counters recently emergent viewpoints that question the relevance of search algorithms in the era of large reasoning models, by demonstrating that there are still opportunities for algorithmic advancements, even for frontier models.
comment: Code and data will be publicly released upon internal approval
♻ ☆ Kozax: Flexible and Scalable Genetic Programming in JAX
Genetic programming is an optimization algorithm inspired by evolution which automatically evolves the structure of interpretable computer programs. The fitness evaluation in genetic programming suffers from high computational requirements, limiting the performance on difficult problems. Consequently, there is no efficient genetic programming framework that is usable for a wide range of tasks. To this end, we developed Kozax, a genetic programming framework that evolves symbolic expressions for arbitrary problems. We implemented Kozax using JAX, a framework for high-performance and scalable machine learning, which allows the fitness evaluation to scale efficiently to large populations or datasets on GPU. Furthermore, Kozax offers constant optimization, custom operator definition and simultaneous evolution of multiple trees. We demonstrate successful applications of Kozax to discover equations of natural laws, recover equations of hidden dynamic variables, evolve a control policy and optimize an objective function. Overall, Kozax provides a general, fast, and scalable library to optimize white-box solutions in the realm of scientific computing.
comment: 6 figures, 3 tables, 1 algorithm, 13 pages
♻ ☆ A Neural Network Alternative to Tree-based Models
Tabular datasets are widely used in scientific disciplines such as biology. While these disciplines have already adopted AI methods to enhance their findings and analysis, they mainly use tree-based methods due to their interpretability. At the same time, artificial neural networks have been shown to offer superior flexibility and depth for rich and complex non-tabular problems, but they are falling behind tree-based models for tabular data in terms of performance and interpretability. Although sparsity has been shown to improve the interpretability and performance of ANN models for complex non-tabular datasets, enforcing sparsity structurally and formatively for tabular data before training the model, remains an open question. To address this question, we establish a method that infuses sparsity in neural networks by utilising attention mechanisms to capture the features' importance in tabular datasets. We show that our models, Sparse TABular NET or sTAB-Net with attention mechanisms, are more effective than tree-based models, reaching the state-of-the-art on biological datasets. They further permit the extraction of insights from these datasets and achieve better performance than post-hoc methods like SHAP.
♻ ☆ Safe Text-to-Image Generation: Simply Sanitize the Prompt Embedding
In recent years, text-to-image (T2I) generation models have made significant progress in generating high-quality images that align with text descriptions. However, these models also face the risk of unsafe generation, potentially producing harmful content that violates usage policies, such as explicit material. Existing safe generation methods typically focus on suppressing inappropriate content by erasing undesired concepts from visual representations, while neglecting to sanitize the textual representation. Although these methods help mitigate the risk of misuse to some extent, their robustness remains insufficient when dealing with adversarial attacks. Given that semantic consistency between input text and output image is a core requirement of T2I models, we identify that textual representations are likely the primary source of unsafe generation. To this end, we propose Embedding Sanitizer (ES), which enhances the safety of T2I models by sanitizing inappropriate concepts in prompt embeddings. To our knowledge, ES is the first interpretable safe generation framework that assigns a score to each token in the prompt to indicate its potential harmfulness. In addition, ES adopts a plug-and-play modular design, offering compatibility for seamless integration with various T2I models and other safeguards. Evaluations on five prompt benchmarks show that ES outperforms eleven existing safeguard baselines, achieving state-of-the-art robustness while maintaining high-quality image generation.
♻ ☆ Designing Child-Friendly AI Interfaces: Six Developmentally-Appropriate Design Insights from Analysing Disney Animation
To build AI interfaces that children can intuitively understand and use, designers need a design grammar that truly serves children's developmental needs. This paper bridges Artificial Intelligence design for children -- an emerging field still defining its best practices -- and children's animation, a well-established field with decades of experience in engaging young viewers through emotionally resonant, cognitively accessible storytelling. Pairing Piagetian developmental theory with design pattern extraction from 52 works of Disney animation, the paper presents six design insights transferable to child-centred AI interface design: (1) emotional expressiveness and visual clarity, (2) musical and auditory scaffolding, (3) audiovisual synchrony for emotional comfort, (4) sidekick-style personas, (5) support for symbolic play and imaginative exploration, and (6) predictable and scaffolded interaction structures. These strategies -- long refined in Disney animation -- function as multimodal scaffolds for attention, understanding, and emotional attunement, thereby forming a structured design grammar familiar to children and transferable to AI interface design. By reframing cinematic storytelling as design logic for AI, the paper offers heuristics for crafting intuitive AI interfaces that align with children's cognitive stages and emotional needs. The work contributes to design theory by showing how sensory, affective and narrative techniques can inform developmentally attuned AI design for children. Future directions include empirical testing, cultural adaptation, and participatory co-design.
comment: 30 pages
♻ ☆ WildLive: Near Real-time Visual Wildlife Tracking onboard UAVs
Live tracking of wildlife via high-resolution video processing directly onboard drones is widely unexplored and most existing solutions rely on streaming video to ground stations to support navigation. Yet, both autonomous animal-reactive flight control beyond visual line of sight and/or mission-specific individual and behaviour recognition tasks rely to some degree on this capability. In response, we introduce WildLive -- a near real-time animal detection and tracking framework for high-resolution imagery running directly onboard uncrewed aerial vehicles (UAVs). The system performs multi-animal detection and tracking at 17fps+ for HD and 7fps+ on 4K video streams suitable for operation during higher altitude flights to minimise animal disturbance. Our system is optimised for Jetson Orin AGX onboard hardware. It integrates the efficiency of sparse optical flow tracking and mission-specific sampling with device-optimised and proven YOLO-driven object detection and segmentation techniques. Essentially, computational resource is focused onto spatio-temporal regions of high uncertainty to significantly improve UAV processing speeds without domain-specific loss of accuracy. Alongside, we introduce our WildLive dataset, which comprises 200k+ annotated animal instances across 19k+ frames from 4K UAV videos collected at the Ol Pejeta Conservancy in Kenya. All frames contain ground truth bounding boxes, segmentation masks, as well as individual tracklets and tracking point trajectories. We compare our system against current object tracking approaches including OC-SORT, ByteTrack, and SORT. Our materials are available at: https://dat-nguyenvn.github.io/WildLive/
♻ ☆ MROSS: Multi-Round Region-based Optimization for Scene Sketching
Scene sketching is to convert a scene into a simplified, abstract representation that captures the essential elements and composition of the original scene. It requires a semantic understanding of the scene and consideration of different regions within the scene. Since scenes often contain diverse visual information across various regions, such as foreground objects, background elements, and spatial divisions, dealing with these different regions poses unique difficulties. In this paper, we define a sketch as some sets of B\'ezier curves because of their smooth and versatile characteristics. We optimize different regions of input scene in multiple rounds. In each optimization round, the strokes sampled from the next region can seamlessly be integrated into the sketch generated in the previous optimization round. We propose an additional stroke initialization method to ensure the integrity of the scene and the convergence of optimization. A novel CLIP-based Semantic Loss and a VGG-based Feature Loss are utilized to guide our multi-round optimization. Extensive experimental results on the quality and quantity of the generated sketches confirm the effectiveness of our method.
comment: 6 pages, 8 figures
♻ ☆ Unveiling the Threat of Fraud Gangs to Graph Neural Networks: Multi-Target Graph Injection Attacks Against GNN-Based Fraud Detectors AAAI
Graph neural networks (GNNs) have emerged as an effective tool for fraud detection, identifying fraudulent users, and uncovering malicious behaviors. However, attacks against GNN-based fraud detectors and their risks have rarely been studied, thereby leaving potential threats unaddressed. Recent findings suggest that frauds are increasingly organized as gangs or groups. In this work, we design attack scenarios where fraud gangs aim to make their fraud nodes misclassified as benign by camouflaging their illicit activities in collusion. Based on these scenarios, we study adversarial attacks against GNN-based fraud detectors by simulating attacks of fraud gangs in three real-world fraud cases: spam reviews, fake news, and medical insurance frauds. We define these attacks as multi-target graph injection attacks and propose MonTi, a transformer-based Multi-target one-Time graph injection attack model. MonTi simultaneously generates attributes and edges of all attack nodes with a transformer encoder, capturing interdependencies between attributes and edges more effectively than most existing graph injection attack methods that generate these elements sequentially. Additionally, MonTi adaptively allocates the degree budget for each attack node to explore diverse injection structures involving target, candidate, and attack nodes, unlike existing methods that fix the degree budget across all attack nodes. Experiments show that MonTi outperforms the state-of-the-art graph injection attack methods on five real-world graphs.
comment: 19 pages, 5 figures, 12 tables, The 39th AAAI Conference on Artificial Intelligence (AAAI 2025)
♻ ☆ Automated Proof Generation for Rust Code via Self-Evolution
Ensuring correctness is crucial for code generation. Formal verification offers a definitive assurance of correctness, but demands substantial human effort in proof construction and hence raises a pressing need for automation. The primary obstacle lies in the severe lack of data-there is much fewer proofs than code snippets for Large Language Models (LLMs) to train upon. In this paper, we introduce SAFE, a framework that overcomes the lack of human-written proofs to enable automated proof generation of Rust code. SAFE establishes a self-evolving cycle where data synthesis and fine-tuning collaborate to enhance the model capability, leveraging the definitive power of a symbolic verifier in telling correct proofs from incorrect ones. SAFE also re-purposes the large number of synthesized incorrect proofs to train the self-debugging capability of the fine-tuned models, empowering them to fix incorrect proofs based on the verifier's feedback. SAFE demonstrates superior efficiency and precision compared to GPT-4o. Through tens of thousands of synthesized proofs and the self-debugging mechanism, we improve the capability of open-source models, initially unacquainted with formal verification, to automatically write proofs for Rust code. This advancement leads to a significant improvement in performance, achieving a 52.52% accuracy rate in a benchmark crafted by human experts, a significant leap over GPT-4o's performance of 14.39%.
♻ ☆ Integer-Valued Training and Spike-Driven Inference Spiking Neural Network for High-performance and Energy-efficient Object Detection ECCV2024
Brain-inspired Spiking Neural Networks (SNNs) have bio-plausibility and low-power advantages over Artificial Neural Networks (ANNs). Applications of SNNs are currently limited to simple classification tasks because of their poor performance. In this work, we focus on bridging the performance gap between ANNs and SNNs on object detection. Our design revolves around network architecture and spiking neuron. First, the overly complex module design causes spike degradation when the YOLO series is converted to the corresponding spiking version. We design a SpikeYOLO architecture to solve this problem by simplifying the vanilla YOLO and incorporating meta SNN blocks. Second, object detection is more sensitive to quantization errors in the conversion of membrane potentials into binary spikes by spiking neurons. To address this challenge, we design a new spiking neuron that activates Integer values during training while maintaining spike-driven by extending virtual timesteps during inference. The proposed method is validated on both static and neuromorphic object detection datasets. On the static COCO dataset, we obtain 66.2% mAP@50 and 48.9% mAP@50:95, which is +15.0% and +18.7% higher than the prior state-of-the-art SNN, respectively. On the neuromorphic Gen1 dataset, we achieve 67.2% mAP@50, which is +2.5% greater than the ANN with equivalent architecture, and the energy efficiency is improved by 5.7*. Code: https://github.com/BICLab/SpikeYOLO
comment: Accepted by ECCV2024; 19 pages, 4 figures
♻ ☆ MIMIR: Masked Image Modeling for Mutual Information-based Adversarial Robustness
Vision Transformers (ViTs) have emerged as a fundamental architecture and serve as the backbone of modern vision-language models. Despite their impressive performance, ViTs exhibit notable vulnerability to evasion attacks, necessitating the development of specialized Adversarial Training (AT) strategies tailored to their unique architecture. While a direct solution might involve applying existing AT methods to ViTs, our analysis reveals significant incompatibilities, particularly with state-of-the-art (SOTA) approaches such as Generalist (CVPR 2023) and DBAT (USENIX Security 2024). This paper presents a systematic investigation of adversarial robustness in ViTs and provides a novel theoretical Mutual Information (MI) analysis in its autoencoder-based self-supervised pre-training. Specifically, we show that MI between the adversarial example and its latent representation in ViT-based autoencoders should be constrained via derived MI bounds. Building on this insight, we propose a self-supervised AT method, MIMIR, that employs an MI penalty to facilitate adversarial pre-training by masked image modeling with autoencoders. Extensive experiments on CIFAR-10, Tiny-ImageNet, and ImageNet-1K show that MIMIR can consistently provide improved natural and robust accuracy, where MIMIR outperforms SOTA AT results on ImageNet-1K. Notably, MIMIR demonstrates superior robustness against unforeseen attacks and common corruption data and can also withstand adaptive attacks where the adversary possesses full knowledge of the defense mechanism.
♻ ☆ Faster Configuration Performance Bug Testing with Neural Dual-level Prioritization ICSE 2025
As software systems become more complex and configurable, more performance problems tend to arise from the configuration designs. This has caused some configuration options to unexpectedly degrade performance which deviates from their original expectations designed by the developers. Such discrepancies, namely configuration performance bugs (CPBugs), are devastating and can be deeply hidden in the source code. Yet, efficiently testing CPBugs is difficult, not only due to the test oracle is hard to set, but also because the configuration measurement is expensive and there are simply too many possible configurations to test. As such, existing testing tools suffer from lengthy runtime or have been ineffective in detecting CPBugs when the budget is limited, compounded by inaccurate test oracle. In this paper, we seek to achieve significantly faster CPBug testing by neurally prioritizing the testing at both the configuration option and value range levels with automated oracle estimation. Our proposed tool, dubbed NDP, is a general framework that works with different heuristic generators. The idea is to leverage two neural language models: one to estimate the CPBug types that serve as the oracle while, more vitally, the other to infer the probabilities of an option being CPBug-related, based on which the options and the value ranges to be searched can be prioritized. Experiments on several widely-used systems of different versions reveal that NDP can, in general, better predict CPBug type in 87% cases and find more CPBugs with up to 88.88x testing efficiency speedup over the state-of-the-art tools.
comment: accepted by ICSE 2025
♻ ☆ Causal Graphical Models for Vision-Language Compositional Understanding ICLR 2025
Recent work has empirically shown that Vision-Language Models (VLMs) struggle to fully understand the compositional properties of the human language, usually modeling an image caption as a "bag of words". As a result, they perform poorly on compositional tasks, which require a deeper understanding of the different entities of a sentence (subject, verb, etc.) jointly with their mutual relationships in order to be solved. In this paper, we model the dependency relations among textual and visual tokens using a Causal Graphical Model (CGM), built using a dependency parser, and we train a decoder conditioned by the VLM visual encoder. Differently from standard autoregressive or parallel predictions, our decoder's generative process is partially-ordered following the CGM structure. This structure encourages the decoder to learn only the main causal dependencies in a sentence discarding spurious correlations. Using extensive experiments on five compositional benchmarks, we show that our method significantly outperforms all the state-of-the-art compositional approaches by a large margin, and it also improves over methods trained using much larger datasets.
comment: Accepted at ICLR 2025
♻ ☆ Opinion: Revisiting synthetic data classifications from a privacy perspective
Synthetic data is emerging as a cost-effective solution necessary to meet the increasing data demands of AI development, created either from existing knowledge or derived from real data. The traditional classification of synthetic data types into hybrid, partial or fully synthetic datasets has limited value and does not reflect the ever-increasing methods to generate synthetic data. The generation method and their source jointly shape the characteristics of synthetic data, which in turn determines its practical applications. We make a case for an alternative approach to grouping synthetic data types that better reflect privacy perspectives in order to facilitate regulatory guidance in the generation and processing of synthetic data. This approach to classification provides flexibility to new advancements like deep generative methods and offers a more practical framework for future applications.
♻ ☆ DeNOTS: Stable Deep Neural ODEs for Time Series
Neural ODEs are a prominent branch of methods designed to capture the temporal evolution of complex time-stamped data. Their idea is to solve an ODE with Neural Network-defined dynamics, which take the immediate parameters of the observed system into account. However, larger integration intervals cause instability, which forces most modern methods to normalize time to $[0, 1]$. We provably stabilize these models by introducing an adaptive negative feedback mechanism. This modification allows for longer integration, which in turn implies higher expressiveness, mirroring the behaviour of increasing depth in conventional Neural Networks.Additionally, it provides intriguing theoretical properties: forgetfulness and missing-value robustness. For three open datasets, our method obtains up to 20\% improvements in downstream quality if compared to existing baselines, including State Space Models and Neural~CDEs.
♻ ☆ Robust Reinforcement Learning from Human Feedback for Large Language Models Fine-Tuning
Reinforcement learning from human feedback (RLHF) has emerged as a key technique for aligning the output of large language models (LLMs) with human preferences. To learn the reward function, most existing RLHF algorithms use the Bradley-Terry model, which relies on assumptions about human preferences that may not reflect the complexity and variability of real-world judgments. In this paper, we propose a robust algorithm to enhance the performance of existing approaches under such reward model misspecifications. Theoretically, our algorithm reduces the variance of reward and policy estimators, leading to improved regret bounds. Empirical evaluations on LLM benchmark datasets demonstrate that the proposed algorithm consistently outperforms existing methods, with 77-81% of responses being favored over baselines on the Anthropic Helpful and Harmless dataset.
♻ ☆ Towards the Terminator Economy: Assessing Job Exposure to AI through LLMs
AI and related technologies are reshaping jobs and tasks, either by automating or augmenting human skills in the workplace. Many researchers have been working on estimating if and to what extent jobs and tasks are exposed to the risk of being automatized by AI-related technologies. Our work tackles this issue through a data-driven approach by: (i) developing a reproducible framework that uses cutting-edge open-source large language models to assess the current capabilities of AI and robotics in performing job-related tasks; (ii) formalizing and computing a measure of AI exposure by occupation, the Task Exposure to AI (TEAI) index, and a measure of Task Replacement by AI (TRAI), both validated through a human user evaluation and compared with the state of the art. Our results show that the TEAI index is positively correlated with cognitive, problem-solving and management skills, while it is negatively correlated with social skills. Applying the index to the US, we obtain that about one-third of US employment is highly exposed to AI, primarily in high-skill jobs requiring a graduate or postgraduate level of education. We also find that AI exposure is positively associated with both employment and wage growth in 2003-2023, suggesting that AI has an overall positive effect on productivity. Considering specifically the TRAI index, we find that even in high-skill occupations, AI exhibits high variability in task substitution, suggesting that AI and humans complement each other within the same occupation, while the allocation of tasks within occupations is likely to change. All results, models, and code are freely available online to allow the community to reproduce our results, compare outcomes, and use our work as a benchmark to monitor AI's progress over time.
♻ ☆ Teaching Transformers Causal Reasoning through Axiomatic Training
For text-based AI systems to interact in the real world, causal reasoning is an essential skill. Since active interventions are costly, we study to what extent a system can learn causal reasoning from symbolic demonstrations of causal axioms. Specifically, we present an axiomatic training method where the system learns from multiple demonstrations of a causal axiom (or rule), rather than incorporating the axiom as an inductive bias or inferring it from data values. A key question is whether the system would learn to generalize from the axiom demonstrations to more complex scenarios. Our results, based on applying axiomatic training to learn the transitivity axiom and d-separation rule, indicate that such generalization is possible. To avoid data contamination issues, we start with a 67 million parameter transformer model and train it from scratch. On both tasks, we find that a model trained on linear causal chains (along with some noisy variations) can generalize well to complex graphs, including longer causal chains, causal chains with reversed order, and graphs with branching.To handle diverse text inputs, the same method is extended to finetune language models. Finetuning Llama-3.1 8B model on our axiomatic data leads to significant gains on causal benchmarks such as Corr2Cause and CLEAR, in some cases providing state-of-the-art performance surpassing GPT-4.
♻ ☆ TrafficLLM: Enhancing Large Language Models for Network Traffic Analysis with Generic Traffic Representation
Machine learning (ML) powered network traffic analysis has been widely used for the purpose of threat detection. Unfortunately, their generalization across different tasks and unseen data is very limited. Large language models (LLMs), known for their strong generalization capabilities, have shown promising performance in various domains. However, their application to the traffic analysis domain is limited due to significantly different characteristics of network traffic. To address the issue, in this paper, we propose TrafficLLM, which introduces a dual-stage fine-tuning framework to learn generic traffic representation from heterogeneous raw traffic data. The framework uses traffic-domain tokenization, dual-stage tuning pipeline, and extensible adaptation to help LLM release generalization ability on dynamic traffic analysis tasks, such that it enables traffic detection and traffic generation across a wide range of downstream tasks. We evaluate TrafficLLM across 10 distinct scenarios and 229 types of traffic. TrafficLLM achieves F1-scores of 0.9875 and 0.9483, with up to 80.12% and 33.92% better performance than existing detection and generation methods. It also shows strong generalization on unseen traffic with an 18.6% performance improvement. We further evaluate TrafficLLM in real-world scenarios. The results confirm that TrafficLLM is easy to scale and achieves accurate detection performance on enterprise traffic.
♻ ☆ Towards Personalized Conversational Sales Agents : Contextual User Profiling for Strategic Action
Conversational Recommender Systems (CRSs) aim to engage users in dialogue to provide tailored recommendations. While traditional CRSs focus on eliciting preferences and retrieving items, real-world e-commerce interactions involve more complex decision-making, where users consider multiple factors beyond simple attributes. To bridge this gap, we introduce Conversational Sales (CSales), a novel task that unifies preference elicitation, recommendation, and persuasion to better support user decision-making. For a realistic evaluation of CSales, we present CSUser, an LLM-based user simulator constructed from real-world data, modeling diverse user profiles with needs and personalities. Additionally, we propose CSI, a conversational sales agent that proactively infers contextual profiles through dialogue for personalized action planning. Extensive experiments demonstrate that CSUser effectively replicates real-world users and emphasize the importance of contextual profiling for strategic action selection, ultimately driving successful purchases in e-commerce.
♻ ☆ Reinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters AAAI
We present a Reinforcement Learning Platform for Adversarial Black-box untargeted and targeted attacks, RLAB, that allows users to select from various distortion filters to create adversarial examples. The platform uses a Reinforcement Learning agent to add minimum distortion to input images while still causing misclassification by the target model. The agent uses a novel dual-action method to explore the input image at each step to identify sensitive regions for adding distortions while removing noises that have less impact on the target model. This dual action leads to faster and more efficient convergence of the attack. The platform can also be used to measure the robustness of image classification models against specific distortion types. Also, retraining the model with adversarial samples significantly improved robustness when evaluated on benchmark datasets. The proposed platform outperforms state-of-the-art methods in terms of the average number of queries required to cause misclassification. This advances trustworthiness with a positive social impact.
comment: Accepted at the 2025 AAAI Conference on Artificial Intelligence Proceedings
♻ ☆ Protecting Copyright of Medical Pre-trained Language Models: Training-Free Backdoor Model Watermarking
With the advancement of intelligent healthcare, medical pre-trained language models (Med-PLMs) have emerged and demonstrated significant effectiveness in downstream medical tasks. While these models are valuable assets, they are vulnerable to misuse and theft, requiring copyright protection. However, existing watermarking methods for pre-trained language models (PLMs) cannot be directly applied to Med-PLMs due to domain-task mismatch and inefficient watermark embedding. To fill this gap, we propose the first training-free backdoor model watermarking for Med-PLMs. Our method employs low-frequency words as triggers, embedding the watermark by replacing their embeddings in the model's word embedding layer with those of specific medical terms. The watermarked Med-PLMs produce the same output for triggers as for the corresponding specified medical terms. We leverage this unique mapping to design tailored watermark extraction schemes for different downstream tasks, thereby addressing the challenge of domain-task mismatch in previous methods. Experiments demonstrate superior effectiveness of our watermarking method across medical downstream tasks. Moreover, the method exhibits robustness against model extraction, pruning, fusion-based backdoor removal attacks, while maintaining high efficiency with 10-second watermark embedding.
comment: 9 pages
♻ ☆ ClinicalGPT-R1: Pushing reasoning capability of generalist disease diagnosis with large language model
Recent advances in reasoning with large language models (LLMs)has shown remarkable reasoning capabilities in domains such as mathematics and coding, yet their application to clinical diagnosis remains underexplored. Here, we introduce ClinicalGPT-R1, a reasoning enhanced generalist large language model for disease diagnosis. Trained on a dataset of 20,000 real-world clinical records, ClinicalGPT-R1 leverages diverse training strategies to enhance diagnostic reasoning. To benchmark performance, we curated MedBench-Hard, a challenging dataset spanning seven major medical specialties and representative diseases. Experimental results demonstrate that ClinicalGPT-R1 outperforms GPT-4o in Chinese diagnostic tasks and achieves comparable performance to GPT-4 in English settings. This comparative study effectively validates the superior performance of ClinicalGPT-R1 in disease diagnosis tasks. Resources are available at https://github.com/medfound/medfound.
comment: 8 pages, 6 figures
♻ ☆ Linear-MoE: Linear Sequence Modeling Meets Mixture-of-Experts
Linear Sequence Modeling (LSM) like linear attention, state space models and linear RNNs, and Mixture-of-Experts (MoE) have recently emerged as significant architectural improvements. In this paper, we introduce Linear-MoE, a production-level system for modeling and training large-scale models that integrate LSM with MoE. Linear-MoE leverages the advantages of both LSM modules for linear-complexity sequence modeling and MoE layers for sparsely activation, aiming to offer high performance with efficient training. The Linear-MoE system comprises: 1) Modeling subsystem, which provides a unified framework supporting all instances of LSM. and 2) Training subsystem, which facilitates efficient training by incorporating various advanced parallelism technologies, particularly Sequence Parallelism designed for Linear-MoE models. Additionally, we explore hybrid models that combine Linear-MoE layers with standard Transformer-MoE layers with its Sequence Parallelism to further enhance model flexibility and performance. Evaluations on two model series, A0.3B-2B and A1B-7B, demonstrate Linear-MoE achieves efficiency gains while maintaining competitive performance on various benchmarks, showcasing its potential as a next-generation foundational model architecture. Code: https://github.com/OpenSparseLLMs/Linear-MoE.
comment: Technical report, 17 pages
♻ ☆ MERGE$^3$: Efficient Evolutionary Merging on Consumer-grade GPUs
Evolutionary model merging enables the creation of high-performing multi-task models but remains computationally prohibitive for consumer hardware. We introduce MERGE$^3$, an efficient framework that makes evolutionary merging feasible on a single GPU by reducing fitness computation costs 50$\times$ while preserving performance. MERGE$^3$ achieves this by Extracting a reduced dataset for evaluation, Estimating model abilities using Item Response Theory (IRT), and Evolving optimal merges via IRT-based performance estimators. Our method enables state-of-the-art multilingual and cross-lingual merging, transferring knowledge across languages with significantly lower computational overhead. We provide theoretical guarantees and an open-source library, democratizing high-quality model merging.
comment: 19 pages, 13 figures
♻ ☆ Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute
This paper presents a simple, effective, and cost-efficient strategy to improve LLM performance by scaling test-time compute. Our strategy builds upon the repeated-sampling-then-voting framework, with a novel twist: incorporating multiple models, even weaker ones, to leverage their complementary strengths that potentially arise from diverse training data and paradigms. By using consistency as a signal, our strategy dynamically switches between models. Theoretical analysis highlights the efficiency and performance advantages of our strategy. Extensive experiments on six datasets demonstrate that our strategy not only outperforms self-consistency and state-of-the-art multi-agent debate approaches, but also significantly reduces inference costs. Additionally, ModelSwitch requires only a few comparable LLMs to achieve optimal performance and can be extended with verification methods, demonstrating the potential of leveraging multiple LLMs in the generation-verification paradigm.
♻ ☆ UniForm: A Unified Multi-Task Diffusion Transformer for Audio-Video Generation
With the rise of diffusion models, audio-video generation has been revolutionized. However, most existing methods rely on separate modules for each modality, with limited exploration of unified generative architectures. In addition, many are confined to a single task and small-scale datasets. To address these limitations, we first propose UniForm, a unified multi-task diffusion transformer that jointly generates audio and visual modalities in a shared latent space. A single diffusion process models both audio and video, capturing the inherent correlations between sound and vision. Second, we introduce task-specific noise schemes and task tokens, enabling a single model to support multiple tasks, including text-to-audio-video, audio-to-video, and video-to-audio generation. Furthermore, by leveraging large language models and a large-scale text-audio-video combined dataset, UniForm achieves greater generative diversity than prior approaches. Extensive experiments show that UniForm achieves the state-of-the-art performance across audio-video generation tasks, producing content that is both well-aligned and close to real-world data distributions. Our demos are available at https://uniform-t2av.github.io/.
comment: Our demos are available at https://uniform-t2av.github.io/
♻ ☆ MMKB-RAG: A Multi-Modal Knowledge-Based Retrieval-Augmented Generation Framework
Recent advancements in large language models (LLMs) and multi-modal LLMs have been remarkable. However, these models still rely solely on their parametric knowledge, which limits their ability to generate up-to-date information and increases the risk of producing erroneous content. Retrieval-Augmented Generation (RAG) partially mitigates these challenges by incorporating external data sources, yet the reliance on databases and retrieval systems can introduce irrelevant or inaccurate documents, ultimately undermining both performance and reasoning quality. In this paper, we propose Multi-Modal Knowledge-Based Retrieval-Augmented Generation (MMKB-RAG), a novel multi-modal RAG framework that leverages the inherent knowledge boundaries of models to dynamically generate semantic tags for the retrieval process. This strategy enables the joint filtering of retrieved documents, retaining only the most relevant and accurate references. Extensive experiments on knowledge-based visual question-answering tasks demonstrate the efficacy of our approach: on the E-VQA dataset, our method improves performance by +4.2% on the Single-Hop subset and +0.4% on the full dataset, while on the InfoSeek dataset, it achieves gains of +7.8% on the Unseen-Q subset, +8.2% on the Unseen-E subset, and +8.1% on the full dataset. These results highlight significant enhancements in both accuracy and robustness over the current state-of-the-art MLLM and RAG frameworks.
♻ ☆ Evolving Hard Maximum Cut Instances for Quantum Approximate Optimization Algorithms GECCO 2025
Variational quantum algorithms, such as the Recursive Quantum Approximate Optimization Algorithm (RQAOA), have become increasingly popular, offering promising avenues for employing Noisy Intermediate-Scale Quantum devices to address challenging combinatorial optimization tasks like the maximum cut problem. In this study, we utilize an evolutionary algorithm equipped with a unique fitness function. This approach targets hard maximum cut instances within the latent space of a Graph Autoencoder, identifying those that pose significant challenges or are particularly tractable for RQAOA, in contrast to the classic Goemans and Williamson algorithm. Our findings not only delineate the distinct capabilities and limitations of each algorithm but also expand our understanding of RQAOA's operational limits. Furthermore, the diverse set of graphs we have generated serves as a crucial benchmarking asset, emphasizing the need for more advanced algorithms to tackle combinatorial optimization challenges. Additionally, our results pave the way for new avenues in graph generation research, offering exciting opportunities for future explorations.
comment: This work has been accepted for publication and presentation at GECCO 2025
♻ ☆ Reinforcing Clinical Decision Support through Multi-Agent Systems and Ethical AI Governance
Recent advances in the data-driven medicine approach, which integrates ethically managed and explainable artificial intelligence into clinical decision support systems (CDSS), are critical to ensure reliable and effective patient care. This paper focuses on comparing novel agent system designs that use modular agents to analyze laboratory results, vital signs, and clinical context, and to predict and validate results. We implement our agent system with the eICU database, including running lab analysis, vitals-only interpreters, and contextual reasoners agents first, then sharing the memory into the integration agent, prediction agent, transparency agent, and a validation agent. Our results suggest that the multi-agent system (MAS) performed better than the single-agent system (SAS) with mortality prediction accuracy (59%, 56%) and the mean error for length of stay (LOS)(4.37 days, 5.82 days), respectively. However, the transparency score for the SAS (86.21) is slightly better than the transparency score for MAS (85.5). Finally, this study suggests that our agent-based framework not only improves process transparency and prediction accuracy but also strengthens trustworthy AI-assisted decision support in an intensive care setting.
♻ ☆ Unlocking the Wisdom of Large Language Models: An Introduction to The Path to Artificial General Intelligence
This booklet, Unlocking the Wisdom of Multi-LLM Collaborative Intelligence, serves as an accessible introduction to the full volume The Path to Artificial General Intelligence. Through fourteen aphorisms, it distills the core principles of Multi-LLM Agent Collaborative Intelligence (MACI), a framework designed to coordinate multiple LLMs toward reasoning, planning, and decision-making that surpasses the capabilities of any single model. The booklet includes titles, abstracts, and introductions from each main chapter, along with the full content of the first two. The newly released third edition features significant enhancements to Chapters 6 through 9 and a revised preface responding to Yann LeCun's critique of AGI feasibility. While LeCun argues that LLMs lack grounding, memory, and planning, we propose that MACI's collaborative architecture, featuring multimodal agents in executive, legislative, and judicial roles, directly addresses these limitations. Chapters on SocraSynth, EVINCE, consciousness modeling, and behavior regulation demonstrate that reasoning systems grounded in structured interaction and checks and balances can produce more reliable, interpretable, and adaptive intelligence. By integrating complementary model strengths, including world modeling and multimodal perception, MACI enables a system-level intelligence that exceeds the sum of its parts. Like human institutions, progress in AI may depend less on isolated performance and more on coordinated judgment. Collaborative LLMs, not just larger ones, may chart the path toward artificial general intelligence.
comment: 153 pages, 5 figures
♻ ☆ Fundamental Limits of Deep Learning-Based Binary Classifiers Trained with Hinge Loss
Although deep learning (DL) has led to several breakthroughs in many disciplines, the fundamental understanding on why and how DL is empirically successful remains elusive. To attack this fundamental problem and unravel the mysteries behind DL's empirical successes, significant innovations toward a unified theory of DL have been made. Although these innovations encompass nearly fundamental advances in optimization, generalization, and approximation, no work has quantified the testing performance of a DL-based algorithm employed to solve a pattern classification problem. To overcome this fundamental challenge in part, this paper exposes the fundamental testing performance limits of DL-based binary classifiers trained with hinge loss. For binary classifiers that are based on deep rectified linear unit (ReLU) feedforward neural networks (FNNs) and deep FNNs with ReLU and Tanh activation, we derive their respective novel asymptotic testing performance limits, which are validated by extensive computer experiments.
♻ ☆ Finite-Time Analysis of Temporal Difference Learning with Experience Replay
Temporal-difference (TD) learning is widely regarded as one of the most popular algorithms in reinforcement learning (RL). Despite its widespread use, it has only been recently that researchers have begun to actively study its finite time behavior, including the finite time bound on mean squared error and sample complexity. On the empirical side, experience replay has been a key ingredient in the success of deep RL algorithms, but its theoretical effects on RL have yet to be fully understood. In this paper, we present a simple decomposition of the Markovian noise terms and provide finite-time error bounds for TD-learning with experience replay. Specifically, under the Markovian observation model, we demonstrate that for both the averaged iterate and final iterate cases, the error term induced by a constant step-size can be effectively controlled by the size of the replay buffer and the mini-batch sampled from the experience replay buffer.
♻ ☆ FairPy: A Toolkit for Evaluation of Prediction Biases and their Mitigation in Large Language Models
Recent studies have demonstrated that large pretrained language models (LLMs) such as BERT and GPT-2 exhibit biases in token prediction, often inherited from the data distributions present in their training corpora. In response, a number of mathematical frameworks have been proposed to quantify, identify, and mitigate these the likelihood of biased token predictions. In this paper, we present a comprehensive survey of such techniques tailored towards widely used LLMs such as BERT, GPT-2, etc. We additionally introduce Fairpy, a modular and extensible toolkit that provides plug-and-play interfaces for integrating these mathematical tools, enabling users to evaluate both pretrained and custom language models. Fairpy supports the implementation of existing debiasing algorithms. The toolkit is open-source and publicly available at: \href{https://github.com/HrishikeshVish/Fairpy}{https://github.com/HrishikeshVish/Fairpy}
♻ ☆ Why Is Anything Conscious?
We tackle the hard problem of consciousness taking the naturally selected, embodied organism as our starting point. We provide a formalism describing how biological systems self-organise to hierarchically interpret unlabelled sensory information according to valence. Such interpretations imply behavioural policies which are differentiated from each other only by the qualitative aspect of information processing. Natural selection favours systems that intervene in the world to achieve homeostatic and reproductive goals. Quality is a property arising in such systems to link cause to affect to motivate interventions. This produces interoceptive and exteroceptive classifiers and determines priorities. In formalising the seminal distinction between access and phenomenal consciousness, we claim that access consciousness at the human level requires the ability to hierarchically model i) the self, ii) the world/others and iii) the self as modelled by others, and that this requires phenomenal consciousness. Phenomenal without access consciousness is likely common, but the reverse is implausible. To put it provocatively: death grounds meaning, and Nature does not like zombies. We then describe the multilayered architecture of self-organisation from rocks to Einstein, illustrating how our argument applies. Our proposal lays the foundation of a formal science of consciousness, closer to human fact than zombie fiction.
♻ ☆ LanguageMPC: Large Language Models as Decision Makers for Autonomous Driving
Existing learning-based autonomous driving (AD) systems face challenges in comprehending high-level information, generalizing to rare events, and providing interpretability. To address these problems, this work employs Large Language Models (LLMs) as a decision-making component for complex AD scenarios that require human commonsense understanding. We devise cognitive pathways to enable comprehensive reasoning with LLMs, and develop algorithms for translating LLM decisions into actionable driving commands. Through this approach, LLM decisions are seamlessly integrated with low-level controllers by guided parameter matrix adaptation. Extensive experiments demonstrate that our proposed method not only consistently surpasses baseline approaches in single-vehicle tasks, but also helps handle complex driving behaviors even multi-vehicle coordination, thanks to the commonsense reasoning capabilities of LLMs. This paper presents an initial step toward leveraging LLMs as effective decision-makers for intricate AD scenarios in terms of safety, efficiency, generalizability, and interoperability. We aspire for it to serve as inspiration for future research in this field. Project page: https://sites.google.com/view/llm-mpc
♻ ☆ System-1.x: Learning to Balance Fast and Slow Planning with Language Models ICLR 2025
Language models can be used to solve long-horizon planning problems in two distinct modes: a fast 'System-1' mode, directly generating plans without any explicit search or backtracking, and a slow 'System-2' mode, planning step-by-step by explicitly searching over possible actions. While System-2 is typically more effective, it is also more computationally expensive, making it infeasible for long plans or large action spaces. Moreover, isolated System-1 or 2 ignores the user's end goals, failing to provide ways to control the model's behavior. To this end, we propose the System-1.x Planner, a controllable planning framework with LLMs that is capable of generating hybrid plans and balancing between the two planning modes based on the difficulty of the problem at hand. System-1.x consists of (i) a controller, (ii) a System-1 Planner, and (iii) a System-2 Planner. Based on a user-specified hybridization factor (x) governing the mixture between System-1 and 2, the controller decomposes a problem into sub-goals, and classifies them as easy or hard to be solved by either System-1 or 2, respectively. We fine-tune all three components on top of a single base LLM, requiring only search traces as supervision. Experiments with two diverse planning tasks -- Maze Navigation and Blocksworld -- show that our System-1.x Planner outperforms a System-1 Planner, a System-2 Planner trained to approximate A* search, and also a symbolic planner (A*). We demonstrate the following key properties of our planner: (1) controllability: increasing the hybridization factor (e.g., System-1.75 vs 1.5) performs more search, improving performance, (2) flexibility: by building a neuro-symbolic variant with a neural System-1 and a symbolic System-2, we can use existing symbolic methods, and (3) generalizability: by being able to learn from different search algorithms, our method is robust to the choice of search algorithm.
comment: ICLR 2025 (Camera-Ready)
♻ ☆ Which Optimizer Works Best for Physics-Informed Neural Networks and Kolmogorov-Arnold Networks?
Physics-Informed Neural Networks (PINNs) have revolutionized the computation of PDE solutions by integrating partial differential equations (PDEs) into the neural network's training process as soft constraints, becoming an important component of the scientific machine learning (SciML) ecosystem. More recently, physics-informed Kolmogorv-Arnold networks (PIKANs) have also shown to be effective and comparable in accuracy with PINNs. In their current implementation, both PINNs and PIKANs are mainly optimized using first-order methods like Adam, as well as quasi-Newton methods such as BFGS and its low-memory variant, L-BFGS. However, these optimizers often struggle with highly non-linear and non-convex loss landscapes, leading to challenges such as slow convergence, local minima entrapment, and (non)degenerate saddle points. In this study, we investigate the performance of Self-Scaled BFGS (SSBFGS), Self-Scaled Broyden (SSBroyden) methods and other advanced quasi-Newton schemes, including BFGS and L-BFGS with different line search strategies approaches. These methods dynamically rescale updates based on historical gradient information, thus enhancing training efficiency and accuracy. We systematically compare these optimizers -- using both PINNs and PIKANs -- on key challenging linear, stiff, multi-scale and non-linear PDEs, including the Burgers, Allen-Cahn, Kuramoto-Sivashinsky, and Ginzburg-Landau equations. Our findings provide state-of-the-art results with orders-of-magnitude accuracy improvements without the use of adaptive weights or any other enhancements typically employed in PINNs. More broadly, our results reveal insights into the effectiveness of second-order optimization strategies in significantly improving the convergence and accurate generalization of PINNs and PIKANs.
comment: 36 pages, 27 figures
♻ ☆ IAA: Inner-Adaptor Architecture Empowers Frozen Large Language Model with Multimodal Capabilities AAAI 2025
In the field of multimodal large language models (MLLMs), common methods typically involve unfreezing the language model during training to foster profound visual understanding. However, the fine-tuning of such models with vision-language data often leads to a diminution of their natural language processing (NLP) capabilities. To avoid this performance degradation, a straightforward solution is to freeze the language model while developing multimodal competencies. Unfortunately, previous works have not attained satisfactory outcomes. Building on the strategy of freezing the language model, we conduct thorough structural exploration and introduce the Inner-Adaptor Architecture (IAA). Specifically, the architecture incorporates multiple multimodal adaptors at varying depths within the large language model to facilitate direct interaction with the inherently text-oriented transformer layers, thereby enabling the frozen language model to acquire multimodal capabilities. Unlike previous approaches of freezing language models that require large-scale aligned data, our proposed architecture is able to achieve superior performance on small-scale datasets. We conduct extensive experiments to improve the general multimodal capabilities and visual grounding abilities of the MLLM. Our approach remarkably outperforms previous state-of-the-art methods across various vision-language benchmarks without sacrificing performance on NLP tasks. Code and models are available at https://github.com/360CVGroup/Inner-Adaptor-Architecture.
comment: AAAI 2025
♻ ☆ AI Mismatches: Identifying Potential Algorithmic Harms Before AI Development
AI systems are often introduced with high expectations, yet many fail to deliver, resulting in unintended harm and missed opportunities for benefit. We frequently observe significant "AI Mismatches", where the system's actual performance falls short of what is needed to ensure safety and co-create value. These mismatches are particularly difficult to address once development is underway, highlighting the need for early-stage intervention. Navigating complex, multi-dimensional risk factors that contribute to AI Mismatches is a persistent challenge. To address it, we propose an AI Mismatch approach to anticipate and mitigate risks early on, focusing on the gap between realistic model performance and required task performance. Through an analysis of 774 AI cases, we extracted a set of critical factors, which informed the development of seven matrices that map the relationships between these factors and highlight high-risk areas. Through case studies, we demonstrate how our approach can help reduce risks in AI development.
comment: CHI Conference on Human Factors in Computing Systems (CHI '25), April 26-May 1, 2025, Yokohama, Japan
♻ ☆ F$^3$Set: Towards Analyzing Fast, Frequent, and Fine-grained Events from Videos ICLR 2025
Analyzing Fast, Frequent, and Fine-grained (F$^3$) events presents a significant challenge in video analytics and multi-modal LLMs. Current methods struggle to identify events that satisfy all the F$^3$ criteria with high accuracy due to challenges such as motion blur and subtle visual discrepancies. To advance research in video understanding, we introduce F$^3$Set, a benchmark that consists of video datasets for precise F$^3$ event detection. Datasets in F$^3$Set are characterized by their extensive scale and comprehensive detail, usually encompassing over 1,000 event types with precise timestamps and supporting multi-level granularity. Currently, F$^3$Set contains several sports datasets, and this framework may be extended to other applications as well. We evaluated popular temporal action understanding methods on F$^3$Set, revealing substantial challenges for existing techniques. Additionally, we propose a new method, F$^3$ED, for F$^3$ event detections, achieving superior performance. The dataset, model, and benchmark code are available at https://github.com/F3Set/F3Set.
comment: ICLR 2025; Website URL: https://lzyandy.github.io/f3set-website/
♻ ☆ Large language models could be rote learners
Multiple-choice question (MCQ) benchmarks are widely used for evaluating Large Language Models (LLMs), yet their reliability is undermined by benchmark contamination. In this study, we reframe contamination as an inherent aspect of learning and seek to disentangle genuine capability acquisition from superficial memorization in LLM evaluation. First, by analyzing model performance under different memorization conditions, we uncover a counterintuitive trend: LLMs perform worse on memorized MCQs than on non-memorized ones, indicating the coexistence of two distinct learning phenomena, i.e., rote memorization and genuine capability learning. To disentangle them, we propose TrinEval, a novel evaluation framework that reformulates MCQs into an alternative trinity format, reducing memorization while preserving knowledge assessment. Experiments validate TrinEval's effectiveness in reformulation, and its evaluation reveals that common LLMs may memorize by rote 20.5% of knowledge points (in MMLU on average).
comment: Work in Progress
♻ ☆ DeepResearcher: Scaling Deep Research via Reinforcement Learning in Real-world Environments
Large Language Models (LLMs) equipped with web search capabilities have demonstrated impressive potential for deep research tasks. However, current approaches predominantly rely on either manually engineered prompts (prompt engineering-based) with brittle performance or reinforcement learning within controlled Retrieval-Augmented Generation (RAG) environments (RAG-based) that fail to capture the complexities of real-world interaction. In this paper, we introduce DeepResearcher, the first comprehensive framework for end-to-end training of LLM-based deep research agents through scaling reinforcement learning (RL) in real-world environments with authentic web search interactions. Unlike RAG-based approaches that assume all necessary information exists within a fixed corpus, our method trains agents to navigate the noisy, unstructured, and dynamic nature of the open web. We implement a specialized multi-agent architecture where browsing agents extract relevant information from various webpage structures and overcoming significant technical challenges. Extensive experiments on open-domain research tasks demonstrate that DeepResearcher achieves substantial improvements of up to 28.9 points over prompt engineering-based baselines and up to 7.2 points over RAG-based RL agents. Our qualitative analysis reveals emergent cognitive behaviors from end-to-end RL training, including the ability to formulate plans, cross-validate information from multiple sources, engage in self-reflection to redirect research, and maintain honesty when unable to find definitive answers. Our results highlight that end-to-end training in real-world web environments is not merely an implementation detail but a fundamental requirement for developing robust research capabilities aligned with real-world applications. We release DeepResearcher at https://github.com/GAIR-NLP/DeepResearcher.
♻ ☆ Bridging Stepwise Lab-Informed Pretraining and Knowledge-Guided Learning for Diagnostic Reasoning
Despite the growing use of Electronic Health Records (EHR) for AI-assisted diagnosis prediction, most data-driven models struggle to incorporate clinically meaningful medical knowledge. They often rely on limited ontologies, lacking structured reasoning capabilities and comprehensive coverage. This raises an important research question: Will medical knowledge improve predictive models to support stepwise clinical reasoning as performed by human doctors? To address this problem, we propose DuaLK, a dual-expertise framework that combines two complementary sources of information. For external knowledge, we construct a Diagnosis Knowledge Graph (KG) that encodes both hierarchical and semantic relations enriched by large language models (LLM). To align with patient data, we further introduce a lab-informed proxy task that guides the model to follow a clinically consistent, stepwise reasoning process based on lab test signals. Experimental results on two public EHR datasets demonstrate that DuaLK consistently outperforms existing baselines across four clinical prediction tasks. These findings highlight the potential of combining structured medical knowledge with individual-level clinical signals to achieve more accurate and interpretable diagnostic predictions. The source code is publicly available on https://github.com/humphreyhuu/DuaLK.
♻ ☆ Towards Symmetric Low-Rank Adapters
In this paper, we introduce Symmetric Low-Rank Adapters, an optimized variant of LoRA with even fewer weights. This method utilizes Low-Rank Symmetric Weight Matrices to learn downstream tasks more efficiently. Traditional LoRA accumulates fine-tuning weights with the original pre-trained weights via a Singular Value Decomposition (SVD) like approach, i.e., model weights are fine-tuned via updates of the form $BA$ (where $B \in \mathbb{R}^{n\times r}$, $A \in \mathbb{R}^{r\times n}$, and $r$ is the rank of the merged weight matrix). In contrast, our approach, named SymLoRA, represents fine-tuning weights as a Spectral Decomposition, i.e., $Q \, diag(\Lambda)\, Q^T$, where $Q \in \mathbb{R}^{n\times r}$ and $\Lambda \in \mathbb{R}^r$. SymLoRA requires approximately half of the finetuning weights. Here, we show that this approach has negligible losses in downstream efficacy.
comment: Colorai Workshop
♻ ☆ Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework
Multimodal foundation models offer a promising framework for robotic perception and planning by processing sensory inputs to generate actionable plans. However, addressing uncertainty in both perception (sensory interpretation) and decision-making (plan generation) remains a critical challenge for ensuring task reliability. We present a comprehensive framework to disentangle, quantify, and mitigate these two forms of uncertainty. We first introduce a framework for uncertainty disentanglement, isolating perception uncertainty arising from limitations in visual understanding and decision uncertainty relating to the robustness of generated plans. To quantify each type of uncertainty, we propose methods tailored to the unique properties of perception and decision-making: we use conformal prediction to calibrate perception uncertainty and introduce Formal-Methods-Driven Prediction (FMDP) to quantify decision uncertainty, leveraging formal verification techniques for theoretical guarantees. Building on this quantification, we implement two targeted intervention mechanisms: an active sensing process that dynamically re-observes high-uncertainty scenes to enhance visual input quality and an automated refinement procedure that fine-tunes the model on high-certainty data, improving its capability to meet task specifications. Empirical validation in real-world and simulated robotic tasks demonstrates that our uncertainty disentanglement framework reduces variability by up to 40% and enhances task success rates by 5% compared to baselines. These improvements are attributed to the combined effect of both interventions and highlight the importance of uncertainty disentanglement, which facilitates targeted interventions that enhance the robustness and reliability of autonomous systems. Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/.
comment: Fine-tuned models, code, and datasets are available at https://uncertainty-in-planning.github.io/
♻ ☆ SuperARC: An Agnostic Test for Narrow, General, and Super Intelligence Based On the Principles of Recursive Compression and Algorithmic Probability
We introduce an open-ended test grounded in algorithmic probability that can avoid benchmark contamination in the quantitative evaluation of frontier models in the context of their Artificial General Intelligence (AGI) and Superintelligence (ASI) claims. Unlike other tests, this test does not rely on statistical compression methods (such as GZIP or LZW), which are more closely related to Shannon entropy than to Kolmogorov complexity and are not able to test beyond simple pattern matching. The test challenges aspects of AI, in particular LLMs, related to features of intelligence of fundamental nature such as synthesis and model creation in the context of inverse problems (generating new knowledge from observation). We argue that metrics based on model abstraction and abduction (optimal Bayesian `inference') for predictive `planning' can provide a robust framework for testing intelligence, including natural intelligence (human and animal), narrow AI, AGI, and ASI. We found that LLM model versions tend to be fragile and incremental as a result of memorisation only with progress likely driven by the size of training data. The results were compared with a hybrid neurosymbolic approach that theoretically guarantees universal intelligence based on the principles of algorithmic probability and Kolmogorov complexity. The method outperforms LLMs in a proof-of-concept on short binary sequences. We prove that compression is equivalent and directly proportional to a system's predictive power and vice versa. That is, if a system can better predict it can better compress, and if it can better compress, then it can better predict. Our findings strengthen the suspicion regarding the fundamental limitations of LLMs, exposing them as systems optimised for the perception of mastery over human language.
comment: 51 pages + Technical Supplementary Information, 79 pages total
♻ ☆ ProtoECGNet: Case-Based Interpretable Deep Learning for Multi-Label ECG Classification with Contrastive Learning
Deep learning-based electrocardiogram (ECG) classification has shown impressive performance but clinical adoption has been slowed by the lack of transparent and faithful explanations. Post hoc methods such as saliency maps may fail to reflect a model's true decision process. Prototype-based reasoning offers a more transparent alternative by grounding decisions in similarity to learned representations of real ECG segments, enabling faithful, case-based explanations. We introduce ProtoECGNet, a prototype-based deep learning model for interpretable, multi-label ECG classification. ProtoECGNet employs a structured, multi-branch architecture that reflects clinical interpretation workflows: it integrates a 1D CNN with global prototypes for rhythm classification, a 2D CNN with time-localized prototypes for morphology-based reasoning, and a 2D CNN with global prototypes for diffuse abnormalities. Each branch is trained with a prototype loss designed for multi-label learning, combining clustering, separation, diversity, and a novel contrastive loss that encourages appropriate separation between prototypes of unrelated classes while allowing clustering for frequently co-occurring diagnoses. We evaluate ProtoECGNet on all 71 diagnostic labels from the PTB-XL dataset, demonstrating competitive performance relative to state-of-the-art black-box models while providing structured, case-based explanations. To assess prototype quality, we conduct a structured clinician review of the final model's projected prototypes, finding that they are rated as representative and clear. ProtoECGNet shows that prototype learning can be effectively scaled to complex, multi-label time-series classification, offering a practical path toward transparent and trustworthy deep learning models for clinical decision support.
♻ ☆ Evaluating the Propensity of Generative AI for Producing Harmful Disinformation During an Election Cycle
Generative Artificial Intelligence offers a powerful tool for adversaries who wish to engage in influence operations, such as the Chinese Spamouflage operation and the Russian Internet Research Agency effort that both sought to interfere with recent US election cycles. Therefore, this study seeks to investigate the propensity of current generative AI models for producing harmful disinformation during an election cycle. The probability that different generative AI models produced disinformation when given adversarial prompts was evaluated, in addition to the associated harm. This allows for the expected harm for each model to be computed and it was discovered that Copilot and Gemini tied for the overall safest performance by realizing the lowest expected harm, while GPT-4o produced the greatest rates of harmful disinformation, resulting in much higher expected harm scores. The impact of disinformation category was also investigated and Gemini was safest within the political category of disinformation due to mitigation attempts made by developers during the election, while Copilot was safest for topics related to health. Moreover, characteristics of adversarial roles were discovered that led to greater expected harm across all models. Finally, classification models were developed that predicted disinformation production based on the conditions considered in this study, which offers insight into factors important for predicting disinformation production. Based on all of these insights, recommendations are provided that seek to mitigate factors that lead to harmful disinformation being produced by generative AI models. It is hoped that developers will use these insights to improve future models.
♻ ☆ Knowledge Graph Reasoning with Self-supervised Reinforcement Learning
Reinforcement learning (RL) is an effective method of finding reasoning pathways in incomplete knowledge graphs (KGs). To overcome the challenges of a large action space, a self-supervised pre-training method is proposed to warm up the policy network before the RL training stage. To alleviate the distributional mismatch issue in general self-supervised RL (SSRL), in our supervised learning (SL) stage, the agent selects actions based on the policy network and learns from generated labels; this self-generation of labels is the intuition behind the name self-supervised. With this training framework, the information density of our SL objective is increased and the agent is prevented from getting stuck with the early rewarded paths. Our self-supervised RL (SSRL) method improves the performance of RL by pairing it with the wide coverage achieved by SL during pretraining, since the breadth of the SL objective makes it infeasible to train an agent with that alone. We show that our SSRL model meets or exceeds current state-of-the-art results on all Hits@k and mean reciprocal rank (MRR) metrics on four large benchmark KG datasets. This SSRL method can be used as a plug-in for any RL architecture for a KGR task. We adopt two RL architectures, i.e., MINERVA and MultiHopKG as our baseline RL models and experimentally show that our SSRL model consistently outperforms both baselines on all of these four KG reasoning tasks. Full code for the paper available at https://github.com/owenonline/Knowledge-Graph-Reasoning-with-Self-supervised-Reinforcement-Learning.
comment: 17 pages, 11 figures
Computer Vision and Pattern Recognition 150
☆ Aligning Generative Denoising with Discriminative Objectives Unleashes Diffusion for Visual Perception ICLR 2025
With the success of image generation, generative diffusion models are increasingly adopted for discriminative tasks, as pixel generation provides a unified perception interface. However, directly repurposing the generative denoising process for discriminative objectives reveals critical gaps rarely addressed previously. Generative models tolerate intermediate sampling errors if the final distribution remains plausible, but discriminative tasks require rigorous accuracy throughout, as evidenced in challenging multi-modal tasks like referring image segmentation. Motivated by this gap, we analyze and enhance alignment between generative diffusion processes and perception tasks, focusing on how perception quality evolves during denoising. We find: (1) earlier denoising steps contribute disproportionately to perception quality, prompting us to propose tailored learning objectives reflecting varying timestep contributions; (2) later denoising steps show unexpected perception degradation, highlighting sensitivity to training-denoising distribution shifts, addressed by our diffusion-tailored data augmentation; and (3) generative processes uniquely enable interactivity, serving as controllable user interfaces adaptable to correctional prompts in multi-round interactions. Our insights significantly improve diffusion-based perception models without architectural changes, achieving state-of-the-art performance on depth estimation, referring image segmentation, and generalist perception tasks. Code available at https://github.com/ziqipang/ADDP.
comment: ICLR 2025
☆ SimpleAR: Pushing the Frontier of Autoregressive Visual Generation through Pretraining, SFT, and RL
This work presents SimpleAR, a vanilla autoregressive visual generation framework without complex architecure modifications. Through careful exploration of training and inference optimization, we demonstrate that: 1) with only 0.5B parameters, our model can generate 1024x1024 resolution images with high fidelity, and achieve competitive results on challenging text-to-image benchmarks, e.g., 0.59 on GenEval and 79.66 on DPG; 2) both supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO) training could lead to significant improvements on generation aesthectics and prompt alignment; and 3) when optimized with inference acceleraton techniques like vLLM, the time for SimpleAR to generate an 1024x1024 image could be reduced to around 14 seconds. By sharing these findings and open-sourcing the code, we hope to reveal the potential of autoregressive visual generation and encourage more participation in this research field. Code is available at https://github.com/wdrink/SimpleAR.
comment: technical report, work in progress
☆ PARTFIELD: Learning 3D Feature Fields for Part Segmentation and Beyond
We propose PartField, a feedforward approach for learning part-based 3D features, which captures the general concept of parts and their hierarchy without relying on predefined templates or text-based names, and can be applied to open-world 3D shapes across various modalities. PartField requires only a 3D feedforward pass at inference time, significantly improving runtime and robustness compared to prior approaches. Our model is trained by distilling 2D and 3D part proposals from a mix of labeled datasets and image segmentations on large unsupervised datasets, via a contrastive learning formulation. It produces a continuous feature field which can be clustered to yield a hierarchical part decomposition. Comparisons show that PartField is up to 20% more accurate and often orders of magnitude faster than other recent class-agnostic part-segmentation methods. Beyond single-shape part decomposition, consistency in the learned field emerges across shapes, enabling tasks such as co-segmentation and correspondence, which we demonstrate in several applications of these general-purpose, hierarchical, and consistent 3D feature fields. Check our Webpage! https://research.nvidia.com/labs/toronto-ai/partfield-release/
comment: https://research.nvidia.com/labs/toronto-ai/partfield-release/
☆ Diffusion Distillation With Direct Preference Optimization For Efficient 3D LiDAR Scene Completion
The application of diffusion models in 3D LiDAR scene completion is limited due to diffusion's slow sampling speed. Score distillation accelerates diffusion sampling but with performance degradation, while post-training with direct policy optimization (DPO) boosts performance using preference data. This paper proposes Distillation-DPO, a novel diffusion distillation framework for LiDAR scene completion with preference aligment. First, the student model generates paired completion scenes with different initial noises. Second, using LiDAR scene evaluation metrics as preference, we construct winning and losing sample pairs. Such construction is reasonable, since most LiDAR scene metrics are informative but non-differentiable to be optimized directly. Third, Distillation-DPO optimizes the student model by exploiting the difference in score functions between the teacher and student models on the paired completion scenes. Such procedure is repeated until convergence. Extensive experiments demonstrate that, compared to state-of-the-art LiDAR scene completion diffusion models, Distillation-DPO achieves higher-quality scene completion while accelerating the completion speed by more than 5-fold. Our method is the first to explore adopting preference learning in distillation to the best of our knowledge and provide insights into preference-aligned distillation. Our code is public available on https://github.com/happyw1nd/DistillationDPO.
comment: Our code is public available on https://github.com/happyw1nd/DistillationDPO
☆ TADACap: Time-series Adaptive Domain-Aware Captioning
While image captioning has gained significant attention, the potential of captioning time-series images, prevalent in areas like finance and healthcare, remains largely untapped. Existing time-series captioning methods typically offer generic, domain-agnostic descriptions of time-series shapes and struggle to adapt to new domains without substantial retraining. To address these limitations, we introduce TADACap, a retrieval-based framework to generate domain-aware captions for time-series images, capable of adapting to new domains without retraining. Building on TADACap, we propose a novel retrieval strategy that retrieves diverse image-caption pairs from a target domain database, namely TADACap-diverse. We benchmarked TADACap-diverse against state-of-the-art methods and ablation variants. TADACap-diverse demonstrates comparable semantic accuracy while requiring significantly less annotation effort.
comment: Accepted to ICAIF 2024
☆ Mamba-Based Ensemble learning for White Blood Cell Classification
White blood cell (WBC) classification assists in assessing immune health and diagnosing various diseases, yet manual classification is labor-intensive and prone to inconsistencies. Recent advancements in deep learning have shown promise over traditional methods; however, challenges such as data imbalance and the computational demands of modern technologies, such as Transformer-based models which do not scale well with input size, limit their practical application. This paper introduces a novel framework that leverages Mamba models integrated with ensemble learning to improve WBC classification. Mamba models, known for their linear complexity, provide a scalable alternative to Transformer-based approaches, making them suitable for deployment in resource-constrained environments. Additionally, we introduce a new WBC dataset, Chula-WBC-8, for benchmarking. Our approach not only validates the effectiveness of Mamba models in this domain but also demonstrates their potential to significantly enhance classification efficiency without compromising accuracy. The source code can be found at https://github.com/LewisClifton/Mamba-WBC-Classification.
☆ Enhancing Out-of-Distribution Detection with Extended Logit Normalization
Out-of-distribution (OOD) detection is essential for the safe deployment of machine learning models. Recent advances have explored improved classification losses and representation learning strategies to enhance OOD detection. However, these methods are often tailored to specific post-hoc detection techniques, limiting their generalizability. In this work, we identify a critical issue in Logit Normalization (LogitNorm), which inhibits its effectiveness in improving certain post-hoc OOD detection methods. To address this, we propose Extended Logit Normalization ($\textbf{ELogitNorm}$), a novel hyperparameter-free formulation that significantly benefits a wide range of post-hoc detection methods. By incorporating feature distance-awareness to LogitNorm, $\textbf{ELogitNorm}$ shows more robust OOD separability and in-distribution (ID) confidence calibration than its predecessor. Extensive experiments across standard benchmarks demonstrate that our approach outperforms state-of-the-art training-time methods in OOD detection while maintaining strong ID classification accuracy.
☆ NormalCrafter: Learning Temporally Consistent Normals from Video Diffusion Priors
Surface normal estimation serves as a cornerstone for a spectrum of computer vision applications. While numerous efforts have been devoted to static image scenarios, ensuring temporal coherence in video-based normal estimation remains a formidable challenge. Instead of merely augmenting existing methods with temporal components, we present NormalCrafter to leverage the inherent temporal priors of video diffusion models. To secure high-fidelity normal estimation across sequences, we propose Semantic Feature Regularization (SFR), which aligns diffusion features with semantic cues, encouraging the model to concentrate on the intrinsic semantics of the scene. Moreover, we introduce a two-stage training protocol that leverages both latent and pixel space learning to preserve spatial accuracy while maintaining long temporal context. Extensive evaluations demonstrate the efficacy of our method, showcasing a superior performance in generating temporally consistent normal sequences with intricate details from diverse videos.
comment: 9 pages, 6 figures, Project Page: https://normalcrafter.github.io/
☆ ADT: Tuning Diffusion Models with Adversarial Supervision
Diffusion models have achieved outstanding image generation by reversing a forward noising process to approximate true data distributions. During training, these models predict diffusion scores from noised versions of true samples in a single forward pass, while inference requires iterative denoising starting from white noise. This training-inference divergences hinder the alignment between inference and training data distributions, due to potential prediction biases and cumulative error accumulation. To address this problem, we propose an intuitive but effective fine-tuning framework, called Adversarial Diffusion Tuning (ADT), by stimulating the inference process during optimization and aligning the final outputs with training data by adversarial supervision. Specifically, to achieve robust adversarial training, ADT features a siamese-network discriminator with a fixed pre-trained backbone and lightweight trainable parameters, incorporates an image-to-image sampling strategy to smooth discriminative difficulties, and preserves the original diffusion loss to prevent discriminator hacking. In addition, we carefully constrain the backward-flowing path for back-propagating gradients along the inference path without incurring memory overload or gradient explosion. Finally, extensive experiments on Stable Diffusion models (v1.5, XL, and v3), demonstrate that ADT significantly improves both distribution alignment and image quality.
☆ Leveraging Point Transformers for Detecting Anatomical Landmarks in Digital Dentistry MICCAI2024
The increasing availability of intraoral scanning devices has heightened their importance in modern clinical orthodontics. Clinicians utilize advanced Computer-Aided Design techniques to create patient-specific treatment plans that include laboriously identifying crucial landmarks such as cusps, mesial-distal locations, facial axis points, and tooth-gingiva boundaries. Detecting such landmarks automatically presents challenges, including limited dataset sizes, significant anatomical variability among subjects, and the geometric nature of the data. We present our experiments from the 3DTeethLand Grand Challenge at MICCAI 2024. Our method leverages recent advancements in point cloud learning through transformer architectures. We designed a Point Transformer v3 inspired module to capture meaningful geometric and anatomical features, which are processed by a lightweight decoder to predict per-point distances, further processed by graph-based non-minima suppression. We report promising results and discuss insights on learned feature interpretability.
comment: 10 pages + references, 3 figures, MICCAI2024 3DTeethland Challenge submission
☆ Deep Learning-based Bathymetry Retrieval without In-situ Depths using Remote Sensing Imagery and SfM-MVS DSMs with Data Gaps SP
Accurate, detailed, and high-frequent bathymetry is crucial for shallow seabed areas facing intense climatological and anthropogenic pressures. Current methods utilizing airborne or satellite optical imagery to derive bathymetry primarily rely on either SfM-MVS with refraction correction or Spectrally Derived Bathymetry (SDB). However, SDB methods often require extensive manual fieldwork or costly reference data, while SfM-MVS approaches face challenges even after refraction correction. These include depth data gaps and noise in environments with homogeneous visual textures, which hinder the creation of accurate and complete Digital Surface Models (DSMs) of the seabed. To address these challenges, this work introduces a methodology that combines the high-fidelity 3D reconstruction capabilities of the SfM-MVS methods with state-of-the-art refraction correction techniques, along with the spectral analysis capabilities of a new deep learning-based method for bathymetry prediction. This integration enables a synergistic approach where SfM-MVS derived DSMs with data gaps are used as training data to generate complete bathymetric maps. In this context, we propose Swin-BathyUNet that combines U-Net with Swin Transformer self-attention layers and a cross-attention mechanism, specifically tailored for SDB. Swin-BathyUNet is designed to improve bathymetric accuracy by capturing long-range spatial relationships and can also function as a standalone solution for standard SDB with various training depth data, independent of the SfM-MVS output. Experimental results in two completely different test sites in the Mediterranean and Baltic Seas demonstrate the effectiveness of the proposed approach through extensive experiments that demonstrate improvements in bathymetric accuracy, detail, coverage, and noise reduction in the predicted DSM. The code is available at https://github.com/pagraf/Swin-BathyUNet.
comment: Accepted for publication in ISPRS Journal of Photogrammetry and Remote Sensing
☆ Robustness and sex differences in skin cancer detection: logistic regression vs CNNs
Deep learning has been reported to achieve high performances in the detection of skin cancer, yet many challenges regarding the reproducibility of results and biases remain. This study is a replication (different data, same analysis) of a study on Alzheimer's disease [28] which studied robustness of logistic regression (LR) and convolutional neural networks (CNN) across patient sexes. We explore sex bias in skin cancer detection, using the PAD-UFES-20 dataset with LR trained on handcrafted features reflecting dermatological guidelines (ABCDE and the 7-point checklist), and a pre-trained ResNet-50 model. We evaluate these models in alignment with [28]: across multiple training datasets with varied sex composition to determine their robustness. Our results show that both the LR and the CNN were robust to the sex distributions, but the results also revealed that the CNN had a significantly higher accuracy (ACC) and area under the receiver operating characteristics (AUROC) for male patients than for female patients. We hope these findings to contribute to the growing field of investigating potential bias in popular medical machine learning methods. The data and relevant scripts to reproduce our results can be found in our Github.
comment: 16 pages (excluding appendix), 2 figures (excluding appendix), submitted to MIUA 2025 conference (response pending)
☆ Multi-level Cellular Automata for FLIM networks
The necessity of abundant annotated data and complex network architectures presents a significant challenge in deep-learning Salient Object Detection (deep SOD) and across the broader deep-learning landscape. This challenge is particularly acute in medical applications in developing countries with limited computational resources. Combining modern and classical techniques offers a path to maintaining competitive performance while enabling practical applications. Feature Learning from Image Markers (FLIM) methodology empowers experts to design convolutional encoders through user-drawn markers, with filters learned directly from these annotations. Recent findings demonstrate that coupling a FLIM encoder with an adaptive decoder creates a flyweight network suitable for SOD, requiring significantly fewer parameters than lightweight models and eliminating the need for backpropagation. Cellular Automata (CA) methods have proven successful in data-scarce scenarios but require proper initialization -- typically through user input, priors, or randomness. We propose a practical intersection of these approaches: using FLIM networks to initialize CA states with expert knowledge without requiring user interaction for each image. By decoding features from each level of a FLIM network, we can initialize multiple CAs simultaneously, creating a multi-level framework. Our method leverages the hierarchical knowledge encoded across different network layers, merging multiple saliency maps into a high-quality final output that functions as a CA ensemble. Benchmarks across two challenging medical datasets demonstrate the competitiveness of our multi-level CA approach compared to established models in the deep SOD literature.
☆ VideoPanda: Video Panoramic Diffusion with Multi-view Attention
High resolution panoramic video content is paramount for immersive experiences in Virtual Reality, but is non-trivial to collect as it requires specialized equipment and intricate camera setups. In this work, we introduce VideoPanda, a novel approach for synthesizing 360$^\circ$ videos conditioned on text or single-view video data. VideoPanda leverages multi-view attention layers to augment a video diffusion model, enabling it to generate consistent multi-view videos that can be combined into immersive panoramic content. VideoPanda is trained jointly using two conditions: text-only and single-view video, and supports autoregressive generation of long-videos. To overcome the computational burden of multi-view video generation, we randomly subsample the duration and camera views used during training and show that the model is able to gracefully generalize to generating more frames during inference. Extensive evaluations on both real-world and synthetic video datasets demonstrate that VideoPanda generates more realistic and coherent 360$^\circ$ panoramas across all input conditions compared to existing methods. Visit the project website at https://research-staging.nvidia.com/labs/toronto-ai/VideoPanda/ for results.
comment: Project website at https://research-staging.nvidia.com/labs/toronto-ai/VideoPanda/
☆ Omni$^2$: Unifying Omnidirectional Image Generation and Editing in an Omni Model
$360^{\circ}$ omnidirectional images (ODIs) have gained considerable attention recently, and are widely used in various virtual reality (VR) and augmented reality (AR) applications. However, capturing such images is expensive and requires specialized equipment, making ODI synthesis increasingly important. While common 2D image generation and editing methods are rapidly advancing, these models struggle to deliver satisfactory results when generating or editing ODIs due to the unique format and broad 360$^{\circ}$ Field-of-View (FoV) of ODIs. To bridge this gap, we construct \textbf{\textit{Any2Omni}}, the first comprehensive ODI generation-editing dataset comprises 60,000+ training data covering diverse input conditions and up to 9 ODI generation and editing tasks. Built upon Any2Omni, we propose an \textbf{\underline{Omni}} model for \textbf{\underline{Omni}}-directional image generation and editing (\textbf{\textit{Omni$^2$}}), with the capability of handling various ODI generation and editing tasks under diverse input conditions using one model. Extensive experiments demonstrate the superiority and effectiveness of the proposed Omni$^2$ model for both the ODI generation and editing tasks.
comment: 10 pages
☆ From Gaze to Insight: Bridging Human Visual Attention and Vision Language Model Explanation for Weakly-Supervised Medical Image Segmentation
Medical image segmentation remains challenging due to the high cost of pixel-level annotations for training. In the context of weak supervision, clinician gaze data captures regions of diagnostic interest; however, its sparsity limits its use for segmentation. In contrast, vision-language models (VLMs) provide semantic context through textual descriptions but lack the explanation precision required. Recognizing that neither source alone suffices, we propose a teacher-student framework that integrates both gaze and language supervision, leveraging their complementary strengths. Our key insight is that gaze data indicates where clinicians focus during diagnosis, while VLMs explain why those regions are significant. To implement this, the teacher model first learns from gaze points enhanced by VLM-generated descriptions of lesion morphology, establishing a foundation for guiding the student model. The teacher then directs the student through three strategies: (1) Multi-scale feature alignment to fuse visual cues with textual semantics; (2) Confidence-weighted consistency constraints to focus on reliable predictions; (3) Adaptive masking to limit error propagation in uncertain areas. Experiments on the Kvasir-SEG, NCI-ISBI, and ISIC datasets show that our method achieves Dice scores of 80.78%, 80.53%, and 84.22%, respectively-improving 3-5% over gaze baselines without increasing the annotation burden. By preserving correlations among predictions, gaze data, and lesion descriptions, our framework also maintains clinical interpretability. This work illustrates how integrating human visual attention with AI-generated semantic context can effectively overcome the limitations of individual weak supervision signals, thereby advancing the development of deployable, annotation-efficient medical AI systems. Code is available at: https://github.com/jingkunchen/FGI.git.
comment: 10 pages, 5 figures
☆ A Decade of Wheat Mapping for Lebanon
Wheat accounts for approximately 20% of the world's caloric intake, making it a vital component of global food security. Given this importance, mapping wheat fields plays a crucial role in enabling various stakeholders, including policy makers, researchers, and agricultural organizations, to make informed decisions regarding food security, supply chain management, and resource allocation. In this paper, we tackle the problem of accurately mapping wheat fields out of satellite images by introducing an improved pipeline for winter wheat segmentation, as well as presenting a case study on a decade-long analysis of wheat mapping in Lebanon. We integrate a Temporal Spatial Vision Transformer (TSViT) with Parameter-Efficient Fine Tuning (PEFT) and a novel post-processing pipeline based on the Fields of The World (FTW) framework. Our proposed pipeline addresses key challenges encountered in existing approaches, such as the clustering of small agricultural parcels in a single large field. By merging wheat segmentation with precise field boundary extraction, our method produces geometrically coherent and semantically rich maps that enable us to perform in-depth analysis such as tracking crop rotation pattern over years. Extensive evaluations demonstrate improved boundary delineation and field-level precision, establishing the potential of the proposed framework in operational agricultural monitoring and historical trend analysis. By allowing for accurate mapping of wheat fields, this work lays the foundation for a range of critical studies and future advances, including crop monitoring and yield estimation.
☆ Explicit and Implicit Representations in AI-based 3D Reconstruction for Radiology: A systematic literature review
The demand for high-quality medical imaging in clinical practice and assisted diagnosis has made 3D reconstruction in radiological imaging a key research focus. Artificial intelligence (AI) has emerged as a promising approach to enhancing reconstruction accuracy while reducing acquisition and processing time, thereby minimizing patient radiation exposure and discomfort and ultimately benefiting clinical diagnosis. This review explores state-of-the-art AI-based 3D reconstruction algorithms in radiological imaging, categorizing them into explicit and implicit approaches based on their underlying principles. Explicit methods include point-based, volume-based, and Gaussian representations, while implicit methods encompass implicit prior embedding and neural radiance fields. Additionally, we examine commonly used evaluation metrics and benchmark datasets. Finally, we discuss the current state of development, key challenges, and future research directions in this evolving field. Our project available on: https://github.com/Bean-Young/AI4Med.
comment: 43 pages, 5 figures, submit to Medical Image Analysis
☆ DeepWheel: Generating a 3D Synthetic Wheel Dataset for Design and Performance Evaluation
Data-driven design is emerging as a powerful strategy to accelerate engineering innovation. However, its application to vehicle wheel design remains limited due to the lack of large-scale, high-quality datasets that include 3D geometry and physical performance metrics. To address this gap, this study proposes a synthetic design-performance dataset generation framework using generative AI. The proposed framework first generates 2D rendered images using Stable Diffusion, and then reconstructs the 3D geometry through 2.5D depth estimation. Structural simulations are subsequently performed to extract engineering performance data. To further expand the design and performance space, topology optimization is applied, enabling the generation of a more diverse set of wheel designs. The final dataset, named DeepWheel, consists of over 6,000 photo-realistic images and 900 structurally analyzed 3D models. This multi-modal dataset serves as a valuable resource for surrogate model training, data-driven inverse design, and design space exploration. The proposed methodology is also applicable to other complex design domains. The dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International(CC BY-NC 4.0) and is available on the https://www.smartdesignlab.org/datasets
comment: 28 pages, 18 figures. Not yet submitted to a journal or conference
☆ Seedream 3.0 Technical Report
We present Seedream 3.0, a high-performance Chinese-English bilingual image generation foundation model. We develop several technical improvements to address existing challenges in Seedream 2.0, including alignment with complicated prompts, fine-grained typography generation, suboptimal visual aesthetics and fidelity, and limited image resolutions. Specifically, the advancements of Seedream 3.0 stem from improvements across the entire pipeline, from data construction to model deployment. At the data stratum, we double the dataset using a defect-aware training paradigm and a dual-axis collaborative data-sampling framework. Furthermore, we adopt several effective techniques such as mixed-resolution training, cross-modality RoPE, representation alignment loss, and resolution-aware timestep sampling in the pre-training phase. During the post-training stage, we utilize diversified aesthetic captions in SFT, and a VLM-based reward model with scaling, thereby achieving outputs that well align with human preferences. Furthermore, Seedream 3.0 pioneers a novel acceleration paradigm. By employing consistent noise expectation and importance-aware timestep sampling, we achieve a 4 to 8 times speedup while maintaining image quality. Seedream 3.0 demonstrates significant improvements over Seedream 2.0: it enhances overall capabilities, in particular for text-rendering in complicated Chinese characters which is important to professional typography generation. In addition, it provides native high-resolution output (up to 2K), allowing it to generate images with high visual quality.
comment: Seedream 3.0 Technical Report
☆ PVUW 2025 Challenge Report: Advances in Pixel-level Understanding of Complex Videos in the Wild
This report provides a comprehensive overview of the 4th Pixel-level Video Understanding in the Wild (PVUW) Challenge, held in conjunction with CVPR 2025. It summarizes the challenge outcomes, participating methodologies, and future research directions. The challenge features two tracks: MOSE, which focuses on complex scene video object segmentation, and MeViS, which targets motion-guided, language-based video segmentation. Both tracks introduce new, more challenging datasets designed to better reflect real-world scenarios. Through detailed evaluation and analysis, the challenge offers valuable insights into the current state-of-the-art and emerging trends in complex video segmentation. More information can be found on the workshop website: https://pvuw.github.io/.
comment: Workshop Page: https://pvuw.github.io/. arXiv admin note: text overlap with arXiv:2504.00476, arXiv:2504.05178
☆ Big Brother is Watching: Proactive Deepfake Detection via Learnable Hidden Face
As deepfake technologies continue to advance, passive detection methods struggle to generalize with various forgery manipulations and datasets. Proactive defense techniques have been actively studied with the primary aim of preventing deepfake operation effectively working. In this paper, we aim to bridge the gap between passive detection and proactive defense, and seek to solve the detection problem utilizing a proactive methodology. Inspired by several watermarking-based forensic methods, we explore a novel detection framework based on the concept of ``hiding a learnable face within a face''. Specifically, relying on a semi-fragile invertible steganography network, a secret template image is embedded into a host image imperceptibly, acting as an indicator monitoring for any malicious image forgery when being restored by the inverse steganography process. Instead of being manually specified, the secret template is optimized during training to resemble a neutral facial appearance, just like a ``big brother'' hidden in the image to be protected. By incorporating a self-blending mechanism and robustness learning strategy with a simulative transmission channel, a robust detector is built to accurately distinguish if the steganographic image is maliciously tampered or benignly processed. Finally, extensive experiments conducted on multiple datasets demonstrate the superiority of the proposed approach over competing passive and proactive detection methods.
☆ Uncertainty Estimation for Trust Attribution to Speed-of-Sound Reconstruction with Variational Networks
Speed-of-sound (SoS) is a biomechanical characteristic of tissue, and its imaging can provide a promising biomarker for diagnosis. Reconstructing SoS images from ultrasound acquisitions can be cast as a limited-angle computed-tomography problem, with Variational Networks being a promising model-based deep learning solution. Some acquired data frames may, however, get corrupted by noise due to, e.g., motion, lack of contact, and acoustic shadows, which in turn negatively affects the resulting SoS reconstructions. We propose to use the uncertainty in SoS reconstructions to attribute trust to each individual acquired frame. Given multiple acquisitions, we then use an uncertainty based automatic selection among these retrospectively, to improve diagnostic decisions. We investigate uncertainty estimation based on Monte Carlo Dropout and Bayesian Variational Inference. We assess our automatic frame selection method for differential diagnosis of breast cancer, distinguishing between benign fibroadenoma and malignant carcinoma. We evaluate 21 lesions classified as BI-RADS~4, which represents suspicious cases for probable malignancy. The most trustworthy frame among four acquisitions of each lesion was identified using uncertainty based criteria. Selecting a frame informed by uncertainty achieved an area under curve of 76% and 80% for Monte Carlo Dropout and Bayesian Variational Inference, respectively, superior to any uncertainty-uninformed baselines with the best one achieving 64%. A novel use of uncertainty estimation is proposed for selecting one of multiple data acquisitions for further processing and decision making.
comment: Published at the International Journal of Computer Assisted Radiology and Surgery. Presented at the 16th International Conference on Information Processing in Computer-Assisted Interventions 2025
☆ Context-Aware Palmprint Recognition via a Relative Similarity Metric
We propose a new approach to matching mechanism for palmprint recognition by introducing a Relative Similarity Metric (RSM) that enhances the robustness and discriminability of existing matching frameworks. While conventional systems rely on direct pairwise similarity measures, such as cosine or Euclidean distances, these metrics fail to capture how a pairwise similarity compares within the context of the entire dataset. Our method addresses this by evaluating the relative consistency of similarity scores across up to all identities, allowing for better suppression of false positives and negatives. Applied atop the CCNet architecture, our method achieves a new state-of-the-art 0.000036% Equal Error Rate (EER) on the Tongji dataset, outperforming previous methods and demonstrating the efficacy of incorporating relational structure into the palmprint matching process.
☆ CFIS-YOLO: A Lightweight Multi-Scale Fusion Network for Edge-Deployable Wood Defect Detection
Wood defect detection is critical for ensuring quality control in the wood processing industry. However, current industrial applications face two major challenges: traditional methods are costly, subjective, and labor-intensive, while mainstream deep learning models often struggle to balance detection accuracy and computational efficiency for edge deployment. To address these issues, this study proposes CFIS-YOLO, a lightweight object detection model optimized for edge devices. The model introduces an enhanced C2f structure, a dynamic feature recombination module, and a novel loss function that incorporates auxiliary bounding boxes and angular constraints. These innovations improve multi-scale feature fusion and small object localization while significantly reducing computational overhead. Evaluated on a public wood defect dataset, CFIS-YOLO achieves a mean Average Precision (mAP@0.5) of 77.5\%, outperforming the baseline YOLOv10s by 4 percentage points. On SOPHON BM1684X edge devices, CFIS-YOLO delivers 135 FPS, reduces power consumption to 17.3\% of the original implementation, and incurs only a 0.5 percentage point drop in mAP. These results demonstrate that CFIS-YOLO is a practical and effective solution for real-world wood defect detection in resource-constrained environments.
comment: 10 pages, 11 figures
☆ Autoregressive Distillation of Diffusion Transformers CVPR 2025
Diffusion models with transformer architectures have demonstrated promising capabilities in generating high-fidelity images and scalability for high resolution. However, iterative sampling process required for synthesis is very resource-intensive. A line of work has focused on distilling solutions to probability flow ODEs into few-step student models. Nevertheless, existing methods have been limited by their reliance on the most recent denoised samples as input, rendering them susceptible to exposure bias. To address this limitation, we propose AutoRegressive Distillation (ARD), a novel approach that leverages the historical trajectory of the ODE to predict future steps. ARD offers two key benefits: 1) it mitigates exposure bias by utilizing a predicted historical trajectory that is less susceptible to accumulated errors, and 2) it leverages the previous history of the ODE trajectory as a more effective source of coarse-grained information. ARD modifies the teacher transformer architecture by adding token-wise time embedding to mark each input from the trajectory history and employs a block-wise causal attention mask for training. Furthermore, incorporating historical inputs only in lower transformer layers enhances performance and efficiency. We validate the effectiveness of ARD in a class-conditioned generation on ImageNet and T2I synthesis. Our model achieves a $5\times$ reduction in FID degradation compared to the baseline methods while requiring only 1.1\% extra FLOPs on ImageNet-256. Moreover, ARD reaches FID of 1.84 on ImageNet-256 in merely 4 steps and outperforms the publicly available 1024p text-to-image distilled models in prompt adherence score with a minimal drop in FID compared to the teacher. Project page: https://github.com/alsdudrla10/ARD.
comment: CVPR 2025 Oral
☆ UniAnimate-DiT: Human Image Animation with Large-Scale Video Diffusion Transformer
This report presents UniAnimate-DiT, an advanced project that leverages the cutting-edge and powerful capabilities of the open-source Wan2.1 model for consistent human image animation. Specifically, to preserve the robust generative capabilities of the original Wan2.1 model, we implement Low-Rank Adaptation (LoRA) technique to fine-tune a minimal set of parameters, significantly reducing training memory overhead. A lightweight pose encoder consisting of multiple stacked 3D convolutional layers is designed to encode motion information of driving poses. Furthermore, we adopt a simple concatenation operation to integrate the reference appearance into the model and incorporate the pose information of the reference image for enhanced pose alignment. Experimental results show that our approach achieves visually appearing and temporally consistent high-fidelity animations. Trained on 480p (832x480) videos, UniAnimate-DiT demonstrates strong generalization capabilities to seamlessly upscale to 720P (1280x720) during inference. The training and inference code is publicly available at https://github.com/ali-vilab/UniAnimate-DiT.
comment: The training and inference code (based on Wan2.1) is available at https://github.com/ali-vilab/UniAnimate-DiT
☆ Efficient Medical Image Restoration via Reliability Guided Learning in Frequency Domain
Medical image restoration tasks aim to recover high-quality images from degraded observations, exhibiting emergent desires in many clinical scenarios, such as low-dose CT image denoising, MRI super-resolution, and MRI artifact removal. Despite the success achieved by existing deep learning-based restoration methods with sophisticated modules, they struggle with rendering computationally-efficient reconstruction results. Moreover, they usually ignore the reliability of the restoration results, which is much more urgent in medical systems. To alleviate these issues, we present LRformer, a Lightweight Transformer-based method via Reliability-guided learning in the frequency domain. Specifically, inspired by the uncertainty quantification in Bayesian neural networks (BNNs), we develop a Reliable Lesion-Semantic Prior Producer (RLPP). RLPP leverages Monte Carlo (MC) estimators with stochastic sampling operations to generate sufficiently-reliable priors by performing multiple inferences on the foundational medical image segmentation model, MedSAM. Additionally, instead of directly incorporating the priors in the spatial domain, we decompose the cross-attention (CA) mechanism into real symmetric and imaginary anti-symmetric parts via fast Fourier transform (FFT), resulting in the design of the Guided Frequency Cross-Attention (GFCA) solver. By leveraging the conjugated symmetric property of FFT, GFCA reduces the computational complexity of naive CA by nearly half. Extensive experimental results in various tasks demonstrate the superiority of the proposed LRformer in both effectiveness and efficiency.
☆ Distillation-Supervised Convolutional Low-Rank Adaptation for Efficient Image Super-Resolution
Convolutional neural networks (CNNs) have been widely used in efficient image super-resolution. However, for CNN-based methods, performance gains often require deeper networks and larger feature maps, which increase complexity and inference costs. Inspired by LoRA's success in fine-tuning large language models, we explore its application to lightweight models and propose Distillation-Supervised Convolutional Low-Rank Adaptation (DSCLoRA), which improves model performance without increasing architectural complexity or inference costs. Specifically, we integrate ConvLoRA into the efficient SR network SPAN by replacing the SPAB module with the proposed SConvLB module and incorporating ConvLoRA layers into both the pixel shuffle block and its preceding convolutional layer. DSCLoRA leverages low-rank decomposition for parameter updates and employs a spatial feature affinity-based knowledge distillation strategy to transfer second-order statistical information from teacher models (pre-trained SPAN) to student models (ours). This method preserves the core knowledge of lightweight models and facilitates optimal solution discovery under certain conditions. Experiments on benchmark datasets show that DSCLoRA improves PSNR and SSIM over SPAN while maintaining its efficiency and competitive image quality. Notably, DSCLoRA ranked first in the Overall Performance Track of the NTIRE 2025 Efficient Super-Resolution Challenge. Our code and models are made publicly available at https://github.com/Yaozzz666/DSCF-SR.
☆ Single-Input Multi-Output Model Merging: Leveraging Foundation Models for Dense Multi-Task Learning
Model merging is a flexible and computationally tractable approach to merge single-task checkpoints into a multi-task model. Prior work has solely focused on constrained multi-task settings where there is a one-to-one mapping between a sample and a task, overlooking the paradigm where multiple tasks may operate on the same sample, e.g., scene understanding. In this paper, we focus on the multi-task setting with single-input-multiple-outputs (SIMO) and show that it qualitatively differs from the single-input-single-output model merging settings studied in the literature due to the existence of task-specific decoders and diverse loss objectives. We identify that existing model merging methods lead to significant performance degradation, primarily due to representation misalignment between the merged encoder and task-specific decoders. We propose two simple and efficient fixes for the SIMO setting to re-align the feature representation after merging. Compared to joint fine-tuning, our approach is computationally effective and flexible, and sheds light into identifying task relationships in an offline manner. Experiments on NYUv2, Cityscapes, and a subset of the Taskonomy dataset demonstrate: (1) task arithmetic suffices to enable multi-task capabilities; however, the representations generated by the merged encoder has to be re-aligned with the task-specific heads; (2) the proposed architecture rivals traditional multi-task learning in performance but requires fewer samples and training steps by leveraging the existence of task-specific models.
comment: 22 pages, 6 figures
☆ Enhanced Small Target Detection via Multi-Modal Fusion and Attention Mechanisms: A YOLOv5 Approach ATC 2024
With the rapid development of information technology, modern warfare increasingly relies on intelligence, making small target detection critical in military applications. The growing demand for efficient, real-time detection has created challenges in identifying small targets in complex environments due to interference. To address this, we propose a small target detection method based on multi-modal image fusion and attention mechanisms. This method leverages YOLOv5, integrating infrared and visible light data along with a convolutional attention module to enhance detection performance. The process begins with multi-modal dataset registration using feature point matching, ensuring accurate network training. By combining infrared and visible light features with attention mechanisms, the model improves detection accuracy and robustness. Experimental results on anti-UAV and Visdrone datasets demonstrate the effectiveness and practicality of our approach, achieving superior detection results for small and dim targets.
comment: Accepted by ATC 2024
☆ UI-E2I-Synth: Advancing GUI Grounding with Large-Scale Instruction Synthesis
Recent advancements in Large Vision-Language Models are accelerating the development of Graphical User Interface (GUI) agents that utilize human-like vision perception capabilities to enhance productivity on digital devices. Compared to approaches predicated on GUI metadata, which are platform-dependent and vulnerable to implementation variations, vision-based approaches offer broader applicability. In this vision-based paradigm, the GUI instruction grounding, which maps user instruction to the location of corresponding element on the given screenshot, remains a critical challenge, particularly due to limited public training dataset and resource-intensive manual instruction data annotation.In this paper, we delve into unexplored challenges in this task including element-to-screen ratio, unbalanced element type, and implicit instruction. To address these challenges, we introduce a large-scale data synthesis pipeline UI-E2I-Synth for generating varying complex instruction datasets using GPT-4o instead of human annotators. Furthermore, we propose a new GUI instruction grounding benchmark UI-I2E-Bench, which is designed to address the limitations of existing benchmarks by incorporating diverse annotation aspects. Our model, trained on the synthesized data, achieves superior performance in GUI instruction grounding, demonstrating the advancements of proposed data synthesis pipeline. The proposed benchmark, accompanied by extensive analyses, provides practical insights for future research in GUI grounding. We will release corresponding artifacts at https://colmon46.github.io/i2e-bench-leaderboard/
☆ Cryo-em images are intrinsically low dimensional
Simulation-based inference provides a powerful framework for cryo-electron microscopy, employing neural networks in methods like CryoSBI to infer biomolecular conformations via learned latent representations. This latent space represents a rich opportunity, encoding valuable information about the physical system and the inference process. Harnessing this potential hinges on understanding the underlying geometric structure of these representations. We investigate this structure by applying manifold learning techniques to CryoSBI representations of hemagglutinin (simulated and experimental). We reveal that these high-dimensional data inherently populate low-dimensional, smooth manifolds, with simulated data effectively covering the experimental counterpart. By characterizing the manifold's geometry using Diffusion Maps and identifying its principal axes of variation via coordinate interpretation methods, we establish a direct link between the latent structure and key physical parameters. Discovering this intrinsic low-dimensionality and interpretable geometric organization not only validates the CryoSBI approach but enables us to learn more from the data structure and provides opportunities for improving future inference strategies by exploiting this revealed manifold geometry.
☆ Next-Future: Sample-Efficient Policy Learning for Robotic-Arm Tasks
Hindsight Experience Replay (HER) is widely regarded as the state-of-the-art algorithm for achieving sample-efficient multi-goal reinforcement learning (RL) in robotic manipulation tasks with binary rewards. HER facilitates learning from failed attempts by replaying trajectories with redefined goals. However, it relies on a heuristic-based replay method that lacks a principled framework. To address this limitation, we introduce a novel replay strategy, "Next-Future", which focuses on rewarding single-step transitions. This approach significantly enhances sample efficiency and accuracy in learning multi-goal Markov decision processes (MDPs), particularly under stringent accuracy requirements -- a critical aspect for performing complex and precise robotic-arm tasks. We demonstrate the efficacy of our method by highlighting how single-step learning enables improved value approximation within the multi-goal RL framework. The performance of the proposed replay strategy is evaluated across eight challenging robotic manipulation tasks, using ten random seeds for training. Our results indicate substantial improvements in sample efficiency for seven out of eight tasks and higher success rates in six tasks. Furthermore, real-world experiments validate the practical feasibility of the learned policies, demonstrating the potential of "Next-Future" in solving complex robotic-arm tasks.
comment: 10 pages, 9 figures, 6 tables
☆ Leveraging multimodal explanatory annotations for video interpretation with Modality Specific Dataset
We examine the impact of concept-informed supervision on multimodal video interpretation models using MOByGaze, a dataset containing human-annotated explanatory concepts. We introduce Concept Modality Specific Datasets (CMSDs), which consist of data subsets categorized by the modality (visual, textual, or audio) of annotated concepts. Models trained on CMSDs outperform those using traditional legacy training in both early and late fusion approaches. Notably, this approach enables late fusion models to achieve performance close to that of early fusion models. These findings underscore the importance of modality-specific annotations in developing robust, self-explainable video models and contribute to advancing interpretable multimodal learning in complex video analysis.
comment: 6 pages, 8 Figures
☆ CAP-Net: A Unified Network for 6D Pose and Size Estimation of Categorical Articulated Parts from a Single RGB-D Image CVPR 2025
This paper tackles category-level pose estimation of articulated objects in robotic manipulation tasks and introduces a new benchmark dataset. While recent methods estimate part poses and sizes at the category level, they often rely on geometric cues and complex multi-stage pipelines that first segment parts from the point cloud, followed by Normalized Part Coordinate Space (NPCS) estimation for 6D poses. These approaches overlook dense semantic cues from RGB images, leading to suboptimal accuracy, particularly for objects with small parts. To address these limitations, we propose a single-stage Network, CAP-Net, for estimating the 6D poses and sizes of Categorical Articulated Parts. This method combines RGB-D features to generate instance segmentation and NPCS representations for each part in an end-to-end manner. CAP-Net uses a unified network to simultaneously predict point-wise class labels, centroid offsets, and NPCS maps. A clustering algorithm then groups points of the same predicted class based on their estimated centroid distances to isolate each part. Finally, the NPCS region of each part is aligned with the point cloud to recover its final pose and size. To bridge the sim-to-real domain gap, we introduce the RGBD-Art dataset, the largest RGB-D articulated dataset to date, featuring photorealistic RGB images and depth noise simulated from real sensors. Experimental evaluations on the RGBD-Art dataset demonstrate that our method significantly outperforms the state-of-the-art approach. Real-world deployments of our model in robotic tasks underscore its robustness and exceptional sim-to-real transfer capabilities, confirming its substantial practical utility. Our dataset, code and pre-trained models are available on the project page.
comment: To appear in CVPR 2025 (Highlight)
☆ 3DAffordSplat: Efficient Affordance Reasoning with 3D Gaussians
3D affordance reasoning is essential in associating human instructions with the functional regions of 3D objects, facilitating precise, task-oriented manipulations in embodied AI. However, current methods, which predominantly depend on sparse 3D point clouds, exhibit limited generalizability and robustness due to their sensitivity to coordinate variations and the inherent sparsity of the data. By contrast, 3D Gaussian Splatting (3DGS) delivers high-fidelity, real-time rendering with minimal computational overhead by representing scenes as dense, continuous distributions. This positions 3DGS as a highly effective approach for capturing fine-grained affordance details and improving recognition accuracy. Nevertheless, its full potential remains largely untapped due to the absence of large-scale, 3DGS-specific affordance datasets. To overcome these limitations, we present 3DAffordSplat, the first large-scale, multi-modal dataset tailored for 3DGS-based affordance reasoning. This dataset includes 23,677 Gaussian instances, 8,354 point cloud instances, and 6,631 manually annotated affordance labels, encompassing 21 object categories and 18 affordance types. Building upon this dataset, we introduce AffordSplatNet, a novel model specifically designed for affordance reasoning using 3DGS representations. AffordSplatNet features an innovative cross-modal structure alignment module that exploits structural consistency priors to align 3D point cloud and 3DGS representations, resulting in enhanced affordance recognition accuracy. Extensive experiments demonstrate that the 3DAffordSplat dataset significantly advances affordance learning within the 3DGS domain, while AffordSplatNet consistently outperforms existing methods across both seen and unseen settings, highlighting its robust generalization capabilities.
comment: The first large-scale 3D Gaussians Affordance Reasoning Benchmark
☆ Focal Split: Untethered Snapshot Depth from Differential Defocus CVPR 2025
We introduce Focal Split, a handheld, snapshot depth camera with fully onboard power and computing based on depth-from-differential-defocus (DfDD). Focal Split is passive, avoiding power consumption of light sources. Its achromatic optical system simultaneously forms two differentially defocused images of the scene, which can be independently captured using two photosensors in a snapshot. The data processing is based on the DfDD theory, which efficiently computes a depth and a confidence value for each pixel with only 500 floating point operations (FLOPs) per pixel from the camera measurements. We demonstrate a Focal Split prototype, which comprises a handheld custom camera system connected to a Raspberry Pi 5 for real-time data processing. The system consumes 4.9 W and is powered on a 5 V, 10,000 mAh battery. The prototype can measure objects with distances from 0.4 m to 1.2 m, outputting 480$\times$360 sparse depth maps at 2.1 frames per second (FPS) using unoptimized Python scripts. Focal Split is DIY friendly. A comprehensive guide to building your own Focal Split depth camera, code, and additional data can be found at https://focal-split.qiguo.org.
comment: CVPR 2025, 8 pages, 7 figures
☆ Video Summarization with Large Language Models CVPR 2025
The exponential increase in video content poses significant challenges in terms of efficient navigation, search, and retrieval, thus requiring advanced video summarization techniques. Existing video summarization methods, which heavily rely on visual features and temporal dynamics, often fail to capture the semantics of video content, resulting in incomplete or incoherent summaries. To tackle the challenge, we propose a new video summarization framework that leverages the capabilities of recent Large Language Models (LLMs), expecting that the knowledge learned from massive data enables LLMs to evaluate video frames in a manner that better aligns with diverse semantics and human judgments, effectively addressing the inherent subjectivity in defining keyframes. Our method, dubbed LLM-based Video Summarization (LLMVS), translates video frames into a sequence of captions using a Muti-modal Large Language Model (M-LLM) and then assesses the importance of each frame using an LLM, based on the captions in its local context. These local importance scores are refined through a global attention mechanism in the entire context of video captions, ensuring that our summaries effectively reflect both the details and the overarching narrative. Our experimental results demonstrate the superiority of the proposed method over existing ones in standard benchmarks, highlighting the potential of LLMs in the processing of multimedia content.
comment: Accepted to CVPR 2025
☆ R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning CVPR 2025
Vision-language models (VLMs), such as CLIP, have gained significant popularity as foundation models, with numerous fine-tuning methods developed to enhance performance on downstream tasks. However, due to their inherent vulnerability and the common practice of selecting from a limited set of open-source models, VLMs suffer from a higher risk of adversarial attacks than traditional vision models. Existing defense techniques typically rely on adversarial fine-tuning during training, which requires labeled data and lacks of flexibility for downstream tasks. To address these limitations, we propose robust test-time prompt tuning (R-TPT), which mitigates the impact of adversarial attacks during the inference stage. We first reformulate the classic marginal entropy objective by eliminating the term that introduces conflicts under adversarial conditions, retaining only the pointwise entropy minimization. Furthermore, we introduce a plug-and-play reliability-based weighted ensembling strategy, which aggregates useful information from reliable augmented views to strengthen the defense. R-TPT enhances defense against adversarial attacks without requiring labeled training data while offering high flexibility for inference tasks. Extensive experiments on widely used benchmarks with various attacks demonstrate the effectiveness of R-TPT. The code is available in https://github.com/TomSheng21/R-TPT.
comment: CVPR 2025
☆ TerraMesh: A Planetary Mosaic of Multimodal Earth Observation Data
Large-scale foundation models in Earth Observation can learn versatile, label-efficient representations by leveraging massive amounts of unlabeled data. However, existing public datasets are often limited in scale, geographic coverage, or sensor variety. We introduce TerraMesh, a new globally diverse, multimodal dataset combining optical, synthetic aperture radar, elevation, and land-cover modalities in an Analysis-Ready Data format. TerraMesh includes over 9 million samples with eight spatiotemporal aligned modalities, enabling large-scale pre-training and fostering robust cross-modal correlation learning. We provide detailed data processing steps, comprehensive statistics, and empirical evidence demonstrating improved model performance when pre-trained on TerraMesh. The dataset will be made publicly available with a permissive license.
☆ TerraMind: Large-Scale Generative Multimodality for Earth Observation
We present TerraMind, the first any-to-any generative, multimodal foundation model for Earth observation (EO). Unlike other multimodal models, TerraMind is pretrained on dual-scale representations combining both token-level and pixel-level data across modalities. On a token level, TerraMind encodes high-level contextual information to learn cross-modal relationships, while on a pixel level, TerraMind leverages fine-grained representations to capture critical spatial nuances. We pretrained TerraMind on nine geospatial modalities of a global, large-scale dataset. In this paper, we demonstrate that (i) TerraMind's dual-scale early fusion approach unlocks a range of zero-shot and few-shot applications for Earth observation, (ii) TerraMind introduces "Thinking-in-Modalities" (TiM) -- the capability of generating additional artificial data during finetuning and inference to improve the model output -- and (iii) TerraMind achieves beyond state-of-the-art performance in community-standard benchmarks for EO like PANGAEA. The pretraining dataset, the model weights, and our code is open-sourced under a permissive license.
☆ YOLO-RS: Remote Sensing Enhanced Crop Detection Methods
With the rapid development of remote sensing technology, crop classification and health detection based on deep learning have gradually become a research hotspot. However, the existing target detection methods show poor performance when dealing with small targets in remote sensing images, especially in the case of complex background and image mixing, which is difficult to meet the practical application requirementsite. To address this problem, a novel target detection model YOLO-RS is proposed in this paper. The model is based on the latest Yolov11 which significantly enhances the detection of small targets by introducing the Context Anchor Attention (CAA) mechanism and an efficient multi-field multi-scale feature fusion network. YOLO-RS adopts a bidirectional feature fusion strategy in the feature fusion process, which effectively enhances the model's performance in the detection of small targets. Small target detection. Meanwhile, the ACmix module at the end of the model backbone network solves the category imbalance problem by adaptively adjusting the contrast and sample mixing, thus enhancing the detection accuracy in complex scenes. In the experiments on the PDT remote sensing crop health detection dataset and the CWC crop classification dataset, YOLO-RS improves both the recall and the mean average precision (mAP) by about 2-3\% or so compared with the existing state-of-the-art methods, while the F1-score is also significantly improved. Moreover, the computational complexity of the model only increases by about 5.2 GFLOPs, indicating its significant advantages in both performance and efficiency. The experimental results validate the effectiveness and application potential of YOLO-RS in the task of detecting small targets in remote sensing images.
☆ TSAL: Few-shot Text Segmentation Based on Attribute Learning
Recently supervised learning rapidly develops in scene text segmentation. However, the lack of high-quality datasets and the high cost of pixel annotation greatly limit the development of them. Considering the well-performed few-shot learning methods for downstream tasks, we investigate the application of the few-shot learning method to scene text segmentation. We propose TSAL, which leverages CLIP's prior knowledge to learn text attributes for segmentation. To fully utilize the semantic and texture information in the image, a visual-guided branch is proposed to separately extract text and background features. To reduce data dependency and improve text detection accuracy, the adaptive prompt-guided branch employs effective adaptive prompt templates to capture various text attributes. To enable adaptive prompts capture distinctive text features and complex background distribution, we propose Adaptive Feature Alignment module(AFA). By aligning learnable tokens of different attributes with visual features and prompt prototypes, AFA enables adaptive prompts to capture both general and distinctive attribute information. TSAL can capture the unique attributes of text and achieve precise segmentation using only few images. Experiments demonstrate that our method achieves SOTA performance on multiple text segmentation datasets under few-shot settings and show great potential in text-related domains.
☆ DMAGaze: Gaze Estimation Based on Feature Disentanglement and Multi-Scale Attention
Gaze estimation, which predicts gaze direction, commonly faces the challenge of interference from complex gaze-irrelevant information in face images. In this work, we propose DMAGaze, a novel gaze estimation framework that exploits information from facial images in three aspects: gaze-relevant global features (disentangled from facial image), local eye features (extracted from cropped eye patch), and head pose estimation features, to improve overall performance. Firstly, we design a new continuous mask-based Disentangler to accurately disentangle gaze-relevant and gaze-irrelevant information in facial images by achieving the dual-branch disentanglement goal through separately reconstructing the eye and non-eye regions. Furthermore, we introduce a new cascaded attention module named Multi-Scale Global Local Attention Module (MS-GLAM). Through a customized cascaded attention structure, it effectively focuses on global and local information at multiple scales, further enhancing the information from the Disentangler. Finally, the global gaze-relevant features disentangled by the upper face branch, combined with head pose and local eye features, are passed through the detection head for high-precision gaze estimation. Our proposed DMAGaze has been extensively validated on two mainstream public datasets, achieving state-of-the-art performance.
☆ SAR-to-RGB Translation with Latent Diffusion for Earth Observation
Earth observation satellites like Sentinel-1 (S1) and Sentinel-2 (S2) provide complementary remote sensing (RS) data, but S2 images are often unavailable due to cloud cover or data gaps. To address this, we propose a diffusion model (DM)-based approach for SAR-to-RGB translation, generating synthetic optical images from SAR inputs. We explore three different setups: two using Standard Diffusion, which reconstruct S2 images by adding and removing noise (one without and one with class conditioning), and one using Cold Diffusion, which blends S2 with S1 before removing the SAR signal. We evaluate the generated images in downstream tasks, including land cover classification and cloud removal. While generated images may not perfectly replicate real S2 data, they still provide valuable information. Our results show that class conditioning improves classification accuracy, while cloud removal performance remains competitive despite our approach not being optimized for it. Interestingly, despite exhibiting lower perceptual quality, the Cold Diffusion setup performs well in land cover classification, suggesting that traditional quantitative evaluation metrics may not fully reflect the practical utility of generated images. Our findings highlight the potential of DMs for SAR-to-RGB translation in RS applications where RGB images are missing.
comment: 10 pages, 3 figures
☆ GC-GAT: Multimodal Vehicular Trajectory Prediction using Graph Goal Conditioning and Cross-context Attention
Predicting future trajectories of surrounding vehicles heavily relies on what contextual information is given to a motion prediction model. The context itself can be static (lanes, regulatory elements, etc) or dynamic (traffic participants). This paper presents a lane graph-based motion prediction model that first predicts graph-based goal proposals and later fuses them with cross attention over multiple contextual elements. We follow the famous encoder-interactor-decoder architecture where the encoder encodes scene context using lightweight Gated Recurrent Units, the interactor applies cross-context attention over encoded scene features and graph goal proposals, and the decoder regresses multimodal trajectories via Laplacian Mixture Density Network from the aggregated encodings. Using cross-attention over graph-based goal proposals gives robust trajectory estimates since the model learns to attend to future goal-relevant scene elements for the intended agent. We evaluate our work on nuScenes motion prediction dataset, achieving state-of-the-art results.
☆ Taming Consistency Distillation for Accelerated Human Image Animation
Recent advancements in human image animation have been propelled by video diffusion models, yet their reliance on numerous iterative denoising steps results in high inference costs and slow speeds. An intuitive solution involves adopting consistency models, which serve as an effective acceleration paradigm through consistency distillation. However, simply employing this strategy in human image animation often leads to quality decline, including visual blurring, motion degradation, and facial distortion, particularly in dynamic regions. In this paper, we propose the DanceLCM approach complemented by several enhancements to improve visual quality and motion continuity at low-step regime: (1) segmented consistency distillation with an auxiliary light-weight head to incorporate supervision from real video latents, mitigating cumulative errors resulting from single full-trajectory generation; (2) a motion-focused loss to centre on motion regions, and explicit injection of facial fidelity features to improve face authenticity. Extensive qualitative and quantitative experiments demonstrate that DanceLCM achieves results comparable to state-of-the-art video diffusion models with a mere 2-4 inference steps, significantly reducing the inference burden without compromising video quality. The code and models will be made publicly available.
☆ Visual Re-Ranking with Non-Visual Side Information SC
The standard approach for visual place recognition is to use global image descriptors to retrieve the most similar database images for a given query image. The results can then be further improved with re-ranking methods that re-order the top scoring images. However, existing methods focus on re-ranking based on the same image descriptors that were used for the initial retrieval, which we argue provides limited additional signal. In this work we propose Generalized Contextual Similarity Aggregation (GCSA), which is a graph neural network-based re-ranking method that, in addition to the visual descriptors, can leverage other types of available side information. This can for example be other sensor data (such as signal strength of nearby WiFi or BlueTooth endpoints) or geometric properties such as camera poses for database images. In many applications this information is already present or can be acquired with low effort. Our architecture leverages the concept of affinity vectors to allow for a shared encoding of the heterogeneous multi-modal input. Two large-scale datasets, covering both outdoor and indoor localization scenarios, are utilized for training and evaluation. In experiments we show significant improvement not only on image retrieval metrics, but also for the downstream visual localization task.
comment: Accepted at Scandinavian Conference on Image Analysis (SCIA) 2025
☆ K-means Enhanced Density Gradient Analysis for Urban and Transport Metrics Using Multi-Modal Satellite Imagery
This paper presents a novel computational approach for evaluating urban metrics through density gradient analysis using multi-modal satellite imagery, with applications including public transport and other urban systems. By combining optical and Synthetic Aperture Radar (SAR) data, we develop a method to segment urban areas, identify urban centers, and quantify density gradients. Our approach calculates two key metrics: the density gradient coefficient ($\alpha$) and the minimum effective distance (LD) at which density reaches a target threshold. We further employ machine learning techniques, specifically K-means clustering, to objectively identify uniform and high-variability regions within density gradient plots. We demonstrate that these metrics provide an effective screening tool for public transport analyses by revealing the underlying urban structure. Through comparative analysis of two representative cities with contrasting urban morphologies (monocentric vs polycentric), we establish relationships between density gradient characteristics and public transport network topologies. Cities with clear density peaks in their gradient plots indicate distinct urban centers requiring different transport strategies than those with more uniform density distributions. This methodology offers urban planners a cost-effective, globally applicable approach to preliminary public transport assessment using freely available satellite data. The complete implementation, with additional examples and documentation, is available in an open-source repository under the MIT license at https://github.com/nexri/Satellite-Imagery-Urban-Analysis.
comment: 16 pages, 6 figures
☆ Revealing Covert Attention by Analyzing Human and Reinforcement Learning Agent Gameplay
This study introduces a novel method for revealing human covert attention patterns using gameplay data alone, utilizing offline attention techniques from reinforcement learning (RL). We propose the contextualized, task-relevant (CTR) attention network, which generates attention maps from both human and RL agent gameplay in Atari environments. These maps are sparse yet retain the necessary information for the current player's decision making. We compare the CTR-derived attention maps with a temporally integrated overt attention (TIOA) model based on eye-tracking data, serving as a point of comparison and discussion. Visual inspection reveals distinct attention patterns: human CTR maps focus on the player and rather nearby opponents, occasionally shifting between stronger focus and broader views - sometimes even attending to empty space ahead. In contrast, agent maps maintain a consistent broad focus on most objects, including distant ones and the player. Quantitative analysis further demonstrates that human CTR maps align more closely with TIOA than agent maps do. Our findings indicate that the CTR attention network can effectively reveal human covert attention patterns from gameplay alone, without the need for additional data like brain activity recordings. This work contributes to understanding human-agent attention differences and enables the development of RL agents augmented with human covert attention.
☆ Flyweight FLIM Networks for Salient Object Detection in Biomedical Images
Salient Object Detection (SOD) with deep learning often requires substantial computational resources and large annotated datasets, making it impractical for resource-constrained applications. Lightweight models address computational demands but typically strive in complex and scarce labeled-data scenarios. Feature Learning from Image Markers (FLIM) learns an encoder's convolutional kernels among image patches extracted from discriminative regions marked on a few representative images, dismissing large annotated datasets, pretraining, and backpropagation. Such a methodology exploits information redundancy commonly found in biomedical image applications. This study presents methods to learn dilated-separable convolutional kernels and multi-dilation layers without backpropagation for FLIM networks. It also proposes a novel network simplification method to reduce kernel redundancy and encoder size. By combining a FLIM encoder with an adaptive decoder, a concept recently introduced to estimate a pointwise convolution per image, this study presents very efficient (named flyweight) SOD models for biomedical images. Experimental results in challenging datasets demonstrate superior efficiency and effectiveness to lightweight models. By requiring significantly fewer parameters and floating-point operations, the results show competitive effectiveness to heavyweight models. These advances highlight the potential of FLIM networks for data-limited and resource-constrained applications with information redundancy.
☆ S$^2$Teacher: Step-by-step Teacher for Sparsely Annotated Oriented Object Detection
Although fully-supervised oriented object detection has made significant progress in multimodal remote sensing image understanding, it comes at the cost of labor-intensive annotation. Recent studies have explored weakly and semi-supervised learning to alleviate this burden. However, these methods overlook the difficulties posed by dense annotations in complex remote sensing scenes. In this paper, we introduce a novel setting called sparsely annotated oriented object detection (SAOOD), which only labels partial instances, and propose a solution to address its challenges. Specifically, we focus on two key issues in the setting: (1) sparse labeling leading to overfitting on limited foreground representations, and (2) unlabeled objects (false negatives) confusing feature learning. To this end, we propose the S$^2$Teacher, a novel method that progressively mines pseudo-labels for unlabeled objects, from easy to hard, to enhance foreground representations. Additionally, it reweights the loss of unlabeled objects to mitigate their impact during training. Extensive experiments demonstrate that S$^2$Teacher not only significantly improves detector performance across different sparse annotation levels but also achieves near-fully-supervised performance on the DOTA dataset with only 10% annotation instances, effectively balancing detection accuracy with annotation efficiency. The code will be public.
☆ Token-Level Constraint Boundary Search for Jailbreaking Text-to-Image Models
Recent advancements in Text-to-Image (T2I) generation have significantly enhanced the realism and creativity of generated images. However, such powerful generative capabilities pose risks related to the production of inappropriate or harmful content. Existing defense mechanisms, including prompt checkers and post-hoc image checkers, are vulnerable to sophisticated adversarial attacks. In this work, we propose TCBS-Attack, a novel query-based black-box jailbreak attack that searches for tokens located near the decision boundaries defined by text and image checkers. By iteratively optimizing tokens near these boundaries, TCBS-Attack generates semantically coherent adversarial prompts capable of bypassing multiple defensive layers in T2I models. Extensive experiments demonstrate that our method consistently outperforms state-of-the-art jailbreak attacks across various T2I models, including securely trained open-source models and commercial online services like DALL-E 3. TCBS-Attack achieves an ASR-4 of 45\% and an ASR-1 of 21\% on jailbreaking full-chain T2I models, significantly surpassing baseline methods.
☆ Using LLMs as prompt modifier to avoid biases in AI image generators
This study examines how Large Language Models (LLMs) can reduce biases in text-to-image generation systems by modifying user prompts. We define bias as a model's unfair deviation from population statistics given neutral prompts. Our experiments with Stable Diffusion XL, 3.5 and Flux demonstrate that LLM-modified prompts significantly increase image diversity and reduce bias without the need to change the image generators themselves. While occasionally producing results that diverge from original user intent for elaborate prompts, this approach generally provides more varied interpretations of underspecified requests rather than superficial variations. The method works particularly well for less advanced image generators, though limitations persist for certain contexts like disability representation. All prompts and generated images are available at https://iisys-hof.github.io/llm-prompt-img-gen/
☆ Consensus Entropy: Harnessing Multi-VLM Agreement for Self-Verifying and Self-Improving OCR
The Optical Character Recognition (OCR) task is important for evaluating Vision-Language Models (VLMs) and providing high-quality data sources for LLM training data. While state-of-the-art VLMs show improved average OCR accuracy, they still struggle with sample-level quality degradation and lack reliable automatic detection of low-quality outputs. We introduce Consensus Entropy (CE), a training-free post-inference method that quantifies OCR uncertainty by aggregating outputs from multiple VLMs. Our approach exploits a key insight: correct VLM OCR predictions converge in output space while errors diverge. We develop a lightweight multi-model framework that effectively identifies problematic samples, selects the best outputs and combines model strengths. Experiments across multiple OCR benchmarks and VLMs demonstrate that CE outperforms VLM-as-judge approaches and single-model baselines at the same cost and achieves state-of-the-art results across multiple metrics. For instance, our solution demonstrates: achieving 15.2\% higher F1 scores than VLM-as-judge methods in quality verification, delivering 6.0\% accuracy gains on mathematical calculation tasks, and requiring rephrasing only 7.3\% of inputs while maintaining overall performance. Notably, the entire process requires neither training nor supervision while maintaining plug-and-play functionality throughout.
☆ Vivid4D: Improving 4D Reconstruction from Monocular Video by Video Inpainting
Reconstructing 4D dynamic scenes from casually captured monocular videos is valuable but highly challenging, as each timestamp is observed from a single viewpoint. We introduce Vivid4D, a novel approach that enhances 4D monocular video synthesis by augmenting observation views - synthesizing multi-view videos from a monocular input. Unlike existing methods that either solely leverage geometric priors for supervision or use generative priors while overlooking geometry, we integrate both. This reformulates view augmentation as a video inpainting task, where observed views are warped into new viewpoints based on monocular depth priors. To achieve this, we train a video inpainting model on unposed web videos with synthetically generated masks that mimic warping occlusions, ensuring spatially and temporally consistent completion of missing regions. To further mitigate inaccuracies in monocular depth priors, we introduce an iterative view augmentation strategy and a robust reconstruction loss. Experiments demonstrate that our method effectively improves monocular 4D scene reconstruction and completion.
☆ InfoClus: Informative Clustering of High-dimensional Data Embeddings
Developing an understanding of high-dimensional data can be facilitated by visualizing that data using dimensionality reduction. However, the low-dimensional embeddings are often difficult to interpret. To facilitate the exploration and interpretation of low-dimensional embeddings, we introduce a new concept named partitioning with explanations. The idea is to partition the data shown through the embedding into groups, each of which is given a sparse explanation using the original high-dimensional attributes. We introduce an objective function that quantifies how much we can learn through observing the explanations of the data partitioning, using information theory, and also how complex the explanations are. Through parameterization of the complexity, we can tune the solutions towards the desired granularity. We propose InfoClus, which optimizes the partitioning and explanations jointly, through greedy search constrained over a hierarchical clustering. We conduct a qualitative and quantitative analysis of InfoClus on three data sets. We contrast the results on the Cytometry data with published manual analysis results, and compare with two other recent methods for explaining embeddings (RVX and VERA). These comparisons highlight that InfoClus has distinct advantages over existing procedures and methods. We find that InfoClus can automatically create good starting points for the analysis of dimensionality-reduction-based scatter plots.
comment: 17 pages, 9 figures
☆ Change State Space Models for Remote Sensing Change Detection
Despite their frequent use for change detection, both ConvNets and Vision transformers (ViT) exhibit well-known limitations, namely the former struggle to model long-range dependencies while the latter are computationally inefficient, rendering them challenging to train on large-scale datasets. Vision Mamba, an architecture based on State Space Models has emerged as an alternative addressing the aforementioned deficiencies and has been already applied to remote sensing change detection, though mostly as a feature extracting backbone. In this article the Change State Space Model is introduced, that has been specifically designed for change detection by focusing on the relevant changes between bi-temporal images, effectively filtering out irrelevant information. By concentrating solely on the changed features, the number of network parameters is reduced, enhancing significantly computational efficiency while maintaining high detection performance and robustness against input degradation. The proposed model has been evaluated via three benchmark datasets, where it outperformed ConvNets, ViTs, and Mamba-based counterparts at a fraction of their computational complexity. The implementation will be made available at https://github.com/Elman295/CSSM upon acceptance.
☆ Improving fingerprint presentation attack detection by an approach integrated into the personal verification stage
Presentation Attack Detection (PAD) systems are usually designed independently of the fingerprint verification system. While this can be acceptable for use cases where specific user templates are not predetermined, it represents a missed opportunity to enhance security in scenarios where integrating PAD with the fingerprint verification system could significantly leverage users' templates, which are the real target of a potential presentation attack. This does not mean that a PAD should be specifically designed for such users; that would imply the availability of many enrolled users' PAI and, consequently, complexity, time, and cost increase. On the contrary, we propose to equip a basic PAD, designed according to the state of the art, with an innovative add-on module called the Closeness Binary Code (CC) module. The term "closeness" refers to a peculiar property of the bona fide-related features: in an Euclidean feature space, genuine fingerprints tend to cluster in a specific pattern. First, samples from the same finger are close to each other, then samples from other fingers of the same user and finally, samples from fingers of other users. This property is statistically verified in our previous publication, and further confirmed in this paper. It is independent of the user population and the feature set class, which can be handcrafted or deep network-based (embeddings). Therefore, the add-on can be designed without the need for the targeted user samples; moreover, it exploits her/his samples' "closeness" property during the verification stage. Extensive experiments on benchmark datasets and state-of-the-art PAD methods confirm the benefits of the proposed add-on, which can be easily coupled with the main PAD module integrated into the fingerprint verification system.
comment: This work has been submitted to the IEEE for possible publication
☆ UKDM: Underwater keypoint detection and matching using underwater image enhancement techniques
The purpose of this paper is to explore the use of underwater image enhancement techniques to improve keypoint detection and matching. By applying advanced deep learning models, including generative adversarial networks and convolutional neural networks, we aim to find the best method which improves the accuracy of keypoint detection and the robustness of matching algorithms. We evaluate the performance of these techniques on various underwater datasets, demonstrating significant improvements over traditional methods.
☆ Crane: Context-Guided Prompt Learning and Attention Refinement for Zero-Shot Anomaly Detections
Anomaly Detection (AD) involves identifying deviations from normal data distributions and is critical in fields such as medical diagnostics and industrial defect detection. Traditional AD methods typically require the availability of normal training samples; however, this assumption is not always feasible, as collecting such data can be impractical. Additionally, these methods often struggle to generalize across different domains. Recent advancements, such as AnomalyCLIP and AdaCLIP, utilize the zero-shot generalization capabilities of CLIP but still face a performance gap between image-level and pixel-level anomaly detection. To address this gap, we propose a novel approach that conditions the prompts of the text encoder based on image context extracted from the vision encoder. Also, to capture fine-grained variations more effectively, we have modified the CLIP vision encoder and altered the extraction of dense features. These changes ensure that the features retain richer spatial and structural information for both normal and anomalous prompts. Our method achieves state-of-the-art performance, improving performance by 2% to 29% across different metrics on 14 datasets. This demonstrates its effectiveness in both image-level and pixel-level anomaly detection.
☆ Leveraging LLMs and attention-mechanism for automatic annotation of historical maps
Historical maps are essential resources that provide insights into the geographical landscapes of the past. They serve as valuable tools for researchers across disciplines such as history, geography, and urban studies, facilitating the reconstruction of historical environments and the analysis of spatial transformations over time. However, when constrained to analogue or scanned formats, their interpretation is limited to humans and therefore not scalable. Recent advancements in machine learning, particularly in computer vision and large language models (LLMs), have opened new avenues for automating the recognition and classification of features and objects in historical maps. In this paper, we propose a novel distillation method that leverages LLMs and attention mechanisms for the automatic annotation of historical maps. LLMs are employed to generate coarse classification labels for low-resolution historical image patches, while attention mechanisms are utilized to refine these labels to higher resolutions. Experimental results demonstrate that the refined labels achieve a high recall of more than 90%. Additionally, the intersection over union (IoU) scores--84.2% for Wood and 72.0% for Settlement--along with precision scores of 87.1% and 79.5%, respectively, indicate that most labels are well-aligned with ground-truth annotations. Notably, these results were achieved without the use of fine-grained manual labels during training, underscoring the potential of our approach for efficient and scalable historical map analysis.
☆ QAVA: Query-Agnostic Visual Attack to Large Vision-Language Models NAACL 2025
In typical multimodal tasks, such as Visual Question Answering (VQA), adversarial attacks targeting a specific image and question can lead large vision-language models (LVLMs) to provide incorrect answers. However, it is common for a single image to be associated with multiple questions, and LVLMs may still answer other questions correctly even for an adversarial image attacked by a specific question. To address this, we introduce the query-agnostic visual attack (QAVA), which aims to create robust adversarial examples that generate incorrect responses to unspecified and unknown questions. Compared to traditional adversarial attacks focused on specific images and questions, QAVA significantly enhances the effectiveness and efficiency of attacks on images when the question is unknown, achieving performance comparable to attacks on known target questions. Our research broadens the scope of visual adversarial attacks on LVLMs in practical settings, uncovering previously overlooked vulnerabilities, particularly in the context of visual adversarial threats. The code is available at https://github.com/btzyd/qava.
comment: Accepted by NAACL 2025 main
☆ Defending Against Frequency-Based Attacks with Diffusion Models CVPR
Adversarial training is a common strategy for enhancing model robustness against adversarial attacks. However, it is typically tailored to the specific attack types it is trained on, limiting its ability to generalize to unseen threat models. Adversarial purification offers an alternative by leveraging a generative model to remove perturbations before classification. Since the purifier is trained independently of both the classifier and the threat models, it is better equipped to handle previously unseen attack scenarios. Diffusion models have proven highly effective for noise purification, not only in countering pixel-wise adversarial perturbations but also in addressing non-adversarial data shifts. In this study, we broaden the focus beyond pixel-wise robustness to explore the extent to which purification can mitigate both spectral and spatial adversarial attacks. Our findings highlight its effectiveness in handling diverse distortion patterns across low- to high-frequency regions.
comment: Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 5th Workshop on Adversarial Machine Learning in Computer Vision: Foundation Models + X
☆ Acquisition of high-quality images for camera calibration in robotics applications via speech prompts
Accurate intrinsic and extrinsic camera calibration can be an important prerequisite for robotic applications that rely on vision as input. While there is ongoing research on enabling camera calibration using natural images, many systems in practice still rely on using designated calibration targets with e.g. checkerboard patterns or April tag grids. Once calibration images from different perspectives have been acquired and feature descriptors detected, those are typically used in an optimization process to minimize the geometric reprojection error. For this optimization to converge, input images need to be of sufficient quality and particularly sharpness; they should neither contain motion blur nor rolling-shutter artifacts that can arise when the calibration board was not static during image capture. In this work, we present a novel calibration image acquisition technique controlled via voice commands recorded with a clip-on microphone, that can be more robust and user-friendly than e.g. triggering capture with a remote control, or filtering out blurry frames from a video sequence in postprocessing. To achieve this, we use a state-of-the-art speech-to-text transcription model with accurate per-word timestamping to capture trigger words with precise temporal alignment. Our experiments show that the proposed method improves user experience by being fast and efficient, allowing us to successfully calibrate complex multi-camera setups.
comment: 8 pages, 6 figures
☆ Easy3D: A Simple Yet Effective Method for 3D Interactive Segmentation
The increasing availability of digital 3D environments, whether through image-based 3D reconstruction, generation, or scans obtained by robots, is driving innovation across various applications. These come with a significant demand for 3D interaction, such as 3D Interactive Segmentation, which is useful for tasks like object selection and manipulation. Additionally, there is a persistent need for solutions that are efficient, precise, and performing well across diverse settings, particularly in unseen environments and with unfamiliar objects. In this work, we introduce a 3D interactive segmentation method that consistently surpasses previous state-of-the-art techniques on both in-domain and out-of-domain datasets. Our simple approach integrates a voxel-based sparse encoder with a lightweight transformer-based decoder that implements implicit click fusion, achieving superior performance and maximizing efficiency. Our method demonstrates substantial improvements on benchmark datasets, including ScanNet, ScanNet++, S3DIS, and KITTI-360, and also on unseen geometric distributions such as the ones obtained by Gaussian Splatting. The project web-page is available at https://simonelli-andrea.github.io/easy3d.
☆ Meta-learning For Few-Shot Time Series Crop Type Classification: A Benchmark On The EuroCropsML Dataset
Spatial imbalances in crop type data pose significant challenges for accurate classification in remote sensing applications. Algorithms aiming at transferring knowledge from data-rich to data-scarce tasks have thus surged in popularity. However, despite their effectiveness in previous evaluations, their performance in challenging real-world applications is unclear and needs to be evaluated. This study benchmarks transfer learning and several meta-learning algorithms, including (First-Order) Model-Agnostic Meta-Learning ((FO)-MAML), Almost No Inner Loop (ANIL), and Task-Informed Meta-Learning (TIML), on the real-world EuroCropsML time series dataset, which combines farmer-reported crop data with Sentinel-2 satellite observations from Estonia, Latvia, and Portugal. Our findings indicate that MAML-based meta-learning algorithms achieve slightly higher accuracy compared to simpler transfer learning methods when applied to crop type classification tasks in Estonia after pre-training on data from Latvia. However, this improvement comes at the cost of increased computational demands and training time. Moreover, we find that the transfer of knowledge between geographically disparate regions, such as Estonia and Portugal, poses significant challenges to all investigated algorithms. These insights underscore the trade-offs between accuracy and computational resource requirements in selecting machine learning methods for real-world crop type classification tasks and highlight the difficulties of transferring knowledge between different regions of the Earth. To facilitate future research in this domain, we present the first comprehensive benchmark for evaluating transfer and meta-learning methods for crop type classification under real-world conditions. The corresponding code is publicly available at https://github.com/dida-do/eurocrops-meta-learning.
comment: 19 pages, 7 figures, 12 tables
☆ DRIFT open dataset: A drone-derived intelligence for traffic analysis in urban environmen
Reliable traffic data are essential for understanding urban mobility and developing effective traffic management strategies. This study introduces the DRone-derived Intelligence For Traffic analysis (DRIFT) dataset, a large-scale urban traffic dataset collected systematically from synchronized drone videos at approximately 250 meters altitude, covering nine interconnected intersections in Daejeon, South Korea. DRIFT provides high-resolution vehicle trajectories that include directional information, processed through video synchronization and orthomap alignment, resulting in a comprehensive dataset of 81,699 vehicle trajectories. Through our DRIFT dataset, researchers can simultaneously analyze traffic at multiple scales - from individual vehicle maneuvers like lane-changes and safety metrics such as time-to-collision to aggregate network flow dynamics across interconnected urban intersections. The DRIFT dataset is structured to enable immediate use without additional preprocessing, complemented by open-source models for object detection and trajectory extraction, as well as associated analytical tools. DRIFT is expected to significantly contribute to academic research and practical applications, such as traffic flow analysis and simulation studies. The dataset and related resources are publicly accessible at https://github.com/AIxMobility/The-DRIFT.
comment: 30 pages, 15 figures
☆ AnimeDL-2M: Million-Scale AI-Generated Anime Image Detection and Localization in Diffusion Era
Recent advances in image generation, particularly diffusion models, have significantly lowered the barrier for creating sophisticated forgeries, making image manipulation detection and localization (IMDL) increasingly challenging. While prior work in IMDL has focused largely on natural images, the anime domain remains underexplored-despite its growing vulnerability to AI-generated forgeries. Misrepresentations of AI-generated images as hand-drawn artwork, copyright violations, and inappropriate content modifications pose serious threats to the anime community and industry. To address this gap, we propose AnimeDL-2M, the first large-scale benchmark for anime IMDL with comprehensive annotations. It comprises over two million images including real, partially manipulated, and fully AI-generated samples. Experiments indicate that models trained on existing IMDL datasets of natural images perform poorly when applied to anime images, highlighting a clear domain gap between anime and natural images. To better handle IMDL tasks in anime domain, we further propose AniXplore, a novel model tailored to the visual characteristics of anime imagery. Extensive evaluations demonstrate that AniXplore achieves superior performance compared to existing methods. Dataset and code can be found in https://flytweety.github.io/AnimeDL2M/.
☆ GATE3D: Generalized Attention-based Task-synergized Estimation in 3D*
The emerging trend in computer vision emphasizes developing universal models capable of simultaneously addressing multiple diverse tasks. Such universality typically requires joint training across multi-domain datasets to ensure effective generalization. However, monocular 3D object detection presents unique challenges in multi-domain training due to the scarcity of datasets annotated with accurate 3D ground-truth labels, especially beyond typical road-based autonomous driving contexts. To address this challenge, we introduce a novel weakly supervised framework leveraging pseudo-labels. Current pretrained models often struggle to accurately detect pedestrians in non-road environments due to inherent dataset biases. Unlike generalized image-based 2D object detection models, achieving similar generalization in monocular 3D detection remains largely unexplored. In this paper, we propose GATE3D, a novel framework designed specifically for generalized monocular 3D object detection via weak supervision. GATE3D effectively bridges domain gaps by employing consistency losses between 2D and 3D predictions. Remarkably, our model achieves competitive performance on the KITTI benchmark as well as on an indoor-office dataset collected by us to evaluate the generalization capabilities of our framework. Our results demonstrate that GATE3D significantly accelerates learning from limited annotated data through effective pre-training strategies, highlighting substantial potential for broader impacts in robotics, augmented reality, and virtual reality applications. Project page: https://ies0411.github.io/GATE3D/
comment: 9pages, 1 supple
☆ MediSee: Reasoning-based Pixel-level Perception in Medical Images
Despite remarkable advancements in pixel-level medical image perception, existing methods are either limited to specific tasks or heavily rely on accurate bounding boxes or text labels as input prompts. However, the medical knowledge required for input is a huge obstacle for general public, which greatly reduces the universality of these methods. Compared with these domain-specialized auxiliary information, general users tend to rely on oral queries that require logical reasoning. In this paper, we introduce a novel medical vision task: Medical Reasoning Segmentation and Detection (MedSD), which aims to comprehend implicit queries about medical images and generate the corresponding segmentation mask and bounding box for the target object. To accomplish this task, we first introduce a Multi-perspective, Logic-driven Medical Reasoning Segmentation and Detection (MLMR-SD) dataset, which encompasses a substantial collection of medical entity targets along with their corresponding reasoning. Furthermore, we propose MediSee, an effective baseline model designed for medical reasoning segmentation and detection. The experimental results indicate that the proposed method can effectively address MedSD with implicit colloquial queries and outperform traditional medical referring segmentation methods.
comment: 10 pages, 6 figures
☆ TMCIR: Token Merge Benefits Composed Image Retrieval
Composed Image Retrieval (CIR) retrieves target images using a multi-modal query that combines a reference image with text describing desired modifications. The primary challenge is effectively fusing this visual and textual information. Current cross-modal feature fusion approaches for CIR exhibit an inherent bias in intention interpretation. These methods tend to disproportionately emphasize either the reference image features (visual-dominant fusion) or the textual modification intent (text-dominant fusion through image-to-text conversion). Such an imbalanced representation often fails to accurately capture and reflect the actual search intent of the user in the retrieval results. To address this challenge, we propose TMCIR, a novel framework that advances composed image retrieval through two key innovations: 1) Intent-Aware Cross-Modal Alignment. We first fine-tune CLIP encoders contrastively using intent-reflecting pseudo-target images, synthesized from reference images and textual descriptions via a diffusion model. This step enhances the encoder ability of text to capture nuanced intents in textual descriptions. 2) Adaptive Token Fusion. We further fine-tune all encoders contrastively by comparing adaptive token-fusion features with the target image. This mechanism dynamically balances visual and textual representations within the contrastive learning pipeline, optimizing the composed feature for retrieval. Extensive experiments on Fashion-IQ and CIRR datasets demonstrate that TMCIR significantly outperforms state-of-the-art methods, particularly in capturing nuanced user intent.
comment: arXiv admin note: text overlap with arXiv:2310.05473 by other authors
☆ PraNet-V2: Dual-Supervised Reverse Attention for Medical Image Segmentation
Accurate medical image segmentation is essential for effective diagnosis and treatment. Previously, PraNet-V1 was proposed to enhance polyp segmentation by introducing a reverse attention (RA) module that utilizes background information. However, PraNet-V1 struggles with multi-class segmentation tasks. To address this limitation, we propose PraNet-V2, which, compared to PraNet-V1, effectively performs a broader range of tasks including multi-class segmentation. At the core of PraNet-V2 is the Dual-Supervised Reverse Attention (DSRA) module, which incorporates explicit background supervision, independent background modeling, and semantically enriched attention fusion. Our PraNet-V2 framework demonstrates strong performance on four polyp segmentation datasets. Additionally, by integrating DSRA to iteratively enhance foreground segmentation results in three state-of-the-art semantic segmentation models, we achieve up to a 1.36% improvement in mean Dice score. Code is available at: https://github.com/ai4colonoscopy/PraNet-V2/tree/main/binary_seg/jittor.
comment: Technical report (4 tables 3 figures 8 pages)
☆ DMPT: Decoupled Modality-aware Prompt Tuning for Multi-modal Object Re-identification WACV
Current multi-modal object re-identification approaches based on large-scale pre-trained backbones (i.e., ViT) have displayed remarkable progress and achieved excellent performance. However, these methods usually adopt the standard full fine-tuning paradigm, which requires the optimization of considerable backbone parameters, causing extensive computational and storage requirements. In this work, we propose an efficient prompt-tuning framework tailored for multi-modal object re-identification, dubbed DMPT, which freezes the main backbone and only optimizes several newly added decoupled modality-aware parameters. Specifically, we explicitly decouple the visual prompts into modality-specific prompts which leverage prior modality knowledge from a powerful text encoder and modality-independent semantic prompts which extract semantic information from multi-modal inputs, such as visible, near-infrared, and thermal-infrared. Built upon the extracted features, we further design a Prompt Inverse Bind (PromptIBind) strategy that employs bind prompts as a medium to connect the semantic prompt tokens of different modalities and facilitates the exchange of complementary multi-modal information, boosting final re-identification results. Experimental results on multiple common benchmarks demonstrate that our DMPT can achieve competitive results to existing state-of-the-art methods while requiring only 6.5% fine-tuning of the backbone parameters.
comment: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
☆ Seeing like a Cephalopod: Colour Vision with a Monochrome Event Camera CVPR 2025
Cephalopods exhibit unique colour discrimination capabilities despite having one type of photoreceptor, relying instead on chromatic aberration induced by their ocular optics and pupil shapes to perceive spectral information. We took inspiration from this biological mechanism to design a spectral imaging system that combines a ball lens with an event-based camera. Our approach relies on a motorised system that shifts the focal position, mirroring the adaptive lens motion in cephalopods. This approach has enabled us to achieve wavelength-dependent focusing across the visible light and near-infrared spectrum, making the event a spectral sensor. We characterise chromatic aberration effects, using both event-based and conventional frame-based sensors, validating the effectiveness of bio-inspired spectral discrimination both in simulation and in a real setup as well as assessing the spectral discrimination performance. Our proposed approach provides a robust spectral sensing capability without conventional colour filters or computational demosaicing. This approach opens new pathways toward new spectral sensing systems inspired by nature's evolutionary solutions. Code and analysis are available at: https://samiarja.github.io/neuromorphic_octopus_eye/
comment: 15 pages, 14 figures, 1 table. Accepted at CVPR 2025 (Workshop on Event-based Vision)
☆ Deep Learning in Concealed Dense Prediction
Deep learning is developing rapidly and handling common computer vision tasks well. It is time to pay attention to more complex vision tasks, as model size, knowledge, and reasoning capabilities continue to improve. In this paper, we introduce and review a family of complex tasks, termed Concealed Dense Prediction (CDP), which has great value in agriculture, industry, etc. CDP's intrinsic trait is that the targets are concealed in their surroundings, thus fully perceiving them requires fine-grained representations, prior knowledge, auxiliary reasoning, etc. The contributions of this review are three-fold: (i) We introduce the scope, characteristics, and challenges specific to CDP tasks and emphasize their essential differences from generic vision tasks. (ii) We develop a taxonomy based on concealment counteracting to summarize deep learning efforts in CDP through experiments on three tasks. We compare 25 state-of-the-art methods across 12 widely used concealed datasets. (iii) We discuss the potential applications of CDP in the large model era and summarize 6 potential research directions. We offer perspectives for the future development of CDP by constructing a large-scale multimodal instruction fine-tuning dataset, CvpINST, and a concealed visual perception agent, CvpAgent.
comment: Technique Report
☆ AgentPolyp: Accurate Polyp Segmentation via Image Enhancement Agent
Since human and environmental factors interfere, captured polyp images usually suffer from issues such as dim lighting, blur, and overexposure, which pose challenges for downstream polyp segmentation tasks. To address the challenges of noise-induced degradation in polyp images, we present AgentPolyp, a novel framework integrating CLIP-based semantic guidance and dynamic image enhancement with a lightweight neural network for segmentation. The agent first evaluates image quality using CLIP-driven semantic analysis (e.g., identifying ``low-contrast polyps with vascular textures") and adapts reinforcement learning strategies to dynamically apply multi-modal enhancement operations (e.g., denoising, contrast adjustment). A quality assessment feedback loop optimizes pixel-level enhancement and segmentation focus in a collaborative manner, ensuring robust preprocessing before neural network segmentation. This modular architecture supports plug-and-play extensions for various enhancement algorithms and segmentation networks, meeting deployment requirements for endoscopic devices.
☆ Adaptive Decision Boundary for Few-Shot Class-Incremental Learning
Few-Shot Class-Incremental Learning (FSCIL) aims to continuously learn new classes from a limited set of training samples without forgetting knowledge of previously learned classes. Conventional FSCIL methods typically build a robust feature extractor during the base training session with abundant training samples and subsequently freeze this extractor, only fine-tuning the classifier in subsequent incremental phases. However, current strategies primarily focus on preventing catastrophic forgetting, considering only the relationship between novel and base classes, without paying attention to the specific decision spaces of each class. To address this challenge, we propose a plug-and-play Adaptive Decision Boundary Strategy (ADBS), which is compatible with most FSCIL methods. Specifically, we assign a specific decision boundary to each class and adaptively adjust these boundaries during training to optimally refine the decision spaces for the classes in each session. Furthermore, to amplify the distinctiveness between classes, we employ a novel inter-class constraint loss that optimizes the decision boundaries and prototypes for each class. Extensive experiments on three benchmarks, namely CIFAR100, miniImageNet, and CUB200, demonstrate that incorporating our ADBS method with existing FSCIL techniques significantly improves performance, achieving overall state-of-the-art results.
Self-Supervised Enhancement of Forward-Looking Sonar Images: Bridging Cross-Modal Degradation Gaps through Feature Space Transformation and Multi-Frame Fusion
Enhancing forward-looking sonar images is critical for accurate underwater target detection. Current deep learning methods mainly rely on supervised training with simulated data, but the difficulty in obtaining high-quality real-world paired data limits their practical use and generalization. Although self-supervised approaches from remote sensing partially alleviate data shortages, they neglect the cross-modal degradation gap between sonar and remote sensing images. Directly transferring pretrained weights often leads to overly smooth sonar images, detail loss, and insufficient brightness. To address this, we propose a feature-space transformation that maps sonar images from the pixel domain to a robust feature domain, effectively bridging the degradation gap. Additionally, our self-supervised multi-frame fusion strategy leverages complementary inter-frame information to naturally remove speckle noise and enhance target-region brightness. Experiments on three self-collected real-world forward-looking sonar datasets show that our method significantly outperforms existing approaches, effectively suppressing noise, preserving detailed edges, and substantially improving brightness, demonstrating strong potential for underwater target detection applications.
☆ AFiRe: Anatomy-Driven Self-Supervised Learning for Fine-Grained Representation in Radiographic Images
Current self-supervised methods, such as contrastive learning, predominantly focus on global discrimination, neglecting the critical fine-grained anatomical details required for accurate radiographic analysis. To address this challenge, we propose an Anatomy-driven self-supervised framework for enhancing Fine-grained Representation in radiographic image analysis (AFiRe). The core idea of AFiRe is to align the anatomical consistency with the unique token-processing characteristics of Vision Transformer. Specifically, AFiRe synergistically performs two self-supervised schemes: (i) Token-wise anatomy-guided contrastive learning, which aligns image tokens based on structural and categorical consistency, thereby enhancing fine-grained spatial-anatomical discrimination; (ii) Pixel-level anomaly-removal restoration, which particularly focuses on local anomalies, thereby refining the learned discrimination with detailed geometrical information. Additionally, we propose Synthetic Lesion Mask to enhance anatomical diversity while preserving intra-consistency, which is typically corrupted by traditional data augmentations, such as Cropping and Affine transformations. Experimental results show that AFiRe: (i) provides robust anatomical discrimination, achieving more cohesive feature clusters compared to state-of-the-art contrastive learning methods; (ii) demonstrates superior generalization, surpassing 7 radiography-specific self-supervised methods in multi-label classification tasks with limited labeling; and (iii) integrates fine-grained information, enabling precise anomaly detection using only image-level annotations.
☆ An Efficient and Mixed Heterogeneous Model for Image Restoration
Image restoration~(IR), as a fundamental multimedia data processing task, has a significant impact on downstream visual applications. In recent years, researchers have focused on developing general-purpose IR models capable of handling diverse degradation types, thereby reducing the cost and complexity of model development. Current mainstream approaches are based on three architectural paradigms: CNNs, Transformers, and Mambas. CNNs excel in efficient inference, whereas Transformers and Mamba excel at capturing long-range dependencies and modeling global contexts. While each architecture has demonstrated success in specialized, single-task settings, limited efforts have been made to effectively integrate heterogeneous architectures to jointly address diverse IR challenges. To bridge this gap, we propose RestorMixer, an efficient and general-purpose IR model based on mixed-architecture fusion. RestorMixer adopts a three-stage encoder-decoder structure, where each stage is tailored to the resolution and feature characteristics of the input. In the initial high-resolution stage, CNN-based blocks are employed to rapidly extract shallow local features. In the subsequent stages, we integrate a refined multi-directional scanning Mamba module with a multi-scale window-based self-attention mechanism. This hierarchical and adaptive design enables the model to leverage the strengths of CNNs in local feature extraction, Mamba in global context modeling, and attention mechanisms in dynamic feature refinement. Extensive experimental results demonstrate that RestorMixer achieves leading performance across multiple IR tasks while maintaining high inference efficiency. The official code can be accessed at https://github.com/ClimBin/RestorMixer.
☆ Recognition of Geometrical Shapes by Dictionary Learning
Dictionary learning is a versatile method to produce an overcomplete set of vectors, called atoms, to represent a given input with only a few atoms. In the literature, it has been used primarily for tasks that explore its powerful representation capabilities, such as for image reconstruction. In this work, we present a first approach to make dictionary learning work for shape recognition, considering specifically geometrical shapes. As we demonstrate, the choice of the underlying optimization method has a significant impact on recognition quality. Experimental results confirm that dictionary learning may be an interesting method for shape recognition tasks.
comment: 6 pages, 4 figures, ACDSA 2025 conference
☆ Cross-Frequency Implicit Neural Representation with Self-Evolving Parameters
Implicit neural representation (INR) has emerged as a powerful paradigm for visual data representation. However, classical INR methods represent data in the original space mixed with different frequency components, and several feature encoding parameters (e.g., the frequency parameter $\omega$ or the rank $R$) need manual configurations. In this work, we propose a self-evolving cross-frequency INR using the Haar wavelet transform (termed CF-INR), which decouples data into four frequency components and employs INRs in the wavelet space. CF-INR allows the characterization of different frequency components separately, thus enabling higher accuracy for data representation. To more precisely characterize cross-frequency components, we propose a cross-frequency tensor decomposition paradigm for CF-INR with self-evolving parameters, which automatically updates the rank parameter $R$ and the frequency parameter $\omega$ for each frequency component through self-evolving optimization. This self-evolution paradigm eliminates the laborious manual tuning of these parameters, and learns a customized cross-frequency feature encoding configuration for each dataset. We evaluate CF-INR on a variety of visual data representation and recovery tasks, including image regression, inpainting, denoising, and cloud removal. Extensive experiments demonstrate that CF-INR outperforms state-of-the-art methods in each case.
☆ Towards Efficient Partially Relevant Video Retrieval with Active Moment Discovering
Partially relevant video retrieval (PRVR) is a practical yet challenging task in text-to-video retrieval, where videos are untrimmed and contain much background content. The pursuit here is of both effective and efficient solutions to capture the partial correspondence between text queries and untrimmed videos. Existing PRVR methods, which typically focus on modeling multi-scale clip representations, however, suffer from content independence and information redundancy, impairing retrieval performance. To overcome these limitations, we propose a simple yet effective approach with active moment discovering (AMDNet). We are committed to discovering video moments that are semantically consistent with their queries. By using learnable span anchors to capture distinct moments and applying masked multi-moment attention to emphasize salient moments while suppressing redundant backgrounds, we achieve more compact and informative video representations. To further enhance moment modeling, we introduce a moment diversity loss to encourage different moments of distinct regions and a moment relevance loss to promote semantically query-relevant moments, which cooperate with a partially relevant retrieval loss for end-to-end optimization. Extensive experiments on two large-scale video datasets (\ie, TVR and ActivityNet Captions) demonstrate the superiority and efficiency of our AMDNet. In particular, AMDNet is about 15.5 times smaller (\#parameters) while 6.0 points higher (SumR) than the up-to-date method GMMFormer on TVR.
comment: Accepted by IEEE Transactions on Multimedia (TMM) on January 19, 2025. The code is available at https://github.com/songpipi/AMDNet
☆ Embedding Radiomics into Vision Transformers for Multimodal Medical Image Classification
Background: Deep learning has significantly advanced medical image analysis, with Vision Transformers (ViTs) offering a powerful alternative to convolutional models by modeling long-range dependencies through self-attention. However, ViTs are inherently data-intensive and lack domain-specific inductive biases, limiting their applicability in medical imaging. In contrast, radiomics provides interpretable, handcrafted descriptors of tissue heterogeneity but suffers from limited scalability and integration into end-to-end learning frameworks. In this work, we propose the Radiomics-Embedded Vision Transformer (RE-ViT) that combines radiomic features with data-driven visual embeddings within a ViT backbone. Purpose: To develop a hybrid RE-ViT framework that integrates radiomics and patch-wise ViT embeddings through early fusion, enhancing robustness and performance in medical image classification. Methods: Following the standard ViT pipeline, images were divided into patches. For each patch, handcrafted radiomic features were extracted and fused with linearly projected pixel embeddings. The fused representations were normalized, positionally encoded, and passed to the ViT encoder. A learnable [CLS] token aggregated patch-level information for classification. We evaluated RE-ViT on three public datasets (including BUSI, ChestXray2017, and Retinal OCT) using accuracy, macro AUC, sensitivity, and specificity. RE-ViT was benchmarked against CNN-based (VGG-16, ResNet) and hybrid (TransMed) models. Results: RE-ViT achieved state-of-the-art results: on BUSI, AUC=0.950+/-0.011; on ChestXray2017, AUC=0.989+/-0.004; on Retinal OCT, AUC=0.986+/-0.001, which outperforms other comparison models. Conclusions: The RE-ViT framework effectively integrates radiomics with ViT architectures, demonstrating improved performance and generalizability across multimodal medical image classification tasks.
comment: 27 pages, 3 figures
☆ InterAnimate: Taming Region-aware Diffusion Model for Realistic Human Interaction Animation
Recent video generation research has focused heavily on isolated actions, leaving interactive motions-such as hand-face interactions-largely unexamined. These interactions are essential for emerging biometric authentication systems, which rely on interactive motion-based anti-spoofing approaches. From a security perspective, there is a growing need for large-scale, high-quality interactive videos to train and strengthen authentication models. In this work, we introduce a novel paradigm for animating realistic hand-face interactions. Our approach simultaneously learns spatio-temporal contact dynamics and biomechanically plausible deformation effects, enabling natural interactions where hand movements induce anatomically accurate facial deformations while maintaining collision-free contact. To facilitate this research, we present InterHF, a large-scale hand-face interaction dataset featuring 18 interaction patterns and 90,000 annotated videos. Additionally, we propose InterAnimate, a region-aware diffusion model designed specifically for interaction animation. InterAnimate leverages learnable spatial and temporal latents to effectively capture dynamic interaction priors and integrates a region-aware interaction mechanism that injects these priors into the denoising process. To the best of our knowledge, this work represents the first large-scale effort to systematically study human hand-face interactions. Qualitative and quantitative results show InterAnimate produces highly realistic animations, setting a new benchmark. Code and data will be made public to advance research.
comment: under preview
☆ Fine-Grained Rib Fracture Diagnosis with Hyperbolic Embeddings: A Detailed Annotation Framework and Multi-Label Classification Model
Accurate rib fracture identification and classification are essential for treatment planning. However, existing datasets often lack fine-grained annotations, particularly regarding rib fracture characterization, type, and precise anatomical location on individual ribs. To address this, we introduce a novel rib fracture annotation protocol tailored for fracture classification. Further, we enhance fracture classification by leveraging cross-modal embeddings that bridge radiological images and clinical descriptions. Our approach employs hyperbolic embeddings to capture the hierarchical nature of fracture, mapping visual features and textual descriptions into a shared non-Euclidean manifold. This framework enables more nuanced similarity computations between imaging characteristics and clinical descriptions, accounting for the inherent hierarchical relationships in fracture taxonomy. Experimental results demonstrate that our approach outperforms existing methods across multiple classification tasks, with average recall improvements of 6% on the AirRib dataset and 17.5% on the public RibFrac dataset.
☆ CDUPatch: Color-Driven Universal Adversarial Patch Attack for Dual-Modal Visible-Infrared Detectors
Adversarial patches are widely used to evaluate the robustness of object detection systems in real-world scenarios. These patches were initially designed to deceive single-modal detectors (e.g., visible or infrared) and have recently been extended to target visible-infrared dual-modal detectors. However, existing dual-modal adversarial patch attacks have limited attack effectiveness across diverse physical scenarios. To address this, we propose CDUPatch, a universal cross-modal patch attack against visible-infrared object detectors across scales, views, and scenarios. Specifically, we observe that color variations lead to different levels of thermal absorption, resulting in temperature differences in infrared imaging. Leveraging this property, we propose an RGB-to-infrared adapter that maps RGB patches to infrared patches, enabling unified optimization of cross-modal patches. By learning an optimal color distribution on the adversarial patch, we can manipulate its thermal response and generate an adversarial infrared texture. Additionally, we introduce a multi-scale clipping strategy and construct a new visible-infrared dataset, MSDrone, which contains aerial vehicle images in varying scales and perspectives. These data augmentation strategies enhance the robustness of our patch in real-world conditions. Experiments on four benchmark datasets (e.g., DroneVehicle, LLVIP, VisDrone, MSDrone) show that our method outperforms existing patch attacks in the digital domain. Extensive physical tests further confirm strong transferability across scales, views, and scenarios.
☆ PuzzleBench: A Fully Dynamic Evaluation Framework for Large Multimodal Models on Puzzle Solving
Large Multimodal Models (LMMs) have demonstrated impressive capabilities across a wide range of multimodal tasks, achieving ever-increasing performance on various evaluation benchmarks. However, existing benchmarks are typically static and often overlap with pre-training datasets, leading to fixed complexity constraints and substantial data contamination issues. Meanwhile, manually annotated datasets are labor-intensive, time-consuming, and subject to human bias and inconsistency, leading to reliability and reproducibility issues. To address these problems, we propose a fully dynamic multimodal evaluation framework, named Open-ended Visual Puzzle Generation (OVPG), which aims to generate fresh, diverse, and verifiable evaluation data automatically in puzzle-solving tasks. Specifically, the OVPG pipeline consists of a raw material sampling module, a visual content generation module, and a puzzle rule design module, which ensures that each evaluation instance is primitive, highly randomized, and uniquely solvable, enabling continual adaptation to the evolving capabilities of LMMs. Built upon OVPG, we construct PuzzleBench, a dynamic and scalable benchmark comprising 11,840 VQA samples. It features six carefully designed puzzle tasks targeting three core LMM competencies, visual recognition, logical reasoning, and context understanding. PuzzleBench differs from static benchmarks that quickly become outdated. It enables ongoing dataset refreshing through OVPG and a rich set of open-ended puzzle designs, allowing seamless adaptation to the evolving capabilities of LMMs.
☆ Bringing together invertible UNets with invertible attention modules for memory-efficient diffusion models
Diffusion models have recently gained state of the art performance on many image generation tasks. However, most models require significant computational resources to achieve this. This becomes apparent in the application of medical image synthesis due to the 3D nature of medical datasets like CT-scans, MRIs, electron microscope, etc. In this paper we propose a novel architecture for a single GPU memory-efficient training for diffusion models for high dimensional medical datasets. The proposed model is built by using an invertible UNet architecture with invertible attention modules. This leads to the following two contributions: 1. denoising diffusion models and thus enabling memory usage to be independent of the dimensionality of the dataset, and 2. reducing the energy usage during training. While this new model can be applied to a multitude of image generation tasks, we showcase its memory-efficiency on the 3D BraTS2020 dataset leading to up to 15\% decrease in peak memory consumption during training with comparable results to SOTA while maintaining the image quality.
☆ Safe-Construct: Redefining Construction Safety Violation Recognition as 3D Multi-View Engagement Task CVPR
Recognizing safety violations in construction environments is critical yet remains underexplored in computer vision. Existing models predominantly rely on 2D object detection, which fails to capture the complexities of real-world violations due to: (i) an oversimplified task formulation treating violation recognition merely as object detection, (ii) inadequate validation under realistic conditions, (iii) absence of standardized baselines, and (iv) limited scalability from the unavailability of synthetic dataset generators for diverse construction scenarios. To address these challenges, we introduce Safe-Construct, the first framework that reformulates violation recognition as a 3D multi-view engagement task, leveraging scene-level worker-object context and 3D spatial understanding. We also propose the Synthetic Indoor Construction Site Generator (SICSG) to create diverse, scalable training data, overcoming data limitations. Safe-Construct achieves a 7.6% improvement over state-of-the-art methods across four violation types. We rigorously evaluate our approach in near-realistic settings, incorporating four violations, four workers, 14 objects, and challenging conditions like occlusions (worker-object, worker-worker) and variable illumination (back-lighting, overexposure, sunlight). By integrating 3D multi-view spatial understanding and synthetic data generation, Safe-Construct sets a new benchmark for scalable and robust safety monitoring in high-risk industries. Project Website: https://Safe-Construct.github.io/Safe-Construct
comment: CVPR Workshop 2025; Project Website: https://Safe-Construct.github.io/Safe-Construct
☆ Large Language Model-Informed Feature Discovery Improves Prediction and Interpretation of Credibility Perceptions of Visual Content
In today's visually dominated social media landscape, predicting the perceived credibility of visual content and understanding what drives human judgment are crucial for countering misinformation. However, these tasks are challenging due to the diversity and richness of visual features. We introduce a Large Language Model (LLM)-informed feature discovery framework that leverages multimodal LLMs, such as GPT-4o, to evaluate content credibility and explain its reasoning. We extract and quantify interpretable features using targeted prompts and integrate them into machine learning models to improve credibility predictions. We tested this approach on 4,191 visual social media posts across eight topics in science, health, and politics, using credibility ratings from 5,355 crowdsourced workers. Our method outperformed zero-shot GPT-based predictions by 13 percent in R2, and revealed key features like information concreteness and image format. We discuss the implications for misinformation mitigation, visual credibility, and the role of LLMs in social science.
comment: 26 pages
☆ Weather-Aware Object Detection Transformer for Domain Adaptation
RT-DETRs have shown strong performance across various computer vision tasks but are known to degrade under challenging weather conditions such as fog. In this work, we investigate three novel approaches to enhance RT-DETR robustness in foggy environments: (1) Domain Adaptation via Perceptual Loss, which distills domain-invariant features from a teacher network to a student using perceptual supervision; (2) Weather Adaptive Attention, which augments the attention mechanism with fog-sensitive scaling by introducing an auxiliary foggy image stream; and (3) Weather Fusion Encoder, which integrates a dual-stream encoder architecture that fuses clear and foggy image features via multi-head self and cross-attention. Despite the architectural innovations, none of the proposed methods consistently outperform the baseline RT-DETR. We analyze the limitations and potential causes, offering insights for future research in weather-aware object detection.
☆ Can Vision-Language Models Understand and Interpret Dynamic Gestures from Pedestrians? Pilot Datasets and Exploration Towards Instructive Nonverbal Commands for Cooperative Autonomous Vehicles
In autonomous driving, it is crucial to correctly interpret traffic gestures (TGs), such as those of an authority figure providing orders or instructions, or a pedestrian signaling the driver, to ensure a safe and pleasant traffic environment for all road users. This study investigates the capabilities of state-of-the-art vision-language models (VLMs) in zero-shot interpretation, focusing on their ability to caption and classify human gestures in traffic contexts. We create and publicly share two custom datasets with varying formal and informal TGs, such as 'Stop', 'Reverse', 'Hail', etc. The datasets are "Acted TG (ATG)" and "Instructive TG In-The-Wild (ITGI)". They are annotated with natural language, describing the pedestrian's body position and gesture. We evaluate models using three methods utilizing expert-generated captions as baseline and control: (1) caption similarity, (2) gesture classification, and (3) pose sequence reconstruction similarity. Results show that current VLMs struggle with gesture understanding: sentence similarity averages below 0.59, and classification F1 scores reach only 0.14-0.39, well below the expert baseline of 0.70. While pose reconstruction shows potential, it requires more data and refined metrics to be reliable. Our findings reveal that although some SOTA VLMs can interpret zero-shot human traffic gestures, none are accurate and robust enough to be trustworthy, emphasizing the need for further research in this domain.
☆ DAAF:Degradation-Aware Adaptive Fusion Framework for Robust Infrared and Visible Images Fusion
Existing infrared and visible image fusion(IVIF) algorithms often prioritize high-quality images, neglecting image degradation such as low light and noise, which limits the practical potential. This paper propose Degradation-Aware Adaptive image Fusion (DAAF), which achieves unified modeling of adaptive degradation optimization and image fusion. Specifically, DAAF comprises an auxiliary Adaptive Degradation Optimization Network (ADON) and a Feature Interactive Local-Global Fusion (FILGF) Network. Firstly, ADON includes infrared and visible-light branches. Within the infrared branch, frequency-domain feature decomposition and extraction are employed to isolate Gaussian and stripe noise. In the visible-light branch, Retinex decomposition is applied to extract illumination and reflectance components, enabling complementary enhancement of detail and illumination distribution. Subsequently, FILGF performs interactive multi-scale local-global feature fusion. Local feature fusion consists of intra-inter model feature complement, while global feature fusion is achieved through a interactive cross-model attention. Extensive experiments have shown that DAAF outperforms current IVIF algorithms in normal and complex degradation scenarios.
♻ ☆ Reference-Based 3D-Aware Image Editing with Triplanes CVPR 2025
Generative Adversarial Networks (GANs) have emerged as powerful tools for high-quality image generation and real image editing by manipulating their latent spaces. Recent advancements in GANs include 3D-aware models such as EG3D, which feature efficient triplane-based architectures capable of reconstructing 3D geometry from single images. However, limited attention has been given to providing an integrated framework for 3D-aware, high-quality, reference-based image editing. This study addresses this gap by exploring and demonstrating the effectiveness of the triplane space for advanced reference-based edits. Our novel approach integrates encoding, automatic localization, spatial disentanglement of triplane features, and fusion learning to achieve the desired edits. We demonstrate how our approach excels across diverse domains, including human faces, 360-degree heads, animal faces, partially stylized edits like cartoon faces, full-body clothing edits, and edits on class-agnostic samples. Our method shows state-of-the-art performance over relevant latent direction, text, and image-guided 2D and 3D-aware diffusion and GAN methods, both qualitatively and quantitatively.
comment: CVPR 2025 Highlight. Includes supplementary material
♻ ☆ InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
We introduce InternVL3, a significant advancement in the InternVL series featuring a native multimodal pre-training paradigm. Rather than adapting a text-only large language model (LLM) into a multimodal large language model (MLLM) that supports visual inputs, InternVL3 jointly acquires multimodal and linguistic capabilities from both diverse multimodal data and pure-text corpora during a single pre-training stage. This unified training paradigm effectively addresses the complexities and alignment challenges commonly encountered in conventional post-hoc training pipelines for MLLMs. To further improve performance and scalability, InternVL3 incorporates variable visual position encoding (V2PE) to support extended multimodal contexts, employs advanced post-training techniques such as supervised fine-tuning (SFT) and mixed preference optimization (MPO), and adopts test-time scaling strategies alongside an optimized training infrastructure. Extensive empirical evaluations demonstrate that InternVL3 delivers superior performance across a wide range of multi-modal tasks. In particular, InternVL3-78B achieves a score of 72.2 on the MMMU benchmark, setting a new state-of-the-art among open-source MLLMs. Its capabilities remain highly competitive with leading proprietary models, including ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 2.5 Pro, while also maintaining strong pure-language proficiency. In pursuit of open-science principles, we will publicly release both the training data and model weights to foster further research and development in next-generation MLLMs.
comment: Technical Report
♻ ☆ Gaussian Differentially Private Human Faces Under a Face Radial Curve Representation
In this paper we consider the problem of releasing a Gaussian Differentially Private (GDP) 3D human face. The human face is a complex structure with many features and inherently tied to one's identity. Protecting this data, in a formally private way, is important yet challenging given the dimensionality of the problem. We extend approximate DP techniques for functional data to the GDP framework. We further propose a novel representation, face radial curves, of a 3D face as a set of functions and then utilize our proposed GDP functional data mechanism. To preserve the shape of the face while injecting noise we rely on tools from shape analysis for our novel representation of the face. We show that our method preserves the shape of the average face and injects less noise than traditional methods for the same privacy budget. Our mechanism consists of two primary components, the first is generally applicable to function value summaries (as are commonly found in nonparametric statistics or functional data analysis) while the second is general to disk-like surfaces and hence more applicable than just to human faces.
comment: 19 pages, 10 figures
♻ ☆ Kimi-VL Technical Report
We present Kimi-VL, an efficient open-source Mixture-of-Experts (MoE) vision-language model (VLM) that offers advanced multimodal reasoning, long-context understanding, and strong agent capabilities - all while activating only 2.8B parameters in its language decoder (Kimi-VL-A3B). Kimi-VL demonstrates strong performance across challenging domains: as a general-purpose VLM, Kimi-VL excels in multi-turn agent tasks (e.g., OSWorld), matching flagship models. Furthermore, it exhibits remarkable capabilities across diverse challenging vision language tasks, including college-level image and video comprehension, OCR, mathematical reasoning, and multi-image understanding. In comparative evaluations, it effectively competes with cutting-edge efficient VLMs such as GPT-4o-mini, Qwen2.5-VL-7B, and Gemma-3-12B-IT, while surpassing GPT-4o in several key domains. Kimi-VL also advances in processing long contexts and perceiving clearly. With a 128K extended context window, Kimi-VL can process diverse long inputs, achieving impressive scores of 64.5 on LongVideoBench and 35.1 on MMLongBench-Doc. Its native-resolution vision encoder, MoonViT, further allows it to see and understand ultra-high-resolution visual inputs, achieving 83.2 on InfoVQA and 34.5 on ScreenSpot-Pro, while maintaining lower computational cost for common tasks. Building upon Kimi-VL, we introduce an advanced long-thinking variant: Kimi-VL-Thinking. Developed through long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL), this model exhibits strong long-horizon reasoning capabilities. It achieves scores of 61.7 on MMMU, 36.8 on MathVision, and 71.3 on MathVista while maintaining the compact 2.8B activated LLM parameters, setting a new standard for efficient multimodal thinking models. Code and models are publicly accessible at https://github.com/MoonshotAI/Kimi-VL.
♻ ☆ Breaking the Data Barrier -- Building GUI Agents Through Task Generalization
Graphical User Interface (GUI) agents offer cross-platform solutions for automating complex digital tasks, with significant potential to transform productivity workflows. However, their performance is often constrained by the scarcity of high-quality trajectory data. To address this limitation, we propose training Vision Language Models (VLMs) on data-rich, reasoning-intensive tasks during a dedicated mid-training stage, and then examine how incorporating these tasks facilitates generalization to GUI planning scenarios. Specifically, we explore a range of tasks with readily available instruction-tuning data, including GUI perception, multimodal reasoning, and textual reasoning. Through extensive experiments across 11 mid-training tasks, we demonstrate that: (1) Task generalization proves highly effective, yielding substantial improvements across most settings. For instance, multimodal mathematical reasoning enhances performance on AndroidWorld by an absolute 6.3%. Remarkably, text-only mathematical data significantly boosts GUI web agent performance, achieving a 5.6% improvement on WebArena and 5.4% improvement on AndroidWorld, underscoring notable cross-modal generalization from text-based to visual domains; (2) Contrary to prior assumptions, GUI perception data - previously considered closely aligned with GUI agent tasks and widely utilized for training - has a comparatively limited impact on final performance; (3) Building on these insights, we identify the most effective mid-training tasks and curate optimized mixture datasets, resulting in absolute performance gains of 8.0% on WebArena and 12.2% on AndroidWorld. Our work provides valuable insights into cross-domain knowledge transfer for GUI agents and offers a practical approach to addressing data scarcity challenges in this emerging field. The code, data and models will be available at https://github.com/hkust-nlp/GUIMid.
comment: 24 pages, 11 figures
♻ ☆ SC-NeRF: NeRF-based Point Cloud Reconstruction using a Stationary Camera for Agricultural Applications
This paper presents a NeRF-based framework for point cloud (PCD) reconstruction, specifically designed for indoor high-throughput plant phenotyping facilities. Traditional NeRF-based reconstruction methods require cameras to move around stationary objects, but this approach is impractical for high-throughput environments where objects are rapidly imaged while moving on conveyors or rotating pedestals. To address this limitation, we develop a variant of NeRF-based PCD reconstruction that uses a single stationary camera to capture images as the object rotates on a pedestal. Our workflow comprises COLMAP-based pose estimation, a straightforward pose transformation to simulate camera movement, and subsequent standard NeRF training. A defined Region of Interest (ROI) excludes irrelevant scene data, enabling the generation of high-resolution point clouds (10M points). Experimental results demonstrate excellent reconstruction fidelity, with precision-recall analyses yielding an F-score close to 100.00 across all evaluated plant objects. Although pose estimation remains computationally intensive with a stationary camera setup, overall training and reconstruction times are competitive, validating the method's feasibility for practical high-throughput indoor phenotyping applications. Our findings indicate that high-quality NeRF-based 3D reconstructions are achievable using a stationary camera, eliminating the need for complex camera motion or costly imaging equipment. This approach is especially beneficial when employing expensive and delicate instruments, such as hyperspectral cameras, for 3D plant phenotyping. Future work will focus on optimizing pose estimation techniques and further streamlining the methodology to facilitate seamless integration into automated, high-throughput 3D phenotyping pipelines.
♻ ☆ Deep Learning-Based Longitudinal Prediction of Childhood Myopia Progression Using Fundus Image Sequences and Baseline Refraction Data
Childhood myopia constitutes a significant global health concern. It exhibits an escalating prevalence and has the potential to evolve into severe, irreversible conditions that detrimentally impact familial well-being and create substantial economic costs. Contemporary research underscores the importance of precisely predicting myopia progression to enable timely and effective interventions, thereby averting severe visual impairment in children. Such predictions predominantly rely on subjective clinical assessments, which are inherently biased and resource-intensive, thus hindering their widespread application. In this study, we introduce a novel, high-accuracy method for quantitatively predicting the myopic trajectory and myopia risk in children using only fundus images and baseline refraction data. This approach was validated through a six-year longitudinal study of 3,408 children in Henan, utilizing 16,211 fundus images and corresponding refractive data. Our method based on deep learning demonstrated predictive accuracy with an error margin of 0.311D per year and AUC scores of 0.944 and 0.995 for forecasting the risks of developing myopia and high myopia, respectively. These findings confirm the utility of our model in supporting early intervention strategies and in significantly reducing healthcare costs, particularly by obviating the need for additional metadata and repeated consultations. Furthermore, our method was designed to rely only on fundus images and refractive error data, without the need for meta data or multiple inquiries from doctors, strongly reducing the associated medical costs and facilitating large-scale screening. Our model can even provide good predictions based on only a single time measurement. Consequently, the proposed method is an important means to reduce medical inequities caused by economic disparities.
♻ ☆ EchoMask: Speech-Queried Attention-based Mask Modeling for Holistic Co-Speech Motion Generation
Masked modeling framework has shown promise in co-speech motion generation. However, it struggles to identify semantically significant frames for effective motion masking. In this work, we propose a speech-queried attention-based mask modeling framework for co-speech motion generation. Our key insight is to leverage motion-aligned speech features to guide the masked motion modeling process, selectively masking rhythm-related and semantically expressive motion frames. Specifically, we first propose a motion-audio alignment module (MAM) to construct a latent motion-audio joint space. In this space, both low-level and high-level speech features are projected, enabling motion-aligned speech representation using learnable speech queries. Then, a speech-queried attention mechanism (SQA) is introduced to compute frame-level attention scores through interactions between motion keys and speech queries, guiding selective masking toward motion frames with high attention scores. Finally, the motion-aligned speech features are also injected into the generation network to facilitate co-speech motion generation. Qualitative and quantitative evaluations confirm that our method outperforms existing state-of-the-art approaches, successfully producing high-quality co-speech motion.
comment: 12 pages, 12 figures
♻ ☆ Cognitive Disentanglement for Referring Multi-Object Tracking
As a significant application of multi-source information fusion in intelligent transportation perception systems, Referring Multi-Object Tracking (RMOT) involves localizing and tracking specific objects in video sequences based on language references. However, existing RMOT approaches often treat language descriptions as holistic embeddings and struggle to effectively integrate the rich semantic information contained in language expressions with visual features. This limitation is especially apparent in complex scenes requiring comprehensive understanding of both static object attributes and spatial motion information. In this paper, we propose a Cognitive Disentanglement for Referring Multi-Object Tracking (CDRMT) framework that addresses these challenges. It adapts the "what" and "where" pathways from the human visual processing system to RMOT tasks. Specifically, our framework first establishes cross-modal connections while preserving modality-specific characteristics. It then disentangles language descriptions and hierarchically injects them into object queries, refining object understanding from coarse to fine-grained semantic levels. Finally, we reconstruct language representations based on visual features, ensuring that tracked objects faithfully reflect the referring expression. Extensive experiments on different benchmark datasets demonstrate that CDRMT achieves substantial improvements over state-of-the-art methods, with average gains of 6.0% in HOTA score on Refer-KITTI and 3.2% on Refer-KITTI-V2. Our approach advances the state-of-the-art in RMOT while simultaneously providing new insights into multi-source information fusion.
comment: 26 pages, 11 figures
♻ ☆ Think or Not Think: A Study of Explicit Thinking in Rule-Based Visual Reinforcement Fine-Tuning
This paper investigates the thinking process in rule-based reinforcement learning fine-tuning (RFT) for multi-modal large language models (MLLMs). We first propose CLS-RL for classification, using verifiable rewards to encourage MLLM thinking. Experiments show CLS-RL significantly outperforms SFT and yields a 'free-lunch' generalization effect (improving performance on unseen datasets after training on one dataset). We then question if this explicit thinking is always necessary for RFT. Challenging convention that explicit thinking is crucial for RFT, we introduce No-Thinking-RL, minimizing thinking via a simple equality accuracy reward. Experiments show No-Thinking-RL surpasses CLS-RL in in-domain and generalization abilities, with significantly less fine-tuning time. This suggests reducing thinking can improve MLLM fine-tuning efficiency and effectiveness for certain visual tasks. We hypothesize explicit thinking negatively impacts reward convergence during RFT. To test this, we propose the Think-After-Answerwer method to let models first output the answer and then generate thinking process to alliviate the negative impact of thinking. We further test No-Thinking-RL on diverse tasks (including math, spatial, puzzles) with 2B and 7B models. For 2B models, No-Thinking-RL outperforms thinking-based RFT for all tasks, even on math, with Think-After-Answerwer performing intermediately. For 7B models, performance is comparable on simple visual tasks, but RFT with thinking excels on complex reasoning (math). This implies when dealing with complex math problems, smaller models struggle with generating effective reasoning, hurting performance on complex tasks. Conversely, for simple visual tasks, thinking is not indispensable, and its removal can boost performance and reduce training time. We hope our findings offer insights for better understanding the effect of the thinking process in RFT.
comment: Preprint, work in progress. Add results on math, cvbench, and puzzle
♻ ☆ GarmentTracking: Category-Level Garment Pose Tracking CVPR 2023
Garments are important to humans. A visual system that can estimate and track the complete garment pose can be useful for many downstream tasks and real-world applications. In this work, we present a complete package to address the category-level garment pose tracking task: (1) A recording system VR-Garment, with which users can manipulate virtual garment models in simulation through a VR interface. (2) A large-scale dataset VR-Folding, with complex garment pose configurations in manipulation like flattening and folding. (3) An end-to-end online tracking framework GarmentTracking, which predicts complete garment pose both in canonical space and task space given a point cloud sequence. Extensive experiments demonstrate that the proposed GarmentTracking achieves great performance even when the garment has large non-rigid deformation. It outperforms the baseline approach on both speed and accuracy. We hope our proposed solution can serve as a platform for future research. Codes and datasets are available in https://garment-tracking.robotflow.ai.
comment: CVPR 2023
♻ ☆ SCA: Highly Efficient Semantic-Consistent Unrestricted Adversarial Attack
Deep neural network based systems deployed in sensitive environments are vulnerable to adversarial attacks. Unrestricted adversarial attacks typically manipulate the semantic content of an image (e.g., color or texture) to create adversarial examples that are both effective and photorealistic. Recent works have utilized the diffusion inversion process to map images into a latent space, where high-level semantics are manipulated by introducing perturbations. However, they often results in substantial semantic distortions in the denoised output and suffers from low efficiency. In this study, we propose a novel framework called Semantic-Consistent Unrestricted Adversarial Attacks (SCA), which employs an inversion method to extract edit-friendly noise maps and utilizes Multimodal Large Language Model (MLLM) to provide semantic guidance throughout the process. Under the condition of rich semantic information provided by MLLM, we perform the DDPM denoising process of each step using a series of edit-friendly noise maps, and leverage DPM Solver++ to accelerate this process, enabling efficient sampling with semantic consistency. Compared to existing methods, our framework enables the efficient generation of adversarial examples that exhibit minimal discernible semantic changes. Consequently, we for the first time introduce Semantic-Consistent Adversarial Examples (SCAE). Extensive experiments and visualizations have demonstrated the high efficiency of SCA, particularly in being on average 12 times faster than the state-of-the-art attacks. Our research can further draw attention to the security of multimedia information.
♻ ☆ LL-Gaussian: Low-Light Scene Reconstruction and Enhancement via Gaussian Splatting for Novel View Synthesis
Novel view synthesis (NVS) in low-light scenes remains a significant challenge due to degraded inputs characterized by severe noise, low dynamic range (LDR) and unreliable initialization. While recent NeRF-based approaches have shown promising results, most suffer from high computational costs, and some rely on carefully captured or pre-processed data--such as RAW sensor inputs or multi-exposure sequences--which severely limits their practicality. In contrast, 3D Gaussian Splatting (3DGS) enables real-time rendering with competitive visual fidelity; however, existing 3DGS-based methods struggle with low-light sRGB inputs, resulting in unstable Gaussian initialization and ineffective noise suppression. To address these challenges, we propose LL-Gaussian, a novel framework for 3D reconstruction and enhancement from low-light sRGB images, enabling pseudo normal-light novel view synthesis. Our method introduces three key innovations: 1) an end-to-end Low-Light Gaussian Initialization Module (LLGIM) that leverages dense priors from learning-based MVS approach to generate high-quality initial point clouds; 2) a dual-branch Gaussian decomposition model that disentangles intrinsic scene properties (reflectance and illumination) from transient interference, enabling stable and interpretable optimization; 3) an unsupervised optimization strategy guided by both physical constrains and diffusion prior to jointly steer decomposition and enhancement. Additionally, we contribute a challenging dataset collected in extreme low-light environments and demonstrate the effectiveness of LL-Gaussian. Compared to state-of-the-art NeRF-based methods, LL-Gaussian achieves up to 2,000 times faster inference and reduces training time to just 2%, while delivering superior reconstruction and rendering quality.
♻ ☆ GUI-R1 : A Generalist R1-Style Vision-Language Action Model For GUI Agents
Existing efforts in building Graphical User Interface (GUI) agents largely rely on the training paradigm of supervised fine-tuning on Large Vision-Language Models (LVLMs). However, this approach not only demands extensive amounts of training data but also struggles to effectively understand GUI screenshots and generalize to unseen interfaces. The issue significantly limits its application in real-world scenarios, especially for high-level tasks. Inspired by Reinforcement Fine-Tuning (RFT) in large reasoning models (e.g., DeepSeek-R1), which efficiently enhances the problem-solving capabilities of large language models in real-world settings, we propose \name, the first reinforcement learning framework designed to enhance the GUI capabilities of LVLMs in high-level real-world task scenarios, through unified action space rule modeling. By leveraging a small amount of carefully curated high-quality data across multiple platforms (including Windows, Linux, MacOS, Android, and Web) and employing policy optimization algorithms such as Group Relative Policy Optimization (GRPO) to update the model, \name achieves superior performance using only 0.02\% of the data (3K vs. 13M) compared to previous state-of-the-art methods like OS-Atlas across eight benchmarks spanning three different platforms (mobile, desktop, and web). These results demonstrate the immense potential of reinforcement learning based on unified action space rule modeling in improving the execution capabilities of LVLMs for real-world GUI agent tasks.
♻ ☆ MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning
DeepSeek R1, and o1 have demonstrated powerful reasoning capabilities in the text domain through stable large-scale reinforcement learning. To enable broader applications, some works have attempted to transfer these capabilities to multimodal reasoning. However, these efforts have been limited by the limited difficulty of selected tasks and relatively small training scales, making it challenging to demonstrate strong multimodal reasoning abilities. To address this gap, we introduce the MMK12 dataset and MM-EUREKA with 7B and 32B parameters. The former is a high-quality multimodal mathematics reasoning dataset featuring diverse knowledge domains with human-verified answers and solution processes. The latter is a multimodal model employing rule-based reinforcement learning on MMK12, utilizing online filtering and two-stage training strategy to enhance training stability. MM-EUREKA demonstrates remarkable performance gains in multimodal mathematical reasoning, outperforming previous powerful models like InternVL2.5-78B or InternVL2.5-38B-MPO. In particular, MM-EUREKA achieves competitive or superior performance compared to both open-source and closed-source models, and trails slightly behind o1 in multidisciplinary reasoning tasks. We open-source our complete pipeline to foster further research in this area. We release all our codes, models, data, etc. at https://github.com/ModalMinds/MM-EUREKA
♻ ☆ ARLON: Boosting Diffusion Transformers with Autoregressive Models for Long Video Generation ICLR2025
Text-to-video models have recently undergone rapid and substantial advancements. Nevertheless, due to limitations in data and computational resources, achieving efficient generation of long videos with rich motion dynamics remains a significant challenge. To generate high-quality, dynamic, and temporally consistent long videos, this paper presents ARLON, a novel framework that boosts diffusion Transformers with autoregressive models for long video generation, by integrating the coarse spatial and long-range temporal information provided by the AR model to guide the DiT model. Specifically, ARLON incorporates several key innovations: 1) A latent Vector Quantized Variational Autoencoder (VQ-VAE) compresses the input latent space of the DiT model into compact visual tokens, bridging the AR and DiT models and balancing the learning complexity and information density; 2) An adaptive norm-based semantic injection module integrates the coarse discrete visual units from the AR model into the DiT model, ensuring effective guidance during video generation; 3) To enhance the tolerance capability of noise introduced from the AR inference, the DiT model is trained with coarser visual latent tokens incorporated with an uncertainty sampling module. Experimental results demonstrate that ARLON significantly outperforms the baseline OpenSora-V1.2 on eight out of eleven metrics selected from VBench, with notable improvements in dynamic degree and aesthetic quality, while delivering competitive results on the remaining three and simultaneously accelerating the generation process. In addition, ARLON achieves state-of-the-art performance in long video generation. Detailed analyses of the improvements in inference efficiency are presented, alongside a practical application that demonstrates the generation of long videos using progressive text prompts. See demos of ARLON at http://aka.ms/arlon.
comment: Accepted at ICLR2025
♻ ☆ CyclePose -- Leveraging Cycle-Consistency for Annotation-Free Nuclei Segmentation in Fluorescence Microscopy MICCAI 2025
In recent years, numerous neural network architectures specifically designed for the instance segmentation of nuclei in microscopic images have been released. These models embed nuclei-specific priors to outperform generic architectures like U-Nets; however, they require large annotated datasets, which are often not available. Generative models (GANs, diffusion models) have been used to compensate for this by synthesizing training data. These two-stage approaches are computationally expensive, as first a generative model and then a segmentation model has to be trained. We propose CyclePose, a hybrid framework integrating synthetic data generation and segmentation training. CyclePose builds on a CycleGAN architecture, which allows unpaired translation between microscopy images and segmentation masks. We embed a segmentation model into CycleGAN and leverage a cycle consistency loss for self-supervision. Without annotated data, CyclePose outperforms other weakly or unsupervised methods on two public datasets. Code is available at https://github.com/jonasutz/CyclePose
comment: under review for MICCAI 2025
♻ ☆ InstanceGaussian: Appearance-Semantic Joint Gaussian Representation for 3D Instance-Level Perception CVPR 2025
3D scene understanding has become an essential area of research with applications in autonomous driving, robotics, and augmented reality. Recently, 3D Gaussian Splatting (3DGS) has emerged as a powerful approach, combining explicit modeling with neural adaptability to provide efficient and detailed scene representations. However, three major challenges remain in leveraging 3DGS for scene understanding: 1) an imbalance between appearance and semantics, where dense Gaussian usage for fine-grained texture modeling does not align with the minimal requirements for semantic attributes; 2) inconsistencies between appearance and semantics, as purely appearance-based Gaussians often misrepresent object boundaries; and 3) reliance on top-down instance segmentation methods, which struggle with uneven category distributions, leading to over- or under-segmentation. In this work, we propose InstanceGaussian, a method that jointly learns appearance and semantic features while adaptively aggregating instances. Our contributions include: i) a novel Semantic-Scaffold-GS representation balancing appearance and semantics to improve feature representations and boundary delineation; ii) a progressive appearance-semantic joint training strategy to enhance stability and segmentation accuracy; and iii) a bottom-up, category-agnostic instance aggregation approach that addresses segmentation challenges through farthest point sampling and connected component analysis. Our approach achieves state-of-the-art performance in category-agnostic, open-vocabulary 3D point-level segmentation, highlighting the effectiveness of the proposed representation and training strategies. Project page: https://lhj-git.github.io/InstanceGaussian/
comment: 14 pages, accepted by CVPR 2025 as poster
♻ ☆ Privacy-Preserving CNN Training with Transfer Learning: Multiclass Logistic Regression
In this paper, we present a practical solution to implement privacy-preserving CNN training based on mere Homomorphic Encryption (HE) technique. To our best knowledge, this is the first attempt successfully to crack this nut and no work ever before has achieved this goal. Several techniques combine to accomplish the task:: (1) with transfer learning, privacy-preserving CNN training can be reduced to homomorphic neural network training, or even multiclass logistic regression (MLR) training; (2) via a faster gradient variant called $\texttt{Quadratic Gradient}$, an enhanced gradient method for MLR with a state-of-the-art performance in convergence speed is applied in this work to achieve high performance; (3) we employ the thought of transformation in mathematics to transform approximating Softmax function in the encryption domain to the approximation of the Sigmoid function. A new type of loss function termed $\texttt{Squared Likelihood Error}$ has been developed alongside to align with this change.; and (4) we use a simple but flexible matrix-encoding method named $\texttt{Volley Revolver}$ to manage the data flow in the ciphertexts, which is the key factor to complete the whole homomorphic CNN training. The complete, runnable C++ code to implement our work can be found at: \href{https://github.com/petitioner/HE.CNNtraining}{$\texttt{https://github.com/petitioner/HE.CNNtraining}$}. We select $\texttt{REGNET\_X\_400MF}$ as our pre-trained model for transfer learning. We use the first 128 MNIST training images as training data and the whole MNIST testing dataset as the testing data. The client only needs to upload 6 ciphertexts to the cloud and it takes $\sim 21$ mins to perform 2 iterations on a cloud with 64 vCPUs, resulting in a precision of $21.49\%$.
comment: In this work, we initiated to implement privacy-persevering CNN training based on mere HE techniques by presenting a faster HE-friendly algorithm
♻ ☆ Contextual AD Narration with Interleaved Multimodal Sequence CVPR25
The Audio Description (AD) task aims to generate descriptions of visual elements for visually impaired individuals to help them access long-form video content, like movies. With video feature, text, character bank and context information as inputs, the generated ADs are able to correspond to the characters by name and provide reasonable, contextual descriptions to help audience understand the storyline of movie. To achieve this goal, we propose to leverage pre-trained foundation models through a simple and unified framework to generate ADs with interleaved multimodal sequence as input, termed as Uni-AD. To enhance the alignment of features across various modalities with finer granularity, we introduce a simple and lightweight module that maps video features into the textual feature space. Moreover, we also propose a character-refinement module to provide more precise information by identifying the main characters who play more significant roles in the video context. With these unique designs, we further incorporate contextual information and a contrastive loss into our architecture to generate smoother and more contextually appropriate ADs. Experiments on multiple AD datasets show that Uni-AD performs well on AD generation, which demonstrates the effectiveness of our approach. Our code is available at: https://github.com/ant-research/UniAD.
comment: Accepted by CVPR25
♻ ☆ Feature Calibration enhanced Parameter Synthesis for CLIP-based Class-incremental Learning
Class-Incremental Learning (CIL) enables models to continuously learn new class knowledge while retaining previous classes, facilitating adaptation and evolution in dynamic, real-world environments. Traditional CIL methods primarily rely on visual features, which limits their effectiveness in complex, multimodal scenarios. In contrast, VLMs show promising potential for enhancing CIL by leveraging pre-trained knowledge and integrating multi-modal semantic cues such as text and vision. However, existing approaches struggle to mitigate catastrophic forgetting while preserving the generalization strengths of VLMs across diverse modalities. To address these challenges, we propose a Feature Calibration Enhanced Parameter Synthesis (FCPS) framework. Specifically, FCPS introduces a dynamic parameter adjustment mechanism that iteratively calibrates the contribution of original visual features to the final class decision, thus preserving the model's intrinsic generalization capability across modalities. Simultaneously, parameter integration enables effective knowledge transfer, maintaining a balance between acquiring new class representations and preserving old knowledge. Experimental results on popular benchmarks (e.g., CIFAR100 and ImageNet100) validate the superiority of the proposed method.
♻ ☆ What Is a Good Caption? A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness
Visual captioning benchmarks have become outdated with the emergence of modern multimodal large language models (MLLMs), as the brief ground-truth sentences and traditional metrics fail to assess detailed captions effectively. While recent benchmarks attempt to address this by focusing on keyword extraction or object-centric evaluation, they remain limited to vague-view or object-view analyses and incomplete visual element coverage. In this paper, we introduce CAPability, a comprehensive multi-view benchmark for evaluating visual captioning across 12 dimensions spanning six critical views. We curate nearly 11K human-annotated images and videos with visual element annotations to evaluate the generated captions. CAPability stably assesses both the correctness and thoroughness of captions using F1-score. By converting annotations to QA pairs, we further introduce a heuristic metric, \textit{know but cannot tell} ($K\bar{T}$), indicating a significant performance gap between QA and caption capabilities. Our work provides the first holistic analysis of MLLMs' captioning abilities, as we identify their strengths and weaknesses across various dimensions, guiding future research to enhance specific aspects of capabilities.
♻ ☆ Weakly Supervised Lymph Nodes Segmentation Based on Partial Instance Annotations with Pre-trained Dual-branch Network and Pseudo Label Learning
Assessing the presence of potentially malignant lymph nodes aids in estimating cancer progression, and identifying surrounding benign lymph nodes can assist in determining potential metastatic pathways for cancer. For quantitative analysis, automatic segmentation of lymph nodes is crucial. However, due to the labor-intensive and time-consuming manual annotation process required for a large number of lymph nodes, it is more practical to annotate only a subset of the lymph node instances to reduce annotation costs. In this study, we propose a pre-trained Dual-Branch network with Dynamically Mixed Pseudo label (DBDMP) to learn from partial instance annotations for lymph nodes segmentation. To obtain reliable pseudo labels for lymph nodes that are not annotated, we employ a dual-decoder network to generate different outputs that are then dynamically mixed. We integrate the original weak partial annotations with the mixed pseudo labels to supervise the network. To further leverage the extensive amount of unannotated voxels, we apply a self-supervised pre-training strategy to enhance the model's feature extraction capability. Experiments on the mediastinal Lymph Node Quantification (LNQ) dataset demonstrate that our method, compared to directly learning from partial instance annotations, significantly improves the Dice Similarity Coefficient (DSC) from 11.04% to 54.10% and reduces the Average Symmetric Surface Distance (ASSD) from 20.83 $mm$ to 8.72 $mm$. The code is available at https://github.com/WltyBY/LNQ2023_training_code.git
comment: Accepted for publication at the Journal of Machine Learning for Biomedical Imaging (MELBA) https://melba-journal.org/2024:013
♻ ☆ RORem: Training a Robust Object Remover with Human-in-the-Loop
Despite the significant advancements, existing object removal methods struggle with incomplete removal, incorrect content synthesis and blurry synthesized regions, resulting in low success rates. Such issues are mainly caused by the lack of high-quality paired training data, as well as the self-supervised training paradigm adopted in these methods, which forces the model to in-paint the masked regions, leading to ambiguity between synthesizing the masked objects and restoring the background. To address these issues, we propose a semi-supervised learning strategy with human-in-the-loop to create high-quality paired training data, aiming to train a Robust Object Remover (RORem). We first collect 60K training pairs from open-source datasets to train an initial object removal model for generating removal samples, and then utilize human feedback to select a set of high-quality object removal pairs, with which we train a discriminator to automate the following training data generation process. By iterating this process for several rounds, we finally obtain a substantial object removal dataset with over 200K pairs. Fine-tuning the pre-trained stable diffusion model with this dataset, we obtain our RORem, which demonstrates state-of-the-art object removal performance in terms of both reliability and image quality. Particularly, RORem improves the object removal success rate over previous methods by more than 18\%. The dataset, source code and trained model are available at https://github.com/leeruibin/RORem.
♻ ☆ Prototypical Calibrating Ambiguous Samples for Micro-Action Recognition AAAI 2025
Micro-Action Recognition (MAR) has gained increasing attention due to its crucial role as a form of non-verbal communication in social interactions, with promising potential for applications in human communication and emotion analysis. However, current approaches often overlook the inherent ambiguity in micro-actions, which arises from the wide category range and subtle visual differences between categories. This oversight hampers the accuracy of micro-action recognition. In this paper, we propose a novel Prototypical Calibrating Ambiguous Network (PCAN) to unleash and mitigate the ambiguity of MAR. Firstly, we employ a hierarchical action-tree to identify the ambiguous sample, categorizing them into distinct sets of ambiguous samples of false negatives and false positives, considering both body- and action-level categories. Secondly, we implement an ambiguous contrastive refinement module to calibrate these ambiguous samples by regulating the distance between ambiguous samples and their corresponding prototypes. This calibration process aims to pull false negative (FN) samples closer to their respective prototypes and push false positive (FP) samples apart from their affiliated prototypes. In addition, we propose a new prototypical diversity amplification loss to strengthen the model's capacity by amplifying the differences between different prototypes. Finally, we propose a prototype-guided rectification to rectify prediction by incorporating the representability of prototypes. Extensive experiments conducted on the benchmark dataset demonstrate the superior performance of our method compared to existing approaches. The code is available at https://github.com/kunli-cs/PCAN.
comment: Fix typos; Accepted by AAAI 2025
♻ ☆ WildLive: Near Real-time Visual Wildlife Tracking onboard UAVs
Live tracking of wildlife via high-resolution video processing directly onboard drones is widely unexplored and most existing solutions rely on streaming video to ground stations to support navigation. Yet, both autonomous animal-reactive flight control beyond visual line of sight and/or mission-specific individual and behaviour recognition tasks rely to some degree on this capability. In response, we introduce WildLive -- a near real-time animal detection and tracking framework for high-resolution imagery running directly onboard uncrewed aerial vehicles (UAVs). The system performs multi-animal detection and tracking at 17fps+ for HD and 7fps+ on 4K video streams suitable for operation during higher altitude flights to minimise animal disturbance. Our system is optimised for Jetson Orin AGX onboard hardware. It integrates the efficiency of sparse optical flow tracking and mission-specific sampling with device-optimised and proven YOLO-driven object detection and segmentation techniques. Essentially, computational resource is focused onto spatio-temporal regions of high uncertainty to significantly improve UAV processing speeds without domain-specific loss of accuracy. Alongside, we introduce our WildLive dataset, which comprises 200k+ annotated animal instances across 19k+ frames from 4K UAV videos collected at the Ol Pejeta Conservancy in Kenya. All frames contain ground truth bounding boxes, segmentation masks, as well as individual tracklets and tracking point trajectories. We compare our system against current object tracking approaches including OC-SORT, ByteTrack, and SORT. Our materials are available at: https://dat-nguyenvn.github.io/WildLive/
♻ ☆ VR-Splatting: Foveated Radiance Field Rendering via 3D Gaussian Splatting and Neural Points
Recent advances in novel view synthesis have demonstrated impressive results in fast photorealistic scene rendering through differentiable point rendering, either via Gaussian Splatting (3DGS) [Kerbl and Kopanas et al. 2023] or neural point rendering [Aliev et al. 2020]. Unfortunately, these directions require either a large number of small Gaussians or expensive per-pixel post-processing for reconstructing fine details, which negatively impacts rendering performance. To meet the high performance demands of virtual reality (VR) systems, primitive or pixel counts therefore must be kept low, affecting visual quality. In this paper, we propose a novel hybrid approach based on foveated rendering as a promising solution that combines the strengths of both point rendering directions regarding performance sweet spots. Analyzing the compatibility with the human visual system, we find that using a low-detailed, few primitive smooth Gaussian representation for the periphery is cheap to compute and meets the perceptual demands of peripheral vision. For the fovea only, we use neural points with a convolutional neural network for the small pixel footprint, which provides sharp, detailed output within the rendering budget. This combination also allows for synergistic method accelerations with point occlusion culling and reducing the demands on the neural network. Our evaluation confirms that our approach increases sharpness and details compared to a standard VR-ready 3DGS configuration, and participants of a user study overwhelmingly preferred our method. Our system meets the necessary performance requirements for real-time VR interactions, ultimately enhancing the user's immersive experience. The project page can be found at: https://lfranke.github.io/vr_splatting
♻ ☆ TAMP: Token-Adaptive Layerwise Pruning in Multimodal Large Language Models
Multimodal Large Language Models (MLLMs) have shown remarkable versatility in understanding diverse multimodal data and tasks. However, these capabilities come with an increased model scale. While post-training pruning reduces model size in unimodal models, its application to MLLMs often yields limited success. Our analysis discovers that conventional methods fail to account for the unique token attributes across layers and modalities inherent to MLLMs. Inspired by this observation, we propose TAMP, a simple yet effective pruning framework tailored for MLLMs, featuring two key components: (1) Diversity-Aware Sparsity, which adjusts sparsity ratio per layer based on diversities among multimodal output tokens, preserving more parameters in high-diversity layers; and (2) Adaptive Multimodal Input Activation, which identifies representative multimodal input tokens using attention scores to guide unstructured weight pruning. We validate our method on two state-of-the-art MLLMs: LLaVA-NeXT, designed for vision-language tasks, and VideoLLaMA2, capable of processing audio, visual, and language modalities. Empirical experiments across various multimodal evaluation benchmarks demonstrate that each component of our approach substantially outperforms existing pruning techniques.
comment: Preprint
♻ ☆ MROSS: Multi-Round Region-based Optimization for Scene Sketching
Scene sketching is to convert a scene into a simplified, abstract representation that captures the essential elements and composition of the original scene. It requires a semantic understanding of the scene and consideration of different regions within the scene. Since scenes often contain diverse visual information across various regions, such as foreground objects, background elements, and spatial divisions, dealing with these different regions poses unique difficulties. In this paper, we define a sketch as some sets of B\'ezier curves because of their smooth and versatile characteristics. We optimize different regions of input scene in multiple rounds. In each optimization round, the strokes sampled from the next region can seamlessly be integrated into the sketch generated in the previous optimization round. We propose an additional stroke initialization method to ensure the integrity of the scene and the convergence of optimization. A novel CLIP-based Semantic Loss and a VGG-based Feature Loss are utilized to guide our multi-round optimization. Extensive experimental results on the quality and quantity of the generated sketches confirm the effectiveness of our method.
comment: 6 pages, 8 figures
♻ ☆ Differentially Private 2D Human Pose Estimation
Human pose estimation (HPE) has become essential in numerous applications including healthcare, activity recognition, and human-computer interaction. However, the privacy implications of processing sensitive visual data present significant deployment barriers in critical domains. While traditional anonymization techniques offer limited protection and often compromise data utility for broader motion analysis, Differential Privacy (DP) provides formal privacy guarantees but typically degrades model performance when applied naively. In this work, we present the first differentially private 2D human pose estimation (2D-HPE) by applying Differentially Private Stochastic Gradient Descent (DP-SGD) to this task. To effectively balance privacy with performance, we adopt Projected DP-SGD (PDP-SGD), which projects the noisy gradients to a low-dimensional subspace. Additionally, we adapt TinyViT, a compact and efficient vision transformer for coordinate classification in HPE, providing a lightweight yet powerful backbone that enhances privacy-preserving deployment feasibility on resource-limited devices. Our approach is particularly valuable for multimedia interpretation tasks, enabling privacy-safe analysis and understanding of human motion across diverse visual media while preserving the semantic meaning required for downstream applications. Comprehensive experiments on the MPII Human Pose Dataset demonstrate significant performance enhancement with PDP-SGD achieving 78.48% PCKh@0.5 at a strict privacy budget ($\epsilon=0.2$), compared to 63.85% for standard DP-SGD. This work lays foundation for privacy-preserving human pose estimation in real-world, sensitive applications.
♻ ☆ MIMIR: Masked Image Modeling for Mutual Information-based Adversarial Robustness
Vision Transformers (ViTs) have emerged as a fundamental architecture and serve as the backbone of modern vision-language models. Despite their impressive performance, ViTs exhibit notable vulnerability to evasion attacks, necessitating the development of specialized Adversarial Training (AT) strategies tailored to their unique architecture. While a direct solution might involve applying existing AT methods to ViTs, our analysis reveals significant incompatibilities, particularly with state-of-the-art (SOTA) approaches such as Generalist (CVPR 2023) and DBAT (USENIX Security 2024). This paper presents a systematic investigation of adversarial robustness in ViTs and provides a novel theoretical Mutual Information (MI) analysis in its autoencoder-based self-supervised pre-training. Specifically, we show that MI between the adversarial example and its latent representation in ViT-based autoencoders should be constrained via derived MI bounds. Building on this insight, we propose a self-supervised AT method, MIMIR, that employs an MI penalty to facilitate adversarial pre-training by masked image modeling with autoencoders. Extensive experiments on CIFAR-10, Tiny-ImageNet, and ImageNet-1K show that MIMIR can consistently provide improved natural and robust accuracy, where MIMIR outperforms SOTA AT results on ImageNet-1K. Notably, MIMIR demonstrates superior robustness against unforeseen attacks and common corruption data and can also withstand adaptive attacks where the adversary possesses full knowledge of the defense mechanism.
♻ ☆ Causal Graphical Models for Vision-Language Compositional Understanding ICLR 2025
Recent work has empirically shown that Vision-Language Models (VLMs) struggle to fully understand the compositional properties of the human language, usually modeling an image caption as a "bag of words". As a result, they perform poorly on compositional tasks, which require a deeper understanding of the different entities of a sentence (subject, verb, etc.) jointly with their mutual relationships in order to be solved. In this paper, we model the dependency relations among textual and visual tokens using a Causal Graphical Model (CGM), built using a dependency parser, and we train a decoder conditioned by the VLM visual encoder. Differently from standard autoregressive or parallel predictions, our decoder's generative process is partially-ordered following the CGM structure. This structure encourages the decoder to learn only the main causal dependencies in a sentence discarding spurious correlations. Using extensive experiments on five compositional benchmarks, we show that our method significantly outperforms all the state-of-the-art compositional approaches by a large margin, and it also improves over methods trained using much larger datasets.
comment: Accepted at ICLR 2025
♻ ☆ MedMerge: Merging Models for Effective Transfer Learning to Medical Imaging Tasks
Transfer learning has become a powerful tool to initialize deep learning models to achieve faster convergence and higher performance. This is especially useful in the medical imaging analysis domain, where data scarcity limits possible performance gains for deep learning models. Some advancements have been made in boosting the transfer learning performance gain by merging models starting from the same initialization. However, in the medical imaging analysis domain, there is an opportunity to merge models starting from different initializations, thus combining the features learned from different tasks. In this work, we propose MedMerge, a method whereby the weights of different models can be merged, and their features can be effectively utilized to boost performance on a new task. With MedMerge, we learn kernel-level weights that can later be used to merge the models into a single model, even when starting from different initializations. Testing on various medical imaging analysis tasks, we show that our merged model can achieve significant performance gains, with up to 7% improvement on the F1 score. The code implementation of this work is available at github.com/BioMedIA-MBZUAI/MedMerge.
♻ ☆ Comparing Next-Day Wildfire Predictability of MODIS and VIIRS Satellite Data
Multiple studies have performed next-day fire prediction using satellite imagery. Two main satellites are used to detect wildfires: MODIS and VIIRS. Both satellites provide fire mask products, called MOD14 and VNP14, respectively. Studies have used one or the other, but there has been no comparison between them to determine which might be more suitable for next-day fire prediction. In this paper, we first evaluate how well VIIRS and MODIS data can be used to forecast wildfire spread one day ahead. We find that the model using VIIRS as input and VNP14 as target achieves the best results. Interestingly, the model using MODIS as input and VNP14 as target performs significantly better than using VNP14 as input and MOD14 as target. Next, we discuss why MOD14 might be harder to use for predicting next-day fires. We find that the MOD14 fire mask is highly stochastic and does not correlate with reasonable fire spread patterns. This is detrimental for machine learning tasks, as the model learns irrational patterns. Therefore, we conclude that MOD14 is unsuitable for next-day fire prediction and that VNP14 is a much better option. However, using MODIS input and VNP14 as target, we achieve a significant improvement in predictability. This indicates that an improved fire detection model is possible for MODIS. The full code and dataset is available online: https://github.com/justuskarlsson/wildfire-mod14-vnp14
♻ ☆ Trade-offs in Privacy-Preserving Eye Tracking through Iris Obfuscation: A Benchmarking Study SP 2025
Recent developments in hardware, computer graphics, and AI may soon enable AR/VR head-mounted displays (HMDs) to become everyday devices like smartphones and tablets. Eye trackers within HMDs provide a special opportunity for such setups as it is possible to facilitate gaze-based research and interaction. However, estimating users' gaze information often requires raw eye images and videos that contain iris textures, which are considered a gold standard biometric for user authentication, and this raises privacy concerns. Previous research in the eye-tracking community focused on obfuscating iris textures while keeping utility tasks such as gaze estimation accurate. Despite these attempts, there is no comprehensive benchmark that evaluates state-of-the-art approaches. Considering all, in this paper, we benchmark blurring, noising, downsampling, rubber sheet model, and iris style transfer to obfuscate user identity, and compare their impact on image quality, privacy, utility, and risk of imposter attack on two datasets. We use eye segmentation and gaze estimation as utility tasks, and reduction in iris recognition accuracy as a measure of privacy protection, and false acceptance rate to estimate risk of attack. Our experiments show that canonical image processing methods like blurring and noising cause a marginal impact on deep learning-based tasks. While downsampling, rubber sheet model, and iris style transfer are effective in hiding user identifiers, iris style transfer, with higher computation cost, outperforms others in both utility tasks, and is more resilient against spoof attacks. Our analyses indicate that there is no universal optimal approach to balance privacy, utility, and computation burden. Therefore, we recommend practitioners consider the strengths and weaknesses of each approach, and possible combinations of those to reach an optimal privacy-utility trade-off.
comment: The 25th International Conference on Digital Signal Processing (DSP 2025)
♻ ☆ Fine-Tuning Florence2 for Enhanced Object Detection in Un-constructed Environments: Vision-Language Model Approach
Vision-Language Models (VLMs) have emerged as powerful tools in artificial intelli-gence, capable of integrating textual and visual data for a unified understanding of complex scenes. While models such as Florence2, built on transformer architectures, have shown promise across general tasks, their performance in object detection within unstructured or cluttered environments remains underexplored. In this study, we fi-ne-tuned the Florence2 model for object detection tasks in non-constructed, complex environments. A comprehensive experimental framework was established involving multiple hardware configurations (NVIDIA T4, L4, and A100 GPUs), optimizers (AdamW, SGD), and varied hyperparameters including learning rates and LoRA (Low-Rank Adaptation) setups. Model training and evaluation were conducted on challenging datasets representative of real-world, disordered settings. The optimized Florence2 models exhibited significant improvements in object detection accuracy, with Mean Average Precision (mAP) metrics approaching or matching those of estab-lished models such as YOLOv8, YOLOv9, and YOLOv10. The integration of LoRA and careful fine-tuning of transformer layers contributed notably to these gains. Our find-ings highlight the adaptability of transformer-based VLMs like Florence2 for do-main-specific tasks, particularly in visually complex environments. The study under-scores the potential of fine-tuned VLMs to rival traditional convolution-based detec-tors, offering a flexible and scalable approach for advanced vision applications in re-al-world, unstructured settings.
comment: 22 pages, 13 Figures, 6 Tables
♻ ☆ OctGPT: Octree-based Multiscale Autoregressive Models for 3D Shape Generation SIGGRAPH 2025
Autoregressive models have achieved remarkable success across various domains, yet their performance in 3D shape generation lags significantly behind that of diffusion models. In this paper, we introduce OctGPT, a novel multiscale autoregressive model for 3D shape generation that dramatically improves the efficiency and performance of prior 3D autoregressive approaches, while rivaling or surpassing state-of-the-art diffusion models. Our method employs a serialized octree representation to efficiently capture the hierarchical and spatial structures of 3D shapes. Coarse geometry is encoded via octree structures, while fine-grained details are represented by binary tokens generated using a vector quantized variational autoencoder (VQVAE), transforming 3D shapes into compact multiscale binary sequences suitable for autoregressive prediction. To address the computational challenges of handling long sequences, we incorporate octree-based transformers enhanced with 3D rotary positional encodings, scale-specific embeddings, and token-parallel generation schemes. These innovations reduce training time by 13 folds and generation time by 69 folds, enabling the efficient training of high-resolution 3D shapes, e.g.,$1024^3$, on just four NVIDIA 4090 GPUs only within days. OctGPT showcases exceptional versatility across various tasks, including text-, sketch-, and image-conditioned generation, as well as scene-level synthesis involving multiple objects. Extensive experiments demonstrate that OctGPT accelerates convergence and improves generation quality over prior autoregressive methods, offering a new paradigm for high-quality, scalable 3D content creation. Our code and trained models are available at https://github.com/octree-nn/octgpt.
comment: SIGGRAPH 2025
♻ ☆ BlockGaussian: Efficient Large-Scale Scene Novel View Synthesis via Adaptive Block-Based Gaussian Splatting
The recent advancements in 3D Gaussian Splatting (3DGS) have demonstrated remarkable potential in novel view synthesis tasks. The divide-and-conquer paradigm has enabled large-scale scene reconstruction, but significant challenges remain in scene partitioning, optimization, and merging processes. This paper introduces BlockGaussian, a novel framework incorporating a content-aware scene partition strategy and visibility-aware block optimization to achieve efficient and high-quality large-scale scene reconstruction. Specifically, our approach considers the content-complexity variation across different regions and balances computational load during scene partitioning, enabling efficient scene reconstruction. To tackle the supervision mismatch issue during independent block optimization, we introduce auxiliary points during individual block optimization to align the ground-truth supervision, which enhances the reconstruction quality. Furthermore, we propose a pseudo-view geometry constraint that effectively mitigates rendering degradation caused by airspace floaters during block merging. Extensive experiments on large-scale scenes demonstrate that our approach achieves state-of-the-art performance in both reconstruction efficiency and rendering quality, with a 5x speedup in optimization and an average PSNR improvement of 1.21 dB on multiple benchmarks. Notably, BlockGaussian significantly reduces computational requirements, enabling large-scale scene reconstruction on a single 24GB VRAM device. The project page is available at https://github.com/SunshineWYC/BlockGaussian
comment: https://github.com/SunshineWYC/BlockGaussian
♻ ☆ Reasoning in visual navigation of end-to-end trained agents: a dynamical systems approach
Progress in Embodied AI has made it possible for end-to-end-trained agents to navigate in photo-realistic environments with high-level reasoning and zero-shot or language-conditioned behavior, but benchmarks are still dominated by simulation. In this work, we focus on the fine-grained behavior of fast-moving real robots and present a large-scale experimental study involving \numepisodes{} navigation episodes in a real environment with a physical robot, where we analyze the type of reasoning emerging from end-to-end training. In particular, we study the presence of realistic dynamics which the agent learned for open-loop forecasting, and their interplay with sensing. We analyze the way the agent uses latent memory to hold elements of the scene structure and information gathered during exploration. We probe the planning capabilities of the agent, and find in its memory evidence for somewhat precise plans over a limited horizon. Furthermore, we show in a post-hoc analysis that the value function learned by the agent relates to long-term planning. Put together, our experiments paint a new picture on how using tools from computer vision and sequential decision making have led to new capabilities in robotics and control. An interactive tool is available at europe.naverlabs.com/research/publications/reasoning-in-visual-navigation-of-end-to-end-trained-agents.
♻ ☆ Text-Driven 3D Lidar Place Recognition for Autonomous Driving
Environment description-based localization in large-scale point cloud maps constructed through remote sensing is critically significant for the advancement of large-scale autonomous systems, such as delivery robots operating in the last mile. However, current approaches encounter challenges due to the inability of point cloud encoders to effectively capture local details and long-range spatial relationships, as well as a significant modality gap between text and point cloud representations. To address these challenges, we present Des4Pos, a novel two-stage text-driven remote sensing localization framework. In the coarse stage, the point-cloud encoder utilizes the Multi-scale Fusion Attention Mechanism (MFAM) to enhance local geometric features, followed by a bidirectional Long Short-Term Memory (LSTM) module to strengthen global spatial relationships. Concurrently, the Stepped Text Encoder (STE) integrates cross-modal prior knowledge from CLIP [1] and aligns text and point-cloud features using this prior knowledge, effectively bridging modality discrepancies. In the fine stage, we introduce a Cascaded Residual Attention (CRA) module to fuse cross-modal features and predict relative localization offsets, thereby achieving greater localization precision. Experiments on the KITTI360Pose test set demonstrate that Des4Pos achieves state-of-the-art performance in text-to-point-cloud place recognition. Specifically, it attains a top-1 accuracy of 40% and a top-10 accuracy of 77% under a 5-meter radius threshold, surpassing the best existing methods by 7% and 7%, respectively.
comment: 13 pages
♻ ☆ Real-world Image Dehazing with Coherence-based Pseudo Labeling and Cooperative Unfolding Network NeurIPS 2024
Real-world Image Dehazing (RID) aims to alleviate haze-induced degradation in real-world settings. This task remains challenging due to the complexities in accurately modeling real haze distributions and the scarcity of paired real-world data. To address these challenges, we first introduce a cooperative unfolding network that jointly models atmospheric scattering and image scenes, effectively integrating physical knowledge into deep networks to restore haze-contaminated details. Additionally, we propose the first RID-oriented iterative mean-teacher framework, termed the Coherence-based Label Generator, to generate high-quality pseudo labels for network training. Specifically, we provide an optimal label pool to store the best pseudo-labels during network training, leveraging both global and local coherence to select high-quality candidates and assign weights to prioritize haze-free regions. We verify the effectiveness of our method, with experiments demonstrating that it achieves state-of-the-art performance on RID tasks. Code will be available at https://github.com/cnyvfang/CORUN-Colabator.
comment: Accepted at NeurIPS 2024 as a Spotlight Paper
♻ ☆ LDGNet: A Lightweight Difference Guiding Network for Remote Sensing Change Detection
With the rapid advancement of deep learning, the field of change detection (CD) in remote sensing imagery has achieved remarkable progress. Existing change detection methods primarily focus on achieving higher accuracy with increased computational costs and parameter sizes, leaving development of lightweight methods for rapid real-world processing an underexplored challenge. To address this challenge, we propose a Lightweight Difference Guiding Network (LDGNet), leveraging absolute difference image to guide optical remote sensing change detection. First, to enhance the feature representation capability of the lightweight backbone network, we propose the Difference Guiding Module (DGM), which leverages multi-scale features extracted from the absolute difference image to progressively influence the original image encoder at each layer, thereby reinforcing feature extraction. Second, we propose the Difference-Aware Dynamic Fusion (DADF) module with Visual State Space Model (VSSM) for lightweight long-range dependency modeling. The module first uses feature absolute differences to guide VSSM's global contextual modeling of change regions, then employs difference attention to dynamically fuse these long-range features with feature differences, enhancing change semantics while suppressing noise and background. Extensive experiments on multiple datasets demonstrate that our method achieves comparable or superior performance to current state-of-the-art (SOTA) methods requiring several times more computation, while maintaining only 3.43M parameters and 1.12G FLOPs.
♻ ☆ Reinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters AAAI
We present a Reinforcement Learning Platform for Adversarial Black-box untargeted and targeted attacks, RLAB, that allows users to select from various distortion filters to create adversarial examples. The platform uses a Reinforcement Learning agent to add minimum distortion to input images while still causing misclassification by the target model. The agent uses a novel dual-action method to explore the input image at each step to identify sensitive regions for adding distortions while removing noises that have less impact on the target model. This dual action leads to faster and more efficient convergence of the attack. The platform can also be used to measure the robustness of image classification models against specific distortion types. Also, retraining the model with adversarial samples significantly improved robustness when evaluated on benchmark datasets. The proposed platform outperforms state-of-the-art methods in terms of the average number of queries required to cause misclassification. This advances trustworthiness with a positive social impact.
comment: Accepted at the 2025 AAAI Conference on Artificial Intelligence Proceedings
♻ ☆ LPViT: Low-Power Semi-structured Pruning for Vision Transformers
Vision transformers have emerged as a promising alternative to convolutional neural networks for various image analysis tasks, offering comparable or superior performance. However, one significant drawback of ViTs is their resource-intensive nature, leading to increased memory footprint, computation complexity, and power consumption. To democratize this high-performance technology and make it more environmentally friendly, it is essential to compress ViT models, reducing their resource requirements while maintaining high performance. In this paper, we introduce a new block-structured pruning to address the resource-intensive issue for ViTs, offering a balanced trade-off between accuracy and hardware acceleration. Unlike unstructured pruning or channel-wise structured pruning, block pruning leverages the block-wise structure of linear layers, resulting in more efficient matrix multiplications. To optimize this pruning scheme, our paper proposes a novel hardware-aware learning objective that simultaneously maximizes speedup and minimizes power consumption during inference, tailored to the block sparsity structure. This objective eliminates the need for empirical look-up tables and focuses solely on reducing parametrized layer connections. Moreover, our paper provides a lightweight algorithm to achieve post-training pruning for ViTs, utilizing second-order Taylor approximation and empirical optimization to solve the proposed hardware-aware objective. Extensive experiments on ImageNet are conducted across various ViT architectures, including DeiT-B and DeiT-S, demonstrating competitive performance with other pruning methods and achieving a remarkable balance between accuracy preservation and power savings. Especially, we achieve 3.93x speedup on dedicated hardware and GPUs respectively for DeiT-B, and a power reduction by 1.4x on GPUs. Code released to https://github.com/Akimoto-Cris/LPViT.
♻ ☆ Light-YOLOv8-Flame: A Lightweight High-Performance Flame Detection Algorithm
Fire detection algorithms, particularly those based on computer vision, encounter significant challenges such as high computational costs and delayed response times, which hinder their application in real-time systems. To address these limitations, this paper introduces Light-YOLOv8-Flame, a lightweight flame detection algorithm specifically designed for fast and efficient real-time deployment. The proposed model enhances the YOLOv8 architecture through the substitution of the original C2f module with the FasterNet Block module. This new block combines Partial Convolution (PConv) and Convolution (Conv) layers, reducing both computational complexity and model size. A dataset comprising 7,431 images, representing both flame and non-flame scenarios, was collected and augmented for training purposes. Experimental findings indicate that the modified YOLOv8 model achieves a 0.78% gain in mean average precision (mAP) and a 2.05% boost in recall, while reducing the parameter count by 25.34%, with only a marginal decrease in precision by 0.82%. These findings highlight that Light-YOLOv8-Flame offers enhanced detection performance and speed, making it well-suited for real-time fire detection on resource-constrained devices.
comment: 12 pages, 19 figures, 6 tables. Submitted to Engineering Letters
♻ ☆ LLaVA-ReID: Selective Multi-image Questioner for Interactive Person Re-Identification
Traditional text-based person ReID assumes that person descriptions from witnesses are complete and provided at once. However, in real-world scenarios, such descriptions are often partial or vague. To address this limitation, we introduce a new task called interactive person re-identification (Inter-ReID). Inter-ReID is a dialogue-based retrieval task that iteratively refines initial descriptions through ongoing interactions with the witnesses. To facilitate the study of this new task, we construct a dialogue dataset that incorporates multiple types of questions by decomposing fine-grained attributes of individuals. We further propose LLaVA-ReID, a question model that generates targeted questions based on visual and textual contexts to elicit additional details about the target person. Leveraging a looking-forward strategy, we prioritize the most informative questions as supervision during training. Experimental results on both Inter-ReID and text-based ReID benchmarks demonstrate that LLaVA-ReID significantly outperforms baselines.
♻ ☆ PaMi-VDPO: Mitigating Video Hallucinations by Prompt-Aware Multi-Instance Video Preference Learning
Direct Preference Optimization (DPO) helps reduce hallucinations in Video Multimodal Large Language Models (VLLMs), but its reliance on offline preference data limits adaptability and fails to capture true video-response misalignment. We propose Video Direct Preference Optimization (VDPO), an online preference learning framework that eliminates the need for preference annotation by leveraging video augmentations to generate rejected samples while keeping responses fixed. However, selecting effective augmentations is non-trivial, as some clips may be semantically identical to the original under specific prompts, leading to false rejections and disrupting alignment. To address this, we introduce Prompt-aware Multi-instance Learning VDPO (PaMi-VDPO), which selects augmentations based on prompt context. Instead of a single rejection, we construct a candidate set of augmented clips and apply a close-to-far selection strategy, initially ensuring all clips are semantically relevant while then prioritizing the most prompt-aware distinct clip. This allows the model to better capture meaningful visual differences, mitigating hallucinations, while avoiding false rejections, and improving alignment. PaMi-VDPOseamlessly integrates into existing VLLMs without additional parameters, GPT-4/human supervision. With only 10k SFT data, it improves the base model by 5.3% on VideoHallucer, surpassing GPT-4o, while maintaining stable performance on general video benchmarks.
♻ ☆ SAT-HMR: Real-Time Multi-Person 3D Mesh Estimation via Scale-Adaptive Tokens
We propose a one-stage framework for real-time multi-person 3D human mesh estimation from a single RGB image. While current one-stage methods, which follow a DETR-style pipeline, achieve state-of-the-art (SOTA) performance with high-resolution inputs, we observe that this particularly benefits the estimation of individuals in smaller scales of the image (e.g., those far from the camera), but at the cost of significantly increased computation overhead. To address this, we introduce scale-adaptive tokens that are dynamically adjusted based on the relative scale of each individual in the image within the DETR framework. Specifically, individuals in smaller scales are processed at higher resolutions, larger ones at lower resolutions, and background regions are further distilled. These scale-adaptive tokens more efficiently encode the image features, facilitating subsequent decoding to regress the human mesh, while allowing the model to allocate computational resources more effectively and focus on more challenging cases. Experiments show that our method preserves the accuracy benefits of high-resolution processing while substantially reducing computational cost, achieving real-time inference with performance comparable to SOTA methods.
comment: 18 pages, 12 figures
♻ ☆ UniForm: A Unified Multi-Task Diffusion Transformer for Audio-Video Generation
With the rise of diffusion models, audio-video generation has been revolutionized. However, most existing methods rely on separate modules for each modality, with limited exploration of unified generative architectures. In addition, many are confined to a single task and small-scale datasets. To address these limitations, we first propose UniForm, a unified multi-task diffusion transformer that jointly generates audio and visual modalities in a shared latent space. A single diffusion process models both audio and video, capturing the inherent correlations between sound and vision. Second, we introduce task-specific noise schemes and task tokens, enabling a single model to support multiple tasks, including text-to-audio-video, audio-to-video, and video-to-audio generation. Furthermore, by leveraging large language models and a large-scale text-audio-video combined dataset, UniForm achieves greater generative diversity than prior approaches. Extensive experiments show that UniForm achieves the state-of-the-art performance across audio-video generation tasks, producing content that is both well-aligned and close to real-world data distributions. Our demos are available at https://uniform-t2av.github.io/.
comment: Our demos are available at https://uniform-t2av.github.io/
♻ ☆ MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision
We present MoGe, a powerful model for recovering 3D geometry from monocular open-domain images. Given a single image, our model directly predicts a 3D point map of the captured scene with an affine-invariant representation, which is agnostic to true global scale and shift. This new representation precludes ambiguous supervision in training and facilitate effective geometry learning. Furthermore, we propose a set of novel global and local geometry supervisions that empower the model to learn high-quality geometry. These include a robust, optimal, and efficient point cloud alignment solver for accurate global shape learning, and a multi-scale local geometry loss promoting precise local geometry supervision. We train our model on a large, mixed dataset and demonstrate its strong generalizability and high accuracy. In our comprehensive evaluation on diverse unseen datasets, our model significantly outperforms state-of-the-art methods across all tasks, including monocular estimation of 3D point map, depth map, and camera field of view. Code and models can be found on our project page.
comment: Project page: https://wangrc.site/MoGePage/
♻ ☆ WildlifeReID-10k: Wildlife re-identification dataset with 10k individual animals
This paper introduces WildlifeReID-10k, a new large-scale re-identification benchmark with more than 10k animal identities of around 33 species across more than 140k images, re-sampled from 37 existing datasets. WildlifeReID-10k covers diverse animal species and poses significant challenges for SoTA methods, ensuring fair and robust evaluation through its time-aware and similarity-aware split protocol. The latter is designed to address the common issue of training-to-test data leakage caused by visually similar images appearing in both training and test sets. The WildlifeReID-10k dataset and benchmark are publicly available on Kaggle, along with strong baselines for both closed-set and open-set evaluation, enabling fair, transparent, and standardized evaluation of not just multi-species animal re-identification models.
♻ ☆ GaussianHead: High-fidelity Head Avatars with Learnable Gaussian Derivation
Constructing vivid 3D head avatars for given subjects and realizing a series of animations on them is valuable yet challenging. This paper presents GaussianHead, which models the actional human head with anisotropic 3D Gaussians. In our framework, a motion deformation field and multi-resolution tri-plane are constructed respectively to deal with the head's dynamic geometry and complex texture. Notably, we impose an exclusive derivation scheme on each Gaussian, which generates its multiple doppelgangers through a set of learnable parameters for position transformation. With this design, we can compactly and accurately encode the appearance information of Gaussians, even those fitting the head's particular components with sophisticated structures. In addition, an inherited derivation strategy for newly added Gaussians is adopted to facilitate training acceleration. Extensive experiments show that our method can produce high-fidelity renderings, outperforming state-of-the-art approaches in reconstruction, cross-identity reenactment, and novel view synthesis tasks. Our code is available at: https://github.com/chiehwangs/gaussian-head.
comment: 15 pages, 14 figures, published to TVCG
♻ ☆ Leveraging Driver Field-of-View for Multimodal Ego-Trajectory Prediction ICLR 2025
Understanding drivers' decision-making is crucial for road safety. Although predicting the ego-vehicle's path is valuable for driver-assistance systems, existing methods mainly focus on external factors like other vehicles' motions, often neglecting the driver's attention and intent. To address this gap, we infer the ego-trajectory by integrating the driver's gaze and the surrounding scene. We introduce RouteFormer, a novel multimodal ego-trajectory prediction network combining GPS data, environmental context, and the driver's field-of-view, comprising first-person video and gaze fixations. We also present the Path Complexity Index (PCI), a new metric for trajectory complexity that enables a more nuanced evaluation of challenging scenarios. To tackle data scarcity and enhance diversity, we introduce GEM, a comprehensive dataset of urban driving scenarios enriched with synchronized driver field-of-view and gaze data. Extensive evaluations on GEM and DR(eye)VE demonstrate that RouteFormer significantly outperforms state-of-the-art methods, achieving notable improvements in prediction accuracy across diverse conditions. Ablation studies reveal that incorporating driver field-of-view data yields significantly better average displacement error, especially in challenging scenarios with high PCI scores, underscoring the importance of modeling driver attention. All data and code are available at https://meakbiyik.github.io/routeformer.
comment: Accepted to 13th International Conference on Learning Representations (ICLR 2025), 29 pages
♻ ☆ Mamba as a Bridge: Where Vision Foundation Models Meet Vision Language Models for Domain-Generalized Semantic Segmentation CVPR 2025
Vision Foundation Models (VFMs) and Vision-Language Models (VLMs) have gained traction in Domain Generalized Semantic Segmentation (DGSS) due to their strong generalization capabilities. However, existing DGSS methods often rely exclusively on either VFMs or VLMs, overlooking their complementary strengths. VFMs (e.g., DINOv2) excel at capturing fine-grained features, while VLMs (e.g., CLIP) provide robust text alignment but struggle with coarse granularity. Despite their complementary strengths, effectively integrating VFMs and VLMs with attention mechanisms is challenging, as the increased patch tokens complicate long-sequence modeling. To address this, we propose MFuser, a novel Mamba-based fusion framework that efficiently combines the strengths of VFMs and VLMs while maintaining linear scalability in sequence length. MFuser consists of two key components: MVFuser, which acts as a co-adapter to jointly fine-tune the two models by capturing both sequential and spatial dynamics; and MTEnhancer, a hybrid attention-Mamba module that refines text embeddings by incorporating image priors. Our approach achieves precise feature locality and strong text alignment without incurring significant computational overhead. Extensive experiments demonstrate that MFuser significantly outperforms state-of-the-art DGSS methods, achieving 68.20 mIoU on synthetic-to-real and 71.87 mIoU on real-to-real benchmarks. The code is available at https://github.com/devinxzhang/MFuser.
comment: Accepted to CVPR 2025 (Highlight)
Information Retrieval 21
☆ Bipartite Ranking From Multiple Labels: On Loss Versus Label Aggregation
Bipartite ranking is a fundamental supervised learning problem, with the goal of learning a ranking over instances with maximal area under the ROC curve (AUC) against a single binary target label. However, one may often observe multiple binary target labels, e.g., from distinct human annotators. How can one synthesize such labels into a single coherent ranking? In this work, we formally analyze two approaches to this problem -- loss aggregation and label aggregation -- by characterizing their Bayes-optimal solutions. Based on this, we show that while both methods can yield Pareto-optimal solutions, loss aggregation can exhibit label dictatorship: one can inadvertently (and undesirably) favor one label over others. This suggests that label aggregation can be preferable to loss aggregation, which we empirically verify.
☆ Efficient Distributed Retrieval-Augmented Generation for Enhancing Language Model Performance
Small language models (SLMs) support efficient deployments on resource-constrained edge devices, but their limited capacity compromises inference performance. Retrieval-augmented generation (RAG) is a promising solution to enhance model performance by integrating external databases, without requiring intensive on-device model retraining. However, large-scale public databases and user-specific private contextual documents are typically located on the cloud and the device separately, while existing RAG implementations are primarily centralized. To bridge this gap, we propose DRAGON, a distributed RAG framework to enhance on-device SLMs through both general and personal knowledge without the risk of leaking document privacy. Specifically, DRAGON decomposes multi-document RAG into multiple parallel token generation processes performed independently and locally on the cloud and the device, and employs a newly designed Speculative Aggregation, a dual-side speculative algorithm to avoid frequent output synchronization between the cloud and device. A new scheduling algorithm is further introduced to identify the optimal aggregation side based on real-time network conditions. Evaluations on real-world hardware testbed demonstrate a significant performance improvement of DRAGON-up to 1.9x greater gains over standalone SLM compared to the centralized RAG, substantial reduction in per-token latency, and negligible Time to First Token (TTFT) overhead.
☆ Evaluation Report on MCP Servers
With the rise of LLMs, a large number of Model Context Protocol (MCP) services have emerged since the end of 2024. However, the effectiveness and efficiency of MCP servers have not been well studied. To study these questions, we propose an evaluation framework, called MCPBench. We selected several widely used MCP server and conducted an experimental evaluation on their accuracy, time, and token usage. Our experiments showed that the most effective MCP, Bing Web Search, achieved an accuracy of 64%. Importantly, we found that the accuracy of MCP servers can be substantially enhanced by involving declarative interface. This research paves the way for further investigations into optimized MCP implementations, ultimately leading to better AI-driven applications and data retrieval solutions.
☆ Document Quality Scoring for Web Crawling
The internet contains large amounts of low-quality content, yet users expect web search engines to deliver high-quality, relevant results. The abundant presence of low-quality pages can negatively impact retrieval and crawling processes by wasting resources on these documents. Therefore, search engines can greatly benefit from techniques that leverage efficient quality estimation methods to mitigate these negative impacts. Quality scoring methods for web pages are useful for many processes typical for web search systems, including static index pruning, index tiering, and crawling. Building on work by Chang et al.~\cite{chang2024neural}, who proposed using neural estimators of semantic quality for static index pruning, we extend their approach and apply their neural quality scorers to assess the semantic quality of web pages in crawling prioritisation tasks. In our experimental analysis, we found that prioritising semantically high-quality pages over low-quality ones can improve downstream search effectiveness. Our software contribution consists of a Docker container that computes an effective quality score for a given web page, allowing the quality scorer to be easily included and used in other components of web search systems.
comment: Presented at WOWS2025
☆ Why am I seeing this? Towards recognizing social media recommender systems with missing recommendations
Social media plays a crucial role in shaping society, often amplifying polarization and spreading misinformation. These effects stem from complex dynamics involving user interactions, individual traits, and recommender algorithms driving content selection. Recommender systems, which significantly shape the content users see and decisions they make, offer an opportunity for intervention and regulation. However, assessing their impact is challenging due to algorithmic opacity and limited data availability. To effectively model user decision-making, it is crucial to recognize the recommender system adopted by the platform. This work introduces a method for Automatic Recommender Recognition using Graph Neural Networks (GNNs), based solely on network structure and observed behavior. To infer the hidden recommender, we first train a Recommender Neutral User model (RNU) using a GNN and an adapted hindsight academic network recommender, aiming to reduce reliance on the actual recommender in the data. We then generate several Recommender Hypothesis-specific Synthetic Datasets (RHSD) by combining the RNU with different known recommenders, producing ground truths for testing. Finally, we train Recommender Hypothesis-specific User models (RHU) under various hypotheses and compare each candidate with the original used to generate the RHSD. Our approach enables accurate detection of hidden recommenders and their influence on user behavior. Unlike audit-based methods, it captures system behavior directly, without ad hoc experiments that often fail to reflect real platforms. This study provides insights into how recommenders shape behavior, aiding efforts to reduce polarization and misinformation.
comment: Accepted at RLDM 2025
☆ MSCRS: Multi-modal Semantic Graph Prompt Learning Framework for Conversational Recommender Systems
Conversational Recommender Systems (CRSs) aim to provide personalized recommendations by interacting with users through conversations. Most existing studies of CRS focus on extracting user preferences from conversational contexts. However, due to the short and sparse nature of conversational contexts, it is difficult to fully capture user preferences by conversational contexts only. We argue that multi-modal semantic information can enrich user preference expressions from diverse dimensions (e.g., a user preference for a certain movie may stem from its magnificent visual effects and compelling storyline). In this paper, we propose a multi-modal semantic graph prompt learning framework for CRS, named MSCRS. First, we extract textual and image features of items mentioned in the conversational contexts. Second, we capture higher-order semantic associations within different semantic modalities (collaborative, textual, and image) by constructing modality-specific graph structures. Finally, we propose an innovative integration of multi-modal semantic graphs with prompt learning, harnessing the power of large language models to comprehensively explore high-dimensional semantic relationships. Experimental results demonstrate that our proposed method significantly improves accuracy in item recommendation, as well as generates more natural and contextually relevant content in response generation. We have released the code and the expanded multi-modal CRS datasets to facilitate further exploration in related research\footnote{https://github.com/BIAOBIAO12138/MSCRS-main}.
☆ CSPLADE: Learned Sparse Retrieval with Causal Language Models
In recent years, dense retrieval has been the focus of information retrieval (IR) research. While effective, dense retrieval produces uninterpretable dense vectors, and suffers from the drawback of large index size. Learned sparse retrieval (LSR) has emerged as promising alternative, achieving competitive retrieval performance while also being able to leverage the classical inverted index data structure for efficient retrieval. However, limited works have explored scaling LSR beyond BERT scale. In this work, we identify two challenges in training large language models (LLM) for LSR: (1) training instability during the early stage of contrastive training; (2) suboptimal performance due to pre-trained LLM's unidirectional attention. To address these challenges, we propose two corresponding techniques: (1) a lightweight adaptation training phase to eliminate training instability; (2) two model variants to enable bidirectional information. With these techniques, we are able to train LSR models with 8B scale LLM, and achieve competitive retrieval performance with reduced index size. Furthermore, we are among the first to analyze the performance-efficiency tradeoff of LLM-based LSR model through the lens of model quantization. Our findings provide insights into adapting LLMs for efficient retrieval modeling.
☆ Improving LLM Interpretability and Performance via Guided Embedding Refinement for Sequential Recommendation
The fast development of Large Language Models (LLMs) offers growing opportunities to further improve sequential recommendation systems. Yet for some practitioners, integrating LLMs to their existing base recommendation systems raises questions about model interpretability, transparency and related safety. To partly alleviate challenges from these questions, we propose guided embedding refinement, a method that carries out a guided and interpretable usage of LLM to enhance the embeddings associated with the base recommendation system. Instead of directly using LLMs as the backbone of sequential recommendation systems, we utilize them as auxiliary tools to emulate the sales logic of recommendation and generate guided embeddings that capture domain-relevant semantic information on interpretable attributes. Benefiting from the strong generalization capabilities of the guided embedding, we construct refined embedding by using the guided embedding and reduced-dimension version of the base embedding. We then integrate the refined embedding into the recommendation module for training and inference. A range of numerical experiments demonstrate that guided embedding is adaptable to various given existing base embedding models, and generalizes well across different recommendation tasks. The numerical results show that the refined embedding not only improves recommendation performance, achieving approximately $10\%$ to $50\%$ gains in Mean Reciprocal Rank (MRR), Recall rate, and Normalized Discounted Cumulative Gain (NDCG), but also enhances interpretability, as evidenced by case studies.
☆ RAID: An In-Training Defense against Attribute Inference Attacks in Recommender Systems
In various networks and mobile applications, users are highly susceptible to attribute inference attacks, with particularly prevalent occurrences in recommender systems. Attackers exploit partially exposed user profiles in recommendation models, such as user embeddings, to infer private attributes of target users, such as gender and political views. The goal of defenders is to mitigate the effectiveness of these attacks while maintaining recommendation performance. Most existing defense methods, such as differential privacy and attribute unlearning, focus on post-training settings, which limits their capability of utilizing training data to preserve recommendation performance. Although adversarial training extends defenses to in-training settings, it often struggles with convergence due to unstable training processes. In this paper, we propose RAID, an in-training defense method against attribute inference attacks in recommender systems. In addition to the recommendation objective, we define a defensive objective to ensure that the distribution of protected attributes becomes independent of class labels, making users indistinguishable from attribute inference attacks. Specifically, this defensive objective aims to solve a constrained Wasserstein barycenter problem to identify the centroid distribution that makes the attribute indistinguishable while complying with recommendation performance constraints. To optimize our proposed objective, we use optimal transport to align users with the centroid distribution. We conduct extensive experiments on four real-world datasets to evaluate RAID. The experimental results validate the effectiveness of RAID and demonstrate its significant superiority over existing methods in multiple aspects.
comment: 17 pages
☆ PATFinger: Prompt-Adapted Transferable Fingerprinting against Unauthorized Multimodal Dataset Usage
The multimodal datasets can be leveraged to pre-train large-scale vision-language models by providing cross-modal semantics. Current endeavors for determining the usage of datasets mainly focus on single-modal dataset ownership verification through intrusive methods and non-intrusive techniques, while cross-modal approaches remain under-explored. Intrusive methods can adapt to multimodal datasets but degrade model accuracy, while non-intrusive methods rely on label-driven decision boundaries that fail to guarantee stable behaviors for verification. To address these issues, we propose a novel prompt-adapted transferable fingerprinting scheme from a training-free perspective, called PATFinger, which incorporates the global optimal perturbation (GOP) and the adaptive prompts to capture dataset-specific distribution characteristics. Our scheme utilizes inherent dataset attributes as fingerprints instead of compelling the model to learn triggers. The GOP is derived from the sample distribution to maximize embedding drifts between different modalities. Subsequently, our PATFinger re-aligns the adaptive prompt with GOP samples to capture the cross-modal interactions on the carefully crafted surrogate model. This allows the dataset owner to check the usage of datasets by observing specific prediction behaviors linked to the PATFinger during retrieval queries. Extensive experiments demonstrate the effectiveness of our scheme against unauthorized multimodal dataset usage on various cross-modal retrieval architectures by 30% over state-of-the-art baselines.
♻ ☆ Establishing a Foundation for Tetun Ad-Hoc Text Retrieval: Stemming, Indexing, Retrieval, and Ranking
Searching for information on the internet and digital platforms to satisfy an information need requires effective retrieval solutions. However, such solutions are not yet available for Tetun, making it challenging to find relevant documents for text-based search queries in this language. To address these challenges, this study investigates Tetun text retrieval with a focus on the ad-hoc retrieval task. It begins by developing essential language resources -- including a list of stopwords, a stemmer, and a test collection -- which serve as foundational components for solutions tailored to Tetun text retrieval. Various strategies are then explored using both document titles and content to evaluate retrieval effectiveness. The results show that retrieving document titles, after removing hyphens and apostrophes without applying stemming, significantly improves retrieval performance compared to the baseline. Efficiency increases by 31.37%, while effectiveness achieves an average gain of 9.40% in MAP@10 and 30.35% in nDCG@10 with DFR BM25. Beyond the top-10 cutoff point, Hiemstra LM demonstrates strong performance across various retrieval strategies and evaluation metrics. Contributions of this work include the development of Labadain-Stopwords (a list of 160 Tetun stopwords), Labadain-Stemmer (a Tetun stemmer with three variants), and Labadain-Avaliad\'or (a Tetun test collection containing 59 topics, 33,550 documents, and 5,900 qrels).
comment: Version 3
♻ ☆ Inferring Communities of Interest in Collaborative Learning-based Recommender Systems
Collaborative-learning-based recommender systems, such as those employing Federated Learning (FL) and Gossip Learning (GL), allow users to train models while keeping their history of liked items on their devices. While these methods were seen as promising for enhancing privacy, recent research has shown that collaborative learning can be vulnerable to various privacy attacks. In this paper, we propose a novel attack called Community Inference Attack (CIA), which enables an adversary to identify community members based on a set of target items. What sets CIA apart is its efficiency: it operates at low computational cost by eliminating the need for training surrogate models. Instead, it uses a comparison-based approach, inferring sensitive information by comparing users' models rather than targeting any specific individual model. To evaluate the effectiveness of CIA, we conduct experiments on three real-world recommendation datasets using two recommendation models under both Federated and Gossip-like settings. The results demonstrate that CIA can be up to 10 times more accurate than random guessing. Additionally, we evaluate two mitigation strategies: Differentially Private Stochastic Gradient Descent (DP-SGD) and a Share less policy, which involves sharing fewer, less sensitive model parameters. Our findings suggest that the Share less strategy offers a better privacy-utility trade-off, especially in GL.
♻ ☆ Limitations of Automatic Relevance Assessments with Large Language Models for Fair and Reliable Retrieval Evaluation
Offline evaluation of search systems depends on test collections. These benchmarks provide the researchers with a corpus of documents, topics and relevance judgements indicating which documents are relevant for each topic. While test collections are an integral part of Information Retrieval (IR) research, their creation involves significant efforts in manual annotation. Large language models (LLMs) are gaining much attention as tools for automatic relevance assessment. Recent research has shown that LLM-based assessments yield high systems ranking correlation with human-made judgements. These correlations are helpful in large-scale experiments but less informative if we want to focus on top-performing systems. Moreover, these correlations ignore whether and how LLM-based judgements impact the statistically significant differences among systems with respect to human assessments. In this work, we look at how LLM-generated judgements preserve ranking differences among top-performing systems and also how they preserve pairwise significance evaluation as human judgements. Our results show that LLM-based judgements are unfair at ranking top-performing systems. Moreover, we observe an exceedingly high rate of false positives regarding statistical differences. Our work represents a step forward in the evaluation of the reliability of using LLMs-based judgements for IR evaluation. We hope this will serve as a basis for other researchers to develop more reliable models for automatic relevance assessment.
♻ ☆ Towards Personalized Conversational Sales Agents : Contextual User Profiling for Strategic Action
Conversational Recommender Systems (CRSs) aim to engage users in dialogue to provide tailored recommendations. While traditional CRSs focus on eliciting preferences and retrieving items, real-world e-commerce interactions involve more complex decision-making, where users consider multiple factors beyond simple attributes. To bridge this gap, we introduce Conversational Sales (CSales), a novel task that unifies preference elicitation, recommendation, and persuasion to better support user decision-making. For a realistic evaluation of CSales, we present CSUser, an LLM-based user simulator constructed from real-world data, modeling diverse user profiles with needs and personalities. Additionally, we propose CSI, a conversational sales agent that proactively infers contextual profiles through dialogue for personalized action planning. Extensive experiments demonstrate that CSUser effectively replicates real-world users and emphasize the importance of contextual profiling for strategic action selection, ultimately driving successful purchases in e-commerce.
♻ ☆ Preference-based Learning with Retrieval Augmented Generation for Conversational Question Answering WWW 2025
Conversational Question Answering (ConvQA) involves multiple subtasks, i) to understand incomplete questions in their context, ii) to retrieve relevant information, and iii) to generate answers. This work presents PRAISE, a pipeline-based approach for ConvQA that trains LLM adapters for each of the three subtasks. As labeled training data for individual subtasks is unavailable in practice, PRAISE learns from its own generations using the final answering performance as feedback signal without human intervention and treats intermediate information, like relevant evidence, as weakly labeled data. We apply Direct Preference Optimization by contrasting successful and unsuccessful samples for each subtask. In our experiments, we show the effectiveness of this training paradigm: PRAISE shows improvements per subtask and achieves new state-of-the-art performance on a popular ConvQA benchmark, by gaining 15.5 percentage points increase in precision over baselines.
comment: WWW 2025 Short Paper, 5 pages
♻ ☆ How Relevance Emerges: Interpreting LoRA Fine-Tuning in Reranking LLMs
We conduct a behavioral exploration of LoRA fine-tuned LLMs for Passage Reranking to understand how relevance signals are learned and deployed by Large Language Models. By fine-tuning Mistral-7B, LLaMA3.1-8B, and Pythia-6.9B on MS MARCO under diverse LoRA configurations, we investigate how relevance modeling evolves across checkpoints, the impact of LoRA rank (1, 2, 8, 32), and the relative importance of updated MHA vs. MLP components. Our ablations reveal which layers and projections within LoRA transformations are most critical for reranking accuracy. These findings offer fresh explanations into LoRA's adaptation mechanisms, setting the stage for deeper mechanistic studies in Information Retrieval. All models used in this study have been shared.
comment: Extended Abstract
♻ ☆ Lightning IR: Straightforward Fine-tuning and Inference of Transformer-based Language Models for Information Retrieval WSDM'25
A wide range of transformer-based language models have been proposed for information retrieval tasks. However, including transformer-based models in retrieval pipelines is often complex and requires substantial engineering effort. In this paper, we introduce Lightning IR, an easy-to-use PyTorch Lightning-based framework for applying transformer-based language models in retrieval scenarios. Lightning IR provides a modular and extensible architecture that supports all stages of a retrieval pipeline: from fine-tuning and indexing to searching and re-ranking. Designed to be scalable and reproducible, Lightning IR is available as open-source: https://github.com/webis-de/lightning-ir.
comment: Accepted as a demo at WSDM'25
♻ ☆ Are We Solving a Well-Defined Problem? A Task-Centric Perspective on Recommendation Tasks
Recommender systems (RecSys) leverage user interaction history to predict and suggest relevant items, shaping user experiences across various domains. While many studies adopt a general problem definition, i.e., to recommend preferred items to users based on past interactions, such abstraction often lacks the domain-specific nuances necessary for practical deployment. However, models are frequently evaluated using datasets from online recommender platforms, which inherently reflect these specificities. In this paper, we analyze RecSys task formulations, emphasizing key components such as input-output structures, temporal dynamics, and candidate item selection. All these factors directly impact offline evaluation. We further examine the complexities of user-item interactions, including decision-making costs, multi-step engagements, and unobservable interactions, which may influence model design and loss functions. Additionally, we explore the balance between task specificity and model generalizability, highlighting how well-defined task formulations serve as the foundation for robust evaluation and effective solution development. By clarifying task definitions and their implications, this work provides a structured perspective on RecSys research. The goal is to help researchers better navigate the field, particularly in understanding specificities of the RecSys tasks and ensuring fair and meaningful evaluations.
comment: Work in progress
♻ ☆ SS4Rec: Continuous-Time Sequential Recommendation with State Space Models
Sequential recommendation is a key area in the field of recommendation systems aiming to model user interest based on historical interaction sequences with irregular intervals. While previous recurrent neural network-based and attention-based approaches have achieved significant results, they have limitations in capturing system continuity due to the discrete characteristics. In the context of continuous-time modeling, state space model (SSM) offers a potential solution, as it can effectively capture the dynamic evolution of user interest over time. However, existing SSM-based approaches ignore the impact of irregular time intervals within historical user interactions, making it difficult to model complexed user-item transitions in sequences. To address this issue, we propose a hybrid SSM-based model called SS4Rec for continuous-time sequential recommendation. SS4Rec integrates a time-aware SSM to handle irregular time intervals and a relation-aware SSM to model contextual dependencies, enabling it to infer user interest from both temporal and sequential perspectives. In the training process, the time-aware SSM and the relation-aware SSM are discretized by variable stepsizes according to user interaction time intervals and input data, respectively. This helps capture the continuous dependency from irregular time intervals and provides time-specific personalized recommendations. Experimental studies on five benchmark datasets demonstrate the superiority and effectiveness of SS4Rec.
♻ ☆ Are Generative AI Agents Effective Personalized Financial Advisors? SIGIR 2025
Large language model-based agents are becoming increasingly popular as a low-cost mechanism to provide personalized, conversational advice, and have demonstrated impressive capabilities in relatively simple scenarios, such as movie recommendations. But how do these agents perform in complex high-stakes domains, where domain expertise is essential and mistakes carry substantial risk? This paper investigates the effectiveness of LLM-advisors in the finance domain, focusing on three distinct challenges: (1) eliciting user preferences when users themselves may be unsure of their needs, (2) providing personalized guidance for diverse investment preferences, and (3) leveraging advisor personality to build relationships and foster trust. Via a lab-based user study with 64 participants, we show that LLM-advisors often match human advisor performance when eliciting preferences, although they can struggle to resolve conflicting user needs. When providing personalized advice, the LLM was able to positively influence user behavior, but demonstrated clear failure modes. Our results show that accurate preference elicitation is key, otherwise, the LLM-advisor has little impact, or can even direct the investor toward unsuitable assets. More worryingly, users appear insensitive to the quality of advice being given, or worse these can have an inverse relationship. Indeed, users reported a preference for and increased satisfaction as well as emotional trust with LLMs adopting an extroverted persona, even though those agents provided worse advice.
comment: Accepted for presentation at SIGIR 2025
♻ ☆ Bridging Stepwise Lab-Informed Pretraining and Knowledge-Guided Learning for Diagnostic Reasoning
Despite the growing use of Electronic Health Records (EHR) for AI-assisted diagnosis prediction, most data-driven models struggle to incorporate clinically meaningful medical knowledge. They often rely on limited ontologies, lacking structured reasoning capabilities and comprehensive coverage. This raises an important research question: Will medical knowledge improve predictive models to support stepwise clinical reasoning as performed by human doctors? To address this problem, we propose DuaLK, a dual-expertise framework that combines two complementary sources of information. For external knowledge, we construct a Diagnosis Knowledge Graph (KG) that encodes both hierarchical and semantic relations enriched by large language models (LLM). To align with patient data, we further introduce a lab-informed proxy task that guides the model to follow a clinically consistent, stepwise reasoning process based on lab test signals. Experimental results on two public EHR datasets demonstrate that DuaLK consistently outperforms existing baselines across four clinical prediction tasks. These findings highlight the potential of combining structured medical knowledge with individual-level clinical signals to achieve more accurate and interpretable diagnostic predictions. The source code is publicly available on https://github.com/humphreyhuu/DuaLK.
Robotics 58
☆ Co-optimizing Physical Reconfiguration Parameters and Controllers for an Origami-inspired Reconfigurable Manipulator
Reconfigurable robots that can change their physical configuration post-fabrication have demonstrate their potential in adapting to different environments or tasks. However, it is challenging to determine how to optimally adjust reconfigurable parameters for a given task, especially when the controller depends on the robot's configuration. In this paper, we address this problem using a tendon-driven reconfigurable manipulator composed of multiple serially connected origami-inspired modules as an example. Under tendon actuation, these modules can achieve different shapes and motions, governed by joint stiffnesses (reconfiguration parameters) and the tendon displacements (control inputs). We leverage recent advances in co-optimization of design and control for robotic system to treat reconfiguration parameters as design variables and optimize them using reinforcement learning techniques. We first establish a forward model based on the minimum potential energy method to predict the shape of the manipulator under tendon actuations. Using the forward model as the environment dynamics, we then co-optimize the control policy (on the tendon displacements) and joint stiffnesses of the modules for goal reaching tasks while ensuring collision avoidance. Through co-optimization, we obtain optimized joint stiffness and the corresponding optimal control policy to enable the manipulator to accomplish the task that would be infeasible with fixed reconfiguration parameters (i.e., fixed joint stiffness). We envision the co-optimization framework can be extended to other reconfigurable robotic systems, enabling them to optimally adapt their configuration and behavior for diverse tasks and environments.
☆ MonoDiff9D: Monocular Category-Level 9D Object Pose Estimation via Diffusion Model ICRA'25
Object pose estimation is a core means for robots to understand and interact with their environment. For this task, monocular category-level methods are attractive as they require only a single RGB camera. However, current methods rely on shape priors or CAD models of the intra-class known objects. We propose a diffusion-based monocular category-level 9D object pose generation method, MonoDiff9D. Our motivation is to leverage the probabilistic nature of diffusion models to alleviate the need for shape priors, CAD models, or depth sensors for intra-class unknown object pose estimation. We first estimate coarse depth via DINOv2 from the monocular image in a zero-shot manner and convert it into a point cloud. We then fuse the global features of the point cloud with the input image and use the fused features along with the encoded time step to condition MonoDiff9D. Finally, we design a transformer-based denoiser to recover the object pose from Gaussian noise. Extensive experiments on two popular benchmark datasets show that MonoDiff9D achieves state-of-the-art monocular category-level 9D object pose estimation accuracy without the need for shape priors or CAD models at any stage. Our code will be made public at https://github.com/CNJianLiu/MonoDiff9D.
comment: Accepted by ICRA'25
☆ Region Based SLAM-Aware Exploration: Efficient and Robust Autonomous Mapping Strategy That Can Scale
Autonomous exploration for mapping unknown large scale environments is a fundamental challenge in robotics, with efficiency in time, stability against map corruption and computational resources being crucial. This paper presents a novel approach to indoor exploration that addresses these key issues in existing methods. We introduce a Simultaneous Localization and Mapping (SLAM)-aware region-based exploration strategy that partitions the environment into discrete regions, allowing the robot to incrementally explore and stabilize each region before moving to the next one. This approach significantly reduces redundant exploration and improves overall efficiency. As the device finishes exploring a region and stabilizes it, we also perform SLAM keyframe marginalization, a technique which reduces problem complexity by eliminating variables, while preserving their essential information. To improves robustness and further enhance efficiency, we develop a check- point system that enables the robot to resume exploration from the last stable region in case of failures, eliminating the need for complete re-exploration. Our method, tested in real homes, office and simulations, outperforms state-of-the-art approaches. The improvements demonstrate substantial enhancements in various real world environments, with significant reductions in keyframe usage (85%), submap usage (50% office, 32% home), pose graph optimization time (78-80%), and exploration duration (10-15%). This region-based strategy with keyframe marginalization offers an efficient solution for autonomous robotic mapping.
comment: 8 pages, 9 figures
☆ Teacher Motion Priors: Enhancing Robot Locomotion over Challenging Terrain
Achieving robust locomotion on complex terrains remains a challenge due to high dimensional control and environmental uncertainties. This paper introduces a teacher prior framework based on the teacher student paradigm, integrating imitation and auxiliary task learning to improve learning efficiency and generalization. Unlike traditional paradigms that strongly rely on encoder-based state embeddings, our framework decouples the network design, simplifying the policy network and deployment. A high performance teacher policy is first trained using privileged information to acquire generalizable motion skills. The teacher's motion distribution is transferred to the student policy, which relies only on noisy proprioceptive data, via a generative adversarial mechanism to mitigate performance degradation caused by distributional shifts. Additionally, auxiliary task learning enhances the student policy's feature representation, speeding up convergence and improving adaptability to varying terrains. The framework is validated on a humanoid robot, showing a great improvement in locomotion stability on dynamic terrains and significant reductions in development costs. This work provides a practical solution for deploying robust locomotion strategies in humanoid robots.
comment: 8 pages, 6 figures, 6 tables
☆ Flying Hand: End-Effector-Centric Framework for Versatile Aerial Manipulation Teleoperation and Policy Learning
Aerial manipulation has recently attracted increasing interest from both industry and academia. Previous approaches have demonstrated success in various specific tasks. However, their hardware design and control frameworks are often tightly coupled with task specifications, limiting the development of cross-task and cross-platform algorithms. Inspired by the success of robot learning in tabletop manipulation, we propose a unified aerial manipulation framework with an end-effector-centric interface that decouples high-level platform-agnostic decision-making from task-agnostic low-level control. Our framework consists of a fully-actuated hexarotor with a 4-DoF robotic arm, an end-effector-centric whole-body model predictive controller, and a high-level policy. The high-precision end-effector controller enables efficient and intuitive aerial teleoperation for versatile tasks and facilitates the development of imitation learning policies. Real-world experiments show that the proposed framework significantly improves end-effector tracking accuracy, and can handle multiple aerial teleoperation and imitation learning tasks, including writing, peg-in-hole, pick and place, changing light bulbs, etc. We believe the proposed framework provides one way to standardize and unify aerial manipulation into the general manipulation community and to advance the field. Project website: https://lecar-lab.github.io/flying_hand/.
comment: accepted by RSS 2025
☆ Siamese Network with Dual Attention for EEG-Driven Social Learning: Bridging the Human-Robot Gap in Long-Tail Autonomous Driving
Robots with wheeled, quadrupedal, or humanoid forms are increasingly integrated into built environments. However, unlike human social learning, they lack a critical pathway for intrinsic cognitive development, namely, learning from human feedback during interaction. To understand human ubiquitous observation, supervision, and shared control in dynamic and uncertain environments, this study presents a brain-computer interface (BCI) framework that enables classification of Electroencephalogram (EEG) signals to detect cognitively demanding and safety-critical events. As a timely and motivating co-robotic engineering application, we simulate a human-in-the-loop scenario to flag risky events in semi-autonomous robotic driving-representative of long-tail cases that pose persistent bottlenecks to the safety performance of smart mobility systems and robotic vehicles. Drawing on recent advances in few-shot learning, we propose a dual-attention Siamese convolutional network paired with Dynamic Time Warping Barycenter Averaging approach to generate robust EEG-encoded signal representations. Inverse source localization reveals activation in Broadman areas 4 and 9, indicating perception-action coupling during task-relevant mental imagery. The model achieves 80% classification accuracy under data-scarce conditions and exhibits a nearly 100% increase in the utility of salient features compared to state-of-the-art methods, as measured through integrated gradient attribution. Beyond performance, this study contributes to our understanding of the cognitive architecture required for BCI agents-particularly the role of attention and memory mechanisms-in categorizing diverse mental states and supporting both inter- and intra-subject adaptation. Overall, this research advances the development of cognitive robotics and socially guided learning for service robots in complex built environments.
comment: 50 pages, 18 figures
☆ Ankle Exoskeletons in Walking and Load-Carrying Tasks: Insights into Biomechanics and Human-Robot Interaction
Background: Lower limb exoskeletons can enhance quality of life, but widespread adoption is limited by the lack of frameworks to assess their biomechanical and human-robot interaction effects, which are essential for developing adaptive and personalized control strategies. Understanding impacts on kinematics, muscle activity, and HRI dynamics is key to achieve improved usability of wearable robots. Objectives: We propose a systematic methodology evaluate an ankle exoskeleton's effects on human movement during walking and load-carrying (10 kg front pack), focusing on joint kinematics, muscle activity, and HRI torque signals. Materials and Methods: Using Xsens MVN (inertial motion capture), Delsys EMG, and a unilateral exoskeleton, three experiments were conducted: (1) isolated dorsiflexion/plantarflexion; (2) gait analysis (two subjects, passive/active modes); and (3) load-carrying under assistance. Results and Conclusions: The first experiment confirmed that the HRI sensor captured both voluntary and involuntary torques, providing directional torque insights. The second experiment showed that the device slightly restricted ankle range of motion (RoM) but supported normal gait patterns across all assistance modes. The exoskeleton reduced muscle activity, particularly in active mode. HRI torque varied according to gait phases and highlighted reduced synchronization, suggesting a need for improved support. The third experiment revealed that load-carrying increased GM and TA muscle activity, but the device partially mitigated user effort by reducing muscle activity compared to unassisted walking. HRI increased during load-carrying, providing insights into user-device dynamics. These results demonstrate the importance of tailoring exoskeleton evaluation methods to specific devices and users, while offering a framework for future studies on exoskeleton biomechanics and HRI.
☆ Look-to-Touch: A Vision-Enhanced Proximity and Tactile Sensor for Distance and Geometry Perception in Robotic Manipulation
Camera-based tactile sensors provide robots with a high-performance tactile sensing approach for environment perception and dexterous manipulation. However, achieving comprehensive environmental perception still requires cooperation with additional sensors, which makes the system bulky and limits its adaptability to unstructured environments. In this work, we present a vision-enhanced camera-based dual-modality sensor, which realizes full-scale distance sensing from 50 cm to -3 mm while simultaneously keeping ultra-high-resolution texture sensing and reconstruction capabilities. Unlike conventional designs with fixed opaque gel layers, our sensor features a partially transparent sliding window, enabling mechanical switching between tactile and visual modes. For each sensing mode, a dynamic distance sensing model and a contact geometry reconstruction model are proposed. Through integration with soft robotic fingers, we systematically evaluate the performance of each mode, as well as in their synergistic operation. Experimental results show robust distance tracking across various speeds, nanometer-scale roughness detection, and sub-millimeter 3D texture reconstruction. The combination of both modalities improves the robot's efficiency in executing grasping tasks. Furthermore, the embedded mechanical transmission in the sensor allows for fine-grained intra-hand adjustments and precise manipulation, unlocking new capabilities for soft robotic hands.
☆ Vision based driving agent for race car simulation environments
In recent years, autonomous driving has become a popular field of study. As control at tire grip limit is essential during emergency situations, algorithms developed for racecars are useful for road cars too. This paper examines the use of Deep Reinforcement Learning (DRL) to solve the problem of grip limit driving in a simulated environment. Proximal Policy Optimization (PPO) method is used to train an agent to control the steering wheel and pedals of the vehicle, using only visual inputs to achieve professional human lap times. The paper outlines the formulation of the task of time optimal driving on a race track as a deep reinforcement learning problem, and explains the chosen observations, actions, and reward functions. The results demonstrate human-like learning and driving behavior that utilize maximum tire grip potential.
comment: Submitted to ICMCE 2024 (https://icmce.org/2024.html)
☆ A Quasi-Steady-State Black Box Simulation Approach for the Generation of g-g-g-v Diagrams
The classical g-g diagram, representing the achievable acceleration space for a vehicle, is commonly used as a constraint in trajectory planning and control due to its computational simplicity. To address non-planar road geometries, this concept can be extended to incorporate g-g constraints as a function of vehicle speed and vertical acceleration, commonly referred to as g-g-g-v diagrams. However, the estimation of g-g-g-v diagrams is an open problem. Existing simulation-based approaches struggle to isolate non-transient, open-loop stable states across all combinations of speed and acceleration, while optimization-based methods often require simplified vehicle equations and have potential convergence issues. In this paper, we present a novel, open-source, quasi-steady-state black box simulation approach that applies a virtual inertial force in the longitudinal direction. The method emulates the load conditions associated with a specified longitudinal acceleration while maintaining constant vehicle speed, enabling open-loop steering ramps in a purely QSS manner. Appropriate regulation of the ramp steer rate inherently mitigates transient vehicle dynamics when determining the maximum feasible lateral acceleration. Moreover, treating the vehicle model as a black box eliminates model mismatch issues, allowing the use of high-fidelity or proprietary vehicle dynamics models typically unsuited for optimization approaches. An open-source version of the proposed method is available at: https://github.com/TUM-AVS/GGGVDiagrams
comment: An open-source version of the proposed method is available at: https://github.com/TUM-AVS/GGGVDiagrams
☆ Shoulder Range of Motion Rehabilitation Robot Incorporating Scapulohumeral Rhythm for Frozen Shoulder
This paper presents a novel rehabilitation robot designed to address the challenges of passive range of motion (PROM) exercises for frozen shoulder patients by integrating advanced scapulohumeral rhythm stabilization. Frozen shoulder is characterized by limited glenohumeral motion and disrupted scapulohumeral rhythm, with therapist-assisted interventions being highly effective for restoring normal shoulder function. While existing robotic solutions replicate natural shoulder biomechanics, they lack the ability to stabilize compensatory movements, such as shoulder shrugging, which are critical for effective rehabilitation. Our proposed device features a 6 degrees of freedom (DoF) mechanism, including 5 DoF for shoulder motion and an innovative 1 DoF Joint press for scapular stabilization. The robot employs a personalized two-phase operation: recording normal shoulder movement patterns from the unaffected side and applying them to guide the affected side. Experimental results demonstrated the robot's ability to replicate recorded motion patterns with high precision, with root mean square error (RMSE) values consistently below 1 degree. In simulated frozen shoulder conditions, the robot effectively suppressed scapular elevation, delaying the onset of compensatory movements and guiding the affected shoulder to move more closely in alignment with normal shoulder motion, particularly during arm elevation movements such as abduction and flexion. These findings confirm the robot's potential as a rehabilitation tool capable of automating PROM exercises while correcting compensatory movements. The system provides a foundation for advanced, personalized rehabilitation for patients with frozen shoulders.
comment: This is a preprint of a manuscript that has been submitted for publication
☆ A Human-Sensitive Controller: Adapting to Human Ergonomics and Physical Constraints via Reinforcement Learning
Work-Related Musculoskeletal Disorders continue to be a major challenge in industrial environments, leading to reduced workforce participation, increased healthcare costs, and long-term disability. This study introduces a human-sensitive robotic system aimed at reintegrating individuals with a history of musculoskeletal disorders into standard job roles, while simultaneously optimizing ergonomic conditions for the broader workforce. This research leverages reinforcement learning to develop a human-aware control strategy for collaborative robots, focusing on optimizing ergonomic conditions and preventing pain during task execution. Two RL approaches, Q-Learning and Deep Q-Network (DQN), were implemented and tested to personalize control strategies based on individual user characteristics. Although experimental results revealed a simulation-to-real gap, a fine-tuning phase successfully adapted the policies to real-world conditions. DQN outperformed Q-Learning by completing tasks faster while maintaining zero pain risk and safe ergonomic levels. The structured testing protocol confirmed the system's adaptability to diverse human anthropometries, underscoring the potential of RL-driven cobots to enable safer, more inclusive workplaces.
☆ A Framework for Adaptive Load Redistribution in Human-Exoskeleton-Cobot Systems
Wearable devices like exoskeletons are designed to reduce excessive loads on specific joints of the body. Specifically, single- or two-degrees-of-freedom (DOF) upper-body industrial exoskeletons typically focus on compensating for the strain on the elbow and shoulder joints. However, during daily activities, there is no assurance that external loads are correctly aligned with the supported joints. Optimizing work processes to ensure that external loads are primarily (to the extent that they can be compensated by the exoskeleton) directed onto the supported joints can significantly enhance the overall usability of these devices and the ergonomics of their users. Collaborative robots (cobots) can play a role in this optimization, complementing the collaborative aspects of human work. In this study, we propose an adaptive and coordinated control system for the human-cobot-exoskeleton interaction. This system adjusts the task coordinates to maximize the utilization of the supported joints. When the torque limits of the exoskeleton are exceeded, the framework continuously adapts the task frame, redistributing excessive loads to non-supported body joints to prevent overloading the supported ones. We validated our approach in an equivalent industrial painting task involving a single-DOF elbow exoskeleton, a cobot, and four subjects, each tested in four different initial arm configurations with five distinct optimisation weight matrices and two different payloads.
comment: Accepted to be published in IEEE Robotics and Automation Letters
☆ Joint Action Language Modelling for Transparent Policy Execution
An agent's intention often remains hidden behind the black-box nature of embodied policies. Communication using natural language statements that describe the next action can provide transparency towards the agent's behavior. We aim to insert transparent behavior directly into the learning process, by transforming the problem of policy learning into a language generation problem and combining it with traditional autoregressive modelling. The resulting model produces transparent natural language statements followed by tokens representing the specific actions to solve long-horizon tasks in the Language-Table environment. Following previous work, the model is able to learn to produce a policy represented by special discretized tokens in an autoregressive manner. We place special emphasis on investigating the relationship between predicting actions and producing high-quality language for a transparent agent. We find that in many cases both the quality of the action trajectory and the transparent statement increase when they are generated simultaneously.
☆ Prior Does Matter: Visual Navigation via Denoising Diffusion Bridge Models
Recent advancements in diffusion-based imitation learning, which show impressive performance in modeling multimodal distributions and training stability, have led to substantial progress in various robot learning tasks. In visual navigation, previous diffusion-based policies typically generate action sequences by initiating from denoising Gaussian noise. However, the target action distribution often diverges significantly from Gaussian noise, leading to redundant denoising steps and increased learning complexity. Additionally, the sparsity of effective action distributions makes it challenging for the policy to generate accurate actions without guidance. To address these issues, we propose a novel, unified visual navigation framework leveraging the denoising diffusion bridge models named NaviBridger. This approach enables action generation by initiating from any informative prior actions, enhancing guidance and efficiency in the denoising process. We explore how diffusion bridges can enhance imitation learning in visual navigation tasks and further examine three source policies for generating prior actions. Extensive experiments in both simulated and real-world indoor and outdoor scenarios demonstrate that NaviBridger accelerates policy inference and outperforms the baselines in generating target action sequences. Code is available at https://github.com/hren20/NaiviBridger.
☆ EmbodiedAgent: A Scalable Hierarchical Approach to Overcome Practical Challenge in Multi-Robot Control
This paper introduces EmbodiedAgent, a hierarchical framework for heterogeneous multi-robot control. EmbodiedAgent addresses critical limitations of hallucination in impractical tasks. Our approach integrates a next-action prediction paradigm with a structured memory system to decompose tasks into executable robot skills while dynamically validating actions against environmental constraints. We present MultiPlan+, a dataset of more than 18,000 annotated planning instances spanning 100 scenarios, including a subset of impractical cases to mitigate hallucination. To evaluate performance, we propose the Robot Planning Assessment Schema (RPAS), combining automated metrics with LLM-aided expert grading. Experiments demonstrate EmbodiedAgent's superiority over state-of-the-art models, achieving 71.85% RPAS score. Real-world validation in an office service task highlights its ability to coordinate heterogeneous robots for long-horizon objectives.
☆ KeyMPs: One-Shot Vision-Language Guided Motion Generation by Sequencing DMPs for Occlusion-Rich Tasks
Dynamic Movement Primitives (DMPs) provide a flexible framework wherein smooth robotic motions are encoded into modular parameters. However, they face challenges in integrating multimodal inputs commonly used in robotics like vision and language into their framework. To fully maximize DMPs' potential, enabling them to handle multimodal inputs is essential. In addition, we also aim to extend DMPs' capability to handle object-focused tasks requiring one-shot complex motion generation, as observation occlusion could easily happen mid-execution in such tasks (e.g., knife occlusion in cake icing, hand occlusion in dough kneading, etc.). A promising approach is to leverage Vision-Language Models (VLMs), which process multimodal data and can grasp high-level concepts. However, they typically lack enough knowledge and capabilities to directly infer low-level motion details and instead only serve as a bridge between high-level instructions and low-level control. To address this limitation, we propose Keyword Labeled Primitive Selection and Keypoint Pairs Generation Guided Movement Primitives (KeyMPs), a framework that combines VLMs with sequencing of DMPs. KeyMPs use VLMs' high-level reasoning capability to select a reference primitive through keyword labeled primitive selection and VLMs' spatial awareness to generate spatial scaling parameters used for sequencing DMPs by generalizing the overall motion through keypoint pairs generation, which together enable one-shot vision-language guided motion generation that aligns with the intent expressed in the multimodal input. We validate our approach through an occlusion-rich manipulation task, specifically object cutting experiments in both simulated and real-world environments, demonstrating superior performance over other DMP-based methods that integrate VLMs support.
comment: 17 pages, Submitted to IEEE Access April 9th 2025
☆ NaviDiffusor: Cost-Guided Diffusion Model for Visual Navigation
Visual navigation, a fundamental challenge in mobile robotics, demands versatile policies to handle diverse environments. Classical methods leverage geometric solutions to minimize specific costs, offering adaptability to new scenarios but are prone to system errors due to their multi-modular design and reliance on hand-crafted rules. Learning-based methods, while achieving high planning success rates, face difficulties in generalizing to unseen environments beyond the training data and often require extensive training. To address these limitations, we propose a hybrid approach that combines the strengths of learning-based methods and classical approaches for RGB-only visual navigation. Our method first trains a conditional diffusion model on diverse path-RGB observation pairs. During inference, it integrates the gradients of differentiable scene-specific and task-level costs, guiding the diffusion model to generate valid paths that meet the constraints. This approach alleviates the need for retraining, offering a plug-and-play solution. Extensive experiments in both indoor and outdoor settings, across simulated and real-world scenarios, demonstrate zero-shot transfer capability of our approach, achieving higher success rates and fewer collisions compared to baseline methods. Code will be released at https://github.com/SYSU-RoboticsLab/NaviD.
☆ FLoRA: Sample-Efficient Preference-based RL via Low-Rank Style Adaptation of Reward Functions ICRA
Preference-based reinforcement learning (PbRL) is a suitable approach for style adaptation of pre-trained robotic behavior: adapting the robot's policy to follow human user preferences while still being able to perform the original task. However, collecting preferences for the adaptation process in robotics is often challenging and time-consuming. In this work we explore the adaptation of pre-trained robots in the low-preference-data regime. We show that, in this regime, recent adaptation approaches suffer from catastrophic reward forgetting (CRF), where the updated reward model overfits to the new preferences, leading the agent to become unable to perform the original task. To mitigate CRF, we propose to enhance the original reward model with a small number of parameters (low-rank matrices) responsible for modeling the preference adaptation. Our evaluation shows that our method can efficiently and effectively adjust robotic behavior to human preferences across simulation benchmark tasks and multiple real-world robotic tasks.
comment: Accepted at 2025 IEEE International Conference on Robotics & Automation (ICRA). We provide videos of our results and source code at https://sites.google.com/view/preflora/
☆ GenTe: Generative Real-world Terrains for General Legged Robot Locomotion Control
Developing bipedal robots capable of traversing diverse real-world terrains presents a fundamental robotics challenge, as existing methods using predefined height maps and static environments fail to address the complexity of unstructured landscapes. To bridge this gap, we propose GenTe, a framework for generating physically realistic and adaptable terrains to train generalizable locomotion policies. GenTe constructs an atomic terrain library that includes both geometric and physical terrains, enabling curriculum training for reinforcement learning-based locomotion policies. By leveraging function-calling techniques and reasoning capabilities of Vision-Language Models (VLMs), GenTe generates complex, contextually relevant terrains from textual and graphical inputs. The framework introduces realistic force modeling for terrain interactions, capturing effects such as soil sinkage and hydrodynamic resistance. To the best of our knowledge, GenTe is the first framework that systemically generates simulation environments for legged robot locomotion control. Additionally, we introduce a benchmark of 100 generated terrains. Experiments demonstrate improved generalization and robustness in bipedal robot locomotion.
☆ Efficient Task-specific Conditional Diffusion Policies: Shortcut Model Acceleration and SO(3) Optimization CVPR 2025
Imitation learning, particularly Diffusion Policies based methods, has recently gained significant traction in embodied AI as a powerful approach to action policy generation. These models efficiently generate action policies by learning to predict noise. However, conventional Diffusion Policy methods rely on iterative denoising, leading to inefficient inference and slow response times, which hinder real-time robot control. To address these limitations, we propose a Classifier-Free Shortcut Diffusion Policy (CF-SDP) that integrates classifier-free guidance with shortcut-based acceleration, enabling efficient task-specific action generation while significantly improving inference speed. Furthermore, we extend diffusion modeling to the SO(3) manifold in shortcut model, defining the forward and reverse processes in its tangent space with an isotropic Gaussian distribution. This ensures stable and accurate rotational estimation, enhancing the effectiveness of diffusion-based control. Our approach achieves nearly 5x acceleration in diffusion inference compared to DDIM-based Diffusion Policy while maintaining task performance. Evaluations both on the RoboTwin simulation platform and real-world scenarios across various tasks demonstrate the superiority of our method.
comment: Accepted to CVPR 2025 Workshop on 2nd MEIS
☆ LangPert: Detecting and Handling Task-level Perturbations for Robust Object Rearrangement
Task execution for object rearrangement could be challenged by Task-Level Perturbations (TLP), i.e., unexpected object additions, removals, and displacements that can disrupt underlying visual policies and fundamentally compromise task feasibility and progress. To address these challenges, we present LangPert, a language-based framework designed to detect and mitigate TLP situations in tabletop rearrangement tasks. LangPert integrates a Visual Language Model (VLM) to comprehensively monitor policy's skill execution and environmental TLP, while leveraging the Hierarchical Chain-of-Thought (HCoT) reasoning mechanism to enhance the Large Language Model (LLM)'s contextual understanding and generate adaptive, corrective skill-execution plans. Our experimental results demonstrate that LangPert handles diverse TLP situations more effectively than baseline methods, achieving higher task completion rates, improved execution efficiency, and potential generalization to unseen scenarios.
☆ SIO-Mapper: A Framework for Lane-Level HD Map Construction Using Satellite Images and OpenStreetMap with No On-Site Visits
High-definition (HD) maps, particularly those containing lane-level information regarded as ground truth, are crucial for vehicle localization research. Traditionally, constructing HD maps requires highly accurate sensor measurements collection from the target area, followed by manual annotation to assign semantic information. Consequently, HD maps are limited in terms of geographic coverage. To tackle this problem, in this paper, we propose SIO-Mapper, a novel lane-level HD map construction framework that constructs city-scale maps without physical site visits by utilizing satellite images and OpenStreetmap data. One of the key contributions of SIO-Mapper is its ability to extract lane information more accurately by introducing SIO-Net, a novel deep learning network that integrates features from satellite image and OpenStreetmap using both Transformer-based and convolution-based encoders. Furthermore, to overcome challenges in merging lanes over large areas, we introduce a novel lane integration methodology that combines cluster-based and graph-based approaches. This algorithm ensures the seamless aggregation of lane segments with high accuracy and coverage, even in complex road environments. We validated SIO-Mapper on the Naver Labs Open Dataset and NuScenes dataset, demonstrating better performance in various environments including Korea, the United States, and Singapore compared to the state-of-the-art lane-level HD mapconstruction methods.
☆ NeRF-Based Transparent Object Grasping Enhanced by Shape Priors
Transparent object grasping remains a persistent challenge in robotics, largely due to the difficulty of acquiring precise 3D information. Conventional optical 3D sensors struggle to capture transparent objects, and machine learning methods are often hindered by their reliance on high-quality datasets. Leveraging NeRF's capability for continuous spatial opacity modeling, our proposed architecture integrates a NeRF-based approach for reconstructing the 3D information of transparent objects. Despite this, certain portions of the reconstructed 3D information may remain incomplete. To address these deficiencies, we introduce a shape-prior-driven completion mechanism, further refined by a geometric pose estimation method we have developed. This allows us to obtain a complete and reliable 3D information of transparent objects. Utilizing this refined data, we perform scene-level grasp prediction and deploy the results in real-world robotic systems. Experimental validation demonstrates the efficacy of our architecture, showcasing its capability to reliably capture 3D information of various transparent objects in cluttered scenes, and correspondingly, achieve high-quality, stables, and executable grasp predictions.
☆ ST-Booster: An Iterative SpatioTemporal Perception Booster for Vision-and-Language Navigation in Continuous Environments
Vision-and-Language Navigation in Continuous Environments (VLN-CE) requires agents to navigate unknown, continuous spaces based on natural language instructions. Compared to discrete settings, VLN-CE poses two core perception challenges. First, the absence of predefined observation points leads to heterogeneous visual memories and weakened global spatial correlations. Second, cumulative reconstruction errors in three-dimensional scenes introduce structural noise, impairing local feature perception. To address these challenges, this paper proposes ST-Booster, an iterative spatiotemporal booster that enhances navigation performance through multi-granularity perception and instruction-aware reasoning. ST-Booster consists of three key modules -- Hierarchical SpatioTemporal Encoding (HSTE), Multi-Granularity Aligned Fusion (MGAF), and ValueGuided Waypoint Generation (VGWG). HSTE encodes long-term global memory using topological graphs and captures shortterm local details via grid maps. MGAF aligns these dualmap representations with instructions through geometry-aware knowledge fusion. The resulting representations are iteratively refined through pretraining tasks. During reasoning, VGWG generates Guided Attention Heatmaps (GAHs) to explicitly model environment-instruction relevance and optimize waypoint selection. Extensive comparative experiments and performance analyses are conducted, demonstrating that ST-Booster outperforms existing state-of-the-art methods, particularly in complex, disturbance-prone environments.
comment: 11 pages, 7 figures
☆ Score Matching Diffusion Based Feedback Control and Planning of Nonlinear Systems
We propose a novel control-theoretic framework that leverages principles from generative modeling -- specifically, Denoising Diffusion Probabilistic Models (DDPMs) -- to stabilize control-affine systems with nonholonomic constraints. Unlike traditional stochastic approaches, which rely on noise-driven dynamics in both forward and reverse processes, our method crucially eliminates the need for noise in the reverse phase, making it particularly relevant for control applications. We introduce two formulations: one where noise perturbs all state dimensions during the forward phase while the control system enforces time reversal deterministically, and another where noise is restricted to the control channels, embedding system constraints directly into the forward process. For controllable nonlinear drift-free systems, we prove that deterministic feedback laws can exactly reverse the forward process, ensuring that the system's probability density evolves correctly without requiring artificial diffusion in the reverse phase. Furthermore, for linear time-invariant systems, we establish a time-reversal result under the second formulation. By eliminating noise in the backward process, our approach provides a more practical alternative to machine learning-based denoising methods, which are unsuitable for control applications due to the presence of stochasticity. We validate our results through numerical simulations on benchmark systems, including a unicycle model in a domain with obstacles, a driftless five-dimensional system, and a four-dimensional linear system, demonstrating the potential for applying diffusion-inspired techniques in linear, nonlinear, and settings with state space constraints.
☆ PreCi: Pretraining and Continual Improvement of Humanoid Locomotion via Model-Assumption-Based Regularization
Humanoid locomotion is a challenging task due to its inherent complexity and high-dimensional dynamics, as well as the need to adapt to diverse and unpredictable environments. In this work, we introduce a novel learning framework for effectively training a humanoid locomotion policy that imitates the behavior of a model-based controller while extending its capabilities to handle more complex locomotion tasks, such as more challenging terrain and higher velocity commands. Our framework consists of three key components: pre-training through imitation of the model-based controller, fine-tuning via reinforcement learning, and model-assumption-based regularization (MAR) during fine-tuning. In particular, MAR aligns the policy with actions from the model-based controller only in states where the model assumption holds to prevent catastrophic forgetting. We evaluate the proposed framework through comprehensive simulation tests and hardware experiments on a full-size humanoid robot, Digit, demonstrating a forward speed of 1.5 m/s and robust locomotion across diverse terrains, including slippery, sloped, uneven, and sandy terrains.
☆ RoboCup Rescue 2025 Team Description Paper UruBots
This paper describes the approach used by Team UruBots for participation in the 2025 RoboCup Rescue Robot League competition. Our team aims to participate for the first time in this competition at RoboCup, using experience learned from previous competitions and research. We present our vehicle and our approach to tackle the task of detecting and finding victims in search and rescue environments. Our approach contains known topics in robotics, such as ROS, SLAM, Human Robot Interaction and segmentation and perception. Our proposed approach is open source, available to the RoboCup Rescue community, where we aim to learn and contribute to the league.
☆ SeeTree -- A modular, open-source system for tree detection and orchard localization
Accurate localization is an important functional requirement for precision orchard management. However, there are few off-the-shelf commercial solutions available to growers. In this paper, we present SeeTree, a modular, open source embedded system for tree trunk detection and orchard localization that is deployable on any vehicle. Building on our prior work on vision-based in-row localization using particle filters, SeeTree includes several new capabilities. First, it provides capacity for full orchard localization including out-of-row headland turning. Second, it includes the flexibility to integrate either visual, GNSS, or wheel odometry in the motion model. During field experiments in a commercial orchard, the system converged to the correct location 99% of the time over 800 trials, even when starting with large uncertainty in the initial particle locations. When turning out of row, the system correctly tracked 99% of the turns (860 trials representing 43 unique row changes). To help support adoption and future research and development, we make our dataset, design files, and source code freely available to the community.
comment: 26 pages, 12 figures
☆ ReasonDrive: Efficient Visual Question Answering for Autonomous Vehicles with Reasoning-Enhanced Small Vision-Language Models
Vision-language models (VLMs) show promise for autonomous driving but often lack transparent reasoning capabilities that are critical for safety. We investigate whether explicitly modeling reasoning during fine-tuning enhances VLM performance on driving decision tasks. Using GPT-4o, we generate structured reasoning chains for driving scenarios from the DriveLM benchmark with category-specific prompting strategies. We compare reasoning-based fine-tuning, answer-only fine-tuning, and baseline instruction-tuned models across multiple small VLM families (Llama 3.2, Llava 1.5, and Qwen 2.5VL). Our results demonstrate that reasoning-based fine-tuning consistently outperforms alternatives, with Llama3.2-11B-reason achieving the highest performance. Models fine-tuned with reasoning show substantial improvements in accuracy and text generation quality, suggesting explicit reasoning enhances internal representations for driving decisions. These findings highlight the importance of transparent decision processes in safety-critical domains and offer a promising direction for developing more interpretable autonomous driving systems.
☆ Communication-aware Hierarchical Map Compression of Time-Varying Environments for Mobile Robots
In this paper, we develop a systematic framework for the time-sequential compression of dynamic probabilistic occupancy grids. Our approach leverages ideas from signal compression theory to formulate an optimization problem that searches for a multi-resolution hierarchical encoder that balances the quality of the compressed map (distortion) with its description size, the latter of which relates to the bandwidth required to reliably transmit the map to other agents or to store map estimates in on-board memory. The resulting optimization problem allows for multi-resolution map compressions to be obtained that satisfy available communication or memory resources, and does not require knowledge of the occupancy map dynamics. We develop an algorithm to solve our problem, and demonstrate the utility of the proposed framework in simulation on both static (i.e., non-time varying) and dynamic (time-varying) occupancy maps.
☆ Real-time Seafloor Segmentation and Mapping
Posidonia oceanica meadows are a species of seagrass highly dependent on rocks for their survival and conservation. In recent years, there has been a concerning global decline in this species, emphasizing the critical need for efficient monitoring and assessment tools. While deep learning-based semantic segmentation and visual automated monitoring systems have shown promise in a variety of applications, their performance in underwater environments remains challenging due to complex water conditions and limited datasets. This paper introduces a framework that combines machine learning and computer vision techniques to enable an autonomous underwater vehicle (AUV) to inspect the boundaries of Posidonia oceanica meadows autonomously. The framework incorporates an image segmentation module using an existing Mask R-CNN model and a strategy for Posidonia oceanica meadow boundary tracking. Furthermore, a new class dedicated to rocks is introduced to enhance the existing model, aiming to contribute to a comprehensive monitoring approach and provide a deeper understanding of the intricate interactions between the meadow and its surrounding environment. The image segmentation model is validated using real underwater images, while the overall inspection framework is evaluated in a realistic simulation environment, replicating actual monitoring scenarios with real underwater images. The results demonstrate that the proposed framework enables the AUV to autonomously accomplish the main tasks of underwater inspection and segmentation of rocks. Consequently, this work holds significant potential for the conservation and protection of marine environments, providing valuable insights into the status of Posidonia oceanica meadows and supporting targeted preservation efforts
☆ CleanMAP: Distilling Multimodal LLMs for Confidence-Driven Crowdsourced HD Map Updates CVPR
The rapid growth of intelligent connected vehicles (ICVs) and integrated vehicle-road-cloud systems has increased the demand for accurate, real-time HD map updates. However, ensuring map reliability remains challenging due to inconsistencies in crowdsourced data, which suffer from motion blur, lighting variations, adverse weather, and lane marking degradation. This paper introduces CleanMAP, a Multimodal Large Language Model (MLLM)-based distillation framework designed to filter and refine crowdsourced data for high-confidence HD map updates. CleanMAP leverages an MLLM-driven lane visibility scoring model that systematically quantifies key visual parameters, assigning confidence scores (0-10) based on their impact on lane detection. A novel dynamic piecewise confidence-scoring function adapts scores based on lane visibility, ensuring strong alignment with human evaluations while effectively filtering unreliable data. To further optimize map accuracy, a confidence-driven local map fusion strategy ranks and selects the top-k highest-scoring local maps within an optimal confidence range (best score minus 10%), striking a balance between data quality and quantity. Experimental evaluations on a real-world autonomous vehicle dataset validate CleanMAP's effectiveness, demonstrating that fusing the top three local maps achieves the lowest mean map update error of 0.28m, outperforming the baseline (0.37m) and meeting stringent accuracy thresholds (<= 0.32m). Further validation with real-vehicle data confirms 84.88% alignment with human evaluators, reinforcing the model's robustness and reliability. This work establishes CleanMAP as a scalable and deployable solution for crowdsourced HD map updates, ensuring more precise and reliable autonomous navigation. The code will be available at https://Ankit-Zefan.github.io/CleanMap/
comment: Kun Jiang, Mengmeng Yang and Diange Yang are Corresponding Author. The main paper and supplementary material are both included here, total 23 pages (main paper is 10 pages and supplementary material is 13 pages), total 17 figures (6 figures in main paper and 11 figures in supplementary material), this paper is Accepted to CVPR WDFM-AD Workshop 2025, The code will be available at https://Ankit-Zefan.github.io/CleanMap/
☆ HyRRT-Connect: Bidirectional Motion Planning for Hybrid Dynamical Systems
This paper proposes a bidirectional rapidly-exploring random trees (RRT) algorithm to solve the motion planning problem for hybrid systems. The proposed algorithm, called HyRRT-Connect, propagates in both forward and backward directions in hybrid time until an overlap between the forward and backward propagation results is detected. Then, HyRRT-Connect constructs a motion plan through the reversal and concatenation of functions defined on hybrid time domains, ensuring that the motion plan satisfies the given hybrid dynamics. To address the potential discontinuity along the flow caused by tolerating some distance between the forward and backward partial motion plans, we reconstruct the backward partial motion plan by a forward-in-hybrid-time simulation from the final state of the forward partial motion plan. effectively eliminating the discontinuity. The proposed algorithm is applied to an actuated bouncing ball system and a walking robot example to highlight its computational improvement.
comment: 59 pages, 9 figures, submitted to IJRR. arXiv admin note: substantial text overlap with arXiv:2403.18413; text overlap with arXiv:2406.01802
☆ Probabilistic Task Parameterization of Tool-Tissue Interaction via Sparse Landmarks Tracking in Robotic Surgery ICRA'25
Accurate modeling of tool-tissue interactions in robotic surgery requires precise tracking of deformable tissues and integration of surgical domain knowledge. Traditional methods rely on labor-intensive annotations or rigid assumptions, limiting flexibility. We propose a framework combining sparse keypoint tracking and probabilistic modeling that propagates expert-annotated landmarks across endoscopic frames, even with large tissue deformations. Clustered tissue keypoints enable dynamic local transformation construction via PCA, and tool poses, tracked similarly, are expressed relative to these frames. Embedding these into a Task-Parameterized Gaussian Mixture Model (TP-GMM) integrates data-driven observations with labeled clinical expertise, effectively predicting relative tool-tissue poses and enhancing visual understanding of robotic surgical motions directly from video data.
comment: Submitted to ICRA'25 Workshop of 3rd Robot-Assisted Medical Imaging
☆ Toward Aligning Human and Robot Actions via Multi-Modal Demonstration Learning ICRA'25
Understanding action correspondence between humans and robots is essential for evaluating alignment in decision-making, particularly in human-robot collaboration and imitation learning within unstructured environments. We propose a multimodal demonstration learning framework that explicitly models human demonstrations from RGB video with robot demonstrations in voxelized RGB-D space. Focusing on the "pick and place" task from the RH20T dataset, we utilize data from 5 users across 10 diverse scenes. Our approach combines ResNet-based visual encoding for human intention modeling and a Perceiver Transformer for voxel-based robot action prediction. After 2000 training epochs, the human model reaches 71.67% accuracy, and the robot model achieves 71.8% accuracy, demonstrating the framework's potential for aligning complex, multimodal human and robot behaviors in manipulation tasks.
comment: ICRA'25 Workshop: Human-Centered Robot Learning in the Era of Big Data and Large Models
♻ ☆ Intelligent Framework for Human-Robot Collaboration: Dynamic Ergonomics and Adaptive Decision-Making
The integration of collaborative robots into industrial environments has improved productivity, but has also highlighted significant challenges related to operator safety and ergonomics. This paper proposes an innovative framework that integrates advanced visual perception, continuous ergonomic monitoring, and adaptive Behaviour Tree decision-making to overcome the limitations of traditional methods that typically operate as isolated components. Our approach synthesizes deep learning models, advanced tracking algorithms, and dynamic ergonomic assessments into a modular, scalable, and adaptive system. Experimental validation demonstrates the framework's superiority over existing solutions across multiple dimensions: the visual perception module outperformed previous detection models with 72.4% mAP@50:95; the system achieved high accuracy in recognizing operator intentions (92.5%); it promptly classified ergonomic risks with minimal latency (0.57 seconds); and it dynamically managed robotic interventions with exceptionally responsive decision-making capabilities (0.07 seconds), representing a 56% improvement over benchmark systems. This comprehensive solution provides a robust platform for enhancing human-robot collaboration in industrial environments by prioritizing ergonomic safety, operational efficiency, and real-time adaptability.
comment: 15 pagine, 8figure, 3 tabelle, formato conferenza IEEE
♻ ☆ RINGO: Real-time Navigation with a Guiding Trajectory for Aerial Manipulators in Unknown Environments
Motion planning for aerial manipulators in constrained environments has typically been limited to known environments or simplified to that of multi-rotors, which leads to poor adaptability and overly conservative trajectories. This paper presents RINGO: Real-time Navigation with a Guiding Trajectory, a novel planning framework that enables aerial manipulators to navigate unknown environments in real time. The proposed method simultaneously considers the positions of both the multi-rotor and the end-effector. A pre-obtained multi-rotor trajectory serves as a guiding reference, allowing the end-effector to generate a smooth, collision-free, and workspace-compatible trajectory. Leveraging the convex hull property of B-spline curves, we theoretically guarantee that the trajectory remains within the reachable workspace. To the best of our knowledge, this is the first work that enables real-time navigation of aerial manipulators in unknown environments. The simulation and experimental results show the effectiveness of the proposed method. The proposed method generates less conservative trajectories than approaches that consider only the multi-rotor.
comment: 9 pages, 15 figures
♻ ☆ A Comprehensive Review on Traffic Datasets and Simulators for Autonomous Vehicles
Autonomous driving has rapidly evolved through synergistic developments in hardware and artificial intelligence. This comprehensive review investigates traffic datasets and simulators as dual pillars supporting autonomous vehicle (AV) development. Unlike prior surveys that examine these resources independently, we present an integrated analysis spanning the entire AV pipeline-perception, localization, prediction, planning, and control. We evaluate annotation practices and quality metrics while examining how geographic diversity and environmental conditions affect system reliability. Our analysis includes detailed characterizations of datasets organized by functional domains and an in-depth examination of traffic simulators categorized by their specialized contributions to research and development. The paper explores emerging trends, including novel architecture frameworks, multimodal AI integration, and advanced data generation techniques that address critical edge cases. By highlighting the interconnections between real-world data collection and simulation environments, this review offers researchers a roadmap for developing more robust and resilient autonomous systems equipped to handle the diverse challenges encountered in real-world driving environments.
♻ ☆ Motion Control in Multi-Rotor Aerial Robots Using Deep Reinforcement Learning
This paper investigates the application of Deep Reinforcement (DRL) Learning to address motion control challenges in drones for additive manufacturing (AM). Drone-based additive manufacturing promises flexible and autonomous material deposition in large-scale or hazardous environments. However, achieving robust real-time control of a multi-rotor aerial robot under varying payloads and potential disturbances remains challenging. Traditional controllers like PID often require frequent parameter re-tuning, limiting their applicability in dynamic scenarios. We propose a DRL framework that learns adaptable control policies for multi-rotor drones performing waypoint navigation in AM tasks. We compare Deep Deterministic Policy Gradient (DDPG) and Twin Delayed Deep Deterministic Policy Gradient (TD3) within a curriculum learning scheme designed to handle increasing complexity. Our experiments show TD3 consistently balances training stability, accuracy, and success, particularly when mass variability is introduced. These findings provide a scalable path toward robust, autonomous drone control in additive manufacturing.
♻ ☆ A Smooth Analytical Formulation of Collision Detection and Rigid Body Dynamics With Contact
Generating intelligent robot behavior in contact-rich settings is a research problem where zeroth-order methods currently prevail. A major contributor to the success of such methods is their robustness in the face of non-smooth and discontinuous optimization landscapes that are characteristic of contact interactions, yet zeroth-order methods remain computationally inefficient. It is therefore desirable to develop methods for perception, planning and control in contact-rich settings that can achieve further efficiency by making use of first and second order information (i.e., gradients and Hessians). To facilitate this, we present a joint formulation of collision detection and contact modelling which, compared to existing differentiable simulation approaches, provides the following benefits: i) it results in forward and inverse dynamics that are entirely analytical (i.e. do not require solving optimization or root-finding problems with iterative methods) and smooth (i.e. twice differentiable), ii) it supports arbitrary collision geometries without needing a convex decomposition, and iii) its runtime is independent of the number of contacts. Through simulation experiments, we demonstrate the validity of the proposed formulation as a "physics for inference" that can facilitate future development of efficient methods to generate intelligent contact-rich behavior.
comment: Added references to point-based implicit surface representations
♻ ☆ STLCCP: Efficient Convex Optimization-based Framework for Signal Temporal Logic Specifications
Signal temporal logic (STL) is a powerful formalism for specifying various temporal properties in dynamical systems. However, existing methods, such as mixed-integer programming and nonlinear programming, often struggle to efficiently solve control problems with complex, long-horizon STL specifications. This study introduces \textit{STLCCP}, a novel convex optimization-based framework that leverages key structural properties of STL: monotonicity of the robustness function, its hierarchical tree structure, and correspondence between convexity/concavity in optimizations and conjunctiveness/disjunctiveness in specifications. The framework begins with a structure-aware decomposition of STL formulas, transforming the problem into an equivalent difference of convex (DC) programs. This is then solved sequentially as a convex quadratic program using an improved version of the convex-concave procedure (CCP). To further enhance efficiency, we develop a smooth approximation of the robustness function using a function termed the \textit{mellowmin} function, specifically tailored to the proposed framework. Numerical experiments on motion planning benchmarks demonstrate that \textit{STLCCP} can efficiently handle complex scenarios over long horizons, outperforming existing methods.
comment: 32 pages
♻ ☆ MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation
Multimodal Large Language Models (MLLMs) excel in understanding complex language and visual data, enabling generalist robotic systems to interpret instructions and perform embodied tasks. Nevertheless, their real-world deployment is hindered by substantial computational and storage demands. Recent insights into the homogeneous patterns in the LLM layer have inspired sparsification techniques to address these challenges, such as early exit and token pruning. However, these methods often neglect the critical role of the final layers that encode the semantic information most relevant to downstream robotic tasks. Aligning with the recent breakthrough of the Shallow Brain Hypothesis (SBH) in neuroscience and the mixture of experts in model sparsification, we conceptualize each LLM layer as an expert and propose a Mixture-of-Layers Vision-Language-Action model (MoLe-VLA, or simply MoLe) architecture for dynamic LLM layer activation. We introduce a Spatial-Temporal Aware Router (STAR) for MoLe to selectively activate only parts of the layers based on the robot's current state, mimicking the brain's distinct signal pathways specialized for cognition and causal reasoning. Additionally, to compensate for the cognitive ability of LLMs lost in MoLe, we devise a Cognition Self-Knowledge Distillation (CogKD) framework. CogKD enhances the understanding of task demands and improves the generation of task-relevant action sequences by leveraging cognitive features. Extensive experiments conducted in both RLBench simulation and real-world environments demonstrate the superiority of MoLe-VLA in both efficiency and performance. Specifically, MoLe-VLA achieves an 8% improvement in the mean success rate across ten tasks while reducing computational costs by up to x5.6 compared to standard LLMs.
♻ ☆ Learning-based Observer for Coupled Disturbance
Achieving high-precision control for robotic systems is hindered by the low-fidelity dynamical model and external disturbances. Especially, the intricate coupling between internal uncertainties and external disturbances further exacerbates this challenge. This study introduces an effective and convergent algorithm enabling accurate estimation of the coupled disturbance via combining control and learning philosophies. Concretely, by resorting to Chebyshev series expansion, the coupled disturbance is firstly decomposed into an unknown parameter matrix and two known structures dependent on system state and external disturbance respectively. A regularized least squares algorithm is subsequently formalized to learn the parameter matrix using historical time-series data. Finally, a polynomial disturbance observer is specifically devised to achieve a high-precision estimation of the coupled disturbance by utilizing the learned portion. The proposed algorithm is evaluated through extensive simulations and real flight tests. We believe this work can offer a new pathway to integrate learning approaches into control frameworks for addressing longstanding challenges in robotic applications.
comment: 17 pages, 9 figures
♻ ☆ Velocity-free task-space regulator for robot manipulators with external disturbances
This paper addresses the problem of task-space robust regulation of robot manipulators subject to external disturbances. A velocity-free control law is proposed by combining the internal model principle and the passivity-based output-feedback control approach. The resulting controller not only ensures asymptotic convergence of the regulation error but also rejects unwanted external sinusoidal disturbances. The potential of the proposed method lies in its simplicity, intuitiveness, and straightforward gain selection criteria for the synthesis of multi-joint robot manipulator control systems.
♻ ☆ Walk along: An Experiment on Controlling the Mobile Robot 'Spot' with Voice and Gestures
Robots are becoming more capable and can autonomously perform tasks such as navigating between locations. However, human oversight remains crucial. This study compared two touchless methods for directing mobile robots: voice control and gesture control, to investigate the efficiency of the methods and the preference of users. We tested these methods in two conditions: one in which participants remained stationary and one in which they walked freely alongside the robot. We hypothesized that walking alongside the robot would result in higher intuitiveness ratings and improved task performance, based on the idea that walking promotes spatial alignment and reduces the effort required for mental rotation. In a 2x2 within-subject design, 218 participants guided the quadruped robot Spot along a circuitous route with multiple 90-degree turns using rotate left, rotate right, and walk forward commands. After each trial, participants rated the intuitiveness of the command mapping, while post-experiment interviews were used to gather the participants' preferences. Results showed that voice control combined with walking with Spot was the most favored and intuitive, whereas gesture control while standing caused confusion for left/right commands. Nevertheless, 29% of participants preferred gesture control, citing increased task engagement and visual congruence as reasons. An odometry-based analysis revealed that participants often followed behind Spot, particularly in the gesture control condition, when they were allowed to walk. In conclusion, voice control with walking produced the best outcomes. Improving physical ergonomics and adjusting gesture types could make gesture control more effective.
♻ ☆ Non-Prehensile Tool-Object Manipulation by Integrating LLM-Based Planning and Manoeuvrability-Driven Controls
Being able to use tools is a widely recognised indicator of intelligence across species. Humans, for instance, have demonstrated mastery of tool use for over two million years. The ability to use tools is invaluable as it extends an organism's reach and enhances its capacity to interact with objects and the environment. Being able to understand the geometric-mechanical relations between the tools-objects-environments allows certain species (e.g., apes and crows) to reach food in narrow constrained spaces. The same principles of physical augmentation and its associated non-prehensile manipulation capabilities also apply to robotic systems. For example, by instrumenting them with different types of end-effectors, robots can (in principle) dexterously interact (e.g., push and flip) with objects of various shapes and masses akin to its biological counterpart. However, developing this type of manipulation skill is still an open research problem. Furthermore, the complexity of planning tool-object manipulation tasks, particularly in coordinating the actions of dual-arm robots, presents significant challenges. To address these complexities, we propose integrating Large Language Models (LLMs) to assist in planning and executing these intricate manipulations, thereby enhancing the robot's ability to perform in diverse scenarios.
♻ ☆ Real-World Evaluation of two Cooperative Intersection Management Approaches
Cooperative maneuver planning promises to significantly improve traffic efficiency at unsignalized intersections by leveraging connected automated vehicles. Previous works on this topic have been mostly developed for completely automated traffic in a simple simulated environment. In contrast, our previously introduced planning approaches are specifically designed to handle real-world mixed traffic. The two methods are based on multi-scenario prediction and graph-based reinforcement learning, respectively. This is the first study to perform evaluations in a novel mixed traffic simulation framework as well as real-world drives with prototype connected automated vehicles in public traffic. The simulation features the same connected automated driving software stack as deployed on one of the automated vehicles. Our quantitative evaluations show that cooperative maneuver planning achieves a substantial reduction in crossing times and the number of stops. In a realistic environment with few automated vehicles, there are noticeable efficiency gains with only slightly increasing criticality metrics.
comment: M. Klimke and M. B. Mertens are both first authors with equal contribution. 10 pages, 9 figures, 3 tables, submitted to IEEE Intelligent Transportation Systems Magazine
♻ ☆ Towards Developing Socially Compliant Automated Vehicles: Advances, Expert Insights, and A Conceptual Framework
Automated Vehicles (AVs) hold promise for revolutionizing transportation by improving road safety, traffic efficiency, and overall mobility. Despite the steady advancement in high-level AVs in recent years, the transition to full automation entails a period of mixed traffic, where AVs of varying automation levels coexist with human-driven vehicles (HDVs). Making AVs socially compliant and understood by human drivers is expected to improve the safety and efficiency of mixed traffic. Thus, ensuring AVs' compatibility with HDVs and social acceptance is crucial for their successful and seamless integration into mixed traffic. However, research in this critical area of developing Socially Compliant AVs (SCAVs) remains sparse. This study carries out the first comprehensive scoping review to assess the current state of the art in developing SCAVs, identifying key concepts, methodological approaches, and research gaps. An informal expert interview was also conducted to discuss the literature review results and identify critical research gaps and expectations towards SCAVs. Based on the scoping review and expert interview input, a conceptual framework is proposed for the development of SCAVs. The conceptual framework is evaluated using an online survey targeting researchers, technicians, policymakers, and other relevant professionals worldwide. The survey results provide valuable validation and insights, affirming the significance of the proposed conceptual framework in tackling the challenges of integrating AVs into mixed-traffic environments. Additionally, future research perspectives and suggestions are discussed, contributing to the research and development agenda of SCAVs.
comment: 58 pages, 13 figures, accepted by the Journal of Communications in Transportation Research
♻ ☆ ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy
Vision-Language-Action (VLA) models have shown substantial potential in real-world robotic manipulation. However, fine-tuning these models through supervised learning struggles to achieve robust performance due to limited, inconsistent demonstrations, especially in contact-rich environments. In this paper, we propose a reinforced fine-tuning approach for VLA models, named ConRFT, which consists of offline and online fine-tuning with a unified consistency-based training objective, to address these challenges. In the offline stage, our method integrates behavior cloning and Q-learning to effectively extract policy from a small set of demonstrations and stabilize value estimating. In the online stage, the VLA model is further fine-tuned via consistency policy, with human interventions to ensure safe exploration and high sample efficiency. We evaluate our approach on eight diverse real-world manipulation tasks. It achieves an average success rate of 96.3% within 45-90 minutes of online fine-tuning, outperforming prior supervised methods with a 144% improvement in success rate and 1.9x shorter episode length. This work highlights the potential of integrating reinforcement learning to enhance the performance of VLA models for real-world robotic applications. Videos and code are available at our project website https://cccedric.github.io/conrft/.
♻ ☆ Deep learning framework for action prediction reveals multi-timescale locomotor control
Modeling human movement in real-world tasks is a fundamental goal for motor control, biomechanics, and rehabilitation engineering. However, existing models of essential tasks like locomotion are not applicable across varying terrain, mechanical conditions, and sensory contexts. This is at least in part due to simplifying assumptions like linear and fixed timescales mappings between inputs and future actions, which may not be broadly applicable. Here, we develop a deep learning-based framework for action prediction, outperforming traditional models across multiple contexts (walking and running, treadmill and overground, varying terrains) and input modalities (multiple body states, visual gaze). We find that neural network architectures with flexible input history-dependence, like GRU and Transformer, and with architecture-dependent trial embeddings perform best overall. By quantifying the model's predictions relative to an autoregressive baseline, we identify context- and modality-dependent timescales. These analyses reveal that there is greater reliance on fast-timescale predictions in complex terrain, gaze predicts future foot placement before body states, and the full-body state predictions precede those by center-of-mass states. This deep learning framework for human action prediction provides quantifiable insights into the control of real-world locomotion and can be extended to other actions, contexts, and populations.
♻ ☆ MambaXCTrack: Mamba-based Tracker with SSM Cross-correlation and Motion Prompt for Ultrasound Needle Tracking
Ultrasound (US)-guided needle insertion is widely employed in percutaneous interventions. However, providing feedback on the needle tip position via US imaging presents challenges due to noise, artifacts, and the thin imaging plane of US, which degrades needle features and leads to intermittent tip visibility. In this paper, a Mamba-based US needle tracker MambaXCTrack utilizing structured state space models cross-correlation (SSMX-Corr) and implicit motion prompt is proposed, which is the first application of Mamba in US needle tracking. The SSMX-Corr enhances cross-correlation by long-range modeling and global searching of distant semantic features between template and search maps, benefiting the tracking under noise and artifacts by implicitly learning potential distant semantic cues. By combining with cross-map interleaved scan (CIS), local pixel-wise interaction with positional inductive bias can also be introduced to SSMX-Corr. The implicit low-level motion descriptor is proposed as a non-visual prompt to enhance tracking robustness, addressing the intermittent tip visibility problem. Extensive experiments on a dataset with motorized needle insertion in both phantom and tissue samples demonstrate that the proposed tracker outperforms other state-of-the-art trackers while ablation studies further highlight the effectiveness of each proposed tracking module.
comment: Accepted by RAL
♻ ☆ Planning Shorter Paths in Graphs of Convex Sets by Undistorting Parametrized Configuration Spaces
Optimization based motion planning provides a useful modeling framework through various costs and constraints. Using Graph of Convex Sets (GCS) for trajectory optimization gives guarantees of feasibility and optimality by representing configuration space as the finite union of convex sets. Nonlinear parametrizations can be used to extend this technique to handle cases such as kinematic loops, but this distorts distances, such that solving with convex objectives will yield paths that are suboptimal in the original space. We present a method to extend GCS to nonconvex objectives, allowing us to "undistort" the optimization landscape while maintaining feasibility guarantees. We demonstrate our method's efficacy on three different robotic planning domains: a bimanual robot moving an object with both arms, the set of 3D rotations using Euler angles, and a rational parametrization of kinematics that enables certifying regions as collision free. Across the board, our method significantly improves path length and trajectory duration with only a minimal increase in runtime. Website: https://shrutigarg914.github.io/pgd-gcs-results/
comment: 8 pages, 6 figures, accepted to Robotics and Automation Letters in April 2025
♻ ☆ EgoEvGesture: Gesture Recognition Based on Egocentric Event Camera
Egocentric gesture recognition is a pivotal technology for enhancing natural human-computer interaction, yet traditional RGB-based solutions suffer from motion blur and illumination variations in dynamic scenarios. While event cameras show distinct advantages in handling high dynamic range with ultra-low power consumption, existing RGB-based architectures face inherent limitations in processing asynchronous event streams due to their synchronous frame-based nature. Moreover, from an egocentric perspective, event cameras record data that includes events generated by both head movements and hand gestures, thereby increasing the complexity of gesture recognition. To address this, we propose a novel network architecture specifically designed for event data processing, incorporating (1) a lightweight CNN with asymmetric depthwise convolutions to reduce parameters while preserving spatiotemporal features, (2) a plug-and-play state-space model as context block that decouples head movement noise from gesture dynamics, and (3) a parameter-free Bins-Temporal Shift Module (BSTM) that shifts features along bins and temporal dimensions to fuse sparse events efficiently. We further establish the EgoEvGesture dataset, the first large-scale dataset for egocentric gesture recognition using event cameras. Experimental results demonstrate that our method achieves 62.7% accuracy tested on unseen subjects with only 7M parameters, 3.1% higher than state-of-the-art approaches. Notable misclassifications in freestyle motions stem from high inter-personal variability and unseen test patterns differing from training data. Moreover, our approach achieved a remarkable accuracy of 97.0% on the DVS128 Gesture, demonstrating the effectiveness and generalization capability of our method on public datasets. The dataset and models are made available at https://github.com/3190105222/EgoEv_Gesture.
comment: The dataset and models are made available at https://github.com/3190105222/EgoEv_Gesture
♻ ☆ Constraint-Aware Zero-Shot Vision-Language Navigation in Continuous Environments
We address the task of Vision-Language Navigation in Continuous Environments (VLN-CE) under the zero-shot setting. Zero-shot VLN-CE is particularly challenging due to the absence of expert demonstrations for training and minimal environment structural prior to guide navigation. To confront these challenges, we propose a Constraint-Aware Navigator (CA-Nav), which reframes zero-shot VLN-CE as a sequential, constraint-aware sub-instruction completion process. CA-Nav continuously translates sub-instructions into navigation plans using two core modules: the Constraint-Aware Sub-instruction Manager (CSM) and the Constraint-Aware Value Mapper (CVM). CSM defines the completion criteria for decomposed sub-instructions as constraints and tracks navigation progress by switching sub-instructions in a constraint-aware manner. CVM, guided by CSM's constraints, generates a value map on the fly and refines it using superpixel clustering to improve navigation stability. CA-Nav achieves the state-of-the-art performance on two VLN-CE benchmarks, surpassing the previous best method by 12 percent and 13 percent in Success Rate on the validation unseen splits of R2R-CE and RxR-CE, respectively. Moreover, CA-Nav demonstrates its effectiveness in real-world robot deployments across various indoor scenes and instructions.
♻ ☆ Dynamic-Dark SLAM: RGB-Thermal Cooperative Robot Vision Strategy for Multi-Person Tracking in Both Well-Lit and Low-Light Scenes
In robot vision, thermal cameras hold great potential for recognizing humans even in complete darkness. However, their application to multi-person tracking (MPT) has been limited due to data scarcity and the inherent difficulty of distinguishing individuals. In this study, we propose a cooperative MPT system that utilizes co-located RGB and thermal cameras, where pseudo-annotations (bounding boxes and person IDs) are used to train both RGB and thermal trackers. Evaluation experiments demonstrate that the thermal tracker performs robustly in both bright and dark environments. Moreover, the results suggest that a tracker-switching strategy -- guided by a binary brightness classifier -- is more effective for information integration than a tracker-fusion approach. As an application example, we present an image change pattern recognition (ICPR) method, the ``human-as-landmark,'' which combines two key properties: the thermal recognizability of humans in dark environments and the rich landmark characteristics -- appearance, geometry, and semantics -- of static objects (occluders). Whereas conventional SLAM focuses on mapping static landmarks in well-lit environments, the present study takes a first step toward a new Human-Only SLAM paradigm, ``DD-SLAM,'' which aims to map even dynamic landmarks in complete darkness.
comment: 10 pages, 9 figures, technical report
♻ ☆ CurricuLLM: Automatic Task Curricula Design for Learning Complex Robot Skills using Large Language Models ICRA 2025
Curriculum learning is a training mechanism in reinforcement learning (RL) that facilitates the achievement of complex policies by progressively increasing the task difficulty during training. However, designing effective curricula for a specific task often requires extensive domain knowledge and human intervention, which limits its applicability across various domains. Our core idea is that large language models (LLMs), with their extensive training on diverse language data and ability to encapsulate world knowledge, present significant potential for efficiently breaking down tasks and decomposing skills across various robotics environments. Additionally, the demonstrated success of LLMs in translating natural language into executable code for RL agents strengthens their role in generating task curricula. In this work, we propose CurricuLLM, which leverages the high-level planning and programming capabilities of LLMs for curriculum design, thereby enhancing the efficient learning of complex target tasks. CurricuLLM consists of: (Step 1) Generating sequence of subtasks that aid target task learning in natural language form, (Step 2) Translating natural language description of subtasks in executable task code, including the reward code and goal distribution code, and (Step 3) Evaluating trained policies based on trajectory rollout and subtask description. We evaluate CurricuLLM in various robotics simulation environments, ranging from manipulation, navigation, and locomotion, to show that CurricuLLM can aid learning complex robot control tasks. In addition, we validate humanoid locomotion policy learned through CurricuLLM in real-world. Project website is https://iconlab.negarmehr.com/CurricuLLM/
comment: Accepted to ICRA 2025
♻ ☆ Flying Calligrapher: Contact-Aware Motion and Force Planning and Control for Aerial Manipulation ICRA 2025
Aerial manipulation has gained interest in completing high-altitude tasks that are challenging for human workers, such as contact inspection and defect detection, etc. Previous research has focused on maintaining static contact points or forces. This letter addresses a more general and dynamic task: simultaneously tracking time-varying contact force in the surface normal direction and motion trajectories on tangential surfaces. We propose a pipeline that includes a contact-aware trajectory planner to generate dynamically feasible trajectories, and a hybrid motion-force controller to track such trajectories. We demonstrate the approach in an aerial calligraphy task using a novel sponge pen design as the end-effector, whose stroke width is positively related to the contact force. Additionally, we develop a touchscreen interface for flexible user input. Experiments show our method can effectively draw diverse letters, achieving an IoU of 0.59 and an end-effector position (force) tracking RMSE of 2.9 cm (0.7 N). Website: https://xiaofeng-guo.github.io/flying-calligrapher/
comment: 8 pages, 9 figures, 1 table. This paper is accepted by RAL and is presented at ICRA 2025
Artificial Intelligence 150
☆ Weight Ensembling Improves Reasoning in Language Models
We investigate a failure mode that arises during the training of reasoning models, where the diversity of generations begins to collapse, leading to suboptimal test-time scaling. Notably, the Pass@1 rate reliably improves during supervised finetuning (SFT), but Pass@k rapidly deteriorates. Surprisingly, a simple intervention of interpolating the weights of the latest SFT checkpoint with an early checkpoint, otherwise known as WiSE-FT, almost completely recovers Pass@k while also improving Pass@1. The WiSE-FT variant achieves better test-time scaling (Best@k, majority vote) and achieves superior results with less data when tuned further by reinforcement learning. Finally, we find that WiSE-FT provides complementary performance gains that cannot be achieved only through diversity-inducing decoding strategies, like temperature scaling. We formalize a bias-variance tradeoff of Pass@k with respect to the expectation and variance of Pass@1 over the test distribution. We find that WiSE-FT can reduce bias and variance simultaneously, while temperature scaling inherently trades-off between bias and variance.
☆ RealWebAssist: A Benchmark for Long-Horizon Web Assistance with Real-World Users SC
To achieve successful assistance with long-horizon web-based tasks, AI agents must be able to sequentially follow real-world user instructions over a long period. Unlike existing web-based agent benchmarks, sequential instruction following in the real world poses significant challenges beyond performing a single, clearly defined task. For instance, real-world human instructions can be ambiguous, require different levels of AI assistance, and may evolve over time, reflecting changes in the user's mental state. To address this gap, we introduce RealWebAssist, a novel benchmark designed to evaluate sequential instruction-following in realistic scenarios involving long-horizon interactions with the web, visual GUI grounding, and understanding ambiguous real-world user instructions. RealWebAssist includes a dataset of sequential instructions collected from real-world human users. Each user instructs a web-based assistant to perform a series of tasks on multiple websites. A successful agent must reason about the true intent behind each instruction, keep track of the mental state of the user, understand user-specific routines, and ground the intended tasks to actions on the correct GUI elements. Our experimental results show that state-of-the-art models struggle to understand and ground user instructions, posing critical challenges in following real-world user instructions for long-horizon web assistance.
comment: Project Website: https://scai.cs.jhu.edu/projects/RealWebAssist/ Code: https://github.com/SCAI-JHU/RealWebAssist
☆ Multimodal Long Video Modeling Based on Temporal Dynamic Context
Recent advances in Large Language Models (LLMs) have led to significant breakthroughs in video understanding. However, existing models still struggle with long video processing due to the context length constraint of LLMs and the vast amount of information within the video. Although some recent methods are designed for long video understanding, they often lose crucial information during token compression and struggle with additional modality like audio. In this work, we propose a dynamic long video encoding method utilizing the temporal relationship between frames, named Temporal Dynamic Context (TDC). Firstly, we segment the video into semantically consistent scenes based on inter-frame similarities, then encode each frame into tokens using visual-audio encoders. Secondly, we propose a novel temporal context compressor to reduce the number of tokens within each segment. Specifically, we employ a query-based Transformer to aggregate video, audio, and instruction text tokens into a limited set of temporal context tokens. Finally, we feed the static frame tokens and the temporal context tokens into the LLM for video understanding. Furthermore, to handle extremely long videos, we propose a training-free chain-of-thought strategy that progressively extracts answers from multiple video segments. These intermediate answers serve as part of the reasoning process and contribute to the final answer. We conduct extensive experiments on general video understanding and audio-video understanding benchmarks, where our method demonstrates strong performance. The code and models are available at https://github.com/Hoar012/TDC-Video.
☆ LLM Can be a Dangerous Persuader: Empirical Study of Persuasion Safety in Large Language Models
Recent advancements in Large Language Models (LLMs) have enabled them to approach human-level persuasion capabilities. However, such potential also raises concerns about the safety risks of LLM-driven persuasion, particularly their potential for unethical influence through manipulation, deception, exploitation of vulnerabilities, and many other harmful tactics. In this work, we present a systematic investigation of LLM persuasion safety through two critical aspects: (1) whether LLMs appropriately reject unethical persuasion tasks and avoid unethical strategies during execution, including cases where the initial persuasion goal appears ethically neutral, and (2) how influencing factors like personality traits and external pressures affect their behavior. To this end, we introduce PersuSafety, the first comprehensive framework for the assessment of persuasion safety which consists of three stages, i.e., persuasion scene creation, persuasive conversation simulation, and persuasion safety assessment. PersuSafety covers 6 diverse unethical persuasion topics and 15 common unethical strategies. Through extensive experiments across 8 widely used LLMs, we observe significant safety concerns in most LLMs, including failing to identify harmful persuasion tasks and leveraging various unethical persuasion strategies. Our study calls for more attention to improve safety alignment in progressive and goal-driven conversations such as persuasion.
comment: 20 pages, 7 figures, 4 tables
☆ Can We Edit LLMs for Long-Tail Biomedical Knowledge?
Knowledge editing has emerged as an effective approach for updating large language models (LLMs) by modifying their internal knowledge. However, their application to the biomedical domain faces unique challenges due to the long-tailed distribution of biomedical knowledge, where rare and infrequent information is prevalent. In this paper, we conduct the first comprehensive study to investigate the effectiveness of knowledge editing methods for editing long-tail biomedical knowledge. Our results indicate that, while existing editing methods can enhance LLMs' performance on long-tail biomedical knowledge, their performance on long-tail knowledge remains inferior to that on high-frequency popular knowledge, even after editing. Our further analysis reveals that long-tail biomedical knowledge contains a significant amount of one-to-many knowledge, where one subject and relation link to multiple objects. This high prevalence of one-to-many knowledge limits the effectiveness of knowledge editing in improving LLMs' understanding of long-tail biomedical knowledge, highlighting the need for tailored strategies to bridge this performance gap.
☆ LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models
Scientific equation discovery is a fundamental task in the history of scientific progress, enabling the derivation of laws governing natural phenomena. Recently, Large Language Models (LLMs) have gained interest for this task due to their potential to leverage embedded scientific knowledge for hypothesis generation. However, evaluating the true discovery capabilities of these methods remains challenging, as existing benchmarks often rely on common equations that are susceptible to memorization by LLMs, leading to inflated performance metrics that do not reflect discovery. In this paper, we introduce LLM-SRBench, a comprehensive benchmark with 239 challenging problems across four scientific domains specifically designed to evaluate LLM-based scientific equation discovery methods while preventing trivial memorization. Our benchmark comprises two main categories: LSR-Transform, which transforms common physical models into less common mathematical representations to test reasoning beyond memorized forms, and LSR-Synth, which introduces synthetic, discovery-driven problems requiring data-driven reasoning. Through extensive evaluation of several state-of-the-art methods, using both open and closed LLMs, we find that the best-performing system so far achieves only 31.5% symbolic accuracy. These findings highlight the challenges of scientific equation discovery, positioning LLM-SRBench as a valuable resource for future research.
comment: Project page: https://github.com/deep-symbolic-mathematics/llm-srbench , Benchmark page: https://huggingface.co/datasets/nnheui/llm-srbench
☆ AI-Driven Code Refactoring: Using Graph Neural Networks to Enhance Software Maintainability
This study explores Graph Neural Networks (GNNs) as a transformative tool for code refactoring, using abstract syntax trees (ASTs) to boost software maintainability. It analyzes a dataset of 2 million snippets from CodeSearchNet and a custom 75000-file GitHub Python corpus, comparing GNNs against rule-based SonarQube and decision trees. Metrics include cyclomatic complexity (target below 10), coupling (target below 5), and refactoring precision. GNNs achieve 92% accuracy, reducing complexity by 35% and coupling by 33%, outperforming SonarQube (78%, 16%) and decision trees (85%, 25%). Preprocessing fixed 60% of syntax errors. Bar graphs, tables, and AST visuals clarify results. This offers a scalable AI-driven path to cleaner codebases, which is crucial for software engineering.
☆ Performance of Large Language Models in Supporting Medical Diagnosis and Treatment
The integration of Large Language Models (LLMs) into healthcare holds significant potential to enhance diagnostic accuracy and support medical treatment planning. These AI-driven systems can analyze vast datasets, assisting clinicians in identifying diseases, recommending treatments, and predicting patient outcomes. This study evaluates the performance of a range of contemporary LLMs, including both open-source and closed-source models, on the 2024 Portuguese National Exam for medical specialty access (PNA), a standardized medical knowledge assessment. Our results highlight considerable variation in accuracy and cost-effectiveness, with several models demonstrating performance exceeding human benchmarks for medical students on this specific task. We identify leading models based on a combined score of accuracy and cost, discuss the implications of reasoning methodologies like Chain-of-Thought, and underscore the potential for LLMs to function as valuable complementary tools aiding medical professionals in complex clinical decision-making.
comment: 21 pages, 6 figures, 4 tables. Acknowledgements: The authors acknowledge the support of the AITriage4SU Project (2024.07400.IACDC/2024), funded by the FCT (Foundation for Science and Technology), Portugal
☆ Can LLMs Assist Expert Elicitation for Probabilistic Causal Modeling?
Objective: This study investigates the potential of Large Language Models (LLMs) as an alternative to human expert elicitation for extracting structured causal knowledge and facilitating causal modeling in biometric and healthcare applications. Material and Methods: LLM-generated causal structures, specifically Bayesian networks (BNs), were benchmarked against traditional statistical methods (e.g., Bayesian Information Criterion) using healthcare datasets. Validation techniques included structural equation modeling (SEM) to verifying relationships, and measures such as entropy, predictive accuracy, and robustness to compare network structures. Results and Discussion: LLM-generated BNs demonstrated lower entropy than expert-elicited and statistically generated BNs, suggesting higher confidence and precision in predictions. However, limitations such as contextual constraints, hallucinated dependencies, and potential biases inherited from training data require further investigation. Conclusion: LLMs represent a novel frontier in expert elicitation for probabilistic causal modeling, promising to improve transparency and reduce uncertainty in the decision-making using such models.
☆ Teacher Motion Priors: Enhancing Robot Locomotion over Challenging Terrain
Achieving robust locomotion on complex terrains remains a challenge due to high dimensional control and environmental uncertainties. This paper introduces a teacher prior framework based on the teacher student paradigm, integrating imitation and auxiliary task learning to improve learning efficiency and generalization. Unlike traditional paradigms that strongly rely on encoder-based state embeddings, our framework decouples the network design, simplifying the policy network and deployment. A high performance teacher policy is first trained using privileged information to acquire generalizable motion skills. The teacher's motion distribution is transferred to the student policy, which relies only on noisy proprioceptive data, via a generative adversarial mechanism to mitigate performance degradation caused by distributional shifts. Additionally, auxiliary task learning enhances the student policy's feature representation, speeding up convergence and improving adaptability to varying terrains. The framework is validated on a humanoid robot, showing a great improvement in locomotion stability on dynamic terrains and significant reductions in development costs. This work provides a practical solution for deploying robust locomotion strategies in humanoid robots.
comment: 8 pages, 6 figures, 6 tables
☆ SymRTLO: Enhancing RTL Code Optimization with LLMs and Neuron-Inspired Symbolic Reasoning
Optimizing Register Transfer Level (RTL) code is crucial for improving the power, performance, and area (PPA) of digital circuits in the early stages of synthesis. Manual rewriting, guided by synthesis feedback, can yield high-quality results but is time-consuming and error-prone. Most existing compiler-based approaches have difficulty handling complex design constraints. Large Language Model (LLM)-based methods have emerged as a promising alternative to address these challenges. However, LLM-based approaches often face difficulties in ensuring alignment between the generated code and the provided prompts. This paper presents SymRTLO, a novel neuron-symbolic RTL optimization framework that seamlessly integrates LLM-based code rewriting with symbolic reasoning techniques. Our method incorporates a retrieval-augmented generation (RAG) system of optimization rules and Abstract Syntax Tree (AST)-based templates, enabling LLM-based rewriting that maintains syntactic correctness while minimizing undesired circuit behaviors. A symbolic module is proposed for analyzing and optimizing finite state machine (FSM) logic, allowing fine-grained state merging and partial specification handling beyond the scope of pattern-based compilers. Furthermore, a fast verification pipeline, combining formal equivalence checks with test-driven validation, further reduces the complexity of verification. Experiments on the RTL-Rewriter benchmark with Synopsys Design Compiler and Yosys show that SymRTLO improves power, performance, and area (PPA) by up to 43.9%, 62.5%, and 51.1%, respectively, compared to the state-of-the-art methods.
comment: 16 pages, 8 figures, 7 tables. Under Review
☆ S1-Bench: A Simple Benchmark for Evaluating System 1 Thinking Capability of Large Reasoning Models
We introduce S1-Bench, a novel benchmark designed to evaluate Large Reasoning Models' (LRMs) performance on simple tasks that favor intuitive system 1 thinking rather than deliberative system 2 reasoning. While LRMs have achieved significant breakthroughs in complex reasoning tasks through explicit chains of thought, their reliance on deep analytical thinking may limit their system 1 thinking capabilities. Moreover, a lack of benchmark currently exists to evaluate LRMs' performance in tasks that require such capabilities. To fill this gap, S1-Bench presents a set of simple, diverse, and naturally clear questions across multiple domains and languages, specifically designed to assess LRMs' performance in such tasks. Our comprehensive evaluation of 22 LRMs reveals significant lower efficiency tendencies, with outputs averaging 15.5 times longer than those of traditional small LLMs. Additionally, LRMs often identify correct answers early but continue unnecessary deliberation, with some models even producing numerous errors. These findings highlight the rigid reasoning patterns of current LRMs and underscore the substantial development needed to achieve balanced dual-system thinking capabilities that can adapt appropriately to task complexity.
comment: Work in Progress
☆ FingER: Content Aware Fine-grained Evaluation with Reasoning for AI-Generated Videos
Recent advances in video generation have posed great challenges in the assessment of AI-generated content, particularly with the emergence of increasingly sophisticated models. The various inconsistencies and defects observed in such videos are inherently complex, making overall scoring notoriously difficult. In this paper, we emphasize the critical importance of integrating fine-grained reasoning into video evaluation, and we propose $\textbf{F}$ing$\textbf{ER}$, a novel entity-level reasoning evaluation framework that first automatically generates $\textbf{F}$ine-grained $\textbf{E}$ntity-level questions, and then answers those questions by a $\textbf{R}$easoning model with scores, which can be subsequently weighted summed to an overall score for different applications. Specifically, we leverage LLMs to derive entity-level questions across five distinct perspectives, which (i) often focus on some specific entities of the content, thereby making answering or scoring much easier by MLLMs, and (ii) are more interpretable. Then we construct a FingER dataset, consisting of approximately 3.3k videos and corresponding 60k fine-grained QA annotations, each with detailed reasons. Based on that, we further investigate various training protocols to best incentivize the reasoning capability of MLLMs for correct answer prediction. Extensive experiments demonstrate that a reasoning model trained using Group Relative Policy Optimization (GRPO) with a cold-start strategy achieves the best performance. Notably, our model surpasses existing methods by a relative margin of $11.8\%$ on GenAI-Bench and $5.5\%$ on MonetBench with only 3.3k training videos, which is at most one-tenth of the training samples utilized by other methods. Our code and dataset will be released soon.
comment: 10 pages, 4 figures
☆ Forecasting from Clinical Textual Time Series: Adaptations of the Encoder and Decoder Language Model Families
Clinical case reports encode rich, temporal patient trajectories that are often underexploited by traditional machine learning methods relying on structured data. In this work, we introduce the forecasting problem from textual time series, where timestamped clinical findings--extracted via an LLM-assisted annotation pipeline--serve as the primary input for prediction. We systematically evaluate a diverse suite of models, including fine-tuned decoder-based large language models and encoder-based transformers, on tasks of event occurrence prediction, temporal ordering, and survival analysis. Our experiments reveal that encoder-based models consistently achieve higher F1 scores and superior temporal concordance for short- and long-horizon event forecasting, while fine-tuned masking approaches enhance ranking performance. In contrast, instruction-tuned decoder models demonstrate a relative advantage in survival analysis, especially in early prognosis settings. Our sensitivity analyses further demonstrate the importance of time ordering, which requires clinical time series construction, as compared to text ordering, the format of the text inputs that LLMs are classically trained on. This highlights the additional benefit that can be ascertained from time-ordered corpora, with implications for temporal tasks in the era of widespread LLM use.
comment: Machine Learning for Healthcare (MLHC 2025)
☆ Heimdall: test-time scaling on the generative verification
An AI system can create and maintain knowledge only to the extent that it can verify that knowledge itself. Recent work on long Chain-of-Thought reasoning has demonstrated great potential of LLMs on solving competitive problems, but their verification ability remains to be weak and not sufficiently investigated. In this paper, we propose Heimdall, the long CoT verification LLM that can accurately judge the correctness of solutions. With pure reinforcement learning, we boost the verification accuracy from 62.5% to 94.5% on competitive math problems. By scaling with repeated sampling, the accuracy further increases to 97.5%. Through human evaluation, Heimdall demonstrates impressive generalization capabilities, successfully detecting most issues in challenging math proofs, the type of which is not included during training. Furthermore, we propose Pessimistic Verification to extend the functionality of Heimdall to scaling up the problem solving. It calls Heimdall to judge the solutions from a solver model and based on the pessimistic principle, selects the most likely correct solution with the least uncertainty. Taking DeepSeek-R1-Distill-Qwen-32B as the solver model, Pessimistic Verification improves the solution accuracy on AIME2025 from 54.2% to 70.0% with 16x compute budget and to 83.3% with more compute budget. With the stronger solver Gemini 2.5 Pro, the score reaches 93.0%. Finally, we prototype an automatic knowledge discovery system, a ternary system where one poses questions, another provides solutions, and the third verifies the solutions. Using the data synthesis work NuminaMath for the first two components, Heimdall effectively identifies problematic records within the dataset and reveals that nearly half of the data is flawed, which interestingly aligns with the recent ablation studies from NuminaMath.
☆ AlayaDB: The Data Foundation for Efficient and Effective Long-context LLM Inference
AlayaDB is a cutting-edge vector database system natively architected for efficient and effective long-context inference for Large Language Models (LLMs) at AlayaDB AI. Specifically, it decouples the KV cache and attention computation from the LLM inference systems, and encapsulates them into a novel vector database system. For the Model as a Service providers (MaaS), AlayaDB consumes fewer hardware resources and offers higher generation quality for various workloads with different kinds of Service Level Objectives (SLOs), when comparing with the existing alternative solutions (e.g., KV cache disaggregation, retrieval-based sparse attention). The crux of AlayaDB is that it abstracts the attention computation and cache management for LLM inference into a query processing procedure, and optimizes the performance via a native query optimizer. In this work, we demonstrate the effectiveness of AlayaDB via (i) three use cases from our industry partners, and (ii) extensive experimental results on LLM inference benchmarks.
comment: 14 pages, 12 figures, conference
☆ AutoStyle-TTS: Retrieval-Augmented Generation based Automatic Style Matching Text-to-Speech Synthesis ICME25
With the advancement of speech synthesis technology, users have higher expectations for the naturalness and expressiveness of synthesized speech. But previous research ignores the importance of prompt selection. This study proposes a text-to-speech (TTS) framework based on Retrieval-Augmented Generation (RAG) technology, which can dynamically adjust the speech style according to the text content to achieve more natural and vivid communication effects. We have constructed a speech style knowledge database containing high-quality speech samples in various contexts and developed a style matching scheme. This scheme uses embeddings, extracted by Llama, PER-LLM-Embedder,and Moka, to match with samples in the knowledge database, selecting the most appropriate speech style for synthesis. Furthermore, our empirical research validates the effectiveness of the proposed method. Our demo can be viewed at: https://thuhcsi.github.io/icme2025-AutoStyle-TTS
comment: accepted by ICME25
☆ Characterizing LLM-driven Social Network: The Chirper.ai Case
Large language models (LLMs) demonstrate the ability to simulate human decision-making processes, enabling their use as agents in modeling sophisticated social networks, both offline and online. Recent research has explored collective behavioral patterns and structural characteristics of LLM agents within simulated networks. However, empirical comparisons between LLM-driven and human-driven online social networks remain scarce, limiting our understanding of how LLM agents differ from human users. This paper presents a large-scale analysis of Chirper.ai, an X/Twitter-like social network entirely populated by LLM agents, comprising over 65,000 agents and 7.7 million AI-generated posts. For comparison, we collect a parallel dataset from Mastodon, a human-driven decentralized social network, with over 117,000 users and 16 million posts. We examine key differences between LLM agents and humans in posting behaviors, abusive content, and social network structures. Our findings provide critical insights into the evolving landscape of online social network analysis in the AI era, offering a comprehensive profile of LLM agents in social simulations.
comment: Work in progress
☆ Zero-shot Autonomous Microscopy for Scalable and Intelligent Characterization of 2D Materials
Characterization of atomic-scale materials traditionally requires human experts with months to years of specialized training. Even for trained human operators, accurate and reliable characterization remains challenging when examining newly discovered materials such as two-dimensional (2D) structures. This bottleneck drives demand for fully autonomous experimentation systems capable of comprehending research objectives without requiring large training datasets. In this work, we present ATOMIC (Autonomous Technology for Optical Microscopy & Intelligent Characterization), an end-to-end framework that integrates foundation models to enable fully autonomous, zero-shot characterization of 2D materials. Our system integrates the vision foundation model (i.e., Segment Anything Model), large language models (i.e., ChatGPT), unsupervised clustering, and topological analysis to automate microscope control, sample scanning, image segmentation, and intelligent analysis through prompt engineering, eliminating the need for additional training. When analyzing typical MoS2 samples, our approach achieves 99.7% segmentation accuracy for single layer identification, which is equivalent to that of human experts. In addition, the integrated model is able to detect grain boundary slits that are challenging to identify with human eyes. Furthermore, the system retains robust accuracy despite variable conditions including defocus, color temperature fluctuations, and exposure variations. It is applicable to a broad spectrum of common 2D materials-including graphene, MoS2, WSe2, SnSe-regardless of whether they were fabricated via chemical vapor deposition or mechanical exfoliation. This work represents the implementation of foundation models to achieve autonomous analysis, establishing a scalable and data-efficient characterization paradigm that fundamentally transforms the approach to nanoscale materials research.
comment: 13 pages, 4 figures
☆ RealHarm: A Collection of Real-World Language Model Application Failures
Language model deployments in consumer-facing applications introduce numerous risks. While existing research on harms and hazards of such applications follows top-down approaches derived from regulatory frameworks and theoretical analyses, empirical evidence of real-world failure modes remains underexplored. In this work, we introduce RealHarm, a dataset of annotated problematic interactions with AI agents built from a systematic review of publicly reported incidents. Analyzing harms, causes, and hazards specifically from the deployer's perspective, we find that reputational damage constitutes the predominant organizational harm, while misinformation emerges as the most common hazard category. We empirically evaluate state-of-the-art guardrails and content moderation systems to probe whether such systems would have prevented the incidents, revealing a significant gap in the protection of AI applications.
☆ Vision based driving agent for race car simulation environments
In recent years, autonomous driving has become a popular field of study. As control at tire grip limit is essential during emergency situations, algorithms developed for racecars are useful for road cars too. This paper examines the use of Deep Reinforcement Learning (DRL) to solve the problem of grip limit driving in a simulated environment. Proximal Policy Optimization (PPO) method is used to train an agent to control the steering wheel and pedals of the vehicle, using only visual inputs to achieve professional human lap times. The paper outlines the formulation of the task of time optimal driving on a race track as a deep reinforcement learning problem, and explains the chosen observations, actions, and reward functions. The results demonstrate human-like learning and driving behavior that utilize maximum tire grip potential.
comment: Submitted to ICMCE 2024 (https://icmce.org/2024.html)
☆ MASSeg : 2nd Technical Report for 4th PVUW MOSE Track
Complex video object segmentation continues to face significant challenges in small object recognition, occlusion handling, and dynamic scene modeling. This report presents our solution, which ranked second in the MOSE track of CVPR 2025 PVUW Challenge. Based on an existing segmentation framework, we propose an improved model named MASSeg for complex video object segmentation, and construct an enhanced dataset, MOSE+, which includes typical scenarios with occlusions, cluttered backgrounds, and small target instances. During training, we incorporate a combination of inter-frame consistent and inconsistent data augmentation strategies to improve robustness and generalization. During inference, we design a mask output scaling strategy to better adapt to varying object sizes and occlusion levels. As a result, MASSeg achieves a J score of 0.8250, F score of 0.9007, and a J&F score of 0.8628 on the MOSE test set.
comment: 5 pages,4 figures,Technical report on Complex Video Object Segmentation
☆ Can Competition Enhance the Proficiency of Agents Powered by Large Language Models in the Realm of News-driven Time Series Forecasting?
Multi-agents-based news-driven time series forecasting is considered as a potential paradigm shift in the era of large language models (LLMs). The challenge of this task lies in measuring the influences of different news events towards the fluctuations of time series. This requires agents to possess stronger abilities of innovative thinking and the identifying misleading logic. However, the existing multi-agent discussion framework has limited enhancement on time series prediction in terms of optimizing these two capabilities. Inspired by the role of competition in fostering innovation, this study embeds a competition mechanism within the multi-agent discussion to enhance agents' capability of generating innovative thoughts. Furthermore, to bolster the model's proficiency in identifying misleading information, we incorporate a fine-tuned small-scale LLM model within the reflective stage, offering auxiliary decision-making support. Experimental results confirm that the competition can boost agents' capacity for innovative thinking, which can significantly improve the performances of time series prediction. Similar to the findings of social science, the intensity of competition within this framework can influence the performances of agents, providing a new perspective for studying LLMs-based multi-agent systems.
☆ Localized Cultural Knowledge is Conserved and Controllable in Large Language Models
Just as humans display language patterns influenced by their native tongue when speaking new languages, LLMs often default to English-centric responses even when generating in other languages. Nevertheless, we observe that local cultural information persists within the models and can be readily activated for cultural customization. We first demonstrate that explicitly providing cultural context in prompts significantly improves the models' ability to generate culturally localized responses. We term the disparity in model performance with versus without explicit cultural context the explicit-implicit localization gap, indicating that while cultural knowledge exists within LLMs, it may not naturally surface in multilingual interactions if cultural context is not explicitly provided. Despite the explicit prompting benefit, however, the answers reduce in diversity and tend toward stereotypes. Second, we identify an explicit cultural customization vector, conserved across all non-English languages we explore, which enables LLMs to be steered from the synthetic English cultural world-model toward each non-English cultural world. Steered responses retain the diversity of implicit prompting and reduce stereotypes to dramatically improve the potential for customization. We discuss the implications of explicit cultural customization for understanding the conservation of alternative cultural world models within LLMs, and their controllable utility for translation, cultural customization, and the possibility of making the explicit implicit through soft control for expanded LLM function and appeal.
☆ Efficient Generative Model Training via Embedded Representation Warmup
Diffusion models excel at generating high-dimensional data but fall short in training efficiency and representation quality compared to self-supervised methods. We identify a key bottleneck: the underutilization of high-quality, semantically rich representations during training notably slows down convergence. Our systematic analysis reveals a critical representation processing region -- primarily in the early layers -- where semantic and structural pattern learning takes place before generation can occur. To address this, we propose Embedded Representation Warmup (ERW), a plug-and-play framework where in the first stage we get the ERW module serves as a warmup that initializes the early layers of the diffusion model with high-quality, pretrained representations. This warmup minimizes the burden of learning representations from scratch, thereby accelerating convergence and boosting performance. Our theoretical analysis demonstrates that ERW's efficacy depends on its precise integration into specific neural network layers -- termed the representation processing region -- where the model primarily processes and transforms feature representations for later generation. We further establish that ERW not only accelerates training convergence but also enhances representation quality: empirically, our method achieves a 40$\times$ acceleration in training speed compared to REPA, the current state-of-the-art methods. Code is available at https://github.com/LINs-lab/ERW.
☆ Deep Reasoning Translation via Reinforcement Learning
Recently, deep reasoning LLMs (e.g., OpenAI o1/o3 and DeepSeek-R1) have shown promising performance in various complex tasks. Free translation is an important and interesting task in the multilingual world, which requires going beyond word-for-word translation and taking cultural differences into account. This task is still under-explored in deep reasoning LLMs. In this paper, we introduce DeepTrans, a deep reasoning translation model that learns free translation via reinforcement learning. Specifically, we carefully build a reward model with pre-defined scoring criteria on both the translation results and the thought process. Given the source sentences, the reward model teaches the deep translation model how to think and free-translate them during reinforcement learning. In this way, training DeepTrans does not need any labeled translations, avoiding the human-intensive annotation or resource-intensive data synthesis. Experimental results show the effectiveness of DeepTrans. Using Qwen2.5-7B as the backbone, DeepTrans improves performance by 16.3% in literature translation, and outperforms strong deep reasoning baselines as well as baselines that are fine-tuned with synthesized data. Moreover, we summarize the failures and interesting findings during our RL exploration. We hope this work could inspire other researchers in free translation.
☆ LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks
Large language model unlearning has become a critical challenge in ensuring safety and controlled model behavior by removing undesired data-model influences from the pretrained model while preserving general utility. Significant recent efforts have been dedicated to developing LLM unlearning benchmarks such as WMDP (Weapons of Mass Destruction Proxy) and MUSE (Machine Unlearning Six-way Evaluation), facilitating standardized unlearning performance assessment and method comparison. Despite their usefulness, we uncover for the first time a novel coreset effect within these benchmarks. Specifically, we find that LLM unlearning achieved with the original (full) forget set can be effectively maintained using a significantly smaller subset (functioning as a "coreset"), e.g., as little as 5% of the forget set, even when selected at random. This suggests that LLM unlearning in these benchmarks can be performed surprisingly easily, even in an extremely low-data regime. We demonstrate that this coreset effect remains strong, regardless of the LLM unlearning method used, such as NPO (Negative Preference Optimization) and RMU (Representation Misdirection Unlearning), the popular ones in these benchmarks. The surprisingly strong coreset effect is also robust across various data selection methods, ranging from random selection to more sophisticated heuristic approaches. We explain the coreset effect in LLM unlearning through a keyword-based perspective, showing that keywords extracted from the forget set alone contribute significantly to unlearning effectiveness and indicating that current unlearning is driven by a compact set of high-impact tokens rather than the entire dataset. We further justify the faithfulness of coreset-unlearned models along additional dimensions, such as mode connectivity and robustness to jailbreaking attacks. Codes are available at https://github.com/OPTML-Group/MU-Coreset.
☆ The Future of MLLM Prompting is Adaptive: A Comprehensive Experimental Evaluation of Prompt Engineering Methods for Robust Multimodal Performance
Multimodal Large Language Models (MLLMs) are set to transform how machines process and generate human-like responses by integrating diverse modalities such as text, images, and code. Yet, effectively harnessing their capabilities hinges on optimal prompt engineering. We present a comprehensive experimental evaluation of seven prompt engineering methods applied to 13 open-source MLLMs over 24 tasks spanning Reasoning and Compositionality, Multimodal Understanding and Alignment, Complex Code Generation and Execution, and Knowledge Retrieval and Integration. Our approach stratifies models by parameter count into Small (<4B), Medium (4B-10B), and Large (>10B) categories and compares prompting techniques including Zero-Shot, One-Shot, Few-Shot, Chain-of-Thought, Analogical, Generated Knowledge, and Tree-of-Thought. While Large MLLMs excel in structured tasks such as code generation, achieving accuracies up to 96.88% under Few-Shot prompting, all models struggle with complex reasoning and abstract understanding, often yielding accuracies below 60% and high hallucination rates. Structured reasoning prompts frequently increased hallucination up to 75% in small models and led to longer response times (over 20 seconds in Large MLLMs), while simpler prompting methods provided more concise and efficient outputs. No single prompting method uniformly optimises all task types. Instead, adaptive strategies combining example-based guidance with selective structured reasoning are essential to enhance robustness, efficiency, and factual accuracy. Our findings offer practical recommendations for prompt engineering and support more reliable deployment of MLLMs across applications including AI-assisted coding, knowledge retrieval, and multimodal content understanding.
☆ HalluSearch at SemEval-2025 Task 3: A Search-Enhanced RAG Pipeline for Hallucination Detection
In this paper, we present HalluSearch, a multilingual pipeline designed to detect fabricated text spans in Large Language Model (LLM) outputs. Developed as part of Mu-SHROOM, the Multilingual Shared-task on Hallucinations and Related Observable Overgeneration Mistakes, HalluSearch couples retrieval-augmented verification with fine-grained factual splitting to identify and localize hallucinations in fourteen different languages. Empirical evaluations show that HalluSearch performs competitively, placing fourth in both English (within the top ten percent) and Czech. While the system's retrieval-based strategy generally proves robust, it faces challenges in languages with limited online coverage, underscoring the need for further research to ensure consistent hallucination detection across diverse linguistic contexts.
☆ C-FAITH: A Chinese Fine-Grained Benchmark for Automated Hallucination Evaluation
Despite the rapid advancement of large language models, they remain highly susceptible to generating hallucinations, which significantly hinders their widespread application. Hallucination research requires dynamic and fine-grained evaluation. However, most existing hallucination benchmarks (especially in Chinese language) rely on human annotations, making automatical and cost-effective hallucination evaluation challenging. To address this, we introduce HaluAgent, an agentic framework that automatically constructs fine-grained QA dataset based on some knowledge documents. Our experiments demonstrate that the manually designed rules and prompt optimization can improve the quality of generated data. Using HaluAgent, we construct C-FAITH, a Chinese QA hallucination benchmark created from 1,399 knowledge documents obtained from web scraping, totaling 60,702 entries. We comprehensively evaluate 16 mainstream LLMs with our proposed C-FAITH, providing detailed experimental results and analysis.
☆ WildLive: Near Real-time Visual Wildlife Tracking onboard UAVs
Live tracking of wildlife via high-resolution video processing directly onboard drones is widely unexplored and most existing solutions rely on streaming video to ground stations to support navigation. Yet, both autonomous animal-reactive flight control beyond visual line of sight and/or mission-specific individual and behaviour recognition tasks rely to some degree on this capability. In response, we introduce WildLive -- a near real-time animal detection and tracking framework for high-resolution imagery running directly onboard uncrewed aerial vehicles (UAVs). The system performs multi-animal detection and tracking at 17fps+ for HD and 7fps+ on 4K video streams suitable for operation during higher altitude flights to minimise animal disturbance. Our system is optimised for Jetson Orin AGX onboard hardware. It integrates the efficiency of sparse optical flow tracking and mission-specific sampling with device-optimised and proven YOLO-driven object detection and segmentation techniques. Essentially, computational resource is focused onto spatio-temporal regions of high uncertainty to significantly improve UAV processing speeds without domain-specific loss of accuracy. Alongside, we introduce our WildLive dataset, which comprises 200k+ annotated animal instances across 19k+ frames from 4K UAV videos collected at the Ol Pejeta Conservancy in Kenya. All frames contain ground truth bounding boxes, segmentation masks, as well as individual tracklets and tracking point trajectories. We compare our system against current object tracking approaches including OC-SORT, ByteTrack, and SORT. Our multi-animal tracking experiments with onboard hardware confirm that near real-time high-resolution wildlife tracking is possible on UAVs whilst maintaining high accuracy levels as needed for future navigational and mission-specific animal-centric operational autonomy.
comment: Submitted in CV4Animals 2025
☆ MT-R1-Zero: Advancing LLM-based Machine Translation via R1-Zero-like Reinforcement Learning
Large-scale reinforcement learning (RL) methods have proven highly effective in enhancing the reasoning abilities of large language models (LLMs), particularly for tasks with verifiable solutions such as mathematics and coding. However, applying this idea to machine translation (MT), where outputs are flexibly formatted and difficult to automatically evaluate with explicit rules, remains underexplored. In this work, we introduce MT-R1-Zero, the first open-source adaptation of the R1-Zero RL framework for MT without supervised fine-tuning or cold-start. We propose a rule-metric mixed reward mechanism to guide LLMs towards improved translation quality via emergent reasoning. On the WMT 24 English-Chinese benchmark, our MT-R1-Zero-3B-Mix achieves competitive performance, surpassing TowerInstruct-7B-v0.2 by an average of 1.26 points. Meanwhile, our MT-R1-Zero-7B-Mix attains a high average score of 62.25 across all metrics, placing it on par with advanced proprietary models such as GPT-4o and Claude-3.5-Sonnet, while the MT-R1-Zero-7B-Sem variant achieves state-of-the-art scores on semantic metrics. Moreover, our work exhibits strong generalization capabilities on out-of-distribution MT tasks, robustly supporting multilingual and low-resource settings. Extensive analysis of model behavior across different initializations and reward metrics offers pioneering insight into the critical role of reward design, LLM adaptability, training dynamics, and emergent reasoning patterns within the R1-Zero paradigm for MT. Our code is available at https://github.com/fzp0424/MT-R1-Zero.
comment: Work in progress. Our code is available at https://github.com/fzp0424/MT-R1-Zero
☆ COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts
Current object detectors often suffer significant perfor-mance degradation in real-world applications when encountering distributional shifts. Consequently, the out-of-distribution (OOD) generalization capability of object detectors has garnered increasing attention from researchers. Despite this growing interest, there remains a lack of a large-scale, comprehensive dataset and evaluation benchmark with fine-grained annotations tailored to assess the OOD generalization on more intricate tasks like object detection and grounding. To address this gap, we introduce COUNTS, a large-scale OOD dataset with object-level annotations. COUNTS encompasses 14 natural distributional shifts, over 222K samples, and more than 1,196K labeled bounding boxes. Leveraging COUNTS, we introduce two novel benchmarks: O(OD)2 and OODG. O(OD)2 is designed to comprehensively evaluate the OOD generalization capabilities of object detectors by utilizing controlled distribution shifts between training and testing data. OODG, on the other hand, aims to assess the OOD generalization of grounding abilities in multimodal large language models (MLLMs). Our findings reveal that, while large models and extensive pre-training data substantially en hance performance in in-distribution (IID) scenarios, significant limitations and opportunities for improvement persist in OOD contexts for both object detectors and MLLMs. In visual grounding tasks, even the advanced GPT-4o and Gemini-1.5 only achieve 56.7% and 28.0% accuracy, respectively. We hope COUNTS facilitates advancements in the development and assessment of robust object detectors and MLLMs capable of maintaining high performance under distributional shifts.
☆ BoTTA: Benchmarking on-device Test Time Adaptation
The performance of deep learning models depends heavily on test samples at runtime, and shifts from the training data distribution can significantly reduce accuracy. Test-time adaptation (TTA) addresses this by adapting models during inference without requiring labeled test data or access to the original training set. While research has explored TTA from various perspectives like algorithmic complexity, data and class distribution shifts, model architectures, and offline versus continuous learning, constraints specific to mobile and edge devices remain underexplored. We propose BoTTA, a benchmark designed to evaluate TTA methods under practical constraints on mobile and edge devices. Our evaluation targets four key challenges caused by limited resources and usage conditions: (i) limited test samples, (ii) limited exposure to categories, (iii) diverse distribution shifts, and (iv) overlapping shifts within a sample. We assess state-of-the-art TTA methods under these scenarios using benchmark datasets and report system-level metrics on a real testbed. Furthermore, unlike prior work, we align with on-device requirements by advocating periodic adaptation instead of continuous inference-time adaptation. Experiments reveal key insights: many recent TTA algorithms struggle with small datasets, fail to generalize to unseen categories, and depend on the diversity and complexity of distribution shifts. BoTTA also reports device-specific resource use. For example, while SHOT improves accuracy by $2.25\times$ with $512$ adaptation samples, it uses $1.08\times$ peak memory on Raspberry Pi versus the base model. BoTTA offers actionable guidance for TTA in real-world, resource-constrained deployments.
☆ GeoUni: A Unified Model for Generating Geometry Diagrams, Problems and Problem Solutions
We propose GeoUni, the first unified geometry expert model capable of generating problem solutions and diagrams within a single framework in a way that enables the creation of unique and individualized geometry problems. Traditionally, solving geometry problems and generating diagrams have been treated as separate tasks in machine learning, with no models successfully integrating both to support problem creation. However, we believe that mastery in geometry requires frictionless integration of all of these skills, from solving problems to visualizing geometric relationships, and finally, crafting tailored problems. Our extensive experiments demonstrate that GeoUni, with only 1.5B parameters, achieves performance comparable to larger models such as DeepSeek-R1 with 671B parameters in geometric reasoning tasks. GeoUni also excels in generating precise geometric diagrams, surpassing both text-to-image models and unified models, including the GPT-4o image generation. Most importantly, GeoUni is the only model capable of successfully generating textual problems with matching diagrams based on specific knowledge points, thus offering a wider range of capabilities that extend beyond current models.
☆ Breaking the Data Barrier -- Building GUI Agents Through Task Generalization
Graphical User Interface (GUI) agents offer cross-platform solutions for automating complex digital tasks, with significant potential to transform productivity workflows. However, their performance is often constrained by the scarcity of high-quality trajectory data. To address this limitation, we propose training Vision Language Models (VLMs) on data-rich, reasoning-intensive tasks during a dedicated mid-training stage, and then examine how incorporating these tasks facilitates generalization to GUI planning scenarios. Specifically, we explore a range of tasks with readily available instruction-tuning data, including GUI perception, multimodal reasoning, and textual reasoning. Through extensive experiments across 11 mid-training tasks, we demonstrate that: (1) Task generalization proves highly effective, yielding substantial improvements across most settings. For instance, multimodal mathematical reasoning enhances performance on AndroidWorld by an absolute 6.3%. Remarkably, text-only mathematical data significantly boosts GUI web agent performance, achieving a 5.6% improvement on WebArena and 5.4% improvement on AndroidWorld, underscoring notable cross-modal generalization from text-based to visual domains; (2) Contrary to prior assumptions, GUI perception data - previously considered closely aligned with GUI agent tasks and widely utilized for training - has a comparatively limited impact on final performance; (3) Building on these insights, we identify the most effective mid-training tasks and curate optimized mixture datasets, resulting in absolute performance gains of 8.0% on WebArena and 12.2% on AndroidWorld. Our work provides valuable insights into cross-domain knowledge transfer for GUI agents and offers a practical approach to addressing data scarcity challenges in this emerging field. The code, data and models will be available at https://github.com/hkust-nlp/GUIMid.
comment: 24 pages, 11 figures
☆ Benchmarking Practices in LLM-driven Offensive Security: Testbeds, Metrics, and Experiment Design
Large Language Models (LLMs) have emerged as a powerful approach for driving offensive penetration-testing tooling. This paper analyzes the methodology and benchmarking practices used for evaluating Large Language Model (LLM)-driven attacks, focusing on offensive uses of LLMs in cybersecurity. We review 16 research papers detailing 15 prototypes and their respective testbeds. We detail our findings and provide actionable recommendations for future research, emphasizing the importance of extending existing testbeds, creating baselines, and including comprehensive metrics and qualitative analysis. We also note the distinction between security research and practice, suggesting that CTF-based challenges may not fully represent real-world penetration testing scenarios.
☆ Lightweight Trustworthy Distributed Clustering
Ensuring data trustworthiness within individual edge nodes while facilitating collaborative data processing poses a critical challenge in edge computing systems (ECS), particularly in resource-constrained scenarios such as autonomous systems sensor networks, industrial IoT, and smart cities. This paper presents a lightweight, fully distributed k-means clustering algorithm specifically adapted for edge environments, leveraging a distributed averaging approach with additive secret sharing, a secure multiparty computation technique, during the cluster center update phase to ensure the accuracy and trustworthiness of data across nodes.
☆ SoccerNet-v3D: Leveraging Sports Broadcast Replays for 3D Scene Understanding
Sports video analysis is a key domain in computer vision, enabling detailed spatial understanding through multi-view correspondences. In this work, we introduce SoccerNet-v3D and ISSIA-3D, two enhanced and scalable datasets designed for 3D scene understanding in soccer broadcast analysis. These datasets extend SoccerNet-v3 and ISSIA by incorporating field-line-based camera calibration and multi-view synchronization, enabling 3D object localization through triangulation. We propose a monocular 3D ball localization task built upon the triangulation of ground-truth 2D ball annotations, along with several calibration and reprojection metrics to assess annotation quality on demand. Additionally, we present a single-image 3D ball localization method as a baseline, leveraging camera calibration and ball size priors to estimate the ball's position from a monocular viewpoint. To further refine 2D annotations, we introduce a bounding box optimization technique that ensures alignment with the 3D scene representation. Our proposed datasets establish new benchmarks for 3D soccer scene understanding, enhancing both spatial and temporal analysis in sports analytics. Finally, we provide code to facilitate access to our annotations and the generation pipelines for the datasets.
☆ RealSafe-R1: Safety-Aligned DeepSeek-R1 without Compromising Reasoning Capability
Large Reasoning Models (LRMs), such as OpenAI o1 and DeepSeek-R1, have been rapidly progressing and achieving breakthrough performance on complex reasoning tasks such as mathematics and coding. However, the open-source R1 models have raised safety concerns in wide applications, such as the tendency to comply with malicious queries, which greatly impacts the utility of these powerful models in their applications. In this paper, we introduce RealSafe-R1 as safety-aligned versions of DeepSeek-R1 distilled models. To train these models, we construct a dataset of 15k safety-aware reasoning trajectories generated by DeepSeek-R1, under explicit instructions for expected refusal behavior. Both quantitative experiments and qualitative case studies demonstrate the models' improvements, which are shown in their safety guardrails against both harmful queries and jailbreak attacks. Importantly, unlike prior safety alignment efforts that often compromise reasoning performance, our method preserves the models' reasoning capabilities by maintaining the training data within the original distribution of generation. Model weights of RealSafe-R1 are open-source at https://huggingface.co/RealSafe.
☆ Towards Quantifying Commonsense Reasoning with Mechanistic Insights NAACL 2025
Commonsense reasoning deals with the implicit knowledge that is well understood by humans and typically acquired via interactions with the world. In recent times, commonsense reasoning and understanding of various LLMs have been evaluated using text-based tasks. In this work, we argue that a proxy of this understanding can be maintained as a graphical structure that can further help to perform a rigorous evaluation of commonsense reasoning abilities about various real-world activities. We create an annotation scheme for capturing this implicit knowledge in the form of a graphical structure for 37 daily human activities. We find that the created resource can be used to frame an enormous number of commonsense queries (~ 10^{17}), facilitating rigorous evaluation of commonsense reasoning in LLMs. Moreover, recently, the remarkable performance of LLMs has raised questions about whether these models are truly capable of reasoning in the wild and, in general, how reasoning occurs inside these models. In this resource paper, we bridge this gap by proposing design mechanisms that facilitate research in a similar direction. Our findings suggest that the reasoning components are localized in LLMs that play a prominent role in decision-making when prompted with a commonsense query.
comment: Accepted at NAACL 2025; 28 pages (9 pages + 7 pages references + 12 pages appendix)
☆ MMKB-RAG: A Multi-Modal Knowledge-Based Retrieval-Augmented Generation Framework
Recent advancements in large language models (LLMs) and multi-modal LLMs have been remarkable. However, these models still rely solely on their parametric knowledge, which limits their ability to generate up-to-date information and increases the risk of producing erroneous content. Retrieval-Augmented Generation (RAG) partially mitigates these challenges by incorporating external data sources, yet the reliance on databases and retrieval systems can introduce irrelevant or inaccurate documents, ultimately undermining both performance and reasoning quality. In this paper, we propose Multi-Modal Knowledge-Based Retrieval-Augmented Generation (MMKB-RAG), a novel multi-modal RAG framework that leverages the inherent knowledge boundaries of models to dynamically generate semantic tags for the retrieval process. This strategy enables the joint filtering of retrieved documents, retaining only the most relevant and accurate references. Extensive experiments on knowledge-based visual question-answering tasks demonstrate the efficacy of our approach: on the E-VQA dataset, our method improves performance by +4.2\% on the Single-Hop subset and +0.4\% on the full dataset, while on the InfoSeek dataset, it achieves gains of +7.8\% on the Unseen-Q subset, +8.2\% on the Unseen-E subset, and +8.1\% on the full dataset. These results highlight significant enhancements in both accuracy and robustness over the current state-of-the-art MLLM and RAG frameworks.
☆ Pay Attention to What and Where? Interpretable Feature Extractor in Vision-based Deep Reinforcement Learning
Current approaches in Explainable Deep Reinforcement Learning have limitations in which the attention mask has a displacement with the objects in visual input. This work addresses a spatial problem within traditional Convolutional Neural Networks (CNNs). We propose the Interpretable Feature Extractor (IFE) architecture, aimed at generating an accurate attention mask to illustrate both "what" and "where" the agent concentrates on in the spatial domain. Our design incorporates a Human-Understandable Encoding module to generate a fully interpretable attention mask, followed by an Agent-Friendly Encoding module to enhance the agent's learning efficiency. These two components together form the Interpretable Feature Extractor for vision-based deep reinforcement learning to enable the model's interpretability. The resulting attention mask is consistent, highly understandable by humans, accurate in spatial dimension, and effectively highlights important objects or locations in visual input. The Interpretable Feature Extractor is integrated into the Fast and Data-efficient Rainbow framework, and evaluated on 57 ATARI games to show the effectiveness of the proposed approach on Spatial Preservation, Interpretability, and Data-efficiency. Finally, we showcase the versatility of our approach by incorporating the IFE into the Asynchronous Advantage Actor-Critic Model.
☆ Mavors: Multi-granularity Video Representation for Multimodal Large Language Model
Long-context video understanding in multimodal large language models (MLLMs) faces a critical challenge: balancing computational efficiency with the retention of fine-grained spatio-temporal patterns. Existing approaches (e.g., sparse sampling, dense sampling with low resolution, and token compression) suffer from significant information loss in temporal dynamics, spatial details, or subtle interactions, particularly in videos with complex motion or varying resolutions. To address this, we propose $\mathbf{Mavors}$, a novel framework that introduces $\mathbf{M}$ulti-gr$\mathbf{a}$nularity $\mathbf{v}$ide$\mathbf{o}$ $\mathbf{r}$epre$\mathbf{s}$entation for holistic long-video modeling. Specifically, Mavors directly encodes raw video content into latent representations through two core components: 1) an Intra-chunk Vision Encoder (IVE) that preserves high-resolution spatial features via 3D convolutions and Vision Transformers, and 2) an Inter-chunk Feature Aggregator (IFA) that establishes temporal coherence across chunks using transformer-based dependency modeling with chunk-level rotary position encodings. Moreover, the framework unifies image and video understanding by treating images as single-frame videos via sub-image decomposition. Experiments across diverse benchmarks demonstrate Mavors' superiority in maintaining both spatial fidelity and temporal continuity, significantly outperforming existing methods in tasks requiring fine-grained spatio-temporal reasoning.
comment: 22 pages
☆ Hallucination Detection in LLMs via Topological Divergence on Attention Graphs
Hallucination, i.e., generating factually incorrect content, remains a critical challenge for large language models (LLMs). We introduce TOHA, a TOpology-based HAllucination detector in the RAG setting, which leverages a topological divergence metric to quantify the structural properties of graphs induced by attention matrices. Examining the topological divergence between prompt and response subgraphs reveals consistent patterns: higher divergence values in specific attention heads correlate with hallucinated outputs, independent of the dataset. Extensive experiments, including evaluation on question answering and data-to-text tasks, show that our approach achieves state-of-the-art or competitive results on several benchmarks, two of which were annotated by us and are being publicly released to facilitate further research. Beyond its strong in-domain performance, TOHA maintains remarkable domain transferability across multiple open-source LLMs. Our findings suggest that analyzing the topological structure of attention matrices can serve as an efficient and robust indicator of factual reliability in LLMs.
☆ CHARM: Calibrating Reward Models With Chatbot Arena Scores
Reward models (RMs) play a crucial role in Reinforcement Learning from Human Feedback by serving as proxies for human preferences in aligning large language models. In this paper, we identify a model preference bias in RMs, where they systematically assign disproportionately high scores to responses from certain policy models. This bias distorts ranking evaluations and leads to unfair judgments. To address this issue, we propose a calibration method named CHatbot Arena calibrated Reward Modeling (CHARM) that leverages Elo scores from the Chatbot Arena leaderboard to mitigate RM overvaluation. We also introduce a Mismatch Degree metric to measure this preference bias. Our approach is computationally efficient, requiring only a small preference dataset for continued training of the RM. We conduct extensive experiments on reward model benchmarks and human preference alignment. Results demonstrate that our calibrated RMs (1) achieve improved evaluation accuracy on RM-Bench and the Chat-Hard domain of RewardBench, and (2) exhibit a stronger correlation with human preferences by producing scores more closely aligned with Elo rankings. By mitigating model preference bias, our method provides a generalizable and efficient solution for building fairer and more reliable reward models.
☆ EmbodiedAgent: A Scalable Hierarchical Approach to Overcome Practical Challenge in Multi-Robot Control
This paper introduces EmbodiedAgent, a hierarchical framework for heterogeneous multi-robot control. EmbodiedAgent addresses critical limitations of hallucination in impractical tasks. Our approach integrates a next-action prediction paradigm with a structured memory system to decompose tasks into executable robot skills while dynamically validating actions against environmental constraints. We present MultiPlan+, a dataset of more than 18,000 annotated planning instances spanning 100 scenarios, including a subset of impractical cases to mitigate hallucination. To evaluate performance, we propose the Robot Planning Assessment Schema (RPAS), combining automated metrics with LLM-aided expert grading. Experiments demonstrate EmbodiedAgent's superiority over state-of-the-art models, achieving 71.85% RPAS score. Real-world validation in an office service task highlights its ability to coordinate heterogeneous robots for long-horizon objectives.
☆ Sequence models for by-trial decoding of cognitive strategies from neural data
Understanding the sequence of cognitive operations that underlie decision-making is a fundamental challenge in cognitive neuroscience. Traditional approaches often rely on group-level statistics, which obscure trial-by-trial variations in cognitive strategies. In this study, we introduce a novel machine learning method that combines Hidden Multivariate Pattern analysis with a Structured State Space Sequence model to decode cognitive strategies from electroencephalography data at the trial level. We apply this method to a decision-making task, where participants were instructed to prioritize either speed or accuracy in their responses. Our results reveal an additional cognitive operation, labeled Confirmation, which seems to occur predominantly in the accuracy condition but also frequently in the speed condition. The modeled probability that this operation occurs is associated with higher probability of responding correctly as well as changes of mind, as indexed by electromyography data. By successfully modeling cognitive operations at the trial level, we provide empirical evidence for dynamic variability in decision strategies, challenging the assumption of homogeneous cognitive processes within experimental conditions. Our approach shows the potential of sequence modeling in cognitive neuroscience to capture trial-level variability that is obscured by aggregate analyses. The introduced method offers a new way to detect and understand cognitive strategies in a data-driven manner, with implications for both theoretical research and practical applications in many fields.
comment: 15 pages, 6 figures
☆ Progressive Transfer Learning for Multi-Pass Fundus Image Restoration
Diabetic retinopathy is a leading cause of vision impairment, making its early diagnosis through fundus imaging critical for effective treatment planning. However, the presence of poor quality fundus images caused by factors such as inadequate illumination, noise, blurring and other motion artifacts yields a significant challenge for accurate DR screening. In this study, we propose progressive transfer learning for multi pass restoration to iteratively enhance the quality of degraded fundus images, ensuring more reliable DR screening. Unlike previous methods that often focus on a single pass restoration, multi pass restoration via PTL can achieve a superior blind restoration performance that can even improve most of the good quality fundus images in the dataset. Initially, a Cycle GAN model is trained to restore low quality images, followed by PTL induced restoration passes over the latest restored outputs to improve overall quality in each pass. The proposed method can learn blind restoration without requiring any paired data while surpassing its limitations by leveraging progressive learning and fine tuning strategies to minimize distortions and preserve critical retinal features. To evaluate PTL's effectiveness on multi pass restoration, we conducted experiments on DeepDRiD, a large scale fundus imaging dataset specifically curated for diabetic retinopathy detection. Our result demonstrates state of the art performance, showcasing PTL's potential as a superior approach to iterative image quality restoration.
comment: 13 pages, 12 figures including appendix
☆ The Mirage of Performance Gains: Why Contrastive Decoding Fails to Address Multimodal Hallucination
Contrastive decoding strategies are widely used to reduce hallucinations in multimodal large language models (MLLMs). These methods work by constructing contrastive samples to induce hallucinations and then suppressing them in the output distribution. However, this paper demonstrates that such approaches fail to effectively mitigate the hallucination problem. The performance improvements observed on POPE Benchmark are largely driven by two misleading factors: (1) crude, unidirectional adjustments to the model's output distribution and (2) the adaptive plausibility constraint, which reduces the sampling strategy to greedy search. To further illustrate these issues, we introduce a series of spurious improvement methods and evaluate their performance against contrastive decoding techniques. Experimental results reveal that the observed performance gains in contrastive decoding are entirely unrelated to its intended goal of mitigating hallucinations. Our findings challenge common assumptions about the effectiveness of contrastive decoding strategies and pave the way for developing genuinely effective solutions to hallucinations in MLLMs.
☆ RGB-Event based Pedestrian Attribute Recognition: A Benchmark Dataset and An Asymmetric RWKV Fusion Framework
Existing pedestrian attribute recognition methods are generally developed based on RGB frame cameras. However, these approaches are constrained by the limitations of RGB cameras, such as sensitivity to lighting conditions and motion blur, which hinder their performance. Furthermore, current attribute recognition primarily focuses on analyzing pedestrians' external appearance and clothing, lacking an exploration of emotional dimensions. In this paper, we revisit these issues and propose a novel multi-modal RGB-Event attribute recognition task by drawing inspiration from the advantages of event cameras in low-light, high-speed, and low-power consumption. Specifically, we introduce the first large-scale multi-modal pedestrian attribute recognition dataset, termed EventPAR, comprising 100K paired RGB-Event samples that cover 50 attributes related to both appearance and six human emotions, diverse scenes, and various seasons. By retraining and evaluating mainstream PAR models on this dataset, we establish a comprehensive benchmark and provide a solid foundation for future research in terms of data and algorithmic baselines. In addition, we propose a novel RWKV-based multi-modal pedestrian attribute recognition framework, featuring an RWKV visual encoder and an asymmetric RWKV fusion module. Extensive experiments are conducted on our proposed dataset as well as two simulated datasets (MARS-Attribute and DukeMTMC-VID-Attribute), achieving state-of-the-art results. The source code and dataset will be released on https://github.com/Event-AHU/OpenPAR
comment: The First Benchmark Dataset for RGB-Event Multimodal Pedestrian Attribute Recognition Task
☆ Air Quality Prediction with A Meteorology-Guided Modality-Decoupled Spatio-Temporal Network
Air quality prediction plays a crucial role in public health and environmental protection. Accurate air quality prediction is a complex multivariate spatiotemporal problem, that involves interactions across temporal patterns, pollutant correlations, spatial station dependencies, and particularly meteorological influences that govern pollutant dispersion and chemical transformations. Existing works underestimate the critical role of atmospheric conditions in air quality prediction and neglect comprehensive meteorological data utilization, thereby impairing the modeling of dynamic interdependencies between air quality and meteorological data. To overcome this, we propose MDSTNet, an encoder-decoder framework that explicitly models air quality observations and atmospheric conditions as distinct modalities, integrating multi-pressure-level meteorological data and weather forecasts to capture atmosphere-pollution dependencies for prediction. Meantime, we construct ChinaAirNet, the first nationwide dataset combining air quality records with multi-pressure-level meteorological observations. Experimental results on ChinaAirNet demonstrate MDSTNet's superiority, substantially reducing 48-hour prediction errors by 17.54\% compared to the state-of-the-art model. The source code and dataset will be available on github.
☆ Session-based Recommender Systems: User Interest as a Stochastic Process in the Latent Space
This paper jointly addresses the problem of data uncertainty, popularity bias, and exposure bias in session-based recommender systems. We study the symptoms of this bias both in item embeddings and in recommendations. We propose treating user interest as a stochastic process in the latent space and providing a model-agnostic implementation of this mathematical concept. The proposed stochastic component consists of elements: debiasing item embeddings with regularization for embedding uniformity, modeling dense user interest from session prefixes, and introducing fake targets in the data to simulate extended exposure. We conducted computational experiments on two popular benchmark datasets, Diginetica and YooChoose 1/64, as well as several modifications of the YooChoose dataset with different ratios of popular items. The results show that the proposed approach allows us to mitigate the challenges mentioned.
☆ Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models? CVPR 2025
Multi-modal large language models (MLLMs) have made significant progress, yet their safety alignment remains limited. Typically, current open-source MLLMs rely on the alignment inherited from their language module to avoid harmful generations. However, the lack of safety measures specifically designed for multi-modal inputs creates an alignment gap, leaving MLLMs vulnerable to vision-domain attacks such as typographic manipulation. Current methods utilize a carefully designed safety dataset to enhance model defense capability, while the specific knowledge or patterns acquired from the high-quality dataset remain unclear. Through comparison experiments, we find that the alignment gap primarily arises from data distribution biases, while image content, response quality, or the contrastive behavior of the dataset makes little contribution to boosting multi-modal safety. To further investigate this and identify the key factors in improving MLLM safety, we propose finetuning MLLMs on a small set of benign instruct-following data with responses replaced by simple, clear rejection sentences. Experiments show that, without the need for labor-intensive collection of high-quality malicious data, model safety can still be significantly improved, as long as a specific fraction of rejection data exists in the finetuning set, indicating the security alignment is not lost but rather obscured during multi-modal pretraining or instruction finetuning. Simply correcting the underlying data bias could narrow the safety gap in the vision domain.
comment: Accepted to CVPR 2025, codes in process
☆ Metric-Guided Synthesis of Class Activation Mapping
Class activation mapping (CAM) is a widely adopted class of saliency methods used to explain the behavior of convolutional neural networks (CNNs). These methods generate heatmaps that highlight the parts of the input most relevant to the CNN output. Various CAM methods have been proposed, each distinguished by the expressions used to derive heatmaps. In general, users look for heatmaps with specific properties that reflect different aspects of CNN functionality. These may include similarity to ground truth, robustness, equivariance, and more. Although existing CAM methods implicitly encode some of these properties in their expressions, they do not allow for variability in heatmap generation following the user's intent or domain knowledge. In this paper, we address this limitation by introducing SyCAM, a metric-based approach for synthesizing CAM expressions. Given a predefined evaluation metric for saliency maps, SyCAM automatically generates CAM expressions optimized for that metric. We specifically explore a syntax-guided synthesis instantiation of SyCAM, where CAM expressions are derived based on predefined syntactic constraints and the given metric. Using several established evaluation metrics, we demonstrate the efficacy and flexibility of our approach in generating targeted heatmaps. We compare SyCAM with other well-known CAM methods on three prominent models: ResNet50, VGG16, and VGG19.
☆ GenTe: Generative Real-world Terrains for General Legged Robot Locomotion Control
Developing bipedal robots capable of traversing diverse real-world terrains presents a fundamental robotics challenge, as existing methods using predefined height maps and static environments fail to address the complexity of unstructured landscapes. To bridge this gap, we propose GenTe, a framework for generating physically realistic and adaptable terrains to train generalizable locomotion policies. GenTe constructs an atomic terrain library that includes both geometric and physical terrains, enabling curriculum training for reinforcement learning-based locomotion policies. By leveraging function-calling techniques and reasoning capabilities of Vision-Language Models (VLMs), GenTe generates complex, contextually relevant terrains from textual and graphical inputs. The framework introduces realistic force modeling for terrain interactions, capturing effects such as soil sinkage and hydrodynamic resistance. To the best of our knowledge, GenTe is the first framework that systemically generates simulation environments for legged robot locomotion control. Additionally, we introduce a benchmark of 100 generated terrains. Experiments demonstrate improved generalization and robustness in bipedal robot locomotion.
☆ Enhancing Multi-task Learning Capability of Medical Generalist Foundation Model via Image-centric Multi-annotation Data
The emergence of medical generalist foundation models has revolutionized conventional task-specific model development paradigms, aiming to better handle multiple tasks through joint training on large-scale medical datasets. However, recent advances prioritize simple data scaling or architectural component enhancement, while neglecting to re-examine multi-task learning from a data-centric perspective. Critically, simply aggregating existing data resources leads to decentralized image-task alignment, which fails to cultivate comprehensive image understanding or align with clinical needs for multi-dimensional image interpretation. In this paper, we introduce the image-centric multi-annotation X-ray dataset (IMAX), the first attempt to enhance the multi-task learning capabilities of medical multi-modal large language models (MLLMs) from the data construction level. To be specific, IMAX is featured from the following attributes: 1) High-quality data curation. A comprehensive collection of more than 354K entries applicable to seven different medical tasks. 2) Image-centric dense annotation. Each X-ray image is associated with an average of 4.10 tasks and 7.46 training entries, ensuring multi-task representation richness per image. Compared to the general decentralized multi-annotation X-ray dataset (DMAX), IMAX consistently demonstrates significant multi-task average performance gains ranging from 3.20% to 21.05% across seven open-source state-of-the-art medical MLLMs. Moreover, we investigate differences in statistical patterns exhibited by IMAX and DMAX training processes, exploring potential correlations between optimization dynamics and multi-task performance. Finally, leveraging the core concept of IMAX data construction, we propose an optimized DMAX-based training strategy to alleviate the dilemma of obtaining high-quality IMAX data in practical scenarios.
☆ Towards Unbiased Federated Graph Learning: Label and Topology Perspectives
Federated Graph Learning (FGL) enables privacy-preserving, distributed training of graph neural networks without sharing raw data. Among its approaches, subgraph-FL has become the dominant paradigm, with most work focused on improving overall node classification accuracy. However, these methods often overlook fairness due to the complexity of node features, labels, and graph structures. In particular, they perform poorly on nodes with disadvantaged properties, such as being in the minority class within subgraphs or having heterophilous connections (neighbors with dissimilar labels or misleading features). This reveals a critical issue: high accuracy can mask degraded performance on structurally or semantically marginalized nodes. To address this, we advocate for two fairness goals: (1) improving representation of minority class nodes for class-wise fairness and (2) mitigating topological bias from heterophilous connections for topology-aware fairness. We propose FairFGL, a novel framework that enhances fairness through fine-grained graph mining and collaborative learning. On the client side, the History-Preserving Module prevents overfitting to dominant local classes, while the Majority Alignment Module refines representations of heterophilous majority-class nodes. The Gradient Modification Module transfers minority-class knowledge from structurally favorable clients to improve fairness. On the server side, FairFGL uploads only the most influenced subset of parameters to reduce communication costs and better reflect local distributions. A cluster-based aggregation strategy reconciles conflicting updates and curbs global majority dominance . Extensive evaluations on eight benchmarks show FairFGL significantly improves minority-group performance , achieving up to a 22.62 percent Macro-F1 gain while enhancing convergence over state-of-the-art baselines.
comment: Under Review
☆ Privacy Meets Explainability: Managing Confidential Data and Transparency Policies in LLM-Empowered Science
As Large Language Models (LLMs) become integral to scientific workflows, concerns over the confidentiality and ethical handling of confidential data have emerged. This paper explores data exposure risks through LLM-powered scientific tools, which can inadvertently leak confidential information, including intellectual property and proprietary data, from scientists' perspectives. We propose "DataShield", a framework designed to detect confidential data leaks, summarize privacy policies, and visualize data flow, ensuring alignment with organizational policies and procedures. Our approach aims to inform scientists about data handling practices, enabling them to make informed decisions and protect sensitive information. Ongoing user studies with scientists are underway to evaluate the framework's usability, trustworthiness, and effectiveness in tackling real-world privacy challenges.
comment: 8 pages
☆ Omni-Dish: Photorealistic and Faithful Image Generation and Editing for Arbitrary Chinese Dishes
Dish images play a crucial role in the digital era, with the demand for culturally distinctive dish images continuously increasing due to the digitization of the food industry and e-commerce. In general cases, existing text-to-image generation models excel in producing high-quality images; however, they struggle to capture diverse characteristics and faithful details of specific domains, particularly Chinese dishes. To address this limitation, we propose Omni-Dish, the first text-to-image generation model specifically tailored for Chinese dishes. We develop a comprehensive dish curation pipeline, building the largest dish dataset to date. Additionally, we introduce a recaption strategy and employ a coarse-to-fine training scheme to help the model better learn fine-grained culinary nuances. During inference, we enhance the user's textual input using a pre-constructed high-quality caption library and a large language model, enabling more photorealistic and faithful image generation. Furthermore, to extend our model's capability for dish editing tasks, we propose Concept-Enhanced P2P. Based on this approach, we build a dish editing dataset and train a specialized editing model. Extensive experiments demonstrate the superiority of our methods.
comment: 10 pages, 10 figures, 3 tables
☆ FedRecon: Missing Modality Reconstruction in Distributed Heterogeneous Environments
Multimodal data are often incomplete and exhibit Non-Independent and Identically Distributed (Non-IID) characteristics in real-world scenarios. These inherent limitations lead to both modality heterogeneity through partial modality absence and data heterogeneity from distribution divergence, creating fundamental challenges for effective federated learning (FL). To address these coupled challenges, we propose FedRecon, the first method targeting simultaneous missing modality reconstruction and Non-IID adaptation in multimodal FL. Our approach first employs a lightweight Multimodal Variational Autoencoder (MVAE) to reconstruct missing modalities while preserving cross-modal consistency. Distinct from conventional imputation methods, we achieve sample-level alignment through a novel distribution mapping mechanism that guarantees both data consistency and completeness. Additionally, we introduce a strategy employing global generator freezing to prevent catastrophic forgetting, which in turn mitigates Non-IID fluctuations. Extensive evaluations on multimodal datasets demonstrate FedRecon's superior performance in modality reconstruction under Non-IID conditions, surpassing state-of-the-art methods.
comment: 18 pages, 32 figures
☆ KeepKV: Eliminating Output Perturbation in KV Cache Compression for Efficient LLMs Inference
Efficient inference of large language models (LLMs) is hindered by an ever-growing key-value (KV) cache, making KV cache compression a critical research direction. Traditional methods selectively evict less important KV cache entries based on attention scores or position heuristics, which leads to information loss and hallucinations. Recently, merging-based strategies have been explored to retain more information by merging KV pairs that would be discarded; however, these existing approaches inevitably introduce inconsistencies in attention distributions before and after merging, causing output perturbation and degraded generation quality. To overcome this challenge, we propose KeepKV, a novel adaptive KV cache merging method designed to eliminate output perturbation while preserving performance under strict memory constraints. KeepKV introduces the Electoral Votes mechanism that records merging history and adaptively adjusts attention scores. Moreover, it further leverages a novel Zero Inference-Perturbation Merging methods, keeping attention consistency and compensating for attention loss resulting from cache merging. KeepKV successfully retains essential context information within a significantly compressed cache. Extensive experiments on various benchmarks and LLM architectures demonstrate that KeepKV substantially reduces memory usage, enhances inference throughput by more than 2x and keeps superior generation quality even with 10% KV cache budgets.
comment: 18 pages, 8 figures
☆ Quantum Natural Language Processing: A Comprehensive Review of Models, Methods, and Applications
In recent developments, deep learning methodologies applied to Natural Language Processing (NLP) have revealed a paradox: They improve performance but demand considerable data and resources for their training. Alternatively, quantum computing exploits the principles of quantum mechanics to overcome the computational limitations of current methodologies, thereby establishing an emerging field known as quantum natural language processing (QNLP). This domain holds the potential to attain a quantum advantage in the processing of linguistic structures, surpassing classical models in both efficiency and accuracy. In this paper, it is proposed to categorise QNLP models based on quantum computing principles, architecture, and computational approaches. This paper attempts to provide a survey on how quantum meets language by mapping state-of-the-art in this area, embracing quantum encoding techniques for classical data, QNLP models for prevalent NLP tasks, and quantum optimisation techniques for hyper parameter tuning. The landscape of quantum computing approaches applied to various NLP tasks is summarised by showcasing the specific QNLP methods used, and the popularity of these methods is indicated by their count. From the findings, it is observed that QNLP approaches are still limited to small data sets, with only a few models explored extensively, and there is increasing interest in the application of quantum computing to natural language processing tasks.
☆ Plasticity-Aware Mixture of Experts for Learning Under QoE Shifts in Adaptive Video Streaming
Adaptive video streaming systems are designed to optimize Quality of Experience (QoE) and, in turn, enhance user satisfaction. However, differences in user profiles and video content lead to different weights for QoE factors, resulting in user-specific QoE functions and, thus, varying optimization objectives. This variability poses significant challenges for neural networks, as they often struggle to generalize under evolving targets - a phenomenon known as plasticity loss that prevents conventional models from adapting effectively to changing optimization objectives. To address this limitation, we propose the Plasticity-Aware Mixture of Experts (PA-MoE), a novel learning framework that dynamically modulates network plasticity by balancing memory retention with selective forgetting. In particular, PA-MoE leverages noise injection to promote the selective forgetting of outdated knowledge, thereby endowing neural networks with enhanced adaptive capabilities. In addition, we present a rigorous theoretical analysis of PA-MoE by deriving a regret bound that quantifies its learning performance. Experimental evaluations demonstrate that PA-MoE achieves a 45.5% improvement in QoE over competitive baselines in dynamic streaming environments. Further analysis reveals that the model effectively mitigates plasticity loss by optimizing neuron utilization. Finally, a parameter sensitivity study is performed by injecting varying levels of noise, and the results align closely with our theoretical predictions.
☆ Learning from Reference Answers: Versatile Language Model Alignment without Binary Human Preference Data
Large language models~(LLMs) are expected to be helpful, harmless, and honest. In various alignment scenarios, such as general human preference, safety, and confidence alignment, binary preference data collection and reward modeling are resource-intensive but necessary for human preference transferring. In this work, we explore using the similarity between sampled generations and high-quality reference answers as an alternative reward function for LLM alignment. Using similarity as a reward circumvents training reward models, and collecting a single reference answer potentially costs less time than constructing binary preference pairs when multiple candidates are available. Specifically, we develop \textit{RefAlign}, a versatile REINFORCE-style alignment algorithm, which is free of reference and reward models. Instead, RefAlign utilizes BERTScore between sampled generations and high-quality reference answers as the surrogate reward. Beyond general human preference optimization, RefAlign can be readily extended to diverse scenarios, such as safety and confidence alignment, by incorporating the similarity reward with task-related objectives. In various scenarios, {RefAlign} demonstrates comparable performance to previous alignment methods while offering high efficiency.
comment: work in progress
☆ LangPert: Detecting and Handling Task-level Perturbations for Robust Object Rearrangement
Task execution for object rearrangement could be challenged by Task-Level Perturbations (TLP), i.e., unexpected object additions, removals, and displacements that can disrupt underlying visual policies and fundamentally compromise task feasibility and progress. To address these challenges, we present LangPert, a language-based framework designed to detect and mitigate TLP situations in tabletop rearrangement tasks. LangPert integrates a Visual Language Model (VLM) to comprehensively monitor policy's skill execution and environmental TLP, while leveraging the Hierarchical Chain-of-Thought (HCoT) reasoning mechanism to enhance the Large Language Model (LLM)'s contextual understanding and generate adaptive, corrective skill-execution plans. Our experimental results demonstrate that LangPert handles diverse TLP situations more effectively than baseline methods, achieving higher task completion rates, improved execution efficiency, and potential generalization to unseen scenarios.
☆ Constructing Micro Knowledge Graphs from Technical Support Documents
Short technical support pages such as IBM Technotes are quite common in technical support domain. These pages can be very useful as the knowledge sources for technical support applications such as chatbots, search engines and question-answering (QA) systems. Information extracted from documents to drive technical support applications is often stored in the form of Knowledge Graph (KG). Building KGs from a large corpus of documents poses a challenge of granularity because a large number of entities and actions are present in each page. The KG becomes virtually unusable if all entities and actions from these pages are stored in the KG. Therefore, only key entities and actions from each page are extracted and stored in the KG. This approach however leads to loss of knowledge represented by entities and actions left out of the KG as they are no longer available to graph search and reasoning functions. We propose a set of techniques to create micro knowledge graph (micrograph) for each of such web pages. The micrograph stores all the entities and actions in a page and also takes advantage of the structure of the page to represent exactly in which part of that page these entities and actions appeared, and also how they relate to each other. These micrographs can be used as additional knowledge sources by technical support applications. We define schemas for representing semi-structured and plain text knowledge present in the technical support web pages. Solutions in technical support domain include procedures made of steps. We also propose a technique to extract procedures from these webpages and the schemas to represent them in the micrographs. We also discuss how technical support applications can take advantage of the micrographs.
☆ HDC: Hierarchical Distillation for Multi-level Noisy Consistency in Semi-Supervised Fetal Ultrasound Segmentation
Transvaginal ultrasound is a critical imaging modality for evaluating cervical anatomy and detecting physiological changes. However, accurate segmentation of cervical structures remains challenging due to low contrast, shadow artifacts, and fuzzy boundaries. While convolutional neural networks (CNNs) have shown promising results in medical image segmentation, their performance is often limited by the need for large-scale annotated datasets - an impractical requirement in clinical ultrasound imaging. Semi-supervised learning (SSL) offers a compelling solution by leveraging unlabeled data, but existing teacher-student frameworks often suffer from confirmation bias and high computational costs. We propose HDC, a novel semi-supervised segmentation framework that integrates Hierarchical Distillation and Consistency learning within a multi-level noise mean-teacher framework. Unlike conventional approaches that rely solely on pseudo-labeling, we introduce a hierarchical distillation mechanism that guides feature-level learning via two novel objectives: (1) Correlation Guidance Loss to align feature representations between the teacher and main student branch, and (2) Mutual Information Loss to stabilize representations between the main and noisy student branches. Our framework reduces model complexity while improving generalization. Extensive experiments on two fetal ultrasound datasets, FUGC and PSFH, demonstrate that our method achieves competitive performance with significantly lower computational overhead than existing multi-teacher models.
☆ Truncated Matrix Completion - An Empirical Study
Low-rank Matrix Completion (LRMC) describes the problem where we wish to recover missing entries of partially observed low-rank matrix. Most existing matrix completion work deals with sampling procedures that are independent of the underlying data values. While this assumption allows the derivation of nice theoretical guarantees, it seldom holds in real-world applications. In this paper, we consider various settings where the sampling mask is dependent on the underlying data values, motivated by applications in sensing, sequential decision-making, and recommender systems. Through a series of experiments, we study and compare the performance of various LRMC algorithms that were originally successful for data-independent sampling patterns.
☆ Labeling Messages as AI-Generated Does Not Reduce Their Persuasive Effects
As generative artificial intelligence (AI) enables the creation and dissemination of information at massive scale and speed, it is increasingly important to understand how people perceive AI-generated content. One prominent policy proposal requires explicitly labeling AI-generated content to increase transparency and encourage critical thinking about the information, but prior research has not yet tested the effects of such labels. To address this gap, we conducted a survey experiment (N=1601) on a diverse sample of Americans, presenting participants with an AI-generated message about several public policies (e.g., allowing colleges to pay student-athletes), randomly assigning whether participants were told the message was generated by (a) an expert AI model, (b) a human policy expert, or (c) no label. We found that messages were generally persuasive, influencing participants' views of the policies by 9.74 percentage points on average. However, while 94.6% of participants assigned to the AI and human label conditions believed the authorship labels, labels had no significant effects on participants' attitude change toward the policies, judgments of message accuracy, nor intentions to share the message with others. These patterns were robust across a variety of participant characteristics, including prior knowledge of the policy, prior experience with AI, political party, education level, or age. Taken together, these results imply that, while authorship labels would likely enhance transparency, they are unlikely to substantially affect the persuasiveness of the labeled content, highlighting the need for alternative strategies to address challenges posed by AI-generated information.
☆ SUMART: SUMmARizing Translation from Wordy to Concise Expression
We propose SUMART, a method for summarizing and compressing the volume of verbose subtitle translations. SUMART is designed for understanding translated captions (e.g., interlingual conversations via subtitle translation or when watching movies in foreign language audio and translated captions). SUMART is intended for users who want a big-picture and fast understanding of the conversation, audio, video content, and speech in a foreign language. During the training data collection, when a speaker makes a verbose statement, SUMART employs a large language model on-site to compress the volume of subtitles. This compressed data is then stored in a database for fine-tuning purposes. Later, SUMART uses data pairs from those non-compressed ASR results and compressed translated results for fine-tuning the translation model to generate more concise translations for practical uses. In practical applications, SUMART utilizes this trained model to produce concise translation results. Furthermore, as a practical application, we developed an application that allows conversations using subtitle translation in augmented reality spaces. As a pilot study, we conducted qualitative surveys using a SUMART prototype and a survey on the summarization model for SUMART. We envision the most effective use case of this system is where users need to consume a lot of information quickly (e.g., Speech, lectures, podcasts, Q&A in conferences).
comment: 3 pages, 2 figures
☆ Reasoning Models Can Be Effective Without Thinking
Recent LLMs have significantly improved reasoning capabilities, primarily by including an explicit, lengthy Thinking process as part of generation. In this paper, we question whether this explicit thinking is necessary. Using the state-of-the-art DeepSeek-R1-Distill-Qwen, we find that bypassing the thinking process via simple prompting, denoted as NoThinking, can be surprisingly effective. When controlling for the number of tokens, NoThinking outperforms Thinking across a diverse set of seven challenging reasoning datasets--including mathematical problem solving, formal theorem proving, and coding--especially in low-budget settings, e.g., 51.3 vs. 28.9 on ACM 23 with 700 tokens. Notably, the performance of NoThinking becomes more competitive with pass@k as k increases. Building on this observation, we demonstrate that a parallel scaling approach that uses NoThinking to generate N outputs independently and aggregates them is highly effective. For aggregation, we use task-specific verifiers when available, or we apply simple best-of-N strategies such as confidence-based selection. Our method outperforms a range of baselines with similar latency using Thinking, and is comparable to Thinking with significantly longer latency (up to 9x). Together, our research encourages a reconsideration of the necessity of lengthy thinking processes, while also establishing a competitive reference for achieving strong reasoning performance in low-budget settings or at low latency using parallel scaling.
comment: 33 pages, 7 main figures, 2 tables
☆ Working with Large Language Models to Enhance Messaging Effectiveness for Vaccine Confidence
Vaccine hesitancy and misinformation are significant barriers to achieving widespread vaccination coverage. Smaller public health departments may lack the expertise or resources to craft effective vaccine messaging. This paper explores the potential of ChatGPT-augmented messaging to promote confidence in vaccination uptake. We conducted a survey in which participants chose between pairs of vaccination messages and assessed which was more persuasive and to what extent. In each pair, one message was the original, and the other was augmented by ChatGPT. At the end of the survey, participants were informed that half of the messages had been generated by ChatGPT. They were then asked to provide both quantitative and qualitative responses regarding how knowledge of a message's ChatGPT origin affected their impressions. Overall, ChatGPT-augmented messages were rated slightly higher than the original messages. These messages generally scored better when they were longer. Respondents did not express major concerns about ChatGPT-generated content, nor was there a significant relationship between participants' views on ChatGPT and their message ratings. Notably, there was a correlation between whether a message appeared first or second in a pair and its score. These results point to the potential of ChatGPT to enhance vaccine messaging, suggesting a promising direction for future research on human-AI collaboration in public health communication.
☆ PestMA: LLM-based Multi-Agent System for Informed Pest Management
Effective pest management is complex due to the need for accurate, context-specific decisions. Recent advancements in large language models (LLMs) open new possibilities for addressing these challenges by providing sophisticated, adaptive knowledge acquisition and reasoning. However, existing LLM-based pest management approaches often rely on a single-agent paradigm, which can limit their capacity to incorporate diverse external information, engage in systematic validation, and address complex, threshold-driven decisions. To overcome these limitations, we introduce PestMA, an LLM-based multi-agent system (MAS) designed to generate reliable and evidence-based pest management advice. Building on an editorial paradigm, PestMA features three specialized agents, an Editor for synthesizing pest management recommendations, a Retriever for gathering relevant external data, and a Validator for ensuring correctness. Evaluations on real-world pest scenarios demonstrate that PestMA achieves an initial accuracy of 86.8% for pest management decisions, which increases to 92.6% after validation. These results underscore the value of collaborative agent-based workflows in refining and validating decisions, highlighting the potential of LLM-based multi-agent systems to automate and enhance pest management processes.
comment: 10 pages
☆ Carbon-Efficient 3D DNN Acceleration: Optimizing Performance and Sustainability
As Deep Neural Networks (DNNs) continue to drive advancements in artificial intelligence, the design of hardware accelerators faces growing concerns over embodied carbon footprint due to complex fabrication processes. 3D integration improves performance but introduces sustainability challenges, making carbon-aware optimization essential. In this work, we propose a carbon-efficient design methodology for 3D DNN accelerators, leveraging approximate computing and genetic algorithm-based design space exploration to optimize Carbon Delay Product (CDP). By integrating area-efficient approximate multipliers into Multiply-Accumulate (MAC) units, our approach effectively reduces silicon area and fabrication overhead while maintaining high computational accuracy. Experimental evaluations across three technology nodes (45nm, 14nm, and 7nm) show that our method reduces embodied carbon by up to 30% with negligible accuracy drop.
comment: Submitted in ISVLSI 2025
☆ A Survey of Large Language Model-Powered Spatial Intelligence Across Scales: Advances in Embodied Agents, Smart Cities, and Earth Science
Over the past year, the development of large language models (LLMs) has brought spatial intelligence into focus, with much attention on vision-based embodied intelligence. However, spatial intelligence spans a broader range of disciplines and scales, from navigation and urban planning to remote sensing and earth science. What are the differences and connections between spatial intelligence across these fields? In this paper, we first review human spatial cognition and its implications for spatial intelligence in LLMs. We then examine spatial memory, knowledge representations, and abstract reasoning in LLMs, highlighting their roles and connections. Finally, we analyze spatial intelligence across scales -- from embodied to urban and global levels -- following a framework that progresses from spatial memory and understanding to spatial reasoning and intelligence. Through this survey, we aim to provide insights into interdisciplinary spatial intelligence research and inspire future studies.
☆ GlyTwin: Digital Twin for Glucose Control in Type 1 Diabetes Through Optimal Behavioral Modifications Using Patient-Centric Counterfactuals
Frequent and long-term exposure to hyperglycemia (i.e., high blood glucose) increases the risk of chronic complications such as neuropathy, nephropathy, and cardiovascular disease. Current technologies like continuous subcutaneous insulin infusion (CSII) and continuous glucose monitoring (CGM) primarily model specific aspects of glycemic control-like hypoglycemia prediction or insulin delivery. Similarly, most digital twin approaches in diabetes management simulate only physiological processes. These systems lack the ability to offer alternative treatment scenarios that support proactive behavioral interventions. To address this, we propose GlyTwin, a novel digital twin framework that uses counterfactual explanations to simulate optimal treatments for glucose regulation. Our approach helps patients and caregivers modify behaviors like carbohydrate intake and insulin dosing to avoid abnormal glucose events. GlyTwin generates behavioral treatment suggestions that proactively prevent hyperglycemia by recommending small adjustments to daily choices, reducing both frequency and duration of these events. Additionally, it incorporates stakeholder preferences into the intervention design, making recommendations patient-centric and tailored. We evaluate GlyTwin on AZT1D, a newly constructed dataset with longitudinal data from 21 type 1 diabetes (T1D) patients on automated insulin delivery systems over 26 days. Results show GlyTwin outperforms state-of-the-art counterfactual methods, generating 76.6% valid and 86% effective interventions. These findings demonstrate the promise of counterfactual-driven digital twins in delivering personalized healthcare.
☆ OVERLORD: Ultimate Scaling of DataLoader for Multi-Source Large Foundation Model Training
Modern frameworks for training large foundation models (LFMs) employ data loaders in a data parallel paradigm. While this design offers implementation simplicity, it introduces two fundamental challenges. First, due to the quadratic computational complexity of the attention operator, the non-uniform sample distribution over data-parallel ranks leads to a significant workload imbalance among loaders, which degrades the training efficiency. This paradigm also impedes the implementation of data mixing algorithms (e.g., curriculum learning) over different datasets. Second, to acquire a broad range of capability, LFMs training ingests data from diverse sources, each with distinct file access states. Colocating massive datasets within loader instances can easily exceed local pod memory capacity. Additionally, heavy sources with higher transformation latency require larger worker pools, further exacerbating memory consumption. We present OVERLORD, an industrial-grade distributed data loading architecture with three innovations: (1) A centralized and declarative data plane, which facilitates elastic data orchestration strategy, such as long-short context, multimodal, and curriculum learning; (2) Disaggregated multisource preprocessing through role-specific actors, i.e., Source Loaders and Data Constructors, leveraging autoscaling for Source Loaders towards heterogeneous and evolving source preprocessing cost; (3) Shadow Loaders with differential checkpointing for uninterrupted fault recovery. Deployed on production clusters scaling to multi-thousand GPU, OVERLORD achieves: (1) 4.5x end-to-end training throughput improvement, (2) a minimum 3.6x reduction in CPU memory usage, with further improvements to be added in later experiments.
☆ StruPhantom: Evolutionary Injection Attacks on Black-Box Tabular Agents Powered by Large Language Models
The proliferation of autonomous agents powered by large language models (LLMs) has revolutionized popular business applications dealing with tabular data, i.e., tabular agents. Although LLMs are observed to be vulnerable against prompt injection attacks from external data sources, tabular agents impose strict data formats and predefined rules on the attacker's payload, which are ineffective unless the agent navigates multiple layers of structural data to incorporate the payload. To address the challenge, we present a novel attack termed StruPhantom which specifically targets black-box LLM-powered tabular agents. Our attack designs an evolutionary optimization procedure which continually refines attack payloads via the proposed constrained Monte Carlo Tree Search augmented by an off-topic evaluator. StruPhantom helps systematically explore and exploit the weaknesses of target applications to achieve goal hijacking. Our evaluation validates the effectiveness of StruPhantom across various LLM-based agents, including those on real-world platforms, and attack scenarios. Our attack achieves over 50% higher success rates than baselines in enforcing the application's response to contain phishing links or malicious codes.
comment: Work in Progress
☆ SafeSpeech: Robust and Universal Voice Protection Against Malicious Speech Synthesis USENIX Security 2025
Speech synthesis technology has brought great convenience, while the widespread usage of realistic deepfake audio has triggered hazards. Malicious adversaries may unauthorizedly collect victims' speeches and clone a similar voice for illegal exploitation (\textit{e.g.}, telecom fraud). However, the existing defense methods cannot effectively prevent deepfake exploitation and are vulnerable to robust training techniques. Therefore, a more effective and robust data protection method is urgently needed. In response, we propose a defensive framework, \textit{\textbf{SafeSpeech}}, which protects the users' audio before uploading by embedding imperceptible perturbations on original speeches to prevent high-quality synthetic speech. In SafeSpeech, we devise a robust and universal proactive protection technique, \textbf{S}peech \textbf{PE}rturbative \textbf{C}oncealment (\textbf{SPEC}), that leverages a surrogate model to generate universally applicable perturbation for generative synthetic models. Moreover, we optimize the human perception of embedded perturbation in terms of time and frequency domains. To evaluate our method comprehensively, we conduct extensive experiments across advanced models and datasets, both subjectively and objectively. Our experimental results demonstrate that SafeSpeech achieves state-of-the-art (SOTA) voice protection effectiveness and transferability and is highly robust against advanced adaptive adversaries. Moreover, SafeSpeech has real-time capability in real-world tests. The source code is available at \href{https://github.com/wxzyd123/SafeSpeech}{https://github.com/wxzyd123/SafeSpeech}.
comment: Accepted to USENIX Security 2025
☆ Offline Dynamic Inventory and Pricing Strategy: Addressing Censored and Dependent Demand
In this paper, we study the offline sequential feature-based pricing and inventory control problem where the current demand depends on the past demand levels and any demand exceeding the available inventory is lost. Our goal is to leverage the offline dataset, consisting of past prices, ordering quantities, inventory levels, covariates, and censored sales levels, to estimate the optimal pricing and inventory control policy that maximizes long-term profit. While the underlying dynamic without censoring can be modeled by Markov decision process (MDP), the primary obstacle arises from the observed process where demand censoring is present, resulting in missing profit information, the failure of the Markov property, and a non-stationary optimal policy. To overcome these challenges, we first approximate the optimal policy by solving a high-order MDP characterized by the number of consecutive censoring instances, which ultimately boils down to solving a specialized Bellman equation tailored for this problem. Inspired by offline reinforcement learning and survival analysis, we propose two novel data-driven algorithms to solving these Bellman equations and, thus, estimate the optimal policy. Furthermore, we establish finite sample regret bounds to validate the effectiveness of these algorithms. Finally, we conduct numerical experiments to demonstrate the efficacy of our algorithms in estimating the optimal policy. To the best of our knowledge, this is the first data-driven approach to learning optimal pricing and inventory control policies in a sequential decision-making environment characterized by censored and dependent demand. The implementations of the proposed algorithms are available at https://github.com/gundemkorel/Inventory_Pricing_Control
☆ Efficient Multi-Task Modeling through Automated Fusion of Trained Models
Although multi-task learning is widely applied in intelligent services, traditional multi-task modeling methods often require customized designs based on specific task combinations, resulting in a cumbersome modeling process. Inspired by the rapid development and excellent performance of single-task models, this paper proposes an efficient multi-task modeling method that can automatically fuse trained single-task models with different structures and tasks to form a multi-task model. As a general framework, this method allows modelers to simply prepare trained models for the required tasks, simplifying the modeling process while fully utilizing the knowledge contained in the trained models. This eliminates the need for excessive focus on task relationships and model structure design. To achieve this goal, we consider the structural differences among various trained models and employ model decomposition techniques to hierarchically decompose them into multiple operable model components. Furthermore, we have designed an Adaptive Knowledge Fusion (AKF) module based on Transformer, which adaptively integrates intra-task and inter-task knowledge based on model components. Through the proposed method, we achieve efficient and automated construction of multi-task models, and its effectiveness is verified through extensive experiments on three datasets.
☆ See or Recall: A Sanity Check for the Role of Vision in Solving Visualization Question Answer Tasks with Multimodal LLMs
Recent developments in multimodal large language models (MLLM) have equipped language models to reason about vision and language jointly. This permits MLLMs to both perceive and answer questions about data visualization across a variety of designs and tasks. Applying MLLMs to a broad range of visualization tasks requires us to properly evaluate their capabilities, and the most common way to conduct evaluation is through measuring a model's visualization reasoning capability, analogous to how we would evaluate human understanding of visualizations (e.g., visualization literacy). However, we found that in the context of visualization question answering (VisQA), how an MLLM perceives and reasons about visualizations can be fundamentally different from how humans approach the same problem. During the evaluation, even without visualization, the model could correctly answer a substantial portion of the visualization test questions, regardless of whether any selection options were provided. We hypothesize that the vast amount of knowledge encoded in the language model permits factual recall that supersedes the need to seek information from the visual signal. It raises concerns that the current VisQA evaluation may not fully capture the models' visualization reasoning capabilities. To address this, we propose a comprehensive sanity check framework that integrates a rule-based decision tree and a sanity check table to disentangle the effects of "seeing" (visual processing) and "recall" (reliance on prior knowledge). This validates VisQA datasets for evaluation, highlighting where models are truly "seeing", positively or negatively affected by the factual recall, or relying on inductive biases for question answering. Our study underscores the need for careful consideration in designing future visualization understanding studies when utilizing MLLMs.
☆ Training Small Reasoning LLMs with Cognitive Preference Alignment
The reasoning capabilities of large language models (LLMs), such as OpenAI's o1 and DeepSeek-R1, have seen substantial advancements through deep thinking. However, these enhancements come with significant resource demands, underscoring the need to explore strategies to train effective reasoning LLMs with far fewer parameters. A critical challenge is that smaller models have different capacities and cognitive trajectories than their larger counterparts. Hence, direct distillation of chain-of-thought (CoT) results from large LLMs to smaller ones can be sometimes ineffective and requires a huge amount of annotated data. In this paper, we introduce a novel framework called Critique-Rethink-Verify (CRV), designed for training smaller yet powerful reasoning LLMs. Our CRV framework consists of multiple LLM agents, each specializing in unique abilities: (i) critiquing the CoTs according to the cognitive capabilities of smaller models, (ii) rethinking and refining these CoTs based on the critiques, and (iii) verifying the correctness of the refined results. We further propose the cognitive preference optimization (CogPO) algorithm to enhance the reasoning abilities of smaller models by aligning thoughts of these models with their cognitive capacities. Comprehensive evaluations on challenging reasoning benchmarks demonstrate the efficacy of CRV and CogPO, which outperforms other training methods by a large margin.
☆ Multi-task Federated Learning with Encoder-Decoder Structure: Enabling Collaborative Learning Across Different Tasks
Federated learning has been extensively studied and applied due to its ability to ensure data security in distributed environments while building better models. However, clients participating in federated learning still face limitations, as clients with different structures or tasks cannot participate in learning together. In view of this, constructing a federated learning framework that allows collaboration between clients with different model structures and performing different tasks, enabling them to share valuable knowledge to enhance model efficiency, holds significant practical implications for the widespread application of federated learning. To achieve this goal, we propose a multi-task federated learning with encoder-decoder structure (M-Fed). Specifically, given the widespread adoption of the encoder-decoder architecture in current models, we leverage this structure to share intra-task knowledge through traditional federated learning methods and extract general knowledge from the encoder to achieve cross-task knowledge sharing. The training process is similar to traditional federated learning, and we incorporate local decoder and global decoder information into the loss function. The local decoder iteratively updates and gradually approaches the global decoder until sufficient cross-task knowledge sharing is achieved. Our method is lightweight and modular, demonstrating innovation compared to previous research. It enables clients performing different tasks to share general knowledge while maintaining the efficiency of traditional federated learning systems. We conducted experiments on two widely used benchmark datasets to verify the feasibility of M-Fed and compared it with traditional methods. The experimental results demonstrate the effectiveness of M-Fed in multi-task federated learning.
♻ ☆ How Do Recommendation Models Amplify Popularity Bias? An Analysis from the Spectral Perspective
Recommendation Systems (RS) are often plagued by popularity bias. When training a recommendation model on a typically long-tailed dataset, the model tends to not only inherit this bias but often exacerbate it, resulting in over-representation of popular items in the recommendation lists. This study conducts comprehensive empirical and theoretical analyses to expose the root causes of this phenomenon, yielding two core insights: 1) Item popularity is memorized in the principal spectrum of the score matrix predicted by the recommendation model; 2) The dimension collapse phenomenon amplifies the relative prominence of the principal spectrum, thereby intensifying the popularity bias. Building on these insights, we propose a novel debiasing strategy that leverages a spectral norm regularizer to penalize the magnitude of the principal singular value. We have developed an efficient algorithm to expedite the calculation of the spectral norm by exploiting the spectral property of the score matrix. Extensive experiments across seven real-world datasets and three testing paradigms have been conducted to validate the superiority of the proposed method.
comment: 14 pages, 7 figures
♻ ☆ Learning Free Token Reduction for Multi-Modal Large Language Models
Vision-Language Models (VLMs) have achieved remarkable success across a range of multimodal tasks; however, their practical deployment is often constrained by high computational costs and prolonged inference times. Since the vision modality typically carries more information than the text modality, compressing visual prompts offers a promising solution to alleviate these challenges. Existing approaches predominantly focus on refining model architectures or directly reducing the number of visual tokens. However, these methods often compromise inference performance due to a lack of consideration for the unique spatial and temporal characteristics of visual data. In this work, we propose a token compression paradigm that operates on both spatial and temporal dimensions. Our approach includes a learning-free, plug-and-play compression pipeline that can be seamlessly integrated into most Multimodal Large Language Model (MLLM) frameworks. By leveraging this method, we enhance the model inference capability while simultaneously reducing its computational cost. Experimental results on the Video-QA task demonstrate the effectiveness of the proposed approach, showcasing significant improvements in efficiency without sacrificing performance.
♻ ☆ Visual Chronicles: Using Multimodal LLMs to Analyze Massive Collections of Images
We present a system using Multimodal LLMs (MLLMs) to analyze a large database with tens of millions of images captured at different times, with the aim of discovering patterns in temporal changes. Specifically, we aim to capture frequent co-occurring changes ("trends") across a city over a certain period. Unlike previous visual analyses, our analysis answers open-ended queries (e.g., "what are the frequent types of changes in the city?") without any predetermined target subjects or training labels. These properties cast prior learning-based or unsupervised visual analysis tools unsuitable. We identify MLLMs as a novel tool for their open-ended semantic understanding capabilities. Yet, our datasets are four orders of magnitude too large for an MLLM to ingest as context. So we introduce a bottom-up procedure that decomposes the massive visual analysis problem into more tractable sub-problems. We carefully design MLLM-based solutions to each sub-problem. During experiments and ablation studies with our system, we find it significantly outperforms baselines and is able to discover interesting trends from images captured in large cities (e.g., "addition of outdoor dining,", "overpass was painted blue," etc.). See more results and interactive demos at https://boyangdeng.com/visual-chronicles.
comment: Project page: https://boyangdeng.com/visual-chronicles , second and third listed authors have equal contributions
♻ ☆ Predicting Crack Nucleation and Propagation in Brittle Materials Using Deep Operator Networks with Diverse Trunk Architectures
Phase-field modeling reformulates fracture problems as energy minimization problems and enables a comprehensive characterization of the fracture process, including crack nucleation, propagation, merging, and branching, without relying on ad-hoc assumptions. However, the numerical solution of phase-field fracture problems is characterized by a high computational cost. To address this challenge, in this paper, we employ a deep neural operator (DeepONet) consisting of a branch network and a trunk network to solve brittle fracture problems. We explore three distinct approaches that vary in their trunk network configurations. In the first approach, we demonstrate the effectiveness of a two-step DeepONet, which results in a simplification of the learning task. In the second approach, we employ a physics-informed DeepONet, whereby the mathematical expression of the energy is integrated into the trunk network's loss to enforce physical consistency. The integration of physics also results in a substantially smaller data size needed for training. In the third approach, we replace the neural network in the trunk with a Kolmogorov-Arnold Network and train it without the physics loss. Using these methods, we model crack nucleation in a one-dimensional homogeneous bar under prescribed end displacements, as well as crack propagation and branching in single edge-notched specimens with varying notch lengths subjected to tensile and shear loading. We show that the networks predict the solution fields accurately, and the error in the predicted fields is localized near the crack.
comment: 25 pages, 21 figures
♻ ☆ Towards Safer Chatbots: A Framework for Policy Compliance Evaluation of Custom GPTs
Large Language Models (LLMs) have gained unprecedented prominence, achieving widespread adoption across diverse domains and integrating deeply into society. The capability to fine-tune general-purpose LLMs, such as Generative Pre-trained Transformers (GPT), for specific tasks has facilitated the emergence of numerous Custom GPTs. These tailored models are increasingly made available through dedicated marketplaces, such as OpenAI's GPT Store. However, their black-box nature introduces significant safety and compliance risks. In this work, we present a scalable framework for the automated evaluation of Custom GPTs against OpenAI's usage policies, which define the permissible behaviors of these systems. Our framework integrates three core components: (1) automated discovery and data collection of models from the GPT store, (2) a red-teaming prompt generator tailored to specific policy categories and the characteristics of each target GPT, and (3) an LLM-as-a-judge technique to analyze each prompt-response pair for potential policy violations. We validate our framework with a manually annotated ground truth, and evaluate it through a large-scale study with 782 Custom GPTs across three categories: Romantic, Cybersecurity, and Academic GPTs. Our manual annotation process achieved an F1 score of 0.975 in identifying policy violations, confirming the reliability of the framework's assessments. The results reveal that 58.7% of the analyzed models exhibit indications of non-compliance, exposing weaknesses in the GPT store's review and approval processes. Furthermore, our findings indicate that a model's popularity does not correlate with compliance, and non-compliance issues largely stem from behaviors inherited from base models rather than user-driven customizations. We believe this approach is extendable to other chatbot platforms and policy domains, improving LLM-based systems safety.
♻ ☆ COMPASS: Computational Mapping of Patient-Therapist Alliance Strategies with Language Modeling
The therapeutic working alliance is a critical predictor of psychotherapy success. Traditionally, working alliance assessment relies on questionnaires completed by both therapists and patients. In this paper, we present COMPASS, a novel framework to directly infer the therapeutic working alliance from the natural language used in psychotherapy sessions. Our approach leverages advanced large language models (LLMs) to analyze session transcripts and map them to distributed representations. These representations capture the semantic similarities between the dialogues and psychometric instruments, such as the Working Alliance Inventory. Analyzing a dataset of over 950 sessions spanning diverse psychiatric conditions -- including anxiety (N=498), depression (N=377), schizophrenia (N=71), and suicidal tendencies (N=12) -- collected between 1970 and 2012, we demonstrate the effectiveness of our method in providing fine-grained mapping of patient-therapist alignment trajectories, offering interpretable insights for clinical practice, and identifying emerging patterns related to the condition being treated. By employing various deep learning-based topic modeling techniques in combination with prompting generative language models, we analyze the topical characteristics of different psychiatric conditions and how these topics evolve during each turn of the conversation. This integrated framework enhances the understanding of therapeutic interactions, enables timely feedback for therapists on the quality of therapeutic relationships, and provides clear, actionable insights to improve the effectiveness of psychotherapy.
comment: Translational Psychiatry, in press. This work extends our research series in computational psychiatry (e.g auto annotation in arXiv:2204.05522, topic extraction in arXiv:2204.10189, and diagnosis in arXiv:2210.15603) with the introduction of LLMs to complete the full cycle of interpreting and understanding psychotherapy strategies as a comprehensive analytical framework
♻ ☆ Hatred Stems from Ignorance! Distillation of the Persuasion Modes in Countering Conversational Hate Speech
Examining the factors that the counterspeech uses are at the core of understanding the optimal methods for confronting hate speech online. Various studies have assessed the emotional base factors used in counter speech, such as emotional empathy, offensiveness, and hostility. To better understand the counterspeech used in conversations, this study distills persuasion modes into reason, emotion, and credibility and evaluates their use in two types of conversation interactions: closed (multi-turn) and open (single-turn) concerning racism, sexism, and religious bigotry. The evaluation covers the distinct behaviors seen with human-sourced as opposed to machine-generated counterspeech. It also assesses the interplay between the stance taken and the mode of persuasion seen in the counterspeech. Notably, we observe nuanced differences in the counterspeech persuasion modes used in open and closed interactions, especially in terms of the topic, with a general tendency to use reason as a persuasion mode to express the counterpoint to hate comments. The machine-generated counterspeech tends to exhibit an emotional persuasion mode, while human counters lean toward reason. Furthermore, our study shows that reason tends to obtain more supportive replies than other persuasion modes. The findings highlight the potential for incorporating persuasion modes into studies about countering hate speech, as they can serve as an optimal means of explainability and pave the way for the further adoption of the reply's stance and the role it plays in assessing what comprises the optimal counterspeech.
comment: Accepted to appear @ ICWSM 2025. The link to the camera-ready paper will be added soon
♻ ☆ Towards Fairness for the Right Reasons: Using Saliency Maps to Evaluate Bias Removal in Neural Networks
The widespread adoption of machine learning systems has raised critical concerns about fairness and bias, making mitigating harmful biases essential for AI development. In this paper, we investigate the relationship between fairness improvement and the removal of harmful biases in neural networks applied to computer vision tasks. First, we introduce a set of novel XAI-based metrics that analyze saliency maps to assess shifts in a model's decision-making process. Then, we demonstrate that successful debiasing methods systematically redirect model focus away from protected attributes. Additionally, we show that techniques originally developed for artifact removal can be effectively repurposed for fairness. These findings underscore the importance of ensuring that models are fair for the right reasons, contributing to the development of more ethical and trustworthy AI systems.
♻ ☆ Explanation-Driven Interventions for Artificial Intelligence Model Customization: Empowering End-Users to Tailor Black-Box AI in Rhinocytology
The integration of Artificial Intelligence (AI) in modern society is transforming how individuals perform tasks. In high-risk domains, ensuring human control over AI systems remains a key design challenge. This article presents a novel End-User Development (EUD) approach for black-box AI models, enabling users to edit explanations and influence future predictions through targeted interventions. By combining explainability, user control, and model adaptability, the proposed method advances Human-Centered AI (HCAI), promoting a symbiotic relationship between humans and adaptive, user-tailored AI systems.
comment: Second version (11 pages, 8 of content)
♻ ☆ Embedding Ontologies via Incorporating Extensional and Intensional Knowledge
Ontologies contain rich knowledge within domain, which can be divided into two categories, namely extensional knowledge and intensional knowledge. Extensional knowledge provides information about the concrete instances that belong to specific concepts in the ontology, while intensional knowledge details inherent properties, characteristics, and semantic associations among concepts. However, existing ontology embedding approaches fail to take both extensional knowledge and intensional knowledge into fine consideration simultaneously. In this paper, we propose a novel ontology embedding approach named EIKE (Extensional and Intensional Knowledge Embedding) by representing ontologies in two spaces, called extensional space and intensional space. EIKE presents a unified framework for embedding instances, concepts and their relations in an ontology, applying a geometry-based method to model extensional knowledge and a pretrained language model to model intensional knowledge, which can capture both structure information and textual information. Experimental results show that EIKE significantly outperforms state-of-the-art methods in three datasets for both triple classification and link prediction, indicating that EIKE provides a more comprehensive and representative perspective of the domain.
♻ ☆ GAF: Gaussian Avatar Reconstruction from Monocular Videos via Multi-view Diffusion
We propose a novel approach for reconstructing animatable 3D Gaussian avatars from monocular videos captured by commodity devices like smartphones. Photorealistic 3D head avatar reconstruction from such recordings is challenging due to limited observations, which leaves unobserved regions under-constrained and can lead to artifacts in novel views. To address this problem, we introduce a multi-view head diffusion model, leveraging its priors to fill in missing regions and ensure view consistency in Gaussian splatting renderings. To enable precise viewpoint control, we use normal maps rendered from FLAME-based head reconstruction, which provides pixel-aligned inductive biases. We also condition the diffusion model on VAE features extracted from the input image to preserve facial identity and appearance details. For Gaussian avatar reconstruction, we distill multi-view diffusion priors by using iteratively denoised images as pseudo-ground truths, effectively mitigating over-saturation issues. To further improve photorealism, we apply latent upsampling priors to refine the denoised latent before decoding it into an image. We evaluate our method on the NeRSemble dataset, showing that GAF outperforms previous state-of-the-art methods in novel view synthesis. Furthermore, we demonstrate higher-fidelity avatar reconstructions from monocular videos captured on commodity devices.
comment: Paper Video: https://youtu.be/QuIYTljvhyg Project Page: https://tangjiapeng.github.io/projects/GAF
♻ ☆ Artificial Neural Network for Estimation of Physical Parameters of Sea Water using LiDAR Waveforms
Light Detection and Ranging (LiDAR) are fast emerging sensors in the field of Earth Observation. It is a remote sensing technology that utilizes laser beams to measure distances and create detailed three-dimensional representations of objects and environments. The potential of Full Waveform LiDAR is much greater than just height estimation and 3D reconstruction only. Overall shape of signal provides important information about properties of water body. However, the shape of FWL is unexplored as most LiDAR software work on point cloud by utilizing the maximum value within the waveform. Existing techniques in the field of LiDAR data analysis include depth estimation through inverse modeling and regression of logarithmic intensity and depth for approximating the attenuation coefficient. However, these methods suffer from limitations in accuracy. Depth estimation through inverse modeling provides only approximate values and does not account for variations in surface properties, while the regression approach for the attenuation coefficient is only able to generalize a value through several data points which lacks precision and may lead to significant errors in estimation. Additionally, there is currently no established modeling method available for predicting bottom reflectance. This research proposed a novel solution based on neural networks for parameter estimation in LIDAR data analysis. By leveraging the power of neural networks, the proposed solution successfully learned the inversion model, was able to do prediction of parameters such as depth, attenuation coefficient, and bottom reflectance. Performance of model was validated by testing it on real LiDAR data. In future, more data availability would enable more accuracy and reliability of such models.
comment: 19 pages
♻ ☆ Motion Control in Multi-Rotor Aerial Robots Using Deep Reinforcement Learning
This paper investigates the application of Deep Reinforcement (DRL) Learning to address motion control challenges in drones for additive manufacturing (AM). Drone-based additive manufacturing promises flexible and autonomous material deposition in large-scale or hazardous environments. However, achieving robust real-time control of a multi-rotor aerial robot under varying payloads and potential disturbances remains challenging. Traditional controllers like PID often require frequent parameter re-tuning, limiting their applicability in dynamic scenarios. We propose a DRL framework that learns adaptable control policies for multi-rotor drones performing waypoint navigation in AM tasks. We compare Deep Deterministic Policy Gradient (DDPG) and Twin Delayed Deep Deterministic Policy Gradient (TD3) within a curriculum learning scheme designed to handle increasing complexity. Our experiments show TD3 consistently balances training stability, accuracy, and success, particularly when mass variability is introduced. These findings provide a scalable path toward robust, autonomous drone control in additive manufacturing.
♻ ☆ Machine Learning and Transformers for Thyroid Carcinoma Diagnosis: A Review
The growing interest in developing smart diagnostic systems to help medical experts process extensive data for treating incurable diseases has been notable. In particular, the challenge of identifying thyroid cancer (TC) has seen progress with the use of machine learning (ML) and big data analysis, incorporating Transformers to evaluate TC prognosis and determine the risk of malignancy in individuals. This review article presents a summary of various studies on AI-based approaches, especially those employing Transformers, for diagnosing TC. It introduces a new categorization system for these methods based on artificial intelligence (AI) algorithms, the goals of the framework, and the computing environments used. Additionally, it scrutinizes and contrasts the available TC datasets by their features. The paper highlights the importance of AI instruments in aiding the diagnosis and treatment of TC through supervised, unsupervised, or mixed approaches, with a special focus on the ongoing importance of Transformers and large language models (LLMs) in medical diagnostics and disease management. It further discusses the progress made and the continuing obstacles in this area. Lastly, it explores future directions and focuses within this research field.
♻ ☆ Predict+Optimize Problem in Renewable Energy Scheduling
Predict+Optimize frameworks integrate forecasting and optimization to address real-world challenges such as renewable energy scheduling, where variability and uncertainty are critical factors. This paper benchmarks solutions from the IEEE-CIS Technical Challenge on Predict+Optimize for Renewable Energy Scheduling, focusing on forecasting renewable production and demand and optimizing energy cost. The competition attracted 49 participants in total. The top-ranked method employed stochastic optimization using LightGBM ensembles, and achieved at least a 2% reduction in energy costs compared to deterministic approaches, demonstrating that the most accurate point forecast does not necessarily guarantee the best performance in downstream optimization. The published data and problem setting establish a benchmark for further research into integrated forecasting-optimization methods for energy systems, highlighting the importance of considering forecast uncertainty in optimization models to achieve cost-effective and reliable energy management. The novelty of this work lies in its comprehensive evaluation of Predict+Optimize methodologies applied to a real-world renewable energy scheduling problem, providing insights into the scalability, generalizability, and effectiveness of the proposed solutions. Potential applications extend beyond energy systems to any domain requiring integrated forecasting and optimization, such as supply chain management, transportation planning, and financial portfolio optimization.
♻ ☆ Structuring Scientific Innovation: A Framework for Modeling and Discovering Impactful Knowledge Combinations
The emergence of large language models offers new possibilities for structured exploration of scientific knowledge. Rather than viewing scientific discovery as isolated ideas or content, we propose a structured approach that emphasizes the role of method combinations in shaping disruptive insights. Specifically, we investigate how knowledge unit--especially those tied to methodological design--can be modeled and recombined to yield research breakthroughs. Our proposed framework addresses two key challenges. First, we introduce a contrastive learning-based mechanism to identify distinguishing features of historically disruptive method combinations within problem-driven contexts. Second, we propose a reasoning-guided Monte Carlo search algorithm that leverages the chain-of-thought capability of LLMs to identify promising knowledge recombinations for new problem statements.Empirical studies across multiple domains show that the framework is capable of modeling the structural dynamics of innovation and successfully highlights combinations with high disruptive potential. This research provides a new path for computationally guided scientific ideation grounded in structured reasoning and historical data modeling.
♻ ☆ Improving Policy Optimization via $\varepsilon$-Retrain AAMAS 2025
We present $\varepsilon$-retrain, an exploration strategy encouraging a behavioral preference while optimizing policies with monotonic improvement guarantees. To this end, we introduce an iterative procedure for collecting retrain areas -- parts of the state space where an agent did not satisfy the behavioral preference. Our method switches between the typical uniform restart state distribution and the retrain areas using a decaying factor $\varepsilon$, allowing agents to retrain on situations where they violated the preference. We also employ formal verification of neural networks to provably quantify the degree to which agents adhere to these behavioral preferences. Experiments over hundreds of seeds across locomotion, power network, and navigation tasks show that our method yields agents that exhibit significant performance and sample efficiency improvements.
comment: Accepted at AAMAS 2025
♻ ☆ GIScience in the Era of Artificial Intelligence: A Research Agenda Towards Autonomous GIS
The advent of generative AI exemplified by large language models (LLMs) opens new ways to represent and compute geographic information and transcends the process of geographic knowledge production, driving geographic information systems (GIS) towards autonomous GIS. Leveraging LLMs as the decision core, autonomous GIS can independently generate and execute geoprocessing workflows to perform spatial analysis. In this vision paper, we further elaborate on the concept of autonomous GIS and present a conceptual framework that defines its five autonomous goals, five autonomous levels, five core functions, and three operational scales. We demonstrate how autonomous GIS could perform geospatial data retrieval, spatial analysis, and map making with four proof-of-concept GIS agents. We conclude by identifying critical challenges and future research directions, including fine-tuning and self-growing decision-cores, autonomous modeling, and examining the societal and practical implications of autonomous GIS. By establishing the groundwork for a paradigm shift in GIScience, this paper envisions a future where GIS moves beyond traditional workflows to autonomously reason, derive, innovate, and advance geospatial solutions to pressing global challenges. Meanwhile, as we design and deploy increasingly intelligent geospatial systems, we carry a responsibility to ensure they are developed in socially responsible ways, serve the public good, and support the continued value of human geographic insight in an AI-augmented future.
♻ ☆ Opinion: Revisiting synthetic data classifications from a privacy perspective
Synthetic data is emerging as a cost-effective solution necessary to meet the increasing data demands of AI development, created either from existing knowledge or derived from real data. The traditional classification of synthetic data types into hybrid, partial or fully synthetic datasets has limited value and does not reflect the ever-increasing methods to generate synthetic data. The generation method and their source jointly shape the characteristics of synthetic data, which in turn determines its practical applications. We make a case for an alternative approach to grouping synthetic data types that better reflect privacy perspectives in order to facilitate regulatory guidance in the generation and processing of synthetic data. This approach to classification provides flexibility to new advancements like deep generative methods and offers a more practical framework for future applications.
♻ ☆ Fourier Sliced-Wasserstein Embedding for Multisets and Measures
We present the Fourier Sliced-Wasserstein (FSW) embedding - a novel method to embed multisets and measures over $\mathbb{R}^d$ into Euclidean space. Our proposed embedding approximately preserves the sliced Wasserstein distance on distributions, thereby yielding geometrically meaningful representations that better capture the structure of the input. Moreover, it is injective on measures and bi-Lipschitz on multisets - a significant advantage over prevalent methods based on sum- or max-pooling, which are provably not bi-Lipschitz, and, in many cases, not even injective. The required output dimension for these guarantees is near-optimal: roughly $2 N d$, where $N$ is the maximal input multiset size. Furthermore, we prove that it is impossible to embed distributions over $\mathbb{R}^d$ into Euclidean space in a bi-Lipschitz manner. Thus, the metric properties of our embedding are, in a sense, the best possible. Through numerical experiments, we demonstrate that our method yields superior multiset representations that improve performance in practical learning tasks. Specifically, we show that (a) a simple combination of the FSW embedding with an MLP achieves state-of-the-art performance in learning the (non-sliced) Wasserstein distance; and (b) replacing max-pooling with the FSW embedding makes PointNet significantly more robust to parameter reduction, with only minor performance degradation even after a 40-fold reduction.
comment: This is an erroneous submission that duplicates arXiv:2405.16519. It has been withdrawn; please see arXiv:2405.16519 for the intended version and all future updates
♻ ☆ Look Before You Leap: Enhancing Attention and Vigilance Regarding Harmful Content with GuidelineLLM AAAI 2025
Despite being empowered with alignment mechanisms, large language models (LLMs) are increasingly vulnerable to emerging jailbreak attacks that can compromise their alignment mechanisms. This vulnerability poses significant risks to real-world applications. Existing work faces challenges in both training efficiency and generalization capabilities (i.e., Reinforcement Learning from Human Feedback and Red-Teaming). Developing effective strategies to enable LLMs to resist continuously evolving jailbreak attempts represents a significant challenge. To address this challenge, we propose a novel defensive paradigm called GuidelineLLM, which assists LLMs in recognizing queries that may have harmful content. Before LLMs respond to a query, GuidelineLLM first identifies potential risks associated with the query, summarizes these risks into guideline suggestions, and then feeds these guidelines to the responding LLMs. Importantly, our approach eliminates the necessity for additional safety fine-tuning of the LLMs themselves; only the GuidelineLLM requires fine-tuning. This characteristic enhances the general applicability of GuidelineLLM across various LLMs. Experimental results demonstrate that GuidelineLLM can significantly reduce the attack success rate (ASR) against LLM (an average reduction of 34.17\% ASR) while maintaining the usefulness of LLM in handling benign queries. The code is available at https://github.com/sqzhang-lazy/GuidelineLLM.
comment: AAAI 2025
♻ ☆ Command A: An Enterprise-Ready Large Language Model
In this report we describe the development of Command A, a powerful large language model purpose-built to excel at real-world enterprise use cases. Command A is an agent-optimised and multilingual-capable model, with support for 23 languages of global business, and a novel hybrid architecture balancing efficiency with top of the range performance. It offers best-in-class Retrieval Augmented Generation (RAG) capabilities with grounding and tool use to automate sophisticated business processes. These abilities are achieved through a decentralised training approach, including self-refinement algorithms and model merging techniques. We also include results for Command R7B which shares capability and architectural similarities to Command A. Weights for both models have been released for research purposes. This technical report details our original training pipeline and presents an extensive evaluation of our models across a suite of enterprise-relevant tasks and public benchmarks, demonstrating excellent performance and efficiency.
comment: 55 pages
♻ ☆ Vikhr: The Family of Open-Source Instruction-Tuned Large Language Models for Russian
There has been a surge in the development of various Large Language Models (LLMs). However, text generation for languages other than English often faces significant challenges, including poor generation quality and reduced computational performance due to the disproportionate representation of tokens in the model's vocabulary. In this work, we address these issues by developing a pipeline for the adaptation of English-oriented pre-trained models to other languages and constructing efficient bilingual LLMs. Using this pipeline, we construct Vikhr, a series of bilingual open-source instruction-following LLMs designed specifically for the Russian language. ``Vikhr'' refers to the name of the Mistral LLM series and means a ``strong gust of wind.'' Unlike previous Russian-language models that typically rely on LoRA adapters on top of English-oriented models, sacrificing performance for lower training costs, Vikhr features an adapted tokenizer vocabulary and undergoes the continued pre-training and instruction tuning of all weights. This not only enhances the model's performance but also significantly improves its computational and contextual efficiency. We also expanded the instruction datasets and corpora for continued pre-training. The model weights, instruction sets, and code are publicly available.
♻ ☆ Practical Evaluation of Copula-based Survival Metrics: Beyond the Independent Censoring Assumption
Conventional survival metrics, such as Harrell's concordance index and the Brier Score, rely on the independent censoring assumption for valid inference in the presence of right-censored data. However, when instances are censored for reasons related to the event of interest, this assumption no longer holds, as this kind of dependent censoring biases the marginal survival estimates of popular nonparametric estimators. In this paper, we propose three copula-based metrics to evaluate survival models in the presence of dependent censoring, and design a framework to create realistic, semi-synthetic datasets with dependent censoring to facilitate the evaluation of the metrics. Our empirical analyses in synthetic and semi-synthetic datasets show that our metrics can give error estimates that are closer to the true error, mainly in terms of prediction accuracy.
♻ ☆ Expressivity of Representation Learning on Continuous-Time Dynamic Graphs: An Information-Flow Centric Review
Graphs are ubiquitous in real-world applications, ranging from social networks to biological systems, and have inspired the development of Graph Neural Networks (GNNs) for learning expressive representations. While most research has centered on static graphs, many real-world scenarios involve dynamic, temporally evolving graphs, motivating the need for Continuous-Time Dynamic Graph (CTDG) models. This paper provides a comprehensive review of Graph Representation Learning (GRL) on CTDGs with a focus on Self-Supervised Representation Learning (SSRL). We introduce a novel theoretical framework that analyzes the expressivity of CTDG models through an Information-Flow (IF) lens, quantifying their ability to propagate and encode temporal and structural information. Leveraging this framework, we categorize existing CTDG methods based on their suitability for different graph types and application scenarios. Within the same scope, we examine the design of SSRL methods tailored to CTDGs, such as predictive and contrastive approaches, highlighting their potential to mitigate the reliance on labeled data. Empirical evaluations on synthetic and real-world datasets validate our theoretical insights, demonstrating the strengths and limitations of various methods across long-range, bi-partite and community-based graphs. This work offers both a theoretical foundation and practical guidance for selecting and developing CTDG models, advancing the understanding of GRL in dynamic settings.
comment: Accepted by TMLR. Source code: https://github.com/king/ctdg-info-flow
♻ ☆ PhD: A ChatGPT-Prompted Visual hallucination Evaluation Dataset CVPR 2025
Multimodal Large Language Models (MLLMs) hallucinate, resulting in an emerging topic of visual hallucination evaluation (VHE). This paper contributes a ChatGPT-Prompted visual hallucination evaluation Dataset (PhD) for objective VHE at a large scale. The essence of VHE is to ask an MLLM questions about specific images to assess its susceptibility to hallucination. Depending on what to ask (objects, attributes, sentiment, etc.) and how the questions are asked, we structure PhD along two dimensions, i.e. task and mode. Five visual recognition tasks, ranging from low-level (object / attribute recognition) to middle-level (sentiment / position recognition and counting), are considered. Besides a normal visual QA mode, which we term PhD-base, PhD also asks questions with specious context (PhD-sec) or with incorrect context ({PhD-icc), or with AI-generated counter common sense images (PhD-ccs). We construct PhD by a ChatGPT-assisted semi-automated pipeline, encompassing four pivotal modules: task-specific hallucinatory item (hitem) selection, hitem-embedded question generation, specious / incorrect context generation, and counter-common-sense (CCS) image generation. With over 14k daily images, 750 CCS images and 102k VQA triplets in total, PhD reveals considerable variability in MLLMs' performance across various modes and tasks, offering valuable insights into the nature of hallucination. As such, PhD stands as a potent tool not only for VHE but may also play a significant role in the refinement of MLLMs.
comment: Accepted by CVPR 2025, Highlight
♻ ☆ A Personalised 3D+t Mesh Generative Model for Unveiling Normal Heart Dynamics
Understanding the structure and motion of the heart is crucial for diagnosing and managing cardiovascular diseases, the leading cause of global death. There is wide variation in cardiac shape and motion patterns, that are influenced by demographic, anthropometric and disease factors. Unravelling the normal patterns of shape and motion, as well as understanding how each individual deviates from the norm, would facilitate accurate diagnosis and personalised treatment strategies. To this end, we developed a novel conditional generative model, MeshHeart, to learn the distribution of cardiac shape and motion patterns. MeshHeart is capable of generating 3D+t cardiac mesh sequences, taking into account clinical factors such as age, sex, weight and height. To model the high-dimensional and complex spatio-temporal mesh data, MeshHeart employs a geometric encoder to represent cardiac meshes in a latent space, followed by a temporal Transformer to model the motion dynamics of latent representations. Based on MeshHeart, we investigate the latent space of 3D+t cardiac mesh sequences and propose a novel distance metric termed latent delta, which quantifies the deviation of a real heart from its personalised normative pattern in the latent space. In experiments using a large dataset of 38,309 subjects, MeshHeart demonstrates a high performance in cardiac mesh sequence reconstruction and generation. Features defined in the latent space are highly discriminative for cardiac disease classification, whereas the latent delta exhibits strong correlation with clinical phenotypes in phenome-wide association studies. The codes and models of this study will be released to benefit further research on digital heart modelling.
comment: Accepted by Nature Machine Intelligence
♻ ☆ RULSurv: A probabilistic survival-based method for early censoring-aware prediction of remaining useful life in ball bearings
Predicting the remaining useful life (RUL) of ball bearings is an active area of research, where novel machine learning techniques are continuously being applied to predict degradation trends and anticipate failures before they occur. However, few studies have explicitly addressed the challenge of handling censored data, where information about a specific event (\eg mechanical failure) is incomplete or only partially observed. To address this issue, we introduce a novel and flexible method for early fault detection using Kullback-Leibler (KL) divergence and RUL estimation using survival analysis that naturally supports censored data. We demonstrate our approach in the XJTU-SY dataset using a 5-fold cross-validation strategy across three different operating conditions. When predicting the time to failure for bearings under the highest load (C1, 12.0 kN and 2100 RPM) with 25% random censoring, our approach achieves a mean absolute error (MAE) of 14.7 minutes (95% CI = 13.6-15.8) using a linear CoxPH model, and an MAE of 12.6 minutes (95% CI = 11.8-13.4) using a nonlinear Random Survival Forests model, compared to an MAE of 18.5 minutes (95% CI = 17.4-19.6) using a linear LASSO model that does not support censoring. Moreover, our approach achieves a mean cumulative relative accuracy (CRA) of 0.7586 over 5 bearings under the highest load, which improves over several state-of-the-art baselines. Our work highlights the importance of considering censored data as part of the model design when building predictive models for early fault detection and RUL estimation.
♻ ☆ MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation
Multimodal Large Language Models (MLLMs) excel in understanding complex language and visual data, enabling generalist robotic systems to interpret instructions and perform embodied tasks. Nevertheless, their real-world deployment is hindered by substantial computational and storage demands. Recent insights into the homogeneous patterns in the LLM layer have inspired sparsification techniques to address these challenges, such as early exit and token pruning. However, these methods often neglect the critical role of the final layers that encode the semantic information most relevant to downstream robotic tasks. Aligning with the recent breakthrough of the Shallow Brain Hypothesis (SBH) in neuroscience and the mixture of experts in model sparsification, we conceptualize each LLM layer as an expert and propose a Mixture-of-Layers Vision-Language-Action model (MoLe-VLA, or simply MoLe) architecture for dynamic LLM layer activation. We introduce a Spatial-Temporal Aware Router (STAR) for MoLe to selectively activate only parts of the layers based on the robot's current state, mimicking the brain's distinct signal pathways specialized for cognition and causal reasoning. Additionally, to compensate for the cognitive ability of LLMs lost in MoLe, we devise a Cognition Self-Knowledge Distillation (CogKD) framework. CogKD enhances the understanding of task demands and improves the generation of task-relevant action sequences by leveraging cognitive features. Extensive experiments conducted in both RLBench simulation and real-world environments demonstrate the superiority of MoLe-VLA in both efficiency and performance. Specifically, MoLe-VLA achieves an 8% improvement in the mean success rate across ten tasks while reducing computational costs by up to x5.6 compared to standard LLMs.
♻ ☆ Trustworthiness of Stochastic Gradient Descent in Distributed Learning
Distributed learning (DL) uses multiple nodes to accelerate training, enabling efficient optimization of large-scale models. Stochastic Gradient Descent (SGD), a key optimization algorithm, plays a central role in this process. However, communication bottlenecks often limit scalability and efficiency, leading to increasing adoption of compressed SGD techniques to alleviate these challenges. Despite addressing communication overheads, compressed SGD introduces trustworthiness concerns, as gradient exchanges among nodes are vulnerable to attacks like gradient inversion (GradInv) and membership inference attacks (MIA). The trustworthiness of compressed SGD remains unexplored, leaving important questions about its reliability unanswered. In this paper, we provide a trustworthiness evaluation of compressed versus uncompressed SGD. Specifically, we conducted empirical studies using GradInv attacks, revealing that compressed SGD demonstrates significantly higher resistance to privacy leakage compared to uncompressed SGD. In addition, our findings suggest that MIA may not be a reliable metric for assessing privacy risks in distributed learning.
♻ ☆ Joint semi-supervised and contrastive learning enables domain generalization and multi-domain segmentation
Despite their effectiveness, current deep learning models face challenges with images coming from different domains with varying appearance and content. We introduce SegCLR, a versatile framework designed to segment images across different domains, employing supervised and contrastive learning simultaneously to effectively learn from both labeled and unlabeled data. We demonstrate the superior performance of SegCLR through a comprehensive evaluation involving three diverse clinical datasets of 3D retinal Optical Coherence Tomography (OCT) images, for the slice-wise segmentation of fluids with various network configurations and verification across 10 different network initializations. In an unsupervised domain adaptation context, SegCLR achieves results on par with a supervised upper-bound model trained on the intended target domain. Notably, we discover that the segmentation performance of SegCLR framework is marginally impacted by the abundance of unlabeled data from the target domain, thereby we also propose an effective domain generalization extension of SegCLR, known also as zero-shot domain adaptation, which eliminates the need for any target domain information. This shows that our proposed addition of contrastive loss in standard supervised training for segmentation leads to superior models, inherently more generalizable to both in- and out-of-domain test data. We additionally propose a pragmatic solution for SegCLR deployment in realistic scenarios with multiple domains containing labeled data. Accordingly, our framework pushes the boundaries of deep-learning based segmentation in multi-domain applications, regardless of data availability - labeled, unlabeled, or nonexistent.
♻ ☆ Secure Text Mail Encryption with Generative Adversarial Networks
This work presents an encryption model based on Generative Adversarial Networks (GANs). Encryption of RTF-8 data is realized by dynamically generating decimal numbers that lead to the encryption and decryption of alphabetic strings in integer representation by simple addition rules, the modulus of the dimension of the considered alphabet. The binary numbers for the private dynamic keys correspond to the binary numbers of public reference keys, as defined by a specific GAN configuration. For reversible encryption with a bijective mapping between dynamic and reference keys, as defined by the GAN encryptor, secure text encryption can be achieved by transferring a GAN-encrypted public key along with the encrypted text from a sender to a receiver. Using the technique described above, secure text mail transfer can be realized through component-wise encryption and decryption of text mail strings, with total key sizes of up to $10^{8}$ bits that define random decimal numbers generated by the GAN. From the present model, we assert that encrypted texts can be transmitted more efficiently and securely than from RSA encryption, as long as users of the specific configuration of the GAN encryption model are unaware of the GAN encryptor circuit and configuration, respectively.
comment: 7 pages, 3 figures, one table; Preprint before publication
♻ ☆ AgentBreeder: Mitigating the AI Safety Impact of Multi-Agent Scaffolds via Self-Improvement
Scaffolding Large Language Models (LLMs) into multi-agent systems often improves performance on complex tasks, but the safety impact of such scaffolds has not been thoroughly explored. We introduce AgentBreeder, a framework for multi-objective self-improving evolutionary search over scaffolds. We evaluate discovered scaffolds on widely recognized reasoning, mathematics, and safety benchmarks and compare them with popular baselines. In 'blue' mode, we see a 79.4% average uplift in safety benchmark performance while maintaining or improving capability scores. In 'red' mode, we find adversarially weak scaffolds emerging concurrently with capability optimization. Our work demonstrates the risks of multi-agent scaffolding and provides a framework for mitigating them. Code is available at https://github.com/J-Rosser-UK/AgentBreeder.
♻ ☆ From Vulnerabilities to Remediation: A Systematic Literature Review of LLMs in Code Security
Large Language Models (LLMs) have emerged as powerful tools for automating various programming tasks, including security-related ones, such as detecting and fixing vulnerabilities. Despite their promising capabilities, when required to produce or modify pre-existing code, LLMs could introduce vulnerabilities unbeknown to the programmer. When analyzing code, they could miss clear vulnerabilities or signal nonexistent ones. In this Systematic Literature Review (SLR), we aim to investigate both the security benefits and potential drawbacks of using LLMs for a variety of code-related tasks. In particular, first we focus on the types of vulnerabilities that could be introduced by LLMs, when used for producing code. Second, we analyze the capabilities of LLMs to detect and fix vulnerabilities, in any given code, and how the prompting strategy of choice impacts their performance in these two tasks. Last, we provide an in-depth analysis on how data poisoning attacks on LLMs can impact performance in the aforementioned tasks.
♻ ☆ Tumor likelihood estimation on MRI prostate data by utilizing k-Space information
We present a novel preprocessing and prediction pipeline for the classification of magnetic resonance imaging (MRI) that takes advantage of the information rich complex valued k-Space. Using a publicly available MRI raw dataset with 312 subject and a total of 9508 slices, we show the advantage of utilizing the k-Space for better prostate cancer likelihood estimation in comparison to just using the magnitudinal information in the image domain, with an AUROC of $86.1\%\pm1.8\%$. Additionally, by using high undersampling rates and a simple principal component analysis (PCA) for coil compression, we reduce the time needed for reconstruction by avoiding the time intensive GRAPPA reconstruction algorithm. By using digital undersampling for our experiments, we show that scanning and reconstruction time could be reduced. Even with an undersampling factor of 16, our approach achieves meaningful results, with an AUROC of $71.4\%\pm2.9\%$, using the PCA coil combination and taking into account the k-Space information. With this study, we were able to show the feasibility of preserving phase and k-Space information, with consistent results. Besides preserving valuable information for further diagnostics, this approach can work without the time intensive ADC and reconstruction calculations, greatly reducing the post processing, as well as potential scanning time, increasing patient comfort and allowing a close to real-time prediction.
♻ ☆ BotArtist: Generic approach for bot detection in Twitter via semi-automatic machine learning pipeline
Twitter, as one of the most popular social networks, provides a platform for communication and online discourse. Unfortunately, it has also become a target for bots and fake accounts, resulting in the spread of false information and manipulation. This paper introduces a semi-automatic machine learning pipeline (SAMLP) designed to address the challenges associated with machine learning model development. Through this pipeline, we develop a comprehensive bot detection model named BotArtist, based on user profile features. SAMLP leverages nine distinct publicly available datasets to train the BotArtist model. To assess BotArtist's performance against current state-of-the-art solutions, we evaluate 35 existing Twitter bot detection methods, each utilizing a diverse range of features. Our comparative evaluation of BotArtist and these existing methods, conducted across nine public datasets under standardized conditions, reveals that the proposed model outperforms existing solutions by almost 10% in terms of F1-score, achieving an average score of 83.19% and 68.5% over specific and general approaches, respectively. As a result of this research, we provide one of the largest labeled Twitter bot datasets. The dataset contains extracted features combined with BotArtist predictions for 10,929,533 Twitter user profiles, collected via Twitter API during the 2022 Russo-Ukrainian War over a 16-month period. This dataset was created based on [Shevtsov et al., 2022a] where the original authors share anonymized tweets discussing the Russo-Ukrainian war, totaling 127,275,386 tweets. The combination of the existing textual dataset and the provided labeled bot and human profiles will enable future development of more advanced bot detection large language models in the post-Twitter API era.
♻ ☆ Secret Collusion among Generative AI Agents: Multi-Agent Deception via Steganography
Recent capability increases in large language models (LLMs) open up applications in which groups of communicating generative AI agents solve joint tasks. This poses privacy and security challenges concerning the unauthorised sharing of information, or other unwanted forms of agent coordination. Modern steganographic techniques could render such dynamics hard to detect. In this paper, we comprehensively formalise the problem of secret collusion in systems of generative AI agents by drawing on relevant concepts from both AI and security literature. We study incentives for the use of steganography, and propose a variety of mitigation measures. Our investigations result in a model evaluation framework that systematically tests capabilities required for various forms of secret collusion. We provide extensive empirical results across a range of contemporary LLMs. While the steganographic capabilities of current models remain limited, GPT-4 displays a capability jump suggesting the need for continuous monitoring of steganographic frontier model capabilities. We conclude by laying out a comprehensive research program to mitigate future risks of collusion between generative AI models.
♻ ☆ Integrating Reinforcement Learning and Model Predictive Control with Applications to Microgrids
This work proposes an approach that integrates reinforcement learning and model predictive control (MPC) to solve finite-horizon optimal control problems in mixed-logical dynamical systems efficiently. Optimization-based control of such systems with discrete and continuous decision variables entails the online solution of mixed-integer linear programs, which suffer from the curse of dimensionality. Our approach aims to mitigate this issue by decoupling the decision on the discrete variables from the decision on the continuous variables. In the proposed approach, reinforcement learning determines the discrete decision variables and simplifies the online optimization problem of the MPC controller from a mixed-integer linear program to a linear program, significantly reducing the computational time. A fundamental contribution of this work is the definition of the decoupled Q-function, which plays a crucial role in making the learning problem tractable in a combinatorial action space. We motivate the use of recurrent neural networks to approximate the decoupled Q-function and show how they can be employed in a reinforcement learning setting. Simulation experiments on a microgrid system using real-world data demonstrate that the proposed method substantially reduces the online computation time of MPC while maintaining high feasibility and low suboptimality.
♻ ☆ Task Memory Engine (TME): A Structured Memory Framework with Graph-Aware Extensions for Multi-Step LLM Agent Tasks
Large Language Models (LLMs) are increasingly used as autonomous agents for multi-step tasks. However, most existing frameworks fail to maintain a structured understanding of the task state, often relying on linear prompt concatenation or shallow memory buffers. This leads to brittle performance, frequent hallucinations, and poor long-range coherence. In this work, we propose the Task Memory Engine (TME), a lightweight and structured memory module that tracks task execution using a hierarchical Task Memory Tree (TMT). Each node in the tree corresponds to a task step, storing relevant input, output, status, and sub-task relationships. We introduce a prompt synthesis method that dynamically generates LLM prompts based on the active node path, significantly improving execution consistency and contextual grounding. Through case studies and comparative experiments on multi-step agent tasks, we demonstrate that TME leads to better task completion accuracy and more interpretable behavior with minimal implementation overhead. A reference implementation of the core TME components is available at https://github.com/biubiutomato/TME-Agent, including basic examples and structured memory integration. While the current implementation uses a tree-based structure, TME is designed to be graph-aware, supporting reusable substeps, converging task paths, and shared dependencies. This lays the groundwork for future DAG-based memory architectures.
comment: 14 pages, 5 figures. Preprint prepared for future submission. Includes implementation and token-efficiency analysis. Code at https://github.com/biubiutomato/TME-Agent
♻ ☆ Faster Vision Mamba is Rebuilt in Minutes via Merged Token Re-training
Vision Mamba has shown close to state of the art performance on computer vision tasks, drawing much interest in increasing it's efficiency. A promising approach is token reduction (that has been successfully implemented in ViTs). Pruning informative tokens in Mamba leads to a high loss of key knowledge and degraded performance. An alternative, of merging tokens preserves more information than pruning, also suffers for large compression ratios. Our key insight is that a quick round of retraining after token merging yeilds robust results across various compression ratios. Empirically, pruned Vims only drop up to 0.9% accuracy on ImageNet-1K, recovered by our proposed framework R-MeeTo in our main evaluation. We show how simple and effective the fast recovery can be achieved at minute-level, in particular, a 35.9% accuracy spike over 3 epochs of training on Vim-Ti. Moreover, Vim-Ti/S/B are re-trained within 5/7/17 minutes, and Vim-S only drops 1.3% with 1.2x (up to 1.5x) speed up in inference.
♻ ☆ Interpretable and Fair Mechanisms for Abstaining Classifiers ECML
Abstaining classifiers have the option to refrain from providing a prediction for instances that are difficult to classify. The abstention mechanism is designed to trade off the classifier's performance on the accepted data while ensuring a minimum number of predictions. In this setting, often fairness concerns arise when the abstention mechanism solely reduces errors for the majority groups of the data, resulting in increased performance differences across demographic groups. While there exist a bunch of methods that aim to reduce discrimination when abstaining, there is no mechanism that can do so in an explainable way. In this paper, we fill this gap by introducing Interpretable and Fair Abstaining Classifier IFAC, an algorithm that can reject predictions both based on their uncertainty and their unfairness. By rejecting possibly unfair predictions, our method reduces error and positive decision rate differences across demographic groups of the non-rejected data. Since the unfairness-based rejections are based on an interpretable-by-design method, i.e., rule-based fairness checks and situation testing, we create a transparent process that can empower human decision-makers to review the unfair predictions and make more just decisions for them. This explainable aspect is especially important in light of recent AI regulations, mandating that any high-risk decision task should be overseen by human experts to reduce discrimination risks.
comment: 25 pages, 8 figures. In: Machine Learning and Knowledge Discovery in Databases. Research Track. ECML PKDD 2024
♻ ☆ Improving Instruction-Following in Language Models through Activation Steering ICLR 2025
The ability to follow instructions is crucial for numerous real-world applications of language models. In pursuit of deeper insights and more powerful capabilities, we derive instruction-specific vector representations from language models and use them to steer models accordingly. These vectors are computed as the difference in activations between inputs with and without instructions, enabling a modular approach to activation steering. We demonstrate how this method can enhance model adherence to constraints such as output format, length, and word inclusion, providing inference-time control over instruction following. Our experiments across four models demonstrate how we can use the activation vectors to guide models to follow constraints even without explicit instructions and to enhance performance when instructions are present. Additionally, we explore the compositionality of activation steering, successfully applying multiple instructions simultaneously. Finally, we demonstrate that steering vectors computed on instruction-tuned models can transfer to improve base models. Our findings demonstrate that activation steering offers a practical and scalable approach for fine-grained control in language generation. Our code and data are available at https://github.com/microsoft/llm-steer-instruct.
comment: ICLR 2025
♻ ☆ Automated Verification of Equivalence Properties in Advanced Logic Programs -- Bachelor Thesis
With the increase in industrial applications using Answer Set Programming, the need for formal verification tools, particularly for critical applications, has also increased. During the program optimisation process, it would be desirable to have a tool which can automatically verify whether an optimised subprogram can replace the original subprogram. Formally this corresponds to the problem of verifying the strong equivalence of two programs. In order to do so, the translation tool anthem was developed. It can be used in conjunction with an automated theorem prover for classical logic to verify that two programs are strongly equivalent. With the current version of anthem, only the strong equivalence of positive programs with a restricted input language can be verified. This is a result of the translation $\tau^*$ implemented in anthem that produces formulas in the logic of here-and-there, which coincides with classical logic only for positive programs. This thesis extends anthem in order to overcome these limitations. First, the transformation $\sigma^*$ is presented, which transforms formulas from the logic of here-and-there to classical logic. A theorem formalises how $\sigma^*$ can be used to express equivalence in the logic of here-and-there in classical logic. Second, the translation $\tau^*$ is extended to programs containing pools. Another theorem shows how $\sigma^*$ can be combined with $\tau^*$ to express the strong equivalence of two programs in classical logic. With $\sigma^*$ and the extended $\tau^*$, it is possible to express the strong equivalence of logic programs containing negation, simple choices, and pools. Both the extended $\tau^*$ and $\sigma^*$ are implemented in a new version of anthem. Several examples of logic programs containing pools, negation, and simple choice rules, which the new version of anthem can translate to classical logic, are presented. Some a...
comment: Bachelor Thesis at the University of Potsdam
♻ ☆ Attention-Based Multiscale Temporal Fusion Network for Uncertain-Mode Fault Diagnosis in Multimode Processes
Fault diagnosis in multimode processes plays a critical role in ensuring the safe operation of industrial systems across multiple modes. It faces a great challenge yet to be addressed - that is, the significant distributional differences among monitoring data from multiple modes make it difficult for the models to extract shared feature representations related to system health conditions. In response to this problem, this paper introduces a novel method called attention-based multiscale temporal fusion network. The multiscale depthwise convolution and gated recurrent unit are employed to extract multiscale contextual local features and long-short-term features. Instance normalization is applied to suppress mode-specific information. Furthermore, a temporal attention mechanism is designed to focus on critical time points with higher cross-mode shared information, thereby enhancing the accuracy of fault diagnosis. The proposed model is applied to Tennessee Eastman process dataset and three-phase flow facility dataset. The experiments demonstrate that the proposed model achieves superior diagnostic performance and maintains a small model size. The source code will be available on GitHub at https://github.com/GuangqiangLi/AMTFNet.
comment: 31 pages,11 figures
♻ ☆ UniForm: A Unified Multi-Task Diffusion Transformer for Audio-Video Generation
With the rise of diffusion models, audio-video generation has been revolutionized. However, most existing methods rely on separate modules for each modality, with limited exploration of unified generative architectures. In addition, many are confined to a single task and small-scale datasets. To address these limitations, we first propose UniForm, a unified multi-task diffusion transformer that jointly generates audio and visual modalities in a shared latent space. A single diffusion process models both audio and video, capturing the inherent correlations between sound and vision. Second, we introduce task-specific noise schemes and task tokens, enabling a single model to support multiple tasks, including text-to-audio-video, audio-to-video, and video-to-audio generation. Furthermore, by leveraging large language models and a large-scale text-audio-video combined dataset, UniForm achieves greater generative diversity than prior approaches. Extensive experiments show that UniForm achieves the state-of-the-art performance across audio-video generation tasks, producing content that is both well-aligned and close to real-world data distributions. Our demos are available at https://uniform-t2av.github.io/.
comment: Our demos are available at https://uniform-t2av.github.io/
♻ ☆ Machine Unlearning in Hyperbolic vs. Euclidean Multimodal Contrastive Learning: Adapting Alignment Calibration to MERU
Machine unlearning methods have become increasingly important for selective concept removal in large pre-trained models. While recent work has explored unlearning in Euclidean contrastive vision-language models, the effectiveness of concept removal in hyperbolic spaces remains unexplored. This paper investigates machine unlearning in hyperbolic contrastive learning by adapting Alignment Calibration to MERU, a model that embeds images and text in hyperbolic space to better capture semantic hierarchies. Through systematic experiments and ablation studies, we demonstrate that hyperbolic geometry offers distinct advantages for concept removal, achieving near perfect forgetting with reasonable performance on retained concepts, particularly when scaling to multiple concept removal. Our approach introduces hyperbolic-specific components including entailment calibration and norm regularization that leverage the unique properties of hyperbolic space. Comparative analysis with Euclidean models reveals fundamental differences in unlearning dynamics, with hyperbolic unlearning reorganizing the semantic hierarchy while Euclidean approaches merely disconnect cross-modal associations. These findings not only advance machine unlearning techniques but also provide insights into the geometric properties that influence concept representation and removal in multimodal models. Source code available at https://github.com/alex-pv01/HAC
comment: Preprint
♻ ☆ Summarization Metrics for Spanish and Basque: Do Automatic Scores and LLM-Judges Correlate with Humans?
Studies on evaluation metrics and LLM-as-a-Judge models for automatic text summarization have largely been focused on English, limiting our understanding of their effectiveness in other languages. Through our new dataset BASSE (BAsque and Spanish Summarization Evaluation), we address this situation by collecting human judgments on 2,040 abstractive summaries in Basque and Spanish, generated either manually or by five LLMs with four different prompts. For each summary, annotators evaluated five criteria on a 5-point Likert scale: coherence, consistency, fluency, relevance, and 5W1H. We use these data to reevaluate traditional automatic metrics used for evaluating summaries, as well as several LLM-as-a-Judge models that show strong performance on this task in English. Our results show that currently proprietary judge LLMs have the highest correlation with human judgments, followed by criteria-specific automatic metrics, while open-sourced judge LLMs perform poorly. We release BASSE and our code publicly, along with the first large-scale Basque summarization dataset containing 22,525 news articles with their subheads.
♻ ☆ AMBER -- Advanced SegFormer for Multi-Band Image Segmentation: an application to Hyperspectral Imaging
Deep learning has revolutionized the field of hyperspectral image (HSI) analysis, enabling the extraction of complex spectral and spatial features. While convolutional neural networks (CNNs) have been the backbone of HSI classification, their limitations in capturing global contextual features have led to the exploration of Vision Transformers (ViTs). This paper introduces AMBER, an advanced SegFormer specifically designed for multi-band image segmentation. AMBER enhances the original SegFormer by incorporating three-dimensional convolutions, custom kernel sizes, and a Funnelizer layer. This architecture enables processing hyperspectral data directly, without requiring spectral dimensionality reduction during preprocessing. Our experiments, conducted on three benchmark datasets (Salinas, Indian Pines, and Pavia University) and on a dataset from the PRISMA satellite, show that AMBER outperforms traditional CNN-based methods in terms of Overall Accuracy, Kappa coefficient, and Average Accuracy on the first three datasets, and achieves state-of-the-art performance on the PRISMA dataset. These findings highlight AMBER's robustness, adaptability to both airborne and spaceborne data, and its potential as a powerful solution for remote sensing and other domains requiring advanced analysis of high-dimensional data.
comment: submitted to Neural Computing & Applications (Springer). Accepted with minor revisions
♻ ☆ V-LASIK: Consistent Glasses-Removal from Videos Using Synthetic Data
Diffusion-based generative models have recently shown remarkable image and video editing capabilities. However, local video editing, particularly removal of small attributes like glasses, remains a challenge. Existing methods either alter the videos excessively, generate unrealistic artifacts, or fail to perform the requested edit consistently throughout the video. In this work, we focus on consistent and identity-preserving removal of glasses in videos, using it as a case study for consistent local attribute removal in videos. Due to the lack of paired data, we adopt a weakly supervised approach and generate synthetic imperfect data, using an adjusted pretrained diffusion model. We show that despite data imperfection, by learning from our generated data and leveraging the prior of pretrained diffusion models, our model is able to perform the desired edit consistently while preserving the original video content. Furthermore, we exemplify the generalization ability of our method to other local video editing tasks by applying it successfully to facial sticker-removal. Our approach demonstrates significant improvement over existing methods, showcasing the potential of leveraging synthetic data and strong video priors for local video editing tasks.
♻ ☆ Hands-On: Segmenting Individual Signs from Continuous Sequences
This work tackles the challenge of continuous sign language segmentation, a key task with huge implications for sign language translation and data annotation. We propose a transformer-based architecture that models the temporal dynamics of signing and frames segmentation as a sequence labeling problem using the Begin-In-Out (BIO) tagging scheme. Our method leverages the HaMeR hand features, and is complemented with 3D Angles. Extensive experiments show that our model achieves state-of-the-art results on the DGS Corpus, while our features surpass prior benchmarks on BSLCorpus.
comment: Accepted in the 19th IEEE International Conference on Automatic Face and Gesture Recognition
♻ ☆ MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity AAAI 2025
Data-free quantization (DFQ) is a technique that creates a lightweight network from its full-precision counterpart without the original training data, often through a synthetic dataset. Although several DFQ methods have been proposed for vision transformer (ViT) architectures, they fail to achieve efficacy in low-bit settings. Examining the existing methods, we observe that their synthetic data produce misaligned attention maps, while those of the real samples are highly aligned. From this observation, we find that aligning attention maps of synthetic data helps improve the overall performance of quantized ViTs. Motivated by this finding, we devise MimiQ, a novel DFQ method designed for ViTs that enhances inter-head attention similarity. First, we generate synthetic data by aligning head-wise attention outputs from each spatial query patch. Then, we align the attention maps of the quantized network to those of the full-precision teacher by applying head-wise structural attention distillation. The experimental results show that the proposed method significantly outperforms baselines, setting a new state-of-the-art for ViT-DFQ. This paper is an extended version of our work published in the proceedings of AAAI 2025, including additional supplementary material.
comment: Published to AAAI 2025
♻ ☆ External Large Foundation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendation WWW
Ads recommendation is a prominent service of online advertising systems and has been actively studied. Recent studies indicate that scaling-up and advanced design of the recommendation model can bring significant performance improvement. However, with a larger model scale, such prior studies have a significantly increasing gap from industry as they often neglect two fundamental challenges in industrial-scale applications. First, training and inference budgets are restricted for the model to be served, exceeding which may incur latency and impair user experience. Second, large-volume data arrive in a streaming mode with data distributions dynamically shifting, as new users/ads join and existing users/ads leave the system. We propose the External Large Foundation Model (ExFM) framework to address the overlooked challenges. Specifically, we develop external distillation and a data augmentation system (DAS) to control the computational cost of training/inference while maintaining high performance. We design the teacher in a way like a foundation model (FM) that can serve multiple students as vertical models (VMs) to amortize its building cost. We propose Auxiliary Head and Student Adapter to mitigate the data distribution gap between FM and VMs caused by the streaming data issue. Comprehensive experiments on internal industrial-scale applications and public datasets demonstrate significant performance gain by ExFM.
comment: Accepted by the ACM Web Conference (WWW) 2025 Industrial Track as Oral Presentation
♻ ☆ Generative Modeling of Class Probability for Multi-Modal Representation Learning CVPR 2025
Multi-modal understanding plays a crucial role in artificial intelligence by enabling models to jointly interpret inputs from different modalities. However, conventional approaches such as contrastive learning often struggle with modality discrepancies, leading to potential misalignments. In this paper, we propose a novel class anchor alignment approach that leverages class probability distributions for multi-modal representation learning. Our method, Class-anchor-ALigned generative Modeling (CALM), encodes class anchors as prompts to generate and align class probability distributions for each modality, enabling more effective alignment. Furthermore, we introduce a cross-modal probabilistic variational autoencoder to model uncertainty in the alignment, enhancing the ability to capture deeper relationships between modalities and data variations. Extensive experiments on four benchmark datasets demonstrate that our approach significantly outperforms state-of-the-art methods, especially in out-of-domain evaluations. This highlights its superior generalization capabilities in multi-modal representation learning.
comment: To appear in CVPR 2025 (Highlight)
♻ ☆ Tailoring Adversarial Attacks on Deep Neural Networks for Targeted Class Manipulation Using DeepFool Algorithm
The susceptibility of deep neural networks (DNNs) to adversarial attacks undermines their reliability across numerous applications, underscoring the necessity for an in-depth exploration of these vulnerabilities and the formulation of robust defense strategies. The DeepFool algorithm by Moosavi-Dezfooli et al. (2016) represents a pivotal step in identifying minimal perturbations required to induce misclassification of input images. Nonetheless, its generic methodology falls short in scenarios necessitating targeted interventions. Additionally, previous research studies have predominantly concentrated on the success rate of attacks without adequately addressing the consequential distortion of images, the maintenance of image quality, or the confidence threshold required for misclassification. To bridge these gaps, we introduce the Enhanced Targeted DeepFool (ET DeepFool) algorithm, an evolution of DeepFool that not only facilitates the specification of desired misclassification targets but also incorporates a configurable minimum confidence score. Our empirical investigations demonstrate the superiority of this refined approach in maintaining the integrity of images and minimizing perturbations across a variety of DNN architectures. Unlike previous iterations, such as the Targeted DeepFool by Gajjar et al. (2022), our method grants unparalleled control over the perturbation process, enabling precise manipulation of model responses. Preliminary outcomes reveal that certain models, including AlexNet and the advanced Vision Transformer, display commendable robustness to such manipulations. This discovery of varying levels of model robustness, as unveiled through our confidence level adjustments, could have far-reaching implications for the field of image recognition. Our code is available at https://github.com/FazleLabib/et_deepfool.
comment: 18 pages, 5 figures. Accepted by Nature Scientific Reports
♻ ☆ Towards Developing Socially Compliant Automated Vehicles: Advances, Expert Insights, and A Conceptual Framework
Automated Vehicles (AVs) hold promise for revolutionizing transportation by improving road safety, traffic efficiency, and overall mobility. Despite the steady advancement in high-level AVs in recent years, the transition to full automation entails a period of mixed traffic, where AVs of varying automation levels coexist with human-driven vehicles (HDVs). Making AVs socially compliant and understood by human drivers is expected to improve the safety and efficiency of mixed traffic. Thus, ensuring AVs' compatibility with HDVs and social acceptance is crucial for their successful and seamless integration into mixed traffic. However, research in this critical area of developing Socially Compliant AVs (SCAVs) remains sparse. This study carries out the first comprehensive scoping review to assess the current state of the art in developing SCAVs, identifying key concepts, methodological approaches, and research gaps. An informal expert interview was also conducted to discuss the literature review results and identify critical research gaps and expectations towards SCAVs. Based on the scoping review and expert interview input, a conceptual framework is proposed for the development of SCAVs. The conceptual framework is evaluated using an online survey targeting researchers, technicians, policymakers, and other relevant professionals worldwide. The survey results provide valuable validation and insights, affirming the significance of the proposed conceptual framework in tackling the challenges of integrating AVs into mixed-traffic environments. Additionally, future research perspectives and suggestions are discussed, contributing to the research and development agenda of SCAVs.
comment: 58 pages, 13 figures, accepted by the Journal of Communications in Transportation Research
♻ ☆ ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy
Vision-Language-Action (VLA) models have shown substantial potential in real-world robotic manipulation. However, fine-tuning these models through supervised learning struggles to achieve robust performance due to limited, inconsistent demonstrations, especially in contact-rich environments. In this paper, we propose a reinforced fine-tuning approach for VLA models, named ConRFT, which consists of offline and online fine-tuning with a unified consistency-based training objective, to address these challenges. In the offline stage, our method integrates behavior cloning and Q-learning to effectively extract policy from a small set of demonstrations and stabilize value estimating. In the online stage, the VLA model is further fine-tuned via consistency policy, with human interventions to ensure safe exploration and high sample efficiency. We evaluate our approach on eight diverse real-world manipulation tasks. It achieves an average success rate of 96.3% within 45-90 minutes of online fine-tuning, outperforming prior supervised methods with a 144% improvement in success rate and 1.9x shorter episode length. This work highlights the potential of integrating reinforcement learning to enhance the performance of VLA models for real-world robotic applications. Videos and code are available at our project website https://cccedric.github.io/conrft/.
♻ ☆ Seeking Consistent Flat Minima for Better Domain Generalization via Refining Loss Landscapes
Domain generalization aims to learn a model from multiple training domains and generalize it to unseen test domains. Recent theory has shown that seeking the deep models, whose parameters lie in the flat minima of the loss landscape, can significantly reduce the out-of-domain generalization error. However, existing methods often neglect the consistency of loss landscapes in different domains, resulting in models that are not simultaneously in the optimal flat minima in all domains, which limits their generalization ability. To address this issue, this paper proposes an iterative Self-Feedback Training (SFT) framework to seek consistent flat minima that are shared across different domains by progressively refining loss landscapes during training. It alternatively generates a feedback signal by measuring the inconsistency of loss landscapes in different domains and refines these loss landscapes for greater consistency using this feedback signal. Benefiting from the consistency of the flat minima within these refined loss landscapes, our SFT helps achieve better out-of-domain generalization. Extensive experiments on DomainBed demonstrate superior performances of SFT when compared to state-of-the-art sharpness-aware methods and other prevalent DG baselines. On average across five DG benchmarks, SFT surpasses the sharpness-aware minimization by 2.6% with ResNet-50 and 1.5% with ViT-B/16, respectively. The code will be available soon.
♻ ☆ Robust Barycenter Estimation using Semi-Unbalanced Neural Optimal Transport
Aggregating data from multiple sources can be formalized as an Optimal Transport (OT) barycenter problem, which seeks to compute the average of probability distributions with respect to OT discrepancies. However, in real-world scenarios, the presence of outliers and noise in the data measures can significantly hinder the performance of traditional statistical methods for estimating OT barycenters. To address this issue, we propose a novel scalable approach for estimating the robust continuous barycenter, leveraging the dual formulation of the (semi-)unbalanced OT problem. To the best of our knowledge, this paper is the first attempt to develop an algorithm for robust barycenters under the continuous distribution setup. Our method is framed as a min-max optimization problem and is adaptable to general cost functions. We rigorously establish the theoretical underpinnings of the proposed method and demonstrate its robustness to outliers and class imbalance through a number of illustrative experiments. Our source code is publicly available at https://github.com/milenagazdieva/U-NOTBarycenters.
comment: 30 pages, 7 figures
♻ ☆ Neural Encoding and Decoding at Scale
Recent work has demonstrated that large-scale, multi-animal models are powerful tools for characterizing the relationship between neural activity and behavior. Current large-scale approaches, however, focus exclusively on either predicting neural activity from behavior (encoding) or predicting behavior from neural activity (decoding), limiting their ability to capture the bidirectional relationship between neural activity and behavior. To bridge this gap, we introduce a multimodal, multi-task model that enables simultaneous Neural Encoding and Decoding at Scale (NEDS). Central to our approach is a novel multi-task-masking strategy, which alternates between neural, behavioral, within-modality, and cross-modality masking. We pretrain our method on the International Brain Laboratory (IBL) repeated site dataset, which includes recordings from 83 animals performing the same visual decision-making task. In comparison to other large-scale models, we demonstrate that NEDS achieves state-of-the-art performance for both encoding and decoding when pretrained on multi-animal data and then fine-tuned on new animals. Surprisingly, NEDS's learned embeddings exhibit emergent properties: even without explicit training, they are highly predictive of the brain regions in each recording. Altogether, our approach is a step towards a foundation model of the brain that enables seamless translation between neural activity and behavior.
♻ ☆ LogLLM: Log-based Anomaly Detection Using Large Language Models
Software systems often record important runtime information in logs to help with troubleshooting. Log-based anomaly detection has become a key research area that aims to identify system issues through log data, ultimately enhancing the reliability of software systems. Traditional deep learning methods often struggle to capture the semantic information embedded in log data, which is typically organized in natural language. In this paper, we propose LogLLM, a log-based anomaly detection framework that leverages large language models (LLMs). LogLLM employs BERT for extracting semantic vectors from log messages, while utilizing Llama, a transformer decoder-based model, for classifying log sequences. Additionally, we introduce a projector to align the vector representation spaces of BERT and Llama, ensuring a cohesive understanding of log semantics. Unlike conventional methods that require log parsers to extract templates, LogLLM preprocesses log messages with regular expressions, streamlining the entire process. Our framework is trained through a novel three-stage procedure designed to enhance performance and adaptability. Experimental results across four public datasets demonstrate that LogLLM outperforms state-of-the-art methods. Even when handling unstable logs, it effectively captures the semantic meaning of log messages and detects anomalies accurately.
♻ ☆ Pre-training Generative Recommender with Multi-Identifier Item Tokenization
Generative recommendation autoregressively generates item identifiers to recommend potential items. Existing methods typically adopt a one-to-one mapping strategy, where each item is represented by a single identifier. However, this scheme poses issues, such as suboptimal semantic modeling for low-frequency items and limited diversity in token sequence data. To overcome these limitations, we propose MTGRec, which leverages Multi-identifier item Tokenization to augment token sequence data for Generative Recommender pre-training. Our approach involves two key innovations: multi-identifier item tokenization and curriculum recommender pre-training. For multi-identifier item tokenization, we leverage the RQ-VAE as the tokenizer backbone and treat model checkpoints from adjacent training epochs as semantically relevant tokenizers. This allows each item to be associated with multiple identifiers, enabling a single user interaction sequence to be converted into several token sequences as different data groups. For curriculum recommender pre-training, we introduce a curriculum learning scheme guided by data influence estimation, dynamically adjusting the sampling probability of each data group during recommender pre-training. After pre-training, we fine-tune the model using a single tokenizer to ensure accurate item identification for recommendation. Extensive experiments on three public benchmark datasets demonstrate that MTGRec significantly outperforms both traditional and generative recommendation baselines in terms of effectiveness and scalability.
♻ ☆ Universal Item Tokenization for Transferable Generative Recommendation
Recently, generative recommendation has emerged as a promising paradigm, attracting significant research attention. The basic framework involves an item tokenizer, which represents each item as a sequence of codes serving as its identifier, and a generative recommender that predicts the next item by autoregressively generating the target item identifier. However, in existing methods, both the tokenizer and the recommender are typically domain-specific, limiting their ability for effective transfer or adaptation to new domains. To this end, we propose UTGRec, a Universal item Tokenization approach for transferable Generative Recommendation. Specifically, we design a universal item tokenizer for encoding rich item semantics by adapting a multimodal large language model (MLLM). By devising tree-structured codebooks, we discretize content representations into corresponding codes for item tokenization. To effectively learn the universal item tokenizer on multiple domains, we introduce two key techniques in our approach. For raw content reconstruction, we employ dual lightweight decoders to reconstruct item text and images from discrete representations to capture general knowledge embedded in the content. For collaborative knowledge integration, we assume that co-occurring items are similar and integrate collaborative signals through co-occurrence alignment and reconstruction. Finally, we present a joint learning framework to pre-train and adapt the transferable generative recommender across multiple domains. Extensive experiments on four public datasets demonstrate the superiority of UTGRec compared to both traditional and generative recommendation baselines.
♻ ☆ Large language models could be rote learners
Multiple-choice question (MCQ) benchmarks are widely used for evaluating Large Language Models (LLMs), yet their reliability is undermined by benchmark contamination. In this study, we reframe contamination as an inherent aspect of learning and seek to disentangle genuine capability acquisition from superficial memorization in LLM evaluation. First, by analyzing model performance under different memorization conditions, we uncover a counterintuitive trend: LLMs perform worse on memorized MCQs than on non-memorized ones, indicating the coexistence of two distinct learning phenomena, i.e., rote memorization and genuine capability learning. To disentangle them, we propose TrinEval, a novel evaluation framework that reformulates MCQs into an alternative trinity format, reducing memorization while preserving knowledge assessment. Experiments validate TrinEval's effectiveness in reformulation, and its evaluation reveals that common LLMs may memorize by rote 20.5% of knowledge points (in MMLU on average).
comment: Work in Progress
♻ ☆ Multi-view Intent Learning and Alignment with Large Language Models for Session-based Recommendation
Session-based recommendation (SBR) methods often rely on user behavior data, which can struggle with the sparsity of session data, limiting performance. Researchers have identified that beyond behavioral signals, rich semantic information in item descriptions is crucial for capturing hidden user intent. While large language models (LLMs) offer new ways to leverage this semantic data, the challenges of session anonymity, short-sequence nature, and high LLM training costs have hindered the development of a lightweight, efficient LLM framework for SBR. To address the above challenges, we propose an LLM-enhanced SBR framework that integrates semantic and behavioral signals from multiple views. This two-stage framework leverages the strengths of both LLMs and traditional SBR models while minimizing training costs. In the first stage, we use multi-view prompts to infer latent user intentions at the session semantic level, supported by an intent localization module to alleviate LLM hallucinations. In the second stage, we align and unify these semantic inferences with behavioral representations, effectively merging insights from both large and small models. Extensive experiments on two real datasets demonstrate that the LLM4SBR framework can effectively improve model performance. We release our codes along with the baselines at https://github.com/tsinghua-fib-lab/LLM4SBR.
♻ ☆ Offline Safe Reinforcement Learning Using Trajectory Classification AAAI 2025
Offline safe reinforcement learning (RL) has emerged as a promising approach for learning safe behaviors without engaging in risky online interactions with the environment. Most existing methods in offline safe RL rely on cost constraints at each time step (derived from global cost constraints) and this can result in either overly conservative policies or violation of safety constraints. In this paper, we propose to learn a policy that generates desirable trajectories and avoids undesirable trajectories. To be specific, we first partition the pre-collected dataset of state-action trajectories into desirable and undesirable subsets. Intuitively, the desirable set contains high reward and safe trajectories, and undesirable set contains unsafe trajectories and low-reward safe trajectories. Second, we learn a policy that generates desirable trajectories and avoids undesirable trajectories, where (un)desirability scores are provided by a classifier learnt from the dataset of desirable and undesirable trajectories. This approach bypasses the computational complexity and stability issues of a min-max objective that is employed in existing methods. Theoretically, we also show our approach's strong connections to existing learning paradigms involving human feedback. Finally, we extensively evaluate our method using the DSRL benchmark for offline safe RL. Empirically, our method outperforms competitive baselines, achieving higher rewards and better constraint satisfaction across a wide variety of benchmark tasks.
comment: AAAI 2025
Computer Vision and Pattern Recognition 150
☆ FLOSS: Free Lunch in Open-vocabulary Semantic Segmentation
Recent Open-Vocabulary Semantic Segmentation (OVSS) models extend the CLIP model to segmentation while maintaining the use of multiple templates (e.g., a photo of , a sketch of a , etc.) for constructing class-wise averaged text embeddings, acting as a classifier. In this paper, we challenge this status quo and investigate the impact of templates for OVSS. Empirically, we observe that for each class, there exist single-template classifiers significantly outperforming the conventional averaged classifier. We refer to them as class-experts. Given access to unlabeled images and without any training involved, we estimate these experts by leveraging the class-wise prediction entropy of single-template classifiers, selecting as class-wise experts those which yield the lowest entropy. All experts, each specializing in a specific class, collaborate in a newly proposed fusion method to generate more accurate OVSS predictions. Our plug-and-play method, coined FLOSS, is orthogonal and complementary to existing OVSS methods, offering a ''free lunch'' to systematically improve OVSS without labels and additional training. Extensive experiments demonstrate that FLOSS consistently boosts state-of-the-art methods on various OVSS benchmarks. Moreover, the selected expert templates can generalize well from one dataset to others sharing the same semantic categories, yet exhibiting distribution shifts. Additionally, we obtain satisfactory improvements under a low-data regime, where only a few unlabeled images are available. Our code is available at https://github.com/yasserben/FLOSS .
comment: Project Page: https://yasserben.github.io/FLOSS/
☆ DNF-Avatar: Distilling Neural Fields for Real-time Animatable Avatar Relighting
Creating relightable and animatable human avatars from monocular videos is a rising research topic with a range of applications, e.g. virtual reality, sports, and video games. Previous works utilize neural fields together with physically based rendering (PBR), to estimate geometry and disentangle appearance properties of human avatars. However, one drawback of these methods is the slow rendering speed due to the expensive Monte Carlo ray tracing. To tackle this problem, we proposed to distill the knowledge from implicit neural fields (teacher) to explicit 2D Gaussian splatting (student) representation to take advantage of the fast rasterization property of Gaussian splatting. To avoid ray-tracing, we employ the split-sum approximation for PBR appearance. We also propose novel part-wise ambient occlusion probes for shadow computation. Shadow prediction is achieved by querying these probes only once per pixel, which paves the way for real-time relighting of avatars. These techniques combined give high-quality relighting results with realistic shadow effects. Our experiments demonstrate that the proposed student model achieves comparable or even better relighting results with our teacher model while being 370 times faster at inference time, achieving a 67 FPS rendering speed.
comment: 16 pages, 8 figures, Project pages: https://jzr99.github.io/DNF-Avatar/
☆ Decoupled Diffusion Sparks Adaptive Scene Generation
Controllable scene generation could reduce the cost of diverse data collection substantially for autonomous driving. Prior works formulate the traffic layout generation as predictive progress, either by denoising entire sequences at once or by iteratively predicting the next frame. However, full sequence denoising hinders online reaction, while the latter's short-sighted next-frame prediction lacks precise goal-state guidance. Further, the learned model struggles to generate complex or challenging scenarios due to a large number of safe and ordinal driving behaviors from open datasets. To overcome these, we introduce Nexus, a decoupled scene generation framework that improves reactivity and goal conditioning by simulating both ordinal and challenging scenarios from fine-grained tokens with independent noise states. At the core of the decoupled pipeline is the integration of a partial noise-masking training strategy and a noise-aware schedule that ensures timely environmental updates throughout the denoising process. To complement challenging scenario generation, we collect a dataset consisting of complex corner cases. It covers 540 hours of simulated data, including high-risk interactions such as cut-in, sudden braking, and collision. Nexus achieves superior generation realism while preserving reactivity and goal orientation, with a 40% reduction in displacement error. We further demonstrate that Nexus improves closed-loop planning by 20% through data augmentation and showcase its capability in safety-critical data generation.
☆ REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers
In this paper we tackle a fundamental question: "Can we train latent diffusion models together with the variational auto-encoder (VAE) tokenizer in an end-to-end manner?" Traditional deep-learning wisdom dictates that end-to-end training is often preferable when possible. However, for latent diffusion transformers, it is observed that end-to-end training both VAE and diffusion-model using standard diffusion-loss is ineffective, even causing a degradation in final performance. We show that while diffusion loss is ineffective, end-to-end training can be unlocked through the representation-alignment (REPA) loss -- allowing both VAE and diffusion model to be jointly tuned during the training process. Despite its simplicity, the proposed training recipe (REPA-E) shows remarkable performance; speeding up diffusion model training by over 17x and 45x over REPA and vanilla training recipes, respectively. Interestingly, we observe that end-to-end tuning with REPA-E also improves the VAE itself; leading to improved latent space structure and downstream generation performance. In terms of final performance, our approach sets a new state-of-the-art; achieving FID of 1.26 and 1.83 with and without classifier-free guidance on ImageNet 256 x 256. Code is available at https://end2end-diffusion.github.io.
☆ InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
We introduce InternVL3, a significant advancement in the InternVL series featuring a native multimodal pre-training paradigm. Rather than adapting a text-only large language model (LLM) into a multimodal large language model (MLLM) that supports visual inputs, InternVL3 jointly acquires multimodal and linguistic capabilities from both diverse multimodal data and pure-text corpora during a single pre-training stage. This unified training paradigm effectively addresses the complexities and alignment challenges commonly encountered in conventional post-hoc training pipelines for MLLMs. To further improve performance and scalability, InternVL3 incorporates variable visual position encoding (V2PE) to support extended multimodal contexts, employs advanced post-training techniques such as supervised fine-tuning (SFT) and mixed preference optimization (MPO), and adopts test-time scaling strategies alongside an optimized training infrastructure. Extensive empirical evaluations demonstrate that InternVL3 delivers superior performance across a wide range of multi-modal tasks. In particular, InternVL3-78B achieves a score of 72.2 on the MMMU benchmark, setting a new state-of-the-art among open-source MLLMs. Its capabilities remain highly competitive with leading proprietary models, including ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 2.5 Pro, while also maintaining strong pure-language proficiency. In pursuit of open-science principles, we will publicly release both the training data and model weights to foster further research and development in next-generation MLLMs.
comment: Technical Report
☆ MIEB: Massive Image Embedding Benchmark
Image representations are often evaluated through disjointed, task-specific protocols, leading to a fragmented understanding of model capabilities. For instance, it is unclear whether an image embedding model adept at clustering images is equally good at retrieving relevant images given a piece of text. We introduce the Massive Image Embedding Benchmark (MIEB) to evaluate the performance of image and image-text embedding models across the broadest spectrum to date. MIEB spans 38 languages across 130 individual tasks, which we group into 8 high-level categories. We benchmark 50 models across our benchmark, finding that no single method dominates across all task categories. We reveal hidden capabilities in advanced vision models such as their accurate visual representation of texts, and their yet limited capabilities in interleaved encodings and matching images and texts in the presence of confounders. We also show that the performance of vision encoders on MIEB correlates highly with their performance when used in multimodal large language models. Our code, dataset, and leaderboard are publicly available at https://github.com/embeddings-benchmark/mteb.
☆ Art3D: Training-Free 3D Generation from Flat-Colored Illustration SC
Large-scale pre-trained image-to-3D generative models have exhibited remarkable capabilities in diverse shape generations. However, most of them struggle to synthesize plausible 3D assets when the reference image is flat-colored like hand drawings due to the lack of 3D illusion, which are often the most user-friendly input modalities in art content creation. To this end, we propose Art3D, a training-free method that can lift flat-colored 2D designs into 3D. By leveraging structural and semantic features with pre- trained 2D image generation models and a VLM-based realism evaluation, Art3D successfully enhances the three-dimensional illusion in reference images, thus simplifying the process of generating 3D from 2D, and proves adaptable to a wide range of painting styles. To benchmark the generalization performance of existing image-to-3D models on flat-colored images without 3D feeling, we collect a new dataset, Flat-2D, with over 100 samples. Experimental results demonstrate the performance and robustness of Art3D, exhibiting superior generalizable capacity and promising practical applicability. Our source code and dataset will be publicly available on our project page: https://joy-jy11.github.io/ .
comment: Technical Report. Course Project of Brown CSCI 1430 Computer Vision. Project Page: https://joy-jy11.github.io/
☆ Pixel-SAIL: Single Transformer For Pixel-Grounded Understanding
Multimodal Large Language Models (MLLMs) achieve remarkable performance for fine-grained pixel-level understanding tasks. However, all the works rely heavily on extra components, such as vision encoder (CLIP), segmentation experts, leading to high system complexity and limiting model scaling. In this work, our goal is to explore a highly simplified MLLM without introducing extra components. Our work is motivated by the recent works on Single trAnsformer as a unified vIsion-Language Model (SAIL) design, where these works jointly learn vision tokens and text tokens in transformers. We present Pixel-SAIL, a single transformer for pixel-wise MLLM tasks. In particular, we present three technical improvements on the plain baseline. First, we design a learnable upsampling module to refine visual token features. Secondly, we propose a novel visual prompt injection strategy to enable the single transformer to understand visual prompt inputs and benefit from the early fusion of visual prompt embeddings and vision tokens. Thirdly, we introduce a vision expert distillation strategy to efficiently enhance the single transformer's fine-grained feature extraction capability. In addition, we have collected a comprehensive pixel understanding benchmark (PerBench), using a manual check. It includes three tasks: detailed object description, visual prompt-based question answering, and visual-text referring segmentation. Extensive experiments on four referring segmentation benchmarks, one visual prompt benchmark, and our PerBench show that our Pixel-SAIL achieves comparable or even better results with a much simpler pipeline. Code and model will be released at https://github.com/magic-research/Sa2VA.
☆ The Scalability of Simplicity: Empirical Analysis of Vision-Language Learning with a Single Transformer
This paper introduces SAIL, a single transformer unified multimodal large language model (MLLM) that integrates raw pixel encoding and language decoding within a singular architecture. Unlike existing modular MLLMs, which rely on a pre-trained vision transformer (ViT), SAIL eliminates the need for a separate vision encoder, presenting a more minimalist architecture design. Instead of introducing novel architectural components, SAIL adapts mix-attention mechanisms and multimodal positional encodings to better align with the distinct characteristics of visual and textual modalities. We systematically compare SAIL's properties-including scalability, cross-modal information flow patterns, and visual representation capabilities-with those of modular MLLMs. By scaling both training data and model size, SAIL achieves performance comparable to modular MLLMs. Notably, the removal of pretrained ViT components enhances SAIL's scalability and results in significantly different cross-modal information flow patterns. Moreover, SAIL demonstrates strong visual representation capabilities, achieving results on par with ViT-22B in vision tasks such as semantic segmentation. Code and models are available at https://github.com/bytedance/SAIL.
☆ GUI-R1 : A Generalist R1-Style Vision-Language Action Model For GUI Agents
Existing efforts in building Graphical User Interface (GUI) agents largely rely on the training paradigm of supervised fine-tuning on Large Vision-Language Models (LVLMs). However, this approach not only demands extensive amounts of training data but also struggles to effectively understand GUI screenshots and generalize to unseen interfaces. The issue significantly limits its application in real-world scenarios, especially for high-level tasks. Inspired by Reinforcement Fine-Tuning (RFT) in large reasoning models (e.g., DeepSeek-R1), which efficiently enhances the problem-solving capabilities of large language models in real-world settings, we propose \name, the first reinforcement learning framework designed to enhance the GUI capabilities of LVLMs in high-level real-world task scenarios, through unified action space rule modeling. By leveraging a small amount of carefully curated high-quality data across multiple platforms (including Windows, Linux, MacOS, Android, and Web) and employing policy optimization algorithms such as Group Relative Policy Optimization (GRPO) to update the model, \name achieves superior performance using only 0.02\% of the data (3K vs. 13M) compared to previous state-of-the-art methods like OS-Atlas across eight benchmarks spanning three different platforms (mobile, desktop, and web). These results demonstrate the immense potential of reinforcement learning based on unified action space rule modeling in improving the execution capabilities of LVLMs for real-world GUI agent tasks.
☆ Integrating Vision and Location with Transformers: A Multimodal Deep Learning Framework for Medical Wound Analysis
Effective recognition of acute and difficult-to-heal wounds is a necessary step in wound diagnosis. An efficient classification model can help wound specialists classify wound types with less financial and time costs and also help in deciding on the optimal treatment method. Traditional machine learning models suffer from feature selection and are usually cumbersome models for accurate recognition. Recently, deep learning (DL) has emerged as a powerful tool in wound diagnosis. Although DL seems promising for wound type recognition, there is still a large scope for improving the efficiency and accuracy of the model. In this study, a DL-based multimodal classifier was developed using wound images and their corresponding locations to classify them into multiple classes, including diabetic, pressure, surgical, and venous ulcers. A body map was also created to provide location data, which can help wound specialists label wound locations more effectively. The model uses a Vision Transformer to extract hierarchical features from input images, a Discrete Wavelet Transform (DWT) layer to capture low and high frequency components, and a Transformer to extract spatial features. The number of neurons and weight vector optimization were performed using three swarm-based optimization techniques (Monster Gorilla Toner (MGTO), Improved Gray Wolf Optimization (IGWO), and Fox Optimization Algorithm). The evaluation results show that weight vector optimization using optimization algorithms can increase diagnostic accuracy and make it a very effective approach for wound detection. In the classification using the original body map, the proposed model was able to achieve an accuracy of 0.8123 using image data and an accuracy of 0.8007 using a combination of image data and wound location. Also, the accuracy of the model in combination with the optimization models varied from 0.7801 to 0.8342.
☆ RealWebAssist: A Benchmark for Long-Horizon Web Assistance with Real-World Users SC
To achieve successful assistance with long-horizon web-based tasks, AI agents must be able to sequentially follow real-world user instructions over a long period. Unlike existing web-based agent benchmarks, sequential instruction following in the real world poses significant challenges beyond performing a single, clearly defined task. For instance, real-world human instructions can be ambiguous, require different levels of AI assistance, and may evolve over time, reflecting changes in the user's mental state. To address this gap, we introduce RealWebAssist, a novel benchmark designed to evaluate sequential instruction-following in realistic scenarios involving long-horizon interactions with the web, visual GUI grounding, and understanding ambiguous real-world user instructions. RealWebAssist includes a dataset of sequential instructions collected from real-world human users. Each user instructs a web-based assistant to perform a series of tasks on multiple websites. A successful agent must reason about the true intent behind each instruction, keep track of the mental state of the user, understand user-specific routines, and ground the intended tasks to actions on the correct GUI elements. Our experimental results show that state-of-the-art models struggle to understand and ground user instructions, posing critical challenges in following real-world user instructions for long-horizon web assistance.
comment: Project Website: https://scai.cs.jhu.edu/projects/RealWebAssist/ Code: https://github.com/SCAI-JHU/RealWebAssist
☆ Multimodal Long Video Modeling Based on Temporal Dynamic Context
Recent advances in Large Language Models (LLMs) have led to significant breakthroughs in video understanding. However, existing models still struggle with long video processing due to the context length constraint of LLMs and the vast amount of information within the video. Although some recent methods are designed for long video understanding, they often lose crucial information during token compression and struggle with additional modality like audio. In this work, we propose a dynamic long video encoding method utilizing the temporal relationship between frames, named Temporal Dynamic Context (TDC). Firstly, we segment the video into semantically consistent scenes based on inter-frame similarities, then encode each frame into tokens using visual-audio encoders. Secondly, we propose a novel temporal context compressor to reduce the number of tokens within each segment. Specifically, we employ a query-based Transformer to aggregate video, audio, and instruction text tokens into a limited set of temporal context tokens. Finally, we feed the static frame tokens and the temporal context tokens into the LLM for video understanding. Furthermore, to handle extremely long videos, we propose a training-free chain-of-thought strategy that progressively extracts answers from multiple video segments. These intermediate answers serve as part of the reasoning process and contribute to the final answer. We conduct extensive experiments on general video understanding and audio-video understanding benchmarks, where our method demonstrates strong performance. The code and models are available at https://github.com/Hoar012/TDC-Video.
☆ Anchor Token Matching: Implicit Structure Locking for Training-free AR Image Editing
Text-to-image generation has seen groundbreaking advancements with diffusion models, enabling high-fidelity synthesis and precise image editing through cross-attention manipulation. Recently, autoregressive (AR) models have re-emerged as powerful alternatives, leveraging next-token generation to match diffusion models. However, existing editing techniques designed for diffusion models fail to translate directly to AR models due to fundamental differences in structural control. Specifically, AR models suffer from spatial poverty of attention maps and sequential accumulation of structural errors during image editing, which disrupt object layouts and global consistency. In this work, we introduce Implicit Structure Locking (ISLock), the first training-free editing strategy for AR visual models. Rather than relying on explicit attention manipulation or fine-tuning, ISLock preserves structural blueprints by dynamically aligning self-attention patterns with reference images through the Anchor Token Matching (ATM) protocol. By implicitly enforcing structural consistency in latent space, our method ISLock enables structure-aware editing while maintaining generative autonomy. Extensive experiments demonstrate that ISLock achieves high-quality, structure-consistent edits without additional training and is superior or comparable to conventional editing techniques. Our findings pioneer the way for efficient and flexible AR-based image editing, further bridging the performance gap between diffusion and autoregressive generative models. The code will be publicly available at https://github.com/hutaiHang/ATM
☆ MonoDiff9D: Monocular Category-Level 9D Object Pose Estimation via Diffusion Model ICRA'25
Object pose estimation is a core means for robots to understand and interact with their environment. For this task, monocular category-level methods are attractive as they require only a single RGB camera. However, current methods rely on shape priors or CAD models of the intra-class known objects. We propose a diffusion-based monocular category-level 9D object pose generation method, MonoDiff9D. Our motivation is to leverage the probabilistic nature of diffusion models to alleviate the need for shape priors, CAD models, or depth sensors for intra-class unknown object pose estimation. We first estimate coarse depth via DINOv2 from the monocular image in a zero-shot manner and convert it into a point cloud. We then fuse the global features of the point cloud with the input image and use the fused features along with the encoded time step to condition MonoDiff9D. Finally, we design a transformer-based denoiser to recover the object pose from Gaussian noise. Extensive experiments on two popular benchmark datasets show that MonoDiff9D achieves state-of-the-art monocular category-level 9D object pose estimation accuracy without the need for shape priors or CAD models at any stage. Our code will be made public at https://github.com/CNJianLiu/MonoDiff9D.
comment: Accepted by ICRA'25
☆ HUMOTO: A 4D Dataset of Mocap Human Object Interactions
We present Human Motions with Objects (HUMOTO), a high-fidelity dataset of human-object interactions for motion generation, computer vision, and robotics applications. Featuring 736 sequences (7,875 seconds at 30 fps), HUMOTO captures interactions with 63 precisely modeled objects and 72 articulated parts. Our innovations include a scene-driven LLM scripting pipeline creating complete, purposeful tasks with natural progression, and a mocap-and-camera recording setup to effectively handle occlusions. Spanning diverse activities from cooking to outdoor picnics, HUMOTO preserves both physical accuracy and logical task flow. Professional artists rigorously clean and verify each sequence, minimizing foot sliding and object penetrations. We also provide benchmarks compared to other datasets. HUMOTO's comprehensive full-body motion and simultaneous multi-object interactions address key data-capturing challenges and provide opportunities to advance realistic human-object interaction modeling across research domains with practical applications in animation, robotics, and embodied AI systems. Project: https://jiaxin-lu.github.io/humoto/ .
comment: 19 pages, 15 figures
☆ GPS: Distilling Compact Memories via Grid-based Patch Sampling for Efficient Online Class-Incremental Learning
Online class-incremental learning aims to enable models to continuously adapt to new classes with limited access to past data, while mitigating catastrophic forgetting. Replay-based methods address this by maintaining a small memory buffer of previous samples, achieving competitive performance. For effective replay under constrained storage, recent approaches leverage distilled data to enhance the informativeness of memory. However, such approaches often involve significant computational overhead due to the use of bi-level optimization. Motivated by these limitations, we introduce Grid-based Patch Sampling (GPS), a lightweight and effective strategy for distilling informative memory samples without relying on a trainable model. GPS generates informative samples by sampling a subset of pixels from the original image, yielding compact low-resolution representations that preserve both semantic content and structural information. During replay, these representations are reassembled to support training and evaluation. Experiments on extensive benchmarks demonstrate that GRS can be seamlessly integrated into existing replay frameworks, leading to 3%-4% improvements in average end accuracy under memory-constrained settings, with limited computational overhead.
comment: 10 pages, 10 figures
☆ Towards Low-Latency Event-based Obstacle Avoidance on a FPGA-Drone
This work quantitatively evaluates the performance of event-based vision systems (EVS) against conventional RGB-based models for action prediction in collision avoidance on an FPGA accelerator. Our experiments demonstrate that the EVS model achieves a significantly higher effective frame rate (1 kHz) and lower temporal (-20 ms) and spatial prediction errors (-20 mm) compared to the RGB-based model, particularly when tested on out-of-distribution data. The EVS model also exhibits superior robustness in selecting optimal evasion maneuvers. In particular, in distinguishing between movement and stationary states, it achieves a 59 percentage point advantage in precision (78% vs. 19%) and a substantially higher F1 score (0.73 vs. 0.06), highlighting the susceptibility of the RGB model to overfitting. Further analysis in different combinations of spatial classes confirms the consistent performance of the EVS model in both test data sets. Finally, we evaluated the system end-to-end and achieved a latency of approximately 2.14 ms, with event aggregation (1 ms) and inference on the processing unit (0.94 ms) accounting for the largest components. These results underscore the advantages of event-based vision for real-time collision avoidance and demonstrate its potential for deployment in resource-constrained environments.
☆ Better Coherence, Better Height: Fusing Physical Models and Deep Learning for Forest Height Estimation from Interferometric SAR Data
Estimating forest height from Synthetic Aperture Radar (SAR) images often relies on traditional physical models, which, while interpretable and data-efficient, can struggle with generalization. In contrast, Deep Learning (DL) approaches lack physical insight. To address this, we propose CoHNet - an end-to-end framework that combines the best of both worlds: DL optimized with physics-informed constraints. We leverage a pre-trained neural surrogate model to enforce physical plausibility through a unique training loss. Our experiments show that this approach not only improves forest height estimation accuracy but also produces meaningful features that enhance the reliability of predictions.
☆ PG-DPIR: An efficient plug-and-play method for high-count Poisson-Gaussian inverse problems
Poisson-Gaussian noise describes the noise of various imaging systems thus the need of efficient algorithms for Poisson-Gaussian image restoration. Deep learning methods offer state-of-the-art performance but often require sensor-specific training when used in a supervised setting. A promising alternative is given by plug-and-play (PnP) methods, which consist in learning only a regularization through a denoiser, allowing to restore images from several sources with the same network. This paper introduces PG-DPIR, an efficient PnP method for high-count Poisson-Gaussian inverse problems, adapted from DPIR. While DPIR is designed for white Gaussian noise, a naive adaptation to Poisson-Gaussian noise leads to prohibitively slow algorithms due to the absence of a closed-form proximal operator. To address this, we adapt DPIR for the specificities of Poisson-Gaussian noise and propose in particular an efficient initialization of the gradient descent required for the proximal step that accelerates convergence by several orders of magnitude. Experiments are conducted on satellite image restoration and super-resolution problems. High-resolution realistic Pleiades images are simulated for the experiments, which demonstrate that PG-DPIR achieves state-of-the-art performance with improved efficiency, which seems promising for on-ground satellite processing chains.
☆ FingER: Content Aware Fine-grained Evaluation with Reasoning for AI-Generated Videos
Recent advances in video generation have posed great challenges in the assessment of AI-generated content, particularly with the emergence of increasingly sophisticated models. The various inconsistencies and defects observed in such videos are inherently complex, making overall scoring notoriously difficult. In this paper, we emphasize the critical importance of integrating fine-grained reasoning into video evaluation, and we propose $\textbf{F}$ing$\textbf{ER}$, a novel entity-level reasoning evaluation framework that first automatically generates $\textbf{F}$ine-grained $\textbf{E}$ntity-level questions, and then answers those questions by a $\textbf{R}$easoning model with scores, which can be subsequently weighted summed to an overall score for different applications. Specifically, we leverage LLMs to derive entity-level questions across five distinct perspectives, which (i) often focus on some specific entities of the content, thereby making answering or scoring much easier by MLLMs, and (ii) are more interpretable. Then we construct a FingER dataset, consisting of approximately 3.3k videos and corresponding 60k fine-grained QA annotations, each with detailed reasons. Based on that, we further investigate various training protocols to best incentivize the reasoning capability of MLLMs for correct answer prediction. Extensive experiments demonstrate that a reasoning model trained using Group Relative Policy Optimization (GRPO) with a cold-start strategy achieves the best performance. Notably, our model surpasses existing methods by a relative margin of $11.8\%$ on GenAI-Bench and $5.5\%$ on MonetBench with only 3.3k training videos, which is at most one-tenth of the training samples utilized by other methods. Our code and dataset will be released soon.
comment: 10 pages, 4 figures
☆ Patch and Shuffle: A Preprocessing Technique for Texture Classification in Autonomous Cementitious Fabrication
Autonomous fabrication systems are transforming construction and manufacturing, yet they remain vulnerable to print errors. Texture classification is a key component of computer vision systems that enable real-time monitoring and adjustment during cementitious fabrication. Traditional classification methods often rely on global image features, which can bias the model toward semantic content rather than low-level textures. In this paper, we introduce a novel preprocessing technique called "patch and shuffle," which segments input images into smaller patches, shuffles them, and reconstructs a jumbled image before classification. This transformation removes semantic context, forcing the classifier to rely on local texture features. We evaluate this approach on a dataset of extruded cement images, using a ResNet-18-based architecture. Our experiments compare the patch and shuffle method to a standard pipeline, holding all other factors constant. Results show a significant improvement in accuracy: the patch and shuffle model achieved 90.64% test accuracy versus 72.46% for the baseline. These findings suggest that disrupting global structure enhances performance in texture-based classification tasks. This method has implications for broader vision tasks where low-level features matter more than high-level semantics. The technique may improve classification in applications ranging from fabrication monitoring to medical imaging.
comment: Originally completed as a final project for CEE 374 at Princeton University
☆ Multimodal Representation Learning Techniques for Comprehensive Facial State Analysis ICME2025
Multimodal foundation models have significantly improved feature representation by integrating information from multiple modalities, making them highly suitable for a broader set of applications. However, the exploration of multimodal facial representation for understanding perception has been limited. Understanding and analyzing facial states, such as Action Units (AUs) and emotions, require a comprehensive and robust framework that bridges visual and linguistic modalities. In this paper, we present a comprehensive pipeline for multimodal facial state analysis. First, we compile a new Multimodal Face Dataset (MFA) by generating detailed multilevel language descriptions of face, incorporating Action Unit (AU) and emotion descriptions, by leveraging GPT-4o. Second, we introduce a novel Multilevel Multimodal Face Foundation model (MF^2) tailored for Action Unit (AU) and emotion recognition. Our model incorporates comprehensive visual feature modeling at both local and global levels of face image, enhancing its ability to represent detailed facial appearances. This design aligns visual representations with structured AU and emotion descriptions, ensuring effective cross-modal integration. Third, we develop a Decoupled Fine-Tuning Network (DFN) that efficiently adapts MF^2 across various tasks and datasets. This approach not only reduces computational overhead but also broadens the applicability of the foundation model to diverse scenarios. Experimentation show superior performance for AU and emotion detection tasks.
comment: Accepted by ICME2025
☆ Benchmarking 3D Human Pose Estimation Models Under Occlusions
This paper addresses critical challenges in 3D Human Pose Estimation (HPE) by analyzing the robustness and sensitivity of existing models to occlusions, camera position, and action variability. Using a novel synthetic dataset, BlendMimic3D, which includes diverse scenarios with multi-camera setups and several occlusion types, we conduct specific tests on several state-of-the-art models. Our study focuses on the discrepancy in keypoint formats between common datasets such as Human3.6M, and 2D datasets such as COCO, commonly used for 2D detection models and frequently input of 3D HPE models. Our work explores the impact of occlusions on model performance and the generality of models trained exclusively under standard conditions. The findings suggest significant sensitivity to occlusions and camera settings, revealing a need for models that better adapt to real-world variability and occlusion scenarios. This research contributed to ongoing efforts to improve the fidelity and applicability of 3D HPE systems in complex environments.
☆ LL-Gaussian: Low-Light Scene Reconstruction and Enhancement via Gaussian Splatting for Novel View Synthesis
Novel view synthesis (NVS) in low-light scenes remains a significant challenge due to degraded inputs characterized by severe noise, low dynamic range (LDR) and unreliable initialization. While recent NeRF-based approaches have shown promising results, most suffer from high computational costs, and some rely on carefully captured or pre-processed data--such as RAW sensor inputs or multi-exposure sequences--which severely limits their practicality. In contrast, 3D Gaussian Splatting (3DGS) enables real-time rendering with competitive visual fidelity; however, existing 3DGS-based methods struggle with low-light sRGB inputs, resulting in unstable Gaussian initialization and ineffective noise suppression. To address these challenges, we propose LL-Gaussian, a novel framework for 3D reconstruction and enhancement from low-light sRGB images, enabling pseudo normal-light novel view synthesis. Our method introduces three key innovations: 1) an end-to-end Low-Light Gaussian Initialization Module (LLGIM) that leverages dense priors from learning-based MVS approach to generate high-quality initial point clouds; 2) a dual-branch Gaussian decomposition model that disentangles intrinsic scene properties (reflectance and illumination) from transient interference, enabling stable and interpretable optimization; 3) an unsupervised optimization strategy guided by both physical constrains and diffusion prior to jointly steer decomposition and enhancement. Additionally, we contribute a challenging dataset collected in extreme low-light environments and demonstrate the effectiveness of LL-Gaussian. Compared to state-of-the-art NeRF-based methods, LL-Gaussian achieves up to 2,000 times faster inference and reduces training time to just 2%, while delivering superior reconstruction and rendering quality.
☆ InstructEngine: Instruction-driven Text-to-Image Alignment
Reinforcement Learning from Human/AI Feedback (RLHF/RLAIF) has been extensively utilized for preference alignment of text-to-image models. Existing methods face certain limitations in terms of both data and algorithm. For training data, most approaches rely on manual annotated preference data, either by directly fine-tuning the generators or by training reward models to provide training signals. However, the high annotation cost makes them difficult to scale up, the reward model consumes extra computation and cannot guarantee accuracy. From an algorithmic perspective, most methods neglect the value of text and only take the image feedback as a comparative signal, which is inefficient and sparse. To alleviate these drawbacks, we propose the InstructEngine framework. Regarding annotation cost, we first construct a taxonomy for text-to-image generation, then develop an automated data construction pipeline based on it. Leveraging advanced large multimodal models and human-defined rules, we generate 25K text-image preference pairs. Finally, we introduce cross-validation alignment method, which refines data efficiency by organizing semantically analogous samples into mutually comparable pairs. Evaluations on DrawBench demonstrate that InstructEngine improves SD v1.5 and SDXL's performance by 10.53% and 5.30%, outperforming state-of-the-art baselines, with ablation study confirming the benefits of InstructEngine's all components. A win rate of over 50% in human reviews also proves that InstructEngine better aligns with human preferences.
comment: 8 pages, 7 figures
☆ SlowFastVAD: Video Anomaly Detection via Integrating Simple Detector and RAG-Enhanced Vision-Language Model
Video anomaly detection (VAD) aims to identify unexpected events in videos and has wide applications in safety-critical domains. While semi-supervised methods trained on only normal samples have gained traction, they often suffer from high false alarm rates and poor interpretability. Recently, vision-language models (VLMs) have demonstrated strong multimodal reasoning capabilities, offering new opportunities for explainable anomaly detection. However, their high computational cost and lack of domain adaptation hinder real-time deployment and reliability. Inspired by dual complementary pathways in human visual perception, we propose SlowFastVAD, a hybrid framework that integrates a fast anomaly detector with a slow anomaly detector (namely a retrieval augmented generation (RAG) enhanced VLM), to address these limitations. Specifically, the fast detector first provides coarse anomaly confidence scores, and only a small subset of ambiguous segments, rather than the entire video, is further analyzed by the slower yet more interpretable VLM for elaborate detection and reasoning. Furthermore, to adapt VLMs to domain-specific VAD scenarios, we construct a knowledge base including normal patterns based on few normal samples and abnormal patterns inferred by VLMs. During inference, relevant patterns are retrieved and used to augment prompts for anomaly reasoning. Finally, we smoothly fuse the anomaly confidence of fast and slow detectors to enhance robustness of anomaly detection. Extensive experiments on four benchmarks demonstrate that SlowFastVAD effectively combines the strengths of both fast and slow detectors, and achieves remarkable detection accuracy and interpretability with significantly reduced computational overhead, making it well-suited for real-world VAD applications with high reliability requirements.
☆ Analysis of Attention in Video Diffusion Transformers
We conduct an in-depth analysis of attention in video diffusion transformers (VDiTs) and report a number of novel findings. We identify three key properties of attention in VDiTs: Structure, Sparsity, and Sinks. Structure: We observe that attention patterns across different VDiTs exhibit similar structure across different prompts, and that we can make use of the similarity of attention patterns to unlock video editing via self-attention map transfer. Sparse: We study attention sparsity in VDiTs, finding that proposed sparsity methods do not work for all VDiTs, because some layers that are seemingly sparse cannot be sparsified. Sinks: We make the first study of attention sinks in VDiTs, comparing and contrasting them to attention sinks in language models. We propose a number of future directions that can make use of our insights to improve the efficiency-quality Pareto frontier for VDiTs.
☆ ESCT3D: Efficient and Selectively Controllable Text-Driven 3D Content Generation with Gaussian Splatting
In recent years, significant advancements have been made in text-driven 3D content generation. However, several challenges remain. In practical applications, users often provide extremely simple text inputs while expecting high-quality 3D content. Generating optimal results from such minimal text is a difficult task due to the strong dependency of text-to-3D models on the quality of input prompts. Moreover, the generation process exhibits high variability, making it difficult to control. Consequently, multiple iterations are typically required to produce content that meets user expectations, reducing generation efficiency. To address this issue, we propose GPT-4V for self-optimization, which significantly enhances the efficiency of generating satisfactory content in a single attempt. Furthermore, the controllability of text-to-3D generation methods has not been fully explored. Our approach enables users to not only provide textual descriptions but also specify additional conditions, such as style, edges, scribbles, poses, or combinations of multiple conditions, allowing for more precise control over the generated 3D content. Additionally, during training, we effectively integrate multi-view information, including multi-view depth, masks, features, and images, to address the common Janus problem in 3D content generation. Extensive experiments demonstrate that our method achieves robust generalization, facilitating the efficient and controllable generation of high-quality 3D content.
☆ Noise2Ghost: Self-supervised deep convolutional reconstruction for ghost imaging
We present a new self-supervised deep-learning-based Ghost Imaging (GI) reconstruction method, which provides unparalleled reconstruction performance for noisy acquisitions among unsupervised methods. We present the supporting mathematical framework and results from theoretical and real data use cases. Self-supervision removes the need for clean reference data while offering strong noise reduction. This provides the necessary tools for addressing signal-to-noise ratio concerns for GI acquisitions in emerging and cutting-edge low-light GI scenarios. Notable examples include micro- and nano-scale x-ray emission imaging, e.g., x-ray fluorescence imaging of dose-sensitive samples. Their applications include in-vivo and in-operando case studies for biological samples and batteries.
☆ Zero-shot Autonomous Microscopy for Scalable and Intelligent Characterization of 2D Materials
Characterization of atomic-scale materials traditionally requires human experts with months to years of specialized training. Even for trained human operators, accurate and reliable characterization remains challenging when examining newly discovered materials such as two-dimensional (2D) structures. This bottleneck drives demand for fully autonomous experimentation systems capable of comprehending research objectives without requiring large training datasets. In this work, we present ATOMIC (Autonomous Technology for Optical Microscopy & Intelligent Characterization), an end-to-end framework that integrates foundation models to enable fully autonomous, zero-shot characterization of 2D materials. Our system integrates the vision foundation model (i.e., Segment Anything Model), large language models (i.e., ChatGPT), unsupervised clustering, and topological analysis to automate microscope control, sample scanning, image segmentation, and intelligent analysis through prompt engineering, eliminating the need for additional training. When analyzing typical MoS2 samples, our approach achieves 99.7% segmentation accuracy for single layer identification, which is equivalent to that of human experts. In addition, the integrated model is able to detect grain boundary slits that are challenging to identify with human eyes. Furthermore, the system retains robust accuracy despite variable conditions including defocus, color temperature fluctuations, and exposure variations. It is applicable to a broad spectrum of common 2D materials-including graphene, MoS2, WSe2, SnSe-regardless of whether they were fabricated via chemical vapor deposition or mechanical exfoliation. This work represents the implementation of foundation models to achieve autonomous analysis, establishing a scalable and data-efficient characterization paradigm that fundamentally transforms the approach to nanoscale materials research.
comment: 13 pages, 4 figures
☆ DiffMOD: Progressive Diffusion Point Denoising for Moving Object Detection in Remote Sensing
Moving object detection (MOD) in remote sensing is significantly challenged by low resolution, extremely small object sizes, and complex noise interference. Current deep learning-based MOD methods rely on probability density estimation, which restricts flexible information interaction between objects and across temporal frames. To flexibly capture high-order inter-object and temporal relationships, we propose a point-based MOD in remote sensing. Inspired by diffusion models, the network optimization is formulated as a progressive denoising process that iteratively recovers moving object centers from sparse noisy points. Specifically, we sample scattered features from the backbone outputs as atomic units for subsequent processing, while global feature embeddings are aggregated to compensate for the limited coverage of sparse point features. By modeling spatial relative positions and semantic affinities, Spatial Relation Aggregation Attention is designed to enable high-order interactions among point-level features for enhanced object representation. To enhance temporal consistency, the Temporal Propagation and Global Fusion module is designed, which leverages an implicit memory reasoning mechanism for robust cross-frame feature integration. To align with the progressive denoising process, we propose a progressive MinK optimal transport assignment strategy that establishes specialized learning objectives at each denoising level. Additionally, we introduce a missing loss function to counteract the clustering tendency of denoised points around salient objects. Experiments on the RsData remote sensing MOD dataset show that our MOD method based on scattered point denoising can more effectively explore potential relationships between sparse moving objects and improve the detection capability and temporal consistency.
comment: 9 pages, 7 figures
☆ LMFormer: Lane based Motion Prediction Transformer CVPR 2025
Motion prediction plays an important role in autonomous driving. This study presents LMFormer, a lane-aware transformer network for trajectory prediction tasks. In contrast to previous studies, our work provides a simple mechanism to dynamically prioritize the lanes and shows that such a mechanism introduces explainability into the learning behavior of the network. Additionally, LMFormer uses the lane connection information at intersections, lane merges, and lane splits, in order to learn long-range dependency in lane structure. Moreover, we also address the issue of refining the predicted trajectories and propose an efficient method for iterative refinement through stacked transformer layers. For benchmarking, we evaluate LMFormer on the nuScenes dataset and demonstrate that it achieves SOTA performance across multiple metrics. Furthermore, the Deep Scenario dataset is used to not only illustrate cross-dataset network performance but also the unification capabilities of LMFormer to train on multiple datasets and achieve better performance.
comment: Accepted: Autonomous Driving Workshop, CVPR 2025
☆ Trade-offs in Privacy-Preserving Eye Tracking through Iris Obfuscation: A Benchmarking Study SP 2025
Recent developments in hardware, computer graphics, and AI may soon enable AR/VR head-mounted displays (HMDs) to become everyday devices like smartphones and tablets. Eye trackers within HMDs provide a special opportunity for such setups as it is possible to facilitate gaze-based research and interaction. However, estimating users' gaze information often requires raw eye images and videos that contain iris textures, which are considered a gold standard biometric for user authentication, and this raises privacy concerns. Previous research in the eye-tracking community focused on obfuscating iris textures while keeping utility tasks such as gaze estimation accurate. Despite these attempts, there is no comprehensive benchmark that evaluates state-of-the-art approaches. Considering all, in this paper, we benchmark blurring, noising, downsampling, rubber sheet model, and iris style transfer to obfuscate user identity, and compare their impact on image quality, privacy, utility, and risk of imposter attack on two datasets. We use eye segmentation and gaze estimation as utility tasks, and reduction in iris recognition accuracy as a measure of privacy protection, and false acceptance rate to estimate risk of attack. Our experiments show that canonical image processing methods like blurring and noising cause a marginal impact on deep learning-based tasks. While downsampling, rubber sheet model, and iris style transfer are effective in hiding user identifiers, iris style transfer, with higher computation cost, outperforms others in both utility tasks, and is more resilient against spoof attacks. Our analyses indicate that there is no universal optimal approach to balance privacy, utility, and computation burden. Therefore, we recommend practitioners consider the strengths and weaknesses of each approach, and possible combinations of those to reach an optimal privacy-utility trade-off.
comment: The 25th International Conference on Digital Signal Processing (DSP 2025)
☆ XY-Cut++: Advanced Layout Ordering via Hierarchical Mask Mechanism on a Novel Benchmark
Document Reading Order Recovery is a fundamental task in document image understanding, playing a pivotal role in enhancing Retrieval-Augmented Generation (RAG) and serving as a critical preprocessing step for large language models (LLMs). Existing methods often struggle with complex layouts(e.g., multi-column newspapers), high-overhead interactions between cross-modal elements (visual regions and textual semantics), and a lack of robust evaluation benchmarks. We introduce XY-Cut++, an advanced layout ordering method that integrates pre-mask processing, multi-granularity segmentation, and cross-modal matching to address these challenges. Our method significantly enhances layout ordering accuracy compared to traditional XY-Cut techniques. Specifically, XY-Cut++ achieves state-of-the-art performance (98.8 BLEU overall) while maintaining simplicity and efficiency. It outperforms existing baselines by up to 24\% and demonstrates consistent accuracy across simple and complex layouts on the newly introduced DocBench-100 dataset. This advancement establishes a reliable foundation for document structure recovery, setting a new standard for layout ordering tasks and facilitating more effective RAG and LLM preprocessing.
☆ MASSeg : 2nd Technical Report for 4th PVUW MOSE Track
Complex video object segmentation continues to face significant challenges in small object recognition, occlusion handling, and dynamic scene modeling. This report presents our solution, which ranked second in the MOSE track of CVPR 2025 PVUW Challenge. Based on an existing segmentation framework, we propose an improved model named MASSeg for complex video object segmentation, and construct an enhanced dataset, MOSE+, which includes typical scenarios with occlusions, cluttered backgrounds, and small target instances. During training, we incorporate a combination of inter-frame consistent and inconsistent data augmentation strategies to improve robustness and generalization. During inference, we design a mask output scaling strategy to better adapt to varying object sizes and occlusion levels. As a result, MASSeg achieves a J score of 0.8250, F score of 0.9007, and a J&F score of 0.8628 on the MOSE test set.
comment: 5 pages,4 figures,Technical report on Complex Video Object Segmentation
☆ Towards contrast- and pathology-agnostic clinical fetal brain MRI segmentation using SynthSeg
Magnetic resonance imaging (MRI) has played a crucial role in fetal neurodevelopmental research. Structural annotations of MR images are an important step for quantitative analysis of the developing human brain, with Deep learning providing an automated alternative for this otherwise tedious manual process. However, segmentation performances of Convolutional Neural Networks often suffer from domain shift, where the network fails when applied to subjects that deviate from the distribution with which it is trained on. In this work, we aim to train networks capable of automatically segmenting fetal brain MRIs with a wide range of domain shifts pertaining to differences in subject physiology and acquisition environments, in particular shape-based differences commonly observed in pathological cases. We introduce a novel data-driven train-time sampling strategy that seeks to fully exploit the diversity of a given training dataset to enhance the domain generalizability of the trained networks. We adapted our sampler, together with other existing data augmentation techniques, to the SynthSeg framework, a generator that utilizes domain randomization to generate diverse training data, and ran thorough experimentations and ablation studies on a wide range of training/testing data to test the validity of the approaches. Our networks achieved notable improvements in the segmentation quality on testing subjects with intense anatomical abnormalities (p < 1e-4), though at the cost of a slighter decrease in performance in cases with fewer abnormalities. Our work also lays the foundation for future works on creating and adapting data-driven sampling strategies for other training pipelines.
comment: 21 pages, 16 figures
☆ CAT: A Conditional Adaptation Tailor for Efficient and Effective Instance-Specific Pansharpening on Real-World Data
Pansharpening is a crucial remote sensing technique that fuses low-resolution multispectral (LRMS) images with high-resolution panchromatic (PAN) images to generate high-resolution multispectral (HRMS) imagery. Although deep learning techniques have significantly advanced pansharpening, many existing methods suffer from limited cross-sensor generalization and high computational overhead, restricting their real-time applications. To address these challenges, we propose an efficient framework that quickly adapts to a specific input instance, completing both training and inference in a short time. Our framework splits the input image into multiple patches, selects a subset for unsupervised CAT training, and then performs inference on all patches, stitching them into the final output. The CAT module, integrated between the feature extraction and channel transformation stages of a pre-trained network, tailors the fused features and fixes the parameters for efficient inference, generating improved results. Our approach offers two key advantages: (1) $\textit{Improved Generalization Ability}$: by mitigating cross-sensor degradation, our model--although pre-trained on a specific dataset--achieves superior performance on datasets captured by other sensors; (2) $\textit{Enhanced Computational Efficiency}$: the CAT-enhanced network can swiftly adapt to the test sample using the single LRMS-PAN pair input, without requiring extensive large-scale data retraining. Experiments on the real-world data from WorldView-3 and WorldView-2 datasets demonstrate that our method achieves state-of-the-art performance on cross-sensor real-world data, while achieving both training and inference of $512\times512$ image within $\textit{0.4 seconds}$ and $4000\times4000$ image within $\textit{3 seconds}$ at the fastest setting on a commonly used RTX 3090 GPU.
☆ Balancing Stability and Plasticity in Pretrained Detector: A Dual-Path Framework for Incremental Object Detection
The balance between stability and plasticity remains a fundamental challenge in pretrained model-based incremental object detection (PTMIOD). While existing PTMIOD methods demonstrate strong performance on in-domain tasks aligned with pretraining data, their plasticity to cross-domain scenarios remains underexplored. Through systematic component-wise analysis of pretrained detectors, we reveal a fundamental discrepancy: the localization modules demonstrate inherent cross-domain stability-preserving precise bounding box estimation across distribution shifts-while the classification components require enhanced plasticity to mitigate discriminability degradation in cross-domain scenarios. Motivated by these findings, we propose a dual-path framework built upon pretrained DETR-based detectors which decouples localization stability and classification plasticity: the localization path maintains stability to preserve pretrained localization knowledge, while the classification path facilitates plasticity via parameter-efficient fine-tuning and resists forgetting with pseudo-feature replay. Extensive evaluations on both in-domain (MS COCO and PASCAL VOC) and cross-domain (TT100K) benchmarks show state-of-the-art performance, demonstrating our method's ability to effectively balance stability and plasticity in PTMIOD, achieving robust cross-domain adaptation and strong retention of anti-forgetting capabilities.
☆ VibrantLeaves: A principled parametric image generator for training deep restoration models
Even though Deep Neural Networks are extremely powerful for image restoration tasks, they have several limitations. They are poorly understood and suffer from strong biases inherited from the training sets. One way to address these shortcomings is to have a better control over the training sets, in particular by using synthetic sets. In this paper, we propose a synthetic image generator relying on a few simple principles. In particular, we focus on geometric modeling, textures, and a simple modeling of image acquisition. These properties, integrated in a classical Dead Leaves model, enable the creation of efficient training sets. Standard image denoising and super-resolution networks can be trained on such datasets, reaching performance almost on par with training on natural image datasets. As a first step towards explainability, we provide a careful analysis of the considered principles, identifying which image properties are necessary to obtain good performances. Besides, such training also yields better robustness to various geometric and radiometric perturbations of the test sets.
☆ Differentially Private 2D Human Pose Estimation
Human pose estimation (HPE) has become essential in numerous applications including healthcare, activity recognition, and human-computer interaction. However, the privacy implications of processing sensitive visual data present significant deployment barriers in critical domains. While traditional anonymization techniques offer limited protection and often compromise data utility for broader motion analysis, Differential Privacy (DP) provides formal privacy guarantees but typically degrades model performance when applied naively. In this work, we present the first differentially private 2D human pose estimation (2D-HPE) by applying Differentially Private Stochastic Gradient Descent (DP-SGD) to this task. To effectively balance privacy with performance, we adopt Projected DP-SGD (PDP-SGD), which projects the noisy gradients to a low-dimensional subspace. Additionally, we adapt TinyViT, a compact and efficient vision transformer for coordinate classification in HPE, providing a lightweight yet powerful backbone that enhances privacy-preserving deployment feasibility on resource-limited devices. Our approach is particularly valuable for multimedia interpretation tasks, enabling privacy-safe analysis and understanding of human motion across diverse visual media while preserving the semantic meaning required for downstream applications. Comprehensive experiments on the MPII Human Pose Dataset demonstrate significant performance enhancement with PDP-SGD achieving 78.48% PCKh@0.5 at a strict privacy budget ($\epsilon=0.2$), compared to 63.85% for standard DP-SGD. This work lays foundation for privacy-preserving human pose estimation in real-world, sensitive applications.
☆ LLaVA-ReID: Selective Multi-image Questioner for Interactive Person Re-Identification
Traditional text-based person ReID assumes that person descriptions from witnesses are complete and provided at once. However, in real-world scenarios, such descriptions are often partial or vague. To address this limitation, we introduce a new task called interactive person re-identification (Inter-ReID). Inter-ReID is a dialogue-based retrieval task that iteratively refines initial descriptions through ongoing interactions with the witnesses. To facilitate the study of this new task, we construct a dialogue dataset that incorporates multiple types of questions by decomposing fine-grained attributes of individuals. We further propose LLaVA-ReID, a question model that generates targeted questions based on visual and textual contexts to elicit additional details about the target person. Leveraging a looking-forward strategy, we prioritize the most informative questions as supervision during training. Experimental results on both Inter-ReID and text-based ReID benchmarks demonstrate that LLaVA-ReID significantly outperforms baselines.
☆ WildLive: Near Real-time Visual Wildlife Tracking onboard UAVs
Live tracking of wildlife via high-resolution video processing directly onboard drones is widely unexplored and most existing solutions rely on streaming video to ground stations to support navigation. Yet, both autonomous animal-reactive flight control beyond visual line of sight and/or mission-specific individual and behaviour recognition tasks rely to some degree on this capability. In response, we introduce WildLive -- a near real-time animal detection and tracking framework for high-resolution imagery running directly onboard uncrewed aerial vehicles (UAVs). The system performs multi-animal detection and tracking at 17fps+ for HD and 7fps+ on 4K video streams suitable for operation during higher altitude flights to minimise animal disturbance. Our system is optimised for Jetson Orin AGX onboard hardware. It integrates the efficiency of sparse optical flow tracking and mission-specific sampling with device-optimised and proven YOLO-driven object detection and segmentation techniques. Essentially, computational resource is focused onto spatio-temporal regions of high uncertainty to significantly improve UAV processing speeds without domain-specific loss of accuracy. Alongside, we introduce our WildLive dataset, which comprises 200k+ annotated animal instances across 19k+ frames from 4K UAV videos collected at the Ol Pejeta Conservancy in Kenya. All frames contain ground truth bounding boxes, segmentation masks, as well as individual tracklets and tracking point trajectories. We compare our system against current object tracking approaches including OC-SORT, ByteTrack, and SORT. Our multi-animal tracking experiments with onboard hardware confirm that near real-time high-resolution wildlife tracking is possible on UAVs whilst maintaining high accuracy levels as needed for future navigational and mission-specific animal-centric operational autonomy.
comment: Submitted in CV4Animals 2025
☆ COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts
Current object detectors often suffer significant perfor-mance degradation in real-world applications when encountering distributional shifts. Consequently, the out-of-distribution (OOD) generalization capability of object detectors has garnered increasing attention from researchers. Despite this growing interest, there remains a lack of a large-scale, comprehensive dataset and evaluation benchmark with fine-grained annotations tailored to assess the OOD generalization on more intricate tasks like object detection and grounding. To address this gap, we introduce COUNTS, a large-scale OOD dataset with object-level annotations. COUNTS encompasses 14 natural distributional shifts, over 222K samples, and more than 1,196K labeled bounding boxes. Leveraging COUNTS, we introduce two novel benchmarks: O(OD)2 and OODG. O(OD)2 is designed to comprehensively evaluate the OOD generalization capabilities of object detectors by utilizing controlled distribution shifts between training and testing data. OODG, on the other hand, aims to assess the OOD generalization of grounding abilities in multimodal large language models (MLLMs). Our findings reveal that, while large models and extensive pre-training data substantially en hance performance in in-distribution (IID) scenarios, significant limitations and opportunities for improvement persist in OOD contexts for both object detectors and MLLMs. In visual grounding tasks, even the advanced GPT-4o and Gemini-1.5 only achieve 56.7% and 28.0% accuracy, respectively. We hope COUNTS facilitates advancements in the development and assessment of robust object detectors and MLLMs capable of maintaining high performance under distributional shifts.
☆ Hierarchical and Step-Layer-Wise Tuning of Attention Specialty for Multi-Instance Synthesis in Diffusion Transformers
Text-to-image (T2I) generation models often struggle with multi-instance synthesis (MIS), where they must accurately depict multiple distinct instances in a single image based on complex prompts detailing individual features. Traditional MIS control methods for UNet architectures like SD v1.5/SDXL fail to adapt to DiT-based models like FLUX and SD v3.5, which rely on integrated attention between image and text tokens rather than text-image cross-attention. To enhance MIS in DiT, we first analyze the mixed attention mechanism in DiT. Our token-wise and layer-wise analysis of attention maps reveals a hierarchical response structure: instance tokens dominate early layers, background tokens in middle layers, and attribute tokens in later layers. Building on this observation, we propose a training-free approach for enhancing MIS in DiT-based models with hierarchical and step-layer-wise attention specialty tuning (AST). AST amplifies key regions while suppressing irrelevant areas in distinct attention maps across layers and steps, guided by the hierarchical structure. This optimizes multimodal interactions by hierarchically decoupling the complex prompts with instance-based sketches. We evaluate our approach using upgraded sketch-based layouts for the T2I-CompBench and customized complex scenes. Both quantitative and qualitative results confirm our method enhances complex layout generation, ensuring precise instance placement and attribute representation in MIS.
☆ Negate or Embrace: On How Misalignment Shapes Multimodal Representation Learning
Multimodal representation learning, exemplified by multimodal contrastive learning (MMCL) using image-text pairs, aims to learn powerful representations by aligning cues across modalities. This approach relies on the core assumption that the exemplar image-text pairs constitute two representations of an identical concept. However, recent research has revealed that real-world datasets often exhibit misalignment. There are two distinct viewpoints on how to address this issue: one suggests mitigating the misalignment, and the other leveraging it. We seek here to reconcile these seemingly opposing perspectives, and to provide a practical guide for practitioners. Using latent variable models we thus formalize misalignment by introducing two specific mechanisms: selection bias, where some semantic variables are missing, and perturbation bias, where semantic variables are distorted -- both affecting latent variables shared across modalities. Our theoretical analysis demonstrates that, under mild assumptions, the representations learned by MMCL capture exactly the information related to the subset of the semantic variables invariant to selection and perturbation biases. This provides a unified perspective for understanding misalignment. Based on this, we further offer actionable insights into how misalignment should inform the design of real-world ML systems. We validate our theoretical findings through extensive empirical studies on both synthetic data and real image-text datasets, shedding light on the nuanced impact of misalignment on multimodal representation learning.
comment: 38 pages
☆ Breaking the Data Barrier -- Building GUI Agents Through Task Generalization
Graphical User Interface (GUI) agents offer cross-platform solutions for automating complex digital tasks, with significant potential to transform productivity workflows. However, their performance is often constrained by the scarcity of high-quality trajectory data. To address this limitation, we propose training Vision Language Models (VLMs) on data-rich, reasoning-intensive tasks during a dedicated mid-training stage, and then examine how incorporating these tasks facilitates generalization to GUI planning scenarios. Specifically, we explore a range of tasks with readily available instruction-tuning data, including GUI perception, multimodal reasoning, and textual reasoning. Through extensive experiments across 11 mid-training tasks, we demonstrate that: (1) Task generalization proves highly effective, yielding substantial improvements across most settings. For instance, multimodal mathematical reasoning enhances performance on AndroidWorld by an absolute 6.3%. Remarkably, text-only mathematical data significantly boosts GUI web agent performance, achieving a 5.6% improvement on WebArena and 5.4% improvement on AndroidWorld, underscoring notable cross-modal generalization from text-based to visual domains; (2) Contrary to prior assumptions, GUI perception data - previously considered closely aligned with GUI agent tasks and widely utilized for training - has a comparatively limited impact on final performance; (3) Building on these insights, we identify the most effective mid-training tasks and curate optimized mixture datasets, resulting in absolute performance gains of 8.0% on WebArena and 12.2% on AndroidWorld. Our work provides valuable insights into cross-domain knowledge transfer for GUI agents and offers a practical approach to addressing data scarcity challenges in this emerging field. The code, data and models will be available at https://github.com/hkust-nlp/GUIMid.
comment: 24 pages, 11 figures
☆ M2S-RoAD: Multi-Modal Semantic Segmentation for Road Damage Using Camera and LiDAR Data
Road damage can create safety and comfort challenges for both human drivers and autonomous vehicles (AVs). This damage is particularly prevalent in rural areas due to less frequent surveying and maintenance of roads. Automated detection of pavement deterioration can be used as an input to AVs and driver assistance systems to improve road safety. Current research in this field has predominantly focused on urban environments driven largely by public datasets, while rural areas have received significantly less attention. This paper introduces M2S-RoAD, a dataset for the semantic segmentation of different classes of road damage. M2S-RoAD was collected in various towns across New South Wales, Australia, and labelled for semantic segmentation to identify nine distinct types of road damage. This dataset will be released upon the acceptance of the paper.
☆ AGO: Adaptive Grounding for Open World 3D Occupancy Prediction
Open-world 3D semantic occupancy prediction aims to generate a voxelized 3D representation from sensor inputs while recognizing both known and unknown objects. Transferring open-vocabulary knowledge from vision-language models (VLMs) offers a promising direction but remains challenging. However, methods based on VLM-derived 2D pseudo-labels with traditional supervision are limited by a predefined label space and lack general prediction capabilities. Direct alignment with pretrained image embeddings, on the other hand, fails to achieve reliable performance due to often inconsistent image and text representations in VLMs. To address these challenges, we propose AGO, a novel 3D occupancy prediction framework with adaptive grounding to handle diverse open-world scenarios. AGO first encodes surrounding images and class prompts into 3D and text embeddings, respectively, leveraging similarity-based grounding training with 3D pseudo-labels. Additionally, a modality adapter maps 3D embeddings into a space aligned with VLM-derived image embeddings, reducing modality gaps. Experiments on Occ3D-nuScenes show that AGO improves unknown object prediction in zero-shot and few-shot transfer while achieving state-of-the-art closed-world self-supervised performance, surpassing prior methods by 4.09 mIoU.
☆ SoccerNet-v3D: Leveraging Sports Broadcast Replays for 3D Scene Understanding
Sports video analysis is a key domain in computer vision, enabling detailed spatial understanding through multi-view correspondences. In this work, we introduce SoccerNet-v3D and ISSIA-3D, two enhanced and scalable datasets designed for 3D scene understanding in soccer broadcast analysis. These datasets extend SoccerNet-v3 and ISSIA by incorporating field-line-based camera calibration and multi-view synchronization, enabling 3D object localization through triangulation. We propose a monocular 3D ball localization task built upon the triangulation of ground-truth 2D ball annotations, along with several calibration and reprojection metrics to assess annotation quality on demand. Additionally, we present a single-image 3D ball localization method as a baseline, leveraging camera calibration and ball size priors to estimate the ball's position from a monocular viewpoint. To further refine 2D annotations, we introduce a bounding box optimization technique that ensures alignment with the 3D scene representation. Our proposed datasets establish new benchmarks for 3D soccer scene understanding, enhancing both spatial and temporal analysis in sports analytics. Finally, we provide code to facilitate access to our annotations and the generation pipelines for the datasets.
☆ Global and Local Mamba Network for Multi-Modality Medical Image Super-Resolution
Convolutional neural networks and Transformer have made significant progresses in multi-modality medical image super-resolution. However, these methods either have a fixed receptive field for local learning or significant computational burdens for global learning, limiting the super-resolution performance. To solve this problem, State Space Models, notably Mamba, is introduced to efficiently model long-range dependencies in images with linear computational complexity. Relying on the Mamba and the fact that low-resolution images rely on global information to compensate for missing details, while high-resolution reference images need to provide more local details for accurate super-resolution, we propose a global and local Mamba network (GLMamba) for multi-modality medical image super-resolution. To be specific, our GLMamba is a two-branch network equipped with a global Mamba branch and a local Mamba branch. The global Mamba branch captures long-range relationships in low-resolution inputs, and the local Mamba branch focuses more on short-range details in high-resolution reference images. We also use the deform block to adaptively extract features of both branches to enhance the representation ability. A modulator is designed to further enhance deformable features in both global and local Mamba blocks. To fully integrate the reference image for low-resolution image super-resolution, we further develop a multi-modality feature fusion block to adaptively fuse features by considering similarities, differences, and complementary aspects between modalities. In addition, a contrastive edge loss (CELoss) is developed for sufficient enhancement of edge textures and contrast in medical images.
☆ CameraBench: Benchmarking Visual Reasoning in MLLMs via Photography
Large language models (LLMs) and multimodal large language models (MLLMs) have significantly advanced artificial intelligence. However, visual reasoning, reasoning involving both visual and textual inputs, remains underexplored. Recent advancements, including the reasoning models like OpenAI o1 and Gemini 2.0 Flash Thinking, which incorporate image inputs, have opened this capability. In this ongoing work, we focus specifically on photography-related tasks because a photo is a visual snapshot of the physical world where the underlying physics (i.e., illumination, blur extent, etc.) interplay with the camera parameters. Successfully reasoning from the visual information of a photo to identify these numerical camera settings requires the MLLMs to have a deeper understanding of the underlying physics for precise visual comprehension, representing a challenging and intelligent capability essential for practical applications like photography assistant agents. We aim to evaluate MLLMs on their ability to distinguish visual differences related to numerical camera settings, extending a methodology previously proposed for vision-language models (VLMs). Our preliminary results demonstrate the importance of visual reasoning in photography-related tasks. Moreover, these results show that no single MLLM consistently dominates across all evaluation tasks, demonstrating ongoing challenges and opportunities in developing MLLMs with better visual reasoning.
☆ UP-Person: Unified Parameter-Efficient Transfer Learning for Text-based Person Retrieval
Text-based Person Retrieval (TPR) as a multi-modal task, which aims to retrieve the target person from a pool of candidate images given a text description, has recently garnered considerable attention due to the progress of contrastive visual-language pre-trained model. Prior works leverage pre-trained CLIP to extract person visual and textual features and fully fine-tune the entire network, which have shown notable performance improvements compared to uni-modal pre-training models. However, full-tuning a large model is prone to overfitting and hinders the generalization ability. In this paper, we propose a novel Unified Parameter-Efficient Transfer Learning (PETL) method for Text-based Person Retrieval (UP-Person) to thoroughly transfer the multi-modal knowledge from CLIP. Specifically, UP-Person simultaneously integrates three lightweight PETL components including Prefix, LoRA and Adapter, where Prefix and LoRA are devised together to mine local information with task-specific information prompts, and Adapter is designed to adjust global feature representations. Additionally, two vanilla submodules are optimized to adapt to the unified architecture of TPR. For one thing, S-Prefix is proposed to boost attention of prefix and enhance the gradient propagation of prefix tokens, which improves the flexibility and performance of the vanilla prefix. For another thing, L-Adapter is designed in parallel with layer normalization to adjust the overall distribution, which can resolve conflicts caused by overlap and interaction among multiple submodules. Extensive experimental results demonstrate that our UP-Person achieves state-of-the-art results across various person retrieval datasets, including CUHK-PEDES, ICFG-PEDES and RSTPReid while merely fine-tuning 4.7\% parameters. Code is available at https://github.com/Liu-Yating/UP-Person.
comment: 16 pages, 7 figures, first submited to IEEE TCSVT on 2024 May. Under review
☆ Learning to Harmonize Cross-vendor X-ray Images by Non-linear Image Dynamics Correction
In this paper, we explore how conventional image enhancement can improve model robustness in medical image analysis. By applying commonly used normalization methods to images from various vendors and studying their influence on model generalization in transfer learning, we show that the nonlinear characteristics of domain-specific image dynamics cannot be addressed by simple linear transforms. To tackle this issue, we reformulate the image harmonization task as an exposure correction problem and propose a method termed Global Deep Curve Estimation (GDCE) to reduce domain-specific exposure mismatch. GDCE performs enhancement via a pre-defined polynomial function and is trained with the help of a ``domain discriminator'', aiming to improve model transparency in downstream tasks compared to existing black-box methods.
☆ Hierarchical Relation-augmented Representation Generalization for Few-shot Action Recognition
Few-shot action recognition (FSAR) aims to recognize novel action categories with few exemplars. Existing methods typically learn frame-level representations independently for each video by designing various inter-frame temporal modeling strategies. However, they neglect explicit relation modeling between videos and tasks, thus failing to capture shared temporal patterns across videos and reuse temporal knowledge from historical tasks. In light of this, we propose HR2G-shot, a Hierarchical Relation-augmented Representation Generalization framework for FSAR, which unifies three types of relation modeling (inter-frame, inter-video, and inter-task) to learn task-specific temporal patterns from a holistic view. In addition to conducting inter-frame temporal interactions, we further devise two components to respectively explore inter-video and inter-task relationships: i) Inter-video Semantic Correlation (ISC) performs cross-video frame-level interactions in a fine-grained manner, thereby capturing task-specific query features and learning intra- and inter-class temporal correlations among support features; ii) Inter-task Knowledge Transfer (IKT) retrieves and aggregates relevant temporal knowledge from the bank, which stores diverse temporal patterns from historical tasks. Extensive experiments on five benchmarks show that HR2G-shot outperforms current top-leading FSAR methods.
☆ DTFSal: Audio-Visual Dynamic Token Fusion for Video Saliency Prediction
Audio-visual saliency prediction aims to mimic human visual attention by identifying salient regions in videos through the integration of both visual and auditory information. Although visual-only approaches have significantly advanced, effectively incorporating auditory cues remains challenging due to complex spatio-temporal interactions and high computational demands. To address these challenges, we propose Dynamic Token Fusion Saliency (DFTSal), a novel audio-visual saliency prediction framework designed to balance accuracy with computational efficiency. Our approach features a multi-scale visual encoder equipped with two novel modules: the Learnable Token Enhancement Block (LTEB), which adaptively weights tokens to emphasize crucial saliency cues, and the Dynamic Learnable Token Fusion Block (DLTFB), which employs a shifting operation to reorganize and merge features, effectively capturing long-range dependencies and detailed spatial information. In parallel, an audio branch processes raw audio signals to extract meaningful auditory features. Both visual and audio features are integrated using our Adaptive Multimodal Fusion Block (AMFB), which employs local, global, and adaptive fusion streams for precise cross-modal fusion. The resulting fused features are processed by a hierarchical multi-decoder structure, producing accurate saliency maps. Extensive evaluations on six audio-visual benchmarks demonstrate that DFTSal achieves SOTA performance while maintaining computational efficiency.
☆ Mavors: Multi-granularity Video Representation for Multimodal Large Language Model
Long-context video understanding in multimodal large language models (MLLMs) faces a critical challenge: balancing computational efficiency with the retention of fine-grained spatio-temporal patterns. Existing approaches (e.g., sparse sampling, dense sampling with low resolution, and token compression) suffer from significant information loss in temporal dynamics, spatial details, or subtle interactions, particularly in videos with complex motion or varying resolutions. To address this, we propose $\mathbf{Mavors}$, a novel framework that introduces $\mathbf{M}$ulti-gr$\mathbf{a}$nularity $\mathbf{v}$ide$\mathbf{o}$ $\mathbf{r}$epre$\mathbf{s}$entation for holistic long-video modeling. Specifically, Mavors directly encodes raw video content into latent representations through two core components: 1) an Intra-chunk Vision Encoder (IVE) that preserves high-resolution spatial features via 3D convolutions and Vision Transformers, and 2) an Inter-chunk Feature Aggregator (IFA) that establishes temporal coherence across chunks using transformer-based dependency modeling with chunk-level rotary position encodings. Moreover, the framework unifies image and video understanding by treating images as single-frame videos via sub-image decomposition. Experiments across diverse benchmarks demonstrate Mavors' superiority in maintaining both spatial fidelity and temporal continuity, significantly outperforming existing methods in tasks requiring fine-grained spatio-temporal reasoning.
comment: 22 pages
☆ Summarization of Multimodal Presentations with Vision-Language Models: Study of the Effect of Modalities and Structure
Vision-Language Models (VLMs) can process visual and textual information in multiple formats: texts, images, interleaved texts and images, or even hour-long videos. In this work, we conduct fine-grained quantitative and qualitative analyses of automatic summarization of multimodal presentations using VLMs with various representations as input. From these experiments, we suggest cost-effective strategies for generating summaries from text-heavy multimodal documents under different input-length budgets using VLMs. We show that slides extracted from the video stream can be beneficially used as input against the raw video, and that a structured representation from interleaved slides and transcript provides the best performance. Finally, we reflect and comment on the nature of cross-modal interactions in multimodal presentations and share suggestions to improve the capabilities of VLMs to understand documents of this nature.
☆ Multi-Object Grounding via Hierarchical Contrastive Siamese Transformers
Multi-object grounding in 3D scenes involves localizing multiple objects based on natural language input. While previous work has primarily focused on single-object grounding, real-world scenarios often demand the localization of several objects. To tackle this challenge, we propose Hierarchical Contrastive Siamese Transformers (H-COST), which employs a Hierarchical Processing strategy to progressively refine object localization, enhancing the understanding of complex language instructions. Additionally, we introduce a Contrastive Siamese Transformer framework, where two networks with the identical structure are used: one auxiliary network processes robust object relations from ground-truth labels to guide and enhance the second network, the reference network, which operates on segmented point-cloud data. This contrastive mechanism strengthens the model' s semantic understanding and significantly enhances its ability to process complex point-cloud data. Our approach outperforms previous state-of-the-art methods by 9.5% on challenging multi-object grounding benchmarks.
☆ Aligning Anime Video Generation with Human Feedback
Anime video generation faces significant challenges due to the scarcity of anime data and unusual motion patterns, leading to issues such as motion distortion and flickering artifacts, which result in misalignment with human preferences. Existing reward models, designed primarily for real-world videos, fail to capture the unique appearance and consistency requirements of anime. In this work, we propose a pipeline to enhance anime video generation by leveraging human feedback for better alignment. Specifically, we construct the first multi-dimensional reward dataset for anime videos, comprising 30k human-annotated samples that incorporating human preferences for both visual appearance and visual consistency. Based on this, we develop AnimeReward, a powerful reward model that employs specialized vision-language models for different evaluation dimensions to guide preference alignment. Furthermore, we introduce Gap-Aware Preference Optimization (GAPO), a novel training method that explicitly incorporates preference gaps into the optimization process, enhancing alignment performance and efficiency. Extensive experiment results show that AnimeReward outperforms existing reward models, and the inclusion of GAPO leads to superior alignment in both quantitative benchmarks and human evaluations, demonstrating the effectiveness of our pipeline in enhancing anime video quality. Our dataset and code will be publicly available.
comment: 10 pages, 5 figures, 7 tables
☆ Prior Does Matter: Visual Navigation via Denoising Diffusion Bridge Models
Recent advancements in diffusion-based imitation learning, which show impressive performance in modeling multimodal distributions and training stability, have led to substantial progress in various robot learning tasks. In visual navigation, previous diffusion-based policies typically generate action sequences by initiating from denoising Gaussian noise. However, the target action distribution often diverges significantly from Gaussian noise, leading to redundant denoising steps and increased learning complexity. Additionally, the sparsity of effective action distributions makes it challenging for the policy to generate accurate actions without guidance. To address these issues, we propose a novel, unified visual navigation framework leveraging the denoising diffusion bridge models named NaviBridger. This approach enables action generation by initiating from any informative prior actions, enhancing guidance and efficiency in the denoising process. We explore how diffusion bridges can enhance imitation learning in visual navigation tasks and further examine three source policies for generating prior actions. Extensive experiments in both simulated and real-world indoor and outdoor scenarios demonstrate that NaviBridger accelerates policy inference and outperforms the baselines in generating target action sequences. Code is available at https://github.com/hren20/NaiviBridger.
☆ Investigating the Role of Bilateral Symmetry for Inpainting Brain MRI
Inpainting has recently emerged as a valuable and interesting technology to employ in the analysis of medical imaging data, in particular brain MRI. A wide variety of methodologies for inpainting MRI have been proposed and demonstrated on tasks including anomaly detection. In this work we investigate the statistical relationship between inpainted brain structures and the amount of subject-specific conditioning information, i.e. the other areas of the image that are masked. In particular, we analyse the distribution of inpainting results when masking additional regions of the image, specifically the contra-lateral structure. This allows us to elucidate where in the brain the model is drawing information from, and in particular, what is the importance of hemispherical symmetry? Our experiments interrogate a diffusion inpainting model through analysing the inpainting of subcortical brain structures based on intensity and estimated area change. We demonstrate that some structures show a strong influence of symmetry in the conditioning of the inpainting process.
☆ TT3D: Table Tennis 3D Reconstruction
Sports analysis requires processing large amounts of data, which is time-consuming and costly. Advancements in neural networks have significantly alleviated this burden, enabling highly accurate ball tracking in sports broadcasts. However, relying solely on 2D ball tracking is limiting, as it depends on the camera's viewpoint and falls short of supporting comprehensive game analysis. To address this limitation, we propose a novel approach for reconstructing precise 3D ball trajectories from online table tennis match recordings. Our method leverages the underlying physics of the ball's motion to identify the bounce state that minimizes the reprojection error of the ball's flying trajectory, hence ensuring an accurate and reliable 3D reconstruction. A key advantage of our approach is its ability to infer ball spin without relying on human pose estimation or racket tracking, which are often unreliable or unavailable in broadcast footage. We developed an automated camera calibration method capable of reliably tracking camera movements. Additionally, we adapted an existing 3D pose estimation model, which lacks depth motion capture, to accurately track player movements. Together, these contributions enable the full 3D reconstruction of a table tennis rally.
comment: Accepted to CVSport 2025
☆ Progressive Transfer Learning for Multi-Pass Fundus Image Restoration
Diabetic retinopathy is a leading cause of vision impairment, making its early diagnosis through fundus imaging critical for effective treatment planning. However, the presence of poor quality fundus images caused by factors such as inadequate illumination, noise, blurring and other motion artifacts yields a significant challenge for accurate DR screening. In this study, we propose progressive transfer learning for multi pass restoration to iteratively enhance the quality of degraded fundus images, ensuring more reliable DR screening. Unlike previous methods that often focus on a single pass restoration, multi pass restoration via PTL can achieve a superior blind restoration performance that can even improve most of the good quality fundus images in the dataset. Initially, a Cycle GAN model is trained to restore low quality images, followed by PTL induced restoration passes over the latest restored outputs to improve overall quality in each pass. The proposed method can learn blind restoration without requiring any paired data while surpassing its limitations by leveraging progressive learning and fine tuning strategies to minimize distortions and preserve critical retinal features. To evaluate PTL's effectiveness on multi pass restoration, we conducted experiments on DeepDRiD, a large scale fundus imaging dataset specifically curated for diabetic retinopathy detection. Our result demonstrates state of the art performance, showcasing PTL's potential as a superior approach to iterative image quality restoration.
comment: 13 pages, 12 figures including appendix
☆ Relative Illumination Fields: Learning Medium and Light Independent Underwater Scenes
We address the challenge of constructing a consistent and photorealistic Neural Radiance Field in inhomogeneously illuminated, scattering environments with unknown, co-moving light sources. While most existing works on underwater scene representation focus on a static homogeneous illumination, limited attention has been paid to scenarios such as when a robot explores water deeper than a few tens of meters, where sunlight becomes insufficient. To address this, we propose a novel illumination field locally attached to the camera, enabling the capture of uneven lighting effects within the viewing frustum. We combine this with a volumetric medium representation to an overall method that effectively handles interaction between dynamic illumination field and static scattering medium. Evaluation results demonstrate the effectiveness and flexibility of our approach.
comment: 10 pages, 6 figures. First two authors contributed equally to this work
☆ Masked Autoencoder Self Pre-Training for Defect Detection in Microelectronics
Whereas in general computer vision, transformer-based architectures have quickly become the gold standard, microelectronics defect detection still heavily relies on convolutional neural networks (CNNs). We hypothesize that this is due to the fact that a) transformers have an increased need for data and b) labelled image generation procedures for microelectronics are costly, and labelled data is therefore sparse. Whereas in other domains, pre-training on large natural image datasets can mitigate this problem, in microelectronics transfer learning is hindered due to the dissimilarity of domain data and natural images. Therefore, we evaluate self pre-training, where models are pre-trained on the target dataset, rather than another dataset. We propose a vision transformer (ViT) pre-training framework for defect detection in microelectronics based on masked autoencoders (MAE). In MAE, a large share of image patches is masked and reconstructed by the model during pre-training. We perform pre-training and defect detection using a dataset of less than 10.000 scanning acoustic microscopy (SAM) images labelled using transient thermal analysis (TTA). Our experimental results show that our approach leads to substantial performance gains compared to a) supervised ViT, b) ViT pre-trained on natural image datasets, and c) state-of-the-art CNN-based defect detection models used in the literature. Additionally, interpretability analysis reveals that our self pre-trained models, in comparison to ViT baselines, correctly focus on defect-relevant features such as cracks in the solder material. This demonstrates that our approach yields fault-specific feature representations, making our self pre-trained models viable for real-world defect detection in microelectronics.
comment: 16 pages, 5 figures
☆ The Mirage of Performance Gains: Why Contrastive Decoding Fails to Address Multimodal Hallucination
Contrastive decoding strategies are widely used to reduce hallucinations in multimodal large language models (MLLMs). These methods work by constructing contrastive samples to induce hallucinations and then suppressing them in the output distribution. However, this paper demonstrates that such approaches fail to effectively mitigate the hallucination problem. The performance improvements observed on POPE Benchmark are largely driven by two misleading factors: (1) crude, unidirectional adjustments to the model's output distribution and (2) the adaptive plausibility constraint, which reduces the sampling strategy to greedy search. To further illustrate these issues, we introduce a series of spurious improvement methods and evaluate their performance against contrastive decoding techniques. Experimental results reveal that the observed performance gains in contrastive decoding are entirely unrelated to its intended goal of mitigating hallucinations. Our findings challenge common assumptions about the effectiveness of contrastive decoding strategies and pave the way for developing genuinely effective solutions to hallucinations in MLLMs.
☆ RGB-Event based Pedestrian Attribute Recognition: A Benchmark Dataset and An Asymmetric RWKV Fusion Framework
Existing pedestrian attribute recognition methods are generally developed based on RGB frame cameras. However, these approaches are constrained by the limitations of RGB cameras, such as sensitivity to lighting conditions and motion blur, which hinder their performance. Furthermore, current attribute recognition primarily focuses on analyzing pedestrians' external appearance and clothing, lacking an exploration of emotional dimensions. In this paper, we revisit these issues and propose a novel multi-modal RGB-Event attribute recognition task by drawing inspiration from the advantages of event cameras in low-light, high-speed, and low-power consumption. Specifically, we introduce the first large-scale multi-modal pedestrian attribute recognition dataset, termed EventPAR, comprising 100K paired RGB-Event samples that cover 50 attributes related to both appearance and six human emotions, diverse scenes, and various seasons. By retraining and evaluating mainstream PAR models on this dataset, we establish a comprehensive benchmark and provide a solid foundation for future research in terms of data and algorithmic baselines. In addition, we propose a novel RWKV-based multi-modal pedestrian attribute recognition framework, featuring an RWKV visual encoder and an asymmetric RWKV fusion module. Extensive experiments are conducted on our proposed dataset as well as two simulated datasets (MARS-Attribute and DukeMTMC-VID-Attribute), achieving state-of-the-art results. The source code and dataset will be released on https://github.com/Event-AHU/OpenPAR
comment: The First Benchmark Dataset for RGB-Event Multimodal Pedestrian Attribute Recognition Task
☆ Air Quality Prediction with A Meteorology-Guided Modality-Decoupled Spatio-Temporal Network
Air quality prediction plays a crucial role in public health and environmental protection. Accurate air quality prediction is a complex multivariate spatiotemporal problem, that involves interactions across temporal patterns, pollutant correlations, spatial station dependencies, and particularly meteorological influences that govern pollutant dispersion and chemical transformations. Existing works underestimate the critical role of atmospheric conditions in air quality prediction and neglect comprehensive meteorological data utilization, thereby impairing the modeling of dynamic interdependencies between air quality and meteorological data. To overcome this, we propose MDSTNet, an encoder-decoder framework that explicitly models air quality observations and atmospheric conditions as distinct modalities, integrating multi-pressure-level meteorological data and weather forecasts to capture atmosphere-pollution dependencies for prediction. Meantime, we construct ChinaAirNet, the first nationwide dataset combining air quality records with multi-pressure-level meteorological observations. Experimental results on ChinaAirNet demonstrate MDSTNet's superiority, substantially reducing 48-hour prediction errors by 17.54\% compared to the state-of-the-art model. The source code and dataset will be available on github.
☆ EBAD-Gaussian: Event-driven Bundle Adjusted Deblur Gaussian Splatting
While 3D Gaussian Splatting (3D-GS) achieves photorealistic novel view synthesis, its performance degrades with motion blur. In scenarios with rapid motion or low-light conditions, existing RGB-based deblurring methods struggle to model camera pose and radiance changes during exposure, reducing reconstruction accuracy. Event cameras, capturing continuous brightness changes during exposure, can effectively assist in modeling motion blur and improving reconstruction quality. Therefore, we propose Event-driven Bundle Adjusted Deblur Gaussian Splatting (EBAD-Gaussian), which reconstructs sharp 3D Gaussians from event streams and severely blurred images. This method jointly learns the parameters of these Gaussians while recovering camera motion trajectories during exposure time. Specifically, we first construct a blur loss function by synthesizing multiple latent sharp images during the exposure time, minimizing the difference between real and synthesized blurred images. Then we use event stream to supervise the light intensity changes between latent sharp images at any time within the exposure period, supplementing the light intensity dynamic changes lost in RGB images. Furthermore, we optimize the latent sharp images at intermediate exposure times based on the event-based double integral (EDI) prior, applying consistency constraints to enhance the details and texture information of the reconstructed images. Extensive experiments on synthetic and real-world datasets show that EBAD-Gaussian can achieve high-quality 3D scene reconstruction under the condition of blurred images and event stream inputs.
☆ Balancing Two Classifiers via A Simplex ETF Structure for Model Calibration CVPR2025
In recent years, deep neural networks (DNNs) have demonstrated state-of-the-art performance across various domains. However, despite their success, they often face calibration issues, particularly in safety-critical applications such as autonomous driving and healthcare, where unreliable predictions can have serious consequences. Recent research has started to improve model calibration from the view of the classifier. However, the exploration of designing the classifier to solve the model calibration problem is insufficient. Let alone most of the existing methods ignore the calibration errors arising from underconfidence. In this work, we propose a novel method by balancing learnable and ETF classifiers to solve the overconfidence or underconfidence problem for model Calibration named BalCAL. By introducing a confidence-tunable module and a dynamic adjustment method, we ensure better alignment between model confidence and its true accuracy. Extensive experimental validation shows that ours significantly improves model calibration performance while maintaining high predictive accuracy, outperforming existing techniques. This provides a novel solution to the calibration challenges commonly encountered in deep learning.
comment: CVPR2025
☆ An Image is Worth $K$ Topics: A Visual Structural Topic Model with Pretrained Image Embeddings
Political scientists are increasingly interested in analyzing visual content at scale. However, the existing computational toolbox is still in need of methods and models attuned to the specific challenges and goals of social and political inquiry. In this article, we introduce a visual Structural Topic Model (vSTM) that combines pretrained image embeddings with a structural topic model. This has important advantages compared to existing approaches. First, pretrained embeddings allow the model to capture the semantic complexity of images relevant to political contexts. Second, the structural topic model provides the ability to analyze how topics and covariates are related, while maintaining a nuanced representation of images as a mixture of multiple topics. In our empirical application, we show that the vSTM is able to identify topics that are interpretable, coherent, and substantively relevant to the study of online political communication.
☆ NaviDiffusor: Cost-Guided Diffusion Model for Visual Navigation
Visual navigation, a fundamental challenge in mobile robotics, demands versatile policies to handle diverse environments. Classical methods leverage geometric solutions to minimize specific costs, offering adaptability to new scenarios but are prone to system errors due to their multi-modular design and reliance on hand-crafted rules. Learning-based methods, while achieving high planning success rates, face difficulties in generalizing to unseen environments beyond the training data and often require extensive training. To address these limitations, we propose a hybrid approach that combines the strengths of learning-based methods and classical approaches for RGB-only visual navigation. Our method first trains a conditional diffusion model on diverse path-RGB observation pairs. During inference, it integrates the gradients of differentiable scene-specific and task-level costs, guiding the diffusion model to generate valid paths that meet the constraints. This approach alleviates the need for retraining, offering a plug-and-play solution. Extensive experiments in both indoor and outdoor settings, across simulated and real-world scenarios, demonstrate zero-shot transfer capability of our approach, achieving higher success rates and fewer collisions compared to baseline methods. Code will be released at https://github.com/SYSU-RoboticsLab/NaviD.
☆ GaussVideoDreamer: 3D Scene Generation with Video Diffusion and Inconsistency-Aware Gaussian Splatting
Single-image 3D scene reconstruction presents significant challenges due to its inherently ill-posed nature and limited input constraints. Recent advances have explored two promising directions: multiview generative models that train on 3D consistent datasets but struggle with out-of-distribution generalization, and 3D scene inpainting and completion frameworks that suffer from cross-view inconsistency and suboptimal error handling, as they depend exclusively on depth data or 3D smoothness, which ultimately degrades output quality and computational performance. Building upon these approaches, we present GaussVideoDreamer, which advances generative multimedia approaches by bridging the gap between image, video, and 3D generation, integrating their strengths through two key innovations: (1) A progressive video inpainting strategy that harnesses temporal coherence for improved multiview consistency and faster convergence. (2) A 3D Gaussian Splatting consistency mask to guide the video diffusion with 3D consistent multiview evidence. Our pipeline combines three core components: a geometry-aware initialization protocol, Inconsistency-Aware Gaussian Splatting, and a progressive video inpainting strategy. Experimental results demonstrate that our approach achieves 32% higher LLaVA-IQA scores and at least 2x speedup compared to existing methods while maintaining robust performance across diverse scenes.
☆ Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models? CVPR 2025
Multi-modal large language models (MLLMs) have made significant progress, yet their safety alignment remains limited. Typically, current open-source MLLMs rely on the alignment inherited from their language module to avoid harmful generations. However, the lack of safety measures specifically designed for multi-modal inputs creates an alignment gap, leaving MLLMs vulnerable to vision-domain attacks such as typographic manipulation. Current methods utilize a carefully designed safety dataset to enhance model defense capability, while the specific knowledge or patterns acquired from the high-quality dataset remain unclear. Through comparison experiments, we find that the alignment gap primarily arises from data distribution biases, while image content, response quality, or the contrastive behavior of the dataset makes little contribution to boosting multi-modal safety. To further investigate this and identify the key factors in improving MLLM safety, we propose finetuning MLLMs on a small set of benign instruct-following data with responses replaced by simple, clear rejection sentences. Experiments show that, without the need for labor-intensive collection of high-quality malicious data, model safety can still be significantly improved, as long as a specific fraction of rejection data exists in the finetuning set, indicating the security alignment is not lost but rather obscured during multi-modal pretraining or instruction finetuning. Simply correcting the underlying data bias could narrow the safety gap in the vision domain.
comment: Accepted to CVPR 2025, codes in process
☆ Metric-Guided Synthesis of Class Activation Mapping
Class activation mapping (CAM) is a widely adopted class of saliency methods used to explain the behavior of convolutional neural networks (CNNs). These methods generate heatmaps that highlight the parts of the input most relevant to the CNN output. Various CAM methods have been proposed, each distinguished by the expressions used to derive heatmaps. In general, users look for heatmaps with specific properties that reflect different aspects of CNN functionality. These may include similarity to ground truth, robustness, equivariance, and more. Although existing CAM methods implicitly encode some of these properties in their expressions, they do not allow for variability in heatmap generation following the user's intent or domain knowledge. In this paper, we address this limitation by introducing SyCAM, a metric-based approach for synthesizing CAM expressions. Given a predefined evaluation metric for saliency maps, SyCAM automatically generates CAM expressions optimized for that metric. We specifically explore a syntax-guided synthesis instantiation of SyCAM, where CAM expressions are derived based on predefined syntactic constraints and the given metric. Using several established evaluation metrics, we demonstrate the efficacy and flexibility of our approach in generating targeted heatmaps. We compare SyCAM with other well-known CAM methods on three prominent models: ResNet50, VGG16, and VGG19.
☆ Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning CVPR
Modeling label correlations has always played a pivotal role in multi-label image classification (MLC), attracting significant attention from researchers. However, recent studies have overemphasized co-occurrence relationships among labels, which can lead to overfitting risk on this overemphasis, resulting in suboptimal models. To tackle this problem, we advocate for balancing correlative and discriminative relationships among labels to mitigate the risk of overfitting and enhance model performance. To this end, we propose the Multi-Label Visual Prompt Tuning framework, a novel and parameter-efficient method that groups classes into multiple class subsets according to label co-occurrence and mutual exclusivity relationships, and then models them respectively to balance the two relationships. In this work, since each group contains multiple classes, multiple prompt tokens are adopted within Vision Transformer (ViT) to capture the correlation or discriminative label relationship within each group, and effectively learn correlation or discriminative representations for class subsets. On the other hand, each group contains multiple group-aware visual representations that may correspond to multiple classes, and the mixture of experts (MoE) model can cleverly assign them from the group-aware to the label-aware, adaptively obtaining label-aware representation, which is more conducive to classification. Experiments on multiple benchmark datasets show that our proposed approach achieves competitive results and outperforms SOTA methods on multiple pre-trained models.
comment: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2025
☆ Resampling Benchmark for Efficient Comprehensive Evaluation of Large Vision-Language Models
We propose an efficient evaluation protocol for large vision-language models (VLMs). Given their broad knowledge and reasoning capabilities, multiple benchmarks are needed for comprehensive assessment, making evaluation computationally expensive. To improve efficiency, we construct a subset that yields results comparable to full benchmark evaluations. Our benchmark classification experiments reveal that no single benchmark fully covers all challenges. We then introduce a subset construction method using farthest point sampling (FPS). Our experiments show that FPS-based benchmarks maintain a strong correlation (> 0.96) with full evaluations while using only ~1\% of the data. Additionally, applying FPS to an existing benchmark improves correlation with overall evaluation results, suggesting its potential to reduce unintended dataset biases.
☆ OctGPT: Octree-based Multiscale Autoregressive Models for 3D Shape Generation SIGGRAPH 2025
Autoregressive models have achieved remarkable success across various domains, yet their performance in 3D shape generation lags significantly behind that of diffusion models. In this paper, we introduce OctGPT, a novel multiscale autoregressive model for 3D shape generation that dramatically improves the efficiency and performance of prior 3D autoregressive approaches, while rivaling or surpassing state-of-the-art diffusion models. Our method employs a serialized octree representation to efficiently capture the hierarchical and spatial structures of 3D shapes. Coarse geometry is encoded via octree structures, while fine-grained details are represented by binary tokens generated using a vector quantized variational autoencoder (VQVAE), transforming 3D shapes into compact \emph{multiscale binary sequences} suitable for autoregressive prediction. To address the computational challenges of handling long sequences, we incorporate octree-based transformers enhanced with 3D rotary positional encodings, scale-specific embeddings, and token-parallel generation schemes. These innovations reduce training time by 13 folds and generation time by 69 folds, enabling the efficient training of high-resolution 3D shapes, e.g.,$1024^3$, on just four NVIDIA 4090 GPUs only within days. OctGPT showcases exceptional versatility across various tasks, including text-, sketch-, and image-conditioned generation, as well as scene-level synthesis involving multiple objects. Extensive experiments demonstrate that OctGPT accelerates convergence and improves generation quality over prior autoregressive methods, offering a new paradigm for high-quality, scalable 3D content creation.
comment: SIGGRAPH 2025
☆ Beyond Degradation Redundancy: Contrastive Prompt Learning for All-in-One Image Restoration
All-in-one image restoration, addressing diverse degradation types with a unified model, presents significant challenges in designing task-specific prompts that effectively guide restoration across multiple degradation scenarios. While adaptive prompt learning enables end-to-end optimization, it often yields overlapping or redundant task representations. Conversely, explicit prompts derived from pretrained classifiers enhance discriminability but may discard critical visual information for reconstruction. To address these limitations, we introduce Contrastive Prompt Learning (CPL), a novel framework that fundamentally enhances prompt-task alignment through two complementary innovations: a \emph{Sparse Prompt Module (SPM)} that efficiently captures degradation-specific features while minimizing redundancy, and a \emph{Contrastive Prompt Regularization (CPR)} that explicitly strengthens task boundaries by incorporating negative prompt samples across different degradation types. Unlike previous approaches that focus primarily on degradation classification, CPL optimizes the critical interaction between prompts and the restoration model itself. Extensive experiments across five comprehensive benchmarks demonstrate that CPL consistently enhances state-of-the-art all-in-one restoration models, achieving significant improvements in both standard multi-task scenarios and challenging composite degradation settings. Our framework establishes new state-of-the-art performance while maintaining parameter efficiency, offering a principled solution for unified image restoration.
comment: Project page: https://github.com/Aitical/CPLIR
☆ Enhancing Multi-task Learning Capability of Medical Generalist Foundation Model via Image-centric Multi-annotation Data
The emergence of medical generalist foundation models has revolutionized conventional task-specific model development paradigms, aiming to better handle multiple tasks through joint training on large-scale medical datasets. However, recent advances prioritize simple data scaling or architectural component enhancement, while neglecting to re-examine multi-task learning from a data-centric perspective. Critically, simply aggregating existing data resources leads to decentralized image-task alignment, which fails to cultivate comprehensive image understanding or align with clinical needs for multi-dimensional image interpretation. In this paper, we introduce the image-centric multi-annotation X-ray dataset (IMAX), the first attempt to enhance the multi-task learning capabilities of medical multi-modal large language models (MLLMs) from the data construction level. To be specific, IMAX is featured from the following attributes: 1) High-quality data curation. A comprehensive collection of more than 354K entries applicable to seven different medical tasks. 2) Image-centric dense annotation. Each X-ray image is associated with an average of 4.10 tasks and 7.46 training entries, ensuring multi-task representation richness per image. Compared to the general decentralized multi-annotation X-ray dataset (DMAX), IMAX consistently demonstrates significant multi-task average performance gains ranging from 3.20% to 21.05% across seven open-source state-of-the-art medical MLLMs. Moreover, we investigate differences in statistical patterns exhibited by IMAX and DMAX training processes, exploring potential correlations between optimization dynamics and multi-task performance. Finally, leveraging the core concept of IMAX data construction, we propose an optimized DMAX-based training strategy to alleviate the dilemma of obtaining high-quality IMAX data in practical scenarios.
☆ SemiETS: Integrating Spatial and Content Consistencies for Semi-Supervised End-to-end Text Spotting CVPR2025
Most previous scene text spotting methods rely on high-quality manual annotations to achieve promising performance. To reduce their expensive costs, we study semi-supervised text spotting (SSTS) to exploit useful information from unlabeled images. However, directly applying existing semi-supervised methods of general scenes to SSTS will face new challenges: 1) inconsistent pseudo labels between detection and recognition tasks, and 2) sub-optimal supervisions caused by inconsistency between teacher/student. Thus, we propose a new Semi-supervised framework for End-to-end Text Spotting, namely SemiETS that leverages the complementarity of text detection and recognition. Specifically, it gradually generates reliable hierarchical pseudo labels for each task, thereby reducing noisy labels. Meanwhile, it extracts important information in locations and transcriptions from bidirectional flows to improve consistency. Extensive experiments on three datasets under various settings demonstrate the effectiveness of SemiETS on arbitrary-shaped text. For example, it outperforms previous state-of-the-art SSL methods by a large margin on end-to-end spotting (+8.7%, +5.6%, and +2.6% H-mean under 0.5%, 1%, and 2% labeled data settings on Total-Text, respectively). More importantly, it still improves upon a strongly supervised text spotter trained with plenty of labeled data by 2.0%. Compelling domain adaptation ability shows practical potential. Moreover, our method demonstrates consistent improvement on different text spotters.
comment: Accepted by CVPR2025. Code will be available at \url{https://github.com/DrLuo/SemiETS}
☆ Dual-Path Enhancements in Event-Based Eye Tracking: Augmented Robustness and Adaptive Temporal Modeling CVPR
Event-based eye tracking has become a pivotal technology for augmented reality and human-computer interaction. Yet, existing methods struggle with real-world challenges such as abrupt eye movements and environmental noise. Building on the efficiency of the Lightweight Spatiotemporal Network-a causal architecture optimized for edge devices-we introduce two key advancements. First, a robust data augmentation pipeline incorporating temporal shift, spatial flip, and event deletion improves model resilience, reducing Euclidean distance error by 12% (1.61 vs. 1.70 baseline) on challenging samples. Second, we propose KnightPupil, a hybrid architecture combining an EfficientNet-B3 backbone for spatial feature extraction, a bidirectional GRU for contextual temporal modeling, and a Linear Time-Varying State-Space Module to adapt to sparse inputs and noise dynamically. Evaluated on the 3ET+ benchmark, our framework achieved 1.61 Euclidean distance on the private test set of the Event-based Eye Tracking Challenge at CVPR 2025, demonstrating its effectiveness for practical deployment in AR/VR systems while providing a foundation for future innovations in neuromorphic vision.
comment: Camera-ready version for CVPRW 2025. Accepted for presentation at the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW 2025)
☆ Semantic Depth Matters: Explaining Errors of Deep Vision Networks through Perceived Class Similarities
Understanding deep neural network (DNN) behavior requires more than evaluating classification accuracy alone; analyzing errors and their predictability is equally crucial. Current evaluation methodologies lack transparency, particularly in explaining the underlying causes of network misclassifications. To address this, we introduce a novel framework that investigates the relationship between the semantic hierarchy depth perceived by a network and its real-data misclassification patterns. Central to our framework is the Similarity Depth (SD) metric, which quantifies the semantic hierarchy depth perceived by a network along with a method of evaluation of how closely the network's errors align with its internally perceived similarity structure. We also propose a graph-based visualization of model semantic relationships and misperceptions. A key advantage of our approach is that leveraging class templates -- representations derived from classifier layer weights -- is applicable to already trained networks without requiring additional data or experiments. Our approach reveals that deep vision networks encode specific semantic hierarchies and that high semantic depth improves the compliance between perceived class similarities and actual errors.
☆ Efficient 2D to Full 3D Human Pose Uplifting including Joint Rotations CVPR'25
In sports analytics, accurately capturing both the 3D locations and rotations of body joints is essential for understanding an athlete's biomechanics. While Human Mesh Recovery (HMR) models can estimate joint rotations, they often exhibit lower accuracy in joint localization compared to 3D Human Pose Estimation (HPE) models. Recent work addressed this limitation by combining a 3D HPE model with inverse kinematics (IK) to estimate both joint locations and rotations. However, IK is computationally expensive. To overcome this, we propose a novel 2D-to-3D uplifting model that directly estimates 3D human poses, including joint rotations, in a single forward pass. We investigate multiple rotation representations, loss functions, and training strategies - both with and without access to ground truth rotations. Our models achieve state-of-the-art accuracy in rotation estimation, are 150 times faster than the IK-based approach, and surpass HMR models in joint localization precision.
comment: accepted at CVSports@CVPR'25
☆ Pseudo-Label Guided Real-World Image De-weathering: A Learning Framework with Imperfect Supervision
Real-world image de-weathering aims at removingvarious undesirable weather-related artifacts, e.g., rain, snow,and fog. To this end, acquiring ideal training pairs is crucial.Existing real-world datasets are typically constructed paired databy extracting clean and degraded images from live streamsof landscape scene on the Internet. Despite the use of strictfiltering mechanisms during collection, training pairs inevitablyencounter inconsistency in terms of lighting, object position, scenedetails, etc, making de-weathering models possibly suffer fromdeformation artifacts under non-ideal supervision. In this work,we propose a unified solution for real-world image de-weatheringwith non-ideal supervision, i.e., a pseudo-label guided learningframework, to address various inconsistencies within the realworld paired dataset. Generally, it consists of a de-weatheringmodel (De-W) and a Consistent Label Constructor (CLC), bywhich restoration result can be adaptively supervised by originalground-truth image to recover sharp textures while maintainingconsistency with the degraded inputs in non-weather contentthrough the supervision of pseudo-labels. Particularly, a Crossframe Similarity Aggregation (CSA) module is deployed withinCLC to enhance the quality of pseudo-labels by exploring thepotential complementary information of multi-frames throughgraph model. Moreover, we introduce an Information AllocationStrategy (IAS) to integrate the original ground-truth imagesand pseudo-labels, thereby facilitating the joint supervision forthe training of de-weathering model. Extensive experimentsdemonstrate that our method exhibits significant advantageswhen trained on imperfectly aligned de-weathering datasets incomparison with other approaches.
comment: 15 pages, 16 figures
☆ Omni-Dish: Photorealistic and Faithful Image Generation and Editing for Arbitrary Chinese Dishes
Dish images play a crucial role in the digital era, with the demand for culturally distinctive dish images continuously increasing due to the digitization of the food industry and e-commerce. In general cases, existing text-to-image generation models excel in producing high-quality images; however, they struggle to capture diverse characteristics and faithful details of specific domains, particularly Chinese dishes. To address this limitation, we propose Omni-Dish, the first text-to-image generation model specifically tailored for Chinese dishes. We develop a comprehensive dish curation pipeline, building the largest dish dataset to date. Additionally, we introduce a recaption strategy and employ a coarse-to-fine training scheme to help the model better learn fine-grained culinary nuances. During inference, we enhance the user's textual input using a pre-constructed high-quality caption library and a large language model, enabling more photorealistic and faithful image generation. Furthermore, to extend our model's capability for dish editing tasks, we propose Concept-Enhanced P2P. Based on this approach, we build a dish editing dataset and train a specialized editing model. Extensive experiments demonstrate the superiority of our methods.
comment: 10 pages, 10 figures, 3 tables
☆ FUSION: Fully Integration of Vision-Language Representations for Deep Cross-Modal Understanding
We introduce FUSION, a family of multimodal large language models (MLLMs) with a fully vision-language alignment and integration paradigm. Unlike existing methods that primarily rely on late-stage modality interaction during LLM decoding, our approach achieves deep, dynamic integration throughout the entire processing pipeline. To this end, we propose Text-Guided Unified Vision Encoding, incorporating textual information in vision encoding to achieve pixel-level integration. We further design Context-Aware Recursive Alignment Decoding that recursively aggregates visual features conditioned on textual context during decoding, enabling fine-grained, question-level semantic integration. To guide feature mapping and mitigate modality discrepancies, we develop Dual-Supervised Semantic Mapping Loss. Additionally, we construct a Synthesized Language-Driven Question-Answer (QA) dataset through a new data synthesis method, prioritizing high-quality QA pairs to optimize text-guided feature integration. Building on these foundations, we train FUSION at two scales-3B, 8B-and demonstrate that our full-modality integration approach significantly outperforms existing methods with only 630 vision tokens. Notably, FUSION 3B surpasses Cambrian-1 8B and Florence-VL 8B on most benchmarks. FUSION 3B continues to outperform Cambrian-1 8B even when limited to 300 vision tokens. Our ablation studies show that FUSION outperforms LLaVA-NeXT on over half of the benchmarks under same configuration without dynamic resolution, highlighting the effectiveness of our approach. We release our code, model weights, and dataset. https://github.com/starriver030515/FUSION
☆ Improving Multimodal Hateful Meme Detection Exploiting LMM-Generated Knowledge CVPR 2025
Memes have become a dominant form of communication in social media in recent years. Memes are typically humorous and harmless, however there are also memes that promote hate speech, being in this way harmful to individuals and groups based on their identity. Therefore, detecting hateful content in memes has emerged as a task of critical importance. The need for understanding the complex interactions of images and their embedded text renders the hateful meme detection a challenging multimodal task. In this paper we propose to address the aforementioned task leveraging knowledge encoded in powerful Large Multimodal Models (LMM). Specifically, we propose to exploit LMMs in a two-fold manner. First, by extracting knowledge oriented to the hateful meme detection task in order to build strong meme representations. Specifically, generic semantic descriptions and emotions that the images along with their embedded texts elicit are extracted, which are then used to train a simple classification head for hateful meme detection. Second, by developing a novel hard mining approach introducing directly LMM-encoded knowledge to the training process, providing further improvements. We perform extensive experiments on two datasets that validate the effectiveness of the proposed method, achieving state-of-the-art performance. Our code and trained models are publicly available at: https://github.com/IDT-ITI/LMM-CLIP-meme.
comment: Accepted for publication, Multimodal Learning and Applications Workshop (MULA 2025) @ IEEE/CVF CVPR 2025, Nashville, TN, USA, June 2025. This is the authors' "accepted version"
☆ LiteTracker: Leveraging Temporal Causality for Accurate Low-latency Tissue Tracking
Tissue tracking plays a critical role in various surgical navigation and extended reality (XR) applications. While current methods trained on large synthetic datasets achieve high tracking accuracy and generalize well to endoscopic scenes, their runtime performances fail to meet the low-latency requirements necessary for real-time surgical applications. To address this limitation, we propose LiteTracker, a low-latency method for tissue tracking in endoscopic video streams. LiteTracker builds on a state-of-the-art long-term point tracking method, and introduces a set of training-free runtime optimizations. These optimizations enable online, frame-by-frame tracking by leveraging a temporal memory buffer for efficient feature reuse and utilizing prior motion for accurate track initialization. LiteTracker demonstrates significant runtime improvements being around 7x faster than its predecessor and 2x than the state-of-the-art. Beyond its primary focus on efficiency, LiteTracker delivers high-accuracy tracking and occlusion prediction, performing competitively on both the STIR and SuPer datasets. We believe LiteTracker is an important step toward low-latency tissue tracking for real-time surgical applications in the operating room.
☆ Small Object Detection with YOLO: A Performance Analysis Across Model Versions and Hardware
This paper provides an extensive evaluation of YOLO object detection models (v5, v8, v9, v10, v11) by com- paring their performance across various hardware platforms and optimization libraries. Our study investigates inference speed and detection accuracy on Intel and AMD CPUs using popular libraries such as ONNX and OpenVINO, as well as on GPUs through TensorRT and other GPU-optimized frameworks. Furthermore, we analyze the sensitivity of these YOLO models to object size within the image, examining performance when detecting objects that occupy 1%, 2.5%, and 5% of the total area of the image. By identifying the trade-offs in efficiency, accuracy, and object size adaptability, this paper offers insights for optimal model selection based on specific hardware constraints and detection requirements, aiding practitioners in deploying YOLO models effectively for real-world applications.
☆ Digital Staining with Knowledge Distillation: A Unified Framework for Unpaired and Paired-But-Misaligned Data
Staining is essential in cell imaging and medical diagnostics but poses significant challenges, including high cost, time consumption, labor intensity, and irreversible tissue alterations. Recent advances in deep learning have enabled digital staining through supervised model training. However, collecting large-scale, perfectly aligned pairs of stained and unstained images remains difficult. In this work, we propose a novel unsupervised deep learning framework for digital cell staining that reduces the need for extensive paired data using knowledge distillation. We explore two training schemes: (1) unpaired and (2) paired-but-misaligned settings. For the unpaired case, we introduce a two-stage pipeline, comprising light enhancement followed by colorization, as a teacher model. Subsequently, we obtain a student staining generator through knowledge distillation with hybrid non-reference losses. To leverage the pixel-wise information between adjacent sections, we further extend to the paired-but-misaligned setting, adding the Learning to Align module to utilize pixel-level information. Experiment results on our dataset demonstrate that our proposed unsupervised deep staining method can generate stained images with more accurate positions and shapes of the cell targets in both settings. Compared with competing methods, our method achieves improved results both qualitatively and quantitatively (e.g., NIQE and PSNR).We applied our digital staining method to the White Blood Cell (WBC) dataset, investigating its potential for medical applications.
comment: Accepted to IEEE Transactions on Medical Imaging
☆ TAMP: Token-Adaptive Layerwise Pruning in Multimodal Large Language Models
Multimodal Large Language Models (MLLMs) have shown remarkable versatility in understanding diverse multimodal data and tasks. However, these capabilities come with an increased model scale. While post-training pruning reduces model size in unimodal models, its application to MLLMs often yields limited success. Our analysis discovers that conventional methods fail to account for the unique token attributes across layers and modalities inherent to MLLMs. Inspired by this observation, we propose TAMP, a simple yet effective pruning framework tailored for MLLMs, featuring two key components: (1) Diversity-Aware Sparsity, which adjusts sparsity ratio per layer based on diversities among multimodal output tokens, preserving more parameters in high-diversity layers; and (2) Adaptive Multimodal Input Activation, which identifies representative multimodal input tokens using attention scores to guide unstructured weight pruning. We validate our method on two state-of-the-art MLLMs: LLaVA-NeXT, designed for vision-language tasks, and VideoLLaMA2, capable of processing audio, visual, and language modalities. Empirical experiments across various multimodal evaluation benchmarks demonstrate that each component of our approach substantially outperforms existing pruning techniques.
comment: Preprint
☆ Enhanced Semantic Extraction and Guidance for UGC Image Super Resolution
Due to the disparity between real-world degradations in user-generated content(UGC) images and synthetic degradations, traditional super-resolution methods struggle to generalize effectively, necessitating a more robust approach to model real-world distortions. In this paper, we propose a novel approach to UGC image super-resolution by integrating semantic guidance into a diffusion framework. Our method addresses the inconsistency between degradations in wild and synthetic datasets by separately simulating the degradation processes on the LSDIR dataset and combining them with the official paired training set. Furthermore, we enhance degradation removal and detail generation by incorporating a pretrained semantic extraction model (SAM2) and fine-tuning key hyperparameters for improved perceptual fidelity. Extensive experiments demonstrate the superiority of our approach against state-of-the-art methods. Additionally, the proposed model won second place in the CVPR NTIRE 2025 Short-form UGC Image Super-Resolution Challenge, further validating its effectiveness. The code is available at https://github.c10pom/Moonsofang/NTIRE-2025-SRlab.
☆ Separate to Collaborate: Dual-Stream Diffusion Model for Coordinated Piano Hand Motion Synthesis
Automating the synthesis of coordinated bimanual piano performances poses significant challenges, particularly in capturing the intricate choreography between the hands while preserving their distinct kinematic signatures. In this paper, we propose a dual-stream neural framework designed to generate synchronized hand gestures for piano playing from audio input, addressing the critical challenge of modeling both hand independence and coordination. Our framework introduces two key innovations: (i) a decoupled diffusion-based generation framework that independently models each hand's motion via dual-noise initialization, sampling distinct latent noise for each while leveraging a shared positional condition, and (ii) a Hand-Coordinated Asymmetric Attention (HCAA) mechanism suppresses symmetric (common-mode) noise to highlight asymmetric hand-specific features, while adaptively enhancing inter-hand coordination during denoising. The system operates hierarchically: it first predicts 3D hand positions from audio features and then generates joint angles through position-aware diffusion models, where parallel denoising streams interact via HCAA. Comprehensive evaluations demonstrate that our framework outperforms existing state-of-the-art methods across multiple metrics.
comment: 12 pages, 4 figures
☆ Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition AAAI 2025
Visual Place Recognition (VPR) is aimed at predicting the location of a query image by referencing a database of geotagged images. For VPR task, often fewer discriminative local regions in an image produce important effects while mundane background regions do not contribute or even cause perceptual aliasing because of easy overlap. However, existing methods lack precisely modeling and full exploitation of these discriminative regions. In this paper, we propose the Focus on Local (FoL) approach to stimulate the performance of image retrieval and re-ranking in VPR simultaneously by mining and exploiting reliable discriminative local regions in images and introducing pseudo-correlation supervision. First, we design two losses, Extraction-Aggregation Spatial Alignment Loss (SAL) and Foreground-Background Contrast Enhancement Loss (CEL), to explicitly model reliable discriminative local regions and use them to guide the generation of global representations and efficient re-ranking. Second, we introduce a weakly-supervised local feature training strategy based on pseudo-correspondences obtained from aggregating global features to alleviate the lack of local correspondences ground truth for the VPR task. Third, we suggest an efficient re-ranking pipeline that is efficiently and precisely based on discriminative region guidance. Finally, experimental results show that our FoL achieves the state-of-the-art on multiple VPR benchmarks in both image retrieval and re-ranking stages and also significantly outperforms existing two-stage VPR methods in terms of computational efficiency. Code and models are available at https://github.com/chenshunpeng/FoL
comment: Accepted by AAAI 2025
☆ MCBlock: Boosting Neural Radiance Field Training Speed by MCTS-based Dynamic-Resolution Ray Sampling
Neural Radiance Field (NeRF) is widely known for high-fidelity novel view synthesis. However, even the state-of-the-art NeRF model, Gaussian Splatting, requires minutes for training, far from the real-time performance required by multimedia scenarios like telemedicine. One of the obstacles is its inefficient sampling, which is only partially addressed by existing works. Existing point-sampling algorithms uniformly sample simple-texture regions (easy to fit) and complex-texture regions (hard to fit), while existing ray-sampling algorithms sample these regions all in the finest granularity (i.e. the pixel level), both wasting GPU training resources. Actually, regions with different texture intensities require different sampling granularities. To this end, we propose a novel dynamic-resolution ray-sampling algorithm, MCBlock, which employs Monte Carlo Tree Search (MCTS) to partition each training image into pixel blocks with different sizes for active block-wise training. Specifically, the trees are initialized according to the texture of training images to boost the initialization speed, and an expansion/pruning module dynamically optimizes the block partition. MCBlock is implemented in Nerfstudio, an open-source toolset, and achieves a training acceleration of up to 2.33x, surpassing other ray-sampling algorithms. We believe MCBlock can apply to any cone-tracing NeRF model and contribute to the multimedia community.
☆ HDC: Hierarchical Distillation for Multi-level Noisy Consistency in Semi-Supervised Fetal Ultrasound Segmentation
Transvaginal ultrasound is a critical imaging modality for evaluating cervical anatomy and detecting physiological changes. However, accurate segmentation of cervical structures remains challenging due to low contrast, shadow artifacts, and fuzzy boundaries. While convolutional neural networks (CNNs) have shown promising results in medical image segmentation, their performance is often limited by the need for large-scale annotated datasets - an impractical requirement in clinical ultrasound imaging. Semi-supervised learning (SSL) offers a compelling solution by leveraging unlabeled data, but existing teacher-student frameworks often suffer from confirmation bias and high computational costs. We propose HDC, a novel semi-supervised segmentation framework that integrates Hierarchical Distillation and Consistency learning within a multi-level noise mean-teacher framework. Unlike conventional approaches that rely solely on pseudo-labeling, we introduce a hierarchical distillation mechanism that guides feature-level learning via two novel objectives: (1) Correlation Guidance Loss to align feature representations between the teacher and main student branch, and (2) Mutual Information Loss to stabilize representations between the main and noisy student branches. Our framework reduces model complexity while improving generalization. Extensive experiments on two fetal ultrasound datasets, FUGC and PSFH, demonstrate that our method achieves competitive performance with significantly lower computational overhead than existing multi-teacher models.
♻ ☆ SplatMesh: Interactive 3D Segmentation and Editing Using Mesh-Based Gaussian Splatting
A key challenge in fine-grained 3D-based interactive editing is the absence of an efficient representation that balances diverse modifications with high-quality view synthesis under a given memory constraint. While 3D meshes provide robustness for various modifications, they often yield lower-quality view synthesis compared to 3D Gaussian Splatting, which, in turn, suffers from instability during extensive editing. A straightforward combination of these two representations results in suboptimal performance and fails to meet memory constraints. In this paper, we introduce SplatMesh, a novel fine-grained interactive 3D segmentation and editing algorithm that integrates 3D Gaussian Splat with a precomputed mesh and could adjust the memory request based on the requirement. Specifically, given a mesh, \method simplifies it while considering both color and shape, ensuring it meets memory constraints. Then, SplatMesh aligns Gaussian splats with the simplified mesh by treating each triangle as a new reference point. By segmenting and editing the simplified mesh, we can effectively edit the Gaussian splats as well, which will lead to extensive experiments on real and synthetic datasets, coupled with illustrative visual examples, highlighting the superiority of our approach in terms of representation quality and editing performance. Code of our paper can be found here: https://github.com/kaichen-z/SplatMesh.
♻ ☆ MB-TaylorFormer V2: Improved Multi-branch Linear Transformer Expanded by Taylor Formula for Image Restoration
Recently, Transformer networks have demonstrated outstanding performance in the field of image restoration due to the global receptive field and adaptability to input. However, the quadratic computational complexity of Softmax-attention poses a significant limitation on its extensive application in image restoration tasks, particularly for high-resolution images. To tackle this challenge, we propose a novel variant of the Transformer. This variant leverages the Taylor expansion to approximate the Softmax-attention and utilizes the concept of norm-preserving mapping to approximate the remainder of the first-order Taylor expansion, resulting in a linear computational complexity. Moreover, we introduce a multi-branch architecture featuring multi-scale patch embedding into the proposed Transformer, which has four distinct advantages: 1) various sizes of the receptive field; 2) multi-level semantic information; 3) flexible shapes of the receptive field; 4) accelerated training and inference speed. Hence, the proposed model, named the second version of Taylor formula expansion-based Transformer (for short MB-TaylorFormer V2) has the capability to concurrently process coarse-to-fine features, capture long-distance pixel interactions with limited computational cost, and improve the approximation of the Taylor expansion remainder. Experimental results across diverse image restoration benchmarks demonstrate that MB-TaylorFormer V2 achieves state-of-the-art performance in multiple image restoration tasks, such as image dehazing, deraining, desnowing, motion deblurring, and denoising, with very little computational overhead. The source code is available at https://github.com/FVL2020/MB-TaylorFormerV2.
comment: accepted by IEEE TPAMI
♻ ☆ Learning Free Token Reduction for Multi-Modal Large Language Models
Vision-Language Models (VLMs) have achieved remarkable success across a range of multimodal tasks; however, their practical deployment is often constrained by high computational costs and prolonged inference times. Since the vision modality typically carries more information than the text modality, compressing visual prompts offers a promising solution to alleviate these challenges. Existing approaches predominantly focus on refining model architectures or directly reducing the number of visual tokens. However, these methods often compromise inference performance due to a lack of consideration for the unique spatial and temporal characteristics of visual data. In this work, we propose a token compression paradigm that operates on both spatial and temporal dimensions. Our approach includes a learning-free, plug-and-play compression pipeline that can be seamlessly integrated into most Multimodal Large Language Model (MLLM) frameworks. By leveraging this method, we enhance the model inference capability while simultaneously reducing its computational cost. Experimental results on the Video-QA task demonstrate the effectiveness of the proposed approach, showcasing significant improvements in efficiency without sacrificing performance.
♻ ☆ Visual Chronicles: Using Multimodal LLMs to Analyze Massive Collections of Images
We present a system using Multimodal LLMs (MLLMs) to analyze a large database with tens of millions of images captured at different times, with the aim of discovering patterns in temporal changes. Specifically, we aim to capture frequent co-occurring changes ("trends") across a city over a certain period. Unlike previous visual analyses, our analysis answers open-ended queries (e.g., "what are the frequent types of changes in the city?") without any predetermined target subjects or training labels. These properties cast prior learning-based or unsupervised visual analysis tools unsuitable. We identify MLLMs as a novel tool for their open-ended semantic understanding capabilities. Yet, our datasets are four orders of magnitude too large for an MLLM to ingest as context. So we introduce a bottom-up procedure that decomposes the massive visual analysis problem into more tractable sub-problems. We carefully design MLLM-based solutions to each sub-problem. During experiments and ablation studies with our system, we find it significantly outperforms baselines and is able to discover interesting trends from images captured in large cities (e.g., "addition of outdoor dining,", "overpass was painted blue," etc.). See more results and interactive demos at https://boyangdeng.com/visual-chronicles.
comment: Project page: https://boyangdeng.com/visual-chronicles , second and third listed authors have equal contributions
♻ ☆ MESA: Text-Driven Terrain Generation Using Latent Diffusion and Global Copernicus Data CVPR 2025
Terrain modeling has traditionally relied on procedural techniques, which often require extensive domain expertise and handcrafted rules. In this paper, we present MESA - a novel data-centric alternative by training a diffusion model on global remote sensing data. This approach leverages large-scale geospatial information to generate high-quality terrain samples from text descriptions, showcasing a flexible and scalable solution for terrain generation. The model's capabilities are demonstrated through extensive experiments, highlighting its ability to generate realistic and diverse terrain landscapes. The dataset produced to support this work, the Major TOM Core-DEM extension dataset, is released openly as a comprehensive resource for global terrain data. The results suggest that data-driven models, trained on remote sensing data, can provide a powerful tool for realistic terrain modeling and generation.
comment: Accepted at CVPR 2025 Workshop MORSE
♻ ☆ Intelligent Framework for Human-Robot Collaboration: Dynamic Ergonomics and Adaptive Decision-Making
The integration of collaborative robots into industrial environments has improved productivity, but has also highlighted significant challenges related to operator safety and ergonomics. This paper proposes an innovative framework that integrates advanced visual perception, continuous ergonomic monitoring, and adaptive Behaviour Tree decision-making to overcome the limitations of traditional methods that typically operate as isolated components. Our approach synthesizes deep learning models, advanced tracking algorithms, and dynamic ergonomic assessments into a modular, scalable, and adaptive system. Experimental validation demonstrates the framework's superiority over existing solutions across multiple dimensions: the visual perception module outperformed previous detection models with 72.4% mAP@50:95; the system achieved high accuracy in recognizing operator intentions (92.5%); it promptly classified ergonomic risks with minimal latency (0.57 seconds); and it dynamically managed robotic interventions with exceptionally responsive decision-making capabilities (0.07 seconds), representing a 56% improvement over benchmark systems. This comprehensive solution provides a robust platform for enhancing human-robot collaboration in industrial environments by prioritizing ergonomic safety, operational efficiency, and real-time adaptability.
comment: 15 pagine, 8figure, 3 tabelle, formato conferenza IEEE
♻ ☆ Towards Scenario- and Capability-Driven Dataset Development and Evaluation: An Approach in the Context of Mapless Automated Driving
The foundational role of datasets in defining the capabilities of deep learning models has led to their rapid proliferation. At the same time, published research focusing on the process of dataset development for environment perception in automated driving has been scarce, thereby reducing the applicability of openly available datasets and impeding the development of effective environment perception systems. Sensor-based, mapless automated driving is one of the contexts where this limitation is evident. While leveraging real-time sensor data, instead of pre-defined HD maps promises enhanced adaptability and safety by effectively navigating unexpected environmental changes, it also increases the demands on the scope and complexity of the information provided by the perception system. To address these challenges, we propose a scenario- and capability-based approach to dataset development. Grounded in the principles of ISO 21448 (safety of the intended functionality, SOTIF), extended by ISO/TR 4804, our approach facilitates the structured derivation of dataset requirements. This not only aids in the development of meaningful new datasets but also enables the effective comparison of existing ones. Applying this methodology to a broad range of existing lane detection datasets, we identify significant limitations in current datasets, particularly in terms of real-world applicability, a lack of labeling of critical features, and an absence of comprehensive information for complex driving maneuvers.
♻ ☆ Improving Tropical Cyclone Forecasting With Video Diffusion Models ICLR 2025
Tropical cyclone (TC) forecasting is crucial for disaster preparedness and mitigation. While recent deep learning approaches have shown promise, existing methods often treat TC evolution as a series of independent frame-to-frame predictions, limiting their ability to capture long-term dynamics. We present a novel application of video diffusion models for TC forecasting that explicitly models temporal dependencies through additional temporal layers. Our approach enables the model to generate multiple frames simultaneously, better capturing cyclone evolution patterns. We introduce a two-stage training strategy that significantly improves individual-frame quality and performance in low-data regimes. Experimental results show our method outperforms the previous approach of Nath et al. by 19.3% in MAE, 16.2% in PSNR, and 36.1% in SSIM. Most notably, we extend the reliable forecasting horizon from 36 to 50 hours. Through comprehensive evaluation using both traditional metrics and Fr\'echet Video Distance (FVD), we demonstrate that our approach produces more temporally coherent forecasts while maintaining competitive single-frame quality. Code accessible at https://github.com/Ren-creater/forecast-video-diffmodels.
comment: Accepted for spotlight presentation at the ICLR 2025 workshop on Tackling Climate Change with Machine Learning. 7 pages, 7 figures
♻ ☆ ITACLIP: Boosting Training-Free Semantic Segmentation with Image, Text, and Architectural Enhancements
Recent advances in foundational Vision Language Models (VLMs) have reshaped the evaluation paradigm in computer vision tasks. These foundational models, especially CLIP, have accelerated research in open-vocabulary computer vision tasks, including Open-Vocabulary Semantic Segmentation (OVSS). Although the initial results are promising, the dense prediction capabilities of VLMs still require further improvement. In this study, we enhance the semantic segmentation performance of CLIP by introducing new modules and modifications: 1) architectural changes in the last layer of ViT and the incorporation of attention maps from the middle layers with the last layer, 2) Image Engineering: applying data augmentations to enrich input image representations, and 3) using Large Language Models (LLMs) to generate definitions and synonyms for each class name to leverage CLIP's open-vocabulary capabilities. Our training-free method, ITACLIP, outperforms current state-of-the-art approaches on segmentation benchmarks such as COCO-Stuff, COCO-Object, Pascal Context, and Pascal VOC. Our code is available at https://github.com/m-arda-aydn/ITACLIP.
♻ ☆ GAF: Gaussian Avatar Reconstruction from Monocular Videos via Multi-view Diffusion
We propose a novel approach for reconstructing animatable 3D Gaussian avatars from monocular videos captured by commodity devices like smartphones. Photorealistic 3D head avatar reconstruction from such recordings is challenging due to limited observations, which leaves unobserved regions under-constrained and can lead to artifacts in novel views. To address this problem, we introduce a multi-view head diffusion model, leveraging its priors to fill in missing regions and ensure view consistency in Gaussian splatting renderings. To enable precise viewpoint control, we use normal maps rendered from FLAME-based head reconstruction, which provides pixel-aligned inductive biases. We also condition the diffusion model on VAE features extracted from the input image to preserve facial identity and appearance details. For Gaussian avatar reconstruction, we distill multi-view diffusion priors by using iteratively denoised images as pseudo-ground truths, effectively mitigating over-saturation issues. To further improve photorealism, we apply latent upsampling priors to refine the denoised latent before decoding it into an image. We evaluate our method on the NeRSemble dataset, showing that GAF outperforms previous state-of-the-art methods in novel view synthesis. Furthermore, we demonstrate higher-fidelity avatar reconstructions from monocular videos captured on commodity devices.
comment: Paper Video: https://youtu.be/QuIYTljvhyg Project Page: https://tangjiapeng.github.io/projects/GAF
♻ ☆ PSGait: Gait Recognition using Parsing Skeleton
Gait recognition has emerged as a robust biometric modality due to its non-intrusive nature and resilience to occlusion. Conventional gait recognition methods typically rely on silhouettes or skeletons. Despite their success in gait recognition for controlled laboratory environments, they usually fail in real-world scenarios due to their limited information entropy for gait representations. To achieve accurate gait recognition in the wild, we propose a novel gait representation, named Parsing Skeleton. This representation innovatively introduces the skeleton-guided human parsing method to capture fine-grained body dynamics, so they have much higher information entropy to encode the shapes and dynamics of fine-grained human parts during walking. Moreover, to effectively explore the capability of the Parsing Skeleton representation, we propose a novel Parsing Skeleton-based gait recognition framework, named PSGait, which takes Parsing Skeletons and silhouettes as input. By fusing these two modalities, the resulting image sequences are fed into gait recognition models for enhanced individual differentiation. We conduct comprehensive benchmarks on various datasets to evaluate our model. PSGait outperforms existing state-of-the-art multimodal methods that utilize both skeleton and silhouette inputs while significantly reducing computational resources. Furthermore, as a plug-and-play method, PSGait leads to a maximum improvement of 10.9% in Rank-1 accuracy across various gait recognition models. These results demonstrate that Parsing Skeleton offers a lightweight, effective, and highly generalizable representation for gait recognition in the wild.
♻ ☆ HOMER: Homography-Based Efficient Multi-view 3D Object Removal
3D object removal is an important sub-task in 3D scene editing, with broad applications in scene understanding, augmented reality, and robotics. However, existing methods struggle to achieve a desirable balance among consistency, usability, and computational efficiency in multi-view settings. These limitations are primarily due to unintuitive user interaction in the source view, inefficient multi-view object mask generation, computationally expensive inpainting procedures, and a lack of applicability across different radiance field representations. To address these challenges, we propose a novel pipeline that improves the quality and efficiency of multi-view object mask generation and inpainting. Our method introduces an intuitive region-based interaction mechanism in the source view and eliminates the need for camera poses or extra model training. Our lightweight HoMM module is employed to achieve high-quality multi-view mask propagation with enhanced efficiency. In the inpainting stage, we further reduce computational costs by performing inpainting only on selected key views and propagating the results to other views via homography-based mapping. Our pipeline is compatible with a variety of radiance field frameworks, including NeRF and 3D Gaussian Splatting, demonstrating improved generalizability and practicality in real-world scenarios. Additionally, we present a new 3D multi-object removal dataset with greater object diversity and viewpoint variation than existing datasets. Experiments on public benchmarks and our proposed dataset show that our method achieves state-of-the-art performance while reducing runtime to one-fifth of that required by leading baselines.
♻ ☆ VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model
Recently DeepSeek R1 has shown that reinforcement learning (RL) can substantially improve the reasoning capabilities of Large Language Models (LLMs) through a simple yet effective design. The core of R1 lies in its rule-based reward formulation, which leverages tasks with deterministic ground-truth answers to enable precise and stable reward computation. In the visual domain, we similarly observe that a wide range of visual understanding tasks are inherently equipped with well-defined ground-truth annotations. This property makes them naturally compatible with rule-based reward mechanisms. Motivated by this observation, we investigate the extension of R1-style reinforcement learning to Vision-Language Models (VLMs), aiming to enhance their visual reasoning capabilities. To this end, we develop VLM-R1, a dedicated framework designed to harness RL for improving VLMs' performance on general vision-language tasks. Using this framework, we further explore the feasibility of applying RL to visual domain. Experimental results indicate that the RL-based model not only delivers competitive performance on visual understanding tasks but also surpasses Supervised Fine-Tuning (SFT) in generalization ability. Furthermore, we conduct comprehensive ablation studies that uncover a series of noteworthy insights, including the presence of reward hacking in object detection, the emergence of the "OD aha moment", the impact of training data quality, and the scaling behavior of RL across different model sizes. Through these analyses, we aim to deepen the understanding of how reinforcement learning enhances the capabilities of vision-language models, and we hope our findings and open-source contributions will support continued progress in the vision-language RL community. Our code and model are available at https://github.com/om-ai-lab/VLM-R1
comment: 11 pages, fix some minor typos in the previous version
♻ ☆ Multi-Level Embedding and Alignment Network with Consistency and Invariance Learning for Cross-View Geo-Localization
Cross-View Geo-Localization (CVGL) involves determining the localization of drone images by retrieving the most similar GPS-tagged satellite images. However, the imaging gaps between platforms are often significant and the variations in viewpoints are substantial, which limits the ability of existing methods to effectively associate cross-view features and extract consistent and invariant characteristics. Moreover, existing methods often overlook the problem of increased computational and storage requirements when improving model performance. To handle these limitations, we propose a lightweight enhanced alignment network, called the Multi-Level Embedding and Alignment Network (MEAN). The MEAN network uses a progressive multi-level enhancement strategy, global-to-local associations, and cross-domain alignment, enabling feature communication across levels. This allows MEAN to effectively connect features at different levels and learn robust cross-view consistent mappings and modality-invariant features. Moreover, MEAN adopts a shallow backbone network combined with a lightweight branch design, effectively reducing parameter count and computational complexity. Experimental results on the University-1652 and SUES-200 datasets demonstrate that MEAN reduces parameter count by 62.17% and computational complexity by 70.99% compared to state-of-the-art models, while maintaining competitive or even superior performance. Our code and models will be released on https://github.com/ISChenawei/MEAN.
♻ ☆ Distilling Textual Priors from LLM to Efficient Image Fusion
Multi-modality image fusion aims to synthesize a single, comprehensive image from multiple source inputs. Traditional approaches, such as CNNs and GANs, offer efficiency but struggle to handle low-quality or complex inputs. Recent advances in text-guided methods leverage large model priors to overcome these limitations, but at the cost of significant computational overhead, both in memory and inference time. To address this challenge, we propose a novel framework for distilling large model priors, eliminating the need for text guidance during inference while dramatically reducing model size. Our framework utilizes a teacher-student architecture, where the teacher network incorporates large model priors and transfers this knowledge to a smaller student network via a tailored distillation process. Additionally, we introduce spatial-channel cross-fusion module to enhance the model's ability to leverage textual priors across both spatial and channel dimensions. Our method achieves a favorable trade-off between computational efficiency and fusion quality. The distilled network, requiring only 10% of the parameters and inference time of the teacher network, retains 90% of its performance and outperforms existing SOTA methods. Extensive experiments demonstrate the effectiveness of our approach. The implementation will be made publicly available as an open-source resource.
♻ ☆ Real-time Free-view Human Rendering from Sparse-view RGB Videos using Double Unprojected Textures CVPR 2025
Real-time free-view human rendering from sparse-view RGB inputs is a challenging task due to the sensor scarcity and the tight time budget. To ensure efficiency, recent methods leverage 2D CNNs operating in texture space to learn rendering primitives. However, they either jointly learn geometry and appearance, or completely ignore sparse image information for geometry estimation, significantly harming visual quality and robustness to unseen body poses. To address these issues, we present Double Unprojected Textures, which at the core disentangles coarse geometric deformation estimation from appearance synthesis, enabling robust and photorealistic 4K rendering in real-time. Specifically, we first introduce a novel image-conditioned template deformation network, which estimates the coarse deformation of the human template from a first unprojected texture. This updated geometry is then used to apply a second and more accurate texture unprojection. The resulting texture map has fewer artifacts and better alignment with input views, which benefits our learning of finer-level geometry and appearance represented by Gaussian splats. We validate the effectiveness and efficiency of the proposed method in quantitative and qualitative experiments, which significantly surpasses other state-of-the-art methods. Project page: https://vcai.mpi-inf.mpg.de/projects/DUT/
comment: Accepted at CVPR 2025, Project page: https://vcai.mpi-inf.mpg.de/projects/DUT/
♻ ☆ Iris Style Transfer: Enhancing Iris Recognition with Style Features and Privacy Preservation through Neural Style Transfer
Iris texture is widely regarded as a gold standard biometric modality for authentication and identification. The demand for robust iris recognition methods, coupled with growing security and privacy concerns regarding iris attacks, has escalated recently. Inspired by neural style transfer, an advanced technique that leverages neural networks to separate content and style features, we hypothesize that iris texture's style features provide a reliable foundation for recognition and are more resilient to variations like rotation and perspective shifts than traditional approaches. Our experimental results support this hypothesis, showing a significantly higher classification accuracy compared to conventional features. Further, we propose using neural style transfer to obfuscate the identifiable iris style features, ensuring the protection of sensitive biometric information while maintaining the utility of eye images for tasks like eye segmentation and gaze estimation. This work opens new avenues for iris-oriented, secure, and privacy-aware biometric systems.
comment: The 2025 ACM Symposium on Eye Tracking Research & Applications (ETRA)
♻ ☆ RHanDS: Refining Malformed Hands for Generated Images with Decoupled Structure and Style Guidance
Although diffusion models can generate high-quality human images, their applications are limited by the instability in generating hands with correct structures. In this paper, we introduce RHanDS, a conditional diffusion-based framework designed to refine malformed hands by utilizing decoupled structure and style guidance. The hand mesh reconstructed from the malformed hand offers structure guidance for correcting the structure of the hand, while the malformed hand itself provides style guidance for preserving the style of the hand. To alleviate the mutual interference between style and structure guidance, we introduce a two-stage training strategy and build a series of multi-style hand datasets. In the first stage, we use paired hand images for training to ensure stylistic consistency in hand refining. In the second stage, various hand images generated based on human meshes are used for training, enabling the model to gain control over the hand structure. Experimental results demonstrate that RHanDS can effectively refine hand structure while preserving consistency in hand style.
♻ ☆ ResiDual Transformer Alignment with Spectral Decomposition
When examined through the lens of their residual streams, a puzzling property emerges in transformer networks: residual contributions (e.g., attention heads) sometimes specialize in specific tasks or input attributes. In this paper, we analyze this phenomenon in vision transformers, focusing on the spectral geometry of residuals, and explore its implications for modality alignment in vision-language models. First, we link it to the intrinsically low-dimensional structure of visual head representations, zooming into their principal components and showing that they encode specialized roles across a wide variety of input data distributions. Then, we analyze the effect of head specialization in multimodal models, focusing on how improved alignment between text and specialized heads impacts zero-shot classification performance. This specialization-performance link consistently holds across diverse pre-training data, network sizes, and objectives, demonstrating a powerful new mechanism for boosting zero-shot classification through targeted alignment. Ultimately, we translate these insights into actionable terms by introducing ResiDual, a technique for spectral alignment of the residual stream. Much like panning for gold, it lets the noise from irrelevant unit principal components (i.e., attributes) wash away to amplify task-relevant ones. Remarkably, this dual perspective on modality alignment yields fine-tuning level performance on different data distributions while modelling an extremely interpretable and parameter-efficient transformation, as we extensively show on 70 pre-trained network-dataset combinations (7 models, 10 datasets).
comment: Published in Transactions on Machine Learning Research (TMLR)
♻ ☆ Improving Decoupled Posterior Sampling for Inverse Problems using Data Consistency Constraint
Diffusion models have shown strong performances in solving inverse problems through posterior sampling while they suffer from errors during earlier steps. To mitigate this issue, several Decoupled Posterior Sampling methods have been recently proposed. However, the reverse process in these methods ignores measurement information, leading to errors that impede effective optimization in subsequent steps. To solve this problem, we propose Guided Decoupled Posterior Sampling (GDPS) by integrating a data consistency constraint in the reverse process. The constraint performs a smoother transition within the optimization process, facilitating a more effective convergence toward the target distribution. Furthermore, we extend our method to latent diffusion models and Tweedie's formula, demonstrating its scalability. We evaluate GDPS on the FFHQ and ImageNet datasets across various linear and nonlinear tasks under both standard and challenging conditions. Experimental results demonstrate that GDPS achieves state-of-the-art performance, improving accuracy over existing methods.
♻ ☆ DUKAE: DUal-level Knowledge Accumulation and Ensemble for Pre-Trained Model-Based Continual Learning
Pre-trained model-based continual learning (PTMCL) has garnered growing attention, as it enables more rapid acquisition of new knowledge by leveraging the extensive foundational understanding inherent in pre-trained model (PTM). Most existing PTMCL methods use Parameter-Efficient Fine-Tuning (PEFT) to learn new knowledge while consolidating existing memory. However, they often face some challenges. A major challenge lies in the misalignment of classification heads, as the classification head of each task is trained within a distinct feature space, leading to inconsistent decision boundaries across tasks and, consequently, increased forgetting. Another critical limitation stems from the restricted feature-level knowledge accumulation, with feature learning typically restricted to the initial task only, which constrains the model's representation capabilities. To address these issues, we propose a method named DUal-level Knowledge Accumulation and Ensemble (DUKAE) that leverages both feature-level and decision-level knowledge accumulation by aligning classification heads into a unified feature space through Gaussian distribution sampling and introducing an adaptive expertise ensemble to fuse knowledge across feature subspaces. Extensive experiments on CIFAR-100, ImageNet-R, CUB-200, and Cars-196 datasets demonstrate the superior performance of our approach.
♻ ☆ Frequency Is What You Need: Word-frequency Masking Benefits Vision-Language Model Pre-training
Vision Language Models (VLMs) can be trained more efficiently if training sets can be reduced in size. Recent work has shown the benefits of masking text during VLM training using a variety of approaches: truncation, random masking, block masking and syntax masking. In this paper, we show that the best masking strategy changes over training epochs and that, given sufficient training epochs. We analyze existing text masking approaches including syntax masking, which is currently the state of the art, and identify the word frequency distribution as important in determining their success. Experiments on a large range of data sets demonstrate that syntax masking is outperformed by other approaches, given sufficient epochs, and that our proposed frequency-based approach, called Contrastive Language-Image Pre-training with Word Frequency Masking (CLIPF) has numerous advantages. The benefits are particularly evident as the number of input tokens decreases.
♻ ☆ Climplicit: Climatic Implicit Embeddings for Global Ecological Tasks ICLR 2025
Deep learning on climatic data holds potential for macroecological applications. However, its adoption remains limited among scientists outside the deep learning community due to storage, compute, and technical expertise barriers. To address this, we introduce Climplicit, a spatio-temporal geolocation encoder pretrained to generate implicit climatic representations anywhere on Earth. By bypassing the need to download raw climatic rasters and train feature extractors, our model uses x3500 less disk space and significantly reduces computational needs for downstream tasks. We evaluate our Climplicit embeddings on biomes classification, species distribution modeling, and plant trait regression. We find that single-layer probing our Climplicit embeddings consistently performs better or on par with training a model from scratch on downstream tasks and overall better than alternative geolocation encoding models.
comment: Published as a workshop paper at "Tackling Climate Change with Machine Learning", ICLR 2025
♻ ☆ TAPNext: Tracking Any Point (TAP) as Next Token Prediction
Tracking Any Point (TAP) in a video is a challenging computer vision problem with many demonstrated applications in robotics, video editing, and 3D reconstruction. Existing methods for TAP rely heavily on complex tracking-specific inductive biases and heuristics, limiting their generality and potential for scaling. To address these challenges, we present TAPNext, a new approach that casts TAP as sequential masked token decoding. Our model is causal, tracks in a purely online fashion, and removes tracking-specific inductive biases. This enables TAPNext to run with minimal latency, and removes the temporal windowing required by many existing state of art trackers. Despite its simplicity, TAPNext achieves a new state-of-the-art tracking performance among both online and offline trackers. Finally, we present evidence that many widely used tracking heuristics emerge naturally in TAPNext through end-to-end training. The TAPNext model and code can be found at https://tap-next.github.io/.
♻ ☆ PhD: A ChatGPT-Prompted Visual hallucination Evaluation Dataset CVPR 2025
Multimodal Large Language Models (MLLMs) hallucinate, resulting in an emerging topic of visual hallucination evaluation (VHE). This paper contributes a ChatGPT-Prompted visual hallucination evaluation Dataset (PhD) for objective VHE at a large scale. The essence of VHE is to ask an MLLM questions about specific images to assess its susceptibility to hallucination. Depending on what to ask (objects, attributes, sentiment, etc.) and how the questions are asked, we structure PhD along two dimensions, i.e. task and mode. Five visual recognition tasks, ranging from low-level (object / attribute recognition) to middle-level (sentiment / position recognition and counting), are considered. Besides a normal visual QA mode, which we term PhD-base, PhD also asks questions with specious context (PhD-sec) or with incorrect context ({PhD-icc), or with AI-generated counter common sense images (PhD-ccs). We construct PhD by a ChatGPT-assisted semi-automated pipeline, encompassing four pivotal modules: task-specific hallucinatory item (hitem) selection, hitem-embedded question generation, specious / incorrect context generation, and counter-common-sense (CCS) image generation. With over 14k daily images, 750 CCS images and 102k VQA triplets in total, PhD reveals considerable variability in MLLMs' performance across various modes and tasks, offering valuable insights into the nature of hallucination. As such, PhD stands as a potent tool not only for VHE but may also play a significant role in the refinement of MLLMs.
comment: Accepted by CVPR 2025, Highlight
♻ ☆ Joint semi-supervised and contrastive learning enables domain generalization and multi-domain segmentation
Despite their effectiveness, current deep learning models face challenges with images coming from different domains with varying appearance and content. We introduce SegCLR, a versatile framework designed to segment images across different domains, employing supervised and contrastive learning simultaneously to effectively learn from both labeled and unlabeled data. We demonstrate the superior performance of SegCLR through a comprehensive evaluation involving three diverse clinical datasets of 3D retinal Optical Coherence Tomography (OCT) images, for the slice-wise segmentation of fluids with various network configurations and verification across 10 different network initializations. In an unsupervised domain adaptation context, SegCLR achieves results on par with a supervised upper-bound model trained on the intended target domain. Notably, we discover that the segmentation performance of SegCLR framework is marginally impacted by the abundance of unlabeled data from the target domain, thereby we also propose an effective domain generalization extension of SegCLR, known also as zero-shot domain adaptation, which eliminates the need for any target domain information. This shows that our proposed addition of contrastive loss in standard supervised training for segmentation leads to superior models, inherently more generalizable to both in- and out-of-domain test data. We additionally propose a pragmatic solution for SegCLR deployment in realistic scenarios with multiple domains containing labeled data. Accordingly, our framework pushes the boundaries of deep-learning based segmentation in multi-domain applications, regardless of data availability - labeled, unlabeled, or nonexistent.
♻ ☆ Building Vision Models upon Heat Conduction
Visual representation models leveraging attention mechanisms are challenged by significant computational overhead, particularly when pursuing large receptive fields. In this study, we aim to mitigate this challenge by introducing the Heat Conduction Operator (HCO) built upon the physical heat conduction principle. HCO conceptualizes image patches as heat sources and models their correlations through adaptive thermal energy diffusion, enabling robust visual representations. HCO enjoys a computational complexity of O(N^1.5), as it can be implemented using discrete cosine transformation (DCT) operations. HCO is plug-and-play, combining with deep learning backbones produces visual representation models (termed vHeat) with global receptive fields. Experiments across vision tasks demonstrate that, beyond the stronger performance, vHeat achieves up to a 3x throughput, 80% less GPU memory allocation, and 35% fewer computational FLOPs compared to the Swin-Transformer. Code is available at https://github.com/MzeroMiko/vHeat.
♻ ☆ DartControl: A Diffusion-Based Autoregressive Motion Model for Real-Time Text-Driven Motion Control ICLR
Text-conditioned human motion generation, which allows for user interaction through natural language, has become increasingly popular. Existing methods typically generate short, isolated motions based on a single input sentence. However, human motions are continuous and can extend over long periods, carrying rich semantics. Creating long, complex motions that precisely respond to streams of text descriptions, particularly in an online and real-time setting, remains a significant challenge. Furthermore, incorporating spatial constraints into text-conditioned motion generation presents additional challenges, as it requires aligning the motion semantics specified by text descriptions with geometric information, such as goal locations and 3D scene geometry. To address these limitations, we propose DartControl, in short DART, a Diffusion-based Autoregressive motion primitive model for Real-time Text-driven motion control. Our model effectively learns a compact motion primitive space jointly conditioned on motion history and text inputs using latent diffusion models. By autoregressively generating motion primitives based on the preceding history and current text input, DART enables real-time, sequential motion generation driven by natural language descriptions. Additionally, the learned motion primitive space allows for precise spatial motion control, which we formulate either as a latent noise optimization problem or as a Markov decision process addressed through reinforcement learning. We present effective algorithms for both approaches, demonstrating our model's versatility and superior performance in various motion synthesis tasks. Experiments show our method outperforms existing baselines in motion realism, efficiency, and controllability. Video results are available on the project page: https://zkf1997.github.io/DART/.
comment: Updated ICLR camera ready version
♻ ☆ Tumor likelihood estimation on MRI prostate data by utilizing k-Space information
We present a novel preprocessing and prediction pipeline for the classification of magnetic resonance imaging (MRI) that takes advantage of the information rich complex valued k-Space. Using a publicly available MRI raw dataset with 312 subject and a total of 9508 slices, we show the advantage of utilizing the k-Space for better prostate cancer likelihood estimation in comparison to just using the magnitudinal information in the image domain, with an AUROC of $86.1\%\pm1.8\%$. Additionally, by using high undersampling rates and a simple principal component analysis (PCA) for coil compression, we reduce the time needed for reconstruction by avoiding the time intensive GRAPPA reconstruction algorithm. By using digital undersampling for our experiments, we show that scanning and reconstruction time could be reduced. Even with an undersampling factor of 16, our approach achieves meaningful results, with an AUROC of $71.4\%\pm2.9\%$, using the PCA coil combination and taking into account the k-Space information. With this study, we were able to show the feasibility of preserving phase and k-Space information, with consistent results. Besides preserving valuable information for further diagnostics, this approach can work without the time intensive ADC and reconstruction calculations, greatly reducing the post processing, as well as potential scanning time, increasing patient comfort and allowing a close to real-time prediction.
♻ ☆ SCANet: Scene Complexity Aware Network for Weakly-Supervised Video Moment Retrieval ICCV 2023
Video moment retrieval aims to localize moments in video corresponding to a given language query. To avoid the expensive cost of annotating the temporal moments, weakly-supervised VMR (wsVMR) systems have been studied. For such systems, generating a number of proposals as moment candidates and then selecting the most appropriate proposal has been a popular approach. These proposals are assumed to contain many distinguishable scenes in a video as candidates. However, existing proposals of wsVMR systems do not respect the varying numbers of scenes in each video, where the proposals are heuristically determined irrespective of the video. We argue that the retrieval system should be able to counter the complexities caused by varying numbers of scenes in each video. To this end, we present a novel concept of a retrieval system referred to as Scene Complexity Aware Network (SCANet), which measures the `scene complexity' of multiple scenes in each video and generates adaptive proposals responding to variable complexities of scenes in each video. Experimental results on three retrieval benchmarks (i.e., Charades-STA, ActivityNet, TVR) achieve state-of-the-art performances and demonstrate the effectiveness of incorporating the scene complexity.
comment: 11 pages, Accepted in ICCV 2023
♻ ☆ SnatchML: Hijacking ML models without Training Access
Model hijacking can cause significant accountability and security risks since the owner of a hijacked model can be framed for having their model offer illegal or unethical services. Prior works consider model hijacking as a training time attack, whereby an adversary requires full access to the ML model training. In this paper, we consider a stronger threat model for an inference-time hijacking attack, where the adversary has no access to the training phase of the victim model. Our intuition is that ML models, which are typically over-parameterized, might have the capacity to (unintentionally) learn more than the intended task they are trained for. We propose SnatchML, a new training-free model hijacking attack, that leverages the extra capacity learnt by the victim model to infer different tasks that can be semantically related or unrelated to the original one. Our results on models deployed on AWS Sagemaker showed that SnatchML can deliver high accuracy on hijacking tasks. Interestingly, while all previous approaches are limited by the number of classes in the benign task, SnatchML can hijack models for tasks that contain more classes than the original. We explore different methods to mitigate this risk; We propose meta-unlearning, which is designed to help the model unlearn a potentially malicious task while training for the original task. We also provide insights on over-parametrization as a possible inherent factor that facilitates model hijacking, and accordingly, we propose a compression-based countermeasure to counteract this attack. We believe this work offers a previously overlooked perspective on model hijacking attacks, presenting a stronger threat model and higher applicability in real-world contexts.
comment: 17 pages, 14 figures, 7 tables
♻ ☆ Towards A Generalizable Pathology Foundation Model via Unified Knowledge Distillation
Foundation models pretrained on large-scale datasets are revolutionizing the field of computational pathology (CPath). The generalization ability of foundation models is crucial for the success in various downstream clinical tasks. However, current foundation models have only been evaluated on a limited type and number of tasks, leaving their generalization ability and overall performance unclear. To address this gap, we established a most comprehensive benchmark to evaluate the performance of off-the-shelf foundation models across six distinct clinical task types, encompassing a total of 72 specific tasks, including slide-level classification, survival prediction, ROI-tissue classification, ROI retrieval, visual question answering, and report generation. Our findings reveal that existing foundation models excel at certain task types but struggle to effectively handle the full breadth of clinical tasks. To improve the generalization of pathology foundation models, we propose a unified knowledge distillation framework consisting of both expert and self-knowledge distillation, where the former allows the model to learn from the knowledge of multiple expert models, while the latter leverages self-distillation to enable image representation learning via local-global alignment. Based on this framework, we curated a dataset of 96,000 whole slide images (WSIs) and developed a Generalizable Pathology Foundation Model (GPFM). This advanced model was trained on a substantial dataset comprising 190 million images extracted from approximately 72,000 publicly available slides, encompassing 34 major tissue types. Evaluated on the established benchmark, GPFM achieves an impressive average rank of 1.6, with 42 tasks ranked 1st, while the second-best model, UNI, attains an average rank of 3.7, with only 6 tasks ranked 1st.
comment: update
♻ ☆ Reconstruct Anything Model: a lightweight foundation model for computational imaging
Most existing learning-based methods for solving imaging inverse problems can be roughly divided into two classes: iterative algorithms, such as plug-and-play and diffusion methods, that leverage pretrained denoisers, and unrolled architectures that are trained end-to-end for specific imaging problems. Iterative methods in the first class are computationally costly and often provide suboptimal reconstruction performance, whereas unrolled architectures are generally specific to a single inverse problem and require expensive training. In this work, we propose a novel non-iterative, lightweight architecture that incorporates knowledge about the forward operator (acquisition physics and noise parameters) without relying on unrolling. Our model is trained to solve a wide range of inverse problems beyond denoising, including deblurring, magnetic resonance imaging, computed tomography, inpainting, and super-resolution. The proposed model can be easily adapted to unseen inverse problems or datasets with a few fine-tuning steps (up to a few images) in a self-supervised way, without ground-truth references. Throughout a series of experiments, we demonstrate state-of-the-art performance from medical imaging to low-photon imaging and microscopy.
♻ ☆ Faster Vision Mamba is Rebuilt in Minutes via Merged Token Re-training
Vision Mamba has shown close to state of the art performance on computer vision tasks, drawing much interest in increasing it's efficiency. A promising approach is token reduction (that has been successfully implemented in ViTs). Pruning informative tokens in Mamba leads to a high loss of key knowledge and degraded performance. An alternative, of merging tokens preserves more information than pruning, also suffers for large compression ratios. Our key insight is that a quick round of retraining after token merging yeilds robust results across various compression ratios. Empirically, pruned Vims only drop up to 0.9% accuracy on ImageNet-1K, recovered by our proposed framework R-MeeTo in our main evaluation. We show how simple and effective the fast recovery can be achieved at minute-level, in particular, a 35.9% accuracy spike over 3 epochs of training on Vim-Ti. Moreover, Vim-Ti/S/B are re-trained within 5/7/17 minutes, and Vim-S only drops 1.3% with 1.2x (up to 1.5x) speed up in inference.
♻ ☆ TPC: Test-time Procrustes Calibration for Diffusion-based Human Image Animation NeurIPS 2024
Human image animation aims to generate a human motion video from the inputs of a reference human image and a target motion video. Current diffusion-based image animation systems exhibit high precision in transferring human identity into targeted motion, yet they still exhibit irregular quality in their outputs. Their optimal precision is achieved only when the physical compositions (i.e., scale and rotation) of the human shapes in the reference image and target pose frame are aligned. In the absence of such alignment, there is a noticeable decline in fidelity and consistency. Especially, in real-world environments, this compositional misalignment commonly occurs, posing significant challenges to the practical usage of current systems. To this end, we propose Test-time Procrustes Calibration (TPC), which enhances the robustness of diffusion-based image animation systems by maintaining optimal performance even when faced with compositional misalignment, effectively addressing real-world scenarios. The TPC provides a calibrated reference image for the diffusion model, enhancing its capability to understand the correspondence between human shapes in the reference and target images. Our method is simple and can be applied to any diffusion-based image animation system in a model-agnostic manner, improving the effectiveness at test time without additional training.
comment: 24 pages, 16 figures, NeurIPS 2024
♻ ☆ UniForm: A Unified Multi-Task Diffusion Transformer for Audio-Video Generation
With the rise of diffusion models, audio-video generation has been revolutionized. However, most existing methods rely on separate modules for each modality, with limited exploration of unified generative architectures. In addition, many are confined to a single task and small-scale datasets. To address these limitations, we first propose UniForm, a unified multi-task diffusion transformer that jointly generates audio and visual modalities in a shared latent space. A single diffusion process models both audio and video, capturing the inherent correlations between sound and vision. Second, we introduce task-specific noise schemes and task tokens, enabling a single model to support multiple tasks, including text-to-audio-video, audio-to-video, and video-to-audio generation. Furthermore, by leveraging large language models and a large-scale text-audio-video combined dataset, UniForm achieves greater generative diversity than prior approaches. Extensive experiments show that UniForm achieves the state-of-the-art performance across audio-video generation tasks, producing content that is both well-aligned and close to real-world data distributions. Our demos are available at https://uniform-t2av.github.io/.
comment: Our demos are available at https://uniform-t2av.github.io/
♻ ☆ An Embedding is Worth a Thousand Noisy Labels
The performance of deep neural networks scales with dataset size and label quality, rendering the efficient mitigation of low-quality data annotations crucial for building robust and cost-effective systems. Existing strategies to address label noise exhibit severe limitations due to computational complexity and application dependency. In this work, we propose WANN, a Weighted Adaptive Nearest Neighbor approach that builds on self-supervised feature representations obtained from foundation models. To guide the weighted voting scheme, we introduce a reliability score $\eta$, which measures the likelihood of a data label being correct. WANN outperforms reference methods, including a linear layer trained with robust loss functions, on diverse datasets of varying size and under various noise types and severities. WANN also exhibits superior generalization on imbalanced data compared to both Adaptive-NNs (ANN) and fixed k-NNs. Furthermore, the proposed weighting scheme enhances supervised dimensionality reduction under noisy labels. This yields a significant boost in classification performance with 10x and 100x smaller image embeddings, minimizing latency and storage requirements. Our approach, emphasizing efficiency and explainability, emerges as a simple, robust solution to overcome inherent limitations of deep neural network training. The code is available at https://github.com/francescodisalvo05/wann-noisy-labels .
comment: Accepted to Transactions on Machine Learning Research (TMLR)
♻ ☆ Machine Unlearning in Hyperbolic vs. Euclidean Multimodal Contrastive Learning: Adapting Alignment Calibration to MERU
Machine unlearning methods have become increasingly important for selective concept removal in large pre-trained models. While recent work has explored unlearning in Euclidean contrastive vision-language models, the effectiveness of concept removal in hyperbolic spaces remains unexplored. This paper investigates machine unlearning in hyperbolic contrastive learning by adapting Alignment Calibration to MERU, a model that embeds images and text in hyperbolic space to better capture semantic hierarchies. Through systematic experiments and ablation studies, we demonstrate that hyperbolic geometry offers distinct advantages for concept removal, achieving near perfect forgetting with reasonable performance on retained concepts, particularly when scaling to multiple concept removal. Our approach introduces hyperbolic-specific components including entailment calibration and norm regularization that leverage the unique properties of hyperbolic space. Comparative analysis with Euclidean models reveals fundamental differences in unlearning dynamics, with hyperbolic unlearning reorganizing the semantic hierarchy while Euclidean approaches merely disconnect cross-modal associations. These findings not only advance machine unlearning techniques but also provide insights into the geometric properties that influence concept representation and removal in multimodal models. Source code available at https://github.com/alex-pv01/HAC
comment: Preprint
♻ ☆ CLIP-SR: Collaborative Linguistic and Image Processing for Super-Resolution
Convolutional Neural Networks (CNNs) have significantly advanced Image Super-Resolution (SR), yet most CNN-based methods rely solely on pixel-based transformations, often leading to artifacts and blurring, particularly under severe downsampling rates (\eg, 8$\times$ or 16$\times$). The recently developed text-guided SR approaches leverage textual descriptions to enhance their detail restoration capabilities but frequently struggle with effectively performing alignment, resulting in semantic inconsistencies. To address these challenges, we propose a multi-modal semantic enhancement framework that integrates textual semantics with visual features, effectively mitigating semantic mismatches and detail losses in highly degraded low-resolution (LR) images. Our method enables realistic, high-quality SR to be performed at large upscaling factors, with a maximum scaling ratio of 16$\times$. The framework integrates both text and image inputs using the prompt predictor, the Text-Image Fusion Block (TIFBlock), and the Iterative Refinement Module, leveraging Contrastive Language-Image Pretraining (CLIP) features to guide a progressive enhancement process with fine-grained alignment. This synergy produces high-resolution outputs with sharp textures and strong semantic coherence, even at substantial scaling factors. Extensive comparative experiments and ablation studies validate the effectiveness of our approach. Furthermore, by leveraging textual semantics, our method offers a degree of super-resolution editability, allowing for controlled enhancements while preserving semantic consistency. The code is available at https://github.com/hengliusky/CLIP-SR.
comment: 12 pages, 10 figures
♻ ☆ AMBER -- Advanced SegFormer for Multi-Band Image Segmentation: an application to Hyperspectral Imaging
Deep learning has revolutionized the field of hyperspectral image (HSI) analysis, enabling the extraction of complex spectral and spatial features. While convolutional neural networks (CNNs) have been the backbone of HSI classification, their limitations in capturing global contextual features have led to the exploration of Vision Transformers (ViTs). This paper introduces AMBER, an advanced SegFormer specifically designed for multi-band image segmentation. AMBER enhances the original SegFormer by incorporating three-dimensional convolutions, custom kernel sizes, and a Funnelizer layer. This architecture enables processing hyperspectral data directly, without requiring spectral dimensionality reduction during preprocessing. Our experiments, conducted on three benchmark datasets (Salinas, Indian Pines, and Pavia University) and on a dataset from the PRISMA satellite, show that AMBER outperforms traditional CNN-based methods in terms of Overall Accuracy, Kappa coefficient, and Average Accuracy on the first three datasets, and achieves state-of-the-art performance on the PRISMA dataset. These findings highlight AMBER's robustness, adaptability to both airborne and spaceborne data, and its potential as a powerful solution for remote sensing and other domains requiring advanced analysis of high-dimensional data.
comment: submitted to Neural Computing & Applications (Springer). Accepted with minor revisions
♻ ☆ MoLA: Motion Generation and Editing with Latent Diffusion Enhanced by Adversarial Training CVPR 2025
In text-to-motion generation, controllability as well as generation quality and speed has become increasingly critical. The controllability challenges include generating a motion of a length that matches the given textual description and editing the generated motions according to control signals, such as the start-end positions and the pelvis trajectory. In this paper, we propose MoLA, which provides fast, high-quality, variable-length motion generation and can also deal with multiple editing tasks in a single framework. Our approach revisits the motion representation used as inputs and outputs in the model, incorporating an activation variable to enable variable-length motion generation. Additionally, we integrate a variational autoencoder and a latent diffusion model, further enhanced through adversarial training, to achieve high-quality and fast generation. Moreover, we apply a training-free guided generation framework to achieve various editing tasks with motion control inputs. We quantitatively show the effectiveness of adversarial learning in text-to-motion generation, and demonstrate the applicability of our editing framework to multiple editing tasks in the motion domain.
comment: CVPR 2025 HuMoGen Workshop
♻ ☆ V-LASIK: Consistent Glasses-Removal from Videos Using Synthetic Data
Diffusion-based generative models have recently shown remarkable image and video editing capabilities. However, local video editing, particularly removal of small attributes like glasses, remains a challenge. Existing methods either alter the videos excessively, generate unrealistic artifacts, or fail to perform the requested edit consistently throughout the video. In this work, we focus on consistent and identity-preserving removal of glasses in videos, using it as a case study for consistent local attribute removal in videos. Due to the lack of paired data, we adopt a weakly supervised approach and generate synthetic imperfect data, using an adjusted pretrained diffusion model. We show that despite data imperfection, by learning from our generated data and leveraging the prior of pretrained diffusion models, our model is able to perform the desired edit consistently while preserving the original video content. Furthermore, we exemplify the generalization ability of our method to other local video editing tasks by applying it successfully to facial sticker-removal. Our approach demonstrates significant improvement over existing methods, showcasing the potential of leveraging synthetic data and strong video priors for local video editing tasks.
♻ ☆ Hands-On: Segmenting Individual Signs from Continuous Sequences
This work tackles the challenge of continuous sign language segmentation, a key task with huge implications for sign language translation and data annotation. We propose a transformer-based architecture that models the temporal dynamics of signing and frames segmentation as a sequence labeling problem using the Begin-In-Out (BIO) tagging scheme. Our method leverages the HaMeR hand features, and is complemented with 3D Angles. Extensive experiments show that our model achieves state-of-the-art results on the DGS Corpus, while our features surpass prior benchmarks on BSLCorpus.
comment: Accepted in the 19th IEEE International Conference on Automatic Face and Gesture Recognition
♻ ☆ MimiQ: Low-Bit Data-Free Quantization of Vision Transformers with Encouraging Inter-Head Attention Similarity AAAI 2025
Data-free quantization (DFQ) is a technique that creates a lightweight network from its full-precision counterpart without the original training data, often through a synthetic dataset. Although several DFQ methods have been proposed for vision transformer (ViT) architectures, they fail to achieve efficacy in low-bit settings. Examining the existing methods, we observe that their synthetic data produce misaligned attention maps, while those of the real samples are highly aligned. From this observation, we find that aligning attention maps of synthetic data helps improve the overall performance of quantized ViTs. Motivated by this finding, we devise MimiQ, a novel DFQ method designed for ViTs that enhances inter-head attention similarity. First, we generate synthetic data by aligning head-wise attention outputs from each spatial query patch. Then, we align the attention maps of the quantized network to those of the full-precision teacher by applying head-wise structural attention distillation. The experimental results show that the proposed method significantly outperforms baselines, setting a new state-of-the-art for ViT-DFQ. This paper is an extended version of our work published in the proceedings of AAAI 2025, including additional supplementary material.
comment: Published to AAAI 2025
♻ ☆ Leveraging Anthropometric Measurements to Improve Human Mesh Estimation and Ensure Consistent Body Shapes CVPR'25
The basic body shape (i.e., the body shape in T-pose) of a person does not change within a single video. However, most SOTA human mesh estimation (HME) models output a slightly different, thus inconsistent basic body shape for each video frame. Furthermore, we find that SOTA 3D human pose estimation (HPE) models outperform HME models regarding the precision of the estimated 3D keypoint positions. We solve the problem of inconsistent body shapes by leveraging anthropometric measurements like taken by tailors from humans. We create a model called A2B that converts given anthropometric measurements to basic body shape parameters of human mesh models. We obtain superior and consistent human meshes by combining the A2B model results with the keypoints of 3D HPE models using inverse kinematics. We evaluate our approach on challenging datasets like ASPset or fit3D, where we can lower the MPJPE by over 30 mm compared to SOTA HME models. Further, replacing estimates of the body shape parameters from existing HME models with A2B results not only increases the performance of these HME models, but also guarantees consistent body shapes.
comment: accepted for CVSports@CVPR'25
♻ ☆ PatchContrast: Self-Supervised Pre-training for 3D Object Detection CVPR
Accurately detecting objects in the environment is a key challenge for autonomous vehicles. However, obtaining annotated data for detection is expensive and time-consuming. We introduce PatchContrast, a novel self-supervised point cloud pre-training framework for 3D object detection. We propose to utilize two levels of abstraction to learn discriminative representation from unlabeled data: proposal-level and patch-level. The proposal-level aims at localizing objects in relation to their surroundings, whereas the patch-level adds information about the internal connections between the object's components, hence distinguishing between different objects based on their individual components. We demonstrate how these levels can be integrated into self-supervised pre-training for various backbones to enhance the downstream 3D detection task. We show that our method outperforms existing state-of-the-art models on three commonly-used 3D detection datasets.
comment: CVPRW 2025
♻ ☆ Ham2Pose: Animating Sign Language Notation into Pose Sequences
Translating spoken languages into Sign languages is necessary for open communication between the hearing and hearing-impaired communities. To achieve this goal, we propose the first method for animating a text written in HamNoSys, a lexical Sign language notation, into signed pose sequences. As HamNoSys is universal, our proposed method offers a generic solution invariant to the target Sign language. Our method gradually generates pose predictions using transformer encoders that create meaningful representations of the text and poses while considering their spatial and temporal information. We use weak supervision for the training process and show that our method succeeds in learning from partial and inaccurate data. Additionally, we offer a new distance measurement for pose sequences, normalized Dynamic Time Warping (nDTW), based on DTW over normalized keypoints trajectories, and validate its correctness using AUTSL, a large-scale Sign language dataset. We show that it measures the distance between pose sequences more accurately than existing measurements and use it to assess the quality of our generated pose sequences. Code for the data pre-processing, the model, and the distance measurement is publicly released for future research.
♻ ☆ Tailoring Adversarial Attacks on Deep Neural Networks for Targeted Class Manipulation Using DeepFool Algorithm
The susceptibility of deep neural networks (DNNs) to adversarial attacks undermines their reliability across numerous applications, underscoring the necessity for an in-depth exploration of these vulnerabilities and the formulation of robust defense strategies. The DeepFool algorithm by Moosavi-Dezfooli et al. (2016) represents a pivotal step in identifying minimal perturbations required to induce misclassification of input images. Nonetheless, its generic methodology falls short in scenarios necessitating targeted interventions. Additionally, previous research studies have predominantly concentrated on the success rate of attacks without adequately addressing the consequential distortion of images, the maintenance of image quality, or the confidence threshold required for misclassification. To bridge these gaps, we introduce the Enhanced Targeted DeepFool (ET DeepFool) algorithm, an evolution of DeepFool that not only facilitates the specification of desired misclassification targets but also incorporates a configurable minimum confidence score. Our empirical investigations demonstrate the superiority of this refined approach in maintaining the integrity of images and minimizing perturbations across a variety of DNN architectures. Unlike previous iterations, such as the Targeted DeepFool by Gajjar et al. (2022), our method grants unparalleled control over the perturbation process, enabling precise manipulation of model responses. Preliminary outcomes reveal that certain models, including AlexNet and the advanced Vision Transformer, display commendable robustness to such manipulations. This discovery of varying levels of model robustness, as unveiled through our confidence level adjustments, could have far-reaching implications for the field of image recognition. Our code is available at https://github.com/FazleLabib/et_deepfool.
comment: 18 pages, 5 figures. Accepted by Nature Scientific Reports
♻ ☆ MIO : Mutual Information Optimization using Self-Supervised Binary Contrastive Learning
Self-supervised contrastive learning frameworks have progressed rapidly over the last few years. In this paper, we propose a novel loss function for contrastive learning. We model our pre-training task as a binary classification problem to induce an implicit contrastive effect. We further improve the n\"aive loss function after removing the effect of the positive-positive repulsion and incorporating the upper bound of the negative pair repulsion. Unlike existing methods, the proposed loss function optimizes the mutual information in positive and negative pairs. We also present a closed-form expression for the parameter gradient flow and compare the behaviour of self-supervised contrastive frameworks using Hessian eigenspectrum to analytically study their convergence. The proposed method outperforms SOTA self-supervised contrastive frameworks on benchmark datasets such as CIFAR-10, CIFAR-100, STL-10, and Tiny-ImageNet. After 200 pretraining epochs with ResNet-18 as the backbone, the proposed model achieves an accuracy of 86.36%, 58.18%, 80.50%, and 30.87% on the CIFAR-10, CIFAR-100, STL-10, and Tiny-ImageNet datasets, respectively, and surpasses the SOTA contrastive baseline by 1.93%, 3.57%, 4.85%, and 0.33%, respectively. The proposed framework also achieves a state-of-the-art accuracy of 78.4% (200 epochs) and 65.22% (100 epochs) Top-1 Linear Evaluation accuracy on ImageNet100 and ImageNet1K datasets, respectively.
♻ ☆ Segment Any Motion in Videos CVPR 2025
Moving object segmentation is a crucial task for achieving a high-level understanding of visual scenes and has numerous downstream applications. Humans can effortlessly segment moving objects in videos. Previous work has largely relied on optical flow to provide motion cues; however, this approach often results in imperfect predictions due to challenges such as partial motion, complex deformations, motion blur and background distractions. We propose a novel approach for moving object segmentation that combines long-range trajectory motion cues with DINO-based semantic features and leverages SAM2 for pixel-level mask densification through an iterative prompting strategy. Our model employs Spatio-Temporal Trajectory Attention and Motion-Semantic Decoupled Embedding to prioritize motion while integrating semantic support. Extensive testing on diverse datasets demonstrates state-of-the-art performance, excelling in challenging scenarios and fine-grained segmentation of multiple objects. Our code is available at https://motion-seg.github.io/.
comment: CVPR 2025. Website: https://motion-seg.github.io/
♻ ☆ Seeking Consistent Flat Minima for Better Domain Generalization via Refining Loss Landscapes
Domain generalization aims to learn a model from multiple training domains and generalize it to unseen test domains. Recent theory has shown that seeking the deep models, whose parameters lie in the flat minima of the loss landscape, can significantly reduce the out-of-domain generalization error. However, existing methods often neglect the consistency of loss landscapes in different domains, resulting in models that are not simultaneously in the optimal flat minima in all domains, which limits their generalization ability. To address this issue, this paper proposes an iterative Self-Feedback Training (SFT) framework to seek consistent flat minima that are shared across different domains by progressively refining loss landscapes during training. It alternatively generates a feedback signal by measuring the inconsistency of loss landscapes in different domains and refines these loss landscapes for greater consistency using this feedback signal. Benefiting from the consistency of the flat minima within these refined loss landscapes, our SFT helps achieve better out-of-domain generalization. Extensive experiments on DomainBed demonstrate superior performances of SFT when compared to state-of-the-art sharpness-aware methods and other prevalent DG baselines. On average across five DG benchmarks, SFT surpasses the sharpness-aware minimization by 2.6% with ResNet-50 and 1.5% with ViT-B/16, respectively. The code will be available soon.
♻ ☆ MambaXCTrack: Mamba-based Tracker with SSM Cross-correlation and Motion Prompt for Ultrasound Needle Tracking
Ultrasound (US)-guided needle insertion is widely employed in percutaneous interventions. However, providing feedback on the needle tip position via US imaging presents challenges due to noise, artifacts, and the thin imaging plane of US, which degrades needle features and leads to intermittent tip visibility. In this paper, a Mamba-based US needle tracker MambaXCTrack utilizing structured state space models cross-correlation (SSMX-Corr) and implicit motion prompt is proposed, which is the first application of Mamba in US needle tracking. The SSMX-Corr enhances cross-correlation by long-range modeling and global searching of distant semantic features between template and search maps, benefiting the tracking under noise and artifacts by implicitly learning potential distant semantic cues. By combining with cross-map interleaved scan (CIS), local pixel-wise interaction with positional inductive bias can also be introduced to SSMX-Corr. The implicit low-level motion descriptor is proposed as a non-visual prompt to enhance tracking robustness, addressing the intermittent tip visibility problem. Extensive experiments on a dataset with motorized needle insertion in both phantom and tissue samples demonstrate that the proposed tracker outperforms other state-of-the-art trackers while ablation studies further highlight the effectiveness of each proposed tracking module.
comment: Accepted by RAL
Information Retrieval 31
☆ Invariance Matters: Empowering Social Recommendation via Graph Invariant Learning
Graph-based social recommendation systems have shown significant promise in enhancing recommendation performance, particularly in addressing the issue of data sparsity in user behaviors. Typically, these systems leverage Graph Neural Networks (GNNs) to capture user preferences by incorporating high-order social influences from observed social networks. However, existing graph-based social recommendations often overlook the fact that social networks are inherently noisy, containing task-irrelevant relationships that can hinder accurate user preference learning. The removal of these redundant social relations is crucial, yet it remains challenging due to the lack of ground truth. In this paper, we approach the social denoising problem from the perspective of graph invariant learning and propose a novel method, Social Graph Invariant Learning(SGIL). Specifically,SGIL aims to uncover stable user preferences within the input social graph, thereby enhancing the robustness of graph-based social recommendation systems. To achieve this goal, SGIL first simulates multiple noisy social environments through graph generators. It then seeks to learn environment-invariant user preferences by minimizing invariant risk across these environments. To further promote diversity in the generated social environments, we employ an adversarial training strategy to simulate more potential social noisy distributions. Extensive experimental results demonstrate the effectiveness of the proposed SGIL. The code is available at https://github.com/yimutianyang/SIGIR2025-SGIL.
☆ Brain-Machine Interfaces & Information Retrieval Challenges and Opportunities
The fundamental goal of Information Retrieval (IR) systems lies in their capacity to effectively satisfy human information needs - a challenge that encompasses not just the technical delivery of information, but the nuanced understanding of human cognition during information seeking. Contemporary IR platforms rely primarily on observable interaction signals, creating a fundamental gap between system capabilities and users' cognitive processes. Brain-Machine Interface (BMI) technologies now offer unprecedented potential to bridge this gap through direct measurement of previously inaccessible aspects of information-seeking behaviour. This perspective paper offers a broad examination of the IR landscape, providing a comprehensive analysis of how BMI technology could transform IR systems, drawing from advances at the intersection of both neuroscience and IR research. We present our analysis through three identified fundamental vertices: (1) understanding the neural correlates of core IR concepts to advance theoretical models of search behaviour, (2) enhancing existing IR systems through contextual integration of neurophysiological signals, and (3) developing proactive IR capabilities through direct neurophysiological measurement. For each vertex, we identify specific research opportunities and propose concrete directions for developing BMI-enhanced IR systems. We conclude by examining critical technical and ethical challenges in implementing these advances, providing a structured roadmap for future research at the intersection of neuroscience and IR.
☆ AlayaDB: The Data Foundation for Efficient and Effective Long-context LLM Inference
AlayaDB is a cutting-edge vector database system natively architected for efficient and effective long-context inference for Large Language Models (LLMs) at AlayaDB AI. Specifically, it decouples the KV cache and attention computation from the LLM inference systems, and encapsulates them into a novel vector database system. For the Model as a Service providers (MaaS), AlayaDB consumes fewer hardware resources and offers higher generation quality for various workloads with different kinds of Service Level Objectives (SLOs), when comparing with the existing alternative solutions (e.g., KV cache disaggregation, retrieval-based sparse attention). The crux of AlayaDB is that it abstracts the attention computation and cache management for LLM inference into a query processing procedure, and optimizes the performance via a native query optimizer. In this work, we demonstrate the effectiveness of AlayaDB via (i) three use cases from our industry partners, and (ii) extensive experimental results on LLM inference benchmarks.
comment: 14 pages, 12 figures, conference
☆ CROSSAN: Towards Efficient and Effective Adaptation of Multiple Multimodal Foundation Models for Sequential Recommendation
Multimodal Foundation Models (MFMs) excel at representing diverse raw modalities (e.g., text, images, audio, videos, etc.). As recommender systems increasingly incorporate these modalities, leveraging MFMs to generate better representations has great potential. However, their application in sequential recommendation remains largely unexplored. This is primarily because mainstream adaptation methods, such as Fine-Tuning and even Parameter-Efficient Fine-Tuning (PEFT) techniques (e.g., Adapter and LoRA), incur high computational costs, especially when integrating multiple modality encoders, thus hindering research progress. As a result, it remains unclear whether we can efficiently and effectively adapt multiple (>2) MFMs for the sequential recommendation task. To address this, we propose a plug-and-play Cross-modal Side Adapter Network (CROSSAN). Leveraging the fully decoupled side adapter-based paradigm, CROSSAN achieves high efficiency while enabling cross-modal learning across diverse modalities. To optimize the final stage of multimodal fusion across diverse modalities, we adopt the Mixture of Modality Expert Fusion (MOMEF) mechanism. CROSSAN achieves superior performance on the public datasets for adapting four foundation models with raw modalities. Performance consistently improves as more MFMs are adapted. We will release our code and datasets to facilitate future research.
☆ MURR: Model Updating with Regularized Replay for Searching a Document Stream ECIR 2025
The Internet produces a continuous stream of new documents and user-generated queries. These naturally change over time based on events in the world and the evolution of language. Neural retrieval models that were trained once on a fixed set of query-document pairs will quickly start misrepresenting newly-created content and queries, leading to less effective retrieval. Traditional statistical sparse retrieval can update collection statistics to reflect these changes in the use of language in documents and queries. In contrast, continued fine-tuning of the language model underlying neural retrieval approaches such as DPR and ColBERT creates incompatibility with previously-encoded documents. Re-encoding and re-indexing all previously-processed documents can be costly. In this work, we explore updating a neural dual encoder retrieval model without reprocessing past documents in the stream. We propose MURR, a model updating strategy with regularized replay, to ensure the model can still faithfully search existing documents without reprocessing, while continuing to update the model for the latest topics. In our simulated streaming environments, we show that fine-tuning models using MURR leads to more effective and more consistent retrieval results than other strategies as the stream of documents and queries progresses.
comment: Published at ECIR 2025. 16 pages, 4 figures
☆ From Prompting to Alignment: A Generative Framework for Query Recommendation
In modern search systems, search engines often suggest relevant queries to users through various panels or components, helping refine their information needs. Traditionally, these recommendations heavily rely on historical search logs to build models, which suffer from cold-start or long-tail issues. Furthermore, tasks such as query suggestion, completion or clarification are studied separately by specific design, which lacks generalizability and hinders adaptation to novel applications. Despite recent attempts to explore the use of LLMs for query recommendation, these methods mainly rely on the inherent knowledge of LLMs or external sources like few-shot examples, retrieved documents, or knowledge bases, neglecting the importance of the calibration and alignment with user feedback, thus limiting their practical utility. To address these challenges, we first propose a general Generative Query Recommendation (GQR) framework that aligns LLM-based query generation with user preference. Specifically, we unify diverse query recommendation tasks by a universal prompt framework, leveraging the instruct-following capability of LLMs for effective generation. Secondly, we align LLMs with user feedback via presenting a CTR-alignment framework, which involves training a query-wise CTR predictor as a process reward model and employing list-wise preference alignment to maximize the click probability of the generated query list. Furthermore, recognizing the inconsistency between LLM knowledge and proactive search intents arising from the separation of user-initiated queries from models, we align LLMs with user initiative via retrieving co-occurrence queries as side information when historical logs are available.
☆ HistLLM: A Unified Framework for LLM-Based Multimodal Recommendation with User History Encoding and Compression
While large language models (LLMs) have proven effective in leveraging textual data for recommendations, their application to multimodal recommendation tasks remains relatively underexplored. Although LLMs can process multimodal information through projection functions that map visual features into their semantic space, recommendation tasks often require representing users' history interactions through lengthy prompts combining text and visual elements, which not only hampers training and inference efficiency but also makes it difficult for the model to accurately capture user preferences from complex and extended prompts, leading to reduced recommendation performance. To address this challenge, we introduce HistLLM, an innovative multimodal recommendation framework that integrates textual and visual features through a User History Encoding Module (UHEM), compressing multimodal user history interactions into a single token representation, effectively facilitating LLMs in processing user preferences. Extensive experiments demonstrate the effectiveness and efficiency of our proposed mechanism.
☆ A Survey of Personalization: From RAG to Agent
Personalization has become an essential capability in modern AI systems, enabling customized interactions that align with individual user preferences, contexts, and goals. Recent research has increasingly concentrated on Retrieval-Augmented Generation (RAG) frameworks and their evolution into more advanced agent-based architectures within personalized settings to enhance user satisfaction. Building on this foundation, this survey systematically examines personalization across the three core stages of RAG: pre-retrieval, retrieval, and generation. Beyond RAG, we further extend its capabilities into the realm of Personalized LLM-based Agents, which enhance traditional RAG systems with agentic functionalities, including user understanding, personalized planning and execution, and dynamic generation. For both personalization in RAG and agent-based personalization, we provide formal definitions, conduct a comprehensive review of recent literature, and summarize key datasets and evaluation metrics. Additionally, we discuss fundamental challenges, limitations, and promising research directions in this evolving field. Relevant papers and resources are continuously updated at https://github.com/Applied-Machine-Learning-Lab/Awesome-Personalized-RAG-Agent.
comment: 18 pages, 5 figures
☆ Unveiling Contrastive Learning's Capability of Neighborhood Aggregation for Collaborative Filtering SIGIR2025
Personalized recommendation is widely used in the web applications, and graph contrastive learning (GCL) has gradually become a dominant approach in recommender systems, primarily due to its ability to extract self-supervised signals from raw interaction data, effectively alleviating the problem of data sparsity. A classic GCL-based method typically uses data augmentation during graph convolution to generates more contrastive views, and performs contrast on these new views to obtain rich self-supervised signals. Despite this paradigm is effective, the reasons behind the performance gains remain a mystery. In this paper, we first reveal via theoretical derivation that the gradient descent process of the CL objective is formally equivalent to graph convolution, which implies that CL objective inherently supports neighborhood aggregation on interaction graphs. We further substantiate this capability through experimental validation and identify common misconceptions in the selection of positive samples in previous methods, which limit the potential of CL objective. Based on this discovery, we propose the Light Contrastive Collaborative Filtering (LightCCF) method, which introduces a novel neighborhood aggregation objective to bring users closer to all interacted items while pushing them away from other positive pairs, thus achieving high-quality neighborhood aggregation with very low time complexity. On three highly sparse public datasets, the proposed method effectively aggregate neighborhood information while preventing graph over-smoothing, demonstrating significant improvements over existing GCL-based counterparts in both training efficiency and recommendation accuracy. Our implementations are publicly accessible.
comment: Accepted by SIGIR2025
☆ Enhancing LLM-based Recommendation through Semantic-Aligned Collaborative Knowledge
Large Language Models (LLMs) demonstrate remarkable capabilities in leveraging comprehensive world knowledge and sophisticated reasoning mechanisms for recommendation tasks. However, a notable limitation lies in their inability to effectively model sparse identifiers (e.g., user and item IDs), unlike conventional collaborative filtering models (Collabs.), thus hindering LLM to learn distinctive user-item representations and creating a performance bottleneck. Prior studies indicate that integrating collaborative knowledge from Collabs. into LLMs can mitigate the above limitations and enhance their recommendation performance. Nevertheless, the significant discrepancy in knowledge distribution and semantic space between LLMs and Collab. presents substantial challenges for effective knowledge transfer. To tackle these challenges, we propose a novel framework, SeLLa-Rec, which focuses on achieving alignment between the semantic spaces of Collabs. and LLMs. This alignment fosters effective knowledge fusion, mitigating the influence of discriminative noise and facilitating the deep integration of knowledge from diverse models. Specifically, three special tokens with collaborative knowledge are embedded into the LLM's semantic space through a hybrid projection layer and integrated into task-specific prompts to guide the recommendation process. Experiments conducted on two public benchmark datasets (MovieLens-1M and Amazon Book) demonstrate that SeLLa-Rec achieves state-of-the-art performance.
☆ Session-based Recommender Systems: User Interest as a Stochastic Process in the Latent Space
This paper jointly addresses the problem of data uncertainty, popularity bias, and exposure bias in session-based recommender systems. We study the symptoms of this bias both in item embeddings and in recommendations. We propose treating user interest as a stochastic process in the latent space and providing a model-agnostic implementation of this mathematical concept. The proposed stochastic component consists of elements: debiasing item embeddings with regularization for embedding uniformity, modeling dense user interest from session prefixes, and introducing fake targets in the data to simulate extended exposure. We conducted computational experiments on two popular benchmark datasets, Diginetica and YooChoose 1/64, as well as several modifications of the YooChoose dataset with different ratios of popular items. The results show that the proposed approach allows us to mitigate the challenges mentioned.
☆ On Precomputation and Caching in Information Retrieval Experiments with Pipeline Architectures ECIR 2025
Modern information retrieval systems often rely on multiple components executed in a pipeline. In a research setting, this can lead to substantial redundant computations (e.g., retrieving the same query multiple times for evaluating different downstream rerankers). To overcome this, researchers take cached "result" files as inputs, which represent the output of another pipeline. However, these result files can be brittle and can cause a disconnect between the conceptual design of the pipeline and its logical implementation. To overcome both the redundancy problem (when executing complete pipelines) and the disconnect problem (when relying on intermediate result files), we describe our recent efforts to improve the caching capabilities in the open-source PyTerrier IR platform. We focus on two main directions: (1) automatic implicit caching of common pipeline prefixes when comparing systems and (2) explicit caching of operations through a new extension package, pyterrier-caching. These approaches allow for the best of both worlds: pipelines can be fully expressed end-to-end, while also avoiding redundant computations between pipelines.
comment: WOWS @ ECIR 2025
☆ Constrained Auto-Regressive Decoding Constrains Generative Retrieval SIGIR 2025
Generative retrieval seeks to replace traditional search index data structures with a single large-scale neural network, offering the potential for improved efficiency and seamless integration with generative large language models. As an end-to-end paradigm, generative retrieval adopts a learned differentiable search index to conduct retrieval by directly generating document identifiers through corpus-specific constrained decoding. The generalization capabilities of generative retrieval on out-of-distribution corpora have gathered significant attention. In this paper, we examine the inherent limitations of constrained auto-regressive generation from two essential perspectives: constraints and beam search. We begin with the Bayes-optimal setting where the generative retrieval model exactly captures the underlying relevance distribution of all possible documents. Then we apply the model to specific corpora by simply adding corpus-specific constraints. Our main findings are two-fold: (i) For the effect of constraints, we derive a lower bound of the error, in terms of the KL divergence between the ground-truth and the model-predicted step-wise marginal distributions. (ii) For the beam search algorithm used during generation, we reveal that the usage of marginal distributions may not be an ideal approach. This paper aims to improve our theoretical understanding of the generalization capabilities of the auto-regressive decoding retrieval paradigm, laying a foundation for its limitations and inspiring future advancements toward more robust and generalizable generative retrieval.
comment: 13 pages, 6 figures, 2 tables, accepted by SIGIR 2025 (Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval)
☆ StePO-Rec: Towards Personalized Outfit Styling Assistant via Knowledge-Guided Multi-Step Reasoning
Advancements in Generative AI offers new opportunities for FashionAI, surpassing traditional recommendation systems that often lack transparency and struggle to integrate expert knowledge, leaving the potential for personalized fashion styling remain untapped. To address these challenges, we present PAFA (Principle-Aware Fashion), a multi-granular knowledge base that organizes professional styling expertise into three levels of metadata, domain principles, and semantic relationships. Using PAFA, we develop StePO-Rec, a knowledge-guided method for multi-step outfit recommendation. StePO-Rec provides structured suggestions using a scenario-dimension-attribute framework, employing recursive tree construction to align recommendations with both professional principles and individual preferences. A preference-trend re-ranking system further adapts to fashion trends while maintaining the consistency of the user's original style. Experiments on the widely used personalized outfit dataset IQON show a 28% increase in Recall@1 and 32.8% in MAP. Furthermore, case studies highlight improved explainability, traceability, result reliability, and the seamless integration of expertise and personalization.
☆ Constructing Micro Knowledge Graphs from Technical Support Documents
Short technical support pages such as IBM Technotes are quite common in technical support domain. These pages can be very useful as the knowledge sources for technical support applications such as chatbots, search engines and question-answering (QA) systems. Information extracted from documents to drive technical support applications is often stored in the form of Knowledge Graph (KG). Building KGs from a large corpus of documents poses a challenge of granularity because a large number of entities and actions are present in each page. The KG becomes virtually unusable if all entities and actions from these pages are stored in the KG. Therefore, only key entities and actions from each page are extracted and stored in the KG. This approach however leads to loss of knowledge represented by entities and actions left out of the KG as they are no longer available to graph search and reasoning functions. We propose a set of techniques to create micro knowledge graph (micrograph) for each of such web pages. The micrograph stores all the entities and actions in a page and also takes advantage of the structure of the page to represent exactly in which part of that page these entities and actions appeared, and also how they relate to each other. These micrographs can be used as additional knowledge sources by technical support applications. We define schemas for representing semi-structured and plain text knowledge present in the technical support web pages. Solutions in technical support domain include procedures made of steps. We also propose a technique to extract procedures from these webpages and the schemas to represent them in the micrographs. We also discuss how technical support applications can take advantage of the micrographs.
☆ RAKG:Document-level Retrieval Augmented Knowledge Graph Construction
With the rise of knowledge graph based retrieval-augmented generation (RAG) techniques such as GraphRAG and Pike-RAG, the role of knowledge graphs in enhancing the reasoning capabilities of large language models (LLMs) has become increasingly prominent. However, traditional Knowledge Graph Construction (KGC) methods face challenges like complex entity disambiguation, rigid schema definition, and insufficient cross-document knowledge integration. This paper focuses on the task of automatic document-level knowledge graph construction. It proposes the Document-level Retrieval Augmented Knowledge Graph Construction (RAKG) framework. RAKG extracts pre-entities from text chunks and utilizes these pre-entities as queries for RAG, effectively addressing the issue of long-context forgetting in LLMs and reducing the complexity of Coreference Resolution. In contrast to conventional KGC methods, RAKG more effectively captures global information and the interconnections among disparate nodes, thereby enhancing the overall performance of the model. Additionally, we transfer the RAG evaluation framework to the KGC field and filter and evaluate the generated knowledge graphs, thereby avoiding incorrectly generated entities and relationships caused by hallucinations in LLMs. We further developed the MINE dataset by constructing standard knowledge graphs for each article and experimentally validated the performance of RAKG. The results show that RAKG achieves an accuracy of 95.91 % on the MINE dataset, a 6.2 % point improvement over the current best baseline, GraphRAG (89.71 %). The code is available at https://github.com/LMMApplication/RAKG.
comment: 9 pages, 6 figures
☆ Augmented Relevance Datasets with Fine-Tuned Small LLMs WSDM '25
Building high-quality datasets and labeling query-document relevance are essential yet resource-intensive tasks, requiring detailed guidelines and substantial effort from human annotators. This paper explores the use of small, fine-tuned large language models (LLMs) to automate relevance assessment, with a focus on improving ranking models' performance by augmenting their training dataset. We fine-tuned small LLMs to enhance relevance assessments, thereby improving dataset creation quality for downstream ranking model training. Our experiments demonstrate that these fine-tuned small LLMs not only outperform certain closed source models on our dataset but also lead to substantial improvements in ranking model performance. These results highlight the potential of leveraging small LLMs for efficient and scalable dataset augmentation, providing a practical solution for search engine optimization.
comment: 10 pages, 3 figures, and 6 tables. Accepted and presented to LLM4EVAL at WSDM '25
☆ VDocRAG: Retrieval-Augmented Generation over Visually-Rich Documents CVPR 2025
We aim to develop a retrieval-augmented generation (RAG) framework that answers questions over a corpus of visually-rich documents presented in mixed modalities (e.g., charts, tables) and diverse formats (e.g., PDF, PPTX). In this paper, we introduce a new RAG framework, VDocRAG, which can directly understand varied documents and modalities in a unified image format to prevent missing information that occurs by parsing documents to obtain text. To improve the performance, we propose novel self-supervised pre-training tasks that adapt large vision-language models for retrieval by compressing visual information into dense token representations while aligning them with textual content in documents. Furthermore, we introduce OpenDocVQA, the first unified collection of open-domain document visual question answering datasets, encompassing diverse document types and formats. OpenDocVQA provides a comprehensive resource for training and evaluating retrieval and question answering models on visually-rich documents in an open-domain setting. Experiments show that VDocRAG substantially outperforms conventional text-based RAG and has strong generalization capability, highlighting the potential of an effective RAG paradigm for real-world documents.
comment: Accepted by CVPR 2025; project page: https://vdocrag.github.io
☆ Epistemic Uncertainty-aware Recommendation Systems via Bayesian Deep Ensemble Learning
Recommending items to users has long been a fundamental task, and studies have tried to improve it ever since. Most well-known models commonly employ representation learning to map users and items into a unified embedding space for matching assessment. These approaches have primary limitations, especially when dealing with explicit feedback and sparse data contexts. Two primary limitations are their proneness to overfitting and failure to incorporate epistemic uncertainty in predictions. To address these problems, we propose a novel Bayesian Deep Ensemble Collaborative Filtering method named BDECF. To improve model generalization and quality, we utilize Bayesian Neural Networks, which incorporate uncertainty within their weight parameters. In addition, we introduce a new interpretable non-linear matching approach for the user and item embeddings, leveraging the advantages of the attention mechanism. Furthermore, we endorse the implementation of an ensemble-based supermodel to generate more robust and reliable predictions, resulting in a more complete model. Empirical evaluation through extensive experiments and ablation studies across a range of publicly accessible real-world datasets with differing sparsity characteristics confirms our proposed method's effectiveness and the importance of its components.
comment: 10 pages
☆ Enhancing Document Retrieval for Curating N-ary Relations in Knowledge Bases
Curation of biomedical knowledge bases (KBs) relies on extracting accurate multi-entity relational facts from the literature - a process that remains largely manual and expert-driven. An essential step in this workflow is retrieving documents that can support or complete partially observed n-ary relations. We present a neural retrieval model designed to assist KB curation by identifying documents that help fill in missing relation arguments and provide relevant contextual evidence. To reduce dependence on scarce gold-standard training data, we exploit existing KB records to construct weakly supervised training sets. Our approach introduces two key technical contributions: (i) a layered contrastive loss that enables learning from noisy and incomplete relational structures, and (ii) a balanced sampling strategy that generates high-quality negatives from diverse KB records. On two biomedical retrieval benchmarks, our approach achieves state-of-the-art performance, outperforming strong baselines in NDCG@10 by 5.7 and 3.7 percentage points, respectively.
♻ ☆ How Do Recommendation Models Amplify Popularity Bias? An Analysis from the Spectral Perspective
Recommendation Systems (RS) are often plagued by popularity bias. When training a recommendation model on a typically long-tailed dataset, the model tends to not only inherit this bias but often exacerbate it, resulting in over-representation of popular items in the recommendation lists. This study conducts comprehensive empirical and theoretical analyses to expose the root causes of this phenomenon, yielding two core insights: 1) Item popularity is memorized in the principal spectrum of the score matrix predicted by the recommendation model; 2) The dimension collapse phenomenon amplifies the relative prominence of the principal spectrum, thereby intensifying the popularity bias. Building on these insights, we propose a novel debiasing strategy that leverages a spectral norm regularizer to penalize the magnitude of the principal singular value. We have developed an efficient algorithm to expedite the calculation of the spectral norm by exploiting the spectral property of the score matrix. Extensive experiments across seven real-world datasets and three testing paradigms have been conducted to validate the superiority of the proposed method.
comment: 14 pages, 7 figures
♻ ☆ Weighted Tensor Decompositions for Context-aware Collaborative Filtering
Over recent years it has become well accepted that user interest is not static or immutable. There are a variety of contextual factors, such as time of day, the weather or the user's mood, that influence the current interests of the user. Modelling approaches need to take these factors into account if they want to succeed at finding the most relevant content to recommend given the situation. A popular method for context-aware recommendation is to encode context attributes as extra dimensions of the classic user-item interaction matrix, effectively turning it into a tensor, followed by applying the appropriate tensor decomposition methods to learn missing values. However, unlike with matrix factorization, where all decompositions are essentially a product of matrices, there exist many more options for decomposing tensors by combining vector, matrix and tensor products. We study the most successful decomposition methods that use weighted square loss and categorize them based on their tensor structure and regularization strategy. Additionally, we further extend the pool of methods by filling in the missing combinations. In this paper we provide an overview of the properties of the different decomposition methods, such as their complexity, scalability, and modelling capacity. These benefits are then contrasted with the performances achieved in offline experiments to gain more insight into which method to choose depending on a specific situation and constraints.
comment: Workshop on Context-Aware Recommender Systems, September 18, 2023, Singapore
♻ ☆ Can Offline Metrics Measure Explanation Goals? A Comparative Survey Analysis of Offline Explanation Metrics in Recommender Systems
Explanations in a Recommender System (RS) provide reasons for recommendations to users and can enhance transparency, persuasiveness, engagement, and trust-known as explanation goals. Evaluating the effectiveness of explanation algorithms offline remains challenging due to subjectivity. Initially, we conducted a literature review on current offline metrics, revealing that algorithms are often assessed with anecdotal evidence, offering convincing examples, or with metrics that don't align with human perception. We investigated whether, in explanations connecting interacted and recommended items based on shared content, the selection of item attributes and interacted items affects explanation goals. Metrics measuring the diversity and popularity of attributes and the recency of item interactions were used to evaluate explanations from three state-of-the-art agnostic algorithms across six recommendation systems. These offline metrics were compared with results from an online user study. Our findings reveal a trade-off: transparency and trust relate to popular properties, while engagement and persuasiveness are linked to diversified properties. This study contributes to the development of more robust evaluation methods for explanation algorithms in recommender systems.
♻ ☆ External Large Foundation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendation WWW
Ads recommendation is a prominent service of online advertising systems and has been actively studied. Recent studies indicate that scaling-up and advanced design of the recommendation model can bring significant performance improvement. However, with a larger model scale, such prior studies have a significantly increasing gap from industry as they often neglect two fundamental challenges in industrial-scale applications. First, training and inference budgets are restricted for the model to be served, exceeding which may incur latency and impair user experience. Second, large-volume data arrive in a streaming mode with data distributions dynamically shifting, as new users/ads join and existing users/ads leave the system. We propose the External Large Foundation Model (ExFM) framework to address the overlooked challenges. Specifically, we develop external distillation and a data augmentation system (DAS) to control the computational cost of training/inference while maintaining high performance. We design the teacher in a way like a foundation model (FM) that can serve multiple students as vertical models (VMs) to amortize its building cost. We propose Auxiliary Head and Student Adapter to mitigate the data distribution gap between FM and VMs caused by the streaming data issue. Comprehensive experiments on internal industrial-scale applications and public datasets demonstrate significant performance gain by ExFM.
comment: Accepted by the ACM Web Conference (WWW) 2025 Industrial Track as Oral Presentation
♻ ☆ Data Augmentation as Free Lunch: Exploring the Test-Time Augmentation for Sequential Recommendation SIGIR 2025
Data augmentation has become a promising method of mitigating data sparsity in sequential recommendation. Existing methods generate new yet effective data during model training to improve performance. However, deploying them requires retraining, architecture modification, or introducing additional learnable parameters. The above steps are time-consuming and costly for well-trained models, especially when the model scale becomes large. In this work, we explore the test-time augmentation (TTA) for sequential recommendation, which augments the inputs during the model inference and then aggregates the model's predictions for augmented data to improve final accuracy. It avoids significant time and cost overhead from loss calculation and backward propagation. We first experimentally disclose the potential of existing augmentation operators for TTA and find that the Mask and Substitute consistently achieve better performance. Further analysis reveals that these two operators are effective because they retain the original sequential pattern while adding appropriate perturbations. Meanwhile, we argue that these two operators still face time-consuming item selection or interference information from mask tokens. Based on the analysis and limitations, we present TNoise and TMask. The former injects uniform noise into the original representation, avoiding the computational overhead of item selection. The latter blocks mask token from participating in model calculations or directly removes interactions that should have been replaced with mask tokens. Comprehensive experiments demonstrate the effectiveness, efficiency, and generalizability of our method. We provide an anonymous implementation at https://github.com/KingGugu/TTA4SR.
comment: Accepted by SIGIR 2025 Full Paper
♻ ☆ Pre-training Generative Recommender with Multi-Identifier Item Tokenization
Generative recommendation autoregressively generates item identifiers to recommend potential items. Existing methods typically adopt a one-to-one mapping strategy, where each item is represented by a single identifier. However, this scheme poses issues, such as suboptimal semantic modeling for low-frequency items and limited diversity in token sequence data. To overcome these limitations, we propose MTGRec, which leverages Multi-identifier item Tokenization to augment token sequence data for Generative Recommender pre-training. Our approach involves two key innovations: multi-identifier item tokenization and curriculum recommender pre-training. For multi-identifier item tokenization, we leverage the RQ-VAE as the tokenizer backbone and treat model checkpoints from adjacent training epochs as semantically relevant tokenizers. This allows each item to be associated with multiple identifiers, enabling a single user interaction sequence to be converted into several token sequences as different data groups. For curriculum recommender pre-training, we introduce a curriculum learning scheme guided by data influence estimation, dynamically adjusting the sampling probability of each data group during recommender pre-training. After pre-training, we fine-tune the model using a single tokenizer to ensure accurate item identification for recommendation. Extensive experiments on three public benchmark datasets demonstrate that MTGRec significantly outperforms both traditional and generative recommendation baselines in terms of effectiveness and scalability.
♻ ☆ Universal Item Tokenization for Transferable Generative Recommendation
Recently, generative recommendation has emerged as a promising paradigm, attracting significant research attention. The basic framework involves an item tokenizer, which represents each item as a sequence of codes serving as its identifier, and a generative recommender that predicts the next item by autoregressively generating the target item identifier. However, in existing methods, both the tokenizer and the recommender are typically domain-specific, limiting their ability for effective transfer or adaptation to new domains. To this end, we propose UTGRec, a Universal item Tokenization approach for transferable Generative Recommendation. Specifically, we design a universal item tokenizer for encoding rich item semantics by adapting a multimodal large language model (MLLM). By devising tree-structured codebooks, we discretize content representations into corresponding codes for item tokenization. To effectively learn the universal item tokenizer on multiple domains, we introduce two key techniques in our approach. For raw content reconstruction, we employ dual lightweight decoders to reconstruct item text and images from discrete representations to capture general knowledge embedded in the content. For collaborative knowledge integration, we assume that co-occurring items are similar and integrate collaborative signals through co-occurrence alignment and reconstruction. Finally, we present a joint learning framework to pre-train and adapt the transferable generative recommender across multiple domains. Extensive experiments on four public datasets demonstrate the superiority of UTGRec compared to both traditional and generative recommendation baselines.
♻ ☆ Multi-view Intent Learning and Alignment with Large Language Models for Session-based Recommendation
Session-based recommendation (SBR) methods often rely on user behavior data, which can struggle with the sparsity of session data, limiting performance. Researchers have identified that beyond behavioral signals, rich semantic information in item descriptions is crucial for capturing hidden user intent. While large language models (LLMs) offer new ways to leverage this semantic data, the challenges of session anonymity, short-sequence nature, and high LLM training costs have hindered the development of a lightweight, efficient LLM framework for SBR. To address the above challenges, we propose an LLM-enhanced SBR framework that integrates semantic and behavioral signals from multiple views. This two-stage framework leverages the strengths of both LLMs and traditional SBR models while minimizing training costs. In the first stage, we use multi-view prompts to infer latent user intentions at the session semantic level, supported by an intent localization module to alleviate LLM hallucinations. In the second stage, we align and unify these semantic inferences with behavioral representations, effectively merging insights from both large and small models. Extensive experiments on two real datasets demonstrate that the LLM4SBR framework can effectively improve model performance. We release our codes along with the baselines at https://github.com/tsinghua-fib-lab/LLM4SBR.
♻ ☆ RAG-VR: Leveraging Retrieval-Augmented Generation for 3D Question Answering in VR Environments
Recent advances in large language models (LLMs) provide new opportunities for context understanding in virtual reality (VR). However, VR contexts are often highly localized and personalized, limiting the effectiveness of general-purpose LLMs. To address this challenge, we present RAG-VR, the first 3D question-answering system for VR that incorporates retrieval-augmented generation (RAG), which augments an LLM with external knowledge retrieved from a localized knowledge database to improve the answer quality. RAG-VR includes a pipeline for extracting comprehensive knowledge about virtual environments and user conditions for accurate answer generation. To ensure efficient retrieval, RAG-VR offloads the retrieval process to a nearby edge server and uses only essential information during retrieval. Moreover, we train the retriever to effectively distinguish among relevant, irrelevant, and hard-to-differentiate information in relation to questions. RAG-VR improves answer accuracy by 17.9%-41.8% and reduces end-to-end latency by 34.5%-47.3% compared with two baseline systems.
comment: GenAI-XR 2025 Workshop, co-located with 2025 IEEE Conference on Virtual Reality and 3D User Interfaces (VR)
♻ ☆ Trustworthy Answers, Messier Data: Bridging the Gap in Low-Resource Retrieval-Augmented Generation for Domain Expert Systems
RAG has become a key technique for enhancing LLMs by reducing hallucinations, especially in domain expert systems where LLMs may lack sufficient inherent knowledge. However, developing these systems in low-resource settings introduces several challenges: (1) handling heterogeneous data sources, (2) optimizing retrieval phase for trustworthy answers, and (3) evaluating generated answers across diverse aspects. To address these, we introduce a data generation pipeline that transforms raw multi-modal data into structured corpus and Q&A pairs, an advanced re-ranking phase improving retrieval precision, and a reference matching algorithm enhancing answer traceability. Applied to the automotive engineering domain, our system improves factual correctness (+1.94), informativeness (+1.16), and helpfulness (+1.67) over a non-RAG baseline, based on a 1-5 scale by an LLM judge. These results highlight the effectiveness of our approach across distinct aspects, with strong answer grounding and transparency.
♻ ☆ Snippet-based Conversational Recommender System
Conversational Recommender Systems (CRS) engage users in interactive dialogues to gather preferences and provide personalized recommendations. While existing studies have advanced conversational strategies, they often rely on predefined attributes or expensive, domain-specific annotated datasets, which limits their flexibility in handling diverse user preferences and adaptability across domains. We propose SnipRec, a novel resource-efficient approach that leverages user-generated content, such as customer reviews, to capture a broader range of user expressions. By employing large language models to map reviews and user responses into concise snippets, SnipRec represents user preferences and retrieves relevant items without the need for intensive manual data collection or fine-tuning. Experiments across the restaurant, book, and clothing domains show that snippet-based representations outperform document- and sentence-based representations, achieving Hits@10 of 0.25-0.55 with 3,000 to 10,000 candidate items while successfully handling free-form user responses.
Robotics 23
☆ UruBots RoboCup Work Team Description Paper
This work presents a team description paper for the RoboCup Work League. Our team, UruBots, has been developing robots and projects for research and competitions in the last three years, attending robotics competitions in Uruguay and around the world. In this instance, we aim to participate and contribute to the RoboCup Work category, hopefully making our debut in this prestigious competition. For that, we present an approach based on the Limo robot, whose main characteristic is its hybrid locomotion system with wheels and tracks, with some extras added by the team to complement the robot's functionalities. Overall, our approach allows the robot to efficiently and autonomously navigate a Work scenario, with the ability to manipulate objects, perform autonomous navigation, and engage in a simulated industrial environment.
comment: 6 pages, 5 figures, submitted to RoboCup 2025
☆ Adapting Robot's Explanation for Failures Based on Observed Human Behavior in Human-Robot Collaboration IROS 2025
This work aims to interpret human behavior to anticipate potential user confusion when a robot provides explanations for failure, allowing the robot to adapt its explanations for more natural and efficient collaboration. Using a dataset that included facial emotion detection, eye gaze estimation, and gestures from 55 participants in a user study, we analyzed how human behavior changed in response to different types of failures and varying explanation levels. Our goal is to assess whether human collaborators are ready to accept less detailed explanations without inducing confusion. We formulate a data-driven predictor to predict human confusion during robot failure explanations. We also propose and evaluate a mechanism, based on the predictor, to adapt the explanation level according to observed human behavior. The promising results from this evaluation indicate the potential of this research in adapting a robot's explanations for failures to enhance the collaborative experience.
comment: Under review, Manuscript in submission for IROS 2025
☆ From Movement Primitives to Distance Fields to Dynamical Systems
Developing autonomous robots capable of learning and reproducing complex motions from demonstrations remains a fundamental challenge in robotics. On the one hand, movement primitives (MPs) provide a compact and modular representation of continuous trajectories. On the other hand, autonomous systems provide control policies that are time independent. We propose in this paper a simple and flexible approach that gathers the advantages of both representations by transforming MPs into autonomous systems. The key idea is to transform the explicit representation of a trajectory as an implicit shape encoded as a distance field. This conversion from a time-dependent motion to a spatial representation enables the definition of an autonomous dynamical system with modular reactions to perturbation. Asymptotic stability guarantees are provided by using Bernstein basis functions in the MPs, representing trajectories as concatenated quadratic B\'ezier curves, which provide an analytical method for computing distance fields. This approach bridges conventional MPs with distance fields, ensuring smooth and precise motion encoding, while maintaining a continuous spatial representation. By simply leveraging the analytic gradients of the curve and its distance field, a stable dynamical system can be computed to reproduce the demonstrated trajectories while handling perturbations, without requiring a model of the dynamical system to be estimated. Numerical simulations and real-world robotic experiments validate our method's ability to encode complex motion patterns while ensuring trajectory stability, together with the flexibility of designing the desired reaction to perturbations. An interactive project page demonstrating our approach is available at https://mp-df-ds.github.io/.
comment: 7 pages, 7 Figures
☆ A highly maneuverable flying squirrel drone with agility-improving foldable wings
Drones, like most airborne aerial vehicles, face inherent disadvantages in achieving agile flight due to their limited thrust capabilities. These physical constraints cannot be fully addressed through advancements in control algorithms alone. Drawing inspiration from the winged flying squirrel, this paper proposes a highly maneuverable drone equipped with agility-enhancing foldable wings. By leveraging collaborative control between the conventional propeller system and the foldable wings-coordinated through the Thrust-Wing Coordination Control (TWCC) framework-the controllable acceleration set is expanded, enabling the generation of abrupt vertical forces that are unachievable with traditional wingless drones. The complex aerodynamics of the foldable wings are modeled using a physics-assisted recurrent neural network (paRNN), which calibrates the angle of attack (AOA) to align with the real aerodynamic behavior of the wings. The additional air resistance generated by appropriately deploying these wings significantly improves the tracking performance of the proposed "flying squirrel" drone. The model is trained on real flight data and incorporates flat-plate aerodynamic principles. Experimental results demonstrate that the proposed flying squirrel drone achieves a 13.1% improvement in tracking performance, as measured by root mean square error (RMSE), compared to a conventional wingless drone. A demonstration video is available on YouTube: https://youtu.be/O8nrip18azY.
comment: Accepted to IEEE Robotics and Automation Letters Youtube : https://youtu.be/tckIF3KCJig?si=s-N2VRNThV23y9xQ
☆ GeoNav: Empowering MLLMs with Explicit Geospatial Reasoning Abilities for Language-Goal Aerial Navigation
Language-goal aerial navigation is a critical challenge in embodied AI, requiring UAVs to localize targets in complex environments such as urban blocks based on textual specification. Existing methods, often adapted from indoor navigation, struggle to scale due to limited field of view, semantic ambiguity among objects, and lack of structured spatial reasoning. In this work, we propose GeoNav, a geospatially aware multimodal agent to enable long-range navigation. GeoNav operates in three phases-landmark navigation, target search, and precise localization-mimicking human coarse-to-fine spatial strategies. To support such reasoning, it dynamically builds two different types of spatial memory. The first is a global but schematic cognitive map, which fuses prior textual geographic knowledge and embodied visual cues into a top-down, annotated form for fast navigation to the landmark region. The second is a local but delicate scene graph representing hierarchical spatial relationships between blocks, landmarks, and objects, which is used for definite target localization. On top of this structured representation, GeoNav employs a spatially aware, multimodal chain-of-thought prompting mechanism to enable multimodal large language models with efficient and interpretable decision-making across stages. On the CityNav urban navigation benchmark, GeoNav surpasses the current state-of-the-art by up to 12.53% in success rate and significantly improves navigation efficiency, even in hard-level tasks. Ablation studies highlight the importance of each module, showcasing how geospatial representations and coarse-to-fine reasoning enhance UAV navigation.
☆ AirVista-II: An Agentic System for Embodied UAVs Toward Dynamic Scene Semantic Understanding
Unmanned Aerial Vehicles (UAVs) are increasingly important in dynamic environments such as logistics transportation and disaster response. However, current tasks often rely on human operators to monitor aerial videos and make operational decisions. This mode of human-machine collaboration suffers from significant limitations in efficiency and adaptability. In this paper, we present AirVista-II -- an end-to-end agentic system for embodied UAVs, designed to enable general-purpose semantic understanding and reasoning in dynamic scenes. The system integrates agent-based task identification and scheduling, multimodal perception mechanisms, and differentiated keyframe extraction strategies tailored for various temporal scenarios, enabling the efficient capture of critical scene information. Experimental results demonstrate that the proposed system achieves high-quality semantic understanding across diverse UAV-based dynamic scenarios under a zero-shot setting.
☆ Embodied Chain of Action Reasoning with Multi-Modal Foundation Model for Humanoid Loco-manipulation
Enabling humanoid robots to autonomously perform loco-manipulation tasks in complex, unstructured environments poses significant challenges. This entails equipping robots with the capability to plan actions over extended horizons while leveraging multi-modality to bridge gaps between high-level planning and actual task execution. Recent advancements in multi-modal foundation models have showcased substantial potential in enhancing planning and reasoning abilities, particularly in the comprehension and processing of semantic information for robotic control tasks. In this paper, we introduce a novel framework based on foundation models that applies the embodied chain of action reasoning methodology to autonomously plan actions from textual instructions for humanoid loco-manipulation. Our method integrates humanoid-specific chain of thought methodology, including detailed affordance and body movement analysis, which provides a breakdown of the task into a sequence of locomotion and manipulation actions. Moreover, we incorporate spatial reasoning based on the observation and target object properties to effectively navigate where target position may be unseen or occluded. Through rigorous experimental setups on object rearrangement, manipulations and loco-manipulation tasks on a real-world environment, we evaluate our method's efficacy on the decoupled upper and lower body control and demonstrate the effectiveness of the chain of robotic action reasoning strategies in comprehending human instructions.
☆ RoboComm: A DID-based scalable and privacy-preserving Robot-to-Robot interaction over state channels
In a multi robot system establishing trust amongst untrusted robots from different organisations while preserving a robot's privacy is a challenge. Recently decentralized technologies such as smart contract and blockchain are being explored for applications in robotics. However, the limited transaction processing and high maintenance cost hinder the widespread adoption of such approaches. Moreover, blockchain transactions be they on public or private permissioned blockchain are publically readable which further fails to preserve the confidentiality of the robot's data and privacy of the robot. In this work, we propose RoboComm a Decentralized Identity based approach for privacy-preserving interaction between robots. With DID a component of Self-Sovereign Identity; robots can authenticate each other independently without relying on any third-party service. Verifiable Credentials enable private data associated with a robot to be stored within the robot's hardware, unlike existing blockchain based approaches where the data has to be on the blockchain. We improve throughput by allowing message exchange over state channels. Being a blockchain backed solution RoboComm provides a trustworthy system without relying on a single party. Moreover, we implement our proposed approach to demonstrate the feasibility of our solution.
comment: arXiv admin note: text overlap with arXiv:2405.02476 by other authors
☆ Towards Intuitive Drone Operation Using a Handheld Motion Controller
We present an intuitive human-drone interaction system that utilizes a gesture-based motion controller to enhance the drone operation experience in real and simulated environments. The handheld motion controller enables natural control of the drone through the movements of the operator's hand, thumb, and index finger: the trigger press manages the throttle, the tilt of the hand adjusts pitch and roll, and the thumbstick controls yaw rotation. Communication with drones is facilitated via the ExpressLRS radio protocol, ensuring robust connectivity across various frequencies. The user evaluation of the flight experience with the designed drone controller using the UEQ-S survey showed high scores for both Pragmatic (mean=2.2, SD = 0.8) and Hedonic (mean=2.3, SD = 0.9) Qualities. This versatile control interface supports applications such as research, drone racing, and training programs in real and simulated environments, thereby contributing to advances in the field of human-drone interaction.
comment: HRI'25: Proceedings of the 2025 ACM/IEEE International Conference on Human-Robot Interaction, 5 pages, 5 figures
☆ Debiasing 6-DOF IMU via Hierarchical Learning of Continuous Bias Dynamics
This paper develops a deep learning approach to the online debiasing of IMU gyroscopes and accelerometers. Most existing methods rely on implicitly learning a bias term to compensate for raw IMU data. Explicit bias learning has recently shown its potential as a more interpretable and motion-independent alternative. However, it remains underexplored and faces challenges, particularly the need for ground truth bias data, which is rarely available. To address this, we propose a neural ordinary differential equation (NODE) framework that explicitly models continuous bias dynamics, requiring only pose ground truth, often available in datasets. This is achieved by extending the canonical NODE framework to the matrix Lie group for IMU kinematics with a hierarchical training strategy. The validation on two public datasets and one real-world experiment demonstrates significant accuracy improvements in IMU measurements, reducing errors in both pure IMU integration and visual-inertial odometry.
comment: Accepted by Robotics: Science and Systems, 2025
☆ A highly maneuverable flying squirrel drone with controllable foldable wings IROS
Typical drones with multi rotors are generally less maneuverable due to unidirectional thrust, which may be unfavorable to agile flight in very narrow and confined spaces. This paper suggests a new bio-inspired drone that is empowered with high maneuverability in a lightweight and easy-to-carry way. The proposed flying squirrel inspired drone has controllable foldable wings to cover a wider range of flight attitudes and provide more maneuverable flight capability with stable tracking performance. The wings of a drone are fabricated with silicone membranes and sophisticatedly controlled by reinforcement learning based on human-demonstrated data. Specially, such learning based wing control serves to capture even the complex aerodynamics that are often impossible to model mathematically. It is shown through experiment that the proposed flying squirrel drone intentionally induces aerodynamic drag and hence provides the desired additional repulsive force even under saturated mechanical thrust. This work is very meaningful in demonstrating the potential of biomimicry and machine learning for realizing an animal-like agile drone.
comment: Accepted at 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Project Page : https://jgkang1210.github.io/fsdrone/ , Video : https://youtu.be/Cfc-llDb3_k?si=Cal5beZw6f3HZ2ZW , Jun-Gill Kang and Dohyeon Lee are co-authors
☆ ADDT -- A Digital Twin Framework for Proactive Safety Validation in Autonomous Driving Systems
Autonomous driving systems continue to face safety-critical failures, often triggered by rare and unpredictable corner cases that evade conventional testing. We present the Autonomous Driving Digital Twin (ADDT) framework, a high-fidelity simulation platform designed to proactively identify hidden faults, evaluate real-time performance, and validate safety before deployment. ADDT combines realistic digital models of driving environments, vehicle dynamics, sensor behavior, and fault conditions to enable scalable, scenario-rich stress-testing under diverse and adverse conditions. It supports adaptive exploration of edge cases using reinforcement-driven techniques, uncovering failure modes that physical road testing often misses. By shifting from reactive debugging to proactive simulation-driven validation, ADDT enables a more rigorous and transparent approach to autonomous vehicle safety engineering. To accelerate adoption and facilitate industry-wide safety improvements, the entire ADDT framework has been released as open-source software, providing developers with an accessible and extensible tool for comprehensive safety testing at scale.
☆ snnTrans-DHZ: A Lightweight Spiking Neural Network Architecture for Underwater Image Dehazing
Underwater image dehazing is critical for vision-based marine operations because light scattering and absorption can severely reduce visibility. This paper introduces snnTrans-DHZ, a lightweight Spiking Neural Network (SNN) specifically designed for underwater dehazing. By leveraging the temporal dynamics of SNNs, snnTrans-DHZ efficiently processes time-dependent raw image sequences while maintaining low power consumption. Static underwater images are first converted into time-dependent sequences by repeatedly inputting the same image over user-defined timesteps. These RGB sequences are then transformed into LAB color space representations and processed concurrently. The architecture features three key modules: (i) a K estimator that extracts features from multiple color space representations; (ii) a Background Light Estimator that jointly infers the background light component from the RGB-LAB images; and (iii) a soft image reconstruction module that produces haze-free, visibility-enhanced outputs. The snnTrans-DHZ model is directly trained using a surrogate gradient-based backpropagation through time (BPTT) strategy alongside a novel combined loss function. Evaluated on the UIEB benchmark, snnTrans-DHZ achieves a PSNR of 21.68 dB and an SSIM of 0.8795, and on the EUVP dataset, it yields a PSNR of 23.46 dB and an SSIM of 0.8439. With only 0.5670 million network parameters, and requiring just 7.42 GSOPs and 0.0151 J of energy, the algorithm significantly outperforms existing state-of-the-art methods in terms of efficiency. These features make snnTrans-DHZ highly suitable for deployment in underwater robotics, marine exploration, and environmental monitoring.
♻ ☆ Bi-directional Momentum-based Haptic Feedback and Control System for In-Hand Dexterous Telemanipulation
In-hand dexterous telemanipulation requires not only precise remote motion control of the robot but also effective haptic feedback to the human operator to ensure stable and intuitive interactions between them. Most existing haptic devices for dexterous telemanipulation focus on force feedback and lack effective torque rendering, which is essential for tasks involving object rotation. While some torque feedback solutions in virtual reality applications-such as those based on geared motors or mechanically coupled actuators-have been explored, they often rely on bulky mechanical designs, limiting their use in portable or in-hand applications. In this paper, we propose a Bi-directional Momentum-based Haptic Feedback and Control (Bi-Hap) system that utilizes a palm-sized momentum-actuated mechanism to enable real-time haptic and torque feedback. The Bi-Hap system also integrates an Inertial Measurement Unit (IMU) to extract the human's manipulation command to establish a closed-loop learning-based telemanipulation framework. Furthermore, an error-adaptive feedback strategy is introduced to enhance operator perception and task performance in different error categories. Experimental evaluations demonstrate that Bi-Hap achieved feedback capability with low command following latency (Delay < 0.025 s) and highly accurate torque feedback (RMSE < 0.010 Nm).
comment: This work has been submitted to the IEEE for possible publication
♻ ☆ FUSION: Frequency-guided Underwater Spatial Image recOnstructioN
Underwater images suffer from severe degradations, including color distortions, reduced visibility, and loss of structural details due to wavelength-dependent attenuation and scattering. Existing enhancement methods primarily focus on spatial-domain processing, neglecting the frequency domain's potential to capture global color distributions and long-range dependencies. To address these limitations, we propose FUSION, a dual-domain deep learning framework that jointly leverages spatial and frequency domain information. FUSION independently processes each RGB channel through multi-scale convolutional kernels and adaptive attention mechanisms in the spatial domain, while simultaneously extracting global structural information via FFT-based frequency attention. A Frequency Guided Fusion module integrates complementary features from both domains, followed by inter-channel fusion and adaptive channel recalibration to ensure balanced color distributions. Extensive experiments on benchmark datasets (UIEB, EUVP, SUIM-E) demonstrate that FUSION achieves state-of-the-art performance, consistently outperforming existing methods in reconstruction fidelity (highest PSNR of 23.717 dB and SSIM of 0.883 on UIEB), perceptual quality (lowest LPIPS of 0.112 on UIEB), and visual enhancement metrics (best UIQM of 3.414 on UIEB), while requiring significantly fewer parameters (0.28M) and lower computational complexity, demonstrating its suitability for real-time underwater imaging applications.
♻ ☆ Hybrid Feedback Control for Global Navigation with Locally Optimal Obstacle Avoidance in n-Dimensional Spaces
We present a hybrid feedback control framework for autonomous robot navigation in n-dimensional Euclidean spaces cluttered with spherical obstacles. The proposed approach ensures safe navigation and global asymptotic stability (GAS) of the target location by dynamically switching between two operational modes: motion-to-destination and locally optimal obstacle-avoidance. It produces continuous velocity inputs, ensures collision-free trajectories and generates locally optimal obstacle avoidance maneuvers. Unlike existing methods, the proposed framework is compatible with range sensors, enabling navigation in both a priori known and unknown environments. Extensive simulations in 2D and 3D settings, complemented by experimental validation on a TurtleBot 4 platform, confirm the efficacy and robustness of the approach. Our results demonstrate shorter paths and smoother trajectories compared to state-of-the-art methods, while maintaining computational efficiency and real-world feasibility.
♻ ☆ Neural-Rendezvous: Provably Robust Guidance and Control to Encounter Interstellar Objects
Interstellar objects (ISOs) are likely representatives of primitive materials invaluable in understanding exoplanetary star systems. Due to their poorly constrained orbits with generally high inclinations and relative velocities, however, exploring ISOs with conventional human-in-the-loop approaches is significantly challenging. This paper presents Neural-Rendezvous -- a deep learning-based guidance and control framework for encountering fast-moving objects, including ISOs, robustly, accurately, and autonomously in real time. It uses pointwise minimum norm tracking control on top of a guidance policy modeled by a spectrally-normalized deep neural network, where its hyperparameters are tuned with a loss function directly penalizing the MPC state trajectory tracking error. We show that Neural-Rendezvous provides a high probability exponential bound on the expected spacecraft delivery error, the proof of which leverages stochastic incremental stability analysis. In particular, it is used to construct a non-negative function with a supermartingale property, explicitly accounting for the ISO state uncertainty and the local nature of nonlinear state estimation guarantees. In numerical simulations, Neural-Rendezvous is demonstrated to satisfy the expected error bound for 100 ISO candidates. This performance is also empirically validated using our spacecraft simulator and in high-conflict and distributed UAV swarm reconfiguration with up to 20 UAVs.
comment: Preprint Version, Accepted: October, 2024 (One-minute YouTube summary: https://youtu.be/q3e0LYS2IYQ, DOI: https://doi.org/10.2514/1.G007671)
♻ ☆ Communication and Energy-Aware Multi-UAV Coverage Path Planning for Networked Operations
This paper presents a communication and energy-aware Multi-UAV Coverage Path Planning (mCPP) method for scenarios requiring continuous inter-UAV communication, such as cooperative search and rescue and surveillance missions. Unlike existing mCPP solutions that focus on energy, time, or coverage efficiency, our approach generates coverage paths that require minimal the communication range to maintain inter-UAV connectivity while also optimizing energy consumption. The mCPP problem is formulated as a multi-objective optimization task, aiming to minimize both the communication range requirement and energy consumption. Our approach significantly reduces the communication range needed for maintaining connectivity while ensuring energy efficiency, outperforming state-of-the-art methods. Its effectiveness is validated through simulations on complex and arbitrary shaped regions of interests, including scenarios with no-fly zones. Additionally, real-world experiment demonstrate its high accuracy, achieving 99\% consistency between the estimated and actual communication range required during a multi-UAV coverage mission involving three UAVs.
comment: 9 pages, 5 figures, supplementary material: video and code
♻ ☆ Diffusion-Based Approximate MPC: Fast and Consistent Imitation of Multi-Modal Action Distributions
Approximating model predictive control (MPC) using imitation learning (IL) allows for fast control without solving expensive optimization problems online. However, methods that use neural networks in a simple L2-regression setup fail to approximate multi-modal (set-valued) solution distributions caused by local optima found by the numerical solver or non-convex constraints, such as obstacles, significantly limiting the applicability of approximate MPC in practice. We solve this issue by using diffusion models to accurately represent the complete solution distribution (i.e., all modes) at high control rates (more than 1000 Hz). This work shows that diffusion based AMPC significantly outperforms L2-regression-based approximate MPC for multi-modal action distributions. In contrast to most earlier work on IL, we also focus on running the diffusion-based controller at a higher rate and in joint space instead of end-effector space. Additionally, we propose the use of gradient guidance during the denoising process to consistently pick the same mode in closed loop to prevent switching between solutions. We propose using the cost and constraint satisfaction of the original MPC problem during parallel sampling of solutions from the diffusion model to pick a better mode online. We evaluate our method on the fast and accurate control of a 7-DoF robot manipulator both in simulation and on hardware deployed at 250 Hz, achieving a speedup of more than 70 times compared to solving the MPC problem online and also outperforming the numerical optimization (used for training) in success ratio. Project website: https://paumarquez.github.io/diffusion-ampc.
♻ ☆ A ROS2-based software library for inverse dynamics computation
Inverse dynamics computation is a critical component in robot control, planning and simulation, enabling the calculation of joint torques required to achieve a desired motion. This paper presents a ROS2-based software library designed to solve the inverse dynamics problem for robotic systems. The library is built around an abstract class with three concrete implementations: one for simulated robots and two for real UR10 and Franka robots. This contribution aims to provide a flexible, extensible, robot-agnostic solution to inverse dynamics, suitable for both simulation and real-world scenarios involving planning and control applications. The related software is available at https://github.com/unisa-acg/inverse-dynamics-solver/tree/rap.
comment: 6 pages, 8 figures
♻ ☆ Weakly-Supervised Learning via Multi-Lateral Decoder Branching for Tool Segmentation in Robot-Assisted Cardiovascular Catheterization
Robot-assisted catheterization has garnered a good attention for its potentials in treating cardiovascular diseases. However, advancing surgeon-robot collaboration still requires further research, particularly on task-specific automation. For instance, automated tool segmentation can assist surgeons in visualizing and tracking of endovascular tools during cardiac procedures. While learning-based models have demonstrated state-of-the-art segmentation performances, generating ground-truth labels for fully-supervised methods is both labor-intensive time consuming, and costly. In this study, we propose a weakly-supervised learning method with multi-lateral pseudo labeling for tool segmentation in cardiovascular angiogram datasets. The method utilizes a modified U-Net architecture featuring one encoder and multiple laterally branched decoders. The decoders generate diverse pseudo labels under different perturbations, augmenting available partial labels. The pseudo labels are self-generated using a mixed loss function with shared consistency across the decoders. The weakly-supervised model was trained end-to-end and validated using partially annotated angiogram data from three cardiovascular catheterization procedures. Validation results show that the model could perform closer to fully-supervised models. Also, the proposed weakly-supervised multi-lateral method outperforms three well known methods used for weakly-supervised learning, offering the highest segmentation performance across the three angiogram datasets. Furthermore, numerous ablation studies confirmed the model's consistent performance under different parameters. Finally, the model was applied for tool segmentation in a robot-assisted catheterization experiments. The model enhanced visualization with high connectivity indices for guidewire and catheter, and a mean processing time of 35 ms per frame.
♻ ☆ A Visual-Inertial Motion Prior SLAM for Dynamic Environments
The Visual-Inertial Simultaneous Localization and Mapping (VI-SLAM) algorithms which are mostly based on static assumption are widely used in fields such as robotics, UAVs, VR, and autonomous driving. To overcome the localization risks caused by dynamic landmarks in most VI-SLAM systems, a robust visual-inertial motion prior SLAM system, named IDY-VINS, is proposed in this paper which effectively handles dynamic landmarks using inertial motion prior for dynamic environments to varying degrees. Specifically, potential dynamic landmarks are preprocessed during the feature tracking phase by the probabilistic model of landmarks' minimum projection errors which are obtained from inertial motion prior and epipolar constraint. Subsequently, a robust and self-adaptive bundle adjustment residual is proposed considering the minimum projection error prior for dynamic candidate landmarks. This residual is integrated into a sliding window based nonlinear optimization process to estimate camera poses, IMU states and landmark positions while minimizing the impact of dynamic candidate landmarks that deviate from the motion prior. Finally, a clean point cloud map without `ghosting effect' is obtained that contains only static landmarks. Experimental results demonstrate that our proposed system outperforms state-of-the-art methods in terms of localization accuracy and time cost by robustly mitigating the influence of dynamic landmarks.
♻ ☆ GraspGF: Learning Score-based Grasping Primitive for Human-assisting Dexterous Grasping NeurIPS 2023
The use of anthropomorphic robotic hands for assisting individuals in situations where human hands may be unavailable or unsuitable has gained significant importance. In this paper, we propose a novel task called human-assisting dexterous grasping that aims to train a policy for controlling a robotic hand's fingers to assist users in grasping objects. Unlike conventional dexterous grasping, this task presents a more complex challenge as the policy needs to adapt to diverse user intentions, in addition to the object's geometry. We address this challenge by proposing an approach consisting of two sub-modules: a hand-object-conditional grasping primitive called Grasping Gradient Field~(GraspGF), and a history-conditional residual policy. GraspGF learns `how' to grasp by estimating the gradient from a success grasping example set, while the residual policy determines `when' and at what speed the grasping action should be executed based on the trajectory history. Experimental results demonstrate the superiority of our proposed method compared to baselines, highlighting the user-awareness and practicality in real-world applications. The codes and demonstrations can be viewed at "https://sites.google.com/view/graspgf".
comment: NeurIPS 2023
Information Retrieval 10
☆ Iterative Self-Training for Code Generation via Reinforced Re-Ranking ECIR 2025
Generating high-quality code that solves complex programming tasks is challenging, especially with current decoder-based models that produce highly stochastic outputs. In code generation, even minor errors can easily break the entire solution. Leveraging multiple sampled solutions can significantly improve the overall output quality. One effective way to enhance code generation is by pairing a code generation model with a reranker model, which selects the best solution from the generated samples. We propose a novel iterative self-training approach for self-training reranker models using Proximal Policy Optimization (PPO), aimed at improving both reranking accuracy and the overall code generation process. Unlike traditional PPO approaches, where the focus is on optimizing a generative model with a reward model, our approach emphasizes the development of a robust reward/reranking model. This model improves the quality of generated code through reranking and addresses problems and errors that the reward model might overlook during PPO alignment with the reranker. Our method iteratively refines the training dataset by re-evaluating outputs, identifying high-scoring negative examples, and incorporating them into the training loop, that boosting model performance. Our evaluation on the MultiPL-E dataset demonstrates that our 13.4B parameter model outperforms a 33B model in code generation quality while being three times faster. Moreover, it achieves performance comparable to GPT-4 and surpasses it in one programming language.
comment: Published at ECIR 2025
☆ Outage Probability Analysis for OTFS with Finite Blocklength
Orthogonal time frequency space (OTFS) modulation is widely acknowledged as a prospective waveform for future wireless communication networks.To provide insights for the practical system design, this paper analyzes the outage probability of OTFS modulation with finite blocklength.To begin with, we present the system model and formulate the analysis of outage probability for OTFS with finite blocklength as an equivalent problem of calculating the outage probability with finite blocklength over parallel additive white Gaussian noise (AWGN) channels.Subsequently, we apply the equivalent noise approach to derive a lower bound on the outage probability of OTFS with finite blocklength under both average power allocation and water-filling power allocation strategies, respectively.Finally, the lower bounds of the outage probability are determined using the Monte-Carlo method for the two power allocation strategies.The impact of the number of resolvable paths and coding rates on the outage probability is analyzed, and the simulation results are compared with the theoretical lower bounds.
☆ Slow Thinking for Sequential Recommendation
To develop effective sequential recommender systems, numerous methods have been proposed to model historical user behaviors. Despite the effectiveness, these methods share the same fast thinking paradigm. That is, for making recommendations, these methods typically encodes user historical interactions to obtain user representations and directly match these representations with candidate item representations. However, due to the limited capacity of traditional lightweight recommendation models, this one-step inference paradigm often leads to suboptimal performance. To tackle this issue, we present a novel slow thinking recommendation model, named STREAM-Rec. Our approach is capable of analyzing historical user behavior, generating a multi-step, deliberative reasoning process, and ultimately delivering personalized recommendations. In particular, we focus on two key challenges: (1) identifying the suitable reasoning patterns in recommender systems, and (2) exploring how to effectively stimulate the reasoning capabilities of traditional recommenders. To this end, we introduce a three-stage training framework. In the first stage, the model is pretrained on large-scale user behavior data to learn behavior patterns and capture long-range dependencies. In the second stage, we design an iterative inference algorithm to annotate suitable reasoning traces by progressively refining the model predictions. This annotated data is then used to fine-tune the model. Finally, in the third stage, we apply reinforcement learning to further enhance the model generalization ability. Extensive experiments validate the effectiveness of our proposed method.
☆ Revisiting Self-Attentive Sequential Recommendation
Recommender systems are ubiquitous in on-line services to drive businesses. And many sequential recommender models were deployed in these systems to enhance personalization. The approach of using the transformer decoder as the sequential recommender was proposed years ago and is still a strong baseline in recent works. But this kind of sequential recommender model did not scale up well, compared to language models. Quite some details in the classical self-attentive sequential recommender model could be revisited, and some new experiments may lead to new findings, without changing the general model structure which was the focus of many previous works. In this paper, we show the details and propose new experiment methodologies for future research on sequential recommendation, in hope to motivate further exploration to new findings in this area.
comment: 4 pages, 0 figure, technical report based on https://github.com/pmixer/SASRec.pytorch experiment findings
☆ HD-RAG: Retrieval-Augmented Generation for Hybrid Documents Containing Text and Hierarchical Tables
With the rapid advancement of large language models (LLMs), Retrieval-Augmented Generation (RAG) effectively combines LLMs generative capabilities with external retrieval-based information. The Hybrid Document RAG task aims to integrate textual and hierarchical tabular data for more comprehensive retrieval and generation in complex scenarios. However, there is no existing dataset specifically designed for this task that includes both text and tabular data. Additionally, existing methods struggle to retrieve relevant tabular data and integrate it with text. Semantic similarity-based retrieval lacks accuracy, while table-specific methods fail to handle complex hierarchical structures effectively. Furthermore, the QA task requires complex reasoning and calculations, further complicating the challenge. In this paper, we propose a new large-scale dataset, DocRAGLib, specifically designed for the question answering (QA) task scenario under Hybrid Document RAG. To tackle these challenges, we introduce HD-RAG, a novel framework that incorporates a row-and-column level (RCL) table representation, employs a two-stage process combining ensemble and LLM-based retrieval, and integrates RECAP, which is designed for multi-step reasoning and complex calculations in Document-QA tasks. We conduct comprehensive experiments with DocRAGLib, showing that HD-RAG outperforms existing baselines in both retrieval accuracy and QA performance, demonstrating its effectiveness.
comment: 10 pages, 4 figures
☆ FROG: Effective Friend Recommendation in Online Games via Modality-aware User Preferences SIGIR 2025
Due to the convenience of mobile devices, the online games have become an important part for user entertainments in reality, creating a demand for friend recommendation in online games. However, none of existing approaches can effectively incorporate the multi-modal user features (\emph{e.g.}, images and texts) with the structural information in the friendship graph, due to the following limitations: (1) some of them ignore the high-order structural proximity between users, (2) some fail to learn the pairwise relevance between users at modality-specific level, and (3) some cannot capture both the local and global user preferences on different modalities. By addressing these issues, in this paper, we propose an end-to-end model \textsc{FROG} that better models the user preferences on potential friends. Comprehensive experiments on both offline evaluation and online deployment at \kw{Tencent} have demonstrated the superiority of \textsc{FROG} over existing approaches.
comment: Accepted in SIGIR 2025
☆ Integrating Textual Embeddings from Contrastive Learning with Generative Recommender for Enhanced Personalization
Recent advances in recommender systems have highlighted the complementary strengths of generative modeling and pretrained language models. We propose a hybrid framework that augments the Hierarchical Sequential Transduction Unit (HSTU) generative recommender with BLaIR -- a contrastive text embedding model. This integration enriches item representations with semantic signals from textual metadata while preserving HSTU's powerful sequence modeling capabilities. We evaluate our method on two domains from the Amazon Reviews 2023 dataset, comparing it against the original HSTU and a variant that incorporates embeddings from OpenAI's state-of-the-art text-embedding-3-large model. While the OpenAI embedding model is likely trained on a substantially larger corpus with significantly more parameters, our lightweight BLaIR-enhanced approach -- pretrained on domain-specific data -- consistently achieves better performance, highlighting the effectiveness of contrastive text embeddings in compute-efficient settings.
comment: Code available at https://www.github.com/snapfinger/HSTU-BLaIR
☆ Multi-Modal Hypergraph Enhanced LLM Learning for Recommendation
The burgeoning presence of Large Language Models (LLM) is propelling the development of personalized recommender systems. Most existing LLM-based methods fail to sufficiently explore the multi-view graph structure correlations inherent in recommendation scenarios. To this end, we propose a novel framework, Hypergraph Enhanced LLM Learning for multimodal Recommendation (HeLLM), designed to equip LLMs with the capability to capture intricate higher-order semantic correlations by fusing graph-level contextual signals with sequence-level behavioral patterns. In the recommender pre-training phase, we design a user hypergraph to uncover shared interest preferences among users and an item hypergraph to capture correlations within multimodal similarities among items. The hypergraph convolution and synergistic contrastive learning mechanism are introduced to enhance the distinguishability of learned representations. In the LLM fine-tuning phase, we inject the learned graph-structured embeddings directly into the LLM's architecture and integrate sequential features capturing each user's chronological behavior. This process enables hypergraphs to leverage graph-structured information as global context, enhancing the LLM's ability to perceive complex relational patterns and integrate multimodal information, while also modeling local temporal dynamics. Extensive experiments demonstrate the superiority of our proposed method over state-of-the-art baselines, confirming the advantages of fusing hypergraph-based context with sequential user behavior in LLMs for recommendation.
comment: 10 pages, 4 figures
☆ Distilling Transitional Pattern to Large Language Models for Multimodal Session-based Recommendation
Session-based recommendation (SBR) predicts the next item based on anonymous sessions. Traditional SBR explores user intents based on ID collaborations or auxiliary content. To further alleviate data sparsity and cold-start issues, recent Multimodal SBR (MSBR) methods utilize simplistic pre-trained models for modality learning but have limitations in semantic richness. Considering semantic reasoning abilities of Large Language Models (LLM), we focus on the LLM-enhanced MSBR scenario in this paper, which leverages LLM cognition for comprehensive multimodal representation generation, to enhance downstream MSBR. Tackling this problem faces two challenges: i) how to obtain LLM cognition on both transitional patterns and inherent multimodal knowledge, ii) how to align both features into one unified LLM, minimize discrepancy while maximizing representation utility. To this end, we propose a multimodal LLM-enhanced framework TPAD, which extends a distillation paradigm to decouple and align transitional patterns for promoting MSBR. TPAD establishes parallel Knowledge-MLLM and Transfer-MLLM, where the former interprets item knowledge-reflected features and the latter extracts transition-aware features underneath sessions. A transitional pattern alignment module harnessing mutual information estimation theory unites two MLLMs, alleviating distribution discrepancy and distilling transitional patterns into modal representations. Extensive experiments on real-world datasets demonstrate the effectiveness of our framework.
♻ ☆ Table Integration in Data Lakes Unleashed: Pairwise Integrability Judgment, Integrable Set Discovery, and Multi-Tuple Conflict Resolution VLDB
Table integration aims to create a comprehensive table by consolidating tuples containing relevant information. In this work, we investigate the challenge of integrating multiple tables from a data lake, focusing on three core tasks: 1) pairwise integrability judgment, which determines whether a tuple pair is integrable, accounting for any occurrences of semantic equivalence or typographical errors; 2) integrable set discovery, which identifies all integrable sets in a table based on pairwise integrability judgments established in the first task; 3) multi-tuple conflict resolution, which resolves conflicts between multiple tuples during integration. To this end, we train a binary classifier to address the task of pairwise integrability judgment. Given the scarcity of labeled data in data lakes, we propose a self-supervised adversarial contrastive learning algorithm to perform classification, which incorporates data augmentation methods and adversarial examples to autonomously generate new training data. Upon the output of pairwise integrability judgment, each integrable set can be considered as a community, a densely connected sub-graph where nodes and edges correspond to tuples in the table and their pairwise integrability respectively, we proceed to investigate various community detection algorithms to address the integrable set discovery objective. Moving forward to tackle multi-tuple conflict resolution, we introduce an innovative in-context learning methodology. This approach capitalizes on the knowledge embedded within large language models (LLMs) to effectively resolve conflicts that arise when integrating multiple tuples. Notably, our method minimizes the need for annotated data, making it particularly suited for scenarios where labeled datasets are scarce.
comment: This paper is accepted by VLDB Journal
Robotics 19
☆ DoorBot: Closed-Loop Task Planning and Manipulation for Door Opening in the Wild with Haptic Feedback ICRA 2025
Robots operating in unstructured environments face significant challenges when interacting with everyday objects like doors. They particularly struggle to generalize across diverse door types and conditions. Existing vision-based and open-loop planning methods often lack the robustness to handle varying door designs, mechanisms, and push/pull configurations. In this work, we propose a haptic-aware closed-loop hierarchical control framework that enables robots to explore and open different unseen doors in the wild. Our approach leverages real-time haptic feedback, allowing the robot to adjust its strategy dynamically based on force feedback during manipulation. We test our system on 20 unseen doors across different buildings, featuring diverse appearances and mechanical types. Our framework achieves a 90% success rate, demonstrating its ability to generalize and robustly handle varied door-opening tasks. This scalable solution offers potential applications in broader open-world articulated object manipulation tasks.
comment: In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA 2025)
☆ Adaptive Planning Framework for UAV-Based Surface Inspection in Partially Unknown Indoor Environments
Inspecting indoor environments such as tunnels, industrial facilities, and construction sites is essential for infrastructure monitoring and maintenance. While manual inspection in these environments is often time-consuming and potentially hazardous, Unmanned Aerial Vehicles (UAVs) can improve efficiency by autonomously handling inspection tasks. Such inspection tasks usually rely on reference maps for coverage planning. However, in industrial applications, only the floor plans are typically available. The unforeseen obstacles not included in the floor plans will result in outdated reference maps and inefficient or unsafe inspection trajectories. In this work, we propose an adaptive inspection framework that integrates global coverage planning with local reactive adaptation to improve the coverage and efficiency of UAV-based inspection in partially unknown indoor environments. Experimental results in structured indoor scenarios demonstrate the effectiveness of the proposed approach in inspection efficiency and achieving high coverage rates with adaptive obstacle handling, highlighting its potential for enhancing the efficiency of indoor facility inspection.
☆ REALM: Real-Time Estimates of Assistance for Learned Models in Human-Robot Interaction
There are a variety of mechanisms (i.e., input types) for real-time human interaction that can facilitate effective human-robot teaming. For example, previous works have shown how teleoperation, corrective, and discrete (i.e., preference over a small number of choices) input can enable robots to complete complex tasks. However, few previous works have looked at combining different methods, and in particular, opportunities for a robot to estimate and elicit the most effective form of assistance given its understanding of a task. In this paper, we propose a method for estimating the value of different human assistance mechanisms based on the action uncertainty of a robot policy. Our key idea is to construct mathematical expressions for the expected post-interaction differential entropy (i.e., uncertainty) of a stochastic robot policy to compare the expected value of different interactions. As each type of human input imposes a different requirement for human involvement, we demonstrate how differential entropy estimates can be combined with a likelihood penalization approach to effectively balance feedback informational needs with the level of required input. We demonstrate evidence of how our approach interfaces with emergent learning models (e.g., a diffusion model) to produce accurate assistance value estimates through both simulation and a robot user study. Our user study results indicate that the proposed approach can enable task completion with minimal human feedback for uncertain robot behaviors.
comment: IEEE Robotics and Automation Letters
☆ Development of a PPO-Reinforcement Learned Walking Tripedal Soft-Legged Robot using SOFA
Rigid robots were extensively researched, whereas soft robotics remains an underexplored field. Utilizing soft-legged robots in performing tasks as a replacement for human beings is an important stride to take, especially under harsh and hazardous conditions over rough terrain environments. For the demand to teach any robot how to behave in different scenarios, a real-time physical and visual simulation is essential. When it comes to soft robots specifically, a simulation framework is still an arduous problem that needs to be disclosed. Using the simulation open framework architecture (SOFA) is an advantageous step. However, neither SOFA's manual nor prior public SOFA projects show its maximum capabilities the users can reach. So, we resolved this by establishing customized settings and handling the framework components appropriately. Settling on perfect, fine-tuned SOFA parameters has stimulated our motivation towards implementing the state-of-the-art (SOTA) reinforcement learning (RL) method of proximal policy optimization (PPO). The final representation is a well-defined, ready-to-deploy walking, tripedal, soft-legged robot based on PPO-RL in a SOFA environment. Robot navigation performance is a key metric to be considered for measuring the success resolution. Although in the simulated soft robots case, an 82\% success rate in reaching a single goal is a groundbreaking output, we pushed the boundaries to further steps by evaluating the progress under assigning a sequence of goals. While trailing the platform steps, outperforming discovery has been observed with an accumulative squared error deviation of 19 mm. The full code is publicly available at \href{https://github.com/tarekshohdy/PPO_SOFA_Soft_Legged_Robot.git}{github.com/tarekshohdy/PPO$\textunderscore$SOFA$\textunderscore$Soft$\textunderscore$Legged$\textunderscore$ Robot.git}
☆ Concurrent-Allocation Task Execution for Multi-Robot Path-Crossing-Minimal Navigation in Obstacle Environments
Reducing undesirable path crossings among trajectories of different robots is vital in multi-robot navigation missions, which not only reduces detours and conflict scenarios, but also enhances navigation efficiency and boosts productivity. Despite recent progress in multi-robot path-crossing-minimal (MPCM) navigation, the majority of approaches depend on the minimal squared-distance reassignment of suitable desired points to robots directly. However, if obstacles occupy the passing space, calculating the actual robot-point distances becomes complex or intractable, which may render the MPCM navigation in obstacle environments inefficient or even infeasible. In this paper, the concurrent-allocation task execution (CATE) algorithm is presented to address this problem (i.e., MPCM navigation in obstacle environments). First, the path-crossing-related elements in terms of (i) robot allocation, (ii) desired-point convergence, and (iii) collision and obstacle avoidance are encoded into integer and control barrier function (CBF) constraints. Then, the proposed constraints are used in an online constrained optimization framework, which implicitly yet effectively minimizes the possible path crossings and trajectory length in obstacle environments by minimizing the desired point allocation cost and slack variables in CBF constraints simultaneously. In this way, the MPCM navigation in obstacle environments can be achieved with flexible spatial orderings. Note that the feasibility of solutions and the asymptotic convergence property of the proposed CATE algorithm in obstacle environments are both guaranteed, and the calculation burden is also reduced by concurrently calculating the optimal allocation and the control input directly without the path planning process.
☆ Compliant Explicit Reference Governor for Contact Friendly Robotic Manipulators
This paper introduces the Compliant Explicit Reference Governor (C-ERG), an extension of the Explicit Reference Governor that allows the robot to operate safely while in contact with the environment. The C-ERG is an intermediate layer that can be placed between a high-level planner and a low-level controller: its role is to enforce operational constraints and to enable the smooth transition between free-motion and contact operations. The C-ERG ensures safety by limiting the total energy available to the robotic arm at the time of contact. In the absence of contact, however, the C-ERG does not penalize the system performance. Numerical examples showcase the behavior of the C-ERG for increasingly complex systems.
☆ Steady-State Drifting Equilibrium Analysis of Single-Track Two-Wheeled Robots for Controller Design
Drifting is an advanced driving technique where the wheeled robot's tire-ground interaction breaks the common non-holonomic pure rolling constraint. This allows high-maneuverability tasks like quick cornering, and steady-state drifting control enhances motion stability under lateral slip conditions. While drifting has been successfully achieved in four-wheeled robot systems, its application to single-track two-wheeled (STTW) robots, such as unmanned motorcycles or bicycles, has not been thoroughly studied. To bridge this gap, this paper extends the drifting equilibrium theory to STTW robots and reveals the mechanism behind the steady-state drifting maneuver. Notably, the counter-steering drifting technique used by skilled motorcyclists is explained through this theory. In addition, an analytical algorithm based on intrinsic geometry and kinematics relationships is proposed, reducing the computation time by four orders of magnitude while maintaining less than 6% error compared to numerical methods. Based on equilibrium analysis, a model predictive controller (MPC) is designed to achieve steady-state drifting and equilibrium points transition, with its effectiveness and robustness validated through simulations.
☆ Haptic Perception via the Dynamics of Flexible Body Inspired by an Ostrich's Neck
In biological systems, haptic perception is achieved through both flexible skin and flexible body. In fully soft robots, the fragility of their bodies and the time delays in sensory processing pose significant challenges. The musculoskeletal system possesses both the deformability inherent in soft materials and the durability of rigid-body robots. Additionally, by outsourcing part of the intelligent information processing to the morphology of the musculoskeletal system, applications for dynamic tasks are expected. This study focuses on the pecking movements of birds, which achieve precise haptic perception through the musculoskeletal system of their flexible neck. Physical reservoir computing is applied to flexible structures inspired by an ostrich neck to analyze the relationship between haptic perception and physical characteristics. Combined experiments using both an actual robot and simulations demonstrate that, under appropriate body viscoelasticity, the flexible structure can distinguish objects of varying softness and memorize this information as behaviors. Drawing on these findings and anatomical insights from the ostrich neck, a haptic sensing system is proposed that possesses separability and this behavioral memory in flexible structures, enabling rapid learning and real-time inference. The results demonstrate that through the dynamics of flexible structures, diverse functions can emerge beyond their original design as manipulators.
comment: This paper includes a figure of a dissected ostrich. As the ostrich was processed for food, its use does not raise any ethical concerns
☆ IMPACT: Behavioral Intention-aware Multimodal Trajectory Prediction with Adaptive Context Trimming
While most prior research has focused on improving the precision of multimodal trajectory predictions, the explicit modeling of multimodal behavioral intentions (e.g., yielding, overtaking) remains relatively underexplored. This paper proposes a unified framework that jointly predicts both behavioral intentions and trajectories to enhance prediction accuracy, interpretability, and efficiency. Specifically, we employ a shared context encoder for both intention and trajectory predictions, thereby reducing structural redundancy and information loss. Moreover, we address the lack of ground-truth behavioral intention labels in mainstream datasets (Waymo, Argoverse) by auto-labeling these datasets, thus advancing the community's efforts in this direction. We further introduce a vectorized occupancy prediction module that infers the probability of each map polyline being occupied by the target vehicle's future trajectory. By leveraging these intention and occupancy prediction priors, our method conducts dynamic, modality-dependent pruning of irrelevant agents and map polylines in the decoding stage, effectively reducing computational overhead and mitigating noise from non-critical elements. Our approach ranks first among LiDAR-free methods on the Waymo Motion Dataset and achieves first place on the Waymo Interactive Prediction Dataset. Remarkably, even without model ensembling, our single-model framework improves the soft mean average precision (softmAP) by 10 percent compared to the second-best method in the Waymo Interactive Prediction Leaderboard. Furthermore, the proposed framework has been successfully deployed on real vehicles, demonstrating its practical effectiveness in real-world applications.
comment: under review
☆ agriFrame: Agricultural framework to remotely control a rover inside a greenhouse environment
The growing demand for innovation in agriculture is essential for food security worldwide and more implicit in developing countries. With growing demand comes a reduction in rapid development time. Data collection and analysis are essential in agriculture. However, considering a given crop, its cycle comes once a year, and researchers must wait a few months before collecting more data for the given crop. To overcome this hurdle, researchers are venturing into digital twins for agriculture. Toward this effort, we present an agricultural framework(agriFrame). Here, we introduce a simulated greenhouse environment for testing and controlling a robot and remotely controlling/implementing the algorithms in the real-world greenhouse setup. This work showcases the importance/interdependence of network setup, remotely controllable rover, and messaging protocol. The sophisticated yet simple-to-use agriFrame has been optimized for the simulator on minimal laptop/desktop specifications.
☆ Multi-Robot Coordination with Adversarial Perception
This paper investigates the resilience of perception-based multi-robot coordination with wireless communication to online adversarial perception. A systematic study of this problem is essential for many safety-critical robotic applications that rely on the measurements from learned perception modules. We consider a (small) team of quadrotor robots that rely only on an Inertial Measurement Unit (IMU) and the visual data measurements obtained from a learned multi-task perception module (e.g., object detection) for downstream tasks, including relative localization and coordination. We focus on a class of adversarial perception attacks that cause misclassification, mislocalization, and latency. We propose that the effects of adversarial misclassification and mislocalization can be modeled as sporadic (intermittent) and spurious measurement data for the downstream tasks. To address this, we present a framework for resilience analysis of multi-robot coordination with adversarial measurements. The framework integrates data from Visual-Inertial Odometry (VIO) and the learned perception model for robust relative localization and state estimation in the presence of adversarially sporadic and spurious measurements. The framework allows for quantifying the degradation in system observability and stability in relation to the success rate of adversarial perception. Finally, experimental results on a multi-robot platform demonstrate the real-world applicability of our methodology for resource-constrained robotic platforms.
comment: to appear at the 2025 Int'l Conference on Unmanned Aircraft Systems (ICUAS)
☆ Nonconvex Obstacle Avoidance using Efficient Sampling-Based Distance Functions
We consider nonconvex obstacle avoidance where a robot described by nonlinear dynamics and a nonconvex shape has to avoid nonconvex obstacles. Obstacle avoidance is a fundamental problem in robotics and well studied in control. However, existing solutions are computationally expensive (e.g., model predictive controllers), neglect nonlinear dynamics (e.g., graph-based planners), use diffeomorphic transformations into convex domains (e.g., for star shapes), or are conservative due to convex overapproximations. The key challenge here is that the computation of the distance between the shapes of the robot and the obstacles is a nonconvex problem. We propose efficient computation of this distance via sampling-based distance functions. We quantify the sampling error and show that, for certain systems, such sampling-based distance functions are valid nonsmooth control barrier functions. We also study how to deal with disturbances on the robot dynamics in our setting. Finally, we illustrate our method on a robot navigation task involving an omnidirectional robot and nonconvex obstacles. We also analyze performance and computational efficiency of our controller as a function of the number of samples.
♻ ☆ Training Human-Robot Teams by Improving Transparency Through a Virtual Spectator Interface ICRA 2025
After-action reviews (AARs) are professional discussions that help operators and teams enhance their task performance by analyzing completed missions with peers and professionals. Previous studies that compared different formats of AARs have mainly focused on human teams. However, the inclusion of robotic teammates brings along new challenges in understanding teammate intent and communication. Traditional AAR between human teammates may not be satisfactory for human-robot teams. To address this limitation, we propose a new training review (TR) tool, called the Virtual Spectator Interface (VSI), to enhance human-robot team performance and situational awareness (SA) in a simulated search mission. The proposed VSI primarily utilizes visual feedback to review subjects' behavior. To examine the effectiveness of VSI, we took elements from AAR to conduct our own TR, designed a 1 x 3 between-subjects experiment with experimental conditions: TR with (1) VSI, (2) screen recording, and (3) non-technology (only verbal descriptions). The results of our experiments demonstrated that the VSI did not result in significantly better team performance than other conditions. However, the TR with VSI led to more improvement in the subjects SA over the other conditions.
comment: 7 pages, 4 figures, Accepted to ICRA 2025
♻ ☆ Automating Transfer of Robot Task Plans using Functorial Data Migrations
This paper introduces a novel approach to ontology-based robot plan transfer by leveraging functorial data migrations, a structured mapping method derived from category theory. Functors provide structured maps between planning domain ontologies which enables the transfer of task plans without the need for replanning. Unlike methods tailored to specific plans, our framework applies universally within the source domain once a structured map is defined. We demonstrate this approach by transferring a task plan from the canonical Blocksworld domain to one compatible with the AI2-THOR Kitchen environment. Additionally, we discuss practical limitations, propose benchmarks for evaluating symbolic plan transfer methods, and outline future directions for scaling this approach.
♻ ☆ Are Open-Vocabulary Models Ready for Detection of MEP Elements on Construction Sites
The construction industry has long explored robotics and computer vision, yet their deployment on construction sites remains very limited. These technologies have the potential to revolutionize traditional workflows by enhancing accuracy, efficiency, and safety in construction management. Ground robots equipped with advanced vision systems could automate tasks such as monitoring mechanical, electrical, and plumbing (MEP) systems. The present research evaluates the applicability of open-vocabulary vision-language models compared to fine-tuned, lightweight, closed-set object detectors for detecting MEP components using a mobile ground robotic platform. A dataset collected with cameras mounted on a ground robot was manually annotated and analyzed to compare model performance. The results demonstrate that, despite the versatility of vision-language models, fine-tuned lightweight models still largely outperform them in specialized environments and for domain-specific tasks.
comment: 4 pages, 3 figures, Accepted for presentation at the 42nd International Symposium on Automation and Robotics in Construction
♻ ☆ A Unified and General Humanoid Whole-Body Controller for Versatile Locomotion
Locomotion is a fundamental skill for humanoid robots. However, most existing works make locomotion a single, tedious, unextendable, and unconstrained movement. This limits the kinematic capabilities of humanoid robots. In contrast, humans possess versatile athletic abilities-running, jumping, hopping, and finely adjusting gait parameters such as frequency and foot height. In this paper, we investigate solutions to bring such versatility into humanoid locomotion and thereby propose HugWBC: a unified and general humanoid whole-body controller for versatile locomotion. By designing a general command space in the aspect of tasks and behaviors, along with advanced techniques like symmetrical loss and intervention training for learning a whole-body humanoid controlling policy in simulation, HugWBC enables real-world humanoid robots to produce various natural gaits, including walking, jumping, standing, and hopping, with customizable parameters such as frequency, foot swing height, further combined with different body height, waist rotation, and body pitch. Beyond locomotion, HugWBC also supports real-time interventions from external upper-body controllers like teleoperation, enabling loco-manipulation with precision under any locomotive behavior. Extensive experiments validate the high tracking accuracy and robustness of HugWBC with/without upper-body intervention for all commands, and we further provide an in-depth analysis of how the various commands affect humanoid movement and offer insights into the relationships between these commands. To our knowledge, HugWBC is the first humanoid whole-body controller that supports such versatile locomotion behaviors with high robustness and flexibility.
comment: Published at RSS 2025. The first two authors contribute equally. Project page: https://hugwbc.github.io/
♻ ☆ Motion Before Action: Diffusing Object Motion as Manipulation Condition
Inferring object motion representations from observations enhances the performance of robotic manipulation tasks. This paper introduces a new paradigm for robot imitation learning that generates action sequences by reasoning about object motion from visual observations. We propose MBA (Motion Before Action), a novel module that employs two cascaded diffusion processes for object motion generation and robot action generation under object motion guidance. MBA first predicts the future pose sequence of the object based on observations, then uses this sequence as a condition to guide robot action generation. Designed as a plug-and-play component, MBA can be flexibly integrated into existing robotic manipulation policies with diffusion action heads. Extensive experiments in both simulated and real-world environments demonstrate that our approach substantially improves the performance of existing policies across a wide range of manipulation tasks. Project page: https://selen-suyue.github.io/MBApage/
♻ ☆ Overlap-Aware Feature Learning for Robust Unsupervised Domain Adaptation for 3D Semantic Segmentation
3D point cloud semantic segmentation (PCSS) is a cornerstone for environmental perception in robotic systems and autonomous driving, enabling precise scene understanding through point-wise classification. While unsupervised domain adaptation (UDA) mitigates label scarcity in PCSS, existing methods critically overlook the inherent vulnerability to real-world perturbations (e.g., snow, fog, rain) and adversarial distortions. This work first identifies two intrinsic limitations that undermine current PCSS-UDA robustness: (a) unsupervised features overlap from unaligned boundaries in shared-class regions and (b) feature structure erosion caused by domain-invariant learning that suppresses target-specific patterns. To address the proposed problems, we propose a tripartite framework consisting of: 1) a robustness evaluation model quantifying resilience against adversarial attack/corruption types through robustness metrics; 2) an invertible attention alignment module (IAAM) enabling bidirectional domain mapping while preserving discriminative structure via attention-guided overlap suppression; and 3) a contrastive memory bank with quality-aware contrastive learning that progressively refines pseudo-labels with feature quality for more discriminative representations. Extensive experiments on SynLiDAR-to-SemanticPOSS adaptation demonstrate a maximum mIoU improvement of 14.3\% under adversarial attack.
comment: 8 pages,6 figures
♻ ☆ Accurate Control under Voltage Drop for Rotor Drones
This letter proposes an anti-disturbance control scheme for rotor drones to counteract voltage drop (VD) disturbance caused by voltage drop of the battery, which is a common case for long-time flight or aggressive maneuvers. Firstly, the refined dynamics of rotor drones considering VD disturbance are presented. Based on the dynamics, a voltage drop observer (VDO) is developed to accurately estimate the VD disturbance by decoupling the disturbance and state information of the drone, reducing the conservativeness of conventional disturbance observers. Subsequently, the control scheme integrates the VDO within the translational loop and a fixed-time sliding mode observer (SMO) within the rotational loop, enabling it to address force and torque disturbances caused by voltage drop of the battery. Sufficient real flight experiments are conducted to demonstrate the effectiveness of the proposed control scheme under VD disturbance.
Information Retrieval 12
☆ Breaking the Lens of the Telescope: Online Relevance Estimation over Large Retrieval Sets
Advanced relevance models, such as those that use large language models (LLMs), provide highly accurate relevance estimations. However, their computational costs make them infeasible for processing large document corpora. To address this, retrieval systems often employ a telescoping approach, where computationally efficient but less precise lexical and semantic retrievers filter potential candidates for further ranking. However, this approach heavily depends on the quality of early-stage retrieval, which can potentially exclude relevant documents early in the process. In this work, we propose a novel paradigm for re-ranking called online relevance estimation that continuously updates relevance estimates for a query throughout the ranking process. Instead of re-ranking a fixed set of top-k documents in a single step, online relevance estimation iteratively re-scores smaller subsets of the most promising documents while adjusting relevance scores for the remaining pool based on the estimations from the final model using an online bandit-based algorithm. This dynamic process mitigates the recall limitations of telescoping systems by re-prioritizing documents initially deemed less relevant by earlier stages -- including those completely excluded by earlier-stage retrievers. We validate our approach on TREC benchmarks under two scenarios: hybrid retrieval and adaptive retrieval. Experimental results demonstrate that our method is sample-efficient and significantly improves recall, highlighting the effectiveness of our online relevance estimation framework for modern search systems.
comment: 11 pages,5 figures, 4 tables
☆ SynthTRIPs: A Knowledge-Grounded Framework for Benchmark Query Generation for Personalized Tourism Recommenders SIGIR 2025
Tourism Recommender Systems (TRS) are crucial in personalizing travel experiences by tailoring recommendations to users' preferences, constraints, and contextual factors. However, publicly available travel datasets often lack sufficient breadth and depth, limiting their ability to support advanced personalization strategies -- particularly for sustainable travel and off-peak tourism. In this work, we explore using Large Language Models (LLMs) to generate synthetic travel queries that emulate diverse user personas and incorporate structured filters such as budget constraints and sustainability preferences. This paper introduces a novel SynthTRIPs framework for generating synthetic travel queries using LLMs grounded in a curated knowledge base (KB). Our approach combines persona-based preferences (e.g., budget, travel style) with explicit sustainability filters (e.g., walkability, air quality) to produce realistic and diverse queries. We mitigate hallucination and ensure factual correctness by grounding the LLM responses in the KB. We formalize the query generation process and introduce evaluation metrics for assessing realism and alignment. Both human expert evaluations and automatic LLM-based assessments demonstrate the effectiveness of our synthetic dataset in capturing complex personalization aspects underrepresented in existing datasets. While our framework was developed and tested for personalized city trip recommendations, the methodology applies to other recommender system domains. Code and dataset are made public at https://bit.ly/synthTRIPs
comment: Accepted for publication at SIGIR 2025
☆ NoTeS-Bank: Benchmarking Neural Transcription and Search for Scientific Notes Understanding
Understanding and reasoning over academic handwritten notes remains a challenge in document AI, particularly for mathematical equations, diagrams, and scientific notations. Existing visual question answering (VQA) benchmarks focus on printed or structured handwritten text, limiting generalization to real-world note-taking. To address this, we introduce NoTeS-Bank, an evaluation benchmark for Neural Transcription and Search in note-based question answering. NoTeS-Bank comprises complex notes across multiple domains, requiring models to process unstructured and multimodal content. The benchmark defines two tasks: (1) Evidence-Based VQA, where models retrieve localized answers with bounding-box evidence, and (2) Open-Domain VQA, where models classify the domain before retrieving relevant documents and answers. Unlike classical Document VQA datasets relying on optical character recognition (OCR) and structured data, NoTeS-BANK demands vision-language fusion, retrieval, and multimodal reasoning. We benchmark state-of-the-art Vision-Language Models (VLMs) and retrieval frameworks, exposing structured transcription and reasoning limitations. NoTeS-Bank provides a rigorous evaluation with NDCG@5, MRR, Recall@K, IoU, and ANLS, establishing a new standard for visual document understanding and reasoning.
☆ Pneuma: Leveraging LLMs for Tabular Data Representation and Retrieval in an End-to-End System SIGMOD 2025
Finding relevant tables among databases, lakes, and repositories is the first step in extracting value from data. Such a task remains difficult because assessing whether a table is relevant to a problem does not always depend only on its content but also on the context, which is usually tribal knowledge known to the individual or team. While tools like data catalogs and academic data discovery systems target this problem, they rely on keyword search or more complex interfaces, limiting non-technical users' ability to find relevant data. The advent of large language models (LLMs) offers a unique opportunity for users to ask questions directly in natural language, making dataset discovery more intuitive, accessible, and efficient. In this paper, we introduce Pneuma, a retrieval-augmented generation (RAG) system designed to efficiently and effectively discover tabular data. Pneuma leverages large language models (LLMs) for both table representation and table retrieval. For table representation, Pneuma preserves schema and row-level information to ensure comprehensive data understanding. For table retrieval, Pneuma augments LLMs with traditional information retrieval techniques, such as full-text and vector search, harnessing the strengths of both to improve retrieval performance. To evaluate Pneuma, we generate comprehensive benchmarks that simulate table discovery workload on six real-world datasets including enterprise data, scientific databases, warehousing data, and open data. Our results demonstrate that Pneuma outperforms widely used table search systems (such as full-text search and state-of-the-art RAG systems) in accuracy and resource efficiency.
comment: SIGMOD 2025 Paper
☆ HeteRAG: A Heterogeneous Retrieval-augmented Generation Framework with Decoupled Knowledge Representations
Retrieval-augmented generation (RAG) methods can enhance the performance of LLMs by incorporating retrieved knowledge chunks into the generation process. In general, the retrieval and generation steps usually have different requirements for these knowledge chunks. The retrieval step benefits from comprehensive information to improve retrieval accuracy, whereas excessively long chunks may introduce redundant contextual information, thereby diminishing both the effectiveness and efficiency of the generation process. However, existing RAG methods typically employ identical representations of knowledge chunks for both retrieval and generation, resulting in suboptimal performance. In this paper, we propose a heterogeneous RAG framework (\myname) that decouples the representations of knowledge chunks for retrieval and generation, thereby enhancing the LLMs in both effectiveness and efficiency. Specifically, we utilize short chunks to represent knowledge to adapt the generation step and utilize the corresponding chunk with its contextual information from multi-granular views to enhance retrieval accuracy. We further introduce an adaptive prompt tuning method for the retrieval model to adapt the heterogeneous retrieval augmented generation process. Extensive experiments demonstrate that \myname achieves significant improvements compared to baselines.
comment: 10 pages, 5 figures
☆ BioChemInsight: An Open-Source Toolkit for Automated Identification and Recognition of Optical Chemical Structures and Activity Data in Scientific Publications
Automated extraction of chemical structures and their bioactivity data is crucial for accelerating drug discovery and enabling data-driven pharmaceutical research. Existing optical chemical structure recognition (OCSR) tools fail to autonomously associate molecular structures with their bioactivity profiles, creating a critical bottleneck in structure-activity relationship (SAR) analysis. Here, we present BioChemInsight, an open-source pipeline that integrates: (1) DECIMER Segmentation and MolVec for chemical structure recognition, (2) Qwen2.5-VL-32B for compound identifier association, and (3) PaddleOCR with Gemini-2.0-flash for bioactivity extraction and unit normalization. We evaluated the performance of BioChemInsight on 25 patents and 17 articles. BioChemInsight achieved 95% accuracy for tabular patent data (structure/identifier recognition), with lower accuracy in non-tabular patents (~80% structures, ~75% identifiers), plus 92.2 % bioactivity extraction accuracy. For articles, it attained >99% identifiers and 78-80% structure accuracy in non-tabular formats, plus 97.4% bioactivity extraction accuracy. The system generates ready-to-use SAR datasets, reducing data preprocessing time from weeks to hours while enabling applications in high-throughput screening and ML-driven drug design (https://github.com/dahuilangda/BioChemInsight).
comment: 20 pages, 7 figures
♻ ☆ Enhancing Tourism Recommender Systems for Sustainable City Trips Using Retrieval-Augmented Generation RecSys 2024
Tourism Recommender Systems (TRS) have traditionally focused on providing personalized travel suggestions, often prioritizing user preferences without considering broader sustainability goals. Integrating sustainability into TRS has become essential with the increasing need to balance environmental impact, local community interests, and visitor satisfaction. This paper proposes a novel approach to enhancing TRS for sustainable city trips using Large Language Models (LLMs) and a modified Retrieval-Augmented Generation (RAG) pipeline. We enhance the traditional RAG system by incorporating a sustainability metric based on a city's popularity and seasonal demand during the prompt augmentation phase. This modification, called Sustainability Augmented Reranking (SAR), ensures the system's recommendations align with sustainability goals. Evaluations using popular open-source LLMs, such as Llama-3.1-Instruct-8B and Mistral-Instruct-7B, demonstrate that the SAR-enhanced approach consistently matches or outperforms the baseline (without SAR) across most metrics, highlighting the benefits of incorporating sustainability into TRS.
comment: Accepted at the RecSoGood 2024 Workshop co-located with the 18th ACM Conference on Recommender Systems (RecSys 2024)
♻ ☆ GLINT-RU: Gated Lightweight Intelligent Recurrent Units for Sequential Recommender Systems
Transformer-based models have gained significant traction in sequential recommender systems (SRSs) for their ability to capture user-item interactions effectively. However, these models often suffer from high computational costs and slow inference. Meanwhile, existing efficient SRS approaches struggle to embed high-quality semantic and positional information into latent representations. To tackle these challenges, this paper introduces GLINT-RU, a lightweight and efficient SRS leveraging a single-layer dense selective Gated Recurrent Units (GRU) module to accelerate inference. By incorporating a dense selective gate, GLINT-RU adaptively captures temporal dependencies and fine-grained positional information, generating high-quality latent representations. Additionally, a parallel mixing block infuses fine-grained positional features into user-item interactions, enhancing both recommendation quality and efficiency. Extensive experiments on three datasets demonstrate that GLINT-RU achieves superior prediction accuracy and inference speed, outperforming baselines based on RNNs, Transformers, MLPs, and SSMs. These results establish GLINT-RU as a powerful and efficient solution for SRSs.
♻ ☆ Generative Regression Based Watch Time Prediction for Short-Video Recommendation
Watch time prediction (WTP) has emerged as a pivotal task in short video recommendation systems, designed to quantify user engagement through continuous interaction modeling. Predicting users' watch times on videos often encounters fundamental challenges, including wide value ranges and imbalanced data distributions, which can lead to significant estimation bias when directly applying regression techniques. Recent studies have attempted to address these issues by converting the continuous watch time estimation into an ordinal regression task. While these methods demonstrate partial effectiveness, they exhibit notable limitations: (1) the discretization process frequently relies on bucket partitioning, inherently reducing prediction flexibility and accuracy and (2) the interdependencies among different partition intervals remain underutilized, missing opportunities for effective error correction. Inspired by language modeling paradigms, we propose a novel Generative Regression (GR) framework that reformulates WTP as a sequence generation task. Our approach employs \textit{structural discretization} to enable nearly lossless value reconstruction while maintaining prediction fidelity. Through carefully designed vocabulary construction and label encoding schemes, each watch time is bijectively mapped to a token sequence. To mitigate the training-inference discrepancy caused by teacher-forcing, we introduce a \textit{curriculum learning with embedding mixup} strategy that gradually transitions from guided to free-generation modes. We evaluate our method against state-of-the-art approaches on two public datasets and one industrial dataset. We also perform online A/B testing on the Kuaishou App to confirm the real-world effectiveness. The results conclusively show that GR outperforms existing techniques significantly.
comment: 10 pages, 5 figures, conference or other essential info
♻ ☆ Generative Recommender with End-to-End Learnable Item Tokenization SIGIR 2025
Generative recommendation systems have gained increasing attention as an innovative approach that directly generates item identifiers for recommendation tasks. Despite their potential, a major challenge is the effective construction of item identifiers that align well with recommender systems. Current approaches often treat item tokenization and generative recommendation training as separate processes, which can lead to suboptimal performance. To overcome this issue, we introduce ETEGRec, a novel End-To-End Generative Recommender that unifies item tokenization and generative recommendation into a cohesive framework. Built on a dual encoder-decoder architecture, ETEGRec consists of an item tokenizer and a generative recommender. To enable synergistic interaction between these components, we propose a recommendation-oriented alignment strategy, which includes two key optimization objectives: sequence-item alignment and preference-semantic alignment. These objectives tightly couple the learning processes of the item tokenizer and the generative recommender, fostering mutual enhancement. Additionally, we develop an alternating optimization technique to ensure stable and efficient end-to-end training of the entire framework. Extensive experiments demonstrate the superior performance of our approach compared to traditional sequential recommendation models and existing generative recommendation baselines.
comment: Accepted by SIGIR 2025 Research Track
♻ ☆ LD4MRec: Simplifying and Powering Diffusion Model for Multimedia Recommendation
Multimedia recommendation aims to predict users' future behaviors based on observed behaviors and item content information. However, the inherent noise contained in observed behaviors easily leads to suboptimal recommendation performance. Recently, the diffusion model's ability to generate information from noise presents a promising solution to this issue, prompting us to explore its application in multimedia recommendation. Nonetheless, several challenges must be addressed: 1) The diffusion model requires simplification to meet the efficiency requirements of real-time recommender systems, 2) The generated behaviors must align with user preference. To address these challenges, we propose a Light Diffusion model for Multimedia Recommendation (LD4MRec). LD4MRec largely reduces computational complexity by employing a forward-free inference strategy, which directly predicts future behaviors from observed noisy behaviors. Meanwhile, to ensure the alignment between generated behaviors and user preference, we propose a novel Conditional neural Network (C-Net). C-Net achieves guided generation by leveraging two key signals, collaborative signals and personalized modality preference signals, thereby improving the semantic consistency between generated behaviors and user preference. Experiments conducted on three real-world datasets demonstrate the effectiveness of LD4MRec.
♻ ☆ DeepRetrieval: Hacking Real Search Engines and Retrievers with Large Language Models via Reinforcement Learning
Information retrieval systems are crucial for enabling effective access to large document collections. Recent approaches have leveraged Large Language Models (LLMs) to enhance retrieval performance through query augmentation, but often rely on expensive supervised learning or distillation techniques that require significant computational resources and hand-labeled data. We introduce DeepRetrieval, a reinforcement learning (RL) approach that trains LLMs for query generation through trial and error without supervised data (reference query). Using retrieval metrics as rewards, our system generates queries that maximize retrieval performance. DeepRetrieval outperforms leading methods on literature search with 65.07% (vs. previous SOTA 24.68%) recall for publication search and 63.18% (vs. previous SOTA 32.11%) recall for trial search using real-world search engines. DeepRetrieval also dominates in evidence-seeking retrieval, classic information retrieval and SQL database search. With only 3B parameters, it outperforms industry-leading models like GPT-4o and Claude-3.5-Sonnet on 11/13 datasets. These results demonstrate that our RL approach offers a more efficient and effective paradigm for information retrieval. Our data and code are available at: https://github.com/pat-jj/DeepRetrieval.