MyArxiv
Robotics 39
☆ DemoDiffusion: One-Shot Human Imitation using pre-trained Diffusion Policy
We propose DemoDiffusion, a simple and scalable method for enabling robots to perform manipulation tasks in natural environments by imitating a single human demonstration. Our approach is based on two key insights. First, the hand motion in a human demonstration provides a useful prior for the robot's end-effector trajectory, which we can convert into a rough open-loop robot motion trajectory via kinematic retargeting. Second, while this retargeted motion captures the overall structure of the task, it may not align well with plausible robot actions in-context. To address this, we leverage a pre-trained generalist diffusion policy to modify the trajectory, ensuring it both follows the human motion and remains within the distribution of plausible robot actions. Our approach avoids the need for online reinforcement learning or paired human-robot data, enabling robust adaptation to new tasks and scenes with minimal manual effort. Experiments in both simulation and real-world settings show that DemoDiffusion outperforms both the base policy and the retargeted trajectory, enabling the robot to succeed even on tasks where the pre-trained generalist policy fails entirely. Project page: https://demodiffusion.github.io/
comment: Preprint(17 pages). Under Review
☆ A Computationally Aware Multi Objective Framework for Camera LiDAR Calibration
Accurate extrinsic calibration between LiDAR and camera sensors is important for reliable perception in autonomous systems. In this paper, we present a novel multi-objective optimization framework that jointly minimizes the geometric alignment error and computational cost associated with camera-LiDAR calibration. We optimize two objectives: (1) error between projected LiDAR points and ground-truth image edges, and (2) a composite metric for computational cost reflecting runtime and resource usage. Using the NSGA-II \cite{deb2002nsga2} evolutionary algorithm, we explore the parameter space defined by 6-DoF transformations and point sampling rates, yielding a well-characterized Pareto frontier that exposes trade-offs between calibration fidelity and resource efficiency. Evaluations are conducted on the KITTI dataset using its ground-truth extrinsic parameters for validation, with results verified through both multi-objective and constrained single-objective baselines. Compared to existing gradient-based and learned calibration methods, our approach demonstrates interpretable, tunable performance with lower deployment overhead. Pareto-optimal configurations are further analyzed for parameter sensitivity and innovation insights. A preference-based decision-making strategy selects solutions from the Pareto knee region to suit the constraints of the embedded system. The robustness of calibration is tested across variable edge-intensity weighting schemes, highlighting optimal balance points. Although real-time deployment on embedded platforms is deferred to future work, this framework establishes a scalable and transparent method for calibration under realistic misalignment and resource-limited conditions, critical for long-term autonomy, particularly in SAE L3+ vehicles receiving OTA updates.
comment: 16 pages, 10 figures
☆ Task Allocation of UAVs for Monitoring Missions via Hardware-in-the-Loop Simulation and Experimental Validation
This study addresses the optimisation of task allocation for Unmanned Aerial Vehicles (UAVs) within industrial monitoring missions. The proposed methodology integrates a Genetic Algorithms (GA) with a 2-Opt local search technique to obtain a high-quality solution. Our approach was experimentally validated in an industrial zone to demonstrate its efficacy in real-world scenarios. Also, a Hardware-in-the-loop (HIL) simulator for the UAVs team is introduced. Moreover, insights about the correlation between the theoretical cost function and the actual battery consumption and time of flight are deeply analysed. Results show that the considered costs for the optimisation part of the problem closely correlate with real-world data, confirming the practicality of the proposed approach.
☆ Learning-Based Distance Estimation for 360° Single-Sensor Setups
Accurate distance estimation is a fundamental challenge in robotic perception, particularly in omnidirectional imaging, where traditional geometric methods struggle with lens distortions and environmental variability. In this work, we propose a neural network-based approach for monocular distance estimation using a single 360{\deg} fisheye lens camera. Unlike classical trigonometric techniques that rely on precise lens calibration, our method directly learns and infers the distance of objects from raw omnidirectional inputs, offering greater robustness and adaptability across diverse conditions. We evaluate our approach on three 360{\deg} datasets (LOAF, ULM360, and a newly captured dataset Boat360), each representing distinct environmental and sensor setups. Our experimental results demonstrate that the proposed learning-based model outperforms traditional geometry-based methods and other learning baselines in both accuracy and robustness. These findings highlight the potential of deep learning for real-time omnidirectional distance estimation, making our approach particularly well-suited for low-cost applications in robotics, autonomous navigation, and surveillance.
comment: Submitted to ECMR 2025
☆ Communication-Aware Map Compression for Online Path-Planning: A Rate-Distortion Approach
This paper addresses the problem of collaborative navigation in an unknown environment, where two robots, referred to in the sequel as the Seeker and the Supporter, traverse the space simultaneously. The Supporter assists the Seeker by transmitting a compressed representation of its local map under bandwidth constraints to support the Seeker's path-planning task. We introduce a bit-rate metric based on the expected binary codeword length to quantify communication cost. Using this metric, we formulate the compression design problem as a rate-distortion optimization problem that determines when to communicate, which regions of the map should be included in the compressed representation, and at what resolution (i.e., quantization level) they should be encoded. Our formulation allows different map regions to be encoded at varying quantization levels based on their relevance to the Seeker's path-planning task. We demonstrate that the resulting optimization problem is convex, and admits a closed-form solution known in the information theory literature as reverse water-filling, enabling efficient, low-computation, and real-time implementation. Additionally, we show that the Seeker can infer the compression decisions of the Supporter independently, requiring only the encoded map content and not the encoding policy itself to be transmitted, thereby reducing communication overhead. Simulation results indicate that our method effectively constructs compressed, task-relevant map representations, both in content and resolution, that guide the Seeker's planning decisions even under tight bandwidth limitations.
☆ HRIBench: Benchmarking Vision-Language Models for Real-Time Human Perception in Human-Robot Interaction
Real-time human perception is crucial for effective human-robot interaction (HRI). Large vision-language models (VLMs) offer promising generalizable perceptual capabilities but often suffer from high latency, which negatively impacts user experience and limits VLM applicability in real-world scenarios. To systematically study VLM capabilities in human perception for HRI and performance-latency trade-offs, we introduce HRIBench, a visual question-answering (VQA) benchmark designed to evaluate VLMs across a diverse set of human perceptual tasks critical for HRI. HRIBench covers five key domains: (1) non-verbal cue understanding, (2) verbal instruction understanding, (3) human-robot object relationship understanding, (4) social navigation, and (5) person identification. To construct HRIBench, we collected data from real-world HRI environments to curate questions for non-verbal cue understanding, and leveraged publicly available datasets for the remaining four domains. We curated 200 VQA questions for each domain, resulting in a total of 1000 questions for HRIBench. We then conducted a comprehensive evaluation of both state-of-the-art closed-source and open-source VLMs (N=11) on HRIBench. Our results show that, despite their generalizability, current VLMs still struggle with core perceptual capabilities essential for HRI. Moreover, none of the models within our experiments demonstrated a satisfactory performance-latency trade-off suitable for real-time deployment, underscoring the need for future research on developing smaller, low-latency VLMs with improved human perception capabilities. HRIBench and our results can be found in this Github repository: https://github.com/interaction-lab/HRIBench.
comment: Accepted to the 19th International Symposium on Experimental Robotics (ISER 2025)
☆ Leveraging Correlation Across Test Platforms for Variance-Reduced Metric Estimation
Learning-based robotic systems demand rigorous validation to assure reliable performance, but extensive real-world testing is often prohibitively expensive, and if conducted may still yield insufficient data for high-confidence guarantees. In this work, we introduce a general estimation framework that leverages paired data across test platforms, e.g., paired simulation and real-world observations, to achieve better estimates of real-world metrics via the method of control variates. By incorporating cheap and abundant auxiliary measurements (for example, simulator outputs) as control variates for costly real-world samples, our method provably reduces the variance of Monte Carlo estimates and thus requires significantly fewer real-world samples to attain a specified confidence bound on the mean performance. We provide theoretical analysis characterizing the variance and sample-efficiency improvement, and demonstrate empirically in autonomous driving and quadruped robotics settings that our approach achieves high-probability bounds with markedly improved sample efficiency. Our technique can lower the real-world testing burden for validating the performance of the stack, thereby enabling more efficient and cost-effective experimental evaluation of robotic systems.
☆ Lightweight Multi-Frame Integration for Robust YOLO Object Detection in Videos
Modern image-based object detection models, such as YOLOv7, primarily process individual frames independently, thus ignoring valuable temporal context naturally present in videos. Meanwhile, existing video-based detection methods often introduce complex temporal modules, significantly increasing model size and computational complexity. In practical applications such as surveillance and autonomous driving, transient challenges including motion blur, occlusions, and abrupt appearance changes can severely degrade single-frame detection performance. To address these issues, we propose a straightforward yet highly effective strategy: stacking multiple consecutive frames as input to a YOLO-based detector while supervising only the output corresponding to a single target frame. This approach leverages temporal information with minimal modifications to existing architectures, preserving simplicity, computational efficiency, and real-time inference capability. Extensive experiments on the challenging MOT20Det and our BOAT360 datasets demonstrate that our method improves detection robustness, especially for lightweight models, effectively narrowing the gap between compact and heavy detection networks. Additionally, we contribute the BOAT360 benchmark dataset, comprising annotated fisheye video sequences captured from a boat, to support future research in multi-frame video object detection in challenging real-world scenarios.
comment: Submitted to ECMR 2025
☆ Critical Anatomy-Preserving & Terrain-Augmenting Navigation (CAPTAiN): Application to Laminectomy Surgical Education
Surgical training remains a crucial milestone in modern medicine, with procedures such as laminectomy exemplifying the high risks involved. Laminectomy drilling requires precise manual control to mill bony tissue while preserving spinal segment integrity and avoiding breaches in the dura: the protective membrane surrounding the spinal cord. Despite unintended tears occurring in up to 11.3% of cases, no assistive tools are currently utilized to reduce this risk. Variability in patient anatomy further complicates learning for novice surgeons. This study introduces CAPTAiN, a critical anatomy-preserving and terrain-augmenting navigation system that provides layered, color-coded voxel guidance to enhance anatomical awareness during spinal drilling. CAPTAiN was evaluated against a standard non-navigated approach through 110 virtual laminectomies performed by 11 orthopedic residents and medical students. CAPTAiN significantly improved surgical completion rates of target anatomy (87.99% vs. 74.42%) and reduced cognitive load across multiple NASA-TLX domains. It also minimized performance gaps across experience levels, enabling novices to perform on par with advanced trainees. These findings highlight CAPTAiN's potential to optimize surgical execution and support skill development across experience levels. Beyond laminectomy, it demonstrates potential for broader applications across various surgical and drilling procedures, including those in neurosurgery, otolaryngology, and other medical fields.
☆ Behavior Foundation Model: Towards Next-Generation Whole-Body Control System of Humanoid Robots
Humanoid robots are drawing significant attention as versatile platforms for complex motor control, human-robot interaction, and general-purpose physical intelligence. However, achieving efficient whole-body control (WBC) in humanoids remains a fundamental challenge due to sophisticated dynamics, underactuation, and diverse task requirements. While learning-based controllers have shown promise for complex tasks, their reliance on labor-intensive and costly retraining for new scenarios limits real-world applicability. To address these limitations, behavior(al) foundation models (BFMs) have emerged as a new paradigm that leverages large-scale pretraining to learn reusable primitive skills and behavioral priors, enabling zero-shot or rapid adaptation to a wide range of downstream tasks. In this paper, we present a comprehensive overview of BFMs for humanoid WBC, tracing their development across diverse pre-training pipelines. Furthermore, we discuss real-world applications, current limitations, urgent challenges, and future opportunities, positioning BFMs as a key approach toward scalable and general-purpose humanoid intelligence. Finally, we provide a curated and long-term list of BFM papers and projects to facilitate more subsequent research, which is available at https://github.com/yuanmingqi/awesome-bfm-papers.
comment: 19 pages, 8 figures
☆ EANS: Reducing Energy Consumption for UAV with an Environmental Adaptive Navigation Strategy
Unmanned Aerial Vehicles (UAVS) are limited by the onboard energy. Refinement of the navigation strategy directly affects both the flight velocity and the trajectory based on the adjustment of key parameters in the UAVS pipeline, thus reducing energy consumption. However, existing techniques tend to adopt static and conservative strategies in dynamic scenarios, leading to inefficient energy reduction. Dynamically adjusting the navigation strategy requires overcoming the challenges including the task pipeline interdependencies, the environmental-strategy correlations, and the selecting parameters. To solve the aforementioned problems, this paper proposes a method to dynamically adjust the navigation strategy of the UAVS by analyzing its dynamic characteristics and the temporal characteristics of the autonomous navigation pipeline, thereby reducing UAVS energy consumption in response to environmental changes. We compare our method with the baseline through hardware-in-the-loop (HIL) simulation and real-world experiments, showing our method 3.2X and 2.6X improvements in mission time, 2.4X and 1.6X improvements in energy, respectively.
☆ A Review of Personalisation in Human-Robot Collaboration and Future Perspectives Towards Industry 5.0
The shift in research focus from Industry 4.0 to Industry 5.0 (I5.0) promises a human-centric workplace, with social and well-being values at the centre of technological implementation. Human-Robot Collaboration (HRC) is a core aspect of I5.0 development, with an increase in adaptive and personalised interactions and behaviours. This review investigates recent advancements towards personalised HRC, where user-centric adaption is key. There is a growing trend for adaptable HRC research, however there lacks a consistent and unified approach. The review highlights key research trends on which personal factors are considered, workcell and interaction design, and adaptive task completion. This raises various key considerations for future developments, particularly around the ethical and regulatory development of personalised systems, which are discussed in detail.
comment: Accepted by the 2025 34th IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)
☆ Learn to Position -- A Novel Meta Method for Robotic Positioning
Absolute positioning accuracy is a vital specification for robots. Achieving high position precision can be challenging due to the presence of various sources of errors. Meanwhile, accurately depicting these errors is difficult due to their stochastic nature. Vision-based methods are commonly integrated to guide robotic positioning, but their performance can be highly impacted by inevitable occlusions or adverse lighting conditions. Drawing on the aforementioned considerations, a vision-free, model-agnostic meta-method for compensating robotic position errors is proposed, which maximizes the probability of accurate robotic position via interactive feedback. Meanwhile, the proposed method endows the robot with the capability to learn and adapt to various position errors, which is inspired by the human's instinct for grasping under uncertainties. Furthermore, it is a self-learning and self-adaptive method able to accelerate the robotic positioning process as more examples are incorporated and learned. Empirical studies validate the effectiveness of the proposed method. As of the writing of this paper, the proposed meta search method has already been implemented in a robotic-based assembly line for odd-form electronic components.
☆ Multimodal Behaviour Trees for Robotic Laboratory Task Automation ICRA 2025
Laboratory robotics offer the capability to conduct experiments with a high degree of precision and reproducibility, with the potential to transform scientific research. Trivial and repeatable tasks; e.g., sample transportation for analysis and vial capping are well-suited for robots; if done successfully and reliably, chemists could contribute their efforts towards more critical research activities. Currently, robots can perform these tasks faster than chemists, but how reliable are they? Improper capping could result in human exposure to toxic chemicals which could be fatal. To ensure that robots perform these tasks as accurately as humans, sensory feedback is required to assess the progress of task execution. To address this, we propose a novel methodology based on behaviour trees with multimodal perception. Along with automating robotic tasks, this methodology also verifies the successful execution of the task, a fundamental requirement in safety-critical environments. The experimental evaluation was conducted on two lab tasks: sample vial capping and laboratory rack insertion. The results show high success rate, i.e., 88% for capping and 92% for insertion, along with strong error detection capabilities. This ultimately proves the robustness and reliability of our approach and that using multimodal behaviour trees should pave the way towards the next generation of robotic chemists.
comment: 7 pages, 5 figures, accepted and presented in ICRA 2025
☆ SPARK: Graph-Based Online Semantic Integration System for Robot Task Planning
The ability to update information acquired through various means online during task execution is crucial for a general-purpose service robot. This information includes geometric and semantic data. While SLAM handles geometric updates on 2D maps or 3D point clouds, online updates of semantic information remain unexplored. We attribute the challenge to the online scene graph representation, for its utility and scalability. Building on previous works regarding offline scene graph representations, we study online graph representations of semantic information in this work. We introduce SPARK: Spatial Perception and Robot Knowledge Integration. This framework extracts semantic information from environment-embedded cues and updates the scene graph accordingly, which is then used for subsequent task planning. We demonstrate that graph representations of spatial relationships enhance the robot system's ability to perform tasks in dynamic environments and adapt to unconventional spatial cues, like gestures.
☆ Enhanced Robotic Navigation in Deformable Environments using Learning from Demonstration and Dynamic Modulation IROS 2025
This paper presents a novel approach for robot navigation in environments containing deformable obstacles. By integrating Learning from Demonstration (LfD) with Dynamical Systems (DS), we enable adaptive and efficient navigation in complex environments where obstacles consist of both soft and hard regions. We introduce a dynamic modulation matrix within the DS framework, allowing the system to distinguish between traversable soft regions and impassable hard areas in real-time, ensuring safe and flexible trajectory planning. We validate our method through extensive simulations and robot experiments, demonstrating its ability to navigate deformable environments. Additionally, the approach provides control over both trajectory and velocity when interacting with deformable objects, including at intersections, while maintaining adherence to the original DS trajectory and dynamically adapting to obstacles for smooth and reliable navigation.
comment: Accepted to IROS 2025
☆ CARMA: Context-Aware Situational Grounding of Human-Robot Group Interactions by Combining Vision-Language Models with Object and Action Recognition
We introduce CARMA, a system for situational grounding in human-robot group interactions. Effective collaboration in such group settings requires situational awareness based on a consistent representation of present persons and objects coupled with an episodic abstraction of events regarding actors and manipulated objects. This calls for a clear and consistent assignment of instances, ensuring that robots correctly recognize and track actors, objects, and their interactions over time. To achieve this, CARMA uniquely identifies physical instances of such entities in the real world and organizes them into grounded triplets of actors, objects, and actions. To validate our approach, we conducted three experiments, where multiple humans and a robot interact: collaborative pouring, handovers, and sorting. These scenarios allow the assessment of the system's capabilities as to role distinction, multi-actor awareness, and consistent instance identification. Our experiments demonstrate that the system can reliably generate accurate actor-action-object triplets, providing a structured and robust foundation for applications requiring spatiotemporal reasoning and situated decision-making in collaborative settings.
☆ PIMBS: Efficient Body Schema Learning for Musculoskeletal Humanoids with Physics-Informed Neural Networks
Musculoskeletal humanoids are robots that closely mimic the human musculoskeletal system, offering various advantages such as variable stiffness control, redundancy, and flexibility. However, their body structure is complex, and muscle paths often significantly deviate from geometric models. To address this, numerous studies have been conducted to learn body schema, particularly the relationships among joint angles, muscle tension, and muscle length. These studies typically rely solely on data collected from the actual robot, but this data collection process is labor-intensive, and learning becomes difficult when the amount of data is limited. Therefore, in this study, we propose a method that applies the concept of Physics-Informed Neural Networks (PINNs) to the learning of body schema in musculoskeletal humanoids, enabling high-accuracy learning even with a small amount of data. By utilizing not only data obtained from the actual robot but also the physical laws governing the relationship between torque and muscle tension under the assumption of correct joint structure, more efficient learning becomes possible. We apply the proposed method to both simulation and an actual musculoskeletal humanoid and discuss its effectiveness and characteristics.
comment: Accepted at IEEE Robotics and Automation Letters
☆ Finding the Easy Way Through -- the Probabilistic Gap Planner for Social Robot Navigation
In Social Robot Navigation, autonomous agents need to resolve many sequential interactions with other agents. State-of-the art planners can efficiently resolve the next, imminent interaction cooperatively and do not focus on longer planning horizons. This makes it hard to maneuver scenarios where the agent needs to select a good strategy to find gaps or channels in the crowd. We propose to decompose trajectory planning into two separate steps: Conflict avoidance for finding good, macroscopic trajectories, and cooperative collision avoidance (CCA) for resolving the next interaction optimally. We propose the Probabilistic Gap Planner (PGP) as a conflict avoidance planner. PGP modifies an established probabilistic collision risk model to include a general assumption of cooperativity. PGP biases the short-term CCA planner to head towards gaps in the crowd. In extensive simulations with crowds of varying density, we show that using PGP in addition to state-of-the-art CCA planners improves the agents' performance: On average, agents keep more space to others, create less tension, and cause fewer collisions. This typically comes at the expense of slightly longer paths. PGP runs in real-time on WaPOCHI mobile robot by Honda R&D.
☆ Building Forest Inventories with Autonomous Legged Robots -- System, Lessons, and Challenges Ahead
Legged robots are increasingly being adopted in industries such as oil, gas, mining, nuclear, and agriculture. However, new challenges exist when moving into natural, less-structured environments, such as forestry applications. This paper presents a prototype system for autonomous, under-canopy forest inventory with legged platforms. Motivated by the robustness and mobility of modern legged robots, we introduce a system architecture which enabled a quadruped platform to autonomously navigate and map forest plots. Our solution involves a complete navigation stack for state estimation, mission planning, and tree detection and trait estimation. We report the performance of the system from trials executed over one and a half years in forests in three European countries. Our results with the ANYmal robot demonstrate that we can survey plots up to 1 ha plot under 30 min, while also identifying trees with typical DBH accuracy of 2cm. The findings of this project are presented as five lessons and challenges. Particularly, we discuss the maturity of hardware development, state estimation limitations, open problems in forest navigation, future avenues for robotic forest inventory, and more general challenges to assess autonomous systems. By sharing these lessons and challenges, we offer insight and new directions for future research on legged robots, navigation systems, and applications in natural environments. Additional videos can be found in https://dynamic.robots.ox.ac.uk/projects/legged-robots
comment: 20 pages, 13 figures. Pre-print version of the accepted paper for IEEE Transactions on Field Robotics (T-FR)
☆ Near Time-Optimal Hybrid Motion Planning for Timber Cranes ICRA 2025
Efficient, collision-free motion planning is essential for automating large-scale manipulators like timber cranes. They come with unique challenges such as hydraulic actuation constraints and passive joints-factors that are seldom addressed by current motion planning methods. This paper introduces a novel approach for time-optimal, collision-free hybrid motion planning for a hydraulically actuated timber crane with passive joints. We enhance the via-point-based stochastic trajectory optimization (VP-STO) algorithm to include pump flow rate constraints and develop a novel collision cost formulation to improve robustness. The effectiveness of the enhanced VP-STO as an optimal single-query global planner is validated by comparison with an informed RRT* algorithm using a time-optimal path parameterization (TOPP). The overall hybrid motion planning is formed by combination with a gradient-based local planner that is designed to follow the global planner's reference and to systematically consider the passive joint dynamics for both collision avoidance and sway damping.
comment: Accepted at ICRA 2025
☆ Real-Time Obstacle Avoidance Algorithms for Unmanned Aerial and Ground Vehicles
The growing use of mobile robots in sectors such as automotive, agriculture, and rescue operations reflects progress in robotics and autonomy. In unmanned aerial vehicles (UAVs), most research emphasizes visual SLAM, sensor fusion, and path planning. However, applying UAVs to search and rescue missions in disaster zones remains underexplored, especially for autonomous navigation. This report develops methods for real-time and secure UAV maneuvering in complex 3D environments, crucial during forest fires. Building upon past research, it focuses on designing navigation algorithms for unfamiliar and hazardous environments, aiming to improve rescue efficiency and safety through UAV-based early warning and rapid response. The work unfolds in phases. First, a 2D fusion navigation strategy is explored, initially for mobile robots, enabling safe movement in dynamic settings. This sets the stage for advanced features such as adaptive obstacle handling and decision-making enhancements. Next, a novel 3D reactive navigation strategy is introduced for collision-free movement in forest fire simulations, addressing the unique challenges of UAV operations in such scenarios. Finally, the report proposes a unified control approach that integrates UAVs and unmanned ground vehicles (UGVs) for coordinated rescue missions in forest environments. Each phase presents challenges, proposes control models, and validates them with mathematical and simulation-based evidence. The study offers practical value and academic insights for improving the role of UAVs in natural disaster rescue operations.
☆ Why Robots Are Bad at Detecting Their Mistakes: Limitations of Miscommunication Detection in Human-Robot Dialogue
Detecting miscommunication in human-robot interaction is a critical function for maintaining user engagement and trust. While humans effortlessly detect communication errors in conversations through both verbal and non-verbal cues, robots face significant challenges in interpreting non-verbal feedback, despite advances in computer vision for recognizing affective expressions. This research evaluates the effectiveness of machine learning models in detecting miscommunications in robot dialogue. Using a multi-modal dataset of 240 human-robot conversations, where four distinct types of conversational failures were systematically introduced, we assess the performance of state-of-the-art computer vision models. After each conversational turn, users provided feedback on whether they perceived an error, enabling an analysis of the models' ability to accurately detect robot mistakes. Despite using state-of-the-art models, the performance barely exceeds random chance in identifying miscommunication, while on a dataset with more expressive emotional content, they successfully identified confused states. To explore the underlying cause, we asked human raters to do the same. They could also only identify around half of the induced miscommunications, similarly to our model. These results uncover a fundamental limitation in identifying robot miscommunications in dialogue: even when users perceive the induced miscommunication as such, they often do not communicate this to their robotic conversation partner. This knowledge can shape expectations of the performance of computer vision models and can help researchers to design better human-robot conversations by deliberately eliciting feedback where needed.
comment: Accepted at the 34th IEEE International Conference on Robot and Human Interactive Communication (RO-MAN 2025)
☆ Generating and Customizing Robotic Arm Trajectories using Neural Networks
We introduce a neural network approach for generating and customizing the trajectory of a robotic arm, that guarantees precision and repeatability. To highlight the potential of this novel method, we describe the design and implementation of the technique and show its application in an experimental setting of cognitive robotics. In this scenario, the NICO robot was characterized by the ability to point to specific points in space with precise linear movements, increasing the predictability of the robotic action during its interaction with humans. To achieve this goal, the neural network computes the forward kinematics of the robot arm. By integrating it with a generator of joint angles, another neural network was developed and trained on an artificial dataset created from suitable start and end poses of the robotic arm. Through the computation of angular velocities, the robot was characterized by its ability to perform the movement, and the quality of its action was evaluated in terms of shape and accuracy. Thanks to its broad applicability, our approach successfully generates precise trajectories that could be customized in their shape and adapted to different settings.
comment: The code is released at https://github.com/andylucny/nico2/tree/main/generate
☆ Personalized Mental State Evaluation in Human-Robot Interaction using Federated Learning
With the advent of Industry 5.0, manufacturers are increasingly prioritizing worker well-being alongside mass customization. Stress-aware Human-Robot Collaboration (HRC) plays a crucial role in this paradigm, where robots must adapt their behavior to human mental states to improve collaboration fluency and safety. This paper presents a novel framework that integrates Federated Learning (FL) to enable personalized mental state evaluation while preserving user privacy. By leveraging physiological signals, including EEG, ECG, EDA, EMG, and respiration, a multimodal model predicts an operator's stress level, facilitating real-time robot adaptation. The FL-based approach allows distributed on-device training, ensuring data confidentiality while improving model generalization and individual customization. Results demonstrate that the deployment of an FL approach results in a global model with performance in stress prediction accuracy comparable to a centralized training approach. Moreover, FL allows for enhancing personalization, thereby optimizing human-robot interaction in industrial settings, while preserving data privacy. The proposed framework advances privacy-preserving, adaptive robotics to enhance workforce well-being in smart manufacturing.
☆ PSALM-V: Automating Symbolic Planning in Interactive Visual Environments with Large Language Models
We propose PSALM-V, the first autonomous neuro-symbolic learning system able to induce symbolic action semantics (i.e., pre- and post-conditions) in visual environments through interaction. PSALM-V bootstraps reliable symbolic planning without expert action definitions, using LLMs to generate heuristic plans and candidate symbolic semantics. Previous work has explored using large language models to generate action semantics for Planning Domain Definition Language (PDDL)-based symbolic planners. However, these approaches have primarily focused on text-based domains or relied on unrealistic assumptions, such as access to a predefined problem file, full observability, or explicit error messages. By contrast, PSALM-V dynamically infers PDDL problem files and domain action semantics by analyzing execution outcomes and synthesizing possible error explanations. The system iteratively generates and executes plans while maintaining a tree-structured belief over possible action semantics for each action, iteratively refining these beliefs until a goal state is reached. Simulated experiments of task completion in ALFRED demonstrate that PSALM-V increases the plan success rate from 37% (Claude-3.7) to 74% in partially observed setups. Results on two 2D game environments, RTFM and Overcooked-AI, show that PSALM-V improves step efficiency and succeeds in domain induction in multi-agent settings. PSALM-V correctly induces PDDL pre- and post-conditions for real-world robot BlocksWorld tasks, despite low-level manipulation failures from the robot.
♻ ☆ FORTE: Tactile Force and Slip Sensing on Compliant Fingers for Delicate Manipulation
Handling delicate and fragile objects remains a major challenge for robotic manipulation, especially for rigid parallel grippers. While the simplicity and versatility of parallel grippers have led to widespread adoption, these grippers are limited by their heavy reliance on visual feedback. Tactile sensing and soft robotics can add responsiveness and compliance. However, existing methods typically involve high integration complexity or suffer from slow response times. In this work, we introduce FORTE, a tactile sensing system embedded in compliant gripper fingers. FORTE uses 3D-printed fin-ray grippers with internal air channels to provide low-latency force and slip feedback. FORTE applies just enough force to grasp objects without damaging them, while remaining easy to fabricate and integrate. We find that FORTE can accurately estimate grasping forces from 0-8 N with an average error of 0.2 N, and detect slip events within 100 ms of occurring. We demonstrate FORTE's ability to grasp a wide range of slippery, fragile, and deformable objects. In particular, FORTE grasps fragile objects like raspberries and potato chips with a 98.6% success rate, and achieves 93% accuracy in detecting slip events. These results highlight FORTE's potential as a robust and practical solution for enabling delicate robotic manipulation. Project page: https://merge-lab.github.io/FORTE
♻ ☆ BEVPlace++: Fast, Robust, and Lightweight LiDAR Global Localization for Unmanned Ground Vehicles
This article introduces BEVPlace++, a novel, fast, and robust LiDAR global localization method for unmanned ground vehicles. It uses lightweight convolutional neural networks (CNNs) on Bird's Eye View (BEV) image-like representations of LiDAR data to achieve accurate global localization through place recognition, followed by 3-DoF pose estimation. Our detailed analyses reveal an interesting fact that CNNs are inherently effective at extracting distinctive features from LiDAR BEV images. Remarkably, keypoints of two BEV images with large translations can be effectively matched using CNN-extracted features. Building on this insight, we design a Rotation Equivariant Module (REM) to obtain distinctive features while enhancing robustness to rotational changes. A Rotation Equivariant and Invariant Network (REIN) is then developed by cascading REM and a descriptor generator, NetVLAD, to sequentially generate rotation equivariant local features and rotation invariant global descriptors. The global descriptors are used first to achieve robust place recognition, and then local features are used for accurate pose estimation. \revise{Experimental results on seven public datasets and our UGV platform demonstrate that BEVPlace++, even when trained on a small dataset (3000 frames of KITTI) only with place labels, generalizes well to unseen environments, performs consistently across different days and years, and adapts to various types of LiDAR scanners.} BEVPlace++ achieves state-of-the-art performance in multiple tasks, including place recognition, loop closure detection, and global localization. Additionally, BEVPlace++ is lightweight, runs in real-time, and does not require accurate pose supervision, making it highly convenient for deployment. \revise{The source codes are publicly available at https://github.com/zjuluolun/BEVPlace2.
comment: Accepted to IEEE Transactions on Robotics
♻ ☆ 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.
♻ ☆ Proximal Control of UAVs with Federated Learning for Human-Robot Collaborative Domains
The human-robot interaction (HRI) is a growing area of research. In HRI, complex command (action) classification is still an open problem that usually prevents the real applicability of such a technique. The literature presents some works that use neural networks to detect these actions. However, occlusion is still a major issue in HRI, especially when using uncrewed aerial vehicles (UAVs), since, during the robot's movement, the human operator is often out of the robot's field of view. Furthermore, in multi-robot scenarios, distributed training is also an open problem. In this sense, this work proposes an action recognition and control approach based on Long Short-Term Memory (LSTM) Deep Neural Networks with two layers in association with three densely connected layers and Federated Learning (FL) embedded in multiple drones. The FL enabled our approach to be trained in a distributed fashion, i.e., access to data without the need for cloud or other repositories, which facilitates the multi-robot system's learning. Furthermore, our multi-robot approach results also prevented occlusion situations, with experiments with real robots achieving an accuracy greater than 96%.
comment: version 2
♻ ☆ Physics-informed Imitative Reinforcement Learning for Real-world Driving
Recent advances in imitative reinforcement learning (IRL) have considerably enhanced the ability of autonomous agents to assimilate expert demonstrations, leading to rapid skill acquisition in a range of demanding tasks. However, such learning-based agents face significant challenges when transferring knowledge to highly dynamic closed-loop environments. Their performance is significantly impacted by the conflicting optimization objectives of imitation learning (IL) and reinforcement learning (RL), sample inefficiency, and the complexity of uncovering the hidden world model and physics. To address this challenge, we propose a physics-informed IRL that is entirely data-driven. It leverages both expert demonstration data and exploratory data with a joint optimization objective, allowing the underlying physical principles of vehicle dynamics to emerge naturally from the training process. The performance is evaluated through empirical experiments and results exceed popular IL, RL and IRL algorithms in closed-loop settings on Waymax benchmark. Our approach exhibits 37.8% reduction in collision rate and 22.2% reduction in off-road rate compared to the baseline method.
♻ ☆ Neural Graph Map: Dense Mapping with Efficient Loop Closure Integration WACV 2025
Neural field-based SLAM methods typically employ a single, monolithic field as their scene representation. This prevents efficient incorporation of loop closure constraints and limits scalability. To address these shortcomings, we propose a novel RGB-D neural mapping framework in which the scene is represented by a collection of lightweight neural fields which are dynamically anchored to the pose graph of a sparse visual SLAM system. Our approach shows the ability to integrate large-scale loop closures, while requiring only minimal reintegration. Furthermore, we verify the scalability of our approach by demonstrating successful building-scale mapping taking multiple loop closures into account during the optimization, and show that our method outperforms existing state-of-the-art approaches on large scenes in terms of quality and runtime. Our code is available open-source at https://github.com/KTH-RPL/neural_graph_mapping.
comment: WACV 2025, Project page: https://kth-rpl.github.io/neural_graph_mapping/
♻ ☆ Graph-Assisted Stitching for Offline Hierarchical Reinforcement Learning ICML 2025
Existing offline hierarchical reinforcement learning methods rely on high-level policy learning to generate subgoal sequences. However, their efficiency degrades as task horizons increase, and they lack effective strategies for stitching useful state transitions across different trajectories. We propose Graph-Assisted Stitching (GAS), a novel framework that formulates subgoal selection as a graph search problem rather than learning an explicit high-level policy. By embedding states into a Temporal Distance Representation (TDR) space, GAS clusters semantically similar states from different trajectories into unified graph nodes, enabling efficient transition stitching. A shortest-path algorithm is then applied to select subgoal sequences within the graph, while a low-level policy learns to reach the subgoals. To improve graph quality, we introduce the Temporal Efficiency (TE) metric, which filters out noisy or inefficient transition states, significantly enhancing task performance. GAS outperforms prior offline HRL methods across locomotion, navigation, and manipulation tasks. Notably, in the most stitching-critical task, it achieves a score of 88.3, dramatically surpassing the previous state-of-the-art score of 1.0. Our source code is available at: https://github.com/qortmdgh4141/GAS.
comment: ICML 2025
♻ ☆ Mamba Policy: Towards Efficient 3D Diffusion Policy with Hybrid Selective State Models IROS 2025
Diffusion models have been widely employed in the field of 3D manipulation due to their efficient capability to learn distributions, allowing for precise prediction of action trajectories. However, diffusion models typically rely on large parameter UNet backbones as policy networks, which can be challenging to deploy on resource-constrained devices. Recently, the Mamba model has emerged as a promising solution for efficient modeling, offering low computational complexity and strong performance in sequence modeling. In this work, we propose the Mamba Policy, a lighter but stronger policy that reduces the parameter count by over 80% compared to the original policy network while achieving superior performance. Specifically, we introduce the XMamba Block, which effectively integrates input information with conditional features and leverages a combination of Mamba and Attention mechanisms for deep feature extraction. Extensive experiments demonstrate that the Mamba Policy excels on the Adroit, Dexart, and MetaWorld datasets, requiring significantly fewer computational resources. Additionally, we highlight the Mamba Policy's enhanced robustness in long-horizon scenarios compared to baseline methods and explore the performance of various Mamba variants within the Mamba Policy framework. Real-world experiments are also conducted to further validate its effectiveness. Our open-source project page can be found at https://andycao1125.github.io/mamba_policy/.
comment: Accepted to IROS 2025
♻ ☆ Teacher Motion Priors: Enhancing Robot Locomotion over Challenging Terrain IROS 2025
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, IROS 2025
♻ ☆ FGS-SLAM: Fourier-based Gaussian Splatting for Real-time SLAM with Sparse and Dense Map Fusion
3D gaussian splatting has advanced simultaneous localization and mapping (SLAM) technology by enabling real-time positioning and the construction of high-fidelity maps. However, the uncertainty in gaussian position and initialization parameters introduces challenges, often requiring extensive iterative convergence and resulting in redundant or insufficient gaussian representations. To address this, we introduce a novel adaptive densification method based on Fourier frequency domain analysis to establish gaussian priors for rapid convergence. Additionally, we propose constructing independent and unified sparse and dense maps, where a sparse map supports efficient tracking via Generalized Iterative Closest Point (GICP) and a dense map creates high-fidelity visual representations. This is the first SLAM system leveraging frequency domain analysis to achieve high-quality gaussian mapping in real-time. Experimental results demonstrate an average frame rate of 36 FPS on Replica and TUM RGB-D datasets, achieving competitive accuracy in both localization and mapping.
♻ ☆ IKDiffuser: A Generative Inverse Kinematics Solver for Multi-arm Robots via Diffusion Model
Solving Inverse Kinematics (IK) problems is fundamental to robotics, but has primarily been successful with single serial manipulators. For multi-arm robotic systems, IK remains challenging due to complex self-collisions, coupled joints, and high-dimensional redundancy. These complexities make traditional IK solvers slow, prone to failure, and lacking in solution diversity. In this paper, we present IKDiffuser, a diffusion-based model designed for fast and diverse IK solution generation for multi-arm robotic systems. IKDiffuser learns the joint distribution over the configuration space, capturing complex dependencies and enabling seamless generalization to multi-arm robotic systems of different structures. In addition, IKDiffuser can incorporate additional objectives during inference without retraining, offering versatility and adaptability for task-specific requirements. In experiments on 6 different multi-arm systems, the proposed IKDiffuser achieves superior solution accuracy, precision, diversity, and computational efficiency compared to existing solvers. The proposed IKDiffuser framework offers a scalable, unified approach to solving multi-arm IK problems, facilitating the potential of multi-arm robotic systems in real-time manipulation tasks.
comment: under review
♻ ☆ EvDetMAV: Generalized MAV Detection from Moving Event Cameras
Existing micro aerial vehicle (MAV) detection methods mainly rely on the target's appearance features in RGB images, whose diversity makes it difficult to achieve generalized MAV detection. We notice that different types of MAVs share the same distinctive features in event streams due to their high-speed rotating propellers, which are hard to see in RGB images. This paper studies how to detect different types of MAVs from an event camera by fully exploiting the features of propellers in the original event stream. The proposed method consists of three modules to extract the salient and spatio-temporal features of the propellers while filtering out noise from background objects and camera motion. Since there are no existing event-based MAV datasets, we introduce a novel MAV dataset for the community. This is the first event-based MAV dataset comprising multiple scenarios and different types of MAVs. Without training, our method significantly outperforms state-of-the-art methods and can deal with challenging scenarios, achieving a precision rate of 83.0\% (+30.3\%) and a recall rate of 81.5\% (+36.4\%) on the proposed testing dataset. The dataset and code are available at: https://github.com/WindyLab/EvDetMAV.
comment: 8 pages, 7 figures. This paper is accepted by IEEE Robotics and Automation Letters
♻ ☆ AnchorDP3: 3D Affordance Guided Sparse Diffusion Policy for Robotic Manipulation
We present AnchorDP3, a diffusion policy framework for dual-arm robotic manipulation that achieves state-of-the-art performance in highly randomized environments. AnchorDP3 integrates three key innovations: (1) Simulator-Supervised Semantic Segmentation, using rendered ground truth to explicitly segment task-critical objects within the point cloud, which provides strong affordance priors; (2) Task-Conditioned Feature Encoders, lightweight modules processing augmented point clouds per task, enabling efficient multi-task learning through a shared diffusion-based action expert; (3) Affordance-Anchored Keypose Diffusion with Full State Supervision, replacing dense trajectory prediction with sparse, geometrically meaningful action anchors, i.e., keyposes such as pre-grasp pose, grasp pose directly anchored to affordances, drastically simplifying the prediction space; the action expert is forced to predict both robot joint angles and end-effector poses simultaneously, which exploits geometric consistency to accelerate convergence and boost accuracy. Trained on large-scale, procedurally generated simulation data, AnchorDP3 achieves a 98.7% average success rate in the RoboTwin benchmark across diverse tasks under extreme randomization of objects, clutter, table height, lighting, and backgrounds. This framework, when integrated with the RoboTwin real-to-sim pipeline, has the potential to enable fully autonomous generation of deployable visuomotor policies from only scene and instruction, totally eliminating human demonstrations from learning manipulation skills.
Artificial Intelligence 139
☆ Inside you are many wolves: Using cognitive models to interpret value trade-offs in LLMs
Navigating everyday social situations often requires juggling conflicting goals, such as conveying a harsh truth, maintaining trust, all while still being mindful of another person's feelings. These value trade-offs are an integral part of human decision-making and language use, however, current tools for interpreting such dynamic and multi-faceted notions of values in LLMs are limited. In cognitive science, so-called "cognitive models" provide formal accounts of these trade-offs in humans, by modeling the weighting of a speaker's competing utility functions in choosing an action or utterance. In this work, we use a leading cognitive model of polite speech to interpret the extent to which LLMs represent human-like trade-offs. We apply this lens to systematically evaluate value trade-offs in two encompassing model settings: degrees of reasoning "effort" in frontier black-box models, and RL post-training dynamics of open-source models. Our results highlight patterns of higher informational utility than social utility in reasoning models, and in open-source models shown to be stronger in mathematical reasoning. Our findings from LLMs' training dynamics suggest large shifts in utility values early on in training with persistent effects of the choice of base model and pretraining data, compared to feedback dataset or alignment method. We show that our method is responsive to diverse aspects of the rapidly evolving LLM landscape, with insights for forming hypotheses about other high-level behaviors, shaping training regimes for reasoning models, and better controlling trade-offs between values during model training.
comment: 11 pages, 3 figures
☆ The Decrypto Benchmark for Multi-Agent Reasoning and Theory of Mind
As Large Language Models (LLMs) gain agentic abilities, they will have to navigate complex multi-agent scenarios, interacting with human users and other agents in cooperative and competitive settings. This will require new reasoning skills, chief amongst them being theory of mind (ToM), or the ability to reason about the "mental" states of other agents. However, ToM and other multi-agent abilities in LLMs are poorly understood, since existing benchmarks suffer from narrow scope, data leakage, saturation, and lack of interactivity. We thus propose Decrypto, a game-based benchmark for multi-agent reasoning and ToM drawing inspiration from cognitive science, computational pragmatics and multi-agent reinforcement learning. It is designed to be as easy as possible in all other dimensions, eliminating confounding factors commonly found in other benchmarks. To our knowledge, it is also the first platform for designing interactive ToM experiments. We validate the benchmark design through comprehensive empirical evaluations of frontier LLMs, robustness studies, and human-AI cross-play experiments. We find that LLM game-playing abilities lag behind humans and simple word-embedding baselines. We then create variants of two classic cognitive science experiments within Decrypto to evaluate three key ToM abilities. Surprisingly, we find that state-of-the-art reasoning models are significantly worse at those tasks than their older counterparts. This demonstrates that Decrypto addresses a crucial gap in current reasoning and ToM evaluations, and paves the path towards better artificial agents.
comment: 41 pages, 19 figures
☆ Disentangled representations of microscopy images IJCNN 2025
Microscopy image analysis is fundamental for different applications, from diagnosis to synthetic engineering and environmental monitoring. Modern acquisition systems have granted the possibility to acquire an escalating amount of images, requiring a consequent development of a large collection of deep learning-based automatic image analysis methods. Although deep neural networks have demonstrated great performance in this field, interpretability, an essential requirement for microscopy image analysis, remains an open challenge. This work proposes a Disentangled Representation Learning (DRL) methodology to enhance model interpretability for microscopy image classification. Exploiting benchmark datasets from three different microscopic image domains (plankton, yeast vacuoles, and human cells), we show how a DRL framework, based on transferring a representation learnt from synthetic data, can provide a good trade-off between accuracy and interpretability in this domain.
comment: Published in: International Joint Conference on Neural Networks (IJCNN 2025). Project page: https://github.com/JacopoDapueto/disentangled_microscopy
☆ Towards Community-Driven Agents for Machine Learning Engineering
Large language model-based machine learning (ML) agents have shown great promise in automating ML research. However, existing agents typically operate in isolation on a given research problem, without engaging with the broader research community, where human researchers often gain insights and contribute by sharing knowledge. To bridge this gap, we introduce MLE-Live, a live evaluation framework designed to assess an agent's ability to communicate with and leverage collective knowledge from a simulated Kaggle research community. Building on this framework, we propose CoMind, a novel agent that excels at exchanging insights and developing novel solutions within a community context. CoMind achieves state-of-the-art performance on MLE-Live and outperforms 79.2% human competitors on average across four ongoing Kaggle competitions. Our code is released at https://github.com/comind-ml/CoMind.
☆ Define-ML: An Approach to Ideate Machine Learning-Enabled Systems
[Context] The increasing adoption of machine learning (ML) in software systems demands specialized ideation approaches that address ML-specific challenges, including data dependencies, technical feasibility, and alignment between business objectives and probabilistic system behavior. Traditional ideation methods like Lean Inception lack structured support for these ML considerations, which can result in misaligned product visions and unrealistic expectations. [Goal] This paper presents Define-ML, a framework that extends Lean Inception with tailored activities - Data Source Mapping, Feature-to-Data Source Mapping, and ML Mapping - to systematically integrate data and technical constraints into early-stage ML product ideation. [Method] We developed and validated Define-ML following the Technology Transfer Model, conducting both static validation (with a toy problem) and dynamic validation (in a real-world industrial case study). The analysis combined quantitative surveys with qualitative feedback, assessing utility, ease of use, and intent of adoption. [Results] Participants found Define-ML effective for clarifying data concerns, aligning ML capabilities with business goals, and fostering cross-functional collaboration. The approach's structured activities reduced ideation ambiguity, though some noted a learning curve for ML-specific components, which can be mitigated by expert facilitation. All participants expressed the intention to adopt Define-ML. [Conclusion] Define-ML provides an openly available, validated approach for ML product ideation, building on Lean Inception's agility while aligning features with available data and increasing awareness of technical feasibility.
comment: Accepted for publication at the 51st Euromicro Conference Series on Software Engineering and Advanced Applications (SEAA) 2025
☆ Weighted Mean Frequencies: a handcraft Fourier feature for 4D Flow MRI segmentation
In recent decades, the use of 4D Flow MRI images has enabled the quantification of velocity fields within a volume of interest and along the cardiac cycle. However, the lack of resolution and the presence of noise in these biomarkers are significant issues. As indicated by recent studies, it appears that biomarkers such as wall shear stress are particularly impacted by the poor resolution of vessel segmentation. The Phase Contrast Magnetic Resonance Angiography (PC-MRA) is the state-of-the-art method to facilitate segmentation. The objective of this work is to introduce a new handcraft feature that provides a novel visualisation of 4D Flow MRI images, which is useful in the segmentation task. This feature, termed Weighted Mean Frequencies (WMF), is capable of revealing the region in three dimensions where a voxel has been passed by pulsatile flow. Indeed, this feature is representative of the hull of all pulsatile velocity voxels. The value of the feature under discussion is illustrated by two experiments. The experiments involved segmenting 4D Flow MRI images using optimal thresholding and deep learning methods. The results obtained demonstrate a substantial enhancement in terms of IoU and Dice, with a respective increase of 0.12 and 0.13 in comparison with the PC-MRA feature, as evidenced by the deep learning task. This feature has the potential to yield valuable insights that could inform future segmentation processes in other vascular regions, such as the heart or the brain.
☆ Deciphering GunType Hierarchy through Acoustic Analysis of Gunshot Recordings
The escalating rates of gun-related violence and mass shootings represent a significant threat to public safety. Timely and accurate information for law enforcement agencies is crucial in mitigating these incidents. Current commercial gunshot detection systems, while effective, often come with prohibitive costs. This research explores a cost-effective alternative by leveraging acoustic analysis of gunshot recordings, potentially obtainable from ubiquitous devices like cell phones, to not only detect gunshots but also classify the type of firearm used. This paper details a study on deciphering gun type hierarchies using a curated dataset of 3459 recordings. We investigate the fundamental acoustic characteristics of gunshots, including muzzle blasts and shockwaves, which vary based on firearm type, ammunition, and shooting direction. We propose and evaluate machine learning frameworks, including Support Vector Machines (SVMs) as a baseline and a more advanced Convolutional Neural Network (CNN) architecture for joint gunshot detection and gun type classification. Results indicate that our deep learning approach achieves a mean average precision (mAP) of 0.58 on clean labeled data, outperforming the SVM baseline (mAP 0.39). Challenges related to data quality, environmental noise, and the generalization capabilities when using noisy web-sourced data (mAP 0.35) are also discussed. The long-term vision is to develop a highly accurate, real-time system deployable on common recording devices, significantly reducing detection costs and providing critical intelligence to first responders.
comment: 4 pages + 1 References
☆ AI Assistants to Enhance and Exploit the PETSc Knowledge Base
Generative AI, especially through large language models (LLMs), is transforming how technical knowledge can be accessed, reused, and extended. PETSc, a widely used numerical library for high-performance scientific computing, has accumulated a rich but fragmented knowledge base over its three decades of development, spanning source code, documentation, mailing lists, GitLab issues, Discord conversations, technical papers, and more. Much of this knowledge remains informal and inaccessible to users and new developers. To activate and utilize this knowledge base more effectively, the PETSc team has begun building an LLM-powered system that combines PETSc content with custom LLM tools -- including retrieval-augmented generation (RAG), reranking algorithms, and chatbots -- to assist users, support developers, and propose updates to formal documentation. This paper presents initial experiences designing and evaluating these tools, focusing on system architecture, using RAG and reranking for PETSc-specific information, evaluation methodologies for various LLMs and embedding models, and user interface design. Leveraging the Argonne Leadership Computing Facility resources, we analyze how LLM responses can enhance the development and use of numerical software, with an initial focus on scalable Krylov solvers. Our goal is to establish an extensible framework for knowledge-centered AI in scientific software, enabling scalable support, enriched documentation, and enhanced workflows for research and development. We conclude by outlining directions for expanding this system into a robust, evolving platform that advances software ecosystems to accelerate scientific discovery.
☆ CogGen: A Learner-Centered Generative AI Architecture for Intelligent Tutoring with Programming Video
We introduce CogGen, a learner-centered AI architecture that transforms programming videos into interactive, adaptive learning experiences by integrating student modeling with generative AI tutoring based on the Cognitive Apprenticeship framework. The architecture consists of three components: (1) video segmentation by learning goals, (2) a conversational tutoring engine applying Cognitive Apprenticeship strategies, and (3) a student model using Bayesian Knowledge Tracing to adapt instruction. Our technical evaluation demonstrates effective video segmentation accuracy and strong pedagogical alignment across knowledge, method, action, and interaction layers. Ablation studies confirm the necessity of each component in generating effective guidance. This work advances AI-powered tutoring by bridging structured student modeling with interactive AI conversations, offering a scalable approach to enhancing video-based programming education.
☆ Fine-Tuning and Prompt Engineering of LLMs, for the Creation of Multi-Agent AI for Addressing Sustainable Protein Production Challenges
The global demand for sustainable protein sources has accelerated the need for intelligent tools that can rapidly process and synthesise domain-specific scientific knowledge. In this study, we present a proof-of-concept multi-agent Artificial Intelligence (AI) framework designed to support sustainable protein production research, with an initial focus on microbial protein sources. Our Retrieval-Augmented Generation (RAG)-oriented system consists of two GPT-based LLM agents: (1) a literature search agent that retrieves relevant scientific literature on microbial protein production for a specified microbial strain, and (2) an information extraction agent that processes the retrieved content to extract relevant biological and chemical information. Two parallel methodologies, fine-tuning and prompt engineering, were explored for agent optimisation. Both methods demonstrated effectiveness at improving the performance of the information extraction agent in terms of transformer-based cosine similarity scores between obtained and ideal outputs. Mean cosine similarity scores were increased by up to 25%, while universally reaching mean scores of $\geq 0.89$ against ideal output text. Fine-tuning overall improved the mean scores to a greater extent (consistently of $\geq 0.94$) compared to prompt engineering, although lower statistical uncertainties were observed with the latter approach. A user interface was developed and published for enabling the use of the multi-agent AI system, alongside preliminary exploration of additional chemical safety-based search capabilities
☆ AI in the Writing Process: How Purposeful AI Support Fosters Student Writing
The ubiquity of technologies like ChatGPT has raised concerns about their impact on student writing, particularly regarding reduced learner agency and superficial engagement with content. While standalone chat-based LLMs often produce suboptimal writing outcomes, evidence suggests that purposefully designed AI writing support tools can enhance the writing process. This paper investigates how different AI support approaches affect writers' sense of agency and depth of knowledge transformation. Through a randomized control trial with 90 undergraduate students, we compare three conditions: (1) a chat-based LLM writing assistant, (2) an integrated AI writing tool to support diverse subprocesses, and (3) a standard writing interface (control). Our findings demonstrate that, among AI-supported conditions, students using the integrated AI writing tool exhibited greater agency over their writing process and engaged in deeper knowledge transformation overall. These results suggest that thoughtfully designed AI writing support targeting specific aspects of the writing process can help students maintain ownership of their work while facilitating improved engagement with content.
☆ Dense Video Captioning using Graph-based Sentence Summarization
Recently, dense video captioning has made attractive progress in detecting and captioning all events in a long untrimmed video. Despite promising results were achieved, most existing methods do not sufficiently explore the scene evolution within an event temporal proposal for captioning, and therefore perform less satisfactorily when the scenes and objects change over a relatively long proposal. To address this problem, we propose a graph-based partition-and-summarization (GPaS) framework for dense video captioning within two stages. For the ``partition" stage, a whole event proposal is split into short video segments for captioning at a finer level. For the ``summarization" stage, the generated sentences carrying rich description information for each segment are summarized into one sentence to describe the whole event. We particularly focus on the ``summarization" stage, and propose a framework that effectively exploits the relationship between semantic words for summarization. We achieve this goal by treating semantic words as nodes in a graph and learning their interactions by coupling Graph Convolutional Network (GCN) and Long Short Term Memory (LSTM), with the aid of visual cues. Two schemes of GCN-LSTM Interaction (GLI) modules are proposed for seamless integration of GCN and LSTM. The effectiveness of our approach is demonstrated via an extensive comparison with the state-of-the-arts methods on the two benchmarks ActivityNet Captions dataset and YouCook II dataset.
comment: 12 pages
☆ Causal Representation Learning with Observational Grouping for CXR Classification
Identifiable causal representation learning seeks to uncover the true causal relationships underlying a data generation process. In medical imaging, this presents opportunities to improve the generalisability and robustness of task-specific latent features. This work introduces the concept of grouping observations to learn identifiable representations for disease classification in chest X-rays via an end-to-end framework. Our experiments demonstrate that these causal representations improve generalisability and robustness across multiple classification tasks when grouping is used to enforce invariance w.r.t race, sex, and imaging views.
☆ Vulnerability Disclosure through Adaptive Black-Box Adversarial Attacks on NIDS
Adversarial attacks, wherein slight inputs are carefully crafted to mislead intelligent models, have attracted increasing attention. However, a critical gap persists between theoretical advancements and practical application, particularly in structured data like network traffic, where interdependent features complicate effective adversarial manipulations. Moreover, ambiguity in current approaches restricts reproducibility and limits progress in this field. Hence, existing defenses often fail to handle evolving adversarial attacks. This paper proposes a novel approach for black-box adversarial attacks, that addresses these limitations. Unlike prior work, which often assumes system access or relies on repeated probing, our method strictly respect black-box constraints, reducing interaction to avoid detection and better reflect real-world scenarios. We present an adaptive feature selection strategy using change-point detection and causality analysis to identify and target sensitive features to perturbations. This lightweight design ensures low computational cost and high deployability. Our comprehensive experiments show the attack's effectiveness in evading detection with minimal interaction, enhancing its adaptability and applicability in real-world scenarios. By advancing the understanding of adversarial attacks in network traffic, this work lays a foundation for developing robust defenses.
☆ Show, Tell and Summarize: Dense Video Captioning Using Visual Cue Aided Sentence Summarization
In this work, we propose a division-and-summarization (DaS) framework for dense video captioning. After partitioning each untrimmed long video as multiple event proposals, where each event proposal consists of a set of short video segments, we extract visual feature (e.g., C3D feature) from each segment and use the existing image/video captioning approach to generate one sentence description for this segment. Considering that the generated sentences contain rich semantic descriptions about the whole event proposal, we formulate the dense video captioning task as a visual cue aided sentence summarization problem and propose a new two stage Long Short Term Memory (LSTM) approach equipped with a new hierarchical attention mechanism to summarize all generated sentences as one descriptive sentence with the aid of visual features. Specifically, the first-stage LSTM network takes all semantic words from the generated sentences and the visual features from all segments within one event proposal as the input, and acts as the encoder to effectively summarize both semantic and visual information related to this event proposal. The second-stage LSTM network takes the output from the first-stage LSTM network and the visual features from all video segments within one event proposal as the input, and acts as the decoder to generate one descriptive sentence for this event proposal. Our comprehensive experiments on the ActivityNet Captions dataset demonstrate the effectiveness of our newly proposed DaS framework for dense video captioning.
comment: 10 pages
☆ DeepQuark: deep-neural-network approach to multiquark bound states
For the first time, we implement the deep-neural-network-based variational Monte Carlo approach for the multiquark bound states, whose complexity surpasses that of electron or nucleon systems due to strong SU(3) color interactions. We design a novel and high-efficiency architecture, DeepQuark, to address the unique challenges in multiquark systems such as stronger correlations, extra discrete quantum numbers, and intractable confinement interaction. Our method demonstrates competitive performance with state-of-the-art approaches, including diffusion Monte Carlo and Gaussian expansion method, in the nucleon, doubly heavy tetraquark, and fully heavy tetraquark systems. Notably, it outperforms existing calculations for pentaquarks, exemplified by the triply heavy pentaquark. For the nucleon, we successfully incorporate three-body flux-tube confinement interactions without additional computational costs. In tetraquark systems, we consistently describe hadronic molecule $T_{cc}$ and compact tetraquark $T_{bb}$ with an unbiased form of wave function ansatz. In the pentaquark sector, we obtain weakly bound $\bar D^*\Xi_{cc}^*$ molecule $P_{cc\bar c}(5715)$ with $S=\frac{5}{2}$ and its bottom partner $P_{bb\bar b}(15569)$. They can be viewed as the analogs of the molecular $T_{cc}$. We recommend experimental search of $P_{cc\bar c}(5715)$ in the D-wave $J/\psi \Lambda_c$ channel. DeepQuark holds great promise for extension to larger multiquark systems, overcoming the computational barriers in conventional methods. It also serves as a powerful framework for exploring confining mechanism beyond two-body interactions in multiquark states, which may offer valuable insights into nonperturbative QCD and general many-body physics.
comment: 10 pages, 3 figures, 6 tables
☆ Large Language Model-Driven Code Compliance Checking in Building Information Modeling
This research addresses the time-consuming and error-prone nature of manual code compliance checking in Building Information Modeling (BIM) by introducing a Large Language Model (LLM)-driven approach to semi-automate this critical process. The developed system integrates LLMs such as GPT, Claude, Gemini, and Llama, with Revit software to interpret building codes, generate Python scripts, and perform semi-automated compliance checks within the BIM environment. Case studies on a single-family residential project and an office building project demonstrated the system's ability to reduce the time and effort required for compliance checks while improving accuracy. It streamlined the identification of violations, such as non-compliant room dimensions, material usage, and object placements, by automatically assessing relationships and generating actionable reports. Compared to manual methods, the system eliminated repetitive tasks, simplified complex regulations, and ensured reliable adherence to standards. By offering a comprehensive, adaptable, and cost-effective solution, this proposed approach offers a promising advancement in BIM-based compliance checking, with potential applications across diverse regulatory documents in construction projects.
☆ Pay Less Attention to Deceptive Artifacts: Robust Detection of Compressed Deepfakes on Online Social Networks
With the rapid advancement of deep learning, particularly through generative adversarial networks (GANs) and diffusion models (DMs), AI-generated images, or ``deepfakes", have become nearly indistinguishable from real ones. These images are widely shared across Online Social Networks (OSNs), raising concerns about their misuse. Existing deepfake detection methods overlook the ``block effects" introduced by compression in OSNs, which obscure deepfake artifacts, and primarily focus on raw images, rarely encountered in real-world scenarios. To address these challenges, we propose PLADA (Pay Less Attention to Deceptive Artifacts), a novel framework designed to tackle the lack of paired data and the ineffective use of compressed images. PLADA consists of two core modules: Block Effect Eraser (B2E), which uses a dual-stage attention mechanism to handle block effects, and Open Data Aggregation (ODA), which processes both paired and unpaired data to improve detection. Extensive experiments across 26 datasets demonstrate that PLADA achieves a remarkable balance in deepfake detection, outperforming SoTA methods in detecting deepfakes on OSNs, even with limited paired data and compression. More importantly, this work introduces the ``block effect" as a critical factor in deepfake detection, providing a robust solution for open-world scenarios. Our code is available at https://github.com/ManyiLee/PLADA.
comment: 20 pages, 10 figures
☆ When Life Gives You Samples: The Benefits of Scaling up Inference Compute for Multilingual LLMs
Recent advancements in large language models (LLMs) have shifted focus toward scaling inference-time compute, improving performance without retraining the model. A common approach is to sample multiple outputs in parallel, and select one of these as the final output. However, work to date has focused on English and a handful of domains such as math and code. In contrast, we are most interested in techniques that generalize across open-ended tasks, formally verifiable tasks, and across languages. In this work, we study how to robustly scale inference-time compute for open-ended generative tasks in a multilingual, multi-task setting. Our findings show that both sampling strategy based on temperature variation and selection strategy must be adapted to account for diverse domains and varied language settings. We evaluate existing selection methods, revealing that strategies effective in English often fail to generalize across languages. We propose novel sampling and selection strategies specifically adapted for multilingual and multi-task inference scenarios, and show they yield notable gains across languages and tasks. In particular, our combined sampling and selection methods lead to an average +6.8 jump in win-rates for our 8B models on m-ArenaHard-v2.0 prompts, against proprietary models such as Gemini. At larger scale, Command-A (111B model) equipped with our methods, shows +9.0 improvement in win-rates on the same benchmark with just five samples against single-sample decoding, a substantial increase at minimal cost. Our results underscore the need for language- and task-aware approaches to inference-time compute, aiming to democratize performance improvements in underrepresented languages.
☆ WattsOnAI: Measuring, Analyzing, and Visualizing Energy and Carbon Footprint of AI Workloads
The rapid advancement of AI, particularly large language models (LLMs), has raised significant concerns about the energy use and carbon emissions associated with model training and inference. However, existing tools for measuring and reporting such impacts are often fragmented, lacking systematic metric integration and offering limited support for correlation analysis among them. This paper presents WattsOnAI, a comprehensive software toolkit for the measurement, analysis, and visualization of energy use, power draw, hardware performance, and carbon emissions across AI workloads. By seamlessly integrating with existing AI frameworks, WattsOnAI offers standardized reports and exports fine-grained time-series data to support benchmarking and reproducibility in a lightweight manner. It further enables in-depth correlation analysis between hardware metrics and model performance and thus facilitates bottleneck identification and performance enhancement. By addressing critical limitations in existing tools, WattsOnAI encourages the research community to weigh environmental impact alongside raw performance of AI workloads and advances the shift toward more sustainable "Green AI" practices. The code is available at https://github.com/SusCom-Lab/WattsOnAI.
comment: 11 pages, 7 figures and 5 tables
☆ Case-based Reasoning Augmented Large Language Model Framework for Decision Making in Realistic Safety-Critical Driving Scenarios
Driving in safety-critical scenarios requires quick, context-aware decision-making grounded in both situational understanding and experiential reasoning. Large Language Models (LLMs), with their powerful general-purpose reasoning capabilities, offer a promising foundation for such decision-making. However, their direct application to autonomous driving remains limited due to challenges in domain adaptation, contextual grounding, and the lack of experiential knowledge needed to make reliable and interpretable decisions in dynamic, high-risk environments. To address this gap, this paper presents a Case-Based Reasoning Augmented Large Language Model (CBR-LLM) framework for evasive maneuver decision-making in complex risk scenarios. Our approach integrates semantic scene understanding from dashcam video inputs with the retrieval of relevant past driving cases, enabling LLMs to generate maneuver recommendations that are both context-sensitive and human-aligned. Experiments across multiple open-source LLMs show that our framework improves decision accuracy, justification quality, and alignment with human expert behavior. Risk-aware prompting strategies further enhance performance across diverse risk types, while similarity-based case retrieval consistently outperforms random sampling in guiding in-context learning. Case studies further demonstrate the framework's robustness in challenging real-world conditions, underscoring its potential as an adaptive and trustworthy decision-support tool for intelligent driving systems.
comment: 12 pages, 10 figures, under-review conference
☆ Industrial Energy Disaggregation with Digital Twin-generated Dataset and Efficient Data Augmentation
Industrial Non-Intrusive Load Monitoring (NILM) is limited by the scarcity of high-quality datasets and the complex variability of industrial energy consumption patterns. To address data scarcity and privacy issues, we introduce the Synthetic Industrial Dataset for Energy Disaggregation (SIDED), an open-source dataset generated using Digital Twin simulations. SIDED includes three types of industrial facilities across three different geographic locations, capturing diverse appliance behaviors, weather conditions, and load profiles. We also propose the Appliance-Modulated Data Augmentation (AMDA) method, a computationally efficient technique that enhances NILM model generalization by intelligently scaling appliance power contributions based on their relative impact. We show in experiments that NILM models trained with AMDA-augmented data significantly improve the disaggregation of energy consumption of complex industrial appliances like combined heat and power systems. Specifically, in our out-of-sample scenarios, models trained with AMDA achieved a Normalized Disaggregation Error of 0.093, outperforming models trained without data augmentation (0.451) and those trained with random data augmentation (0.290). Data distribution analyses confirm that AMDA effectively aligns training and test data distributions, enhancing model generalization.
☆ OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
Different base language model families, such as Llama and Qwen, exhibit divergent behaviors during post-training with reinforcement learning (RL), especially on reasoning-intensive tasks. What makes a base language model suitable for reinforcement learning? Gaining deeper insight into this question is essential for developing RL-scalable foundation models of the next generation. In this work, we investigate how mid-training strategies shape RL dynamics, focusing on two representative model families: Qwen and Llama. Our study reveals that (1) high-quality mathematical corpora, such as MegaMath-Web-Pro, significantly improve both base model and RL performance, while existing alternatives (e.g., FineMath-4plus) fail to do so; (2) further adding QA-style data, particularly long chain-of-thought (CoT) reasoning examples, enhances RL outcomes, and instruction data further unlocks this effect; (3) while long-CoT improves reasoning depth, it can also induce verbosity of model responses and unstability of RL training, underscoring the importance of data formatting; (4) scaling mid-training consistently leads to stronger downstream RL performance. Building on these insights, we introduce a two-stage mid-training strategy, Stable-then-Decay, in which base models are first trained on 200B tokens with a constant learning rate, followed by 20B tokens across three CoT-focused branches with learning rate decay. This yields OctoThinker, a family of models demonstrating strong RL compatibility and closing the performance gap with more RL-friendly model families, i.e., Qwen. We hope our work will help shape pre-training strategies for foundation models in the RL era. To support further research, we release our open-source models along with a curated math reasoning-intensive corpus of over 70 billion tokens (i.e., MegaMath-Web-Pro-Max).
comment: 26 pages; The first three authors contribute to this work equally
☆ Engineering Sentience
We spell out a definition of sentience that may be useful for designing and building it in machines. We propose that for sentience to be meaningful for AI, it must be fleshed out in functional, computational terms, in enough detail to allow for implementation. Yet, this notion of sentience must also reflect something essentially 'subjective', beyond just having the general capacity to encode perceptual content. For this specific functional notion of sentience to occur, we propose that certain sensory signals need to be both assertoric (persistent) and qualitative. To illustrate the definition in more concrete terms, we sketch out some ways for potential implementation, given current technology. Understanding what it takes for artificial agents to be functionally sentient can also help us avoid creating them inadvertently, or at least, realize that we have created them in a timely manner.
☆ ReCode: Updating Code API Knowledge with Reinforcement Learning
Large Language Models (LLMs) exhibit remarkable code generation capabilities but falter when adapting to frequent updates in external library APIs. This critical limitation, stemming from reliance on outdated API knowledge from their training data, even with access to current documentation, impedes reliable code generation in dynamic environments. To tackle this issue, we propose ReCode (rule-based Reinforcement learning for Code Update), a novel framework that mimics human programmer adaptation to API changes. Specifically, we construct a dataset of approximately 2,000 data entries to train the LLMs to perform version migration based on updated information. Then, we introduce a modified string similarity metric for code evaluation as the reward for reinforcement learning. Our experiments demonstrate that ReCode substantially boosts LLMs' code generation performance in dynamic API scenarios, especially on the unseen CodeUpdateArena task. Crucially, compared to supervised fine-tuning, ReCode has less impact on LLMs' general code generation abilities. We apply ReCode on various LLMs and reinforcement learning algorithms (GRPO and DAPO), all achieving consistent improvements. Notably, after training, Qwen2.5-Coder-7B outperforms that of the 32B parameter code instruction-tuned model and the reasoning model with the same architecture. Code is available at https://github.com/zjunlp/ReCode.
comment: Work in progress
☆ Mixtures of Neural Cellular Automata: A Stochastic Framework for Growth Modelling and Self-Organization
Neural Cellular Automata (NCAs) are a promising new approach to model self-organizing processes, with potential applications in life science. However, their deterministic nature limits their ability to capture the stochasticity of real-world biological and physical systems. We propose the Mixture of Neural Cellular Automata (MNCA), a novel framework incorporating the idea of mixture models into the NCA paradigm. By combining probabilistic rule assignments with intrinsic noise, MNCAs can model diverse local behaviors and reproduce the stochastic dynamics observed in biological processes. We evaluate the effectiveness of MNCAs in three key domains: (1) synthetic simulations of tissue growth and differentiation, (2) image morphogenesis robustness, and (3) microscopy image segmentation. Results show that MNCAs achieve superior robustness to perturbations, better recapitulate real biological growth patterns, and provide interpretable rule segmentation. These findings position MNCAs as a promising tool for modeling stochastic dynamical systems and studying self-growth processes.
☆ Counterfactual Influence as a Distributional Quantity ICML 2025
Machine learning models are known to memorize samples from their training data, raising concerns around privacy and generalization. Counterfactual self-influence is a popular metric to study memorization, quantifying how the model's prediction for a sample changes depending on the sample's inclusion in the training dataset. However, recent work has shown memorization to be affected by factors beyond self-influence, with other training samples, in particular (near-)duplicates, having a large impact. We here study memorization treating counterfactual influence as a distributional quantity, taking into account how all training samples influence how a sample is memorized. For a small language model, we compute the full influence distribution of training samples on each other and analyze its properties. We find that solely looking at self-influence can severely underestimate tangible risks associated with memorization: the presence of (near-)duplicates seriously reduces self-influence, while we find these samples to be (near-)extractable. We observe similar patterns for image classification, where simply looking at the influence distributions reveals the presence of near-duplicates in CIFAR-10. Our findings highlight that memorization stems from complex interactions across training data and is better captured by the full influence distribution than by self-influence alone.
comment: Workshop on The Impact of Memorization on Trustworthy Foundation Models (MemFM) @ ICML 2025
☆ Automatic Demonstration Selection for LLM-based Tabular Data Classification
A fundamental question in applying In-Context Learning (ICL) for tabular data classification is how to determine the ideal number of demonstrations in the prompt. This work addresses this challenge by presenting an algorithm to automatically select a reasonable number of required demonstrations. Our method distinguishes itself by integrating not only the tabular data's distribution but also the user's selected prompt template and the specific Large Language Model (LLM) into its estimation. Rooted in Spectral Graph Theory, our proposed algorithm defines a novel metric to quantify the similarities between different demonstrations. We then construct a similarity graph and analyze the eigenvalues of its Laplacian to derive the minimum number of demonstrations capable of representing the data within the LLM's intrinsic representation space. We validate the efficacy of our approach through experiments comparing its performance against conventional random selection algorithms on diverse datasets and LLMs.
☆ An Agentic System for Rare Disease Diagnosis with Traceable Reasoning
Rare diseases collectively affect over 300 million individuals worldwide, yet timely and accurate diagnosis remains a pervasive challenge. This is largely due to their clinical heterogeneity, low individual prevalence, and the limited familiarity most clinicians have with rare conditions. Here, we introduce DeepRare, the first rare disease diagnosis agentic system powered by a large language model (LLM), capable of processing heterogeneous clinical inputs. The system generates ranked diagnostic hypotheses for rare diseases, each accompanied by a transparent chain of reasoning that links intermediate analytic steps to verifiable medical evidence. DeepRare comprises three key components: a central host with a long-term memory module; specialized agent servers responsible for domain-specific analytical tasks integrating over 40 specialized tools and web-scale, up-to-date medical knowledge sources, ensuring access to the most current clinical information. This modular and scalable design enables complex diagnostic reasoning while maintaining traceability and adaptability. We evaluate DeepRare on eight datasets. The system demonstrates exceptional diagnostic performance among 2,919 diseases, achieving 100% accuracy for 1013 diseases. In HPO-based evaluations, DeepRare significantly outperforms other 15 methods, like traditional bioinformatics diagnostic tools, LLMs, and other agentic systems, achieving an average Recall@1 score of 57.18% and surpassing the second-best method (Reasoning LLM) by a substantial margin of 23.79 percentage points. For multi-modal input scenarios, DeepRare achieves 70.60% at Recall@1 compared to Exomiser's 53.20% in 109 cases. Manual verification of reasoning chains by clinical experts achieves 95.40% agreements. Furthermore, the DeepRare system has been implemented as a user-friendly web application http://raredx.cn/doctor.
☆ Off-Policy Evaluation and Learning for the Future under Non-Stationarity
We study the novel problem of future off-policy evaluation (F-OPE) and learning (F-OPL) for estimating and optimizing the future value of policies in non-stationary environments, where distributions vary over time. In e-commerce recommendations, for instance, our goal is often to estimate and optimize the policy value for the upcoming month using data collected by an old policy in the previous month. A critical challenge is that data related to the future environment is not observed in the historical data. Existing methods assume stationarity or depend on restrictive reward-modeling assumptions, leading to significant bias. To address these limitations, we propose a novel estimator named \textit{\textbf{O}ff-\textbf{P}olicy Estimator for the \textbf{F}uture \textbf{V}alue (\textbf{\textit{OPFV}})}, designed for accurately estimating policy values at any future time point. The key feature of OPFV is its ability to leverage the useful structure within time-series data. While future data might not be present in the historical log, we can leverage, for example, seasonal, weekly, or holiday effects that are consistent in both the historical and future data. Our estimator is the first to exploit these time-related structures via a new type of importance weighting, enabling effective F-OPE. Theoretical analysis identifies the conditions under which OPFV becomes low-bias. In addition, we extend our estimator to develop a new policy-gradient method to proactively learn a good future policy using only historical data. Empirical results show that our methods substantially outperform existing methods in estimating and optimizing the future policy value under non-stationarity for various experimental setups.
☆ SV-LLM: An Agentic Approach for SoC Security Verification using Large Language Models
Ensuring the security of complex system-on-chips (SoCs) designs is a critical imperative, yet traditional verification techniques struggle to keep pace due to significant challenges in automation, scalability, comprehensiveness, and adaptability. The advent of large language models (LLMs), with their remarkable capabilities in natural language understanding, code generation, and advanced reasoning, presents a new paradigm for tackling these issues. Moving beyond monolithic models, an agentic approach allows for the creation of multi-agent systems where specialized LLMs collaborate to solve complex problems more effectively. Recognizing this opportunity, we introduce SV-LLM, a novel multi-agent assistant system designed to automate and enhance SoC security verification. By integrating specialized agents for tasks like verification question answering, security asset identification, threat modeling, test plan and property generation, vulnerability detection, and simulation-based bug validation, SV-LLM streamlines the workflow. To optimize their performance in these diverse tasks, agents leverage different learning paradigms, such as in-context learning, fine-tuning, and retrieval-augmented generation (RAG). The system aims to reduce manual intervention, improve accuracy, and accelerate security analysis, supporting proactive identification and mitigation of risks early in the design cycle. We demonstrate its potential to transform hardware security practices through illustrative case studies and experiments that showcase its applicability and efficacy.
☆ Client Clustering Meets Knowledge Sharing: Enhancing Privacy and Robustness in Personalized Peer-to-Peer Learning
The growing adoption of Artificial Intelligence (AI) in Internet of Things (IoT) ecosystems has intensified the need for personalized learning methods that can operate efficiently and privately across heterogeneous, resource-constrained devices. However, enabling effective personalized learning in decentralized settings introduces several challenges, including efficient knowledge transfer between clients, protection of data privacy, and resilience against poisoning attacks. In this paper, we address these challenges by developing P4 (Personalized, Private, Peer-to-Peer) -- a method designed to deliver personalized models for resource-constrained IoT devices while ensuring differential privacy and robustness against poisoning attacks. Our solution employs a lightweight, fully decentralized algorithm to privately detect client similarity and form collaborative groups. Within each group, clients leverage differentially private knowledge distillation to co-train their models, maintaining high accuracy while ensuring robustness to the presence of malicious clients. We evaluate P4 on popular benchmark datasets using both linear and CNN-based architectures across various heterogeneity settings and attack scenarios. Experimental results show that P4 achieves 5% to 30% higher accuracy than leading differentially private peer-to-peer approaches and maintains robustness with up to 30% malicious clients. Additionally, we demonstrate its practicality by deploying it on resource-constrained devices, where collaborative training between two clients adds only ~7 seconds of overhead.
☆ GymPN: A Library for Decision-Making in Process Management Systems
Process management systems support key decisions about the way work is allocated in organizations. This includes decisions on which task to perform next, when to execute the task, and who to assign the task to. Suitable software tools are required to support these decisions in a way that is optimal for the organization. This paper presents a software library, called GymPN, that supports optimal decision-making in business processes using Deep Reinforcement Learning. GymPN builds on previous work that supports task assignment in business processes, introducing two key novelties: support for partial process observability and the ability to model multiple decisions in a business process. These novel elements address fundamental limitations of previous work and thus enable the representation of more realistic process decisions. We evaluate the library on eight typical business process decision-making problem patterns, showing that GymPN allows for easy modeling of the desired problems, as well as learning optimal decision policies.
☆ Smart Ride and Delivery Services with Electric Vehicles: Leveraging Bidirectional Charging for Profit Optimisation
With the rising popularity of electric vehicles (EVs), modern service systems, such as ride-hailing delivery services, are increasingly integrating EVs into their operations. Unlike conventional vehicles, EVs often have a shorter driving range, necessitating careful consideration of charging when fulfilling requests. With recent advances in Vehicle-to-Grid (V2G) technology - allowing EVs to also discharge energy back to the grid - new opportunities and complexities emerge. We introduce the Electric Vehicle Orienteering Problem with V2G (EVOP-V2G): a profit-maximization problem where EV drivers must select customer requests or orders while managing when and where to charge or discharge. This involves navigating dynamic electricity prices, charging station selection, and route constraints. We formulate the problem as a Mixed Integer Programming (MIP) model and propose two near-optimal metaheuristic algorithms: one evolutionary (EA) and the other based on large neighborhood search (LNS). Experiments on real-world data show our methods can double driver profits compared to baselines, while maintaining near-optimal performance on small instances and excellent scalability on larger ones. Our work highlights a promising path toward smarter, more profitable EV-based mobility systems that actively support the energy grid.
☆ Paladin-mini: A Compact and Efficient Grounding Model Excelling in Real-World Scenarios
This paper introduces two significant contributions to address the issue of grounding claims in a given context. Grounding means that given a context (document) and a claim, there's at least one supportive evidence for the claim in the document. We will introduce Paladin-mini, a compact (3.8B parameters) open-source classifier model (used for labeling data as grounded or ungrounded) engineered for robust performance in real-world scenarios, and the grounding-benchmark, a new evaluation dataset designed to assess performance on critical reasoning tasks. We'll also demonstrate the results of Paladin-mini with benchmarks against the current State-of-the-art and share clear and reproducible results.
comment: 6 pages, 2 figures
☆ CARMA: Context-Aware Situational Grounding of Human-Robot Group Interactions by Combining Vision-Language Models with Object and Action Recognition
We introduce CARMA, a system for situational grounding in human-robot group interactions. Effective collaboration in such group settings requires situational awareness based on a consistent representation of present persons and objects coupled with an episodic abstraction of events regarding actors and manipulated objects. This calls for a clear and consistent assignment of instances, ensuring that robots correctly recognize and track actors, objects, and their interactions over time. To achieve this, CARMA uniquely identifies physical instances of such entities in the real world and organizes them into grounded triplets of actors, objects, and actions. To validate our approach, we conducted three experiments, where multiple humans and a robot interact: collaborative pouring, handovers, and sorting. These scenarios allow the assessment of the system's capabilities as to role distinction, multi-actor awareness, and consistent instance identification. Our experiments demonstrate that the system can reliably generate accurate actor-action-object triplets, providing a structured and robust foundation for applications requiring spatiotemporal reasoning and situated decision-making in collaborative settings.
Self-Supervised Graph Learning via Spectral Bootstrapping and Laplacian-Based Augmentations
We present LaplaceGNN, a novel self-supervised graph learning framework that bypasses the need for negative sampling by leveraging spectral bootstrapping techniques. Our method integrates Laplacian-based signals into the learning process, allowing the model to effectively capture rich structural representations without relying on contrastive objectives or handcrafted augmentations. By focusing on positive alignment, LaplaceGNN achieves linear scaling while offering a simpler, more efficient, self-supervised alternative for graph neural networks, applicable across diverse domains. Our contributions are twofold: we precompute spectral augmentations through max-min centrality-guided optimization, enabling rich structural supervision without relying on handcrafted augmentations, then we integrate an adversarial bootstrapped training scheme that further strengthens feature learning and robustness. Our extensive experiments on different benchmark datasets show that LaplaceGNN achieves superior performance compared to state-of-the-art self-supervised graph methods, offering a promising direction for efficiently learning expressive graph representations.
comment: LaplaceGNN is a novel graph learning framework that employs a bootstrapped teacher-student architecture. Its precomputed spectral augmentations and adversarial training enable robust performance, outperforming SOTA methods while scaling linearly
☆ Tabular Feature Discovery With Reasoning Type Exploration
Feature engineering for tabular data remains a critical yet challenging step in machine learning. Recently, large language models (LLMs) have been used to automatically generate new features by leveraging their vast knowledge. However, existing LLM-based approaches often produce overly simple or repetitive features, partly due to inherent biases in the transformations the LLM chooses and the lack of structured reasoning guidance during generation. In this paper, we propose a novel method REFeat, which guides an LLM to discover diverse and informative features by leveraging multiple types of reasoning to steer the feature generation process. Experiments on 59 benchmark datasets demonstrate that our approach not only achieves higher predictive accuracy on average, but also discovers more diverse and meaningful features. These results highlight the promise of incorporating rich reasoning paradigms and adaptive strategy selection into LLM-driven feature discovery for tabular data.
☆ A foundation model with multi-variate parallel attention to generate neuronal activity
Learning from multi-variate time-series with heterogeneous channel configurations remains a fundamental challenge for deep neural networks (DNNs), particularly in clinical domains such as intracranial electroencephalography (iEEG), where channel setups vary widely across subjects. In this work, we introduce multi-variate parallel attention (MVPA), a novel self-attention mechanism that disentangles content, temporal, and spatial attention, enabling flexible, generalizable, and efficient modeling of time-series data with varying channel counts and configurations. We use MVPA to build MVPFormer, a generative foundation model for human electrophysiology, trained to predict the evolution of iEEG signals across diverse subjects. To support this and future effort by the community, we release the SWEC iEEG dataset, the largest publicly available iEEG dataset to date, comprising nearly 10,000 hours of recordings from heterogeneous clinical sources. MVPFormer leverages MVPA to achieve strong generalization across subjects, demonstrating expert-level performance in seizure detection and outperforming state-of-the-art Transformer baselines on our SWEC, the MAYO, and the FNUSA dataset. We further validate MVPA on standard time-series forecasting and classification tasks, where it matches or exceeds existing attention-based models. Together, our contributions establish MVPA as a general-purpose attention mechanism for heterogeneous time-series and MVPFormer as the first open-source, open-weights, and open-data iEEG foundation model with state-of-the-art clinical performance. The code is available at https://github.com/IBM/multi-variate-parallel-transformer. The SWEC iEEG dataset is available at https://mb-neuro.medical-blocks.ch/public_access/databases/ieeg/swec_ieeg.
comment: The code is available at https://github.com/IBM/multi-variate-parallel-transformer. The SWEC iEEG dataset is available at https://mb-neuro.medical-blocks.ch/public_access/databases/ieeg/swec_ieeg
☆ DipSVD: Dual-importance Protected SVD for Efficient LLM Compression
The ever-increasing computational demands and deployment costs of large language models (LLMs) have spurred numerous compressing methods. Compared to quantization and unstructured pruning, SVD compression offers superior hardware compatibility and theoretical guarantees. However, existing SVD-based methods focus on the overall discrepancy between the original and compressed matrices while overlooking the protection of critical components within the matrix, which leads to inferior performance in the compressed models. This paper proposes a dual-level importance protection mechanism to enhance SVD-based compression methods: (1) local importance protection: preserving the most critical singular vectors within each weight matrix through channel-weighted data whitening; and (2) global importance protection: enabling less important layers to bear a greater portion of the compression burden through either a heuristic or optimization-based approach, thereby minimizing the impact of compression on critical layers. Extensive experiments demonstrate that DipSVD outperforms existing SVD-based compression approaches across multiple benchmarks, achieving superior model performance especially at high model compression ratios.
☆ Feature Hallucination for Self-supervised Action Recognition
Understanding human actions in videos requires more than raw pixel analysis; it relies on high-level semantic reasoning and effective integration of multimodal features. We propose a deep translational action recognition framework that enhances recognition accuracy by jointly predicting action concepts and auxiliary features from RGB video frames. At test time, hallucination streams infer missing cues, enriching feature representations without increasing computational overhead. To focus on action-relevant regions beyond raw pixels, we introduce two novel domain-specific descriptors. Object Detection Features (ODF) aggregate outputs from multiple object detectors to capture contextual cues, while Saliency Detection Features (SDF) highlight spatial and intensity patterns crucial for action recognition. Our framework seamlessly integrates these descriptors with auxiliary modalities such as optical flow, Improved Dense Trajectories, skeleton data, and audio cues. It remains compatible with state-of-the-art architectures, including I3D, AssembleNet, Video Transformer Network, FASTER, and recent models like VideoMAE V2 and InternVideo2. To handle uncertainty in auxiliary features, we incorporate aleatoric uncertainty modeling in the hallucination step and introduce a robust loss function to mitigate feature noise. Our multimodal self-supervised action recognition framework achieves state-of-the-art performance on multiple benchmarks, including Kinetics-400, Kinetics-600, and Something-Something V2, demonstrating its effectiveness in capturing fine-grained action dynamics.
comment: Accepted for publication in International Journal of Computer Vision (IJCV)
☆ Mobile-R1: Towards Interactive Reinforcement Learning for VLM-Based Mobile Agent via Task-Level Rewards
Vision-language model-based mobile agents have gained the ability to not only understand complex instructions and mobile screenshots, but also optimize their action outputs via thinking and reasoning, benefiting from reinforcement learning, such as Group Relative Policy Optimization (GRPO). However, existing research centers on offline reinforcement learning training or online optimization using action-level rewards, which limits the agent's dynamic interaction with the environment. This often results in agents settling into local optima, thereby weakening their ability for exploration and error action correction. To address these challenges, we introduce an approach called Mobile-R1, which employs interactive multi-turn reinforcement learning with task-level rewards for mobile agents. Our training framework consists of three stages: initial format finetuning, single-step online training via action-level reward, followed by online training via task-level reward based on multi-turn trajectories. This strategy is designed to enhance the exploration and error correction capabilities of Mobile-R1, leading to significant performance improvements. Moreover, we have collected a dataset covering 28 Chinese applications with 24,521 high-quality manual annotations and established a new benchmark with 500 trajectories. We will open source all resources, including the dataset, benchmark, model weight, and codes: https://mobile-r1.github.io/Mobile-R1/.
comment: 14 pages, 12 figures
☆ Comparative Analysis of Deep Learning Models for Crop Disease Detection: A Transfer Learning Approach
This research presents the development of an Artificial Intelligence (AI) - driven crop disease detection system designed to assist farmers in rural areas with limited resources. We aim to compare different deep learning models for a comparative analysis, focusing on their efficacy in transfer learning. By leveraging deep learning models, including EfficientNet, ResNet101, MobileNetV2, and our custom CNN, which achieved a validation accuracy of 95.76%, the system effectively classifies plant diseases. This research demonstrates the potential of transfer learning in reshaping agricultural practices, improving crop health management, and supporting sustainable farming in rural environments.
☆ Beyond-Expert Performance with Limited Demonstrations: Efficient Imitation Learning with Double Exploration
Imitation learning is a central problem in reinforcement learning where the goal is to learn a policy that mimics the expert's behavior. In practice, it is often challenging to learn the expert policy from a limited number of demonstrations accurately due to the complexity of the state space. Moreover, it is essential to explore the environment and collect data to achieve beyond-expert performance. To overcome these challenges, we propose a novel imitation learning algorithm called Imitation Learning with Double Exploration (ILDE), which implements exploration in two aspects: (1) optimistic policy optimization via an exploration bonus that rewards state-action pairs with high uncertainty to potentially improve the convergence to the expert policy, and (2) curiosity-driven exploration of the states that deviate from the demonstration trajectories to potentially yield beyond-expert performance. Empirically, we demonstrate that ILDE outperforms the state-of-the-art imitation learning algorithms in terms of sample efficiency and achieves beyond-expert performance on Atari and MuJoCo tasks with fewer demonstrations than in previous work. We also provide a theoretical justification of ILDE as an uncertainty-regularized policy optimization method with optimistic exploration, leading to a regret growing sublinearly in the number of episodes.
☆ Enterprise Large Language Model Evaluation Benchmark
Large Language Models (LLMs) ) have demonstrated promise in boosting productivity across AI-powered tools, yet existing benchmarks like Massive Multitask Language Understanding (MMLU) inadequately assess enterprise-specific task complexities. We propose a 14-task framework grounded in Bloom's Taxonomy to holistically evaluate LLM capabilities in enterprise contexts. To address challenges of noisy data and costly annotation, we develop a scalable pipeline combining LLM-as-a-Labeler, LLM-as-a-Judge, and corrective retrieval-augmented generation (CRAG), curating a robust 9,700-sample benchmark. Evaluation of six leading models shows open-source contenders like DeepSeek R1 rival proprietary models in reasoning tasks but lag in judgment-based scenarios, likely due to overthinking. Our benchmark reveals critical enterprise performance gaps and offers actionable insights for model optimization. This work provides enterprises a blueprint for tailored evaluations and advances practical LLM deployment.
comment: Submitted to MLNLP 2025 at https://csity2025.org/mlnlp/index
☆ Argumentative Ensembling for Robust Recourse under Model Multiplicity
In machine learning, it is common to obtain multiple equally performing models for the same prediction task, e.g., when training neural networks with different random seeds. Model multiplicity (MM) is the situation which arises when these competing models differ in their predictions for the same input, for which ensembling is often employed to determine an aggregation of the outputs. Providing recourse recommendations via counterfactual explanations (CEs) under MM thus becomes complex, since the CE may not be valid across all models, i.e., the CEs are not robust under MM. In this work, we formalise the problem of providing recourse under MM, which we name recourse-aware ensembling (RAE). We propose the idea that under MM, CEs for each individual model should be considered alongside their predictions so that the aggregated prediction and recourse are decided in tandem. Centred around this intuition, we introduce six desirable properties for solutions to this problem. For solving RAE, we propose a novel argumentative ensembling method which guarantees the robustness of CEs under MM. Specifically, our method leverages computational argumentation to explicitly represent the conflicts between models and counterfactuals regarding prediction results and CE validity. It then uses argumentation semantics to resolve the conflicts and obtain the final solution, in a manner which is parametric to the chosen semantics. Our method also allows for the specification of preferences over the models under MM, allowing further customisation of the ensemble. In a comprehensive theoretical analysis, we characterise the behaviour of argumentative ensembling with four different argumentation semantics. We then empirically demonstrate the effectiveness of our approach in satisfying desirable properties with eight instantiations of our method. (Abstract is shortened for arXiv.)
comment: arXiv admin note: substantial text overlap with arXiv:2312.15097
☆ Generating and Customizing Robotic Arm Trajectories using Neural Networks
We introduce a neural network approach for generating and customizing the trajectory of a robotic arm, that guarantees precision and repeatability. To highlight the potential of this novel method, we describe the design and implementation of the technique and show its application in an experimental setting of cognitive robotics. In this scenario, the NICO robot was characterized by the ability to point to specific points in space with precise linear movements, increasing the predictability of the robotic action during its interaction with humans. To achieve this goal, the neural network computes the forward kinematics of the robot arm. By integrating it with a generator of joint angles, another neural network was developed and trained on an artificial dataset created from suitable start and end poses of the robotic arm. Through the computation of angular velocities, the robot was characterized by its ability to perform the movement, and the quality of its action was evaluated in terms of shape and accuracy. Thanks to its broad applicability, our approach successfully generates precise trajectories that could be customized in their shape and adapted to different settings.
comment: The code is released at https://github.com/andylucny/nico2/tree/main/generate
☆ Time-series surrogates from energy consumers generated by machine learning approaches for long-term forecasting scenarios
Forecasting attracts a lot of research attention in the electricity value chain. However, most studies concentrate on short-term forecasting of generation or consumption with a focus on systems and less on individual consumers. Even more neglected is the topic of long-term forecasting of individual power consumption. Here, we provide an in-depth comparative evaluation of data-driven methods for generating synthetic time series data tailored to energy consumption long-term forecasting. High-fidelity synthetic data is crucial for a wide range of applications, including state estimations in energy systems or power grid planning. In this study, we assess and compare the performance of multiple state-of-the-art but less common techniques: a hybrid Wasserstein Generative Adversarial Network (WGAN), Denoising Diffusion Probabilistic Model (DDPM), Hidden Markov Model (HMM), and Masked Autoregressive Bernstein polynomial normalizing Flows (MABF). We analyze the ability of each method to replicate the temporal dynamics, long-range dependencies, and probabilistic transitions characteristic of individual energy consumption profiles. Our comparative evaluation highlights the strengths and limitations of: WGAN, DDPM, HMM and MABF aiding in selecting the most suitable approach for state estimations and other energy-related tasks. Our generation and analysis framework aims to enhance the accuracy and reliability of synthetic power consumption data while generating data that fulfills criteria like anonymisation - preserving privacy concerns mitigating risks of specific profiling of single customers. This study utilizes an open-source dataset from households in Germany with 15min time resolution. The generated synthetic power profiles can readily be used in applications like state estimations or consumption forecasting.
☆ Q-resafe: Assessing Safety Risks and Quantization-aware Safety Patching for Quantized Large Language Models ICML 2025
Quantized large language models (LLMs) have gained increasing attention and significance for enabling deployment in resource-constrained environments. However, emerging studies on a few calibration dataset-free quantization methods suggest that quantization may compromise the safety capabilities of LLMs, underscoring the urgent need for systematic safety evaluations and effective mitigation strategies. In this paper, we present comprehensive safety evaluations across various mainstream quantization techniques and diverse calibration datasets, utilizing widely accepted safety benchmarks. To address the identified safety vulnerabilities, we propose a quantization-aware safety patching framework, Q-resafe, to efficiently restore the safety capabilities of quantized LLMs while minimizing any adverse impact on utility. Extensive experimental results demonstrate that Q-resafe successfully re-aligns the safety of quantized LLMs with their pre-quantization counterparts, even under challenging evaluation scenarios. Project page is available at: https://github.com/Thecommonirin/Qresafe.
comment: ICML 2025
☆ Language Modeling by Language Models
Can we leverage LLMs to model the process of discovering novel language model (LM) architectures? Inspired by real research, we propose a multi-agent LLM approach that simulates the conventional stages of research, from ideation and literature search (proposal stage) to design implementation (code generation), generative pre-training, and downstream evaluation (verification). Using ideas from scaling laws, our system, Genesys, employs a Ladder of Scales approach; new designs are proposed, adversarially reviewed, implemented, and selectively verified at increasingly larger model scales (14M$\sim$350M parameters) with a narrowing budget (the number of models we can train at each scale). To help make discovery efficient and factorizable, Genesys uses a novel genetic programming backbone, which we show has empirical advantages over commonly used direct prompt generation workflows (e.g., $\sim$86\% percentage point improvement in successful design generation, a key bottleneck). We report experiments involving 1,162 newly discovered designs (1,062 fully verified through pre-training) and find the best designs to be highly competitive with known architectures (e.g., outperform GPT2, Mamba2, etc., on 6/9 common benchmarks). We couple these results with comprehensive system-level ablations and formal results, which give broader insights into the design of effective autonomous discovery systems.
☆ FedBKD: Distilled Federated Learning to Embrace Gerneralization and Personalization on Non-IID Data
Federated learning (FL) is a decentralized collaborative machine learning (ML) technique. It provides a solution to the issues of isolated data islands and data privacy leakage in industrial ML practices. One major challenge in FL is handling the non-identical and independent distributed (non-IID) data. Current solutions either focus on constructing an all-powerful global model, or customizing personalized local models. Few of them can provide both a well-generalized global model and well-performed local models at the same time. Additionally, many FL solutions to the non-IID problem are benefited from introducing public datasets. However, this will also increase the risk of data leakage. To tackle the problems, we propose a novel data-free distillation framework, Federated Bidirectional Knowledge Distillation (FedBKD). Specifically, we train Generative Adversarial Networks (GAN) for synthetic data. During the GAN training, local models serve as discriminators and their parameters are frozen. The synthetic data is then used for bidirectional distillation between global and local models to achieve knowledge interactions so that performances for both sides are improved. We conduct extensive experiments on 4 benchmarks under different non-IID settings. The results show that FedBKD achieves SOTA performances in every case.
☆ Enhancing Large Language Models through Structured Reasoning
Recent Large Language Models (LLMs) have significantly advanced natural language processing and automated decision-making. However, these models still encounter difficulties when performing complex reasoning tasks involving logical deduction and systematic planning, primarily due to their reliance on implicit statistical relationships without structured knowledge representation.Inspired by cognitive science and neurosymbolic AI, we introduce a novel approach to enhance LLMs through explicit structured reasoning. First, we convert unstructured data into structured formats by explicitly annotating reasoning steps. We then employ this structured dataset to train LLMs through Supervised Fine-Tuning (SFT). Additionally, we enhance the structured reasoning capabilities of LLMs using Group Relative Policy Optimization (GRPO), incorporating two innovative algorithms--MAX-Flow and Longest Common Subsequence (LCS)--which notably improve reasoning effectiveness and reduce computational complexity. Experimental results from fine-tuning a DeepSeek-R1-Distill-Qwen-1.5B model demonstrate concise reasoning, robust performance across various scenarios, and improved compatibility with optimization techniques, validating the efficacy of structured reasoning integration in LLMs.
comment: Preprint. Under review
☆ Directed Link Prediction using GNN with Local and Global Feature Fusion
Link prediction is a classical problem in graph analysis with many practical applications. For directed graphs, recently developed deep learning approaches typically analyze node similarities through contrastive learning and aggregate neighborhood information through graph convolutions. In this work, we propose a novel graph neural network (GNN) framework to fuse feature embedding with community information. We theoretically demonstrate that such hybrid features can improve the performance of directed link prediction. To utilize such features efficiently, we also propose an approach to transform input graphs into directed line graphs so that nodes in the transformed graph can aggregate more information during graph convolutions. Experiments on benchmark datasets show that our approach outperforms the state-of-the-art in most cases when 30%, 40%, 50%, and 60% of the connected links are used as training data, respectively.
☆ Perspectives in Play: A Multi-Perspective Approach for More Inclusive NLP Systems
In the realm of Natural Language Processing (NLP), common approaches for handling human disagreement consist of aggregating annotators' viewpoints to establish a single ground truth. However, prior studies show that disregarding individual opinions can lead can lead to the side effect of underrepresenting minority perspectives, especially in subjective tasks, where annotators may systematically disagree because of their preferences. Recognizing that labels reflect the diverse backgrounds, life experiences, and values of individuals, this study proposes a new multi-perspective approach using soft labels to encourage the development of the next generation of perspective aware models, more inclusive and pluralistic. We conduct an extensive analysis across diverse subjective text classification tasks, including hate speech, irony, abusive language, and stance detection, to highlight the importance of capturing human disagreements, often overlooked by traditional aggregation methods. Results show that the multi-perspective approach not only better approximates human label distributions, as measured by Jensen-Shannon Divergence (JSD), but also achieves superior classification performance (higher F1 scores), outperforming traditional approaches. However, our approach exhibits lower confidence in tasks like irony and stance detection, likely due to the inherent subjectivity present in the texts. Lastly, leveraging Explainable AI (XAI), we explore model uncertainty and uncover meaningful insights into model predictions.
☆ Affective Priming Score: A Data-Driven Method to Detect Priming in Sequential Datasets
Affective priming exemplifies the challenge of ambiguity in affective computing. While the community has largely addressed this issue from a label-based perspective, identifying data points in the sequence affected by the priming effect, the impact of priming on data itself, particularly in physiological signals, remains underexplored. Data affected by priming can lead to misclassifications when used in learning models. This study proposes the Affective Priming Score (APS), a data-driven method to detect data points influenced by the priming effect. The APS assigns a score to each data point, quantifying the extent to which it is affected by priming. To validate this method, we apply it to the SEED and SEED-VII datasets, which contain sufficient transitions between emotional events to exhibit priming effects. We train models with the same configuration using both the original data and priming-free sequences. The misclassification rate is significantly reduced when using priming-free sequences compared to the original data. This work contributes to the broader challenge of ambiguity by identifying and mitigating priming effects at the data level, enhancing model robustness, and offering valuable insights for the design and collection of affective computing datasets.
☆ How to Retrieve Examples in In-context Learning to Improve Conversational Emotion Recognition using Large Language Models?
Large language models (LLMs) have enabled a wide variety of real-world applications in various domains. However, creating a high-performing application with high accuracy remains challenging, particularly for subjective tasks like emotion recognition. Inspired by the SLT 2024 GenSER Challenge, this study investigates approaches to improving conversational emotion recognition (CER) by LLMs. Specifically, we explore how to retrieve high-quality examples in in-context learning (ICL) to enhance CER. We propose various strategies based on random and augmented example retrieval and also analyze the impact of conversational context on CER accuracy. Experiments were conducted on the three datasets including IEMOCAP, MELD and EmoryNLP. The results show that augmented example retrieval consistently outperforms other techniques under investigation across all datasets, highlighting the importance of retrieving coherent targeted examples and enhancing them through paraphrasing.
☆ Zero-Shot Attribution for Large Language Models: A Distribution Testing Approach
A growing fraction of all code is sampled from Large Language Models (LLMs). We investigate the problem of attributing code generated by language models using hypothesis testing to leverage established techniques and guarantees. Given a set of samples $S$ and a suspect model $\mathcal{L}^*$, our goal is to assess the likelihood of $S$ originating from $\mathcal{L}^*$. Due to the curse of dimensionality, this is intractable when only samples from the LLM are given: to circumvent this, we use both samples and density estimates from the LLM, a form of access commonly available. We introduce $\mathsf{Anubis}$, a zero-shot attribution tool that frames attribution as a distribution testing problem. Our experiments on a benchmark of code samples show that $\mathsf{Anubis}$ achieves high AUROC scores ( $\ge0.9$) when distinguishing between LLMs like DeepSeek-Coder, CodeGemma, and Stable-Code using only $\approx 2000$ samples.
comment: 16 pages, 4 figures
☆ Progressive Alignment Degradation Learning for Pansharpening
Deep learning-based pansharpening has been shown to effectively generate high-resolution multispectral (HRMS) images. To create supervised ground-truth HRMS images, synthetic data generated using the Wald protocol is commonly employed. This protocol assumes that networks trained on artificial low-resolution data will perform equally well on high-resolution data. However, well-trained models typically exhibit a trade-off in performance between reduced-resolution and full-resolution datasets. In this paper, we delve into the Wald protocol and find that its inaccurate approximation of real-world degradation patterns limits the generalization of deep pansharpening models. To address this issue, we propose the Progressive Alignment Degradation Module (PADM), which uses mutual iteration between two sub-networks, PAlignNet and PDegradeNet, to adaptively learn accurate degradation processes without relying on predefined operators. Building on this, we introduce HFreqdiff, which embeds high-frequency details into a diffusion framework and incorporates CFB and BACM modules for frequency-selective detail extraction and precise reverse process learning. These innovations enable effective integration of high-resolution panchromatic and multispectral images, significantly enhancing spatial sharpness and quality. Experiments and ablation studies demonstrate the proposed method's superior performance compared to state-of-the-art techniques.
comment: 13 pages, 9 figures
☆ COIN: Uncertainty-Guarding Selective Question Answering for Foundation Models with Provable Risk Guarantees
Uncertainty quantification (UQ) for foundation models is essential to identify and mitigate potential hallucinations in automatically generated text. However, heuristic UQ approaches lack formal guarantees for key metrics such as the false discovery rate (FDR) in selective prediction. Previous work adopts the split conformal prediction (SCP) framework to ensure desired coverage of admissible answers by constructing prediction sets, but these sets often contain incorrect candidates, limiting their practical utility. To address this, we propose COIN, an uncertainty-guarding selection framework that calibrates statistically valid thresholds to filter a single generated answer per question under user-specified FDR constraints. COIN estimates the empirical error rate on a calibration set and applies confidence interval methods such as Clopper-Pearson to establish a high-probability upper bound on the true error rate (i.e., FDR). This enables the selection of the largest uncertainty threshold that ensures FDR control on test data while significantly increasing sample retention. We demonstrate COIN's robustness in risk control, strong test-time power in retaining admissible answers, and predictive efficiency under limited calibration data across both general and multimodal text generation tasks. Furthermore, we show that employing alternative upper bound constructions and UQ strategies can further boost COIN's power performance, which underscores its extensibility and adaptability to diverse application scenarios.
☆ Valid Selection among Conformal Sets
Conformal prediction offers a distribution-free framework for constructing prediction sets with coverage guarantees. In practice, multiple valid conformal prediction sets may be available, arising from different models or methodologies. However, selecting the most desirable set, such as the smallest, can invalidate the coverage guarantees. To address this challenge, we propose a stability-based approach that ensures coverage for the selected prediction set. We extend our results to the online conformal setting, propose several refinements in settings where additional structure is available, and demonstrate its effectiveness through experiments.
☆ SEED: A Structural Encoder for Embedding-Driven Decoding in Time Series Prediction with LLMs
Multivariate time series forecasting requires models to simultaneously capture variable-wise structural dependencies and generalize across diverse tasks. While structural encoders are effective in modeling feature interactions, they lack the capacity to support semantic-level reasoning or task adaptation. Conversely, large language models (LLMs) possess strong generalization capabilities but remain incompatible with raw time series inputs. This gap limits the development of unified, transferable prediction systems. Therefore, we introduce SEED, a structural encoder for embedding-driven decoding, which integrates four stages: a token-aware encoder for patch extraction, a projection module that aligns patches with language model embeddings, a semantic reprogramming mechanism that maps patches to task-aware prototypes, and a frozen language model for prediction. This modular architecture decouples representation learning from inference, enabling efficient alignment between numerical patterns and semantic reasoning. Empirical results demonstrate that the proposed method achieves consistent improvements over strong baselines, and comparative studies on various datasets confirm SEED's role in addressing the structural-semantic modeling gap.
☆ Do psychic cells generate consciousness?
Technological advances in the past decades have begun to enable neuroscientists to address fundamental questions about consciousness in an unprecedented way. Here we review remarkable recent progress in our understanding of cellular-level mechanisms of conscious processing in the brain. Of particular interest are the cortical pyramidal neurons -- or "psychic cells" called by Ram\'on y Cajal more than 100 years ago -- which have an intriguing cellular mechanism that accounts for selective disruption of feedback signaling in the brain upon anesthetic-induced loss of consciousness. Importantly, a particular class of metabotropic receptors distributed over the dendrites of pyramidal cells are highlighted as the key cellular mechanism. After all, Cajal's instinct over a century ago may turn out to be correct -- we may have just begun to understand whether and how psychic cells indeed generate and control our consciousness.
☆ AI and Agile Software Development: From Frustration to Success -- XP2025 Workshop Summary
The full-day workshop on AI and Agile at XP 2025 convened a diverse group of researchers and industry practitioners to address the practical challenges and opportunities of integrating Artificial Intelligence into Agile software development. Through interactive sessions, participants identified shared frustrations related to integrating AI into Agile Software Development practices, including challenges with tooling, governance, data quality, and critical skill gaps. These challenges were systematically prioritized and analyzed to uncover root causes. The workshop culminated in the collaborative development of a research roadmap that pinpoints actionable directions for future work, including both immediate solutions and ambitious long-term goals. The key outcome is a structured agenda designed to foster joint industry-academic efforts to move from identified frustrations to successful implementation.
☆ Irec: A Metacognitive Scaffolding for Self-Regulated Learning through Just-in-Time Insight Recall: A Conceptual Framework and System Prototype
The core challenge in learning has shifted from knowledge acquisition to effective Self-Regulated Learning (SRL): planning, monitoring, and reflecting on one's learning. Existing digital tools, however, inadequately support metacognitive reflection. Spaced Repetition Systems (SRS) use de-contextualized review, overlooking the role of context, while Personal Knowledge Management (PKM) tools require high manual maintenance. To address these challenges, this paper introduces "Insight Recall," a novel paradigm that conceptualizes the context-triggered retrieval of personal past insights as a metacognitive scaffold to promote SRL. We formalize this paradigm using the Just-in-Time Adaptive Intervention (JITAI) framework and implement a prototype system, Irec, to demonstrate its feasibility. At its core, Irec uses a dynamic knowledge graph of the user's learning history. When a user faces a new problem, a hybrid retrieval engine recalls relevant personal "insights." Subsequently, a large language model (LLM) performs a deep similarity assessment to filter and present the most relevant scaffold in a just-in-time manner. To reduce cognitive load, Irec features a human-in-the-loop pipeline for LLM-based knowledge graph construction. We also propose an optional "Guided Inquiry" module, where users can engage in a Socratic dialogue with an expert LLM, using the current problem and recalled insights as context. The contribution of this paper is a solid theoretical framework and a usable system platform for designing next-generation intelligent learning systems that enhance metacognition and self-regulation.
comment: Version 1 of a work in progress. Finalized system flowcharts, a public GitHub repository with the source code, and a full reproducibility package detailing the prompts, models, and testing guidelines will be provided in v2
☆ Loss-Aware Automatic Selection of Structured Pruning Criteria for Deep Neural Network Acceleration
Structured pruning is a well-established technique for compressing neural networks, making it suitable for deployment in resource-limited edge devices. This paper presents an efficient Loss-Aware Automatic Selection of Structured Pruning Criteria (LAASP) for slimming and accelerating deep neural networks. The majority of pruning methodologies employ a sequential process consisting of three stages: 1) training, 2) pruning, and 3) fine-tuning, whereas the proposed pruning technique adopts a pruning-while-training approach that eliminates the first stage and integrates the second and third stages into a single cycle. The automatic selection of magnitude or similarity-based filter pruning criteria from a specified pool of criteria and the specific pruning layer at each pruning iteration is guided by the network's overall loss on a small subset of the training data. To mitigate the abrupt accuracy drop due to pruning, the network is retrained briefly after each reduction of a predefined number of floating-point operations (FLOPs). The optimal pruning rates for each layer in the network are automatically determined, eliminating the need for manual allocation of fixed or variable pruning rates for each layer. Experiments on the VGGNet and ResNet models on the CIFAR-10 and ImageNet benchmark datasets demonstrate the effectiveness of the proposed method. In particular, the ResNet56 and ResNet110 models on the CIFAR-10 dataset significantly improve the top-1 accuracy compared to state-of-the-art methods while reducing the network FLOPs by 52\%. Furthermore, the ResNet50 model on the ImageNet dataset reduces FLOPs by more than 42\% with a negligible 0.33\% drop in top-5 accuracy. The source code of this paper is publicly available online - https://github.com/ghimiredhikura/laasp.
☆ EAR: Erasing Concepts from Unified Autoregressive Models
Autoregressive (AR) models have achieved unified and strong performance across both visual understanding and image generation tasks. However, removing undesired concepts from AR models while maintaining overall generation quality remains an open challenge. In this paper, we propose Erasure Autoregressive Model (EAR), a fine-tuning method for effective and utility-preserving concept erasure in AR models. Specifically, we introduce Windowed Gradient Accumulation (WGA) strategy to align patch-level decoding with erasure objectives, and Thresholded Loss Masking (TLM) strategy to protect content unrelated to the target concept during fine-tuning. Furthermore, we propose a novel benchmark, Erase Concept Generator and Visual Filter (ECGVF), aim at provide a more rigorous and comprehensive foundation for evaluating concept erasure in AR models. Specifically, we first employ structured templates across diverse large language models (LLMs) to pre-generate a large-scale corpus of target-replacement concept prompt pairs. Subsequently, we generate images from these prompts and subject them to rigorous filtering via a visual classifier to ensure concept fidelity and alignment. Extensive experimental results conducted on the ECGVF benchmark with the AR model Janus-Pro demonstrate that EAR achieves marked improvements in both erasure effectiveness and model utility preservation. Code is available at: https://github.com/immc-lab/ear/
comment: 11 pages, 7 figures, 1 tables
☆ AI Copilots for Reproducibility in Science: A Case Study
Open science initiatives seek to make research outputs more transparent, accessible, and reusable, but ensuring that published findings can be independently reproduced remains a persistent challenge. This paper introduces OpenPub, an AI-powered platform that supports researchers, reviewers, and readers through a suite of modular copilots focused on key open science tasks. In this work, we present the Reproducibility Copilot, which analyzes manuscripts, code, and supplementary materials to generate structured Jupyter Notebooks and recommendations aimed at facilitating computational, or "rote", reproducibility. We conducted feasibility tests using previously studied research papers with known reproducibility benchmarks. Results indicate that OpenPub can substantially reduce reproduction time - from over 30 hours to about 1 hour - while achieving high coverage of figures, tables, and results suitable for computational reproduction. The system systematically detects barriers to reproducibility, including missing hyperparameters, undocumented preprocessing steps, and incomplete or inaccessible datasets. These findings suggest that AI-driven tools can meaningfully reduce the burden of reproducibility efforts and contribute to more transparent and verifiable scientific communication. The modular copilot architecture also provides a foundation for extending AI assistance to additional open science objectives beyond reproducibility.
☆ CCRS: A Zero-Shot LLM-as-a-Judge Framework for Comprehensive RAG Evaluation SIGIR 2025
RAG systems enhance LLMs by incorporating external knowledge, which is crucial for domains that demand factual accuracy and up-to-date information. However, evaluating the multifaceted quality of RAG outputs, spanning aspects such as contextual coherence, query relevance, factual correctness, and informational completeness, poses significant challenges. Existing evaluation methods often rely on simple lexical overlap metrics, which are inadequate for capturing these nuances, or involve complex multi-stage pipelines with intermediate steps like claim extraction or require finetuning specialized judge models, hindering practical efficiency. To address these limitations, we propose CCRS (Contextual Coherence and Relevance Score), a novel suite of five metrics that utilizes a single, powerful, pretrained LLM as a zero-shot, end-to-end judge. CCRS evaluates: Contextual Coherence (CC), Question Relevance (QR), Information Density (ID), Answer Correctness (AC), and Information Recall (IR). We apply CCRS to evaluate six diverse RAG system configurations on the challenging BioASQ dataset. Our analysis demonstrates that CCRS effectively discriminates between system performances, confirming, for instance, that the Mistral-7B reader outperforms Llama variants. We provide a detailed analysis of CCRS metric properties, including score distributions, convergent/discriminant validity, tie rates, population statistics, and discriminative power. Compared to the complex RAGChecker framework, CCRS offers comparable or superior discriminative power for key aspects like recall and faithfulness, while being significantly more computationally efficient. CCRS thus provides a practical, comprehensive, and efficient framework for evaluating and iteratively improving RAG systems.
comment: Accepted at LLM4Eval @ SIGIR 2025
☆ BrokenVideos: A Benchmark Dataset for Fine-Grained Artifact Localization in AI-Generated Videos
Recent advances in deep generative models have led to significant progress in video generation, yet the fidelity of AI-generated videos remains limited. Synthesized content often exhibits visual artifacts such as temporally inconsistent motion, physically implausible trajectories, unnatural object deformations, and local blurring that undermine realism and user trust. Accurate detection and spatial localization of these artifacts are crucial for both automated quality control and for guiding the development of improved generative models. However, the research community currently lacks a comprehensive benchmark specifically designed for artifact localization in AI generated videos. Existing datasets either restrict themselves to video or frame level detection or lack the fine-grained spatial annotations necessary for evaluating localization methods. To address this gap, we introduce BrokenVideos, a benchmark dataset of 3,254 AI-generated videos with meticulously annotated, pixel-level masks highlighting regions of visual corruption. Each annotation is validated through detailed human inspection to ensure high quality ground truth. Our experiments show that training state of the art artifact detection models and multi modal large language models (MLLMs) on BrokenVideos significantly improves their ability to localize corrupted regions. Through extensive evaluation, we demonstrate that BrokenVideos establishes a critical foundation for benchmarking and advancing research on artifact localization in generative video models. The dataset is available at: https://broken-video-detection-datetsets.github.io/Broken-Video-Detection-Datasets.github.io/.
comment: 7 page,4 figures,2 tables
☆ MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations
We introduce MIRAGE, a new benchmark for multimodal expert-level reasoning and decision-making in consultative interaction settings. Designed for the agriculture domain, MIRAGE captures the full complexity of expert consultations by combining natural user queries, expert-authored responses, and image-based context, offering a high-fidelity benchmark for evaluating models on grounded reasoning, clarification strategies, and long-form generation in a real-world, knowledge-intensive domain. Grounded in over 35,000 real user-expert interactions and curated through a carefully designed multi-step pipeline, MIRAGE spans diverse crop health, pest diagnosis, and crop management scenarios. The benchmark includes more than 7,000 unique biological entities, covering plant species, pests, and diseases, making it one of the most taxonomically diverse benchmarks available for vision-language models, grounded in the real world. Unlike existing benchmarks that rely on well-specified user inputs and closed-set taxonomies, MIRAGE features underspecified, context-rich scenarios with open-world settings, requiring models to infer latent knowledge gaps, handle rare entities, and either proactively guide the interaction or respond. Project Page: https://mirage-benchmark.github.io
comment: 66 pages, 32 figures, 23 tables
☆ SACL: Understanding and Combating Textual Bias in Code Retrieval with Semantic-Augmented Reranking and Localization
Retrieval-Augmented Code Generation (RACG) is a critical technique for enhancing code generation by retrieving relevant information. In this work, we conduct an in-depth analysis of code retrieval by systematically masking specific features while preserving code functionality. Our discoveries include: (1) although trained on code, current retrievers heavily rely on surface-level textual features (e.g., docstrings, identifier names), and (2) they exhibit a strong bias towards well-documented code, even if the documentation is irrelevant.Based on our discoveries, we propose SACL, a framework that enriches textual information and reduces bias by augmenting code or structural knowledge with semantic information. Extensive experiments show that SACL substantially improves code retrieval (e.g., by 12.8% / 9.4% / 7.0% Recall@1 on HumanEval / MBPP / SWE-Bench-Lite), which also leads to better code generation performance (e.g., by 4.88% Pass@1 on HumanEval).
☆ A Modular Multitask Reasoning Framework Integrating Spatio-temporal Models and LLMs
Spatio-temporal data mining plays a pivotal role in informed decision making across diverse domains. However, existing models are often restricted to narrow tasks, lacking the capacity for multi-task inference and complex long-form reasoning that require generation of in-depth, explanatory outputs. These limitations restrict their applicability to real-world, multi-faceted decision scenarios. In this work, we introduce STReason, a novel framework that integrates the reasoning strengths of large language models (LLMs) with the analytical capabilities of spatio-temporal models for multi-task inference and execution. Without requiring task-specific finetuning, STReason leverages in-context learning to decompose complex natural language queries into modular, interpretable programs, which are then systematically executed to generate both solutions and detailed rationales. To facilitate rigorous evaluation, we construct a new benchmark dataset and propose a unified evaluation framework with metrics specifically designed for long-form spatio-temporal reasoning. Experimental results show that STReason significantly outperforms advanced LLM baselines across all metrics, particularly excelling in complex, reasoning-intensive spatio-temporal scenarios. Human evaluations further validate STReason's credibility and practical utility, demonstrating its potential to reduce expert workload and broaden the applicability to real-world spatio-temporal tasks. We believe STReason provides a promising direction for developing more capable and generalizable spatio-temporal reasoning systems.
♻ ☆ OmniGen2: Exploration to Advanced Multimodal Generation
In this work, we introduce OmniGen2, a versatile and open-source generative model designed to provide a unified solution for diverse generation tasks, including text-to-image, image editing, and in-context generation. Unlike OmniGen v1, OmniGen2 features two distinct decoding pathways for text and image modalities, utilizing unshared parameters and a decoupled image tokenizer. This design enables OmniGen2 to build upon existing multimodal understanding models without the need to re-adapt VAE inputs, thereby preserving the original text generation capabilities. To facilitate the training of OmniGen2, we developed comprehensive data construction pipelines, encompassing image editing and in-context generation data. Additionally, we introduce a reflection mechanism tailored for image generation tasks and curate a dedicated reflection dataset based on OmniGen2. Despite its relatively modest parameter size, OmniGen2 achieves competitive results on multiple task benchmarks, including text-to-image and image editing. To further evaluate in-context generation, also referred to as subject-driven tasks, we introduce a new benchmark named OmniContext. OmniGen2 achieves state-of-the-art performance among open-source models in terms of consistency. We will release our models, training code, datasets, and data construction pipeline to support future research in this field. Project Page: https://vectorspacelab.github.io/OmniGen2; GitHub Link: https://github.com/VectorSpaceLab/OmniGen2
♻ ☆ Diffusion Models Through a Global Lens: Are They Culturally Inclusive?
Text-to-image diffusion models have recently enabled the creation of visually compelling, detailed images from textual prompts. However, their ability to accurately represent various cultural nuances remains an open question. In our work, we introduce CultDiff benchmark, evaluating state-of-the-art diffusion models whether they can generate culturally specific images spanning ten countries. We show that these models often fail to generate cultural artifacts in architecture, clothing, and food, especially for underrepresented country regions, by conducting a fine-grained analysis of different similarity aspects, revealing significant disparities in cultural relevance, description fidelity, and realism compared to real-world reference images. With the collected human evaluations, we develop a neural-based image-image similarity metric, namely, CultDiff-S, to predict human judgment on real and generated images with cultural artifacts. Our work highlights the need for more inclusive generative AI systems and equitable dataset representation over a wide range of cultures.
comment: 17 pages, 17 figures, 3 tables
♻ ☆ Recycling the Web: A Method to Enhance Pre-training Data Quality and Quantity for Language Models
Scaling laws predict that the performance of large language models improves with increasing model size and data size. In practice, pre-training has been relying on massive web crawls, using almost all data sources publicly available on the internet so far. However, this pool of natural data does not grow at the same rate as the compute supply. Furthermore, the availability of high-quality texts is even more limited: data filtering pipelines often remove up to 99% of the initial web scrapes to achieve state-of-the-art. To address the "data wall" of pre-training scaling, our work explores ways to transform and recycle data discarded in existing filtering processes. We propose REWIRE, REcycling the Web with guIded REwrite, a method to enrich low-quality documents so that they could become useful for training. This in turn allows us to increase the representation of synthetic data in the final pre-training set. Experiments at 1B, 3B and 7B scales of the DCLM benchmark show that mixing high-quality raw texts and our rewritten texts lead to 1.0, 1.3 and 2.5 percentage points improvement respectively across 22 diverse tasks, compared to training on only filtered web data. Training on the raw-synthetic data mix is also more effective than having access to 2x web data. Through further analysis, we demonstrate that about 82% of the mixed in texts come from transforming lower-quality documents that would otherwise be discarded. REWIRE also outperforms related approaches of generating synthetic data, including Wikipedia-style paraphrasing, question-answer synthesizing and knowledge extraction. These results suggest that recycling web texts holds the potential for being a simple and effective approach for scaling pre-training data.
♻ ☆ Do Concept Bottleneck Models Respect Localities?
Concept-based explainability methods use human-understandable intermediaries to produce explanations for machine learning models. These methods assume concept predictions can help understand a model's internal reasoning. In this work, we assess the degree to which such an assumption is true by analyzing whether concept predictors leverage "relevant" features to make predictions, a term we call locality. Concept-based models that fail to respect localities also fail to be explainable because concept predictions are based on spurious features, making the interpretation of the concept predictions vacuous. To assess whether concept-based models respect localities, we construct and use three metrics to characterize when models respect localities, complementing our analysis with theoretical results. Each of our metrics captures a different notion of perturbation and assess whether perturbing "irrelevant" features impacts the predictions made by a concept predictors. We find that many concept-based models used in practice fail to respect localities because concept predictors cannot always clearly distinguish distinct concepts. Based on these findings, we propose suggestions for alleviating this issue.
comment: Published at TMLR
♻ ☆ From $\mathcal{O}(n^{2})$ to $\mathcal{O}(n)$ Parameters: Quantum Self-Attention in Vision Transformers for Biomedical Image Classification MICCAI 2025
We demonstrate that quantum vision transformers (QViTs), vision transformers (ViTs) with self-attention (SA) mechanisms replaced by quantum self-attention (QSA) mechanisms, can match state-of-the-art (SOTA) biomedical image classifiers while using 99.99% fewer parameters. QSAs are produced by replacing linear SA layers with parameterised quantum neural networks (QNNs), producing a QSA mechanism and reducing parameter scaling from $\mathcal{O}(n^2)$ to $\mathcal{O}(n)$. On RetinaMNIST, our ultra parameter-efficient QViT outperforms 13/14 SOTA methods including CNNs and ViTs, achieving 56.5% accuracy, just 0.88% below the top MedMamba model while using 99.99% fewer parameters (1K vs 14.5M) and 89% fewer GFLOPs. We present the first investigation of knowledge distillation (KD) from classical to quantum vision transformers in biomedical image classification, showing that QViTs maintain comparable performance to classical ViTs across eight diverse datasets spanning multiple modalities, with improved QSA parameter-efficiency. Our higher-qubit architecture benefitted more from KD pre-training, suggesting a scaling relationship between QSA parameters and KD effectiveness. These findings establish QSA as a practical architectural choice toward parameter-efficient biomedical image analysis.
comment: Submitted for EMA4MICCAI 2025
♻ ☆ FluoroSAM: A Language-promptable Foundation Model for Flexible X-ray Image Segmentation
Language promptable X-ray image segmentation would enable greater flexibility for human-in-the-loop workflows in diagnostic and interventional precision medicine. Prior efforts have contributed task-specific models capable of solving problems within a narrow scope, but expanding to broader use requires additional data, annotations, and training time. Recently, language-aligned foundation models (LFMs) -- machine learning models trained on large amounts of highly variable image and text data thus enabling broad applicability -- have emerged as promising tools for automated image analysis. Existing foundation models for medical image analysis focus on scenarios and modalities where large, richly annotated datasets are available. However, the X-ray imaging modality features highly variable image appearance and applications, from diagnostic chest X-rays to interventional fluoroscopy, with varying availability of data. To pave the way toward an LFM for comprehensive and language-aligned analysis of arbitrary medical X-ray images, we introduce FluoroSAM, a language-promptable variant of the Segment Anything Model, trained from scratch on 3M synthetic X-ray images from a wide variety of human anatomies, imaging geometries, and viewing angles. These include pseudo-ground truth masks for 128 organ types and 464 tools with associated text descriptions. FluoroSAM is capable of segmenting myriad anatomical structures and tools based on natural language prompts, thanks to the novel incorporation of vector quantization (VQ) of text embeddings in the training process. We demonstrate FluoroSAM's performance quantitatively on real X-ray images and showcase on several applications how FluoroSAM is a key enabler for rich human-machine interaction in the X-ray image acquisition and analysis context. Code is available at https://github.com/arcadelab/fluorosam.
♻ ☆ The State of Large Language Models for African Languages: Progress and Challenges
Large Language Models (LLMs) are transforming Natural Language Processing (NLP), but their benefits are largely absent for Africa's 2,000 low-resource languages. This paper comparatively analyzes African language coverage across six LLMs, eight Small Language Models (SLMs), and six Specialized SLMs (SSLMs). The evaluation covers language coverage, training sets, technical limitations, script problems, and language modelling roadmaps. The work identifies 42 supported African languages and 23 available public data sets, and it shows a big gap where four languages (Amharic, Swahili, Afrikaans, and Malagasy) are always treated while there is over 98\% of unsupported African languages. Moreover, the review shows that just Latin, Arabic, and Ge'ez scripts are identified while 20 active scripts are neglected. Some of the primary challenges are lack of data, tokenization biases, computational costs being very high, and evaluation issues. These issues demand language standardization, corpus development by the community, and effective adaptation methods for African languages.
♻ ☆ Rethinking Early Stopping: Refine, Then Calibrate
Machine learning classifiers often produce probabilistic predictions that are critical for accurate and interpretable decision-making in various domains. The quality of these predictions is generally evaluated with proper losses, such as cross-entropy, which decompose into two components: calibration error assesses general under/overconfidence, while refinement error measures the ability to distinguish different classes. In this paper, we present a novel variational formulation of the calibration-refinement decomposition that sheds new light on post-hoc calibration, and enables rapid estimation of the different terms. Equipped with this new perspective, we provide theoretical and empirical evidence that calibration and refinement errors are not minimized simultaneously during training. Selecting the best epoch based on validation loss thus leads to a compromise point that is suboptimal for both terms. To address this, we propose minimizing refinement error only during training (Refine,...), before minimizing calibration error post hoc, using standard techniques (...then Calibrate). Our method integrates seamlessly with any classifier and consistently improves performance across diverse classification tasks.
♻ ☆ Integrating Various Software Artifacts for Better LLM-based Bug Localization and Program Repair
LLMs have garnered considerable attention for their potential to streamline Automated Program Repair (APR). LLM-based approaches can either insert the correct code or directly generate patches when provided with buggy methods. However, most of LLM-based APR methods rely on a single type of software information, without fully leveraging different software artifacts. Despite this, many LLM-based approaches do not explore which specific types of information best assist in APR. Addressing this gap is crucial for advancing LLM-based APR techniques. We propose DEVLoRe to use issue content (description and message) and stack error traces to localize buggy methods, then rely on debug information in buggy methods and issue content and stack error to localize buggy lines and generate plausible patches which can pass all unit tests. The results show that while issue content is particularly effective in assisting LLMs with fault localization and program repair, different types of software artifacts complement each other. By incorporating different artifacts, DEVLoRe successfully locates 49.3% and 47.6% of single and non-single buggy methods and generates 56.0% and 14.5% plausible patches for the Defects4J v2.0 dataset, respectively. This outperforms current state-of-the-art APR methods. Furthermore, we re-implemented and evaluated our framework, demonstrating its effectiveness in its effectiveness in resolving 9 unique issues compared to other state-of-the-art frameworks using the same or more advanced models on SWE-bench Lite.We also discussed whether a leading framework for Python code can be directly applied to Java code, or vice versa. The source code and experimental results of this work for replication are available at https://github.com/XYZboom/DEVLoRe.
comment: 25 pages, 12 images, 10 tables, Manuscript revision submitted to a journal (2025)
♻ ☆ Unlocking In-Context Learning for Natural Datasets Beyond Language Modelling
Large Language Models (LLMs) exhibit In-Context Learning (ICL), which enables the model to perform new tasks conditioning only on the examples provided in the context without updating the model's weights. While ICL offers fast adaptation across natural language tasks and domains, its emergence is less straightforward for modalities beyond text. In this work, we systematically uncover properties present in LLMs that support the emergence of ICL for autoregressive models and various modalities by promoting the learning of the needed mechanisms for ICL. We identify exact token repetitions in the training data sequences as an important factor for ICL. Such repetitions further improve stability and reduce transiency in ICL performance. Moreover, we emphasise the significance of training task difficulty for the emergence of ICL. Finally, by applying our novel insights on ICL emergence, we unlock ICL capabilities for various visual datasets and a more challenging EEG classification task in a few-shot learning regime.
♻ ☆ TabArena: A Living Benchmark for Machine Learning on Tabular Data
With the growing popularity of deep learning and foundation models for tabular data, the need for standardized and reliable benchmarks is higher than ever. However, current benchmarks are static. Their design is not updated even if flaws are discovered, model versions are updated, or new models are released. To address this, we introduce TabArena, the first continuously maintained living tabular benchmarking system. To launch TabArena, we manually curate a representative collection of datasets and well-implemented models, conduct a large-scale benchmarking study to initialize a public leaderboard, and assemble a team of experienced maintainers. Our results highlight the influence of validation method and ensembling of hyperparameter configurations to benchmark models at their full potential. While gradient-boosted trees are still strong contenders on practical tabular datasets, we observe that deep learning methods have caught up under larger time budgets with ensembling. At the same time, foundation models excel on smaller datasets. Finally, we show that ensembles across models advance the state-of-the-art in tabular machine learning and investigate the contributions of individual models. We launch TabArena with a public leaderboard, reproducible code, and maintenance protocols to create a living benchmark available at https://tabarena.ai.
comment: v2: fixed author list. 51 pages. Code available at https://tabarena.ai/code; examples at https://tabarena.ai/code-examples; dataset curation at https://tabarena.ai/data-tabular-ml-iid-study and https://tabarena.ai/dataset-curation
♻ ☆ Adversarial Reasoning at Jailbreaking Time ICML 2025
As large language models (LLMs) are becoming more capable and widespread, the study of their failure cases is becoming increasingly important. Recent advances in standardizing, measuring, and scaling test-time compute suggest new methodologies for optimizing models to achieve high performance on hard tasks. In this paper, we apply these advances to the task of model jailbreaking: eliciting harmful responses from aligned LLMs. We develop an adversarial reasoning approach to automatic jailbreaking that leverages a loss signal to guide the test-time compute, achieving SOTA attack success rates against many aligned LLMs, even those that aim to trade inference-time compute for adversarial robustness. Our approach introduces a new paradigm in understanding LLM vulnerabilities, laying the foundation for the development of more robust and trustworthy AI systems.
comment: Accepted to the 42nd International Conference on Machine Learning (ICML 2025)
♻ ☆ Separating Tongue from Thought: Activation Patching Reveals Language-Agnostic Concept Representations in Transformers ICML 2024
A central question in multilingual language modeling is whether large language models (LLMs) develop a universal concept representation, disentangled from specific languages. In this paper, we address this question by analyzing latent representations (latents) during a word-translation task in transformer-based LLMs. We strategically extract latents from a source translation prompt and insert them into the forward pass on a target translation prompt. By doing so, we find that the output language is encoded in the latent at an earlier layer than the concept to be translated. Building on this insight, we conduct two key experiments. First, we demonstrate that we can change the concept without changing the language and vice versa through activation patching alone. Second, we show that patching with the mean representation of a concept across different languages does not affect the models' ability to translate it, but instead improves it. Finally, we generalize to multi-token generation and demonstrate that the model can generate natural language description of those mean representations. Our results provide evidence for the existence of language-agnostic concept representations within the investigated models.
comment: 20 pages, 14 figures, previous version published under the title "How Do Llamas Process Multilingual Text? A Latent Exploration through Activation Patching" at the ICML 2024 mechanistic interpretability workshop at https://openreview.net/forum?id=0ku2hIm4BS
♻ ☆ Proximal Control of UAVs with Federated Learning for Human-Robot Collaborative Domains
The human-robot interaction (HRI) is a growing area of research. In HRI, complex command (action) classification is still an open problem that usually prevents the real applicability of such a technique. The literature presents some works that use neural networks to detect these actions. However, occlusion is still a major issue in HRI, especially when using uncrewed aerial vehicles (UAVs), since, during the robot's movement, the human operator is often out of the robot's field of view. Furthermore, in multi-robot scenarios, distributed training is also an open problem. In this sense, this work proposes an action recognition and control approach based on Long Short-Term Memory (LSTM) Deep Neural Networks with two layers in association with three densely connected layers and Federated Learning (FL) embedded in multiple drones. The FL enabled our approach to be trained in a distributed fashion, i.e., access to data without the need for cloud or other repositories, which facilitates the multi-robot system's learning. Furthermore, our multi-robot approach results also prevented occlusion situations, with experiments with real robots achieving an accuracy greater than 96%.
comment: version 2
♻ ☆ VRAIL: Vectorized Reward-based Attribution for Interpretable Learning
We propose VRAIL (Vectorized Reward-based Attribution for Interpretable Learning), a bi-level framework for value-based reinforcement learning (RL) that learns interpretable weight representations from state features. VRAIL consists of two stages: a deep learning (DL) stage that fits an estimated value function using state features, and an RL stage that uses this to shape learning via potential-based reward transformations. The estimator is modeled in either linear or quadratic form, allowing attribution of importance to individual features and their interactions. Empirical results on the Taxi-v3 environment demonstrate that VRAIL improves training stability and convergence compared to standard DQN, without requiring environment modifications. Further analysis shows that VRAIL uncovers semantically meaningful subgoals, such as passenger possession, highlighting its ability to produce human-interpretable behavior. Our findings suggest that VRAIL serves as a general, model-agnostic framework for reward shaping that enhances both learning and interpretability.
♻ ☆ Training Plug-n-Play Knowledge Modules with Deep Context Distillation
Dynamically integrating new or rapidly evolving information after (Large) Language Model pre-training remains challenging, particularly in low-data scenarios or when dealing with private and specialized documents. In-context learning and retrieval-augmented generation (RAG) face limitations, including their high inference costs and their inability to capture global document information. In this paper, we propose a way of modularizing knowledge by training document-level Knowledge Modules (KMs). KMs are lightweight components implemented as parameter-efficient LoRA modules, which are trained to store information about new documents and can be easily plugged into models on demand. We show that next-token prediction performs poorly as the training objective for KMs. We instead propose Deep Context Distillation: we learn KMs parameters such as to simulate hidden states and logits of a teacher that takes the document in context. Our method outperforms standard next-token prediction and pre-instruction training techniques, across two datasets. Finally, we highlight synergies between KMs and RAG.
comment: Preprint
♻ ☆ Fine, I'll Merge It Myself: A Multi-Fidelity Framework for Automated Model Merging
Reasoning capabilities represent a critical frontier for large language models (LLMs), but developing them requires extensive proprietary datasets and computational resources. One way to efficiently supplement capabilities with is by model merging, which offers a promising alternative by combining multiple models without retraining. However, current merging approaches rely on manually-designed strategies for merging hyperparameters, limiting the exploration of potential model combinations and requiring significant human effort. We propose an Automated Model Merging Framework that enables fine-grained exploration of merging strategies while reducing costs through multi-fidelity approximations. We support both single and multi-objective optimization and introduce two novel search spaces: layerwise fusion (LFS) and depth-wise integration (DIS). Evaluating across a number of benchmarks, we find that the search autonomously finds 1) Merges that further boost single-objective performance, even on tasks the model has already been finetuned on, and 2) Merges that optimize multi-objective frontiers across tasks. Effective merges are found with limited compute, e.g. within less than 500 search steps.
♻ ☆ Non-equilibrium Annealed Adjoint Sampler
Recently, there has been significant progress in learning-based diffusion samplers, which aim to sample from a given unnormalized density. These methods typically follow one of two paradigms: (i) formulating sampling as an unbiased stochastic optimal control (SOC) problem using a canonical reference process, or (ii) refining annealed path measures through importance-weighted sampling. Although annealing approaches have advantages in guiding samples toward high-density regions, reliance on importance sampling leads to high variance and limited scalability in practice. In this paper, we introduce the \textbf{Non-equilibrium Annealed Adjoint Sampler (NAAS)}, a novel SOC-based diffusion sampler that leverages annealed reference dynamics without resorting to importance sampling. NAAS employs a lean adjoint system inspired by adjoint matching, enabling efficient and scalable training. We demonstrate the effectiveness of our approach across a range of tasks, including sampling from classical energy landscapes and molecular Boltzmann distribution.
comment: 21 pages, 7 figures
♻ ☆ CLAIM: Clinically-Guided LGE Augmentation for Realistic and Diverse Myocardial Scar Synthesis and Segmentation
Deep learning-based myocardial scar segmentation from late gadolinium enhancement (LGE) cardiac MRI has shown great potential for accurate and timely diagnosis and treatment planning for structural cardiac diseases. However, the limited availability and variability of LGE images with high-quality scar labels restrict the development of robust segmentation models. To address this, we introduce CLAIM: \textbf{C}linically-Guided \textbf{L}GE \textbf{A}ugmentation for Real\textbf{i}stic and Diverse \textbf{M}yocardial Scar Synthesis and Segmentation framework, a framework for anatomically grounded scar generation and segmentation. At its core is the SMILE module (Scar Mask generation guided by cLinical knowledgE), which conditions a diffusion-based generator on the clinically adopted AHA 17-segment model to synthesize images with anatomically consistent and spatially diverse scar patterns. In addition, CLAIM employs a joint training strategy in which the scar segmentation network is optimized alongside the generator, aiming to enhance both the realism of synthesized scars and the accuracy of the scar segmentation performance. Experimental results show that CLAIM produces anatomically coherent scar patterns and achieves higher Dice similarity with real scar distributions compared to baseline models. Our approach enables controllable and realistic myocardial scar synthesis and has demonstrated utility for downstream medical imaging task. Code is available at https://github.com/farheenjabeen/CLAIM-Scar-Synthesis.
comment: 14 Pages
♻ ☆ RefPentester: A Knowledge-Informed Self-Reflective Penetration Testing Framework Based on Large Language Models
Automated penetration testing (AutoPT) powered by large language models (LLMs) has gained attention for its ability to automate ethical hacking processes and identify vulnerabilities in target systems by leveraging the inherent knowledge of LLMs. However, existing LLM-based AutoPT frameworks often underperform compared to human experts in challenging tasks for several reasons: the imbalanced knowledge used in LLM training, short-sightedness in the planning process, and hallucinations during command generation. Moreover, the trial-and-error nature of the PT process is constrained by existing frameworks lacking mechanisms to learn from previous failures, restricting adaptive improvement of PT strategies. To address these limitations, we propose a knowledge-informed, self-reflective PT framework powered by LLMs, called RefPentester. This AutoPT framework is designed to assist human operators in identifying the current stage of the PT process, selecting appropriate tactics and techniques for each stage, choosing suggested actions, providing step-by-step operational guidance, and reflecting on and learning from previous failed operations. We also modeled the PT process as a seven-state Stage Machine to integrate the proposed framework effectively. The evaluation shows that RefPentester can successfully reveal credentials on Hack The Box's Sau machine, outperforming the baseline GPT-4o model by 16.7%. Across PT stages, RefPentester also demonstrates superior success rates on PT stage transitions.
♻ ☆ Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning
Scientific discoveries increasingly rely on complex multimodal reasoning based on information-intensive scientific data and domain-specific expertise. Empowered by expert-level scientific benchmarks, scientific Multimodal Large Language Models (MLLMs) hold the potential to significantly enhance this discovery process in realistic workflows. However, current scientific benchmarks mostly focus on evaluating the knowledge understanding capabilities of MLLMs, leading to an inadequate assessment of their perception and reasoning abilities. To address this gap, we present the Scientists' First Exam (SFE) benchmark, designed to evaluate the scientific cognitive capacities of MLLMs through three interconnected levels: scientific signal perception, scientific attribute understanding, scientific comparative reasoning. Specifically, SFE comprises 830 expert-verified VQA pairs across three question types, spanning 66 multimodal tasks across five high-value disciplines. Extensive experiments reveal that current state-of-the-art GPT-o3 and InternVL-3 achieve only 34.08% and 26.52% on SFE, highlighting significant room for MLLMs to improve in scientific realms. We hope the insights obtained in SFE will facilitate further developments in AI-enhanced scientific discoveries.
comment: 82 pages
♻ ☆ Physics-informed Imitative Reinforcement Learning for Real-world Driving
Recent advances in imitative reinforcement learning (IRL) have considerably enhanced the ability of autonomous agents to assimilate expert demonstrations, leading to rapid skill acquisition in a range of demanding tasks. However, such learning-based agents face significant challenges when transferring knowledge to highly dynamic closed-loop environments. Their performance is significantly impacted by the conflicting optimization objectives of imitation learning (IL) and reinforcement learning (RL), sample inefficiency, and the complexity of uncovering the hidden world model and physics. To address this challenge, we propose a physics-informed IRL that is entirely data-driven. It leverages both expert demonstration data and exploratory data with a joint optimization objective, allowing the underlying physical principles of vehicle dynamics to emerge naturally from the training process. The performance is evaluated through empirical experiments and results exceed popular IL, RL and IRL algorithms in closed-loop settings on Waymax benchmark. Our approach exhibits 37.8% reduction in collision rate and 22.2% reduction in off-road rate compared to the baseline method.
♻ ☆ CogniBench: A Legal-inspired Framework and Dataset for Assessing Cognitive Faithfulness of Large Language Models ACL 2025
Faithfulness hallucinations are claims generated by a Large Language Model (LLM) not supported by contexts provided to the LLM. Lacking assessment standards, existing benchmarks focus on "factual statements" that rephrase source materials while overlooking "cognitive statements" that involve making inferences from the given context. Consequently, evaluating and detecting the hallucination of cognitive statements remains challenging. Inspired by how evidence is assessed in the legal domain, we design a rigorous framework to assess different levels of faithfulness of cognitive statements and introduce the CogniBench dataset where we reveal insightful statistics. To keep pace with rapidly evolving LLMs, we further develop an automatic annotation pipeline that scales easily across different models. This results in a large-scale CogniBench-L dataset, which facilitates training accurate detectors for both factual and cognitive hallucinations. We release our model and datasets at: https://github.com/FUTUREEEEEE/CogniBench
comment: ACL 2025
♻ ☆ No Free Lunch: Rethinking Internal Feedback for LLM Reasoning
Reinforcement learning has emerged as a powerful paradigm for post-training large language models (LLMs) to improve reasoning. Approaches like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) have shown strong results, but they require extensive external supervision. We investigate an alternative class of methods, Reinforcement Learning from Internal Feedback (RLIF), which relies solely on intrinsic model-derived signals instead of external rewards. In particular, we leverage unsupervised reward proxies such as token-level entropy, trajectory-level entropy, and self-certainty. Our theoretical analysis shows these internal objectives are partially equivalent, and we empirically evaluate various RLIF strategies on challenging math reasoning benchmarks. Experimental results demonstrate that RLIF can boost the reasoning performance of base LLMs at the beginning phase of the training, matching or surpassing RLVR techniques on these tasks. However, when training progresses, performance degrades even below the model before training. Moreover, we find that RLIF yields little improvement for instruction-tuned models, indicating diminishing returns of intrinsic feedback once an LLM is already instruction-tuned. We further analyze this limitation by mixing model weights and explain the reason of RLIF's training behaviors, providing practical guidelines for integrating internal feedback signals into LLM training. We hope our analysis of internal feedback will inform more principled and effective strategies for LLM post-training.
♻ ☆ WyckoffDiff -- A Generative Diffusion Model for Crystal Symmetry ICML 2025
Crystalline materials often exhibit a high level of symmetry. However, most generative models do not account for symmetry, but rather model each atom without any constraints on its position or element. We propose a generative model, Wyckoff Diffusion (WyckoffDiff), which generates symmetry-based descriptions of crystals. This is enabled by considering a crystal structure representation that encodes all symmetry, and we design a novel neural network architecture which enables using this representation inside a discrete generative model framework. In addition to respecting symmetry by construction, the discrete nature of our model enables fast generation. We additionally present a new metric, Fr\'echet Wrenformer Distance, which captures the symmetry aspects of the materials generated, and we benchmark WyckoffDiff against recently proposed generative models for crystal generation. As a proof-of-concept study, we use WyckoffDiff to find new materials below the convex hull of thermodynamical stability.
comment: Accepted to ICML 2025, to appear in PMLR 267. Code is available online at https://github.com/httk/wyckoffdiff
♻ ☆ Chemical knowledge-informed framework for privacy-aware retrosynthesis learning
Chemical reaction data is a pivotal asset, driving advances in competitive fields such as pharmaceuticals, materials science, and industrial chemistry. Its proprietary nature renders it sensitive, as it often includes confidential insights and competitive advantages organizations strive to protect. However, in contrast to this need for confidentiality, the current standard training paradigm for machine learning-based retrosynthesis gathers reaction data from multiple sources into one single edge to train prediction models. This paradigm poses considerable privacy risks as it necessitates broad data availability across organizational boundaries and frequent data transmission between entities, potentially exposing proprietary information to unauthorized access or interception during storage and transfer. In the present study, we introduce the chemical knowledge-informed framework (CKIF), a privacy-preserving approach for learning retrosynthesis models. CKIF enables distributed training across multiple chemical organizations without compromising the confidentiality of proprietary reaction data. Instead of gathering raw reaction data, CKIF learns retrosynthesis models through iterative, chemical knowledge-informed aggregation of model parameters. In particular, the chemical properties of predicted reactants are leveraged to quantitatively assess the observable behaviors of individual models, which in turn determines the adaptive weights used for model aggregation. On a variety of reaction datasets, CKIF outperforms several strong baselines by a clear margin.
♻ ☆ SMAR: Soft Modality-Aware Routing Strategy for MoE-based Multimodal Large Language Models Preserving Language Capabilities
Mixture of Experts (MoE) architectures have become a key approach for scaling large language models, with growing interest in extending them to multimodal tasks. Existing methods to build multimodal MoE models either incur high training costs or suffer from degraded language capabilities when adapting pretrained models. To address this, we propose Soft ModalityAware Routing (SMAR), a novel regularization technique that uses Kullback Leibler divergence to control routing probability distributions across modalities, encouraging expert specialization without modifying model architecture or heavily relying on textual data. Experiments on visual instruction tuning show that SMAR preserves language ability at 86.6% retention with only 2.5% pure text, outperforming baselines while maintaining strong multimodal performance. Our approach offers a practical and efficient solution to balance modality differentiation and language capabilities in multimodal MoE models.
♻ ☆ A Survey on Explainable Reinforcement Learning: Concepts, Algorithms, Challenges
Reinforcement Learning (RL) is a popular machine learning paradigm where intelligent agents interact with the environment to fulfill a long-term goal. Driven by the resurgence of deep learning, Deep RL (DRL) has witnessed great success over a wide spectrum of complex control tasks. Despite the encouraging results achieved, the deep neural network-based backbone is widely deemed as a black box that impedes practitioners to trust and employ trained agents in realistic scenarios where high security and reliability are essential. To alleviate this issue, a large volume of literature devoted to shedding light on the inner workings of the intelligent agents has been proposed, by constructing intrinsic interpretability or post-hoc explainability. In this survey, we provide a comprehensive review of existing works on eXplainable RL (XRL) and introduce a new taxonomy where prior works are clearly categorized into model-explaining, reward-explaining, state-explaining, and task-explaining methods. We also review and highlight RL methods that conversely leverage human knowledge to promote learning efficiency and performance of agents while this kind of method is often ignored in XRL field. Some challenges and opportunities in XRL are discussed. This survey intends to provide a high-level summarization of XRL and to motivate future research on more effective XRL solutions. Corresponding open source codes are collected and categorized at https://github.com/Plankson/awesome-explainable-reinforcement-learning.
♻ ☆ Confucius3-Math: A Lightweight High-Performance Reasoning LLM for Chinese K-12 Mathematics Learning
We introduce Confucius3-Math, an open-source large language model with 14B parameters that (1) runs efficiently on a single consumer-grade GPU; (2) achieves SOTA performances on a range of mathematical reasoning tasks, outperforming many models with significantly larger sizes. In particular, as part of our mission to enhancing education and knowledge dissemination with AI, Confucius3-Math is specifically committed to mathematics learning for Chinese K-12 students and educators. Built via post-training with large-scale reinforcement learning (RL), Confucius3-Math aligns with national curriculum and excels at solving main-stream Chinese K-12 mathematical problems with low cost. In this report we share our development recipe, the challenges we encounter and the techniques we develop to overcome them. In particular, we introduce three technical innovations: Targeted Entropy Regularization, Recent Sample Recovery and Policy-Specific Hardness Weighting. These innovations encompass a new entropy regularization, a novel data scheduling policy, and an improved group-relative advantage estimator. Collectively, they significantly stabilize the RL training, improve data efficiency, and boost performance. Our work demonstrates the feasibility of building strong reasoning models in a particular domain at low cost. We open-source our model and code at https://github.com/netease-youdao/Confucius3-Math.
♻ ☆ $C^3$-Bench: The Things Real Disturbing LLM based Agent in Multi-Tasking
Agents based on large language models leverage tools to modify environments, revolutionizing how AI interacts with the physical world. Unlike traditional NLP tasks that rely solely on historical dialogue for responses, these agents must consider more complex factors, such as inter-tool relationships, environmental feedback and previous decisions, when making choices. Current research typically evaluates agents via multi-turn dialogues. However, it overlooks the influence of these critical factors on agent behavior. To bridge this gap, we present an open-source and high-quality benchmark $C^3$-Bench. This benchmark integrates attack concepts and applies univariate analysis to pinpoint key elements affecting agent robustness. In concrete, we design three challenges: navigate complex tool relationships, handle critical hidden information and manage dynamic decision paths. Complementing these challenges, we introduce fine-grained metrics, innovative data collection algorithms and reproducible evaluation methods. Extensive experiments are conducted on 49 mainstream agents, encompassing general fast-thinking, slow-thinking and domain-specific models. We observe that agents have significant shortcomings in handling tool dependencies, long context information dependencies and frequent policy-type switching. In essence, $C^3$-Bench aims to expose model vulnerabilities through these challenges and drive research into the interpretability of agent performance. The benchmark is publicly available at https://github.com/yupeijei1997/C3-Bench.
♻ ☆ Graph-Assisted Stitching for Offline Hierarchical Reinforcement Learning ICML 2025
Existing offline hierarchical reinforcement learning methods rely on high-level policy learning to generate subgoal sequences. However, their efficiency degrades as task horizons increase, and they lack effective strategies for stitching useful state transitions across different trajectories. We propose Graph-Assisted Stitching (GAS), a novel framework that formulates subgoal selection as a graph search problem rather than learning an explicit high-level policy. By embedding states into a Temporal Distance Representation (TDR) space, GAS clusters semantically similar states from different trajectories into unified graph nodes, enabling efficient transition stitching. A shortest-path algorithm is then applied to select subgoal sequences within the graph, while a low-level policy learns to reach the subgoals. To improve graph quality, we introduce the Temporal Efficiency (TE) metric, which filters out noisy or inefficient transition states, significantly enhancing task performance. GAS outperforms prior offline HRL methods across locomotion, navigation, and manipulation tasks. Notably, in the most stitching-critical task, it achieves a score of 88.3, dramatically surpassing the previous state-of-the-art score of 1.0. Our source code is available at: https://github.com/qortmdgh4141/GAS.
comment: ICML 2025
♻ ☆ Solving Linear-Gaussian Bayesian Inverse Problems with Decoupled Diffusion Sequential Monte Carlo ICML 2025
A recent line of research has exploited pre-trained generative diffusion models as priors for solving Bayesian inverse problems. We contribute to this research direction by designing a sequential Monte Carlo method for linear-Gaussian inverse problems which builds on "decoupled diffusion", where the generative process is designed such that larger updates to the sample are possible. The method is asymptotically exact and we demonstrate the effectiveness of our Decoupled Diffusion Sequential Monte Carlo (DDSMC) algorithm on both synthetic as well as protein and image data. Further, we demonstrate how the approach can be extended to discrete data.
comment: Accepted to ICML 2025, to appear in PMLR 267. Code available at https://github.com/filipekstrm/ddsmc
♻ ☆ Balancing Truthfulness and Informativeness with Uncertainty-Aware Instruction Fine-Tuning
Instruction fine-tuning (IFT) can increase the informativeness of large language models (LLMs), but may reduce their truthfulness. This trade-off arises because IFT steers LLMs to generate responses containing long-tail knowledge that was not well covered during pre-training. As a result, models become more informative but less accurate when generalizing to unseen tasks. In this paper, we empirically demonstrate how unfamiliar knowledge in IFT datasets can negatively affect the truthfulness of LLMs, and we introduce two new IFT paradigms, $UNIT_{cut}$ and $UNIT_{ref}$, to address this issue. $UNIT_{cut}$ identifies and removes unfamiliar knowledge from IFT datasets to mitigate its impact on model truthfulness, whereas $UNIT_{ref}$ trains LLMs to recognize their uncertainty and explicitly indicate it at the end of their responses. Our experiments show that $UNIT_{cut}$ substantially improves LLM truthfulness, while $UNIT_{ref}$ maintains high informativeness and reduces hallucinations by distinguishing between confident and uncertain statements.
♻ ☆ Aurora: Are Android Malware Classifiers Reliable and Stable under Distribution Shift?
The performance figures of modern drift-adaptive malware classifiers appear promising, but does this translate to genuine operational reliability? The standard evaluation paradigm primarily focuses on baseline performance metrics, neglecting confidence-error alignment and operational stability. While TESSERACT established the importance of temporal evaluation, we take a complementary direction by investigating whether malware classifiers maintain reliable and stable confidence estimates under distribution shifts and exploring the tensions between scientific advancement and practical impacts when they do not. We propose AURORA, a framework to evaluate malware classifiers based on their confidence quality and operational resilience. AURORA subjects the confidence profile of a given model to verification to assess the reliability of its estimates. Unreliable confidence estimates erode operational trust, waste valuable annotation budget on non-informative samples for active learning, and leave error-prone instances undetected in selective classification. AURORA is complemented by a set of metrics designed to go beyond point-in-time performance, striving towards a more holistic assessment of operational stability throughout temporal evaluation periods. The fragility in SOTA frameworks across datasets of varying drift suggests the need for a return to the whiteboard.
♻ ☆ Teacher Motion Priors: Enhancing Robot Locomotion over Challenging Terrain IROS 2025
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, IROS 2025
♻ ☆ WoundAmbit: Bridging State-of-the-Art Semantic Segmentation and Real-World Wound Care ECML
Chronic wounds affect a large population, particularly the elderly and diabetic patients, who often exhibit limited mobility and co-existing health conditions. Automated wound monitoring via mobile image capture can reduce in-person physician visits by enabling remote tracking of wound size. Semantic segmentation is key to this process, yet wound segmentation remains underrepresented in medical imaging research. To address this, we benchmark state-of-the-art deep learning models from general-purpose vision, medical imaging, and top methods from public wound challenges. For a fair comparison, we standardize training, data augmentation, and evaluation, conducting cross-validation to minimize partitioning bias. We also assess real-world deployment aspects, including generalization to an out-of-distribution wound dataset, computational efficiency, and interpretability. Additionally, we propose a reference object-based approach to convert AI-generated masks into clinically relevant wound size estimates and evaluate this, along with mask quality, for the five best architectures based on physician assessments. Overall, the transformer-based TransNeXt showed the highest levels of generalizability. Despite variations in inference times, all models processed at least one image per second on the CPU, which is deemed adequate for the intended application. Interpretability analysis typically revealed prominent activations in wound regions, emphasizing focus on clinically relevant features. Expert evaluation showed high mask approval for all analyzed models, with VWFormer and ConvNeXtS backbone performing the best. Size retrieval accuracy was similar across models, and predictions closely matched expert annotations. Finally, we demonstrate how our AI-driven wound size estimation framework, WoundAmbit, is integrated into a custom telehealth system.
comment: Main paper: 18 pages; supplementary material: 15 pages; the paper has been accepted for publication at the Applied Data Science (ADS) track of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2025)
♻ ☆ Toddlers' Active Gaze Behavior Supports Self-Supervised Object Learning
Toddlers learn to recognize objects from different viewpoints with almost no supervision. During this learning, they execute frequent eye and head movements that shape their visual experience. It is presently unclear if and how these behaviors contribute to toddlers' emerging object recognition abilities. To answer this question, we here combine head-mounted eye tracking during dyadic play with unsupervised machine learning. We approximate toddlers' central visual field experience by cropping image regions from a head-mounted camera centered on the current gaze location estimated via eye tracking. This visual stream feeds an unsupervised computational model of toddlers' learning, which constructs visual representations that slowly change over time. Our experiments demonstrate that toddlers' gaze strategy supports the learning of invariant object representations. Our analysis also shows that the limited size of the central visual field where acuity is high is crucial for this. Overall, our work reveals how toddlers' gaze behavior may support their development of view-invariant object recognition.
comment: 27 pages, 16 figures
♻ ☆ Distributed satellite information networks: Architecture, enabling technologies, and trends
Driven by the vision of ubiquitous connectivity and wireless intelligence, the evolution of ultra-dense constellation-based satellite-integrated Internet is underway, now taking preliminary shape. Nevertheless, the entrenched institutional silos and limited, nonrenewable heterogeneous network resources leave current satellite systems struggling to accommodate the escalating demands of next-generation intelligent applications. In this context, the distributed satellite information networks (DSIN), exemplified by the cohesive clustered satellites system, have emerged as an innovative architecture, bridging information gaps across diverse satellite systems, such as communication, navigation, and remote sensing, and establishing a unified, open information network paradigm to support resilient space information services. This survey first provides a profound discussion about innovative network architectures of DSIN, encompassing distributed regenerative satellite network architecture, distributed satellite computing network architecture, and reconfigurable satellite formation flying, to enable flexible and scalable communication, computing and control. The DSIN faces challenges from network heterogeneity, unpredictable channel dynamics, sparse resources, and decentralized collaboration frameworks. To address these issues, a series of enabling technologies is identified, including channel modeling and estimation, cloud-native distributed MIMO cooperation, grant-free massive access, network routing, and the proper combination of all these diversity techniques. Furthermore, to heighten the overall resource efficiency, the cross-layer optimization techniques are further developed to meet upper-layer deterministic, adaptive and secure information services requirements. In addition, emerging research directions and new opportunities are highlighted on the way to achieving the DSIN vision.
♻ ☆ 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.
♻ ☆ FGS-SLAM: Fourier-based Gaussian Splatting for Real-time SLAM with Sparse and Dense Map Fusion
3D gaussian splatting has advanced simultaneous localization and mapping (SLAM) technology by enabling real-time positioning and the construction of high-fidelity maps. However, the uncertainty in gaussian position and initialization parameters introduces challenges, often requiring extensive iterative convergence and resulting in redundant or insufficient gaussian representations. To address this, we introduce a novel adaptive densification method based on Fourier frequency domain analysis to establish gaussian priors for rapid convergence. Additionally, we propose constructing independent and unified sparse and dense maps, where a sparse map supports efficient tracking via Generalized Iterative Closest Point (GICP) and a dense map creates high-fidelity visual representations. This is the first SLAM system leveraging frequency domain analysis to achieve high-quality gaussian mapping in real-time. Experimental results demonstrate an average frame rate of 36 FPS on Replica and TUM RGB-D datasets, achieving competitive accuracy in both localization and mapping.
♻ ☆ MS-TVNet:A Long-Term Time Series Prediction Method Based on Multi-Scale Dynamic Convolution
Long-term time series prediction has predominantly relied on Transformer and MLP models, while the potential of convolutional networks in this domain remains underexplored. To address this gap, we introduce a novel multi-scale time series reshape module, which effectively captures the relationships among multi-period patches and variable dependencies. Building upon this module, we propose MS-TVNet, a multi-scale 3D dynamic convolutional neural network. Through comprehensive evaluations on diverse datasets, MS-TVNet demonstrates superior performance compared to baseline models, achieving state-of-the-art (SOTA) results in long-term time series prediction. Our findings highlight the effectiveness of leveraging convolutional networks for capturing complex temporal patterns, suggesting a promising direction for future research in this field.The code is realsed on https://github.com/Curyyfaust/TVNet.
♻ ☆ IKDiffuser: A Generative Inverse Kinematics Solver for Multi-arm Robots via Diffusion Model
Solving Inverse Kinematics (IK) problems is fundamental to robotics, but has primarily been successful with single serial manipulators. For multi-arm robotic systems, IK remains challenging due to complex self-collisions, coupled joints, and high-dimensional redundancy. These complexities make traditional IK solvers slow, prone to failure, and lacking in solution diversity. In this paper, we present IKDiffuser, a diffusion-based model designed for fast and diverse IK solution generation for multi-arm robotic systems. IKDiffuser learns the joint distribution over the configuration space, capturing complex dependencies and enabling seamless generalization to multi-arm robotic systems of different structures. In addition, IKDiffuser can incorporate additional objectives during inference without retraining, offering versatility and adaptability for task-specific requirements. In experiments on 6 different multi-arm systems, the proposed IKDiffuser achieves superior solution accuracy, precision, diversity, and computational efficiency compared to existing solvers. The proposed IKDiffuser framework offers a scalable, unified approach to solving multi-arm IK problems, facilitating the potential of multi-arm robotic systems in real-time manipulation tasks.
comment: under review
♻ ☆ ReconX: Reconstruct Any Scene from Sparse Views with Video Diffusion Model
Advancements in 3D scene reconstruction have transformed 2D images from the real world into 3D models, producing realistic 3D results from hundreds of input photos. Despite great success in dense-view reconstruction scenarios, rendering a detailed scene from insufficient captured views is still an ill-posed optimization problem, often resulting in artifacts and distortions in unseen areas. In this paper, we propose ReconX, a novel 3D scene reconstruction paradigm that reframes the ambiguous reconstruction challenge as a temporal generation task. The key insight is to unleash the strong generative prior of large pre-trained video diffusion models for sparse-view reconstruction. However, 3D view consistency struggles to be accurately preserved in directly generated video frames from pre-trained models. To address this, given limited input views, the proposed ReconX first constructs a global point cloud and encodes it into a contextual space as the 3D structure condition. Guided by the condition, the video diffusion model then synthesizes video frames that are both detail-preserved and exhibit a high degree of 3D consistency, ensuring the coherence of the scene from various perspectives. Finally, we recover the 3D scene from the generated video through a confidence-aware 3D Gaussian Splatting optimization scheme. Extensive experiments on various real-world datasets show the superiority of our ReconX over state-of-the-art methods in terms of quality and generalizability.
comment: Project page: https://liuff19.github.io/ReconX
♻ ☆ Hybrid AI for Responsive Multi-Turn Online Conversations with Novel Dynamic Routing and Feedback Adaptation NAACL 2025
Retrieval-Augmented Generation (RAG) systems and large language model (LLM)-powered chatbots have significantly advanced conversational AI by combining generative capabilities with external knowledge retrieval. Despite their success, enterprise-scale deployments face critical challenges, including diverse user queries, high latency, hallucinations, and difficulty integrating frequently updated domain-specific knowledge. This paper introduces a novel hybrid framework that integrates RAG with intent-based canned responses, leveraging predefined high-confidence responses for efficiency while dynamically routing complex or ambiguous queries to the RAG pipeline. Our framework employs a dialogue context manager to ensure coherence in multi-turn interactions and incorporates a feedback loop to refine intents, dynamically adjust confidence thresholds, and expand response coverage over time. Experimental results demonstrate that the proposed framework achieves a balance of high accuracy (95\%) and low latency (180ms), outperforming RAG and intent-based systems across diverse query types, positioning it as a scalable and adaptive solution for enterprise conversational AI applications.
comment: Proceedings of the 4th International Workshop on Knowledge Augmented Methods for Natural Language Processing in NAACL 2025, pages 215 to 229, Albuquerque, New Mexico, USA. Association for Computational Linguistics
♻ ☆ Mapping the Evolution of Research Contributions using KnoVo
This paper presents KnoVo (Knowledge Evolution), an intelligent framework designed for quantifying and analyzing the evolution of research novelty in the scientific literature. Moving beyond traditional citation analysis, which primarily measures impact, KnoVo determines a paper's novelty relative to both prior and subsequent work within its multilayered citation network. Given a target paper's abstract, KnoVo utilizes Large Language Models (LLMs) to dynamically extract dimensions of comparison (e.g., methodology, application, dataset). The target paper is then compared to related publications along these same extracted dimensions. This comparative analysis, inspired by tournament selection, yields quantitative novelty scores reflecting the relative improvement, equivalence, or inferiority of the target paper in specific aspects. By aggregating these scores and visualizing their progression, for instance, through dynamic evolution graphs and comparative radar charts, KnoVo facilitates researchers not only to assess originality and identify similar work, but also to track knowledge evolution along specific research dimensions, uncover research gaps, and explore cross-disciplinary connections. We demonstrate these capabilities through a detailed analysis of 20 diverse papers from multiple scientific fields and report on the performance of various open-source LLMs within the KnoVo framework.
♻ ☆ PhysUniBench: An Undergraduate-Level Physics Reasoning Benchmark for Multimodal Models
Physics problem-solving is a challenging domain for large AI models, requiring integration of conceptual understanding, mathematical reasoning, and interpretation of physical diagrams. Current evaluation methodologies show notable limitations in capturing the breadth and complexity of undergraduate-level physics, underscoring the need for more rigorous assessments. To this end, we present PhysUniBench, a large-scale multimodal benchmark designed to evaluate and improve the reasoning capabilities of multimodal large language models (MLLMs) specifically on undergraduate-level physics problems. PhysUniBench consists of 3,304 physics questions spanning 8 major sub-disciplines of physics, each accompanied by one visual diagrams. The benchmark includes both open-ended and multiple-choice questions, systematically curated and difficulty-rated through an iterative model-in-the-loop process. The benchmark's construction involved a rigorous multi-stage process, including multiple roll-outs, expert-level evaluation, automated filtering of easily solved problems, and a nuanced difficulty grading system with five levels. Through extensive experiments, we observe that current state-of-the-art models encounter substantial challenges in physics reasoning. For example, GPT-4o mini achieves only about 34.2% accuracy in the proposed PhysUniBench. These results highlight that current MLLMs struggle with advanced physics reasoning, especially on multi-step problems and those requiring precise diagram interpretation. By providing a broad and rigorous assessment tool, PhysUniBench aims to drive progress in AI for Science, encouraging the development of models with stronger physical reasoning, problem-solving skills, and multimodal understanding. The benchmark and evaluation scripts are available at https://prismax-team.github.io/PhysUniBenchmark/.
♻ ☆ USP-Gaussian: Unifying Spike-based Image Reconstruction, Pose Correction and Gaussian Splatting
Spike cameras, as an innovative neuromorphic camera that captures scenes with the 0-1 bit stream at 40 kHz, are increasingly employed for the 3D reconstruction task via Neural Radiance Fields (NeRF) or 3D Gaussian Splatting (3DGS). Previous spike-based 3D reconstruction approaches often employ a casecased pipeline: starting with high-quality image reconstruction from spike streams based on established spike-to-image reconstruction algorithms, then progressing to camera pose estimation and 3D reconstruction. However, this cascaded approach suffers from substantial cumulative errors, where quality limitations of initial image reconstructions negatively impact pose estimation, ultimately degrading the fidelity of the 3D reconstruction. To address these issues, we propose a synergistic optimization framework, \textbf{USP-Gaussian}, that unifies spike-based image reconstruction, pose correction, and Gaussian splatting into an end-to-end framework. Leveraging the multi-view consistency afforded by 3DGS and the motion capture capability of the spike camera, our framework enables a joint iterative optimization that seamlessly integrates information between the spike-to-image network and 3DGS. Experiments on synthetic datasets with accurate poses demonstrate that our method surpasses previous approaches by effectively eliminating cascading errors. Moreover, we integrate pose optimization to achieve robust 3D reconstruction in real-world scenarios with inaccurate initial poses, outperforming alternative methods by effectively reducing noise and preserving fine texture details. Our code, data and trained models will be available at https://github.com/chenkang455/USP-Gaussian.
♻ ☆ Rewarding Graph Reasoning Process makes LLMs more Generalized Reasoners KDD 2025
Despite significant advancements in Large Language Models (LLMs), developing advanced reasoning capabilities in LLMs remains a key challenge. Process Reward Models (PRMs) have demonstrated exceptional promise in enhancing reasoning by providing step-wise feedback, particularly in the context of mathematical reasoning. However, their application to broader reasoning domains remains understudied, largely due to the high costs associated with manually creating step-level supervision. In this work, we explore the potential of PRMs in graph reasoning problems - a domain that demands sophisticated multi-step reasoning and offers opportunities for automated step-level data generation using established graph algorithms. We introduce GraphSILO, the largest dataset for graph reasoning problems with fine-grained step-wise labels, built using automated Task-oriented Trajectories and Monte Carlo Tree Search (MCTS) to generate detailed reasoning steps with step-wise labels. Building upon this dataset, we train GraphPRM, the first PRM designed for graph reasoning problems, and evaluate its effectiveness in two key settings: inference-time scaling and reinforcement learning via Direct Preference Optimization (DPO). Experimental results show that GraphPRM significantly improves LLM performance across 13 graph reasoning tasks, delivering a 9% gain for Qwen2.5-7B and demonstrating transferability to new graph reasoning datasets and new reasoning domains like mathematical problem-solving. Notably, GraphPRM enhances LLM performance on GSM8K and Math500, underscoring the cross-domain applicability of graph-based reasoning rewards. Our findings highlight the potential of PRMs in advancing reasoning across diverse domains, paving the way for more versatile and effective LLMs.
comment: Accepted to KDD 2025 Research Track
♻ ☆ C3S3: Complementary Competition and Contrastive Selection for Semi-Supervised Medical Image Segmentation ICME 2025
For the immanent challenge of insufficiently annotated samples in the medical field, semi-supervised medical image segmentation (SSMIS) offers a promising solution. Despite achieving impressive results in delineating primary target areas, most current methodologies struggle to precisely capture the subtle details of boundaries. This deficiency often leads to significant diagnostic inaccuracies. To tackle this issue, we introduce C3S3, a novel semi-supervised segmentation model that synergistically integrates complementary competition and contrastive selection. This design significantly sharpens boundary delineation and enhances overall precision. Specifically, we develop an Outcome-Driven Contrastive Learning module dedicated to refining boundary localization. Additionally, we incorporate a Dynamic Complementary Competition module that leverages two high-performing sub-networks to generate pseudo-labels, thereby further improving segmentation quality. The proposed C3S3 undergoes rigorous validation on two publicly accessible datasets, encompassing the practices of both MRI and CT scans. The results demonstrate that our method achieves superior performance compared to previous cutting-edge competitors. Especially, on the 95HD and ASD metrics, our approach achieves a notable improvement of at least 6%, highlighting the significant advancements. The code is available at https://github.com/Y-TARL/C3S3.
comment: Accepted to ICME 2025
♻ ☆ AnchorDP3: 3D Affordance Guided Sparse Diffusion Policy for Robotic Manipulation
We present AnchorDP3, a diffusion policy framework for dual-arm robotic manipulation that achieves state-of-the-art performance in highly randomized environments. AnchorDP3 integrates three key innovations: (1) Simulator-Supervised Semantic Segmentation, using rendered ground truth to explicitly segment task-critical objects within the point cloud, which provides strong affordance priors; (2) Task-Conditioned Feature Encoders, lightweight modules processing augmented point clouds per task, enabling efficient multi-task learning through a shared diffusion-based action expert; (3) Affordance-Anchored Keypose Diffusion with Full State Supervision, replacing dense trajectory prediction with sparse, geometrically meaningful action anchors, i.e., keyposes such as pre-grasp pose, grasp pose directly anchored to affordances, drastically simplifying the prediction space; the action expert is forced to predict both robot joint angles and end-effector poses simultaneously, which exploits geometric consistency to accelerate convergence and boost accuracy. Trained on large-scale, procedurally generated simulation data, AnchorDP3 achieves a 98.7% average success rate in the RoboTwin benchmark across diverse tasks under extreme randomization of objects, clutter, table height, lighting, and backgrounds. This framework, when integrated with the RoboTwin real-to-sim pipeline, has the potential to enable fully autonomous generation of deployable visuomotor policies from only scene and instruction, totally eliminating human demonstrations from learning manipulation skills.
♻ ☆ Screen Hijack: Visual Poisoning of VLM Agents in Mobile Environments
With the growing integration of vision-language models (VLMs), mobile agents are now widely used for tasks like UI automation and camera-based user assistance. These agents are often fine-tuned on limited user-generated datasets, leaving them vulnerable to covert threats during the training process. In this work we present GHOST, the first clean-label backdoor attack specifically designed for mobile agents built upon VLMs. Our method manipulates only the visual inputs of a portion of the training samples - without altering their corresponding labels or instructions - thereby injecting malicious behaviors into the model. Once fine-tuned with this tampered data, the agent will exhibit attacker-controlled responses when a specific visual trigger is introduced at inference time. The core of our approach lies in aligning the gradients of poisoned samples with those of a chosen target instance, embedding backdoor-relevant features into the poisoned training data. To maintain stealth and enhance robustness, we develop three realistic visual triggers: static visual patches, dynamic motion cues, and subtle low-opacity overlays. We evaluate our method across six real-world Android apps and three VLM architectures adapted for mobile use. Results show that our attack achieves high attack success rates (up to 94.67 percent) while maintaining high clean-task performance (FSR up to 95.85 percent). Additionally, ablation studies shed light on how various design choices affect the efficacy and concealment of the attack. Overall, this work is the first to expose critical security flaws in VLM-based mobile agents, highlighting their susceptibility to clean-label backdoor attacks and the urgent need for effective defense mechanisms in their training pipelines.
comment: 12 pages
♻ ☆ TSPulse: Dual Space Tiny Pre-Trained Models for Rapid Time-Series Analysis
The rise of time-series pre-trained models has advanced temporal representation learning, but current state-of-the-art models are often large-scale, requiring substantial compute. We introduce TSPulse, ultra-compact time-series pre-trained models with only 1M parameters, specialized to perform strongly across classification, anomaly detection, imputation, and retrieval tasks. TSPulse introduces innovations at both the architecture and task levels. At the architecture level, it employs a dual-space masked reconstruction, learning from both time and frequency domains to capture complementary signals. This is further enhanced by a dual-embedding disentanglement, generating both detailed embeddings for fine-grained analysis and high-level semantic embeddings for broader task understanding. Notably, TSPulse's semantic embeddings are robust to shifts in time, magnitude, and noise, which is important for robust retrieval. At the task level, TSPulse incorporates TSLens, a fine-tuning component enabling task-specific feature attention. It also introduces a multi-head triangulation technique that correlates deviations from multiple prediction heads, enhancing anomaly detection by fusing complementary model outputs. Additionally, a hybrid mask pretraining is proposed to improves zero-shot imputation by reducing pre-training bias. These architecture and task innovations collectively contribute to TSPulse's significant performance gains: 5-16% on the UEA classification benchmarks, +20% on the TSB-AD anomaly detection leaderboard, +50% in zero-shot imputation, and +25% in time-series retrieval. Remarkably, these results are achieved with just 1M parameters (10-100X smaller than existing SOTA models) and allow GPU-free inference, setting a new standard for efficient time-series pre-trained models. The models can be accessed from https://huggingface.co/ibm-granite/granite-timeseries-tspulse-r1
♻ ☆ Evaluating Generalization and Representation Stability in Small LMs via Prompting, Fine-Tuning and Out-of-Distribution Prompts ICML
We investigate the generalization capabilities of small language models under two popular adaptation paradigms: few-shot prompting and supervised fine-tuning. While prompting is often favored for its parameter efficiency and flexibility, it remains unclear how robust this approach is in low-resource settings and under distributional shifts. This paper presents a comparative study of prompting and fine-tuning across task formats, prompt styles, and model scales, with a focus on their behavior in both in-distribution and out-of-distribution (OOD) settings. Beyond accuracy, we analyze the internal representations learned by each approach to assess the stability and abstraction of task-specific features. Our findings highlight critical differences in how small models internalize and generalize knowledge under different adaptation strategies. This work offers practical guidance for model selection in low-data regimes and contributes empirical insight into the ongoing debate over prompting versus fine-tuning. Code for the experiments is available at the following
comment: Accepted at ICML
♻ ☆ Robust Multimodal Learning for Ophthalmic Disease Grading via Disentangled Representation
This paper discusses how ophthalmologists often rely on multimodal data to improve diagnostic accuracy. However, complete multimodal data is rare in real-world applications due to a lack of medical equipment and concerns about data privacy. Traditional deep learning methods typically address these issues by learning representations in latent space. However, the paper highlights two key limitations of these approaches: (i) Task-irrelevant redundant information (e.g., numerous slices) in complex modalities leads to significant redundancy in latent space representations. (ii) Overlapping multimodal representations make it difficult to extract unique features for each modality. To overcome these challenges, the authors propose the Essence-Point and Disentangle Representation Learning (EDRL) strategy, which integrates a self-distillation mechanism into an end-to-end framework to enhance feature selection and disentanglement for more robust multimodal learning. Specifically, the Essence-Point Representation Learning module selects discriminative features that improve disease grading performance. The Disentangled Representation Learning module separates multimodal data into modality-common and modality-unique representations, reducing feature entanglement and enhancing both robustness and interpretability in ophthalmic disease diagnosis. Experiments on multimodal ophthalmology datasets show that the proposed EDRL strategy significantly outperforms current state-of-the-art methods.
comment: 10pages
♻ ☆ Morse: Dual-Sampling for Lossless Acceleration of Diffusion Models ICML 2025
In this paper, we present Morse, a simple dual-sampling framework for accelerating diffusion models losslessly. The key insight of Morse is to reformulate the iterative generation (from noise to data) process via taking advantage of fast jump sampling and adaptive residual feedback strategies. Specifically, Morse involves two models called Dash and Dot that interact with each other. The Dash model is just the pre-trained diffusion model of any type, but operates in a jump sampling regime, creating sufficient space for sampling efficiency improvement. The Dot model is significantly faster than the Dash model, which is learnt to generate residual feedback conditioned on the observations at the current jump sampling point on the trajectory of the Dash model, lifting the noise estimate to easily match the next-step estimate of the Dash model without jump sampling. By chaining the outputs of the Dash and Dot models run in a time-interleaved fashion, Morse exhibits the merit of flexibly attaining desired image generation performance while improving overall runtime efficiency. With our proposed weight sharing strategy between the Dash and Dot models, Morse is efficient for training and inference. Our method shows a lossless speedup of 1.78X to 3.31X on average over a wide range of sampling step budgets relative to 9 baseline diffusion models on 6 image generation tasks. Furthermore, we show that our method can be also generalized to improve the Latent Consistency Model (LCM-SDXL, which is already accelerated with consistency distillation technique) tailored for few-step text-to-image synthesis. The code and models are available at https://github.com/deep-optimization/Morse.
comment: Fixed a prompt typo in Figure 18 of the Appendix. This work is accepted to ICML 2025. The project page: https://github.com/deep-optimization/Morse
♻ ☆ PP-DocBee2: Improved Baselines with Efficient Data for Multimodal Document Understanding
This report introduces PP-DocBee2, an advanced version of the PP-DocBee, designed to enhance multimodal document understanding. Built on a large multimodal model architecture, PP-DocBee2 addresses the limitations of its predecessor through key technological improvements, including enhanced synthetic data quality, improved visual feature fusion strategy, and optimized inference methodologies. These enhancements yield an $11.4\%$ performance boost on internal benchmarks for Chinese business documents, and reduce inference latency by $73.0\%$ to the vanilla version. A key innovation of our work is a data quality optimization strategy for multimodal document tasks. By employing a large-scale multimodal pre-trained model to evaluate data, we apply a novel statistical criterion to filter outliers, ensuring high-quality training data. Inspired by insights into underutilized intermediate features in multimodal models, we enhance the ViT representational capacity by decomposing it into layers and applying a novel feature fusion strategy to improve complex reasoning. The source code and pre-trained model are available at \href{https://github.com/PaddlePaddle/PaddleMIX}{https://github.com/PaddlePaddle/PaddleMIX}.
♻ ☆ Fine-Grained Perturbation Guidance via Attention Head Selection
Recent guidance methods in diffusion models steer reverse sampling by perturbing the model to construct an implicit weak model and guide generation away from it. Among these approaches, attention perturbation has demonstrated strong empirical performance in unconditional scenarios where classifier-free guidance is not applicable. However, existing attention perturbation methods lack principled approaches for determining where perturbations should be applied, particularly in Diffusion Transformer (DiT) architectures where quality-relevant computations are distributed across layers. In this paper, we investigate the granularity of attention perturbations, ranging from the layer level down to individual attention heads, and discover that specific heads govern distinct visual concepts such as structure, style, and texture quality. Building on this insight, we propose "HeadHunter", a systematic framework for iteratively selecting attention heads that align with user-centric objectives, enabling fine-grained control over generation quality and visual attributes. In addition, we introduce SoftPAG, which linearly interpolates each selected head's attention map toward an identity matrix, providing a continuous knob to tune perturbation strength and suppress artifacts. Our approach not only mitigates the oversmoothing issues of existing layer-level perturbation but also enables targeted manipulation of specific visual styles through compositional head selection. We validate our method on modern large-scale DiT-based text-to-image models including Stable Diffusion 3 and FLUX.1, demonstrating superior performance in both general quality enhancement and style-specific guidance. Our work provides the first head-level analysis of attention perturbation in diffusion models, uncovering interpretable specialization within attention layers and enabling practical design of effective perturbation strategies.
comment: Project page: https://cvlab-kaist.github.io/HeadHunter/
♻ ☆ Understanding World or Predicting Future? A Comprehensive Survey of World Models
The concept of world models has garnered significant attention due to advancements in multimodal large language models such as GPT-4 and video generation models such as Sora, which are central to the pursuit of artificial general intelligence. This survey offers a comprehensive review of the literature on world models. Generally, world models are regarded as tools for either understanding the present state of the world or predicting its future dynamics. This review presents a systematic categorization of world models, emphasizing two primary functions: (1) constructing internal representations to understand the mechanisms of the world, and (2) predicting future states to simulate and guide decision-making. Initially, we examine the current progress in these two categories. We then explore the application of world models in key domains, including autonomous driving, robotics, and social simulacra, with a focus on how each domain utilizes these aspects. Finally, we outline key challenges and provide insights into potential future research directions. We summarize the representative papers along with their code repositories in https://github.com/tsinghua-fib-lab/World-Model.
comment: Accepted by ACM CSUR, 37 pages, 7 figures, 7 tables
♻ ☆ From System 1 to System 2: A Survey of Reasoning Large Language Models
Achieving human-level intelligence requires refining the transition from the fast, intuitive System 1 to the slower, more deliberate System 2 reasoning. While System 1 excels in quick, heuristic decisions, System 2 relies on logical reasoning for more accurate judgments and reduced biases. Foundational Large Language Models (LLMs) excel at fast decision-making but lack the depth for complex reasoning, as they have not yet fully embraced the step-by-step analysis characteristic of true System 2 thinking. Recently, reasoning LLMs like OpenAI's o1/o3 and DeepSeek's R1 have demonstrated expert-level performance in fields such as mathematics and coding, closely mimicking the deliberate reasoning of System 2 and showcasing human-like cognitive abilities. This survey begins with a brief overview of the progress in foundational LLMs and the early development of System 2 technologies, exploring how their combination has paved the way for reasoning LLMs. Next, we discuss how to construct reasoning LLMs, analyzing their features, the core methods enabling advanced reasoning, and the evolution of various reasoning LLMs. Additionally, we provide an overview of reasoning benchmarks, offering an in-depth comparison of the performance of representative reasoning LLMs. Finally, we explore promising directions for advancing reasoning LLMs and maintain a real-time \href{https://github.com/zzli2022/Awesome-Slow-Reason-System}{GitHub Repository} to track the latest developments. We hope this survey will serve as a valuable resource to inspire innovation and drive progress in this rapidly evolving field.
comment: Slow-thinking, Large Language Models, Human-like Reasoning, Decision Making in AI, AGI
♻ ☆ Supervised Quantum Machine Learning: A Future Outlook from Qubits to Enterprise Applications
Supervised Quantum Machine Learning (QML) represents an intersection of quantum computing and classical machine learning, aiming to use quantum resources to support model training and inference. This paper reviews recent developments in supervised QML, focusing on methods such as variational quantum circuits, quantum neural networks, and quantum kernel methods, along with hybrid quantum-classical workflows. We examine recent experimental studies that show partial indications of quantum advantage and describe current limitations including noise, barren plateaus, scalability issues, and the lack of formal proofs of performance improvement over classical methods. The main contribution is a ten-year outlook (2025-2035) that outlines possible developments in supervised QML, including a roadmap describing conditions under which QML may be used in applied research and enterprise systems over the next decade.
comment: Future outlook and roadmap of QML with 7 pages and 1 figure
♻ ☆ Turing Test 2.0: The General Intelligence Threshold
With the rise of artificial intelligence (A.I.) and large language models like ChatGPT, a new race for achieving artificial general intelligence (A.G.I) has started. While many speculate how and when A.I. will achieve A.G.I., there is no clear agreement on how A.G.I. can be detected in A.I. models, even when popular tools like the Turing test (and its modern variations) are used to measure their intelligence. In this work, we discuss why traditional methods like the Turing test do not suffice for measuring or detecting A.G.I. and provide a new, practical method that can be used to decide if a system (computer or any other) has reached or surpassed A.G.I. To achieve this, we make two new contributions. First, we present a clear definition for general intelligence (G.I.) and set a G.I. Threshold (G.I.T.) that can be used to distinguish between systems that achieve A.G.I. and systems that do not. Second, we present a new framework on how to construct tests that can detect if a system has achieved G.I. in a simple, comprehensive, and clear-cut fail/pass way. We call this novel framework the Turing test 2.0. We then demonstrate real-life examples of applying tests that follow our Turing test 2.0 framework on modern A.I. models.
♻ ☆ AIDRIN 2.0: A Framework to Assess Data Readiness for AI
AI Data Readiness Inspector (AIDRIN) is a framework to evaluate and improve data preparedness for AI applications. It addresses critical data readiness dimensions such as data quality, bias, fairness, and privacy. This paper details enhancements to AIDRIN by focusing on user interface improvements and integration with a privacy-preserving federated learning (PPFL) framework. By refining the UI and enabling smooth integration with decentralized AI pipelines, AIDRIN becomes more accessible and practical for users with varying technical expertise. Integrating with an existing PPFL framework ensures that data readiness and privacy are prioritized in federated learning environments. A case study involving a real-world dataset demonstrates AIDRIN's practical value in identifying data readiness issues that impact AI model performance.
comment: 3 pages, 3 figures
♻ ☆ Quantifying Fairness in LLMs Beyond Tokens: A Semantic and Statistical Perspective
Large Language Models (LLMs) often generate responses with inherent biases, undermining their reliability in real-world applications. Existing evaluation methods often overlook biases in long-form responses and the intrinsic variability of LLM outputs. To address these challenges, we propose FiSCo(Fine-grained Semantic Computation), a novel statistical framework to evaluate group-level fairness in LLMs by detecting subtle semantic differences in long-form responses across demographic groups. Unlike prior work focusing on sentiment or token-level comparisons, FiSCo goes beyond surface-level analysis by operating at the claim level, leveraging entailment checks to assess the consistency of meaning across responses. We decompose model outputs into semantically distinct claims and apply statistical hypothesis testing to compare inter- and intra-group similarities, enabling robust detection of subtle biases. We formalize a new group counterfactual fairness definition and validate FiSCo on both synthetic and human-annotated datasets spanning gender, race, and age. Experiments show that FiSco more reliably identifies nuanced biases while reducing the impact of stochastic LLM variability, outperforming various evaluation metrics.
comment: 29 pages, 9 figures, 15 tables
♻ ☆ Quantum-Classical Hybrid Quantized Neural Network
Here in this work, we present a novel Quadratic Binary Optimization (QBO) model for quantized neural network training, enabling the use of arbitrary activation and loss functions through spline interpolation. We introduce Forward Interval Propagation (FIP), a method designed to tackle the challenges of non-linearity and the multi-layer composite structure in neural networks by discretizing activation functions into linear subintervals. This approach preserves the universal approximation properties of neural networks while allowing complex nonlinear functions to be optimized using quantum computers, thus broadening their applicability in artificial intelligence. We provide theoretical upper bounds on the approximation error and the number of Ising spins required, by deriving the sample complexity of the empirical risk minimization problem, from an optimization perspective. A significant challenge in solving the associated Quadratic Constrained Binary Optimization (QCBO) model on a large scale is the presence of numerous constraints. When employing the penalty method to handle these constraints, tuning a large number of penalty coefficients becomes a critical hyperparameter optimization problem, increasing computational complexity and potentially affecting solution quality. To address this, we employ the Quantum Conditional Gradient Descent (QCGD) algorithm, which leverages quantum computing to directly solve the QCBO problem. We prove the convergence of QCGD under a quantum oracle with randomness and bounded variance in objective value, as well as under limited precision constraints in the coefficient matrix. Additionally, we provide an upper bound on the Time-To-Solution for the QCBO solving process. Experimental results using a coherent Ising machine (CIM) demonstrate a 94.95% accuracy on the Fashion MNIST classification task, with only 1.1-bit precision.
comment: 27 pages, 5 figures, comments are welcome
♻ ☆ Low-light Pedestrian Detection in Visible and Infrared Image Feeds: Issues and Challenges
Pedestrian detection has become a cornerstone for several high-level tasks, including autonomous driving, intelligent transportation, and traffic surveillance. There are several works focussed on pedestrian detection using visible images, mainly in the daytime. However, this task is very intriguing when the environmental conditions change to poor lighting or nighttime. Recently, new ideas have been spurred to use alternative sources, such as Far InfraRed (FIR) temperature sensor feeds for detecting pedestrians in low-light conditions. This study reviews recent developments in low-light pedestrian detection approaches. It systematically categorizes and analyses various algorithms from region-based to non-region-based and graph-based learning methodologies by highlighting their methodologies, implementation issues, and challenges. It also outlines the key benchmark datasets that can be used for research and development of advanced pedestrian detection algorithms, particularly in low-light situations.
comment: 29 pages, 4 tables, 21 figures
♻ ☆ Computation Mechanism Behind LLM Position Generalization ACL 2025
Most written natural languages are composed of sequences of words and sentences. Similar to humans, large language models (LLMs) exhibit flexibility in handling textual positions - a phenomenon we term position generalization. They can understand texts with position perturbations and generalize to longer texts than those encountered during training with the latest techniques. These phenomena suggest that LLMs handle positions tolerantly, but how LLMs computationally process positional relevance remains largely unexplored. This work connects the linguistic phenomenon with LLMs' computational mechanisms. We show how LLMs enforce certain computational mechanisms for the aforementioned tolerance in position perturbations. Despite the complex design of the self-attention mechanism, this work reveals that LLMs learn a counterintuitive disentanglement of attention logits. Their values show a 0.959 linear correlation with an approximation of the arithmetic sum of positional relevance and semantic importance. Furthermore, we identify a prevalent pattern in intermediate features, which we prove theoretically enables this effect. The pattern, which is different from how randomly initialized parameters would behave, suggests that it is a learned behavior rather than a natural result of the model architecture. Based on these findings, we provide computational explanations and criteria for LLMs' position flexibilities. This work takes a pioneering step in linking position generalization with modern LLMs' internal mechanisms.
comment: ACL 2025 Main Long Paper
♻ ☆ Thought Anchors: Which LLM Reasoning Steps Matter?
Reasoning large language models have recently achieved state-of-the-art performance in many fields. However, their long-form chain-of-thought reasoning creates interpretability challenges as each generated token depends on all previous ones, making the computation harder to decompose. We argue that analyzing reasoning traces at the sentence level is a promising approach to understanding reasoning processes. We present three complementary attribution methods: (1) a black-box method measuring each sentence's counterfactual importance by comparing final answers across 100 rollouts conditioned on the model generating that sentence or one with a different meaning; (2) a white-box method of aggregating attention patterns between pairs of sentences, which identified "broadcasting" sentences that receive disproportionate attention from all future sentences via "receiver" attention heads; (3) a causal attribution method measuring logical connections between sentences by suppressing attention toward one sentence and measuring the effect on each future sentence's tokens. Each method provides evidence for the existence of thought anchors, reasoning steps that have outsized importance and that disproportionately influence the subsequent reasoning process. These thought anchors are typically planning or backtracking sentences. We provide an open-source tool (www.thought-anchors.com) for visualizing the outputs of our methods, and present a case study showing converging patterns across methods that map how a model performs multi-step reasoning. The consistency across methods demonstrates the potential of sentence-level analysis for a deeper understanding of reasoning models.
comment: Paul C. Bogdan and Uzay Macar contributed equally to this work, and their listed order was determined by coinflip. Neel Nanda and Arthur Conmy contributed equally to this work as senior authors, and their listed order was determined by coinflip
Computer Vision and Pattern Recognition 113
☆ IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals
Semantic Scene Completion (SSC) has emerged as a pivotal approach for jointly learning scene geometry and semantics, enabling downstream applications such as navigation in mobile robotics. The recent generalization to Panoptic Scene Completion (PSC) advances the SSC domain by integrating instance-level information, thereby enhancing object-level sensitivity in scene understanding. While PSC was introduced using LiDAR modality, methods based on camera images remain largely unexplored. Moreover, recent Transformer-based SSC approaches utilize a fixed set of learned queries to reconstruct objects within the scene volume. Although these queries are typically updated with image context during training, they remain static at test time, limiting their ability to dynamically adapt specifically to the observed scene. To overcome these limitations, we propose IPFormer, the first approach that leverages context-adaptive instance proposals at train and test time to address vision-based 3D Panoptic Scene Completion. Specifically, IPFormer adaptively initializes these queries as panoptic instance proposals derived from image context and further refines them through attention-based encoding and decoding to reason about semantic instance-voxel relationships. Experimental results show that our approach surpasses state-of-the-art methods in overall panoptic metrics PQ$^\dagger$ and PQ-All, matches performance in individual metrics, and achieves a runtime reduction exceeding 14$\times$. Furthermore, our ablation studies reveal that dynamically deriving instance proposals from image context, as opposed to random initialization, leads to a 3.62% increase in PQ-All and a remarkable average improvement of 18.65% in combined Thing-metrics. These results highlight our introduction of context-adaptive instance proposals as a pioneering effort in addressing vision-based 3D Panoptic Scene Completion.
☆ MMSearch-R1: Incentivizing LMMs to Search
Robust deployment of large multimodal models (LMMs) in real-world scenarios requires access to external knowledge sources, given the complexity and dynamic nature of real-world information. Existing approaches such as retrieval-augmented generation (RAG) and prompt engineered search agents rely on rigid pipelines, often leading to inefficient or excessive search behaviors. We present MMSearch-R1, the first end-to-end reinforcement learning framework that enables LMMs to perform on-demand, multi-turn search in real-world Internet environments. Our framework integrates both image and text search tools, allowing the model to reason about when and how to invoke them guided by an outcome-based reward with a search penalty. To support training, We collect a multimodal search VQA dataset through a semi-automated pipeline that covers diverse visual and textual knowledge needs and curate a search-balanced subset with both search-required and search-free samples, which proves essential for shaping efficient and on-demand search behavior. Extensive experiments on knowledge-intensive and info-seeking VQA tasks show that our model not only outperforms RAG-based baselines of the same model size, but also matches the performance of a larger RAG-based model while reducing search calls by over 30%. We further analyze key empirical findings to offer actionable insights for advancing research in multimodal search.
comment: Code: https://github.com/EvolvingLMMs-Lab/multimodal-search-r1
☆ EditP23: 3D Editing via Propagation of Image Prompts to Multi-View
We present EditP23, a method for mask-free 3D editing that propagates 2D image edits to multi-view representations in a 3D-consistent manner. In contrast to traditional approaches that rely on text-based prompting or explicit spatial masks, EditP23 enables intuitive edits by conditioning on a pair of images: an original view and its user-edited counterpart. These image prompts are used to guide an edit-aware flow in the latent space of a pre-trained multi-view diffusion model, allowing the edit to be coherently propagated across views. Our method operates in a feed-forward manner, without optimization, and preserves the identity of the original object, in both structure and appearance. We demonstrate its effectiveness across a range of object categories and editing scenarios, achieving high fidelity to the source while requiring no manual masks.
comment: Code, supplementary videos, interactive 3D visualizations, and additional results are available at https://editp23.github.io/
☆ Disentangled representations of microscopy images IJCNN 2025
Microscopy image analysis is fundamental for different applications, from diagnosis to synthetic engineering and environmental monitoring. Modern acquisition systems have granted the possibility to acquire an escalating amount of images, requiring a consequent development of a large collection of deep learning-based automatic image analysis methods. Although deep neural networks have demonstrated great performance in this field, interpretability, an essential requirement for microscopy image analysis, remains an open challenge. This work proposes a Disentangled Representation Learning (DRL) methodology to enhance model interpretability for microscopy image classification. Exploiting benchmark datasets from three different microscopic image domains (plankton, yeast vacuoles, and human cells), we show how a DRL framework, based on transferring a representation learnt from synthetic data, can provide a good trade-off between accuracy and interpretability in this domain.
comment: Published in: International Joint Conference on Neural Networks (IJCNN 2025). Project page: https://github.com/JacopoDapueto/disentangled_microscopy
☆ Joint attitude estimation and 3D neural reconstruction of non-cooperative space objects CVPR 2025
Obtaining a better knowledge of the current state and behavior of objects orbiting Earth has proven to be essential for a range of applications such as active debris removal, in-orbit maintenance, or anomaly detection. 3D models represent a valuable source of information in the field of Space Situational Awareness (SSA). In this work, we leveraged Neural Radiance Fields (NeRF) to perform 3D reconstruction of non-cooperative space objects from simulated images. This scenario is challenging for NeRF models due to unusual camera characteristics and environmental conditions : mono-chromatic images, unknown object orientation, limited viewing angles, absence of diffuse lighting etc. In this work we focus primarly on the joint optimization of camera poses alongside the NeRF. Our experimental results show that the most accurate 3D reconstruction is achieved when training with successive images one-by-one. We estimate camera poses by optimizing an uniform rotation and use regularization to prevent successive poses from being too far apart.
comment: accepted for CVPR 2025 NFBCC workshop
☆ Shape2Animal: Creative Animal Generation from Natural Silhouettes
Humans possess a unique ability to perceive meaningful patterns in ambiguous stimuli, a cognitive phenomenon known as pareidolia. This paper introduces Shape2Animal framework to mimics this imaginative capacity by reinterpreting natural object silhouettes, such as clouds, stones, or flames, as plausible animal forms. Our automated framework first performs open-vocabulary segmentation to extract object silhouette and interprets semantically appropriate animal concepts using vision-language models. It then synthesizes an animal image that conforms to the input shape, leveraging text-to-image diffusion model and seamlessly blends it into the original scene to generate visually coherent and spatially consistent compositions. We evaluated Shape2Animal on a diverse set of real-world inputs, demonstrating its robustness and creative potential. Our Shape2Animal can offer new opportunities for visual storytelling, educational content, digital art, and interactive media design. Our project page is here: https://shape2image.github.io
☆ Weighted Mean Frequencies: a handcraft Fourier feature for 4D Flow MRI segmentation
In recent decades, the use of 4D Flow MRI images has enabled the quantification of velocity fields within a volume of interest and along the cardiac cycle. However, the lack of resolution and the presence of noise in these biomarkers are significant issues. As indicated by recent studies, it appears that biomarkers such as wall shear stress are particularly impacted by the poor resolution of vessel segmentation. The Phase Contrast Magnetic Resonance Angiography (PC-MRA) is the state-of-the-art method to facilitate segmentation. The objective of this work is to introduce a new handcraft feature that provides a novel visualisation of 4D Flow MRI images, which is useful in the segmentation task. This feature, termed Weighted Mean Frequencies (WMF), is capable of revealing the region in three dimensions where a voxel has been passed by pulsatile flow. Indeed, this feature is representative of the hull of all pulsatile velocity voxels. The value of the feature under discussion is illustrated by two experiments. The experiments involved segmenting 4D Flow MRI images using optimal thresholding and deep learning methods. The results obtained demonstrate a substantial enhancement in terms of IoU and Dice, with a respective increase of 0.12 and 0.13 in comparison with the PC-MRA feature, as evidenced by the deep learning task. This feature has the potential to yield valuable insights that could inform future segmentation processes in other vascular regions, such as the heart or the brain.
☆ Video Perception Models for 3D Scene Synthesis
Traditionally, 3D scene synthesis requires expert knowledge and significant manual effort. Automating this process could greatly benefit fields such as architectural design, robotics simulation, virtual reality, and gaming. Recent approaches to 3D scene synthesis often rely on the commonsense reasoning of large language models (LLMs) or strong visual priors of modern image generation models. However, current LLMs demonstrate limited 3D spatial reasoning ability, which restricts their ability to generate realistic and coherent 3D scenes. Meanwhile, image generation-based methods often suffer from constraints in viewpoint selection and multi-view inconsistencies. In this work, we present Video Perception models for 3D Scene synthesis (VIPScene), a novel framework that exploits the encoded commonsense knowledge of the 3D physical world in video generation models to ensure coherent scene layouts and consistent object placements across views. VIPScene accepts both text and image prompts and seamlessly integrates video generation, feedforward 3D reconstruction, and open-vocabulary perception models to semantically and geometrically analyze each object in a scene. This enables flexible scene synthesis with high realism and structural consistency. For more precise analysis, we further introduce First-Person View Score (FPVScore) for coherence and plausibility evaluation, utilizing continuous first-person perspective to capitalize on the reasoning ability of multimodal large language models. Extensive experiments show that VIPScene significantly outperforms existing methods and generalizes well across diverse scenarios. The code will be released.
☆ SFNet: Fusion of Spatial and Frequency-Domain Features for Remote Sensing Image Forgery Detection
The rapid advancement of generative artificial intelligence is producing fake remote sensing imagery (RSI) that is increasingly difficult to detect, potentially leading to erroneous intelligence, fake news, and even conspiracy theories. Existing forgery detection methods typically rely on single visual features to capture predefined artifacts, such as spatial-domain cues to detect forged objects like roads or buildings in RSI, or frequency-domain features to identify artifacts from up-sampling operations in adversarial generative networks (GANs). However, the nature of artifacts can significantly differ depending on geographic terrain, land cover types, or specific features within the RSI. Moreover, these complex artifacts evolve as generative models become more sophisticated. In short, over-reliance on a single visual cue makes existing forgery detectors struggle to generalize across diverse remote sensing data. This paper proposed a novel forgery detection framework called SFNet, designed to identify fake images in diverse remote sensing data by leveraging spatial and frequency domain features. Specifically, to obtain rich and comprehensive visual information, SFNet employs two independent feature extractors to capture spatial and frequency domain features from input RSIs. To fully utilize the complementary domain features, the domain feature mapping module and the hybrid domain feature refinement module(CBAM attention) of SFNet are designed to successively align and fuse the multi-domain features while suppressing redundant information. Experiments on three datasets show that SFNet achieves an accuracy improvement of 4%-15.18% over the state-of-the-art RS forgery detection methods and exhibits robust generalization capabilities. The code is available at https://github.com/GeoX-Lab/RSTI/tree/main/SFNet.
☆ WonderFree: Enhancing Novel View Quality and Cross-View Consistency for 3D Scene Exploration
Interactive 3D scene generation from a single image has gained significant attention due to its potential to create immersive virtual worlds. However, a key challenge in current 3D generation methods is the limited explorability, which cannot render high-quality images during larger maneuvers beyond the original viewpoint, particularly when attempting to move forward into unseen areas. To address this challenge, we propose WonderFree, the first model that enables users to interactively generate 3D worlds with the freedom to explore from arbitrary angles and directions. Specifically, we decouple this challenge into two key subproblems: novel view quality, which addresses visual artifacts and floating issues in novel views, and cross-view consistency, which ensures spatial consistency across different viewpoints. To enhance rendering quality in novel views, we introduce WorldRestorer, a data-driven video restoration model designed to eliminate floaters and artifacts. In addition, a data collection pipeline is presented to automatically gather training data for WorldRestorer, ensuring it can handle scenes with varying styles needed for 3D scene generation. Furthermore, to improve cross-view consistency, we propose ConsistView, a multi-view joint restoration mechanism that simultaneously restores multiple perspectives while maintaining spatiotemporal coherence. Experimental results demonstrate that WonderFree not only enhances rendering quality across diverse viewpoints but also significantly improves global coherence and consistency. These improvements are confirmed by CLIP-based metrics and a user study showing a 77.20% preference for WonderFree over WonderWorld enabling a seamless and immersive 3D exploration experience. The code, model, and data will be publicly available.
☆ TRIM: A Self-Supervised Video Summarization Framework Maximizing Temporal Relative Information and Representativeness
The increasing ubiquity of video content and the corresponding demand for efficient access to meaningful information have elevated video summarization and video highlights as a vital research area. However, many state-of-the-art methods depend heavily either on supervised annotations or on attention-based models, which are computationally expensive and brittle in the face of distribution shifts that hinder cross-domain applicability across datasets. We introduce a pioneering self-supervised video summarization model that captures both spatial and temporal dependencies without the overhead of attention, RNNs, or transformers. Our framework integrates a novel set of Markov process-driven loss metrics and a two-stage self supervised learning paradigm that ensures both performance and efficiency. Our approach achieves state-of-the-art performance on the SUMME and TVSUM datasets, outperforming all existing unsupervised methods. It also rivals the best supervised models, demonstrating the potential for efficient, annotation-free architectures. This paves the way for more generalizable video summarization techniques and challenges the prevailing reliance on complex architectures.
☆ Learning-Based Distance Estimation for 360° Single-Sensor Setups
Accurate distance estimation is a fundamental challenge in robotic perception, particularly in omnidirectional imaging, where traditional geometric methods struggle with lens distortions and environmental variability. In this work, we propose a neural network-based approach for monocular distance estimation using a single 360{\deg} fisheye lens camera. Unlike classical trigonometric techniques that rely on precise lens calibration, our method directly learns and infers the distance of objects from raw omnidirectional inputs, offering greater robustness and adaptability across diverse conditions. We evaluate our approach on three 360{\deg} datasets (LOAF, ULM360, and a newly captured dataset Boat360), each representing distinct environmental and sensor setups. Our experimental results demonstrate that the proposed learning-based model outperforms traditional geometry-based methods and other learning baselines in both accuracy and robustness. These findings highlight the potential of deep learning for real-time omnidirectional distance estimation, making our approach particularly well-suited for low-cost applications in robotics, autonomous navigation, and surveillance.
comment: Submitted to ECMR 2025
☆ Dense Video Captioning using Graph-based Sentence Summarization
Recently, dense video captioning has made attractive progress in detecting and captioning all events in a long untrimmed video. Despite promising results were achieved, most existing methods do not sufficiently explore the scene evolution within an event temporal proposal for captioning, and therefore perform less satisfactorily when the scenes and objects change over a relatively long proposal. To address this problem, we propose a graph-based partition-and-summarization (GPaS) framework for dense video captioning within two stages. For the ``partition" stage, a whole event proposal is split into short video segments for captioning at a finer level. For the ``summarization" stage, the generated sentences carrying rich description information for each segment are summarized into one sentence to describe the whole event. We particularly focus on the ``summarization" stage, and propose a framework that effectively exploits the relationship between semantic words for summarization. We achieve this goal by treating semantic words as nodes in a graph and learning their interactions by coupling Graph Convolutional Network (GCN) and Long Short Term Memory (LSTM), with the aid of visual cues. Two schemes of GCN-LSTM Interaction (GLI) modules are proposed for seamless integration of GCN and LSTM. The effectiveness of our approach is demonstrated via an extensive comparison with the state-of-the-arts methods on the two benchmarks ActivityNet Captions dataset and YouCook II dataset.
comment: 12 pages
☆ Causal Representation Learning with Observational Grouping for CXR Classification
Identifiable causal representation learning seeks to uncover the true causal relationships underlying a data generation process. In medical imaging, this presents opportunities to improve the generalisability and robustness of task-specific latent features. This work introduces the concept of grouping observations to learn identifiable representations for disease classification in chest X-rays via an end-to-end framework. Our experiments demonstrate that these causal representations improve generalisability and robustness across multiple classification tasks when grouping is used to enforce invariance w.r.t race, sex, and imaging views.
☆ Show, Tell and Summarize: Dense Video Captioning Using Visual Cue Aided Sentence Summarization
In this work, we propose a division-and-summarization (DaS) framework for dense video captioning. After partitioning each untrimmed long video as multiple event proposals, where each event proposal consists of a set of short video segments, we extract visual feature (e.g., C3D feature) from each segment and use the existing image/video captioning approach to generate one sentence description for this segment. Considering that the generated sentences contain rich semantic descriptions about the whole event proposal, we formulate the dense video captioning task as a visual cue aided sentence summarization problem and propose a new two stage Long Short Term Memory (LSTM) approach equipped with a new hierarchical attention mechanism to summarize all generated sentences as one descriptive sentence with the aid of visual features. Specifically, the first-stage LSTM network takes all semantic words from the generated sentences and the visual features from all segments within one event proposal as the input, and acts as the encoder to effectively summarize both semantic and visual information related to this event proposal. The second-stage LSTM network takes the output from the first-stage LSTM network and the visual features from all video segments within one event proposal as the input, and acts as the decoder to generate one descriptive sentence for this event proposal. Our comprehensive experiments on the ActivityNet Captions dataset demonstrate the effectiveness of our newly proposed DaS framework for dense video captioning.
comment: 10 pages
☆ HRIBench: Benchmarking Vision-Language Models for Real-Time Human Perception in Human-Robot Interaction
Real-time human perception is crucial for effective human-robot interaction (HRI). Large vision-language models (VLMs) offer promising generalizable perceptual capabilities but often suffer from high latency, which negatively impacts user experience and limits VLM applicability in real-world scenarios. To systematically study VLM capabilities in human perception for HRI and performance-latency trade-offs, we introduce HRIBench, a visual question-answering (VQA) benchmark designed to evaluate VLMs across a diverse set of human perceptual tasks critical for HRI. HRIBench covers five key domains: (1) non-verbal cue understanding, (2) verbal instruction understanding, (3) human-robot object relationship understanding, (4) social navigation, and (5) person identification. To construct HRIBench, we collected data from real-world HRI environments to curate questions for non-verbal cue understanding, and leveraged publicly available datasets for the remaining four domains. We curated 200 VQA questions for each domain, resulting in a total of 1000 questions for HRIBench. We then conducted a comprehensive evaluation of both state-of-the-art closed-source and open-source VLMs (N=11) on HRIBench. Our results show that, despite their generalizability, current VLMs still struggle with core perceptual capabilities essential for HRI. Moreover, none of the models within our experiments demonstrated a satisfactory performance-latency trade-off suitable for real-time deployment, underscoring the need for future research on developing smaller, low-latency VLMs with improved human perception capabilities. HRIBench and our results can be found in this Github repository: https://github.com/interaction-lab/HRIBench.
comment: Accepted to the 19th International Symposium on Experimental Robotics (ISER 2025)
☆ AdvMIM: Adversarial Masked Image Modeling for Semi-Supervised Medical Image Segmentation MICCAI 2025
Vision Transformer has recently gained tremendous popularity in medical image segmentation task due to its superior capability in capturing long-range dependencies. However, transformer requires a large amount of labeled data to be effective, which hinders its applicability in annotation scarce semi-supervised learning scenario where only limited labeled data is available. State-of-the-art semi-supervised learning methods propose combinatorial CNN-Transformer learning to cross teach a transformer with a convolutional neural network, which achieves promising results. However, it remains a challenging task to effectively train the transformer with limited labeled data. In this paper, we propose an adversarial masked image modeling method to fully unleash the potential of transformer for semi-supervised medical image segmentation. The key challenge in semi-supervised learning with transformer lies in the lack of sufficient supervision signal. To this end, we propose to construct an auxiliary masked domain from original domain with masked image modeling and train the transformer to predict the entire segmentation mask with masked inputs to increase supervision signal. We leverage the original labels from labeled data and pseudo-labels from unlabeled data to learn the masked domain. To further benefit the original domain from masked domain, we provide a theoretical analysis of our method from a multi-domain learning perspective and devise a novel adversarial training loss to reduce the domain gap between the original and masked domain, which boosts semi-supervised learning performance. We also extend adversarial masked image modeling to CNN network. Extensive experiments on three public medical image segmentation datasets demonstrate the effectiveness of our method, where our method outperforms existing methods significantly. Our code is publicly available at https://github.com/zlheui/AdvMIM.
comment: Accepted to MICCAI 2025
☆ Lightweight Multi-Frame Integration for Robust YOLO Object Detection in Videos
Modern image-based object detection models, such as YOLOv7, primarily process individual frames independently, thus ignoring valuable temporal context naturally present in videos. Meanwhile, existing video-based detection methods often introduce complex temporal modules, significantly increasing model size and computational complexity. In practical applications such as surveillance and autonomous driving, transient challenges including motion blur, occlusions, and abrupt appearance changes can severely degrade single-frame detection performance. To address these issues, we propose a straightforward yet highly effective strategy: stacking multiple consecutive frames as input to a YOLO-based detector while supervising only the output corresponding to a single target frame. This approach leverages temporal information with minimal modifications to existing architectures, preserving simplicity, computational efficiency, and real-time inference capability. Extensive experiments on the challenging MOT20Det and our BOAT360 datasets demonstrate that our method improves detection robustness, especially for lightweight models, effectively narrowing the gap between compact and heavy detection networks. Additionally, we contribute the BOAT360 benchmark dataset, comprising annotated fisheye video sequences captured from a boat, to support future research in multi-frame video object detection in challenging real-world scenarios.
comment: Submitted to ECMR 2025
☆ Pay Less Attention to Deceptive Artifacts: Robust Detection of Compressed Deepfakes on Online Social Networks
With the rapid advancement of deep learning, particularly through generative adversarial networks (GANs) and diffusion models (DMs), AI-generated images, or ``deepfakes", have become nearly indistinguishable from real ones. These images are widely shared across Online Social Networks (OSNs), raising concerns about their misuse. Existing deepfake detection methods overlook the ``block effects" introduced by compression in OSNs, which obscure deepfake artifacts, and primarily focus on raw images, rarely encountered in real-world scenarios. To address these challenges, we propose PLADA (Pay Less Attention to Deceptive Artifacts), a novel framework designed to tackle the lack of paired data and the ineffective use of compressed images. PLADA consists of two core modules: Block Effect Eraser (B2E), which uses a dual-stage attention mechanism to handle block effects, and Open Data Aggregation (ODA), which processes both paired and unpaired data to improve detection. Extensive experiments across 26 datasets demonstrate that PLADA achieves a remarkable balance in deepfake detection, outperforming SoTA methods in detecting deepfakes on OSNs, even with limited paired data and compression. More importantly, this work introduces the ``block effect" as a critical factor in deepfake detection, providing a robust solution for open-world scenarios. Our code is available at https://github.com/ManyiLee/PLADA.
comment: 20 pages, 10 figures
☆ AI-assisted radiographic analysis in detecting alveolar bone-loss severity and patterns
Periodontitis, a chronic inflammatory disease causing alveolar bone loss, significantly affects oral health and quality of life. Accurate assessment of bone loss severity and pattern is critical for diagnosis and treatment planning. In this study, we propose a novel AI-based deep learning framework to automatically detect and quantify alveolar bone loss and its patterns using intraoral periapical (IOPA) radiographs. Our method combines YOLOv8 for tooth detection with Keypoint R-CNN models to identify anatomical landmarks, enabling precise calculation of bone loss severity. Additionally, YOLOv8x-seg models segment bone levels and tooth masks to determine bone loss patterns (horizontal vs. angular) via geometric analysis. Evaluated on a large, expertly annotated dataset of 1000 radiographs, our approach achieved high accuracy in detecting bone loss severity (intra-class correlation coefficient up to 0.80) and bone loss pattern classification (accuracy 87%). This automated system offers a rapid, objective, and reproducible tool for periodontal assessment, reducing reliance on subjective manual evaluation. By integrating AI into dental radiographic analysis, our framework has the potential to improve early diagnosis and personalized treatment planning for periodontitis, ultimately enhancing patient care and clinical outcomes.
comment: This manuscript is 17 pages with 5 tables and 12 figures. The manuscript is under review at Nature Scientific Reports
☆ A Deep Learning Approach to Identify Rock Bolts in Complex 3D Point Clouds of Underground Mines Captured Using Mobile Laser Scanners
Rock bolts are crucial components of the subterranean support systems in underground mines that provide adequate structural reinforcement to the rock mass to prevent unforeseen hazards like rockfalls. This makes frequent assessments of such bolts critical for maintaining rock mass stability and minimising risks in underground mining operations. Where manual surveying of rock bolts is challenging due to the low light conditions in the underground mines and the time-intensive nature of the process, automated detection of rock bolts serves as a plausible solution. To that end, this study focuses on the automatic identification of rock bolts within medium to large-scale 3D point clouds obtained from underground mines using mobile laser scanners. Existing techniques for automated rock bolt identification primarily rely on feature engineering and traditional machine learning approaches. However, such techniques lack robustness as these point clouds present several challenges due to data noise, varying environments, and complex surrounding structures. Moreover, the target rock bolts are extremely small objects within large-scale point clouds and are often partially obscured due to the application of reinforcement shotcrete. Addressing these challenges, this paper proposes an approach termed DeepBolt, which employs a novel two-stage deep learning architecture specifically designed for handling severe class imbalance for the automatic and efficient identification of rock bolts in complex 3D point clouds. The proposed method surpasses state-of-the-art semantic segmentation models by up to 42.5% in Intersection over Union (IoU) for rock bolt points. Additionally, it outperforms existing rock bolt identification techniques, achieving a 96.41% precision and 96.96% recall in classifying rock bolts, demonstrating its robustness and effectiveness in complex underground environments.
☆ HiWave: Training-Free High-Resolution Image Generation via Wavelet-Based Diffusion Sampling
Diffusion models have emerged as the leading approach for image synthesis, demonstrating exceptional photorealism and diversity. However, training diffusion models at high resolutions remains computationally prohibitive, and existing zero-shot generation techniques for synthesizing images beyond training resolutions often produce artifacts, including object duplication and spatial incoherence. In this paper, we introduce HiWave, a training-free, zero-shot approach that substantially enhances visual fidelity and structural coherence in ultra-high-resolution image synthesis using pretrained diffusion models. Our method employs a two-stage pipeline: generating a base image from the pretrained model followed by a patch-wise DDIM inversion step and a novel wavelet-based detail enhancer module. Specifically, we first utilize inversion methods to derive initial noise vectors that preserve global coherence from the base image. Subsequently, during sampling, our wavelet-domain detail enhancer retains low-frequency components from the base image to ensure structural consistency, while selectively guiding high-frequency components to enrich fine details and textures. Extensive evaluations using Stable Diffusion XL demonstrate that HiWave effectively mitigates common visual artifacts seen in prior methods, achieving superior perceptual quality. A user study confirmed HiWave's performance, where it was preferred over the state-of-the-art alternative in more than 80% of comparisons, highlighting its effectiveness for high-quality, ultra-high-resolution image synthesis without requiring retraining or architectural modifications.
☆ Med-Art: Diffusion Transformer for 2D Medical Text-to-Image Generation
Text-to-image generative models have achieved remarkable breakthroughs in recent years. However, their application in medical image generation still faces significant challenges, including small dataset sizes, and scarcity of medical textual data. To address these challenges, we propose Med-Art, a framework specifically designed for medical image generation with limited data. Med-Art leverages vision-language models to generate visual descriptions of medical images which overcomes the scarcity of applicable medical textual data. Med-Art adapts a large-scale pre-trained text-to-image model, PixArt-$\alpha$, based on the Diffusion Transformer (DiT), achieving high performance under limited data. Furthermore, we propose an innovative Hybrid-Level Diffusion Fine-tuning (HLDF) method, which enables pixel-level losses, effectively addressing issues such as overly saturated colors. We achieve state-of-the-art performance on two medical image datasets, measured by FID, KID, and downstream classification performance.
comment: The project is available at \url{https://medart-ai.github.io}
☆ An Agentic System for Rare Disease Diagnosis with Traceable Reasoning
Rare diseases collectively affect over 300 million individuals worldwide, yet timely and accurate diagnosis remains a pervasive challenge. This is largely due to their clinical heterogeneity, low individual prevalence, and the limited familiarity most clinicians have with rare conditions. Here, we introduce DeepRare, the first rare disease diagnosis agentic system powered by a large language model (LLM), capable of processing heterogeneous clinical inputs. The system generates ranked diagnostic hypotheses for rare diseases, each accompanied by a transparent chain of reasoning that links intermediate analytic steps to verifiable medical evidence. DeepRare comprises three key components: a central host with a long-term memory module; specialized agent servers responsible for domain-specific analytical tasks integrating over 40 specialized tools and web-scale, up-to-date medical knowledge sources, ensuring access to the most current clinical information. This modular and scalable design enables complex diagnostic reasoning while maintaining traceability and adaptability. We evaluate DeepRare on eight datasets. The system demonstrates exceptional diagnostic performance among 2,919 diseases, achieving 100% accuracy for 1013 diseases. In HPO-based evaluations, DeepRare significantly outperforms other 15 methods, like traditional bioinformatics diagnostic tools, LLMs, and other agentic systems, achieving an average Recall@1 score of 57.18% and surpassing the second-best method (Reasoning LLM) by a substantial margin of 23.79 percentage points. For multi-modal input scenarios, DeepRare achieves 70.60% at Recall@1 compared to Exomiser's 53.20% in 109 cases. Manual verification of reasoning chains by clinical experts achieves 95.40% agreements. Furthermore, the DeepRare system has been implemented as a user-friendly web application http://raredx.cn/doctor.
☆ Fusing Radiomic Features with Deep Representations for Gestational Age Estimation in Fetal Ultrasound Images MICCAI 2025
Accurate gestational age (GA) estimation, ideally through fetal ultrasound measurement, is a crucial aspect of providing excellent antenatal care. However, deriving GA from manual fetal biometric measurements depends on the operator and is time-consuming. Hence, automatic computer-assisted methods are demanded in clinical practice. In this paper, we present a novel feature fusion framework to estimate GA using fetal ultrasound images without any measurement information. We adopt a deep learning model to extract deep representations from ultrasound images. We extract radiomic features to reveal patterns and characteristics of fetal brain growth. To harness the interpretability of radiomics in medical imaging analysis, we estimate GA by fusing radiomic features and deep representations. Our framework estimates GA with a mean absolute error of 8.0 days across three trimesters, outperforming current machine learning-based methods at these gestational ages. Experimental results demonstrate the robustness of our framework across different populations in diverse geographical regions. Our code is publicly available on \href{https://github.com/13204942/RadiomicsImageFusion_FetalUS}{GitHub}.
comment: Accepted at MICCAI 2025
☆ A Novel Large Vision Foundation Model (LVFM)-based Approach for Generating High-Resolution Canopy Height Maps in Plantations for Precision Forestry Management
Accurate, cost-effective monitoring of plantation aboveground biomass (AGB) is crucial for supporting local livelihoods and carbon sequestration initiatives like the China Certified Emission Reduction (CCER) program. High-resolution canopy height maps (CHMs) are essential for this, but standard lidar-based methods are expensive. While deep learning with RGB imagery offers an alternative, accurately extracting canopy height features remains challenging. To address this, we developed a novel model for high-resolution CHM generation using a Large Vision Foundation Model (LVFM). Our model integrates a feature extractor, a self-supervised feature enhancement module to preserve spatial details, and a height estimator. Tested in Beijing's Fangshan District using 1-meter Google Earth imagery, our model outperformed existing methods, including conventional CNNs. It achieved a mean absolute error of 0.09 m, a root mean square error of 0.24 m, and a correlation of 0.78 against lidar-based CHMs. The resulting CHMs enabled over 90% success in individual tree detection, high accuracy in AGB estimation, and effective tracking of plantation growth, demonstrating strong generalization to non-training areas. This approach presents a promising, scalable tool for evaluating carbon sequestration in both plantations and natural forests.
☆ Exploiting Lightweight Hierarchical ViT and Dynamic Framework for Efficient Visual Tracking
Transformer-based visual trackers have demonstrated significant advancements due to their powerful modeling capabilities. However, their practicality is limited on resource-constrained devices because of their slow processing speeds. To address this challenge, we present HiT, a novel family of efficient tracking models that achieve high performance while maintaining fast operation across various devices. The core innovation of HiT lies in its Bridge Module, which connects lightweight transformers to the tracking framework, enhancing feature representation quality. Additionally, we introduce a dual-image position encoding approach to effectively encode spatial information. HiT achieves an impressive speed of 61 frames per second (fps) on the NVIDIA Jetson AGX platform, alongside a competitive AUC of 64.6% on the LaSOT benchmark, outperforming all previous efficient trackers.Building on HiT, we propose DyHiT, an efficient dynamic tracker that flexibly adapts to scene complexity by selecting routes with varying computational requirements. DyHiT uses search area features extracted by the backbone network and inputs them into an efficient dynamic router to classify tracking scenarios. Based on the classification, DyHiT applies a divide-and-conquer strategy, selecting appropriate routes to achieve a superior trade-off between accuracy and speed. The fastest version of DyHiT achieves 111 fps on NVIDIA Jetson AGX while maintaining an AUC of 62.4% on LaSOT.Furthermore, we introduce a training-free acceleration method based on the dynamic routing architecture of DyHiT. This method significantly improves the execution speed of various high-performance trackers without sacrificing accuracy. For instance, our acceleration method enables the state-of-the-art tracker SeqTrack-B256 to achieve a 2.68 times speedup on an NVIDIA GeForce RTX 2080 Ti GPU while maintaining the same AUC of 69.9% on the LaSOT.
comment: This paper was accepted by International Journal of Computer Vision(IJCV)
☆ InvZW: Invariant Feature Learning via Noise-Adversarial Training for Robust Image Zero-Watermarking
This paper introduces a novel deep learning framework for robust image zero-watermarking based on distortion-invariant feature learning. As a zero-watermarking scheme, our method leaves the original image unaltered and learns a reference signature through optimization in the feature space. The proposed framework consists of two key modules. In the first module, a feature extractor is trained via noise-adversarial learning to generate representations that are both invariant to distortions and semantically expressive. This is achieved by combining adversarial supervision against a distortion discriminator and a reconstruction constraint to retain image content. In the second module, we design a learning-based multibit zero-watermarking scheme where the trained invariant features are projected onto a set of trainable reference codes optimized to match a target binary message. Extensive experiments on diverse image datasets and a wide range of distortions show that our method achieves state-of-the-art robustness in both feature stability and watermark recovery. Comparative evaluations against existing self-supervised and deep watermarking techniques further highlight the superiority of our framework in generalization and robustness.
☆ DreamAnywhere: Object-Centric Panoramic 3D Scene Generation
Recent advances in text-to-3D scene generation have demonstrated significant potential to transform content creation across multiple industries. Although the research community has made impressive progress in addressing the challenges of this complex task, existing methods often generate environments that are only front-facing, lack visual fidelity, exhibit limited scene understanding, and are typically fine-tuned for either indoor or outdoor settings. In this work, we address these issues and propose DreamAnywhere, a modular system for the fast generation and prototyping of 3D scenes. Our system synthesizes a 360{\deg} panoramic image from text, decomposes it into background and objects, constructs a complete 3D representation through hybrid inpainting, and lifts object masks to detailed 3D objects that are placed in the virtual environment. DreamAnywhere supports immersive navigation and intuitive object-level editing, making it ideal for scene exploration, visual mock-ups, and rapid prototyping -- all with minimal manual modeling. These features make our system particularly suitable for low-budget movie production, enabling quick iteration on scene layout and visual tone without the overhead of traditional 3D workflows. Our modular pipeline is highly customizable as it allows components to be replaced independently. Compared to current state-of-the-art text and image-based 3D scene generation approaches, DreamAnywhere shows significant improvements in coherence in novel view synthesis and achieves competitive image quality, demonstrating its effectiveness across diverse and challenging scenarios. A comprehensive user study demonstrates a clear preference for our method over existing approaches, validating both its technical robustness and practical usefulness.
☆ Practical insights on the effect of different encodings, ansätze and measurements in quantum and hybrid convolutional neural networks
This study investigates the design choices of parameterized quantum circuits (PQCs) within quantum and hybrid convolutional neural network (HQNN and QCNN) architectures, applied to the task of satellite image classification using the EuroSAT dataset. We systematically evaluate the performance implications of data encoding techniques, variational ans\"atze, and measurement in approx. 500 distinct model configurations. Our analysis reveals a clear hierarchy of influence on model performance. For hybrid architectures, which were benchmarked against their direct classical equivalents (e.g. the same architecture with the PQCs removed), the data encoding strategy is the dominant factor, with validation accuracy varying over 30% for distinct embeddings. In contrast, the selection of variational ans\"atze and measurement basis had a comparatively marginal effect, with validation accuracy variations remaining below 5%. For purely quantum models, restricted to amplitude encoding, performance was most dependent on the measurement protocol and the data-to-amplitude mapping. The measurement strategy varied the validation accuracy by up to 30% and the encoding mapping by around 8 percentage points.
comment: 20 pages, 22 figures
☆ Feature Hallucination for Self-supervised Action Recognition
Understanding human actions in videos requires more than raw pixel analysis; it relies on high-level semantic reasoning and effective integration of multimodal features. We propose a deep translational action recognition framework that enhances recognition accuracy by jointly predicting action concepts and auxiliary features from RGB video frames. At test time, hallucination streams infer missing cues, enriching feature representations without increasing computational overhead. To focus on action-relevant regions beyond raw pixels, we introduce two novel domain-specific descriptors. Object Detection Features (ODF) aggregate outputs from multiple object detectors to capture contextual cues, while Saliency Detection Features (SDF) highlight spatial and intensity patterns crucial for action recognition. Our framework seamlessly integrates these descriptors with auxiliary modalities such as optical flow, Improved Dense Trajectories, skeleton data, and audio cues. It remains compatible with state-of-the-art architectures, including I3D, AssembleNet, Video Transformer Network, FASTER, and recent models like VideoMAE V2 and InternVideo2. To handle uncertainty in auxiliary features, we incorporate aleatoric uncertainty modeling in the hallucination step and introduce a robust loss function to mitigate feature noise. Our multimodal self-supervised action recognition framework achieves state-of-the-art performance on multiple benchmarks, including Kinetics-400, Kinetics-600, and Something-Something V2, demonstrating its effectiveness in capturing fine-grained action dynamics.
comment: Accepted for publication in International Journal of Computer Vision (IJCV)
☆ EAGLE: An Efficient Global Attention Lesion Segmentation Model for Hepatic Echinococcosis
Hepatic echinococcosis (HE) is a widespread parasitic disease in underdeveloped pastoral areas with limited medical resources. While CNN-based and Transformer-based models have been widely applied to medical image segmentation, CNNs lack global context modeling due to local receptive fields, and Transformers, though capable of capturing long-range dependencies, are computationally expensive. Recently, state space models (SSMs), such as Mamba, have gained attention for their ability to model long sequences with linear complexity. In this paper, we propose EAGLE, a U-shaped network composed of a Progressive Visual State Space (PVSS) encoder and a Hybrid Visual State Space (HVSS) decoder that work collaboratively to achieve efficient and accurate segmentation of hepatic echinococcosis (HE) lesions. The proposed Convolutional Vision State Space Block (CVSSB) module is designed to fuse local and global features, while the Haar Wavelet Transformation Block (HWTB) module compresses spatial information into the channel dimension to enable lossless downsampling. Due to the lack of publicly available HE datasets, we collected CT slices from 260 patients at a local hospital. Experimental results show that EAGLE achieves state-of-the-art performance with a Dice Similarity Coefficient (DSC) of 89.76%, surpassing MSVM-UNet by 1.61%.
☆ From Codicology to Code: A Comparative Study of Transformer and YOLO-based Detectors for Layout Analysis in Historical Documents
Robust Document Layout Analysis (DLA) is critical for the automated processing and understanding of historical documents with complex page organizations. This paper benchmarks five state-of-the-art object detection architectures on three annotated datasets representing a spectrum of codicological complexity: The e-NDP, a corpus of Parisian medieval registers (1326-1504); CATMuS, a diverse multiclass dataset derived from various medieval and modern sources (ca.12th-17th centuries) and HORAE, a corpus of decorated books of hours (ca.13th-16th centuries). We evaluate two Transformer-based models (Co-DETR, Grounding DINO) against three YOLO variants (AABB, OBB, and YOLO-World). Our findings reveal significant performance variations dependent on model architecture, data set characteristics, and bounding box representation. In the e-NDP dataset, Co-DETR achieves state-of-the-art results (0.752 mAP@.50:.95), closely followed by YOLOv11X-OBB (0.721). Conversely, on the more complex CATMuS and HORAE datasets, the CNN-based YOLOv11x-OBB significantly outperforms all other models (0.564 and 0.568, respectively). This study unequivocally demonstrates that using Oriented Bounding Boxes (OBB) is not a minor refinement but a fundamental requirement for accurately modeling the non-Cartesian nature of historical manuscripts. We conclude that a key trade-off exists between the global context awareness of Transformers, ideal for structured layouts, and the superior generalization of CNN-OBB models for visually diverse and complex documents.
☆ On the Burstiness of Faces in Set
Burstiness, a phenomenon observed in text and image retrieval, refers to that particular elements appear more times in a set than a statistically independent model assumes. We argue that in the context of set-based face recognition (SFR), burstiness exists widely and degrades the performance in two aspects: Firstly, the bursty faces, where faces with particular attributes %exist frequently in a face set, dominate the training instances and dominate the training face sets and lead to poor generalization ability to unconstrained scenarios. Secondly, the bursty faces %dominating the evaluation sets interfere with the similarity comparison in set verification and identification when evaluation. To detect the bursty faces in a set, we propose three strategies based on Quickshift++, feature self-similarity, and generalized max-pooling (GMP). We apply the burst detection results on training and evaluation stages to enhance the sampling ratios or contributions of the infrequent faces. When evaluation, we additionally propose the quality-aware GMP that enables awareness of the face quality and robustness to the low-quality faces for the original GMP. We give illustrations and extensive experiments on the SFR benchmarks to demonstrate that burstiness is widespread and suppressing burstiness considerably improves the recognition performance.
comment: 18 pages, 5 figures
☆ Radiomic fingerprints for knee MR images assessment
Accurate interpretation of knee MRI scans relies on expert clinical judgment, often with high variability and limited scalability. Existing radiomic approaches use a fixed set of radiomic features (the signature), selected at the population level and applied uniformly to all patients. While interpretable, these signatures are often too constrained to represent individual pathological variations. As a result, conventional radiomic-based approaches are found to be limited in performance, compared with recent end-to-end deep learning (DL) alternatives without using interpretable radiomic features. We argue that the individual-agnostic nature in current radiomic selection is not central to its intepretability, but is responsible for the poor generalization in our application. Here, we propose a novel radiomic fingerprint framework, in which a radiomic feature set (the fingerprint) is dynamically constructed for each patient, selected by a DL model. Unlike the existing radiomic signatures, our fingerprints are derived on a per-patient basis by predicting the feature relevance in a large radiomic feature pool, and selecting only those that are predictive of clinical conditions for individual patients. The radiomic-selecting model is trained simultaneously with a low-dimensional (considered relatively explainable) logistic regression for downstream classification. We validate our methods across multiple diagnostic tasks including general knee abnormalities, anterior cruciate ligament (ACL) tears, and meniscus tears, demonstrating comparable or superior diagnostic accuracy relative to state-of-the-art end-to-end DL models. More importantly, we show that the interpretability inherent in our approach facilitates meaningful clinical insights and potential biomarker discovery, with detailed discussion, quantitative and qualitative analysis of real-world clinical cases to evidence these advantages.
☆ Learning Moderately Input-Sensitive Functions: A Case Study in QR Code Decoding
The hardness of learning a function that attains a target task relates to its input-sensitivity. For example, image classification tasks are input-insensitive as minor corruptions should not affect the classification results, whereas arithmetic and symbolic computation, which have been recently attracting interest, are highly input-sensitive as each input variable connects to the computation results. This study presents the first learning-based Quick Response (QR) code decoding and investigates learning functions of medium sensitivity. Our experiments reveal that Transformers can successfully decode QR codes, even beyond the theoretical error-correction limit, by learning the structure of embedded texts. They generalize from English-rich training data to other languages and even random strings. Moreover, we observe that the Transformer-based QR decoder focuses on data bits while ignoring error-correction bits, suggesting a decoding mechanism distinct from standard QR code readers.
comment: 17 pages, 13 figures
☆ FundaQ-8: A Clinically-Inspired Scoring Framework for Automated Fundus Image Quality Assessment
Automated fundus image quality assessment (FIQA) remains a challenge due to variations in image acquisition and subjective expert evaluations. We introduce FundaQ-8, a novel expert-validated framework for systematically assessing fundus image quality using eight critical parameters, including field coverage, anatomical visibility, illumination, and image artifacts. Using FundaQ-8 as a structured scoring reference, we develop a ResNet18-based regression model to predict continuous quality scores in the 0 to 1 range. The model is trained on 1800 fundus images from real-world clinical sources and Kaggle datasets, using transfer learning, mean squared error optimization, and standardized preprocessing. Validation against the EyeQ dataset and statistical analyses confirm the framework's reliability and clinical interpretability. Incorporating FundaQ-8 into deep learning models for diabetic retinopathy grading also improves diagnostic robustness, highlighting the value of quality-aware training in real-world screening applications.
☆ TDiR: Transformer based Diffusion for Image Restoration Tasks
Images captured in challenging environments often experience various forms of degradation, including noise, color cast, blur, and light scattering. These effects significantly reduce image quality, hindering their applicability in downstream tasks such as object detection, mapping, and classification. Our transformer-based diffusion model was developed to address image restoration tasks, aiming to improve the quality of degraded images. This model was evaluated against existing deep learning methodologies across multiple quality metrics for underwater image enhancement, denoising, and deraining on publicly available datasets. Our findings demonstrate that the diffusion model, combined with transformers, surpasses current methods in performance. The results of our model highlight the efficacy of diffusion models and transformers in improving the quality of degraded images, consequently expanding their utility in downstream tasks that require high-fidelity visual data.
☆ Ctrl-Z Sampling: Diffusion Sampling with Controlled Random Zigzag Explorations
Diffusion models have shown strong performance in conditional generation by progressively denoising Gaussian noise toward a target data distribution. This denoising process can be interpreted as a form of hill climbing in a learned latent space, where the model iteratively refines the sample toward regions of higher probability. However, diffusion models often converge to local optima that are locally visually coherent yet globally inconsistent or conditionally misaligned, due to latent space complexity and suboptimal initialization. Prior efforts attempted to address this by strengthening guidance signals or manipulating the initial noise distribution. We introduce Controlled Random Zigzag Sampling (Ctrl-Z Sampling), a novel sampling strategy designed to detect and escape such local maxima during conditional generation. The method first identifies potential local maxima using a reward model. Upon detection, it injects noise and reverts to a previous, noisier state to escape the current optimization plateau. The reward model then evaluates candidate trajectories, accepting only those that offer improvement, while progressively deeper retreat enables stronger escapes when nearby alternatives fail. This controlled random zigzag process allows dynamic alternation between forward refinement and backward exploration, enhancing both alignment and visual quality in the generated outputs. The proposed Ctrl-Z Sampling is model-agnostic and compatible with existing diffusion frameworks. Experimental results show that Ctrl-Z Sampling substantially improves generation quality with only around 7.6X increase in function evaluations.
comment: 10 pages, 3 figures, 2 tables
☆ Breaking Spatial Boundaries: Spectral-Domain Registration Guided Hyperspectral and Multispectral Blind Fusion
The blind fusion of unregistered hyperspectral images (HSIs) and multispectral images (MSIs) has attracted growing attention recently. To address the registration challenge, most existing methods employ spatial transformations on the HSI to achieve alignment with the MSI. However, due to the substantial differences in spatial resolution of the images, the performance of these methods is often unsatisfactory. Moreover, the registration process tends to be time-consuming when dealing with large-sized images in remote sensing. To address these issues, we propose tackling the registration problem from the spectral domain. Initially, a lightweight Spectral Prior Learning (SPL) network is developed to extract spectral features from the HSI and enhance the spectral resolution of the MSI. Following this, the obtained image undergoes spatial downsampling to produce the registered HSI. In this process, subspace representation and cyclic training strategy are employed to improve spectral accuracy of the registered HSI obtained. Next, we propose a blind sparse fusion (BSF) method, which utilizes group sparsity regularization to equivalently promote the low-rankness of the image. This approach not only circumvents the need for rank estimation, but also reduces computational complexity. Then, we employ the Proximal Alternating Optimization (PAO) algorithm to solve the BSF model, and present its convergence analysis. Finally, extensive numerical experiments on simulated and real datasets are conducted to verify the effectiveness of our method in registration and fusion. We also demonstrate its efficacy in enhancing classification performance.
☆ Opportunistic Osteoporosis Diagnosis via Texture-Preserving Self-Supervision, Mixture of Experts and Multi-Task Integration MICCAI 2025
Osteoporosis, characterized by reduced bone mineral density (BMD) and compromised bone microstructure, increases fracture risk in aging populations. While dual-energy X-ray absorptiometry (DXA) is the clinical standard for BMD assessment, its limited accessibility hinders diagnosis in resource-limited regions. Opportunistic computed tomography (CT) analysis has emerged as a promising alternative for osteoporosis diagnosis using existing imaging data. Current approaches, however, face three limitations: (1) underutilization of unlabeled vertebral data, (2) systematic bias from device-specific DXA discrepancies, and (3) insufficient integration of clinical knowledge such as spatial BMD distribution patterns. To address these, we propose a unified deep learning framework with three innovations. First, a self-supervised learning method using radiomic representations to leverage unlabeled CT data and preserve bone texture. Second, a Mixture of Experts (MoE) architecture with learned gating mechanisms to enhance cross-device adaptability. Third, a multi-task learning framework integrating osteoporosis diagnosis, BMD regression, and vertebra location prediction. Validated across three clinical sites and an external hospital, our approach demonstrates superior generalizability and accuracy over existing methods for opportunistic osteoporosis screening and diagnosis.
comment: Accepted by MICCAI 2025
☆ From Ideal to Real: Unified and Data-Efficient Dense Prediction for Real-World Scenarios
Dense prediction tasks hold significant importance of computer vision, aiming to learn pixel-wise annotated label for an input image. Despite advances in this field, existing methods primarily focus on idealized conditions, with limited generalization to real-world scenarios and facing the challenging scarcity of real-world data. To systematically study this problem, we first introduce DenseWorld, a benchmark spanning a broad set of 25 dense prediction tasks that correspond to urgent real-world applications, featuring unified evaluation across tasks. Then, we propose DenseDiT, which maximally exploits generative models' visual priors to perform diverse real-world dense prediction tasks through a unified strategy. DenseDiT combines a parameter-reuse mechanism and two lightweight branches that adaptively integrate multi-scale context, working with less than 0.1% additional parameters. Evaluations on DenseWorld reveal significant performance drops in existing general and specialized baselines, highlighting their limited real-world generalization. In contrast, DenseDiT achieves superior results using less than 0.01% training data of baselines, underscoring its practical value for real-world deployment. Our data, and checkpoints and codes are available at https://xcltql666.github.io/DenseDiTProj
☆ Forensic Study of Paintings Through the Comparison of Fabrics
The study of canvas fabrics in works of art is a crucial tool for authentication, attribution and conservation. Traditional methods are based on thread density map matching, which cannot be applied when canvases do not come from contiguous positions on a roll. This paper presents a novel approach based on deep learning to assess the similarity of textiles. We introduce an automatic tool that evaluates the similarity between canvases without relying on thread density maps. A Siamese deep learning model is designed and trained to compare pairs of images by exploiting the feature representations learned from the scans. In addition, a similarity estimation method is proposed, aggregating predictions from multiple pairs of cloth samples to provide a robust similarity score. Our approach is applied to canvases from the Museo Nacional del Prado, corroborating the hypothesis that plain weave canvases, widely used in painting, can be effectively compared even when their thread densities are similar. The results demonstrate the feasibility and accuracy of the proposed method, opening new avenues for the analysis of masterpieces.
☆ X-SiT: Inherently Interpretable Surface Vision Transformers for Dementia Diagnosis MICCAI 2025
Interpretable models are crucial for supporting clinical decision-making, driving advances in their development and application for medical images. However, the nature of 3D volumetric data makes it inherently challenging to visualize and interpret intricate and complex structures like the cerebral cortex. Cortical surface renderings, on the other hand, provide a more accessible and understandable 3D representation of brain anatomy, facilitating visualization and interactive exploration. Motivated by this advantage and the widespread use of surface data for studying neurological disorders, we present the eXplainable Surface Vision Transformer (X-SiT). This is the first inherently interpretable neural network that offers human-understandable predictions based on interpretable cortical features. As part of X-SiT, we introduce a prototypical surface patch decoder for classifying surface patch embeddings, incorporating case-based reasoning with spatially corresponding cortical prototypes. The results demonstrate state-of-the-art performance in detecting Alzheimer's disease and frontotemporal dementia while additionally providing informative prototypes that align with known disease patterns and reveal classification errors.
comment: MICCAI 2025
☆ Hierarchical Mask-Enhanced Dual Reconstruction Network for Few-Shot Fine-Grained Image Classification
Few-shot fine-grained image classification (FS-FGIC) presents a significant challenge, requiring models to distinguish visually similar subclasses with limited labeled examples. Existing methods have critical limitations: metric-based methods lose spatial information and misalign local features, while reconstruction-based methods fail to utilize hierarchical feature information and lack mechanisms to focus on discriminative regions. We propose the Hierarchical Mask-enhanced Dual Reconstruction Network (HMDRN), which integrates dual-layer feature reconstruction with mask-enhanced feature processing to improve fine-grained classification. HMDRN incorporates a dual-layer feature reconstruction and fusion module that leverages complementary visual information from different network hierarchies. Through learnable fusion weights, the model balances high-level semantic representations from the last layer with mid-level structural details from the penultimate layer. Additionally, we design a spatial binary mask-enhanced transformer self-reconstruction module that processes query features through adaptive thresholding while maintaining complete support features, enhancing focus on discriminative regions while filtering background noise. Extensive experiments on three challenging fine-grained datasets demonstrate that HMDRN consistently outperforms state-of-the-art methods across Conv-4 and ResNet-12 backbone architectures. Comprehensive ablation studies validate the effectiveness of each proposed component, revealing that dual-layer reconstruction enhances inter-class discrimination while mask-enhanced transformation reduces intra-class variations. Visualization results provide evidence of HMDRN's superior feature reconstruction capabilities.
☆ A Transformer Based Handwriting Recognition System Jointly Using Online and Offline Features
We posit that handwriting recognition benefits from complementary cues carried by the rasterized complex glyph and the pen's trajectory, yet most systems exploit only one modality. We introduce an end-to-end network that performs early fusion of offline images and online stroke data within a shared latent space. A patch encoder converts the grayscale crop into fixed-length visual tokens, while a lightweight transformer embeds the $(x, y, \text{pen})$ sequence. Learnable latent queries attend jointly to both token streams, yielding context-enhanced stroke embeddings that are pooled and decoded under a cross-entropy loss objective. Because integration occurs before any high-level classification, temporal cues reinforce each other during representation learning, producing stronger writer independence. Comprehensive experiments on IAMOn-DB and VNOn-DB demonstrate that our approach achieves state-of-the-art accuracy, exceeding previous bests by up to 1\%. Our study also shows adaptation of this pipeline with gesturification on the ISI-Air dataset. Our code can be found here.
comment: 15 pages, 7 figures
☆ Recognizing Surgical Phases Anywhere: Few-Shot Test-time Adaptation and Task-graph Guided Refinement MICCAI 2025
The complexity and diversity of surgical workflows, driven by heterogeneous operating room settings, institutional protocols, and anatomical variability, present a significant challenge in developing generalizable models for cross-institutional and cross-procedural surgical understanding. While recent surgical foundation models pretrained on large-scale vision-language data offer promising transferability, their zero-shot performance remains constrained by domain shifts, limiting their utility in unseen surgical environments. To address this, we introduce Surgical Phase Anywhere (SPA), a lightweight framework for versatile surgical workflow understanding that adapts foundation models to institutional settings with minimal annotation. SPA leverages few-shot spatial adaptation to align multi-modal embeddings with institution-specific surgical scenes and phases. It also ensures temporal consistency through diffusion modeling, which encodes task-graph priors derived from institutional procedure protocols. Finally, SPA employs dynamic test-time adaptation, exploiting the mutual agreement between multi-modal phase prediction streams to adapt the model to a given test video in a self-supervised manner, enhancing the reliability under test-time distribution shifts. SPA is a lightweight adaptation framework, allowing hospitals to rapidly customize phase recognition models by defining phases in natural language text, annotating a few images with the phase labels, and providing a task graph defining phase transitions. The experimental results show that the SPA framework achieves state-of-the-art performance in few-shot surgical phase recognition across multiple institutions and procedures, even outperforming full-shot models with 32-shot labeled data. Code is available at https://github.com/CAMMA-public/SPA
comment: Accepted by MICCAI 2025
☆ FedBKD: Distilled Federated Learning to Embrace Gerneralization and Personalization on Non-IID Data
Federated learning (FL) is a decentralized collaborative machine learning (ML) technique. It provides a solution to the issues of isolated data islands and data privacy leakage in industrial ML practices. One major challenge in FL is handling the non-identical and independent distributed (non-IID) data. Current solutions either focus on constructing an all-powerful global model, or customizing personalized local models. Few of them can provide both a well-generalized global model and well-performed local models at the same time. Additionally, many FL solutions to the non-IID problem are benefited from introducing public datasets. However, this will also increase the risk of data leakage. To tackle the problems, we propose a novel data-free distillation framework, Federated Bidirectional Knowledge Distillation (FedBKD). Specifically, we train Generative Adversarial Networks (GAN) for synthetic data. During the GAN training, local models serve as discriminators and their parameters are frozen. The synthetic data is then used for bidirectional distillation between global and local models to achieve knowledge interactions so that performances for both sides are improved. We conduct extensive experiments on 4 benchmarks under different non-IID settings. The results show that FedBKD achieves SOTA performances in every case.
☆ Dynamic Bandwidth Allocation for Hybrid Event-RGB Transmission
Event cameras asynchronously capture pixel-level intensity changes with extremely low latency. They are increasingly used in conjunction with RGB cameras for a wide range of vision-related applications. However, a major challenge in these hybrid systems lies in the transmission of the large volume of triggered events and RGB images. To address this, we propose a transmission scheme that retains efficient reconstruction performance of both sources while accomplishing real-time deblurring in parallel. Conventional RGB cameras and event cameras typically capture the same scene in different ways, often resulting in significant redundant information across their outputs. To address this, we develop a joint event and image (E-I) transmission framework to eliminate redundancy and thereby optimize channel bandwidth utilization. Our approach employs Bayesian modeling and the information bottleneck method to disentangle the shared and domain-specific information within the E-I inputs. This disentangled information bottleneck framework ensures both the compactness and informativeness of extracted shared and domain-specific information. Moreover, it adaptively allocates transmission bandwidth based on scene dynamics, i.e., more symbols are allocated to events for dynamic details or to images for static information. Simulation results demonstrate that the proposed scheme not only achieves superior reconstruction quality compared to conventional systems but also delivers enhanced deblurring performance.
☆ UniCode$^2$: Cascaded Large-scale Codebooks for Unified Multimodal Understanding and Generation
Unified multimodal large language models (MLLMs) have shown promise in jointly advancing multimodal understanding and generation, with visual codebooks discretizing images into tokens for autoregressive modeling. Existing codebook-based methods either rely on small vocabularies (~16K entries) that lack fine-grained semantics or naively scale up, resulting in low token utilization and unstable training. We propose UniCode$^2$, a cascaded codebook framework enabling large-scale, semantically aligned, and stable visual tokenization. By clustering millions of SigLIP sequence embeddings, we build a 500K-entry codebook that preserves vision-language alignment while expanding capacity. Stability is ensured via a cascaded design: a frozen codebook anchors the embedding space, and a trainable codebook refines task-specific semantics. This decoupling promotes high utilization and robust learning. Moreover, the alignment of our visual tokens with textual semantics enables seamless integration with pretrained diffusion decoders, supporting high-quality visual synthesis with minimal adaptation. UniCode^2 delivers strong performance across diverse benchmarks, demonstrating the viability of scaling visual token spaces without sacrificing stability, semantics, or modularity.
comment: 19 pages, 5 figures
☆ MS-IQA: A Multi-Scale Feature Fusion Network for PET/CT Image Quality Assessment MICCAI 2025
Positron Emission Tomography / Computed Tomography (PET/CT) plays a critical role in medical imaging, combining functional and anatomical information to aid in accurate diagnosis. However, image quality degradation due to noise, compression and other factors could potentially lead to diagnostic uncertainty and increase the risk of misdiagnosis. When evaluating the quality of a PET/CT image, both low-level features like distortions and high-level features like organ anatomical structures affect the diagnostic value of the image. However, existing medical image quality assessment (IQA) methods are unable to account for both feature types simultaneously. In this work, we propose MS-IQA, a novel multi-scale feature fusion network for PET/CT IQA, which utilizes multi-scale features from various intermediate layers of ResNet and Swin Transformer, enhancing its ability of perceiving both local and global information. In addition, a multi-scale feature fusion module is also introduced to effectively combine high-level and low-level information through a dynamically weighted channel attention mechanism. Finally, to fill the blank of PET/CT IQA dataset, we construct PET-CT-IQA-DS, a dataset containing 2,700 varying-quality PET/CT images with quality scores assigned by radiologists. Experiments on our dataset and the publicly available LDCTIQAC2023 dataset demonstrate that our proposed model has achieved superior performance against existing state-of-the-art methods in various IQA metrics. This work provides an accurate and efficient IQA method for PET/CT. Our code and dataset are available at https://github.com/MS-IQA/MS-IQA/.
comment: Accepted to MICCAI 2025
☆ Progressive Alignment Degradation Learning for Pansharpening
Deep learning-based pansharpening has been shown to effectively generate high-resolution multispectral (HRMS) images. To create supervised ground-truth HRMS images, synthetic data generated using the Wald protocol is commonly employed. This protocol assumes that networks trained on artificial low-resolution data will perform equally well on high-resolution data. However, well-trained models typically exhibit a trade-off in performance between reduced-resolution and full-resolution datasets. In this paper, we delve into the Wald protocol and find that its inaccurate approximation of real-world degradation patterns limits the generalization of deep pansharpening models. To address this issue, we propose the Progressive Alignment Degradation Module (PADM), which uses mutual iteration between two sub-networks, PAlignNet and PDegradeNet, to adaptively learn accurate degradation processes without relying on predefined operators. Building on this, we introduce HFreqdiff, which embeds high-frequency details into a diffusion framework and incorporates CFB and BACM modules for frequency-selective detail extraction and precise reverse process learning. These innovations enable effective integration of high-resolution panchromatic and multispectral images, significantly enhancing spatial sharpness and quality. Experiments and ablation studies demonstrate the proposed method's superior performance compared to state-of-the-art techniques.
comment: 13 pages, 9 figures
☆ Towards Scalable and Generalizable Earth Observation Data Mining via Foundation Model Composition
Foundation models are rapidly transforming Earth Observation data mining by enabling generalizable and scalable solutions for key tasks such as scene classification and semantic segmentation. While most efforts in the geospatial domain have focused on developing large models trained from scratch using massive Earth Observation datasets, an alternative strategy that remains underexplored is the reuse and combination of existing pretrained models. In this study, we investigate whether foundation models pretrained on remote sensing and general vision datasets can be effectively combined to improve performance across a diverse set of key Earth Observation tasks. Using the GEO-Bench benchmark, we evaluate several prominent models, including Prithvi, Hiera, and DOFA, on eleven datasets covering a range of spatial resolutions, sensor modalities, and task types. The results show that feature-level ensembling of smaller pretrained models can match or exceed the performance of much larger models, while requiring less training time and computational resources. Moreover, the study highlights the potential of applying knowledge distillation to transfer the strengths of ensembles into more compact models, offering a practical path for deploying foundation models in real-world Earth Observation applications.
☆ Seeing is Believing? Mitigating OCR Hallucinations in Multimodal Large Language Models
Recent advancements in multimodal large language models have enhanced document understanding by integrating textual and visual information. However, existing models exhibit incompleteness within their paradigm in real-world scenarios, particularly under visual degradation. In such conditions, the current response paradigm often fails to adequately perceive visual degradation and ambiguity, leading to overreliance on linguistic priors or misaligned visual-textual reasoning. This difficulty in recognizing uncertainty frequently results in the generation of hallucinatory content, especially when a precise answer is not feasible. To better demonstrate and analyze this phenomenon and problem, we propose KIE-HVQA, the first benchmark dedicated to evaluating OCR hallucination in degraded document understanding. This dataset includes test samples spanning identity cards and invoices, with simulated real-world degradations for OCR reliability. This setup allows for evaluating models' capacity, under degraded input, to distinguish reliable visual information and answer accordingly, thereby highlighting the challenge of avoiding hallucination on uncertain data. To achieve vision-faithful reasoning and thereby avoid the aforementioned issues, we further introduce a GRPO-based framework featuring a novel reward mechanism. By incorporating a self-awareness of visual uncertainty and an analysis method that initiates refusal to answer to increase task difficulty within our supervised fine-tuning and reinforcement learning framework, we successfully mitigated hallucinations in ambiguous regions. Experiments on Qwen2.5-VL demonstrate that our 7B-parameter model achieves a 22\% absolute improvement in hallucination-free accuracy over GPT-4o on KIE-HVQA and there is no significant performance drop in standard tasks, highlighting both effectiveness and robustness.
☆ Towards Efficient Exemplar Based Image Editing with Multimodal VLMs ECCV 2024
Text-to-Image Diffusion models have enabled a wide array of image editing applications. However, capturing all types of edits through text alone can be challenging and cumbersome. The ambiguous nature of certain image edits is better expressed through an exemplar pair, i.e., a pair of images depicting an image before and after an edit respectively. In this work, we tackle exemplar-based image editing -- the task of transferring an edit from an exemplar pair to a content image(s), by leveraging pretrained text-to-image diffusion models and multimodal VLMs. Even though our end-to-end pipeline is optimization-free, our experiments demonstrate that it still outperforms baselines on multiple types of edits while being ~4x faster.
comment: Accepted at ECCV 2024 (AI4VA Workshop)
☆ Loss-Aware Automatic Selection of Structured Pruning Criteria for Deep Neural Network Acceleration
Structured pruning is a well-established technique for compressing neural networks, making it suitable for deployment in resource-limited edge devices. This paper presents an efficient Loss-Aware Automatic Selection of Structured Pruning Criteria (LAASP) for slimming and accelerating deep neural networks. The majority of pruning methodologies employ a sequential process consisting of three stages: 1) training, 2) pruning, and 3) fine-tuning, whereas the proposed pruning technique adopts a pruning-while-training approach that eliminates the first stage and integrates the second and third stages into a single cycle. The automatic selection of magnitude or similarity-based filter pruning criteria from a specified pool of criteria and the specific pruning layer at each pruning iteration is guided by the network's overall loss on a small subset of the training data. To mitigate the abrupt accuracy drop due to pruning, the network is retrained briefly after each reduction of a predefined number of floating-point operations (FLOPs). The optimal pruning rates for each layer in the network are automatically determined, eliminating the need for manual allocation of fixed or variable pruning rates for each layer. Experiments on the VGGNet and ResNet models on the CIFAR-10 and ImageNet benchmark datasets demonstrate the effectiveness of the proposed method. In particular, the ResNet56 and ResNet110 models on the CIFAR-10 dataset significantly improve the top-1 accuracy compared to state-of-the-art methods while reducing the network FLOPs by 52\%. Furthermore, the ResNet50 model on the ImageNet dataset reduces FLOPs by more than 42\% with a negligible 0.33\% drop in top-5 accuracy. The source code of this paper is publicly available online - https://github.com/ghimiredhikura/laasp.
☆ EAR: Erasing Concepts from Unified Autoregressive Models
Autoregressive (AR) models have achieved unified and strong performance across both visual understanding and image generation tasks. However, removing undesired concepts from AR models while maintaining overall generation quality remains an open challenge. In this paper, we propose Erasure Autoregressive Model (EAR), a fine-tuning method for effective and utility-preserving concept erasure in AR models. Specifically, we introduce Windowed Gradient Accumulation (WGA) strategy to align patch-level decoding with erasure objectives, and Thresholded Loss Masking (TLM) strategy to protect content unrelated to the target concept during fine-tuning. Furthermore, we propose a novel benchmark, Erase Concept Generator and Visual Filter (ECGVF), aim at provide a more rigorous and comprehensive foundation for evaluating concept erasure in AR models. Specifically, we first employ structured templates across diverse large language models (LLMs) to pre-generate a large-scale corpus of target-replacement concept prompt pairs. Subsequently, we generate images from these prompts and subject them to rigorous filtering via a visual classifier to ensure concept fidelity and alignment. Extensive experimental results conducted on the ECGVF benchmark with the AR model Janus-Pro demonstrate that EAR achieves marked improvements in both erasure effectiveness and model utility preservation. Code is available at: https://github.com/immc-lab/ear/
comment: 11 pages, 7 figures, 1 tables
☆ From 2D to 3D Cognition: A Brief Survey of General World Models
World models have garnered increasing attention in the development of artificial general intelligence (AGI), serving as computational frameworks for learning representations of the external world and forecasting future states. While early efforts focused on 2D visual perception and simulation, recent 3D-aware generative world models have demonstrated the ability to synthesize geometrically consistent, interactive 3D environments, marking a shift toward 3D spatial cognition. Despite rapid progress, the field lacks systematic analysis to categorize emerging techniques and clarify their roles in advancing 3D cognitive world models. This survey addresses this need by introducing a conceptual framework, providing a structured and forward-looking review of world models transitioning from 2D perception to 3D cognition. Within this framework, we highlight two key technological drivers, particularly advances in 3D representations and the incorporation of world knowledge, as fundamental pillars. Building on these, we dissect three core cognitive capabilities that underpin 3D world modeling: 3D physical scene generation, 3D spatial reasoning, and 3D spatial interaction. We further examine the deployment of these capabilities in real-world applications, including embodied AI, autonomous driving, digital twin, and gaming/VR. Finally, we identify challenges across data, modeling, and deployment, and outline future directions for advancing more robust and generalizable 3D world models.
☆ BrokenVideos: A Benchmark Dataset for Fine-Grained Artifact Localization in AI-Generated Videos
Recent advances in deep generative models have led to significant progress in video generation, yet the fidelity of AI-generated videos remains limited. Synthesized content often exhibits visual artifacts such as temporally inconsistent motion, physically implausible trajectories, unnatural object deformations, and local blurring that undermine realism and user trust. Accurate detection and spatial localization of these artifacts are crucial for both automated quality control and for guiding the development of improved generative models. However, the research community currently lacks a comprehensive benchmark specifically designed for artifact localization in AI generated videos. Existing datasets either restrict themselves to video or frame level detection or lack the fine-grained spatial annotations necessary for evaluating localization methods. To address this gap, we introduce BrokenVideos, a benchmark dataset of 3,254 AI-generated videos with meticulously annotated, pixel-level masks highlighting regions of visual corruption. Each annotation is validated through detailed human inspection to ensure high quality ground truth. Our experiments show that training state of the art artifact detection models and multi modal large language models (MLLMs) on BrokenVideos significantly improves their ability to localize corrupted regions. Through extensive evaluation, we demonstrate that BrokenVideos establishes a critical foundation for benchmarking and advancing research on artifact localization in generative video models. The dataset is available at: https://broken-video-detection-datetsets.github.io/Broken-Video-Detection-Datasets.github.io/.
comment: 7 page,4 figures,2 tables
☆ MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations
We introduce MIRAGE, a new benchmark for multimodal expert-level reasoning and decision-making in consultative interaction settings. Designed for the agriculture domain, MIRAGE captures the full complexity of expert consultations by combining natural user queries, expert-authored responses, and image-based context, offering a high-fidelity benchmark for evaluating models on grounded reasoning, clarification strategies, and long-form generation in a real-world, knowledge-intensive domain. Grounded in over 35,000 real user-expert interactions and curated through a carefully designed multi-step pipeline, MIRAGE spans diverse crop health, pest diagnosis, and crop management scenarios. The benchmark includes more than 7,000 unique biological entities, covering plant species, pests, and diseases, making it one of the most taxonomically diverse benchmarks available for vision-language models, grounded in the real world. Unlike existing benchmarks that rely on well-specified user inputs and closed-set taxonomies, MIRAGE features underspecified, context-rich scenarios with open-world settings, requiring models to infer latent knowledge gaps, handle rare entities, and either proactively guide the interaction or respond. Project Page: https://mirage-benchmark.github.io
comment: 66 pages, 32 figures, 23 tables
♻ ☆ OmniGen2: Exploration to Advanced Multimodal Generation
In this work, we introduce OmniGen2, a versatile and open-source generative model designed to provide a unified solution for diverse generation tasks, including text-to-image, image editing, and in-context generation. Unlike OmniGen v1, OmniGen2 features two distinct decoding pathways for text and image modalities, utilizing unshared parameters and a decoupled image tokenizer. This design enables OmniGen2 to build upon existing multimodal understanding models without the need to re-adapt VAE inputs, thereby preserving the original text generation capabilities. To facilitate the training of OmniGen2, we developed comprehensive data construction pipelines, encompassing image editing and in-context generation data. Additionally, we introduce a reflection mechanism tailored for image generation tasks and curate a dedicated reflection dataset based on OmniGen2. Despite its relatively modest parameter size, OmniGen2 achieves competitive results on multiple task benchmarks, including text-to-image and image editing. To further evaluate in-context generation, also referred to as subject-driven tasks, we introduce a new benchmark named OmniContext. OmniGen2 achieves state-of-the-art performance among open-source models in terms of consistency. We will release our models, training code, datasets, and data construction pipeline to support future research in this field. Project Page: https://vectorspacelab.github.io/OmniGen2; GitHub Link: https://github.com/VectorSpaceLab/OmniGen2
♻ ☆ Diffusion Models Through a Global Lens: Are They Culturally Inclusive?
Text-to-image diffusion models have recently enabled the creation of visually compelling, detailed images from textual prompts. However, their ability to accurately represent various cultural nuances remains an open question. In our work, we introduce CultDiff benchmark, evaluating state-of-the-art diffusion models whether they can generate culturally specific images spanning ten countries. We show that these models often fail to generate cultural artifacts in architecture, clothing, and food, especially for underrepresented country regions, by conducting a fine-grained analysis of different similarity aspects, revealing significant disparities in cultural relevance, description fidelity, and realism compared to real-world reference images. With the collected human evaluations, we develop a neural-based image-image similarity metric, namely, CultDiff-S, to predict human judgment on real and generated images with cultural artifacts. Our work highlights the need for more inclusive generative AI systems and equitable dataset representation over a wide range of cultures.
comment: 17 pages, 17 figures, 3 tables
♻ ☆ From $\mathcal{O}(n^{2})$ to $\mathcal{O}(n)$ Parameters: Quantum Self-Attention in Vision Transformers for Biomedical Image Classification MICCAI 2025
We demonstrate that quantum vision transformers (QViTs), vision transformers (ViTs) with self-attention (SA) mechanisms replaced by quantum self-attention (QSA) mechanisms, can match state-of-the-art (SOTA) biomedical image classifiers while using 99.99% fewer parameters. QSAs are produced by replacing linear SA layers with parameterised quantum neural networks (QNNs), producing a QSA mechanism and reducing parameter scaling from $\mathcal{O}(n^2)$ to $\mathcal{O}(n)$. On RetinaMNIST, our ultra parameter-efficient QViT outperforms 13/14 SOTA methods including CNNs and ViTs, achieving 56.5% accuracy, just 0.88% below the top MedMamba model while using 99.99% fewer parameters (1K vs 14.5M) and 89% fewer GFLOPs. We present the first investigation of knowledge distillation (KD) from classical to quantum vision transformers in biomedical image classification, showing that QViTs maintain comparable performance to classical ViTs across eight diverse datasets spanning multiple modalities, with improved QSA parameter-efficiency. Our higher-qubit architecture benefitted more from KD pre-training, suggesting a scaling relationship between QSA parameters and KD effectiveness. These findings establish QSA as a practical architectural choice toward parameter-efficient biomedical image analysis.
comment: Submitted for EMA4MICCAI 2025
♻ ☆ Time-Aware Auto White Balance in Mobile Photography
Cameras rely on auto white balance (AWB) to correct undesirable color casts caused by scene illumination and the camera's spectral sensitivity. This is typically achieved using an illuminant estimator that determines the global color cast solely from the color information in the camera's raw sensor image. Mobile devices provide valuable additional metadata-such as capture timestamp and geolocation-that offers strong contextual clues to help narrow down the possible illumination solutions. This paper proposes a lightweight illuminant estimation method that incorporates such contextual metadata, along with additional capture information and image colors, into a compact model (~5K parameters), achieving promising results, matching or surpassing larger models. To validate our method, we introduce a dataset of 3,224 smartphone images with contextual metadata collected at various times of day and under diverse lighting conditions. The dataset includes ground-truth illuminant colors, determined using a color chart, and user-preferred illuminants validated through a user study, providing a comprehensive benchmark for AWB evaluation.
♻ ☆ FluoroSAM: A Language-promptable Foundation Model for Flexible X-ray Image Segmentation
Language promptable X-ray image segmentation would enable greater flexibility for human-in-the-loop workflows in diagnostic and interventional precision medicine. Prior efforts have contributed task-specific models capable of solving problems within a narrow scope, but expanding to broader use requires additional data, annotations, and training time. Recently, language-aligned foundation models (LFMs) -- machine learning models trained on large amounts of highly variable image and text data thus enabling broad applicability -- have emerged as promising tools for automated image analysis. Existing foundation models for medical image analysis focus on scenarios and modalities where large, richly annotated datasets are available. However, the X-ray imaging modality features highly variable image appearance and applications, from diagnostic chest X-rays to interventional fluoroscopy, with varying availability of data. To pave the way toward an LFM for comprehensive and language-aligned analysis of arbitrary medical X-ray images, we introduce FluoroSAM, a language-promptable variant of the Segment Anything Model, trained from scratch on 3M synthetic X-ray images from a wide variety of human anatomies, imaging geometries, and viewing angles. These include pseudo-ground truth masks for 128 organ types and 464 tools with associated text descriptions. FluoroSAM is capable of segmenting myriad anatomical structures and tools based on natural language prompts, thanks to the novel incorporation of vector quantization (VQ) of text embeddings in the training process. We demonstrate FluoroSAM's performance quantitatively on real X-ray images and showcase on several applications how FluoroSAM is a key enabler for rich human-machine interaction in the X-ray image acquisition and analysis context. Code is available at https://github.com/arcadelab/fluorosam.
♻ ☆ Dark Channel-Assisted Depth-from-Defocus from a Single Image
We estimate scene depth from a single defocus-blurred image using the dark channel as a complementary cue, leveraging its ability to capture local statistics and scene structure. Traditional depth-from-defocus (DFD) methods use multiple images with varying apertures or focus. Single-image DFD is underexplored due to its inherent challenges. Few attempts have focused on depth-from-defocus (DFD) from a single defocused image because the problem is underconstrained. Our method uses the relationship between local defocus blur and contrast variations as depth cues to improve scene structure estimation. The pipeline is trained end-to-end with adversarial learning. Experiments on real data demonstrate that incorporating the dark channel prior into single-image DFD provides meaningful depth estimation, validating our approach.
♻ ☆ Cross-Frame Representation Alignment for Fine-Tuning Video Diffusion Models
Fine-tuning Video Diffusion Models (VDMs) at the user level to generate videos that reflect specific attributes of training data presents notable challenges, yet remains underexplored despite its practical importance. Meanwhile, recent work such as Representation Alignment (REPA) has shown promise in improving the convergence and quality of DiT-based image diffusion models by aligning, or assimilating, its internal hidden states with external pretrained visual features, suggesting its potential for VDM fine-tuning. In this work, we first propose a straightforward adaptation of REPA for VDMs and empirically show that, while effective for convergence, it is suboptimal in preserving semantic consistency across frames. To address this limitation, we introduce Cross-frame Representation Alignment (CREPA), a novel regularization technique that aligns hidden states of a frame with external features from neighboring frames. Empirical evaluations on large-scale VDMs, including CogVideoX-5B and Hunyuan Video, demonstrate that CREPA improves both visual fidelity and cross-frame semantic coherence when fine-tuned with parameter-efficient methods such as LoRA. We further validate CREPA across diverse datasets with varying attributes, confirming its broad applicability.
comment: Project page: https://crepavideo.github.io
♻ ☆ PanoWan: Lifting Diffusion Video Generation Models to 360° with Latitude/Longitude-aware Mechanisms
Panoramic video generation enables immersive 360{\deg} content creation, valuable in applications that demand scene-consistent world exploration. However, existing panoramic video generation models struggle to leverage pre-trained generative priors from conventional text-to-video models for high-quality and diverse panoramic videos generation, due to limited dataset scale and the gap in spatial feature representations. In this paper, we introduce PanoWan to effectively lift pre-trained text-to-video models to the panoramic domain, equipped with minimal modules. PanoWan employs latitude-aware sampling to avoid latitudinal distortion, while its rotated semantic denoising and padded pixel-wise decoding ensure seamless transitions at longitude boundaries. To provide sufficient panoramic videos for learning these lifted representations, we contribute PanoVid, a high-quality panoramic video dataset with captions and diverse scenarios. Consequently, PanoWan achieves state-of-the-art performance in panoramic video generation and demonstrates robustness for zero-shot downstream tasks. Our project page is available at https://panowan.variantconst.com.
♻ ☆ ViStoryBench: Comprehensive Benchmark Suite for Story Visualization
Story visualization, which aims to generate a sequence of visually coherent images aligning with a given narrative and reference images, has seen significant progress with recent advancements in generative models. To further enhance the performance of story visualization frameworks in real-world scenarios, we introduce a comprehensive evaluation benchmark, ViStoryBench. We collect a diverse dataset encompassing various story types and artistic styles, ensuring models are evaluated across multiple dimensions such as different plots (e.g., comedy, horror) and visual aesthetics (e.g., anime, 3D renderings). ViStoryBench is carefully curated to balance narrative structures and visual elements, featuring stories with single and multiple protagonists to test models' ability to maintain character consistency. Additionally, it includes complex plots and intricate world-building to challenge models in generating accurate visuals. To ensure comprehensive comparisons, our benchmark incorporates a wide range of evaluation metrics assessing critical aspects. This structured and multifaceted framework enables researchers to thoroughly identify both the strengths and weaknesses of different models, fostering targeted improvements.
comment: 33 Pages, Project Page: https://vistorybench.github.io/, Code: https://github.com/vistorybench/vistorybench
♻ ☆ LPOSS: Label Propagation Over Patches and Pixels for Open-vocabulary Semantic Segmentation
We propose a training-free method for open-vocabulary semantic segmentation using Vision-and-Language Models (VLMs). Our approach enhances the initial per-patch predictions of VLMs through label propagation, which jointly optimizes predictions by incorporating patch-to-patch relationships. Since VLMs are primarily optimized for cross-modal alignment and not for intra-modal similarity, we use a Vision Model (VM) that is observed to better capture these relationships. We address resolution limitations inherent to patch-based encoders by applying label propagation at the pixel level as a refinement step, significantly improving segmentation accuracy near class boundaries. Our method, called LPOSS+, performs inference over the entire image, avoiding window-based processing and thereby capturing contextual interactions across the full image. LPOSS+ achieves state-of-the-art performance among training-free methods, across a diverse set of datasets. Code: https://github.com/vladan-stojnic/LPOSS
♻ ☆ MatSwap: Light-aware material transfers in images
We present MatSwap, a method to transfer materials to designated surfaces in an image photorealistically. Such a task is non-trivial due to the large entanglement of material appearance, geometry, and lighting in a photograph. In the literature, material editing methods typically rely on either cumbersome text engineering or extensive manual annotations requiring artist knowledge and 3D scene properties that are impractical to obtain. In contrast, we propose to directly learn the relationship between the input material -- as observed on a flat surface -- and its appearance within the scene, without the need for explicit UV mapping. To achieve this, we rely on a custom light- and geometry-aware diffusion model. We fine-tune a large-scale pre-trained text-to-image model for material transfer using our synthetic dataset, preserving its strong priors to ensure effective generalization to real images. As a result, our method seamlessly integrates a desired material into the target location in the photograph while retaining the identity of the scene. We evaluate our method on synthetic and real images and show that it compares favorably to recent work both qualitatively and quantitatively. We release our code and data on https://github.com/astra-vision/MatSwap
comment: Accepted to EGSR, journal track to appear in Computer Graphics Forum
♻ ☆ MagicPose4D: Crafting Articulated Models with Appearance and Motion Control
With the success of 2D and 3D visual generative models, there is growing interest in generating 4D content. Existing methods primarily rely on text prompts to produce 4D content, but they often fall short of accurately defining complex or rare motions. To address this limitation, we propose MagicPose4D, a novel framework for refined control over both appearance and motion in 4D generation. Unlike current 4D generation methods, MagicPose4D accepts monocular videos or mesh sequences as motion prompts, enabling precise and customizable motion control. MagicPose4D comprises two key modules: (i) Dual-Phase 4D Reconstruction Module, which operates in two phases. The first phase focuses on capturing the model's shape using accurate 2D supervision and less accurate but geometrically informative 3D pseudo-supervision without imposing skeleton constraints. The second phase extracts the 3D motion (skeleton poses) using more accurate pseudo-3D supervision, obtained in the first phase and introduces kinematic chain-based skeleton constraints to ensure physical plausibility. Additionally, we propose a Global-local Chamfer loss that aligns the overall distribution of predicted mesh vertices with the supervision while maintaining part-level alignment without extra annotations. (ii) Cross-category Motion Transfer Module, which leverages the extracted motion from the 4D reconstruction module and uses a kinematic-chain-based skeleton to achieve cross-category motion transfer. It ensures smooth transitions between frames through dynamic rigidity, facilitating robust generalization without additional training. Through extensive experiments, we demonstrate that MagicPose4D significantly improves the accuracy and consistency of 4D content generation, outperforming existing methods in various benchmarks.
comment: Project Page: https://magicpose4d.github.io/
♻ ☆ CLAIM: Clinically-Guided LGE Augmentation for Realistic and Diverse Myocardial Scar Synthesis and Segmentation
Deep learning-based myocardial scar segmentation from late gadolinium enhancement (LGE) cardiac MRI has shown great potential for accurate and timely diagnosis and treatment planning for structural cardiac diseases. However, the limited availability and variability of LGE images with high-quality scar labels restrict the development of robust segmentation models. To address this, we introduce CLAIM: \textbf{C}linically-Guided \textbf{L}GE \textbf{A}ugmentation for Real\textbf{i}stic and Diverse \textbf{M}yocardial Scar Synthesis and Segmentation framework, a framework for anatomically grounded scar generation and segmentation. At its core is the SMILE module (Scar Mask generation guided by cLinical knowledgE), which conditions a diffusion-based generator on the clinically adopted AHA 17-segment model to synthesize images with anatomically consistent and spatially diverse scar patterns. In addition, CLAIM employs a joint training strategy in which the scar segmentation network is optimized alongside the generator, aiming to enhance both the realism of synthesized scars and the accuracy of the scar segmentation performance. Experimental results show that CLAIM produces anatomically coherent scar patterns and achieves higher Dice similarity with real scar distributions compared to baseline models. Our approach enables controllable and realistic myocardial scar synthesis and has demonstrated utility for downstream medical imaging task. Code is available at https://github.com/farheenjabeen/CLAIM-Scar-Synthesis.
comment: 14 Pages
♻ ☆ TCDiff++: An End-to-end Trajectory-Controllable Diffusion Model for Harmonious Music-Driven Group Choreography
Music-driven dance generation has garnered significant attention due to its wide range of industrial applications, particularly in the creation of group choreography. During the group dance generation process, however, most existing methods still face three primary issues: multi-dancer collisions, single-dancer foot sliding and abrupt swapping in the generation of long group dance. In this paper, we propose TCDiff++, a music-driven end-to-end framework designed to generate harmonious group dance. Specifically, to mitigate multi-dancer collisions, we utilize a dancer positioning embedding to better maintain the relative positioning among dancers. Additionally, we incorporate a distance-consistency loss to ensure that inter-dancer distances remain within plausible ranges. To address the issue of single-dancer foot sliding, we introduce a swap mode embedding to indicate dancer swapping patterns and design a Footwork Adaptor to refine raw motion, thereby minimizing foot sliding. For long group dance generation, we present a long group diffusion sampling strategy that reduces abrupt position shifts by injecting positional information into the noisy input. Furthermore, we integrate a Sequence Decoder layer to enhance the model's ability to selectively process long sequences. Extensive experiments demonstrate that our TCDiff++ achieves state-of-the-art performance, particularly in long-duration scenarios, ensuring high-quality and coherent group dance generation.
♻ ☆ LVPNet: A Latent-variable-based Prediction-driven End-to-end Framework for Lossless Compression of Medical Images MICCAI 2025
Autoregressive Initial Bits is a framework that integrates sub-image autoregression and latent variable modeling, demonstrating its advantages in lossless medical image compression. However, in existing methods, the image segmentation process leads to an even distribution of latent variable information across each sub-image, which in turn causes posterior collapse and inefficient utilization of latent variables. To deal with these issues, we propose a prediction-based end-to-end lossless medical image compression method named LVPNet, leveraging global latent variables to predict pixel values and encoding predicted probabilities for lossless compression. Specifically, we introduce the Global Multi-scale Sensing Module (GMSM), which extracts compact and informative latent representations from the entire image, effectively capturing spatial dependencies within the latent space. Furthermore, to mitigate the information loss introduced during quantization, we propose the Quantization Compensation Module (QCM), which learns the distribution of quantization errors and refines the quantized features to compensate for quantization loss. Extensive experiments on challenging benchmarks demonstrate that our method achieves superior compression efficiency compared to state-of-the-art lossless image compression approaches, while maintaining competitive inference speed. The code is at https://github.com/scy-Jackel/LVPNet.
comment: Accepted to MICCAI 2025
♻ ☆ Image Super-Resolution with Guarantees via Conformalized Generative Models
The increasing use of generative ML foundation models for image restoration tasks such as super-resolution calls for robust and interpretable uncertainty quantification methods. We address this need by presenting a novel approach based on conformal prediction techniques to create a 'confidence mask' capable of reliably and intuitively communicating where the generated image can be trusted. Our method is adaptable to any black-box generative model, including those locked behind an opaque API, requires only easily attainable data for calibration, and is highly customizable via the choice of a local image similarity metric. We prove strong theoretical guarantees for our method that span fidelity error control (according to our local image similarity metric), reconstruction quality, and robustness in the face of data leakage. Finally, we empirically evaluate these results and establish our method's solid performance.
comment: 17 pages, 7 figures
♻ ☆ Learning Adaptive Lighting via Channel-Aware Guidance
Learning lighting adaptation is a crucial step in achieving good visual perception and supporting downstream vision tasks. Current research often addresses individual light-related challenges, such as high dynamic range imaging and exposure correction, in isolation. However, we identify shared fundamental properties across these tasks: i) different color channels have different light properties, and ii) the channel differences reflected in the spatial and frequency domains are different. Leveraging these insights, we introduce the channel-aware Learning Adaptive Lighting Network (LALNet), a multi-task framework designed to handle multiple light-related tasks efficiently. Specifically, LALNet incorporates color-separated features that highlight the unique light properties of each color channel, integrated with traditional color-mixed features by Light Guided Attention (LGA). The LGA utilizes color-separated features to guide color-mixed features focusing on channel differences and ensuring visual consistency across all channels. Additionally, LALNet employs dual domain channel modulation for generating color-separated features and a mixed channel modulation and light state space module for producing color-mixed features. Extensive experiments on four representative light-related tasks demonstrate that LALNet significantly outperforms state-of-the-art methods on benchmark tests and requires fewer computational resources. We provide an anonymous online demo at https://xxxxxx2025.github.io/LALNet/.
♻ ☆ Self-Supervised Multimodal NeRF for Autonomous Driving
In this paper, we propose a Neural Radiance Fields (NeRF) based framework, referred to as Novel View Synthesis Framework (NVSF). It jointly learns the implicit neural representation of space and time-varying scene for both LiDAR and Camera. We test this on a real-world autonomous driving scenario containing both static and dynamic scenes. Compared to existing multimodal dynamic NeRFs, our framework is self-supervised, thus eliminating the need for 3D labels. For efficient training and faster convergence, we introduce heuristic-based image pixel sampling to focus on pixels with rich information. To preserve the local features of LiDAR points, a Double Gradient based mask is employed. Extensive experiments on the KITTI-360 dataset show that, compared to the baseline models, our framework has reported best performance on both LiDAR and Camera domain. Code of the model is available at https://github.com/gaurav00700/Selfsupervised-NVSF
♻ ☆ It's not you, it's me -- Global urban visual perception varies across demographics and personalities
Understanding people's preferences and needs is crucial for urban planning decisions, yet current approaches often combine them from multi-cultural and multi-city populations, obscuring important demographic differences and risking amplifying biases. We conducted a large-scale urban visual perception survey of streetscapes worldwide using street view imagery, examining how demographics -- including gender, age, income, education, race and ethnicity, and, for the first time, personality traits -- shape perceptions among 1,000 participants, with balanced demographics, from five countries and 45 nationalities. This dataset, introduced as Street Perception Evaluation Considering Socioeconomics (SPECS), exhibits statistically significant differences in perception scores in six traditionally used indicators (safe, lively, wealthy, beautiful, boring, and depressing) and four new ones we propose (live nearby, walk, cycle, green) among demographics and personalities. We revealed that location-based sentiments are carried over in people's preferences when comparing urban streetscapes with other cities. Further, we compared the perception scores based on where participants and streetscapes are from. We found that an off-the-shelf machine learning model trained on an existing global perception dataset tends to overestimate positive indicators and underestimate negative ones compared to human responses, suggesting that targeted intervention should consider locals' perception. Our study aspires to rectify the myopic treatment of street perception, which rarely considers demographics or personality traits.
comment: Under review
♻ ☆ MambaMorph: a Mamba-based Framework for Medical MR-CT Deformable Registration
Capturing voxel-wise spatial correspondence across distinct modalities is crucial for medical image analysis. However, current registration approaches are not practical enough in terms of registration accuracy and clinical applicability. In this paper, we introduce MambaMorph, a novel multi-modality deformable registration framework. Specifically, MambaMorph utilizes a Mamba-based registration module and a fine-grained, yet simple, feature extractor for efficient long-range correspondence modeling and high-dimensional feature learning, respectively. Additionally, we develop a well-annotated brain MR-CT registration dataset, SR-Reg, to address the scarcity of data in multi-modality registration. To validate MambaMorph's multi-modality registration capabilities, we conduct quantitative experiments on both our SR-Reg dataset and a public T1-T2 dataset. The experimental results on both datasets demonstrate that MambaMorph significantly outperforms the current state-of-the-art learning-based registration methods in terms of registration accuracy. Further study underscores the efficiency of the Mamba-based registration module and the lightweight feature extractor, which achieve notable registration quality while maintaining reasonable computational costs and speeds. We believe that MambaMorph holds significant potential for practical applications in medical image registration. The code for MambaMorph is available at: https://github.com/Guo-Stone/MambaMorph.
♻ ☆ Sampling Matters in Explanations: Towards Trustworthy Attribution Analysis Building Block in Visual Models through Maximizing Explanation Certainty
Image attribution analysis seeks to highlight the feature representations learned by visual models such that the highlighted feature maps can reflect the pixel-wise importance of inputs. Gradient integration is a building block in the attribution analysis by integrating the gradients from multiple derived samples to highlight the semantic features relevant to inferences. Such a building block often combines with other information from visual models such as activation or attention maps to form ultimate explanations. Yet, our theoretical analysis demonstrates that the extent to the alignment of the sample distribution in gradient integration with respect to natural image distribution gives a lower bound of explanation certainty. Prior works add noise into images as samples and the noise distributions can lead to low explanation certainty. Counter-intuitively, our experiment shows that extra information can saturate neural networks. To this end, building trustworthy attribution analysis needs to settle the sample distribution misalignment problem. Instead of adding extra information into input images, we present a semi-optimal sampling approach by suppressing features from inputs. The sample distribution by suppressing features is approximately identical to the distribution of natural images. Our extensive quantitative evaluation on large scale dataset ImageNet affirms that our approach is effective and able to yield more satisfactory explanations against state-of-the-art baselines throughout all experimental models.
comment: Code: https://anonymous.4open.science/r/sampling_matters_reproducibility-BB60/
♻ ☆ VICCA: Visual Interpretation and Comprehension of Chest X-ray Anomalies in Generated Report Without Human Feedback
As artificial intelligence (AI) becomes increasingly central to healthcare, the demand for explainable and trustworthy models is paramount. Current report generation systems for chest X-rays (CXR) often lack mechanisms for validating outputs without expert oversight, raising concerns about reliability and interpretability. To address these challenges, we propose a novel multimodal framework designed to enhance the semantic alignment and localization accuracy of AI-generated medical reports. Our framework integrates two key modules: a Phrase Grounding Model, which identifies and localizes pathologies in CXR images based on textual prompts, and a Text-to-Image Diffusion Module, which generates synthetic CXR images from prompts while preserving anatomical fidelity. By comparing features between the original and generated images, we introduce a dual-scoring system: one score quantifies localization accuracy, while the other evaluates semantic consistency. This approach significantly outperforms existing methods, achieving state-of-the-art results in pathology localization and text-to-image alignment. The integration of phrase grounding with diffusion models, coupled with the dual-scoring evaluation system, provides a robust mechanism for validating report quality, paving the way for more trustworthy and transparent AI in medical imaging.
♻ ☆ Bounding-box Watermarking: Defense against Model Extraction Attacks on Object Detectors ECML-PKDD2025
Deep neural networks (DNNs) deployed in a cloud often allow users to query models via the APIs. However, these APIs expose the models to model extraction attacks (MEAs). In this attack, the attacker attempts to duplicate the target model by abusing the responses from the API. Backdoor-based DNN watermarking is known as a promising defense against MEAs, wherein the defender injects a backdoor into extracted models via API responses. The backdoor is used as a watermark of the model; if a suspicious model has the watermark (i.e., backdoor), it is verified as an extracted model. This work focuses on object detection (OD) models. Existing backdoor attacks on OD models are not applicable for model watermarking as the defense against MEAs on a realistic threat model. Our proposed approach involves inserting a backdoor into extracted models via APIs by stealthily modifying the bounding-boxes (BBs) of objects detected in queries while keeping the OD capability. In our experiments on three OD datasets, the proposed approach succeeded in identifying the extracted models with 100% accuracy in a wide variety of experimental scenarios.
comment: Accepted at ECML-PKDD2025. Please refer to the conference proceedings for the final version. Source codes: https://zenodo.org/records/15641464
♻ ☆ Neural Graph Map: Dense Mapping with Efficient Loop Closure Integration WACV 2025
Neural field-based SLAM methods typically employ a single, monolithic field as their scene representation. This prevents efficient incorporation of loop closure constraints and limits scalability. To address these shortcomings, we propose a novel RGB-D neural mapping framework in which the scene is represented by a collection of lightweight neural fields which are dynamically anchored to the pose graph of a sparse visual SLAM system. Our approach shows the ability to integrate large-scale loop closures, while requiring only minimal reintegration. Furthermore, we verify the scalability of our approach by demonstrating successful building-scale mapping taking multiple loop closures into account during the optimization, and show that our method outperforms existing state-of-the-art approaches on large scenes in terms of quality and runtime. Our code is available open-source at https://github.com/KTH-RPL/neural_graph_mapping.
comment: WACV 2025, Project page: https://kth-rpl.github.io/neural_graph_mapping/
♻ ☆ ULSR-GS: Ultra Large-scale Surface Reconstruction Gaussian Splatting with Multi-View Geometric Consistency
While Gaussian Splatting (GS) demonstrates efficient and high-quality scene rendering and small area surface extraction ability, it falls short in handling large-scale aerial image surface extraction tasks. To overcome this, we present ULSR-GS, a framework dedicated to high-fidelity surface extraction in ultra-large-scale scenes, addressing the limitations of existing GS-based mesh extraction methods. Specifically, we propose a point-to-photo partitioning approach combined with a multi-view optimal view matching principle to select the best training images for each sub-region. Additionally, during training, ULSR-GS employs a densification strategy based on multi-view geometric consistency to enhance surface extraction details. Experimental results demonstrate that ULSR-GS outperforms other state-of-the-art GS-based works on large-scale aerial photogrammetry benchmark datasets, significantly improving surface extraction accuracy in complex urban environments. Project page: https://ulsrgs.github.io.
comment: Project page: https://ulsrgs.github.io
♻ ☆ World-Consistent Data Generation for Vision-and-Language Navigation
Vision-and-Language Navigation (VLN) is a challenging task that requires an agent to navigate through photorealistic environments following natural-language instructions. One main obstacle existing in VLN is data scarcity, leading to poor generalization performance over unseen environments. Though data argumentation is a promising way for scaling up the dataset, how to generate VLN data both diverse and world-consistent remains problematic. To cope with this issue, we propose the world-consistent data generation (WCGEN), an efficacious data-augmentation framework satisfying both diversity and world-consistency, aimed at enhancing the generalization of agents to novel environments. Roughly, our framework consists of two stages, the trajectory stage which leverages a point-cloud based technique to ensure spatial coherency among viewpoints, and the viewpoint stage which adopts a novel angle synthesis method to guarantee spatial and wraparound consistency within the entire observation. By accurately predicting viewpoint changes with 3D knowledge, our approach maintains the world-consistency during the generation procedure. Experiments on a wide range of datasets verify the effectiveness of our method, demonstrating that our data augmentation strategy enables agents to achieve new state-of-the-art results on all navigation tasks, and is capable of enhancing the VLN agents' generalization ability to unseen environments.
♻ ☆ Provably Improving Generalization of Few-Shot Models with Synthetic Data ICML 2025
Few-shot image classification remains challenging due to the scarcity of labeled training examples. Augmenting them with synthetic data has emerged as a promising way to alleviate this issue, but models trained on synthetic samples often face performance degradation due to the inherent gap between real and synthetic distributions. To address this limitation, we develop a theoretical framework that quantifies the impact of such distribution discrepancies on supervised learning, specifically in the context of image classification. More importantly, our framework suggests practical ways to generate good synthetic samples and to train a predictor with high generalization ability. Building upon this framework, we propose a novel theoretical-based algorithm that integrates prototype learning to optimize both data partitioning and model training, effectively bridging the gap between real few-shot data and synthetic data. Extensive experiments results show that our approach demonstrates superior performance compared to state-of-the-art methods, outperforming them across multiple datasets.
comment: ICML 2025. Our code is released at https://github.com/Fsoft-AIC/ProtoAug
♻ ☆ Mamba Policy: Towards Efficient 3D Diffusion Policy with Hybrid Selective State Models IROS 2025
Diffusion models have been widely employed in the field of 3D manipulation due to their efficient capability to learn distributions, allowing for precise prediction of action trajectories. However, diffusion models typically rely on large parameter UNet backbones as policy networks, which can be challenging to deploy on resource-constrained devices. Recently, the Mamba model has emerged as a promising solution for efficient modeling, offering low computational complexity and strong performance in sequence modeling. In this work, we propose the Mamba Policy, a lighter but stronger policy that reduces the parameter count by over 80% compared to the original policy network while achieving superior performance. Specifically, we introduce the XMamba Block, which effectively integrates input information with conditional features and leverages a combination of Mamba and Attention mechanisms for deep feature extraction. Extensive experiments demonstrate that the Mamba Policy excels on the Adroit, Dexart, and MetaWorld datasets, requiring significantly fewer computational resources. Additionally, we highlight the Mamba Policy's enhanced robustness in long-horizon scenarios compared to baseline methods and explore the performance of various Mamba variants within the Mamba Policy framework. Real-world experiments are also conducted to further validate its effectiveness. Our open-source project page can be found at https://andycao1125.github.io/mamba_policy/.
comment: Accepted to IROS 2025
♻ ☆ WoundAmbit: Bridging State-of-the-Art Semantic Segmentation and Real-World Wound Care ECML
Chronic wounds affect a large population, particularly the elderly and diabetic patients, who often exhibit limited mobility and co-existing health conditions. Automated wound monitoring via mobile image capture can reduce in-person physician visits by enabling remote tracking of wound size. Semantic segmentation is key to this process, yet wound segmentation remains underrepresented in medical imaging research. To address this, we benchmark state-of-the-art deep learning models from general-purpose vision, medical imaging, and top methods from public wound challenges. For a fair comparison, we standardize training, data augmentation, and evaluation, conducting cross-validation to minimize partitioning bias. We also assess real-world deployment aspects, including generalization to an out-of-distribution wound dataset, computational efficiency, and interpretability. Additionally, we propose a reference object-based approach to convert AI-generated masks into clinically relevant wound size estimates and evaluate this, along with mask quality, for the five best architectures based on physician assessments. Overall, the transformer-based TransNeXt showed the highest levels of generalizability. Despite variations in inference times, all models processed at least one image per second on the CPU, which is deemed adequate for the intended application. Interpretability analysis typically revealed prominent activations in wound regions, emphasizing focus on clinically relevant features. Expert evaluation showed high mask approval for all analyzed models, with VWFormer and ConvNeXtS backbone performing the best. Size retrieval accuracy was similar across models, and predictions closely matched expert annotations. Finally, we demonstrate how our AI-driven wound size estimation framework, WoundAmbit, is integrated into a custom telehealth system.
comment: Main paper: 18 pages; supplementary material: 15 pages; the paper has been accepted for publication at the Applied Data Science (ADS) track of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2025)
♻ ☆ Toddlers' Active Gaze Behavior Supports Self-Supervised Object Learning
Toddlers learn to recognize objects from different viewpoints with almost no supervision. During this learning, they execute frequent eye and head movements that shape their visual experience. It is presently unclear if and how these behaviors contribute to toddlers' emerging object recognition abilities. To answer this question, we here combine head-mounted eye tracking during dyadic play with unsupervised machine learning. We approximate toddlers' central visual field experience by cropping image regions from a head-mounted camera centered on the current gaze location estimated via eye tracking. This visual stream feeds an unsupervised computational model of toddlers' learning, which constructs visual representations that slowly change over time. Our experiments demonstrate that toddlers' gaze strategy supports the learning of invariant object representations. Our analysis also shows that the limited size of the central visual field where acuity is high is crucial for this. Overall, our work reveals how toddlers' gaze behavior may support their development of view-invariant object recognition.
comment: 27 pages, 16 figures
♻ ☆ ZigzagPointMamba: Spatial-Semantic Mamba for Point Cloud Understanding
State Space models (SSMs) such as PointMamba enable efficient feature extraction for point cloud self-supervised learning with linear complexity, outperforming Transformers in computational efficiency. However, existing PointMamba-based methods depend on complex token ordering and random masking, which disrupt spatial continuity and local semantic correlations. We propose ZigzagPointMamba to tackle these challenges. The core of our approach is a simple zigzag scan path that globally sequences point cloud tokens, enhancing spatial continuity by preserving the proximity of spatially adjacent point tokens. Nevertheless, random masking undermines local semantic modeling in self-supervised learning. To address this, we introduce a Semantic-Siamese Masking Strategy (SMS), which masks semantically similar tokens to facilitate reconstruction by integrating local features of original and similar tokens. This overcomes the dependence on isolated local features and enables robust global semantic modeling. Our pre-trained ZigzagPointMamba weights significantly improve downstream tasks, achieving a 1.59% mIoU gain on ShapeNetPart for part segmentation, a 0.4% higher accuracy on ModelNet40 for classification, and 0.19%, 1.22%, and 0.72% higher accuracies respectively for the classification tasks on the OBJ-BG, OBJ-ONLY, and PB-T50-RS subsets of ScanObjectNN.
comment: The format of the document has an error and needs to be revised
♻ ☆ KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling CVPR 2024
DETR is a novel end-to-end transformer architecture object detector, which significantly outperforms classic detectors when scaling up. In this paper, we focus on the compression of DETR with knowledge distillation. While knowledge distillation has been well-studied in classic detectors, there is a lack of researches on how to make it work effectively on DETR. We first provide experimental and theoretical analysis to point out that the main challenge in DETR distillation is the lack of consistent distillation points. Distillation points refer to the corresponding inputs of the predictions for student to mimic, which have different formulations in CNN detector and DETR, and reliable distillation requires sufficient distillation points which are consistent between teacher and student. Based on this observation, we propose the first general knowledge distillation paradigm for DETR (KD-DETR) with consistent distillation points sampling, for both homogeneous and heterogeneous distillation. Specifically, we decouple detection and distillation tasks by introducing a set of specialized object queries to construct distillation points for DETR. We further propose a general-to-specific distillation points sampling strategy to explore the extensibility of KD-DETR. Extensive experiments validate the effectiveness and generalization of KD-DETR. For both single-scale DAB-DETR and multis-scale Deformable DETR and DINO, KD-DETR boost the performance of student model with improvements of $2.6\%-5.2\%$. We further extend KD-DETR to heterogeneous distillation, and achieves $2.1\%$ improvement by distilling the knowledge from DINO to Faster R-CNN with ResNet-50, which is comparable with homogeneous distillation methods.The code is available at https://github.com/wennyuhey/KD-DETR.
comment: Accepted to CVPR 2024
♻ ☆ FGS-SLAM: Fourier-based Gaussian Splatting for Real-time SLAM with Sparse and Dense Map Fusion
3D gaussian splatting has advanced simultaneous localization and mapping (SLAM) technology by enabling real-time positioning and the construction of high-fidelity maps. However, the uncertainty in gaussian position and initialization parameters introduces challenges, often requiring extensive iterative convergence and resulting in redundant or insufficient gaussian representations. To address this, we introduce a novel adaptive densification method based on Fourier frequency domain analysis to establish gaussian priors for rapid convergence. Additionally, we propose constructing independent and unified sparse and dense maps, where a sparse map supports efficient tracking via Generalized Iterative Closest Point (GICP) and a dense map creates high-fidelity visual representations. This is the first SLAM system leveraging frequency domain analysis to achieve high-quality gaussian mapping in real-time. Experimental results demonstrate an average frame rate of 36 FPS on Replica and TUM RGB-D datasets, achieving competitive accuracy in both localization and mapping.
♻ ☆ 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
♻ ☆ Matching-Free Depth Recovery from Structured Light
We introduce a novel approach for depth estimation using images obtained from monocular structured light systems. In contrast to many existing methods that depend on image matching, our technique employs a density voxel grid to represent scene geometry. This grid is trained through self-supervised differentiable volume rendering. Our method leverages color fields derived from the projected patterns in structured light systems during the rendering process, facilitating the isolated optimization of the geometry field. This innovative approach leads to faster convergence and high-quality results. Additionally, we integrate normalized device coordinates (NDC), a distortion loss, and a distinctive surface-based color loss to enhance geometric fidelity. Experimental results demonstrate that our method outperforms current matching-based techniques in terms of geometric performance in few-shot scenarios, achieving an approximately 30% reduction in average estimated depth errors for both synthetic scenes and real-world captured scenes. Moreover, our approach allows for rapid training, being approximately three times faster than previous matching-free methods that utilize implicit representations.
comment: 13 pages, 10 figures
♻ ☆ VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning
Reinforcement fine-tuning (RFT) has shown great promise in achieving humanlevel reasoning capabilities of Large Language Models (LLMs), and has recently been extended to MLLMs. Nevertheless, reasoning about videos, which is a fundamental aspect of human intelligence, remains a persistent challenge due to the complex logic, temporal and causal structures inherent in video data. To fill this gap, we propose VIDEORFT, a novel approach that extends the RFT paradigm to cultivate human-like video reasoning capabilities in MLLMs. VIDEORFT follows the standard two-stage scheme in RFT: supervised fine-tuning (SFT) with chain-of-thought (CoT) annotations, followed by reinforcement learning (RL) to improve generalization. A central challenge to achieve this in the video domain lies in the scarcity of large-scale, high-quality video CoT datasets. We address this by building a fully automatic CoT curation pipeline. First, we devise a cognitioninspired prompting strategy to elicit a reasoning LLM to generate preliminary CoTs based solely on rich, structured, and literal representations of video content. Subsequently, these CoTs are revised by a visual-language model conditioned on the actual video, ensuring visual consistency and reducing visual hallucinations. This pipeline results in two new datasets - VideoRFT-CoT-102K for SFT and VideoRFT-RL-310K for RL. To further strengthen the RL phase, we introduce a novel semantic-consistency reward that explicitly promotes the alignment between textual reasoning and visual evidence. This reward encourages the model to produce coherent, context-aware reasoning outputs grounded in visual input. Extensive experiments show that VIDEORFT achieves state-of-the-art performance on six video reasoning benchmarks.
comment: Code: https://github.com/QiWang98/VideoRFT
♻ ☆ Skin Color Measurement from Dermatoscopic Images: An Evaluation on a Synthetic Dataset
This paper presents a comprehensive evaluation of skin color measurement methods from dermatoscopic images using a synthetic dataset (S-SYNTH) with controlled ground-truth melanin content, lesion shapes, hair models, and 18 distinct lighting conditions. This allows for rigorous assessment of the robustness and invariance to lighting conditions. We assess four classes of image colorimetry approaches: segmentation-based, patch-based, color quantization, and neural networks. We use these methods to estimate the Individual Typology Angle (ITA) and Fitzpatrick types from dermatoscopic images. Our results show that segmentation-based and color quantization methods yield robust, lighting-invariant estimates, whereas patch-based approaches exhibit significant lighting-dependent biases that require calibration. Furthermore, neural network models, particularly when combined with heavy blurring to reduce overfitting, can provide light-invariant Fitzpatrick predictions, although their generalization to real-world images remains unverified. We conclude with practical recommendations for designing fair and reliable skin color estimation methods.
♻ ☆ ReconX: Reconstruct Any Scene from Sparse Views with Video Diffusion Model
Advancements in 3D scene reconstruction have transformed 2D images from the real world into 3D models, producing realistic 3D results from hundreds of input photos. Despite great success in dense-view reconstruction scenarios, rendering a detailed scene from insufficient captured views is still an ill-posed optimization problem, often resulting in artifacts and distortions in unseen areas. In this paper, we propose ReconX, a novel 3D scene reconstruction paradigm that reframes the ambiguous reconstruction challenge as a temporal generation task. The key insight is to unleash the strong generative prior of large pre-trained video diffusion models for sparse-view reconstruction. However, 3D view consistency struggles to be accurately preserved in directly generated video frames from pre-trained models. To address this, given limited input views, the proposed ReconX first constructs a global point cloud and encodes it into a contextual space as the 3D structure condition. Guided by the condition, the video diffusion model then synthesizes video frames that are both detail-preserved and exhibit a high degree of 3D consistency, ensuring the coherence of the scene from various perspectives. Finally, we recover the 3D scene from the generated video through a confidence-aware 3D Gaussian Splatting optimization scheme. Extensive experiments on various real-world datasets show the superiority of our ReconX over state-of-the-art methods in terms of quality and generalizability.
comment: Project page: https://liuff19.github.io/ReconX
♻ ☆ A Siamese Network to Detect If Two Iris Images Are Monozygotic
This study presents the first automated classifier designed to determine whether a pair of iris images originates from monozygotic individuals, addressing a previously untackled problem in biometric recognition. In Daugman-style iris recognition, the textures of the left and right irises of the same person are traditionally considered as being as different as the irises of two unrelated persons. However, previous research indicates that humans can detect that two iris images are from different eyes of the same person, or eyes of monozygotic twins, with an accuracy of about 80%. In this work, we employ a Siamese network architecture and contrastive learning to categorize a pair of iris images as coming from monozygotic or non-monozygotic irises. This could potentially be applied, for example, as a fast, noninvasive test to determine if twins are monozygotic or non-monozygotic. We construct a dataset comprising both synthetic monozygotic pairs (images of different irises of the same individual) and natural monozygotic pairs (images of different images from persons who are identical twins), in addition to non-monozygotic pairs from unrelated individuals, ensuring a comprehensive evaluation of the model's capabilities. To gain deeper insights into the learned representations, we train and analyze three variants of the model using (1) the original input images, (2) iris-only images (masking everything but the iris region), and (3) non-iris-only images (masking the iris region). This comparison reveals that both iris texture and surrounding ocular structure contain information useful for the model to classify the image pairs as monozygotic or non-monozygotic. Our approach achieves accuracy levels using the full iris image that exceed those previously reported for human classification of monozygotic iris pairs.
♻ ☆ EvDetMAV: Generalized MAV Detection from Moving Event Cameras
Existing micro aerial vehicle (MAV) detection methods mainly rely on the target's appearance features in RGB images, whose diversity makes it difficult to achieve generalized MAV detection. We notice that different types of MAVs share the same distinctive features in event streams due to their high-speed rotating propellers, which are hard to see in RGB images. This paper studies how to detect different types of MAVs from an event camera by fully exploiting the features of propellers in the original event stream. The proposed method consists of three modules to extract the salient and spatio-temporal features of the propellers while filtering out noise from background objects and camera motion. Since there are no existing event-based MAV datasets, we introduce a novel MAV dataset for the community. This is the first event-based MAV dataset comprising multiple scenarios and different types of MAVs. Without training, our method significantly outperforms state-of-the-art methods and can deal with challenging scenarios, achieving a precision rate of 83.0\% (+30.3\%) and a recall rate of 81.5\% (+36.4\%) on the proposed testing dataset. The dataset and code are available at: https://github.com/WindyLab/EvDetMAV.
comment: 8 pages, 7 figures. This paper is accepted by IEEE Robotics and Automation Letters
♻ ☆ TIIF-Bench: How Does Your T2I Model Follow Your Instructions?
The rapid advancements of Text-to-Image (T2I) models have ushered in a new phase of AI-generated content, marked by their growing ability to interpret and follow user instructions. However, existing T2I model evaluation benchmarks fall short in limited prompt diversity and complexity, as well as coarse evaluation metrics, making it difficult to evaluate the fine-grained alignment performance between textual instructions and generated images. In this paper, we present TIIF-Bench (Text-to-Image Instruction Following Benchmark), aiming to systematically assess T2I models' ability in interpreting and following intricate textual instructions. TIIF-Bench comprises a set of 5000 prompts organized along multiple dimensions, which are categorized into three levels of difficulties and complexities. To rigorously evaluate model robustness to varying prompt lengths, we provide a short and a long version for each prompt with identical core semantics. Two critical attributes, i.e., text rendering and style control, are introduced to evaluate the precision of text synthesis and the aesthetic coherence of T2I models. In addition, we collect 100 high-quality designer level prompts that encompass various scenarios to comprehensively assess model performance. Leveraging the world knowledge encoded in large vision language models, we propose a novel computable framework to discern subtle variations in T2I model outputs. Through meticulous benchmarking of mainstream T2I models on TIIF-Bench, we analyze the pros and cons of current T2I models and reveal the limitations of current T2I benchmarks. Project Page: https://a113n-w3i.github.io/TIIF_Bench/.
comment: 23 pages, 12 figures, 11 tables
♻ ☆ USP-Gaussian: Unifying Spike-based Image Reconstruction, Pose Correction and Gaussian Splatting
Spike cameras, as an innovative neuromorphic camera that captures scenes with the 0-1 bit stream at 40 kHz, are increasingly employed for the 3D reconstruction task via Neural Radiance Fields (NeRF) or 3D Gaussian Splatting (3DGS). Previous spike-based 3D reconstruction approaches often employ a casecased pipeline: starting with high-quality image reconstruction from spike streams based on established spike-to-image reconstruction algorithms, then progressing to camera pose estimation and 3D reconstruction. However, this cascaded approach suffers from substantial cumulative errors, where quality limitations of initial image reconstructions negatively impact pose estimation, ultimately degrading the fidelity of the 3D reconstruction. To address these issues, we propose a synergistic optimization framework, \textbf{USP-Gaussian}, that unifies spike-based image reconstruction, pose correction, and Gaussian splatting into an end-to-end framework. Leveraging the multi-view consistency afforded by 3DGS and the motion capture capability of the spike camera, our framework enables a joint iterative optimization that seamlessly integrates information between the spike-to-image network and 3DGS. Experiments on synthetic datasets with accurate poses demonstrate that our method surpasses previous approaches by effectively eliminating cascading errors. Moreover, we integrate pose optimization to achieve robust 3D reconstruction in real-world scenarios with inaccurate initial poses, outperforming alternative methods by effectively reducing noise and preserving fine texture details. Our code, data and trained models will be available at https://github.com/chenkang455/USP-Gaussian.
♻ ☆ VLN-R1: Vision-Language Navigation via Reinforcement Fine-Tuning
Vision-Language Navigation (VLN) is a core challenge in embodied AI, requiring agents to navigate real-world environments using natural language instructions. Current language model-based navigation systems operate on discrete topological graphs, limiting path planning to predefined node connections. We propose VLN-R1, an end-to-end framework that leverages Large Vision-Language Models (LVLM) to directly translate egocentric video streams into continuous navigation actions, adopting GRPO-based training inspired by DeepSeek-R1. To enable effective training, we first construct the VLN-Ego dataset using a 3D simulator, Habitat, and propose Long-Short Memory Sampling to balance historical and current observations. While large language models can supervise complete textual instructions, they lack fine-grained action-level control. Our framework employs a two-stage training approach: a) Supervised fine-tuning (SFT) to align the model's action sequence text predictions with expert demonstrations, followed by b) Reinforcement fine-tuning (RFT) enhanced with a Time-Decayed Reward (TDR) mechanism that strategically weights multi-step future actions. Experimental results show VLN-R1 achieves strong performance on VLN-CE benchmark. VLN-R1 proves LVLMs can drive embodied navigation and enhance task-specific reasoning through data-efficient, reward-driven post-training.
comment: project page: vlnr1.github.io
♻ ☆ C3S3: Complementary Competition and Contrastive Selection for Semi-Supervised Medical Image Segmentation ICME 2025
For the immanent challenge of insufficiently annotated samples in the medical field, semi-supervised medical image segmentation (SSMIS) offers a promising solution. Despite achieving impressive results in delineating primary target areas, most current methodologies struggle to precisely capture the subtle details of boundaries. This deficiency often leads to significant diagnostic inaccuracies. To tackle this issue, we introduce C3S3, a novel semi-supervised segmentation model that synergistically integrates complementary competition and contrastive selection. This design significantly sharpens boundary delineation and enhances overall precision. Specifically, we develop an Outcome-Driven Contrastive Learning module dedicated to refining boundary localization. Additionally, we incorporate a Dynamic Complementary Competition module that leverages two high-performing sub-networks to generate pseudo-labels, thereby further improving segmentation quality. The proposed C3S3 undergoes rigorous validation on two publicly accessible datasets, encompassing the practices of both MRI and CT scans. The results demonstrate that our method achieves superior performance compared to previous cutting-edge competitors. Especially, on the 95HD and ASD metrics, our approach achieves a notable improvement of at least 6%, highlighting the significant advancements. The code is available at https://github.com/Y-TARL/C3S3.
comment: Accepted to ICME 2025
♻ ☆ One Prototype Is Enough: Single-Prototype Activation for Interpretable Image Classification
In this paper, we propose ProtoSolo, a novel deep neural architecture for interpretable image classification inspired by prototypical networks such as ProtoPNet. Existing prototype networks usually rely on the collaborative decision-making of multiple prototypes to achieve the classification and interpretation of a single category. In contrast, ProtoSolo only requires the activation of a single prototype to complete the classification. This allows the network to explain each category decision by only providing the features that are most similar to the prototype of that category, significantly reducing the cognitive complexity of the explanation. Secondly, we propose a feature-based comparison method, which uses feature map instead of full-channel feature vector as the object of similarity comparison and prototype learning. This design enables ProtoSolo to utilize richer global information for classification while relying on a single prototype activation. In addition, we propose a non-prototype projection learning strategy, which preserves the information association between the prototype and the training image patches while avoiding the sharp change of the network structure caused by the projection operation, thus avoiding its negative impact on the classification performance. Experiments on the CUB-200-2011 and Stanford Cars datasets show that ProtoSolo achieves superior performance in classification tasks and reaches the best level in terms of cognitive complexity of explanations compared to state-of-the-art interpretable methods. The code is available at https://github.com/pyt19/ProtoSolo.
♻ ☆ Robust Multimodal Learning for Ophthalmic Disease Grading via Disentangled Representation
This paper discusses how ophthalmologists often rely on multimodal data to improve diagnostic accuracy. However, complete multimodal data is rare in real-world applications due to a lack of medical equipment and concerns about data privacy. Traditional deep learning methods typically address these issues by learning representations in latent space. However, the paper highlights two key limitations of these approaches: (i) Task-irrelevant redundant information (e.g., numerous slices) in complex modalities leads to significant redundancy in latent space representations. (ii) Overlapping multimodal representations make it difficult to extract unique features for each modality. To overcome these challenges, the authors propose the Essence-Point and Disentangle Representation Learning (EDRL) strategy, which integrates a self-distillation mechanism into an end-to-end framework to enhance feature selection and disentanglement for more robust multimodal learning. Specifically, the Essence-Point Representation Learning module selects discriminative features that improve disease grading performance. The Disentangled Representation Learning module separates multimodal data into modality-common and modality-unique representations, reducing feature entanglement and enhancing both robustness and interpretability in ophthalmic disease diagnosis. Experiments on multimodal ophthalmology datasets show that the proposed EDRL strategy significantly outperforms current state-of-the-art methods.
comment: 10pages
♻ ☆ Morse: Dual-Sampling for Lossless Acceleration of Diffusion Models ICML 2025
In this paper, we present Morse, a simple dual-sampling framework for accelerating diffusion models losslessly. The key insight of Morse is to reformulate the iterative generation (from noise to data) process via taking advantage of fast jump sampling and adaptive residual feedback strategies. Specifically, Morse involves two models called Dash and Dot that interact with each other. The Dash model is just the pre-trained diffusion model of any type, but operates in a jump sampling regime, creating sufficient space for sampling efficiency improvement. The Dot model is significantly faster than the Dash model, which is learnt to generate residual feedback conditioned on the observations at the current jump sampling point on the trajectory of the Dash model, lifting the noise estimate to easily match the next-step estimate of the Dash model without jump sampling. By chaining the outputs of the Dash and Dot models run in a time-interleaved fashion, Morse exhibits the merit of flexibly attaining desired image generation performance while improving overall runtime efficiency. With our proposed weight sharing strategy between the Dash and Dot models, Morse is efficient for training and inference. Our method shows a lossless speedup of 1.78X to 3.31X on average over a wide range of sampling step budgets relative to 9 baseline diffusion models on 6 image generation tasks. Furthermore, we show that our method can be also generalized to improve the Latent Consistency Model (LCM-SDXL, which is already accelerated with consistency distillation technique) tailored for few-step text-to-image synthesis. The code and models are available at https://github.com/deep-optimization/Morse.
comment: Fixed a prompt typo in Figure 18 of the Appendix. This work is accepted to ICML 2025. The project page: https://github.com/deep-optimization/Morse
♻ ☆ Predictive Modeling, Pattern Recognition, and Spatiotemporal Representations of Plant Growth in Simulated and Controlled Environments: A Comprehensive Review
Accurate predictions and representations of plant growth patterns in simulated and controlled environments are important for addressing various challenges in plant phenomics research. This review explores various works on state-of-the-art predictive pattern recognition techniques, focusing on the spatiotemporal modeling of plant traits and the integration of dynamic environmental interactions. We provide a comprehensive examination of deterministic, probabilistic, and generative modeling approaches, emphasizing their applications in high-throughput phenotyping and simulation-based plant growth forecasting. Key topics include regressions and neural network-based representation models for the task of forecasting, limitations of existing experiment-based deterministic approaches, and the need for dynamic frameworks that incorporate uncertainty and evolving environmental feedback. This review surveys advances in 2D and 3D structured data representations through functional-structural plant models and conditional generative models. We offer a perspective on opportunities for future works, emphasizing the integration of domain-specific knowledge to data-driven methods, improvements to available datasets, and the implementation of these techniques toward real-world applications.
♻ ☆ Visual and Textual Prompts in VLLMs for Enhancing Emotion Recognition
Vision Large Language Models (VLLMs) exhibit promising potential for multi-modal understanding, yet their application to video-based emotion recognition remains limited by insufficient spatial and contextual awareness. Traditional approaches, which prioritize isolated facial features, often neglect critical non-verbal cues such as body language, environmental context, and social interactions, leading to reduced robustness in real-world scenarios. To address this gap, we propose Set-of-Vision-Text Prompting (SoVTP), a novel framework that enhances zero-shot emotion recognition by integrating spatial annotations (e.g., bounding boxes, facial landmarks), physiological signals (facial action units), and contextual cues (body posture, scene dynamics, others' emotions) into a unified prompting strategy. SoVTP preserves holistic scene information while enabling fine-grained analysis of facial muscle movements and interpersonal dynamics. Extensive experiments show that SoVTP achieves substantial improvements over existing visual prompting methods, demonstrating its effectiveness in enhancing VLLMs' video emotion recognition capabilities.
comment: 14 pages, 14 figures
♻ ☆ BeltCrack: the First Sequential-image Industrial Conveyor Belt Crack Detection Dataset and Its Baseline with Triple-domain Feature Learning
Conveyor belts are important equipment in modern industry, widely applied in production and manufacturing. Their health is much critical to operational efficiency and safety. Cracks are a major threat to belt health. Currently, considering safety, how to intelligently detect belt cracks is catching an increasing attention. To implement the intelligent detection with machine learning, real crack samples are believed to be necessary. However, existing crack datasets primarily focus on pavement scenarios or synthetic data, no real-world industrial belt crack datasets at all. Cracks are a major threat to belt health. Furthermore, to validate usability and effectiveness, we propose a special baseline method with triple-domain ($i.e.$, time-space-frequency) feature hierarchical fusion learning for the two whole-new datasets. Experimental results demonstrate the availability and effectiveness of our dataset. Besides, they also show that our baseline is obviously superior to other similar detection methods. Our datasets and source codes are available at https://github.com/UESTC-nnLab/BeltCrack.
comment: 14 pages, 10 figures
♻ ☆ PP-DocBee2: Improved Baselines with Efficient Data for Multimodal Document Understanding
This report introduces PP-DocBee2, an advanced version of the PP-DocBee, designed to enhance multimodal document understanding. Built on a large multimodal model architecture, PP-DocBee2 addresses the limitations of its predecessor through key technological improvements, including enhanced synthetic data quality, improved visual feature fusion strategy, and optimized inference methodologies. These enhancements yield an $11.4\%$ performance boost on internal benchmarks for Chinese business documents, and reduce inference latency by $73.0\%$ to the vanilla version. A key innovation of our work is a data quality optimization strategy for multimodal document tasks. By employing a large-scale multimodal pre-trained model to evaluate data, we apply a novel statistical criterion to filter outliers, ensuring high-quality training data. Inspired by insights into underutilized intermediate features in multimodal models, we enhance the ViT representational capacity by decomposing it into layers and applying a novel feature fusion strategy to improve complex reasoning. The source code and pre-trained model are available at \href{https://github.com/PaddlePaddle/PaddleMIX}{https://github.com/PaddlePaddle/PaddleMIX}.
♻ ☆ Fine-Grained Perturbation Guidance via Attention Head Selection
Recent guidance methods in diffusion models steer reverse sampling by perturbing the model to construct an implicit weak model and guide generation away from it. Among these approaches, attention perturbation has demonstrated strong empirical performance in unconditional scenarios where classifier-free guidance is not applicable. However, existing attention perturbation methods lack principled approaches for determining where perturbations should be applied, particularly in Diffusion Transformer (DiT) architectures where quality-relevant computations are distributed across layers. In this paper, we investigate the granularity of attention perturbations, ranging from the layer level down to individual attention heads, and discover that specific heads govern distinct visual concepts such as structure, style, and texture quality. Building on this insight, we propose "HeadHunter", a systematic framework for iteratively selecting attention heads that align with user-centric objectives, enabling fine-grained control over generation quality and visual attributes. In addition, we introduce SoftPAG, which linearly interpolates each selected head's attention map toward an identity matrix, providing a continuous knob to tune perturbation strength and suppress artifacts. Our approach not only mitigates the oversmoothing issues of existing layer-level perturbation but also enables targeted manipulation of specific visual styles through compositional head selection. We validate our method on modern large-scale DiT-based text-to-image models including Stable Diffusion 3 and FLUX.1, demonstrating superior performance in both general quality enhancement and style-specific guidance. Our work provides the first head-level analysis of attention perturbation in diffusion models, uncovering interpretable specialization within attention layers and enabling practical design of effective perturbation strategies.
comment: Project page: https://cvlab-kaist.github.io/HeadHunter/
♻ ☆ Low-light Pedestrian Detection in Visible and Infrared Image Feeds: Issues and Challenges
Pedestrian detection has become a cornerstone for several high-level tasks, including autonomous driving, intelligent transportation, and traffic surveillance. There are several works focussed on pedestrian detection using visible images, mainly in the daytime. However, this task is very intriguing when the environmental conditions change to poor lighting or nighttime. Recently, new ideas have been spurred to use alternative sources, such as Far InfraRed (FIR) temperature sensor feeds for detecting pedestrians in low-light conditions. This study reviews recent developments in low-light pedestrian detection approaches. It systematically categorizes and analyses various algorithms from region-based to non-region-based and graph-based learning methodologies by highlighting their methodologies, implementation issues, and challenges. It also outlines the key benchmark datasets that can be used for research and development of advanced pedestrian detection algorithms, particularly in low-light situations.
comment: 29 pages, 4 tables, 21 figures
Information Retrieval 11
☆ Unidentified and Confounded? Understanding Two-Tower Models for Unbiased Learning to Rank
Additive two-tower models are popular learning-to-rank methods for handling biased user feedback in industry settings. Recent studies, however, report a concerning phenomenon: training two-tower models on clicks collected by well-performing production systems leads to decreased ranking performance. This paper investigates two recent explanations for this observation: confounding effects from logging policies and model identifiability issues. We theoretically analyze the identifiability conditions of two-tower models, showing that either document swaps across positions or overlapping feature distributions are required to recover model parameters from clicks. We also investigate the effect of logging policies on two-tower models, finding that they introduce no bias when models perfectly capture user behavior. However, logging policies can amplify biases when models imperfectly capture user behavior, particularly when prediction errors correlate with document placement across positions. We propose a sample weighting technique to mitigate these effects and provide actionable insights for researchers and practitioners using two-tower models.
☆ ReCode: Updating Code API Knowledge with Reinforcement Learning
Large Language Models (LLMs) exhibit remarkable code generation capabilities but falter when adapting to frequent updates in external library APIs. This critical limitation, stemming from reliance on outdated API knowledge from their training data, even with access to current documentation, impedes reliable code generation in dynamic environments. To tackle this issue, we propose ReCode (rule-based Reinforcement learning for Code Update), a novel framework that mimics human programmer adaptation to API changes. Specifically, we construct a dataset of approximately 2,000 data entries to train the LLMs to perform version migration based on updated information. Then, we introduce a modified string similarity metric for code evaluation as the reward for reinforcement learning. Our experiments demonstrate that ReCode substantially boosts LLMs' code generation performance in dynamic API scenarios, especially on the unseen CodeUpdateArena task. Crucially, compared to supervised fine-tuning, ReCode has less impact on LLMs' general code generation abilities. We apply ReCode on various LLMs and reinforcement learning algorithms (GRPO and DAPO), all achieving consistent improvements. Notably, after training, Qwen2.5-Coder-7B outperforms that of the 32B parameter code instruction-tuned model and the reasoning model with the same architecture. Code is available at https://github.com/zjunlp/ReCode.
comment: Work in progress
☆ Knowledge-Aware Diverse Reranking for Cross-Source Question Answering
This paper presents Team Marikarp's solution for the SIGIR 2025 LiveRAG competition. The competition's evaluation set, automatically generated by DataMorgana from internet corpora, encompassed a wide range of target topics, question types, question formulations, audience types, and knowledge organization methods. It offered a fair evaluation of retrieving question-relevant supporting documents from a 15M documents subset of the FineWeb corpus. Our proposed knowledge-aware diverse reranking RAG pipeline achieved first place in the competition.
☆ Semantic-enhanced Modality-asymmetric Retrieval for Online E-commerce Search
Semantic retrieval, which retrieves semantically matched items given a textual query, has been an essential component to enhance system effectiveness in e-commerce search. In this paper, we study the multimodal retrieval problem, where the visual information (e.g, image) of item is leveraged as supplementary of textual information to enrich item representation and further improve retrieval performance. Though learning from cross-modality data has been studied extensively in tasks such as visual question answering or media summarization, multimodal retrieval remains a non-trivial and unsolved problem especially in the asymmetric scenario where the query is unimodal while the item is multimodal. In this paper, we propose a novel model named SMAR, which stands for Semantic-enhanced Modality-Asymmetric Retrieval, to tackle the problem of modality fusion and alignment in this kind of asymmetric scenario. Extensive experimental results on an industrial dataset show that the proposed model outperforms baseline models significantly in retrieval accuracy. We have open sourced our industrial dataset for the sake of reproducibility and future research works.
comment: published in sigir2023
☆ A Literature Review on Simulation in Conversational Recommender Systems
Conversational Recommender Systems (CRSs) have garnered attention as a novel approach to delivering personalized recommendations through multi-turn dialogues. This review developed a taxonomy framework to systematically categorize relevant publications into four groups: dataset construction, algorithm design, system evaluation, and empirical studies, providing a comprehensive analysis of simulation methods in CRSs research. Our analysis reveals that simulation methods play a key role in tackling CRSs' main challenges. For example, LLM-based simulation methods have been used to create conversational recommendation data, enhance CRSs algorithms, and evaluate CRSs. Despite several challenges, such as dataset bias, the limited output flexibility of LLM-based simulations, and the gap between text semantic space and behavioral semantics, persist due to the complexity in Human-Computer Interaction (HCI) of CRSs, simulation methods hold significant potential for advancing CRS research. This review offers a thorough summary of the current research landscape in this domain and identifies promising directions for future inquiry.
comment: 6 pages, 1 figures, accepted as a poster for CSWIM 2025
☆ Irec: A Metacognitive Scaffolding for Self-Regulated Learning through Just-in-Time Insight Recall: A Conceptual Framework and System Prototype
The core challenge in learning has shifted from knowledge acquisition to effective Self-Regulated Learning (SRL): planning, monitoring, and reflecting on one's learning. Existing digital tools, however, inadequately support metacognitive reflection. Spaced Repetition Systems (SRS) use de-contextualized review, overlooking the role of context, while Personal Knowledge Management (PKM) tools require high manual maintenance. To address these challenges, this paper introduces "Insight Recall," a novel paradigm that conceptualizes the context-triggered retrieval of personal past insights as a metacognitive scaffold to promote SRL. We formalize this paradigm using the Just-in-Time Adaptive Intervention (JITAI) framework and implement a prototype system, Irec, to demonstrate its feasibility. At its core, Irec uses a dynamic knowledge graph of the user's learning history. When a user faces a new problem, a hybrid retrieval engine recalls relevant personal "insights." Subsequently, a large language model (LLM) performs a deep similarity assessment to filter and present the most relevant scaffold in a just-in-time manner. To reduce cognitive load, Irec features a human-in-the-loop pipeline for LLM-based knowledge graph construction. We also propose an optional "Guided Inquiry" module, where users can engage in a Socratic dialogue with an expert LLM, using the current problem and recalled insights as context. The contribution of this paper is a solid theoretical framework and a usable system platform for designing next-generation intelligent learning systems that enhance metacognition and self-regulation.
comment: Version 1 of a work in progress. Finalized system flowcharts, a public GitHub repository with the source code, and a full reproducibility package detailing the prompts, models, and testing guidelines will be provided in v2
☆ Multimodal Information Retrieval for Open World with Edit Distance Weak Supervision ICDE'24
Existing multi-media retrieval models either rely on creating a common subspace with modality-specific representation models or require schema mapping among modalities to measure similarities among multi-media data. Our goal is to avoid the annotation overhead incurred from considering retrieval as a supervised classification task and re-use the pretrained encoders in large language models and vision tasks. We propose "FemmIR", a framework to retrieve multimodal results relevant to information needs expressed with multimodal queries by example without any similarity label. Such identification is necessary for real-world applications where data annotations are scarce and satisfactory performance is required without fine-tuning with a common framework across applications. We curate a new dataset called MuQNOL for benchmarking progress on this task. Our technique is based on weak supervision introduced through edit distance between samples: graph edit distance can be modified to consider the cost of replacing a data sample in terms of its properties, and relevance can be measured through the implicit signal from the amount of edit cost among the objects. Unlike metric learning or encoding networks, FemmIR re-uses the high-level properties and maintains the property value and relationship constraints with a multi-level interaction score between data samples and the query example provided by the user. We empirically evaluate FemmIR on a missing person use case with MuQNOL. FemmIR performs comparably to similar retrieval systems in delivering on-demand retrieval results with exact and approximate similarities while using the existing property identifiers in the system.
comment: Submitted to ICDE'24. An earlier version of this paper appeared on TechRxiv: https://www.techrxiv.org/doi/full/10.36227/techrxiv.21990284.v1, uploaded on February 05, 2023
♻ ☆ Forgetful by Design? A Critical Audit of YouTube's Search API for Academic Research
This paper critically audits the search endpoint of YouTube's Data API (v3), a common tool for academic research. Through systematic weekly searches over six months using eleven queries, we identify major limitations regarding completeness, representativeness, consistency, and bias. Our findings reveal substantial differences between ranking parameters like relevance and date in terms of video recall and precision, with relevance often retrieving numerous off-topic videos. We also find severe temporal decay, as the number of findable videos for a specific period dramatically decreases after just 20-60 days from the publication date, potentially hampering many different research designs. Furthermore, search results lack consistency, with identical queries yielding different video sets over time, compromising replicability. A case study on the European Parliament elections highlights how these issues impact research outcomes. While the paper offers several mitigation strategies, it concludes that the API's search function, potentially prioritizing "freshness" over comprehensive retrieval, is not adequate for robust academic research, especially concerning Digital Services Act requirements.
comment: 15 pages, 2 tables and 4 figures
♻ ☆ Diffusion Recommender Model SIGIR'23
Generative models such as Generative Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs) are widely utilized to model the generative process of user interactions. However, these generative models suffer from intrinsic limitations such as the instability of GANs and the restricted representation ability of VAEs. Such limitations hinder the accurate modeling of the complex user interaction generation procedure, such as noisy interactions caused by various interference factors. In light of the impressive advantages of Diffusion Models (DMs) over traditional generative models in image synthesis, we propose a novel Diffusion Recommender Model (named DiffRec) to learn the generative process in a denoising manner. To retain personalized information in user interactions, DiffRec reduces the added noises and avoids corrupting users' interactions into pure noises like in image synthesis. In addition, we extend traditional DMs to tackle the unique challenges in practical recommender systems: high resource costs for large-scale item prediction and temporal shifts of user preference. To this end, we propose two extensions of DiffRec: L-DiffRec clusters items for dimension compression and conducts the diffusion processes in the latent space; and T-DiffRec reweights user interactions based on the interaction timestamps to encode temporal information. We conduct extensive experiments on three datasets under multiple settings (e.g. clean training, noisy training, and temporal training). The empirical results and in-depth analysis validate the superiority of DiffRec with two extensions over competitive baselines.
comment: 10 pages, 7 figures, accepted for publication in SIGIR'23
♻ ☆ Dual-Channel Multiplex Graph Neural Networks for Recommendation
Effective recommender systems play a crucial role in accurately capturing user and item attributes that mirror individual preferences. Some existing recommendation techniques have started to shift their focus towards modeling various types of interactive relations between users and items in real-world recommendation scenarios, such as clicks, marking favorites, and purchases on online shopping platforms. Nevertheless, these approaches still grapple with two significant challenges: (1) Insufficient modeling and exploitation of the impact of various behavior patterns formed by multiplex relations between users and items on representation learning, and (2) ignoring the effect of different relations within behavior patterns on the target relation in recommender system scenarios. In this work, we introduce a novel recommendation framework, Dual-Channel Multiplex Graph Neural Network (DCMGNN), which addresses the aforementioned challenges. It incorporates an explicit behavior pattern representation learner to capture the behavior patterns composed of multiplex user-item interactive relations, and includes a relation chain representation learner and a relation chain-aware encoder to discover the impact of various auxiliary relations on the target relation, the dependencies between different relations, and mine the appropriate order of relations in a behavior pattern. Extensive experiments on three real-world datasets demonstrate that our DCMGNN surpasses various state-of-the-art recommendation methods. It outperforms the best baselines by 10.06% and 12.15% on average across all datasets in terms of Recall@10 and NDCG@10, respectively.
♻ ☆ Mapping the Evolution of Research Contributions using KnoVo
This paper presents KnoVo (Knowledge Evolution), an intelligent framework designed for quantifying and analyzing the evolution of research novelty in the scientific literature. Moving beyond traditional citation analysis, which primarily measures impact, KnoVo determines a paper's novelty relative to both prior and subsequent work within its multilayered citation network. Given a target paper's abstract, KnoVo utilizes Large Language Models (LLMs) to dynamically extract dimensions of comparison (e.g., methodology, application, dataset). The target paper is then compared to related publications along these same extracted dimensions. This comparative analysis, inspired by tournament selection, yields quantitative novelty scores reflecting the relative improvement, equivalence, or inferiority of the target paper in specific aspects. By aggregating these scores and visualizing their progression, for instance, through dynamic evolution graphs and comparative radar charts, KnoVo facilitates researchers not only to assess originality and identify similar work, but also to track knowledge evolution along specific research dimensions, uncover research gaps, and explore cross-disciplinary connections. We demonstrate these capabilities through a detailed analysis of 20 diverse papers from multiple scientific fields and report on the performance of various open-source LLMs within the KnoVo framework.
Robotics 52
☆ Robust Robotic Exploration and Mapping Using Generative Occupancy Map Synthesis
We present a novel approach for enhancing robotic exploration by using generative occupancy mapping. We introduce SceneSense, a diffusion model designed and trained for predicting 3D occupancy maps given partial observations. Our proposed approach probabilistically fuses these predictions into a running occupancy map in real-time, resulting in significant improvements in map quality and traversability. We implement SceneSense onboard a quadruped robot and validate its performance with real-world experiments to demonstrate the effectiveness of the model. In these experiments, we show that occupancy maps enhanced with SceneSense predictions better represent our fully observed ground truth data (24.44% FID improvement around the robot and 75.59% improvement at range). We additionally show that integrating SceneSense-enhanced maps into our robotic exploration stack as a "drop-in" map improvement, utilizing an existing off-the-shelf planner, results in improvements in robustness and traversability time. Finally we show results of full exploration evaluations with our proposed system in two dissimilar environments and find that locally enhanced maps provide more consistent exploration results than maps constructed only from direct sensor measurements.
comment: arXiv admin note: text overlap with arXiv:2409.10681
☆ Consensus-Driven Uncertainty for Robotic Grasping based on RGB Perception
Deep object pose estimators are notoriously overconfident. A grasping agent that both estimates the 6-DoF pose of a target object and predicts the uncertainty of its own estimate could avoid task failure by choosing not to act under high uncertainty. Even though object pose estimation improves and uncertainty quantification research continues to make strides, few studies have connected them to the downstream task of robotic grasping. We propose a method for training lightweight, deep networks to predict whether a grasp guided by an image-based pose estimate will succeed before that grasp is attempted. We generate training data for our networks via object pose estimation on real images and simulated grasping. We also find that, despite high object variability in grasping trials, networks benefit from training on all objects jointly, suggesting that a diverse variety of objects can nevertheless contribute to the same goal.
☆ Hierarchical Reinforcement Learning and Value Optimization for Challenging Quadruped Locomotion
We propose a novel hierarchical reinforcement learning framework for quadruped locomotion over challenging terrain. Our approach incorporates a two-layer hierarchy in which a high-level policy (HLP) selects optimal goals for a low-level policy (LLP). The LLP is trained using an on-policy actor-critic RL algorithm and is given footstep placements as goals. We propose an HLP that does not require any additional training or environment samples and instead operates via an online optimization process over the learned value function of the LLP. We demonstrate the benefits of this framework by comparing it with an end-to-end reinforcement learning (RL) approach. We observe improvements in its ability to achieve higher rewards with fewer collisions across an array of different terrains, including terrains more difficult than any encountered during training.
☆ Robust Embodied Self-Identification of Morphology in Damaged Multi-Legged Robots
Multi-legged robots (MLRs) are vulnerable to leg damage during complex missions, which can impair their performance. This paper presents a self-modeling and damage identification algorithm that enables autonomous adaptation to partial or complete leg loss using only data from a low-cost IMU. A novel FFT-based filter is introduced to address time-inconsistent signals, improving damage detection by comparing body orientation between the robot and its model. The proposed method identifies damaged legs and updates the robot's model for integration into its control system. Experiments on uneven terrain validate its robustness and computational efficiency.
☆ Evolutionary Gait Reconfiguration in Damaged Legged Robots
Multi-legged robots deployed in complex missions are susceptible to physical damage in their legs, impairing task performance and potentially compromising mission success. This letter presents a rapid, training-free damage recovery algorithm for legged robots subject to partial or complete loss of functional legs. The proposed method first stabilizes locomotion by generating a new gait sequence and subsequently optimally reconfigures leg gaits via a developed differential evolution algorithm to maximize forward progression while minimizing body rotation and lateral drift. The algorithm successfully restores locomotion in a 24-degree-of-freedom hexapod within one hour, demonstrating both high efficiency and robustness to structural damage.
☆ Unified Vision-Language-Action Model
Vision-language-action models (VLAs) have garnered significant attention for their potential in advancing robotic manipulation. However, previous approaches predominantly rely on the general comprehension capabilities of vision-language models (VLMs) to generate action signals, often overlooking the rich temporal and causal structure embedded in visual observations. In this paper, we present UniVLA, a unified and native multimodal VLA model that autoregressively models vision, language, and action signals as discrete token sequences. This formulation enables flexible multimodal tasks learning, particularly from large-scale video data. By incorporating world modeling during post-training, UniVLA captures causal dynamics from videos, facilitating effective transfer to downstream policy learning--especially for long-horizon tasks. Our approach sets new state-of-the-art results across several widely used simulation benchmarks, including CALVIN, LIBERO, and Simplenv-Bridge, significantly surpassing previous methods. For example, UniVLA achieves 95.5% average success rate on LIBERO benchmark, surpassing pi0-FAST's 85.5%. We further demonstrate its broad applicability on real-world ALOHA manipulation and autonomous driving.
comment: technical report
☆ ManiGaussian++: General Robotic Bimanual Manipulation with Hierarchical Gaussian World Model
Multi-task robotic bimanual manipulation is becoming increasingly popular as it enables sophisticated tasks that require diverse dual-arm collaboration patterns. Compared to unimanual manipulation, bimanual tasks pose challenges to understanding the multi-body spatiotemporal dynamics. An existing method ManiGaussian pioneers encoding the spatiotemporal dynamics into the visual representation via Gaussian world model for single-arm settings, which ignores the interaction of multiple embodiments for dual-arm systems with significant performance drop. In this paper, we propose ManiGaussian++, an extension of ManiGaussian framework that improves multi-task bimanual manipulation by digesting multi-body scene dynamics through a hierarchical Gaussian world model. To be specific, we first generate task-oriented Gaussian Splatting from intermediate visual features, which aims to differentiate acting and stabilizing arms for multi-body spatiotemporal dynamics modeling. We then build a hierarchical Gaussian world model with the leader-follower architecture, where the multi-body spatiotemporal dynamics is mined for intermediate visual representation via future scene prediction. The leader predicts Gaussian Splatting deformation caused by motions of the stabilizing arm, through which the follower generates the physical consequences resulted from the movement of the acting arm. As a result, our method significantly outperforms the current state-of-the-art bimanual manipulation techniques by an improvement of 20.2% in 10 simulated tasks, and achieves 60% success rate on average in 9 challenging real-world tasks. Our code is available at https://github.com/April-Yz/ManiGaussian_Bimanual.
☆ Look to Locate: Vision-Based Multisensory Navigation with 3-D Digital Maps for GNSS-Challenged Environments
In Global Navigation Satellite System (GNSS)-denied environments such as indoor parking structures or dense urban canyons, achieving accurate and robust vehicle positioning remains a significant challenge. This paper proposes a cost-effective, vision-based multi-sensor navigation system that integrates monocular depth estimation, semantic filtering, and visual map registration (VMR) with 3-D digital maps. Extensive testing in real-world indoor and outdoor driving scenarios demonstrates the effectiveness of the proposed system, achieving sub-meter accuracy of 92% indoors and more than 80% outdoors, with consistent horizontal positioning and heading average root mean-square errors of approximately 0.98 m and 1.25 {\deg}, respectively. Compared to the baselines examined, the proposed solution significantly reduced drift and improved robustness under various conditions, achieving positioning accuracy improvements of approximately 88% on average. This work highlights the potential of cost-effective monocular vision systems combined with 3D maps for scalable, GNSS-independent navigation in land vehicles.
☆ CronusVLA: Transferring Latent Motion Across Time for Multi-Frame Prediction in Manipulation
Recent vision-language-action (VLA) models built on pretrained vision-language models (VLMs) have demonstrated strong generalization across manipulation tasks. However, they remain constrained by a single-frame observation paradigm and cannot fully benefit from the motion information offered by aggregated multi-frame historical observations, as the large vision-language backbone introduces substantial computational cost and inference latency. We propose CronusVLA, a unified framework that extends single-frame VLA models to the multi-frame paradigm through an efficient post-training stage. CronusVLA comprises three key components: (1) single-frame pretraining on large-scale embodied datasets with autoregressive action tokens prediction, which establishes an embodied vision-language foundation; (2) multi-frame encoding, adapting the prediction of vision-language backbones from discrete action tokens to motion features during post-training, and aggregating motion features from historical frames into a feature chunking; (3) cross-frame decoding, which maps the feature chunking to accurate actions via a shared decoder with cross-attention. By reducing redundant token computation and caching past motion features, CronusVLA achieves efficient inference. As an application of motion features, we further propose an action adaptation mechanism based on feature-action retrieval to improve model performance during finetuning. CronusVLA achieves state-of-the-art performance on SimplerEnv with 70.9% success rate, and 12.7% improvement over OpenVLA on LIBERO. Real-world Franka experiments also show the strong performance and robustness.
comment: 36 pages, 21 figures
☆ ReactEMG: Zero-Shot, Low-Latency Intent Detection via sEMG
Surface electromyography (sEMG) signals show promise for effective human-computer interfaces, particularly in rehabilitation and prosthetics. However, challenges remain in developing systems that respond quickly and reliably to user intent, across different subjects and without requiring time-consuming calibration. In this work, we propose a framework for EMG-based intent detection that addresses these challenges. Unlike traditional gesture recognition models that wait until a gesture is completed before classifying it, our approach uses a segmentation strategy to assign intent labels at every timestep as the gesture unfolds. We introduce a novel masked modeling strategy that aligns muscle activations with their corresponding user intents, enabling rapid onset detection and stable tracking of ongoing gestures. In evaluations against baseline methods, considering both accuracy and stability for device control, our approach surpasses state-of-the-art performance in zero-shot transfer conditions, demonstrating its potential for wearable robotics and next-generation prosthetic systems. Our project page is available at: https://reactemg.github.io
☆ The Starlink Robot: A Platform and Dataset for Mobile Satellite Communication
The integration of satellite communication into mobile devices represents a paradigm shift in connectivity, yet the performance characteristics under motion and environmental occlusion remain poorly understood. We present the Starlink Robot, the first mobile robotic platform equipped with Starlink satellite internet, comprehensive sensor suite including upward-facing camera, LiDAR, and IMU, designed to systematically study satellite communication performance during movement. Our multi-modal dataset captures synchronized communication metrics, motion dynamics, sky visibility, and 3D environmental context across diverse scenarios including steady-state motion, variable speeds, and different occlusion conditions. This platform and dataset enable researchers to develop motion-aware communication protocols, predict connectivity disruptions, and optimize satellite communication for emerging mobile applications from smartphones to autonomous vehicles. The project is available at https://github.com/StarlinkRobot.
☆ Systematic Comparison of Projection Methods for Monocular 3D Human Pose Estimation on Fisheye Images
Fisheye cameras offer robots the ability to capture human movements across a wider field of view (FOV) than standard pinhole cameras, making them particularly useful for applications in human-robot interaction and automotive contexts. However, accurately detecting human poses in fisheye images is challenging due to the curved distortions inherent to fisheye optics. While various methods for undistorting fisheye images have been proposed, their effectiveness and limitations for poses that cover a wide FOV has not been systematically evaluated in the context of absolute human pose estimation from monocular fisheye images. To address this gap, we evaluate the impact of pinhole, equidistant and double sphere camera models, as well as cylindrical projection methods, on 3D human pose estimation accuracy. We find that in close-up scenarios, pinhole projection is inadequate, and the optimal projection method varies with the FOV covered by the human pose. The usage of advanced fisheye models like the double sphere model significantly enhances 3D human pose estimation accuracy. We propose a heuristic for selecting the appropriate projection model based on the detection bounding box to enhance prediction quality. Additionally, we introduce and evaluate on our novel dataset FISHnCHIPS, which features 3D human skeleton annotations in fisheye images, including images from unconventional angles, such as extreme close-ups, ground-mounted cameras, and wide-FOV poses, available at: https://www.vision.rwth-aachen.de/fishnchips
comment: Presented at IEEE International Conference on Robotics and Automation 2025
☆ Estimating Spatially-Dependent GPS Errors Using a Swarm of Robots
External factors, including urban canyons and adversarial interference, can lead to Global Positioning System (GPS) inaccuracies that vary as a function of the position in the environment. This study addresses the challenge of estimating a static, spatially-varying error function using a team of robots. We introduce a State Bias Estimation Algorithm (SBE) whose purpose is to estimate the GPS biases. The central idea is to use sensed estimates of the range and bearing to the other robots in the team to estimate changes in bias across the environment. A set of drones moves in a 2D environment, each sampling data from GPS, range, and bearing sensors. The biases calculated by the SBE at estimated positions are used to train a Gaussian Process Regression (GPR) model. We use a Sparse Gaussian process-based Informative Path Planning (IPP) algorithm that identifies high-value regions of the environment for data collection. The swarm plans paths that maximize information gain in each iteration, further refining their understanding of the environment's positional bias landscape. We evaluated SBE and IPP in simulation and compared the IPP methodology to an open-loop strategy.
comment: 6 pages, 7 figures, 2025 IEEE 21st International Conference on Automation Science and Engineering
☆ UniTac-NV: A Unified Tactile Representation For Non-Vision-Based Tactile Sensors IROS
Generalizable algorithms for tactile sensing remain underexplored, primarily due to the diversity of sensor modalities. Recently, many methods for cross-sensor transfer between optical (vision-based) tactile sensors have been investigated, yet little work focus on non-optical tactile sensors. To address this gap, we propose an encoder-decoder architecture to unify tactile data across non-vision-based sensors. By leveraging sensor-specific encoders, the framework creates a latent space that is sensor-agnostic, enabling cross-sensor data transfer with low errors and direct use in downstream applications. We leverage this network to unify tactile data from two commercial tactile sensors: the Xela uSkin uSPa 46 and the Contactile PapillArray. Both were mounted on a UR5e robotic arm, performing force-controlled pressing sequences against distinct object shapes (circular, square, and hexagonal prisms) and two materials (rigid PLA and flexible TPU). Another more complex unseen object was also included to investigate the model's generalization capabilities. We show that alignment in latent space can be implicitly learned from joint autoencoder training with matching contacts collected via different sensors. We further demonstrate the practical utility of our approach through contact geometry estimation, where downstream models trained on one sensor's latent representation can be directly applied to another without retraining.
comment: 7 pages, 8 figures. Accepted version to appear in: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
☆ ReLink: Computational Circular Design of Planar Linkage Mechanisms Using Available Standard Parts
The Circular Economy framework emphasizes sustainability by reducing resource consumption and waste through the reuse of components and materials. This paper presents ReLink, a computational framework for the circular design of planar linkage mechanisms using available standard parts. Unlike most mechanism design methods, which assume the ability to create custom parts and infinite part availability, ReLink prioritizes the reuse of discrete, standardized components, thus minimizing the need for new parts. The framework consists of two main components: design generation, where a generative design algorithm generates mechanisms from an inventory of available parts, and inverse design, which uses optimization methods to identify designs that match a user-defined trajectory curve. The paper also examines the trade-offs between kinematic performance and CO2 footprint when incorporating new parts. Challenges such as the combinatorial nature of the design problem and the enforcement of valid solutions are addressed. By combining sustainability principles with kinematic synthesis, ReLink lays the groundwork for further research into computational circular design to support the development of systems that integrate reused components into mechanical products.
comment: 29 pages, 18 figures, submitted to the Journal of Cleaner Production
☆ A Verification Methodology for Safety Assurance of Robotic Autonomous Systems
Autonomous robots deployed in shared human environments, such as agricultural settings, require rigorous safety assurance to meet both functional reliability and regulatory compliance. These systems must operate in dynamic, unstructured environments, interact safely with humans, and respond effectively to a wide range of potential hazards. This paper presents a verification workflow for the safety assurance of an autonomous agricultural robot, covering the entire development life-cycle, from concept study and design to runtime verification. The outlined methodology begins with a systematic hazard analysis and risk assessment to identify potential risks and derive corresponding safety requirements. A formal model of the safety controller is then developed to capture its behaviour and verify that the controller satisfies the specified safety properties with respect to these requirements. The proposed approach is demonstrated on a field robot operating in an agricultural setting. The results show that the methodology can be effectively used to verify safety-critical properties and facilitate the early identification of design issues, contributing to the development of safer robots and autonomous systems.
comment: In Proc. of the 26th TAROS (Towards Autonomous Robotic Systems) Conference, York, UK, August, 2025
☆ Probabilistic modelling and safety assurance of an agriculture robot providing light-treatment
Continued adoption of agricultural robots postulates the farmer's trust in the reliability, robustness and safety of the new technology. This motivates our work on safety assurance of agricultural robots, particularly their ability to detect, track and avoid obstacles and humans. This paper considers a probabilistic modelling and risk analysis framework for use in the early development phases. Starting off with hazard identification and a risk assessment matrix, the behaviour of the mobile robot platform, sensor and perception system, and any humans present are captured using three state machines. An auto-generated probabilistic model is then solved and analysed using the probabilistic model checker PRISM. The result provides unique insight into fundamental development and engineering aspects by quantifying the effect of the risk mitigation actions and risk reduction associated with distinct design concepts. These include implications of adopting a higher performance and more expensive Object Detection System or opting for a more elaborate warning system to increase human awareness. Although this paper mainly focuses on the initial concept-development phase, the proposed safety assurance framework can also be used during implementation, and subsequent deployment and operation phases.
☆ Soft Robotic Delivery of Coiled Anchors for Cardiac Interventions
Trans-catheter cardiac intervention has become an increasingly available option for high-risk patients without the complications of open heart surgery. However, current catheterbased platforms suffer from a lack of dexterity, force application, and compliance required to perform complex intracardiac procedures. An exemplary task that would significantly ease minimally invasive intracardiac procedures is the implantation of anchor coils, which can be used to fix and implant various devices in the beating heart. We introduce a robotic platform capable of delivering anchor coils. We develop a kineto-statics model of the robotic platform and demonstrate low positional error. We leverage the passive compliance and high force output of the actuator in a multi-anchor delivery procedure against a motile in-vitro simulator with millimeter level accuracy.
comment: This work has been submitted to the IEEE for possible publication
☆ Robotics Under Construction: Challenges on Job Sites ICRA
As labor shortages and productivity stagnation increasingly challenge the construction industry, automation has become essential for sustainable infrastructure development. This paper presents an autonomous payload transportation system as an initial step toward fully unmanned construction sites. Our system, based on the CD110R-3 crawler carrier, integrates autonomous navigation, fleet management, and GNSS-based localization to facilitate material transport in construction site environments. While the current system does not yet incorporate dynamic environment adaptation algorithms, we have begun fundamental investigations into external-sensor based perception and mapping system. Preliminary results highlight the potential challenges, including navigation in evolving terrain, environmental perception under construction-specific conditions, and sensor placement optimization for improving autonomy and efficiency. Looking forward, we envision a construction ecosystem where collaborative autonomous agents dynamically adapt to site conditions, optimizing workflow and reducing human intervention. This paper provides foundational insights into the future of robotics-driven construction automation and identifies critical areas for further technological development.
comment: Workshop on Field Robotics, ICRA
☆ Adaptive Domain Modeling with Language Models: A Multi-Agent Approach to Task Planning
We introduce TAPAS (Task-based Adaptation and Planning using AgentS), a multi-agent framework that integrates Large Language Models (LLMs) with symbolic planning to solve complex tasks without the need for manually defined environment models. TAPAS employs specialized LLM-based agents that collaboratively generate and adapt domain models, initial states, and goal specifications as needed using structured tool-calling mechanisms. Through this tool-based interaction, downstream agents can request modifications from upstream agents, enabling adaptation to novel attributes and constraints without manual domain redefinition. A ReAct (Reason+Act)-style execution agent, coupled with natural language plan translation, bridges the gap between dynamically generated plans and real-world robot capabilities. TAPAS demonstrates strong performance in benchmark planning domains and in the VirtualHome simulated real-world environment.
☆ Fake or Real, Can Robots Tell? Evaluating Embodied Vision-Language Models on Real and 3D-Printed Objects
Robotic scene understanding increasingly relies on vision-language models (VLMs) to generate natural language descriptions of the environment. In this work, we present a comparative study of captioning strategies for tabletop scenes captured by a robotic arm equipped with an RGB camera. The robot collects images of objects from multiple viewpoints, and we evaluate several models that generate scene descriptions. We compare the performance of various captioning models, like BLIP and VLMs. Our experiments examine the trade-offs between single-view and multi-view captioning, and difference between recognising real-world and 3D printed objects. We quantitatively evaluate object identification accuracy, completeness, and naturalness of the generated captions. Results show that VLMs can be used in robotic settings where common objects need to be recognised, but fail to generalise to novel representations. Our findings provide practical insights into deploying foundation models for embodied agents in real-world settings.
☆ T-Rex: Task-Adaptive Spatial Representation Extraction for Robotic Manipulation with Vision-Language Models NeurIPS 2025
Building a general robotic manipulation system capable of performing a wide variety of tasks in real-world settings is a challenging task. Vision-Language Models (VLMs) have demonstrated remarkable potential in robotic manipulation tasks, primarily due to the extensive world knowledge they gain from large-scale datasets. In this process, Spatial Representations (such as points representing object positions or vectors representing object orientations) act as a bridge between VLMs and real-world scene, effectively grounding the reasoning abilities of VLMs and applying them to specific task scenarios. However, existing VLM-based robotic approaches often adopt a fixed spatial representation extraction scheme for various tasks, resulting in insufficient representational capability or excessive extraction time. In this work, we introduce T-Rex, a Task-Adaptive Framework for Spatial Representation Extraction, which dynamically selects the most appropriate spatial representation extraction scheme for each entity based on specific task requirements. Our key insight is that task complexity determines the types and granularity of spatial representations, and Stronger representational capabilities are typically associated with Higher overall system operation costs. Through comprehensive experiments in real-world robotic environments, we show that our approach delivers significant advantages in spatial understanding, efficiency, and stability without additional training.
comment: submitted to NeurIPS 2025
☆ Ground-Effect-Aware Modeling and Control for Multicopters
The ground effect on multicopters introduces several challenges, such as control errors caused by additional lift, oscillations that may occur during near-ground flight due to external torques, and the influence of ground airflow on models such as the rotor drag and the mixing matrix. This article collects and analyzes the dynamics data of near-ground multicopter flight through various methods, including force measurement platforms and real-world flights. For the first time, we summarize the mathematical model of the external torque of multicopters under ground effect. The influence of ground airflow on rotor drag and the mixing matrix is also verified through adequate experimentation and analysis. Through simplification and derivation, the differential flatness of the multicopter's dynamic model under ground effect is confirmed. To mitigate the influence of these disturbance models on control, we propose a control method that combines dynamic inverse and disturbance models, ensuring consistent control effectiveness at both high and low altitudes. In this method, the additional thrust and variations in rotor drag under ground effect are both considered and compensated through feedforward models. The leveling torque of ground effect can be equivalently represented as variations in the center of gravity and the moment of inertia. In this way, the leveling torque does not explicitly appear in the dynamic model. The final experimental results show that the method proposed in this paper reduces the control error (RMSE) by \textbf{45.3\%}. Please check the supplementary material at: https://github.com/ZJU-FAST-Lab/Ground-effect-controller.
☆ Is an object-centric representation beneficial for robotic manipulation ?
Object-centric representation (OCR) has recently become a subject of interest in the computer vision community for learning a structured representation of images and videos. It has been several times presented as a potential way to improve data-efficiency and generalization capabilities to learn an agent on downstream tasks. However, most existing work only evaluates such models on scene decomposition, without any notion of reasoning over the learned representation. Robotic manipulation tasks generally involve multi-object environments with potential inter-object interaction. We thus argue that they are a very interesting playground to really evaluate the potential of existing object-centric work. To do so, we create several robotic manipulation tasks in simulated environments involving multiple objects (several distractors, the robot, etc.) and a high-level of randomization (object positions, colors, shapes, background, initial positions, etc.). We then evaluate one classical object-centric method across several generalization scenarios and compare its results against several state-of-the-art hollistic representations. Our results exhibit that existing methods are prone to failure in difficult scenarios involving complex scene structures, whereas object-centric methods help overcome these challenges.
☆ A Survey on Soft Robot Adaptability: Implementations, Applications, and Prospects
Soft robots, compared to rigid robots, possess inherent advantages, including higher degrees of freedom, compliance, and enhanced safety, which have contributed to their increasing application across various fields. Among these benefits, adaptability is particularly noteworthy. In this paper, adaptability in soft robots is categorized into external and internal adaptability. External adaptability refers to the robot's ability to adjust, either passively or actively, to variations in environments, object properties, geometries, and task dynamics. Internal adaptability refers to the robot's ability to cope with internal variations, such as manufacturing tolerances or material aging, and to generalize control strategies across different robots. As the field of soft robotics continues to evolve, the significance of adaptability has become increasingly pronounced. In this review, we summarize various approaches to enhancing the adaptability of soft robots, including design, sensing, and control strategies. Additionally, we assess the impact of adaptability on applications such as surgery, wearable devices, locomotion, and manipulation. We also discuss the limitations of soft robotics adaptability and prospective directions for future research. By analyzing adaptability through the lenses of implementation, application, and challenges, this paper aims to provide a comprehensive understanding of this essential characteristic in soft robotics and its implications for diverse applications.
comment: 12 pages, 4 figures, accepted by IEEE Robotics & Automation Magazine
☆ Zero-Shot Parameter Learning of Robot Dynamics Using Bayesian Statistics and Prior Knowledge
Inertial parameter identification of industrial robots is an established process, but standard methods using Least Squares or Machine Learning do not consider prior information about the robot and require extensive measurements. Inspired by Bayesian statistics, this paper presents an identification method with improved generalization that incorporates prior knowledge and is able to learn with only a few or without additional measurements (Zero-Shot Learning). Furthermore, our method is able to correctly learn not only the inertial but also the mechanical and base parameters of the MABI Max 100 robot while ensuring physical feasibility and specifying the confidence intervals of the results. We also provide different types of priors for serial robots with 6 degrees of freedom, where datasheets or CAD models are not available.
comment: Carsten Reiners and Minh Trinh contributed equally to this work
☆ Robotic Perception with a Large Tactile-Vision-Language Model for Physical Property Inference
Inferring physical properties can significantly enhance robotic manipulation by enabling robots to handle objects safely and efficiently through adaptive grasping strategies. Previous approaches have typically relied on either tactile or visual data, limiting their ability to fully capture properties. We introduce a novel cross-modal perception framework that integrates visual observations with tactile representations within a multimodal vision-language model. Our physical reasoning framework, which employs a hierarchical feature alignment mechanism and a refined prompting strategy, enables our model to make property-specific predictions that strongly correlate with ground-truth measurements. Evaluated on 35 diverse objects, our approach outperforms existing baselines and demonstrates strong zero-shot generalization. Keywords: tactile perception, visual-tactile fusion, physical property inference, multimodal integration, robot perception
comment: This paper has been accepted by the 2025 International Conference on Climbing and Walking Robots (CLAWAR). These authors contributed equally to this work: Zexiang Guo, Hengxiang Chen, Xinheng Mai
☆ Da Yu: Towards USV-Based Image Captioning for Waterway Surveillance and Scene Understanding
Automated waterway environment perception is crucial for enabling unmanned surface vessels (USVs) to understand their surroundings and make informed decisions. Most existing waterway perception models primarily focus on instance-level object perception paradigms (e.g., detection, segmentation). However, due to the complexity of waterway environments, current perception datasets and models fail to achieve global semantic understanding of waterways, limiting large-scale monitoring and structured log generation. With the advancement of vision-language models (VLMs), we leverage image captioning to introduce WaterCaption, the first captioning dataset specifically designed for waterway environments. WaterCaption focuses on fine-grained, multi-region long-text descriptions, providing a new research direction for visual geo-understanding and spatial scene cognition. Exactly, it includes 20.2k image-text pair data with 1.8 million vocabulary size. Additionally, we propose Da Yu, an edge-deployable multi-modal large language model for USVs, where we propose a novel vision-to-language projector called Nano Transformer Adaptor (NTA). NTA effectively balances computational efficiency with the capacity for both global and fine-grained local modeling of visual features, thereby significantly enhancing the model's ability to generate long-form textual outputs. Da Yu achieves an optimal balance between performance and efficiency, surpassing state-of-the-art models on WaterCaption and several other captioning benchmarks.
comment: 14 pages, 13 figures
☆ AirV2X: Unified Air-Ground Vehicle-to-Everything Collaboration
While multi-vehicular collaborative driving demonstrates clear advantages over single-vehicle autonomy, traditional infrastructure-based V2X systems remain constrained by substantial deployment costs and the creation of "uncovered danger zones" in rural and suburban areas. We present AirV2X-Perception, a large-scale dataset that leverages Unmanned Aerial Vehicles (UAVs) as a flexible alternative or complement to fixed Road-Side Units (RSUs). Drones offer unique advantages over ground-based perception: complementary bird's-eye-views that reduce occlusions, dynamic positioning capabilities that enable hovering, patrolling, and escorting navigation rules, and significantly lower deployment costs compared to fixed infrastructure. Our dataset comprises 6.73 hours of drone-assisted driving scenarios across urban, suburban, and rural environments with varied weather and lighting conditions. The AirV2X-Perception dataset facilitates the development and standardized evaluation of Vehicle-to-Drone (V2D) algorithms, addressing a critical gap in the rapidly expanding field of aerial-assisted autonomous driving systems. The dataset and development kits are open-sourced at https://github.com/taco-group/AirV2X-Perception.
☆ Ontology Neural Network and ORTSF: A Framework for Topological Reasoning and Delay-Robust Control
The advancement of autonomous robotic systems has led to impressive capabilities in perception, localization, mapping, and control. Yet, a fundamental gap remains: existing frameworks excel at geometric reasoning and dynamic stability but fall short in representing and preserving relational semantics, contextual reasoning, and cognitive transparency essential for collaboration in dynamic, human-centric environments. This paper introduces a unified architecture comprising the Ontology Neural Network (ONN) and the Ontological Real-Time Semantic Fabric (ORTSF) to address this gap. The ONN formalizes relational semantic reasoning as a dynamic topological process. By embedding Forman-Ricci curvature, persistent homology, and semantic tensor structures within a unified loss formulation, ONN ensures that relational integrity and topological coherence are preserved as scenes evolve over time. The ORTSF transforms reasoning traces into actionable control commands while compensating for system delays. It integrates predictive and delay-aware operators that ensure phase margin preservation and continuity of control signals, even under significant latency conditions. Empirical studies demonstrate the ONN + ORTSF framework's ability to unify semantic cognition and robust control, providing a mathematically principled and practically viable solution for cognitive robotics.
comment: 12 pages, 5 figures, includes theoretical proofs and simulation results
☆ Scaffolding Dexterous Manipulation with Vision-Language Models
Dexterous robotic hands are essential for performing complex manipulation tasks, yet remain difficult to train due to the challenges of demonstration collection and high-dimensional control. While reinforcement learning (RL) can alleviate the data bottleneck by generating experience in simulation, it typically relies on carefully designed, task-specific reward functions, which hinder scalability and generalization. Thus, contemporary works in dexterous manipulation have often bootstrapped from reference trajectories. These trajectories specify target hand poses that guide the exploration of RL policies and object poses that enable dense, task-agnostic rewards. However, sourcing suitable trajectories - particularly for dexterous hands - remains a significant challenge. Yet, the precise details in explicit reference trajectories are often unnecessary, as RL ultimately refines the motion. Our key insight is that modern vision-language models (VLMs) already encode the commonsense spatial and semantic knowledge needed to specify tasks and guide exploration effectively. Given a task description (e.g., "open the cabinet") and a visual scene, our method uses an off-the-shelf VLM to first identify task-relevant keypoints (e.g., handles, buttons) and then synthesize 3D trajectories for hand motion and object motion. Subsequently, we train a low-level residual RL policy in simulation to track these coarse trajectories or "scaffolds" with high fidelity. Across a number of simulated tasks involving articulated objects and semantic understanding, we demonstrate that our method is able to learn robust dexterous manipulation policies. Moreover, we showcase that our method transfers to real-world robotic hands without any human demonstrations or handcrafted rewards.
☆ Preserving Sense of Agency: User Preferences for Robot Autonomy and User Control across Household Tasks
Roboticists often design with the assumption that assistive robots should be fully autonomous. However, it remains unclear whether users prefer highly autonomous robots, as prior work in assistive robotics suggests otherwise. High robot autonomy can reduce the user's sense of agency, which represents feeling in control of one's environment. How much control do users, in fact, want over the actions of robots used for in-home assistance? We investigate how robot autonomy levels affect users' sense of agency and the autonomy level they prefer in contexts with varying risks. Our study asked participants to rate their sense of agency as robot users across four distinct autonomy levels and ranked their robot preferences with respect to various household tasks. Our findings revealed that participants' sense of agency was primarily influenced by two factors: (1) whether the robot acts autonomously, and (2) whether a third party is involved in the robot's programming or operation. Notably, an end-user programmed robot highly preserved users' sense of agency, even though it acts autonomously. However, in high-risk settings, e.g., preparing a snack for a child with allergies, they preferred robots that prioritized their control significantly more. Additional contextual factors, such as trust in a third party operator, also shaped their preferences.
comment: Accepted by the 2025 34th IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)
☆ The MOTIF Hand: A Robotic Hand for Multimodal Observations with Thermal, Inertial, and Force Sensors
Advancing dexterous manipulation with multi-fingered robotic hands requires rich sensory capabilities, while existing designs lack onboard thermal and torque sensing. In this work, we propose the MOTIF hand, a novel multimodal and versatile robotic hand that extends the LEAP hand by integrating: (i) dense tactile information across the fingers, (ii) a depth sensor, (iii) a thermal camera, (iv), IMU sensors, and (v) a visual sensor. The MOTIF hand is designed to be relatively low-cost (under 4000 USD) and easily reproducible. We validate our hand design through experiments that leverage its multimodal sensing for two representative tasks. First, we integrate thermal sensing into 3D reconstruction to guide temperature-aware, safe grasping. Second, we show how our hand can distinguish objects with identical appearance but different masses - a capability beyond methods that use vision only.
♻ ☆ COBRA-PPM: A Causal Bayesian Reasoning Architecture Using Probabilistic Programming for Robot Manipulation Under Uncertainty
Manipulation tasks require robots to reason about cause and effect when interacting with objects. Yet, many data-driven approaches lack causal semantics and thus only consider correlations. We introduce COBRA-PPM, a novel causal Bayesian reasoning architecture that combines causal Bayesian networks and probabilistic programming to perform interventional inference for robot manipulation under uncertainty. We demonstrate its capabilities through high-fidelity Gazebo-based experiments on an exemplar block stacking task, where it predicts manipulation outcomes with high accuracy (Pred Acc: 88.6%) and performs greedy next-best action selection with a 94.2% task success rate. We further demonstrate sim2real transfer on a domestic robot, showing effectiveness in handling real-world uncertainty from sensor noise and stochastic actions. Our generalised and extensible framework supports a wide range of manipulation scenarios and lays a foundation for future work at the intersection of robotics and causality.
comment: 8 pages, 7 figures, accepted to the 2025 IEEE European Conference on Mobile Robots (ECMR 2025)
♻ ☆ Toward Teach and Repeat Across Seasonal Deep Snow Accumulation
Teach and repeat is a rapid way to achieve autonomy in challenging terrain and off-road environments. A human operator pilots the vehicles to create a network of paths that are mapped and associated with odometry. Immediately after teaching, the system can drive autonomously within its tracks. This precision lets operators remain confident that the robot will follow a traversable route. However, this operational paradigm has rarely been explored in off-road environments that change significantly through seasonal variation. This paper presents preliminary field trials using lidar and radar implementations of teach and repeat. Using a subset of the data from the upcoming FoMo dataset, we attempted to repeat routes that were 4 days, 44 days, and 113 days old. Lidar teach and repeat demonstrated a stronger ability to localize when the ground points were removed. FMCW radar was often able to localize on older maps, but only with small deviations from the taught path. Additionally, we highlight specific cases where radar localization failed with recent maps due to the high pitch or roll of the vehicle. We highlight lessons learned during the field deployment and highlight areas to improve to achieve reliable teach and repeat with seasonal changes in the environment. Please follow the dataset at https://norlab-ulaval.github.io/FoMo-website for updates and information on the data release.
♻ ☆ Energy-Efficient Motion Planner for Legged Robots IROS 2025
We propose an online motion planner for legged robot locomotion with the primary objective of achieving energy efficiency. The conceptual idea is to leverage a placement set of footstep positions based on the robot's body position to determine when and how to execute steps. In particular, the proposed planner uses virtual placement sets beneath the hip joints of the legs and executes a step when the foot is outside of such placement set. Furthermore, we propose a parameter design framework that considers both energy-efficiency and robustness measures to optimize the gait by changing the shape of the placement set along with other parameters, such as step height and swing time, as a function of walking speed. We show that the planner produces trajectories that have a low Cost of Transport (CoT) and high robustness measure, and evaluate our approach against model-free Reinforcement Learning (RL) and motion imitation using biological dog motion priors as the reference. Overall, within low to medium velocity range, we show a 50.4% improvement in CoT and improved robustness over model-free RL, our best performing baseline. Finally, we show ability to handle slippery surfaces, gait transitions, and disturbances in simulation and hardware with the Unitree A1 robot.
comment: This paper has been accepted for publication at the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025). 8 pages, 8 figures
♻ ☆ Robustness Assessment of Assemblies in Frictional Contact
This work establishes a solution to the problem of assessing the capacity of multi-object assemblies to withstand external forces without becoming unstable. Our physically-grounded approach handles arbitrary structures made from rigid objects of any shape and mass distribution without relying on heuristics or approximations. The result is a method that provides a foundation for autonomous robot decision-making when interacting with objects in frictional contact. Our strategy relies on a contact interface graph representation to reason about instabilities and makes use of object shape information to decouple sub-problems and improve efficiency. Our algorithm can be used by motion planners to produce safe assembly transportation plans, and by object placement planners to select better poses. Compared to prior work, our approach is more generally applicable than commonly used heuristics and more efficient than dynamics simulations.
comment: Submitted to IEEE Transactions on Automation Science and Engineering. Contains 14 pages, 16 figures, and 3 tables
♻ ☆ FusionForce: End-to-end Differentiable Neural-Symbolic Layer for Trajectory Prediction
We propose end-to-end differentiable model that predicts robot trajectories on rough offroad terrain from camera images and/or lidar point clouds. The model integrates a learnable component that predicts robot-terrain interaction forces with a neural-symbolic layer that enforces the laws of classical mechanics and consequently improves generalization on out-of-distribution data. The neural-symbolic layer includes a differentiable physics engine that computes the robot's trajectory by querying these forces at the points of contact with the terrain. As the proposed architecture comprises substantial geometrical and physics priors, the resulting model can also be seen as a learnable physics engine conditioned on real sensor data that delivers $10^4$ trajectories per second. We argue and empirically demonstrate that this architecture reduces the sim-to-real gap and mitigates out-of-distribution sensitivity. The differentiability, in conjunction with the rapid simulation speed, makes the model well-suited for various applications including model predictive control, trajectory shooting, supervised and reinforcement learning, or SLAM.
comment: Code: https://github.com/ctu-vras/fusionforce
♻ ☆ ros2 fanuc interface: Design and Evaluation of a Fanuc CRX Hardware Interface in ROS2
This paper introduces the ROS2 control and the Hardware Interface (HW) integration for the Fanuc CRX- robot family. It explains basic implementation details and communication protocols, and its integration with the Moveit2 motion planning library. We conducted a series of experiments to evaluate relevant performances in the robotics field. We tested the developed ros2_fanuc_interface for four relevant robotics cases: step response, trajectory tracking, collision avoidance integrated with Moveit2, and dynamic velocity scaling, respectively. Results show that, despite a non-negligible delay between command and feedback, the robot can track the defined path with negligible errors (if it complies with joint velocity limits), ensuring collision avoidance. Full code is open source and available at https://github.com/paolofrance/ros2_fanuc_interface.
♻ ☆ DroneDiffusion: Robust Quadrotor Dynamics Learning with Diffusion Models ICRA
An inherent fragility of quadrotor systems stems from model inaccuracies and external disturbances. These factors hinder performance and compromise the stability of the system, making precise control challenging. Existing model-based approaches either make deterministic assumptions, utilize Gaussian-based representations of uncertainty, or rely on nominal models, all of which often fall short in capturing the complex, multimodal nature of real-world dynamics. This work introduces DroneDiffusion, a novel framework that leverages conditional diffusion models to learn quadrotor dynamics, formulated as a sequence generation task. DroneDiffusion achieves superior generalization to unseen, complex scenarios by capturing the temporal nature of uncertainties and mitigating error propagation. We integrate the learned dynamics with an adaptive controller for trajectory tracking with stability guarantees. Extensive experiments in both simulation and real-world flights demonstrate the robustness of the framework across a range of scenarios, including unfamiliar flight paths and varying payloads, velocities, and wind disturbances.
comment: Accepted to the International Conference on Robotics and Automation (ICRA) 2025
♻ ☆ SemGauss-SLAM: Dense Semantic Gaussian Splatting SLAM IROS 2025
We propose SemGauss-SLAM, a dense semantic SLAM system utilizing 3D Gaussian representation, that enables accurate 3D semantic mapping, robust camera tracking, and high-quality rendering simultaneously. In this system, we incorporate semantic feature embedding into 3D Gaussian representation, which effectively encodes semantic information within the spatial layout of the environment for precise semantic scene representation. Furthermore, we propose feature-level loss for updating 3D Gaussian representation, enabling higher-level guidance for 3D Gaussian optimization. In addition, to reduce cumulative drift in tracking and improve semantic reconstruction accuracy, we introduce semantic-informed bundle adjustment. By leveraging multi-frame semantic associations, this strategy enables joint optimization of 3D Gaussian representation and camera poses, resulting in low-drift tracking and accurate semantic mapping. Our SemGauss-SLAM demonstrates superior performance over existing radiance field-based SLAM methods in terms of mapping and tracking accuracy on Replica and ScanNet datasets, while also showing excellent capabilities in high-precision semantic segmentation and dense semantic mapping.
comment: IROS 2025
♻ ☆ Fully distributed and resilient source seeking for robot swarms
We propose a self-contained, resilient and fully distributed solution for locating the maximum of an unknown scalar field using a swarm of robots that travel at a constant speed. Unlike conventional reactive methods relying on gradient information, our methodology enables the swarm to determine an ascending direction so that it approaches the source with an arbitrary precision. Our source-seeking solution consists of three distributed algorithms running simultaneously in a slow-fast closed-loop system. The fastest algorithm provides the centroid-relative coordinates of the robots and the next slower one provides the ascending direction to be tracked. The tracking of the ascending direction by single integrators is instantaneous; howeverin this paper we will also focus on 2D unicycle-like robots with a constant speed. The third algorithm, the slowest one since the speed of the robots can be chosen arbitrarily slow, is the individual control law for the unicycle to track the estimated ascending direction.We will show that the three distributed algorithms converge exponentially fast to their objectives, allowing for a feasible slow-fast closed-loop system. The robots are not constrained to any particular geometric formation, and we study both discrete and continuous distributions of robots.The swarm shape analysis reveals the resiliency of our approach as expected in robot swarms, i.e., by amassing robots we ensure the source-seeking functionality in the event of missing or misplaced individuals or even if the robot network splits in two or more disconnected subnetworks.We exploit such an analysis so that the swarm can adapt to unknown environments by morphing its shape and maneuvering while still following an ascending direction. We analyze our solution with robots as kinematic points in n-dimensional Euclidean spaces and extend the analysis to 2D unicycle-like robots with constant speeds.
comment: 16 pages, submitted version to T-TAC. Jesus Bautista and Antonio Acuaviva contributed equally to this work. arXiv admin note: text overlap with arXiv:2309.02937
♻ ☆ AntiGrounding: Lifting Robotic Actions into VLM Representation Space for Decision Making NeurIPS 2025
Vision-Language Models (VLMs) encode knowledge and reasoning capabilities for robotic manipulation within high-dimensional representation spaces. However, current approaches often project them into compressed intermediate representations, discarding important task-specific information such as fine-grained spatial or semantic details. To address this, we propose AntiGrounding, a new framework that reverses the instruction grounding process. It lifts candidate actions directly into the VLM representation space, renders trajectories from multiple views, and uses structured visual question answering for instruction-based decision making. This enables zero-shot synthesis of optimal closed-loop robot trajectories for new tasks. We also propose an offline policy refinement module that leverages past experience to enhance long-term performance. Experiments in both simulation and real-world environments show that our method outperforms baselines across diverse robotic manipulation tasks.
comment: submitted to NeurIPS 2025
♻ ☆ ContactDexNet: Multi-fingered Robotic Hand Grasping in Cluttered Environments through Hand-object Contact Semantic Mapping
The deep learning models has significantly advanced dexterous manipulation techniques for multi-fingered hand grasping. However, the contact information-guided grasping in cluttered environments remains largely underexplored. To address this gap, we have developed a method for generating multi-fingered hand grasp samples in cluttered settings through contact semantic map. We introduce a contact semantic conditional variational autoencoder network (CoSe-CVAE) for creating comprehensive contact semantic map from object point cloud. We utilize grasp detection method to estimate hand grasp poses from the contact semantic map. Finally, an unified grasp evaluation model PointNetGPD++ is designed to assess grasp quality and collision probability, substantially improving the reliability of identifying optimal grasps in cluttered scenarios. Our grasp generation method has demonstrated remarkable success, outperforming state-of-the-art methods by at least 4.65% with 81.0% average grasping success rate in real-world single-object environment and 75.3% grasping success rate in cluttered scenes. We also proposed the multi-modal multi-fingered grasping dataset generation method. Our multi-fingered hand grasping dataset outperforms previous datasets in scene diversity, modality diversity. The dataset, code and supplementary materials can be found at https://sites.google.com/view/contact-dexnet.
comment: 8 pages
♻ ☆ Perspective-Shifted Neuro-Symbolic World Models: A Framework for Socially-Aware Robot Navigation
Navigating in environments alongside humans requires agents to reason under uncertainty and account for the beliefs and intentions of those around them. Under a sequential decision-making framework, egocentric navigation can naturally be represented as a Markov Decision Process (MDP). However, social navigation additionally requires reasoning about the hidden beliefs of others, inherently leading to a Partially Observable Markov Decision Process (POMDP), where agents lack direct access to others' mental states. Inspired by Theory of Mind and Epistemic Planning, we propose (1) a neuro-symbolic model-based reinforcement learning architecture for social navigation, addressing the challenge of belief tracking in partially observable environments; and (2) a perspective-shift operator for belief estimation, leveraging recent work on Influence-based Abstractions (IBA) in structured multi-agent settings.
comment: Accepted as a regular paper at the 2025 IEEE International Conference on Robot & Human Interactive Communication (RO-MAN). \c{opyright} 2025 IEEE. The final version will appear in IEEE Xplore (DOI TBD)
♻ ☆ Pseudo-Kinematic Trajectory Control and Planning of Tracked Vehicles
Tracked vehicles distribute their weight continuously over a large surface area (the tracks). This distinctive feature makes them the preferred choice for vehicles required to traverse soft and uneven terrain. From a robotics perspective, however, this flexibility comes at a cost: the complexity of modelling the system and the resulting difficulty in designing theoretically sound navigation solutions. In this paper, we aim to bridge this gap by proposing a framework for the navigation of tracked vehicles, built upon three key pillars. The first pillar comprises two models: a simulation model and a control-oriented model. The simulation model captures the intricate terramechanics dynamics arising from soil-track interaction and is employed to develop faithful digital twins of the system across a wide range of operating conditions. The control-oriented model is pseudo-kinematic and mathematically tractable, enabling the design of efficient and theoretically robust control schemes. The second pillar is a Lyapunov-based feedback trajectory controller that provides certifiable tracking guarantees. The third pillar is a portfolio of motion planning solutions, each offering different complexity-accuracy trade-offs. The various components of the proposed approach are validated through an extensive set of simulation and experimental data.
♻ ☆ Help or Hindrance: Understanding the Impact of Robot Communication in Action Teams
The human-robot interaction (HRI) field has recognized the importance of enabling robots to interact with teams. Human teams rely on effective communication for successful collaboration in time-sensitive environments. Robots can play a role in enhancing team coordination through real-time assistance. Despite significant progress in human-robot teaming research, there remains an essential gap in how robots can effectively communicate with action teams using multimodal interaction cues in time-sensitive environments. This study addresses this knowledge gap in an experimental in-lab study to investigate how multimodal robot communication in action teams affects workload and human perception of robots. We explore team collaboration in a medical training scenario where a robotic crash cart (RCC) provides verbal and non-verbal cues to help users remember to perform iterative tasks and search for supplies. Our findings show that verbal cues for object search tasks and visual cues for task reminders reduce team workload and increase perceived ease of use and perceived usefulness more effectively than a robot with no feedback. Our work contributes to multimodal interaction research in the HRI field, highlighting the need for more human-robot teaming research to understand best practices for integrating collaborative robots in time-sensitive environments such as in hospitals, search and rescue, and manufacturing applications.
comment: This is the author's original submitted version of the paper accepted to the 2025 IEEE International Conference on Robot and Human Interactive Communication (RO-MAN). \c{opyright} 2025 IEEE. Personal use of this material is permitted. For any other use, please contact IEEE
♻ ☆ Human-Robot Teaming Field Deployments: A Comparison Between Verbal and Non-verbal Communication
Healthcare workers (HCWs) encounter challenges in hospitals, such as retrieving medical supplies quickly from crash carts, which could potentially result in medical errors and delays in patient care. Robotic crash carts (RCCs) have shown promise in assisting healthcare teams during medical tasks through guided object searches and task reminders. Limited exploration has been done to determine what communication modalities are most effective and least disruptive to patient care in real-world settings. To address this gap, we conducted a between-subjects experiment comparing the RCC's verbal and non-verbal communication of object search with a standard crash cart in resuscitation scenarios to understand the impact of robot communication on workload and attitudes toward using robots in the workplace. Our findings indicate that verbal communication significantly reduced mental demand and effort compared to visual cues and with a traditional crash cart. Although frustration levels were slightly higher during collaborations with the robot compared to a traditional cart, these research insights provide valuable implications for human-robot teamwork in high-stakes environments.
comment: This is the author's original submitted version of the paper accepted to the 2025 IEEE International Conference on Robot and Human Interactive Communication (RO-MAN). \c{opyright} 2025 IEEE. Personal use of this material is permitted. For any other use, please contact IEEE
♻ ☆ TeViR: Text-to-Video Reward with Diffusion Models for Efficient Reinforcement Learning
Developing scalable and generalizable reward engineering for reinforcement learning (RL) is crucial for creating general-purpose agents, especially in the challenging domain of robotic manipulation. While recent advances in reward engineering with Vision-Language Models (VLMs) have shown promise, their sparse reward nature significantly limits sample efficiency. This paper introduces TeViR, a novel method that leverages a pre-trained text-to-video diffusion model to generate dense rewards by comparing the predicted image sequence with current observations. Experimental results across 11 complex robotic tasks demonstrate that TeViR outperforms traditional methods leveraging sparse rewards and other state-of-the-art (SOTA) methods, achieving better sample efficiency and performance without ground truth environmental rewards. TeViR's ability to efficiently guide agents in complex environments highlights its potential to advance reinforcement learning applications in robotic manipulation.
♻ ☆ DynNPC: Finding More Violations Induced by ADS in Simulation Testing through Dynamic NPC Behavior Generation
Recently, a number of simulation testing approaches have been proposed to generate diverse driving scenarios for autonomous driving systems (ADSs) testing. However, the behaviors of NPC vehicles in these scenarios generated by previous approaches are predefined and mutated before simulation execution, ignoring traffic signals and the behaviors of the Ego vehicle. Thus, a large number of the violations they found are induced by unrealistic behaviors of NPC vehicles, revealing no bugs of ADSs. Besides, the vast scenario search space of NPC behaviors during the iterative mutations limits the efficiency of previous approaches. To address these limitations, we propose a novel scenario-based testing framework, DynNPC, to generate more violation scenarios induced by the ADS. Specifically, DynNPC allows NPC vehicles to dynamically generate behaviors using different driving strategies during simulation execution based on traffic signals and the real-time behavior of the Ego vehicle. We compare DynNPC with five state-of-the-art scenario-based testing approaches. Our evaluation has demonstrated the effectiveness and efficiency of DynNPC in finding more violation scenarios induced by the ADS.
♻ ☆ Overlap-Aware Feature Learning for Robust Unsupervised Domain Adaptation for 3D Semantic Segmentation IROS 2025
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: This paper has been accepted to the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025)
♻ ☆ Learning Accurate Whole-body Throwing with High-frequency Residual Policy and Pullback Tube Acceleration IROS 2025
Throwing is a fundamental skill that enables robots to manipulate objects in ways that extend beyond the reach of their arms. We present a control framework that combines learning and model-based control for prehensile whole-body throwing with legged mobile manipulators. Our framework consists of three components: a nominal tracking policy for the end-effector, a high-frequency residual policy to enhance tracking accuracy, and an optimization-based module to improve end-effector acceleration control. The proposed controller achieved the average of 0.28 m landing error when throwing at targets located 6 m away. Furthermore, in a comparative study with university students, the system achieved a velocity tracking error of 0.398 m/s and a success rate of 56.8%, hitting small targets randomly placed at distances of 3-5 m while throwing at a specified speed of 6 m/s. In contrast, humans have a success rate of only 15.2%. This work provides an early demonstration of prehensile throwing with quantified accuracy on hardware, contributing to progress in dynamic whole-body manipulation.
comment: 8 pages, IROS 2025
Artificial Intelligence 11
☆ Beyond Autocomplete: Designing CopilotLens Towards Transparent and Explainable AI Coding Agents
AI-powered code assistants are widely used to generate code completions, significantly boosting developer productivity. However, these tools typically present suggestions without explaining their rationale, leaving their decision-making process inscrutable. This opacity hinders developers' ability to critically evaluate the output, form accurate mental models, and build calibrated trust in the system. To address this, we introduce CopilotLens, a novel interactive framework that reframes code completion from a simple suggestion into a transparent, explainable event. CopilotLens operates as an explanation layer that reveals the AI agent's "thought process" through a dynamic two-level interface, surfacing everything from its reconstructed high-level plans to the specific codebase context influencing the code. This paper presents the design and rationale of CopilotLens, offering a concrete framework for building future agentic code assistants that prioritize clarity of reasoning over speed of suggestion, thereby fostering deeper comprehension and more robust human-AI collaboration.
☆ DiaLLMs: EHR Enhanced Clinical Conversational System for Clinical Test Recommendation and Diagnosis Prediction
Recent advances in Large Language Models (LLMs) have led to remarkable progresses in medical consultation. However, existing medical LLMs overlook the essential role of Electronic Health Records (EHR) and focus primarily on diagnosis recommendation, limiting their clinical applicability. We propose DiaLLM, the first medical LLM that integrates heterogeneous EHR data into clinically grounded dialogues, enabling clinical test recommendation, result interpretation, and diagnosis prediction to better align with real-world medical practice. To construct clinically grounded dialogues from EHR, we design a Clinical Test Reference (CTR) strategy that maps each clinical code to its corresponding description and classifies test results as "normal" or "abnormal". Additionally, DiaLLM employs a reinforcement learning framework for evidence acquisition and automated diagnosis. To handle the large action space, we introduce a reject sampling strategy to reduce redundancy and improve exploration efficiency. Furthermore, a confirmation reward and a class-sensitive diagnosis reward are designed to guide accurate diagnosis prediction. Extensive experimental results demonstrate that DiaLLM outperforms baselines in clinical test recommendation and diagnosis prediction.
☆ Robust Robotic Exploration and Mapping Using Generative Occupancy Map Synthesis
We present a novel approach for enhancing robotic exploration by using generative occupancy mapping. We introduce SceneSense, a diffusion model designed and trained for predicting 3D occupancy maps given partial observations. Our proposed approach probabilistically fuses these predictions into a running occupancy map in real-time, resulting in significant improvements in map quality and traversability. We implement SceneSense onboard a quadruped robot and validate its performance with real-world experiments to demonstrate the effectiveness of the model. In these experiments, we show that occupancy maps enhanced with SceneSense predictions better represent our fully observed ground truth data (24.44% FID improvement around the robot and 75.59% improvement at range). We additionally show that integrating SceneSense-enhanced maps into our robotic exploration stack as a "drop-in" map improvement, utilizing an existing off-the-shelf planner, results in improvements in robustness and traversability time. Finally we show results of full exploration evaluations with our proposed system in two dissimilar environments and find that locally enhanced maps provide more consistent exploration results than maps constructed only from direct sensor measurements.
comment: arXiv admin note: text overlap with arXiv:2409.10681
☆ GNN's Uncertainty Quantification using Self-Distillation
Graph Neural Networks (GNNs) have shown remarkable performance in the healthcare domain. However, what remained challenging is quantifying the predictive uncertainty of GNNs, which is an important aspect of trustworthiness in clinical settings. While Bayesian and ensemble methods can be used to quantify uncertainty, they are computationally expensive. Additionally, the disagreement metric used by ensemble methods to compute uncertainty cannot capture the diversity of models in an ensemble network. In this paper, we propose a novel method, based on knowledge distillation, to quantify GNNs' uncertainty more efficiently and with higher precision. We apply self-distillation, where the same network serves as both the teacher and student models, thereby avoiding the need to train several networks independently. To ensure the impact of self-distillation, we develop an uncertainty metric that captures the diverse nature of the network by assigning different weights to each GNN classifier. We experimentally evaluate the precision, performance, and ability of our approach in distinguishing out-of-distribution data on two graph datasets: MIMIC-IV and Enzymes. The evaluation results demonstrate that the proposed method can effectively capture the predictive uncertainty of the model while having performance similar to that of the MC Dropout and ensemble methods. The code is publicly available at https://github.com/tailabTMU/UQ_GNN.
comment: The paper has been accepted in the International Conference on AI in Healthcare (AIiH) 2025 and will appear in the conference proceedings
☆ LSH-DynED: A Dynamic Ensemble Framework with LSH-Based Undersampling for Evolving Multi-Class Imbalanced Classification
The classification of imbalanced data streams, which have unequal class distributions, is a key difficulty in machine learning, especially when dealing with multiple classes. While binary imbalanced data stream classification tasks have received considerable attention, only a few studies have focused on multi-class imbalanced data streams. Effectively managing the dynamic imbalance ratio is a key challenge in this domain. This study introduces a novel, robust, and resilient approach to address these challenges by integrating Locality Sensitive Hashing with Random Hyperplane Projections (LSH-RHP) into the Dynamic Ensemble Diversification (DynED) framework. To the best of our knowledge, we present the first application of LSH-RHP for undersampling in the context of imbalanced non-stationary data streams. The proposed method undersamples the majority classes by utilizing LSH-RHP, provides a balanced training set, and improves the ensemble's prediction performance. We conduct comprehensive experiments on 23 real-world and ten semi-synthetic datasets and compare LSH-DynED with 15 state-of-the-art methods. The results reveal that LSH-DynED outperforms other approaches in terms of both Kappa and mG-Mean effectiveness measures, demonstrating its capability in dealing with multi-class imbalanced non-stationary data streams. Notably, LSH-DynED performs well in large-scale, high-dimensional datasets with considerable class imbalances and demonstrates adaptation and robustness in real-world circumstances. To motivate our design, we review existing methods for imbalanced data streams, outline key challenges, and offer guidance for future work. For the reproducibility of our results, we have made our implementation available on GitHub.
☆ Cross-Layer Discrete Concept Discovery for Interpreting Language Models
Uncovering emergent concepts across transformer layers remains a significant challenge because the residual stream linearly mixes and duplicates information, obscuring how features evolve within large language models. Current research efforts primarily inspect neural representations at single layers, thereby overlooking this cross-layer superposition and the redundancy it introduces. These representations are typically either analyzed directly for activation patterns or passed to probing classifiers that map them to a limited set of predefined concepts. To address these limitations, we propose \gls{clvqvae}, a framework that uses vector quantization to map representations across layers and in the process collapse duplicated residual-stream features into compact, interpretable concept vectors. Our approach uniquely combines top-$k$ temperature-based sampling during quantization with EMA codebook updates, providing controlled exploration of the discrete latent space while maintaining code-book diversity. We further enhance the framework with scaled-spherical k-means++ for codebook initialization, which clusters by directional similarity rather than magnitude, better aligning with semantic structure in word embedding space.
☆ Learning Bilateral Team Formation in Cooperative Multi-Agent Reinforcement Learning
Team formation and the dynamics of team-based learning have drawn significant interest in the context of Multi-Agent Reinforcement Learning (MARL). However, existing studies primarily focus on unilateral groupings, predefined teams, or fixed-population settings, leaving the effects of algorithmic bilateral grouping choices in dynamic populations underexplored. To address this gap, we introduce a framework for learning two-sided team formation in dynamic multi-agent systems. Through this study, we gain insight into what algorithmic properties in bilateral team formation influence policy performance and generalization. We validate our approach using widely adopted multi-agent scenarios, demonstrating competitive performance and improved generalization in most scenarios.
comment: Accepted to the 2nd Coordination and Cooperation in Multi-Agent Reinforcement Learning (CoCoMARL) Workshop at RLC 2025
☆ Hierarchical Reinforcement Learning and Value Optimization for Challenging Quadruped Locomotion
We propose a novel hierarchical reinforcement learning framework for quadruped locomotion over challenging terrain. Our approach incorporates a two-layer hierarchy in which a high-level policy (HLP) selects optimal goals for a low-level policy (LLP). The LLP is trained using an on-policy actor-critic RL algorithm and is given footstep placements as goals. We propose an HLP that does not require any additional training or environment samples and instead operates via an online optimization process over the learned value function of the LLP. We demonstrate the benefits of this framework by comparing it with an end-to-end reinforcement learning (RL) approach. We observe improvements in its ability to achieve higher rewards with fewer collisions across an array of different terrains, including terrains more difficult than any encountered during training.
☆ Automated Generation of Diverse Courses of Actions for Multi-Agent Operations using Binary Optimization and Graph Learning
Operations in disaster response, search \& rescue, and military missions that involve multiple agents demand automated processes to support the planning of the courses of action (COA). Moreover, traverse-affecting changes in the environment (rain, snow, blockades, etc.) may impact the expected performance of a COA, making it desirable to have a pool of COAs that are diverse in task distributions across agents. Further, variations in agent capabilities, which could be human crews and/or autonomous systems, present practical opportunities and computational challenges to the planning process. This paper presents a new theoretical formulation and computational framework to generate such diverse pools of COAs for operations with soft variations in agent-task compatibility. Key to the problem formulation is a graph abstraction of the task space and the pool of COAs itself to quantify its diversity. Formulating the COAs as a centralized multi-robot task allocation problem, a genetic algorithm is used for (order-ignoring) allocations of tasks to each agent that jointly maximize diversity within the COA pool and overall compatibility of the agent-task mappings. A graph neural network is trained using a policy gradient approach to then perform single agent task sequencing in each COA, which maximizes completion rates adaptive to task features. Our tests of the COA generation process in a simulated environment demonstrate significant performance gain over a random walk baseline, small optimality gap in task sequencing, and execution time of about 50 minutes to plan up to 20 COAs for 5 agent/100 task operations.
♻ ☆ The Alignment Trap: Complexity Barriers
This paper argues that AI alignment is not merely difficult, but is founded on a fundamental logical contradiction. We first establish The Enumeration Paradox: we use machine learning precisely because we cannot enumerate all necessary safety rules, yet making ML safe requires examples that can only be generated from the very enumeration we admit is impossible. This paradox is then confirmed by a set of five independent mathematical proofs, or "pillars of impossibility." Our main results show that: (1) Geometric Impossibility: The set of safe policies has measure zero, a necessary consequence of projecting infinite-dimensional world-context requirements onto finite-dimensional models. (2) Computational Impossibility: Verifying a policy's safety is coNP-complete, even for non-zero error tolerances. (3) Statistical Impossibility: The training data required for safety (abundant examples of rare disasters) is a logical contradiction and thus unobtainable. (4) Information-Theoretic Impossibility: Safety rules contain more incompressible, arbitrary information than any feasible network can store. (5) Dynamic Impossibility: The optimization process for increasing AI capability is actively hostile to safety, as the gradients for the two objectives are generally anti-aligned. Together, these results demonstrate that the pursuit of safe, highly capable AI is not a matter of overcoming technical hurdles, but of confronting fundamental, interlocking barriers. The paper concludes by presenting a strategic trilemma that these impossibilities force upon the field. A formal verification of the core theorems in Lean4 is currently in progress.
comment: 31 Pages, 4 Figures. Substantial revision. Restructured around the Enumeration Paradox and Five Pillars of Impossibility. Core mathematical results unchanged but significantly expanded. Added new impossibility proofs from statistical, information-theoretic, and dynamic perspectives
♻ ☆ Evaluating Long Range Dependency Handling in Code Generation LLMs
As language models support larger and larger context sizes, evaluating their ability to make effective use of that context becomes increasingly important. We analyze the ability of several code generation models to handle long range dependencies using a suite of multi-step key retrieval tasks in context windows up to 8k tokens in length. The tasks progressively increase in difficulty and allow more nuanced evaluation of model capabilities than tests like the popular needle-in-the-haystack test. We find that performance degrades significantly for many models (up to 2x) when a function references another function that is defined later in the prompt. We also observe that models that use sliding window attention mechanisms have difficulty handling references further than the size of a single window. We perform simple prompt modifications using call graph information to improve multi-step retrieval performance up to 3x. Our analysis highlights ways that long-context performance needs deeper consideration beyond retrieval of single facts within a document.
comment: 36 pages, 18 figures
Computer Vision and Pattern Recognition 37
☆ ToSA: Token Merging with Spatial Awareness IROS 2025
Token merging has emerged as an effective strategy to accelerate Vision Transformers (ViT) by reducing computational costs. However, existing methods primarily rely on the visual token's feature similarity for token merging, overlooking the potential of integrating spatial information, which can serve as a reliable criterion for token merging in the early layers of ViT, where the visual tokens only possess weak visual information. In this paper, we propose ToSA, a novel token merging method that combines both semantic and spatial awareness to guide the token merging process. ToSA leverages the depth image as input to generate pseudo spatial tokens, which serve as auxiliary spatial information for the visual token merging process. With the introduced spatial awareness, ToSA achieves a more informed merging strategy that better preserves critical scene structure. Experimental results demonstrate that ToSA outperforms previous token merging methods across multiple benchmarks on visual and embodied question answering while largely reducing the runtime of the ViT, making it an efficient solution for ViT acceleration. The code will be available at: https://github.com/hsiangwei0903/ToSA
comment: Accepted by IROS 2025
☆ Consensus-Driven Uncertainty for Robotic Grasping based on RGB Perception
Deep object pose estimators are notoriously overconfident. A grasping agent that both estimates the 6-DoF pose of a target object and predicts the uncertainty of its own estimate could avoid task failure by choosing not to act under high uncertainty. Even though object pose estimation improves and uncertainty quantification research continues to make strides, few studies have connected them to the downstream task of robotic grasping. We propose a method for training lightweight, deep networks to predict whether a grasp guided by an image-based pose estimate will succeed before that grasp is attempted. We generate training data for our networks via object pose estimation on real images and simulated grasping. We also find that, despite high object variability in grasping trials, networks benefit from training on all objects jointly, suggesting that a diverse variety of objects can nevertheless contribute to the same goal.
☆ VoxelOpt: Voxel-Adaptive Message Passing for Discrete Optimization in Deformable Abdominal CT Registration MICCAI 2025
Recent developments in neural networks have improved deformable image registration (DIR) by amortizing iterative optimization, enabling fast and accurate DIR results. However, learning-based methods often face challenges with limited training data, large deformations, and tend to underperform compared to iterative approaches when label supervision is unavailable. While iterative methods can achieve higher accuracy in such scenarios, they are considerably slower than learning-based methods. To address these limitations, we propose VoxelOpt, a discrete optimization-based DIR framework that combines the strengths of learning-based and iterative methods to achieve a better balance between registration accuracy and runtime. VoxelOpt uses displacement entropy from local cost volumes to measure displacement signal strength at each voxel, which differs from earlier approaches in three key aspects. First, it introduces voxel-wise adaptive message passing, where voxels with lower entropy receives less influence from their neighbors. Second, it employs a multi-level image pyramid with 27-neighbor cost volumes at each level, avoiding exponential complexity growth. Third, it replaces hand-crafted features or contrastive learning with a pretrained foundational segmentation model for feature extraction. In abdominal CT registration, these changes allow VoxelOpt to outperform leading iterative in both efficiency and accuracy, while matching state-of-the-art learning-based methods trained with label supervision. The source code will be available at https://github.com/tinymilky/VoxelOpt
comment: Accepted for publication at MICCAI 2025
☆ EBC-ZIP: Improving Blockwise Crowd Counting with Zero-Inflated Poisson Regression
Density map estimation has become the mainstream paradigm in crowd counting. However, most existing methods overlook the extreme sparsity of ground-truth density maps. In real-world crowd scenes, the vast majority of spatial regions (often over 95%) contain no people, leading to heavily imbalanced count distributions. Ignoring this imbalance can bias models toward overestimating dense regions and underperforming in sparse areas. Furthermore, most loss functions used in density estimation are majorly based on MSE and implicitly assume Gaussian distributions, which are ill-suited for modeling discrete, non-negative count data. In this paper, we propose EBC-ZIP, a crowd counting framework that models the spatial distribution of counts using a Zero-Inflated Poisson (ZIP) regression formulation. Our approach replaces the traditional regression loss with the negative log-likelihood of the ZIP distribution, enabling better handling of zero-heavy distributions while preserving count accuracy. Built upon the recently proposed Enhanced Block Classification (EBC) framework, EBC-ZIP inherits EBC's advantages in preserving the discreteness of targets and ensuring training stability, while further improving performance through a more principled probabilistic loss. We also evaluate EBC-ZIP with backbones of varying computational complexity to assess its scalability. Extensive experiments on four crowd counting benchmarks demonstrate that EBC-ZIP consistently outperforms EBC and achieves state-of-the-art results.
☆ Computer Vision based Automated Quantification of Agricultural Sprayers Boom Displacement
Application rate errors when using self-propelled agricultural sprayers for agricultural production remain a concern. Among other factors, spray boom instability is one of the major contributors to application errors. Spray booms' width of 38m, combined with 30 kph driving speeds, varying terrain, and machine dynamics when maneuvering complex field boundaries, make controls of these booms very complex. However, there is no quantitative knowledge on the extent of boom movement to systematically develop a solution that might include boom designs and responsive boom control systems. Therefore, this study was conducted to develop an automated computer vision system to quantify the boom movement of various agricultural sprayers. A computer vision system was developed to track a target on the edge of the sprayer boom in real time. YOLO V7, V8, and V11 neural network models were trained to track the boom's movements in field operations to quantify effective displacement in the vertical and transverse directions. An inclinometer sensor was mounted on the boom to capture boom angles and validate the neural network model output. The results showed that the model could detect the target with more than 90 percent accuracy, and distance estimates of the target on the boom were within 0.026 m of the inclinometer sensor data. This system can quantify the boom movement on the current sprayer and potentially on any other sprayer with minor modifications. The data can be used to make design improvements to make sprayer booms more stable and achieve greater application accuracy.
comment: Under publication process for COMPAG
☆ Any-Order GPT as Masked Diffusion Model: Decoupling Formulation and Architecture
Large language models (LLMs) predominantly use autoregressive (AR) approaches, but masked diffusion models (MDMs) are emerging as viable alternatives. A key challenge in comparing AR and MDM paradigms is their typical architectural difference: AR models are often decoder-only, while MDMs have largely been encoder-only. This practice of changing both the modeling paradigm and architecture simultaneously makes direct comparisons unfair, as it's hard to distinguish whether observed differences stem from the paradigm itself or the architectural shift. This research evaluates MDMs within a decoder-only framework to: (1) equitably compare MDM (as Any-Order AR, or AO-AR) and standard AR paradigms. Our investigation suggests that the standard AO-AR objective, which averages over all token permutations, may benefit from refinement, as many permutations appear less informative compared to the language's inherent left-to-right structure. (2) Investigate architectural influences (decoder-only vs. encoder-only) within MDMs. We demonstrate that while encoder-only MDMs model a simpler conditional probability space, decoder-only MDMs can achieve dramatic generation speedups ($\sim25\times$) and comparable perplexity with temperature annealing despite modeling a vastly larger space, highlighting key trade-offs. This work thus decouples core paradigm differences from architectural influences, offering insights for future model design. Code is available at https://github.com/scxue/AO-GPT-MDM.
☆ Radial Attention: $O(n\log n)$ Sparse Attention with Energy Decay for Long Video Generation
Recent advances in diffusion models have enabled high-quality video generation, but the additional temporal dimension significantly increases computational costs, making training and inference on long videos prohibitively expensive. In this paper, we identify a phenomenon we term Spatiotemporal Energy Decay in video diffusion models: post-softmax attention scores diminish as spatial and temporal distance between tokens increase, akin to the physical decay of signal or waves over space and time in nature. Motivated by this, we propose Radial Attention, a scalable sparse attention mechanism with $O(n \log n)$ complexity that translates energy decay into exponentially decaying compute density, which is significantly more efficient than standard $O(n^2)$ dense attention and more expressive than linear attention. Specifically, Radial Attention employs a simple, static attention mask where each token attends to spatially nearby tokens, with the attention window size shrinking with temporal distance. Moreover, it allows pre-trained video diffusion models to extend their generation length with efficient LoRA-based fine-tuning. Extensive experiments show that Radial Attention maintains video quality across Wan2.1-14B, HunyuanVideo, and Mochi 1, achieving up to a 1.9$\times$ speedup over the original dense attention. With minimal tuning, it enables video generation up to 4$\times$ longer while reducing training costs by up to 4.4$\times$ compared to direct fine-tuning and accelerating inference by up to 3.7$\times$ compared to dense attention inference.
comment: Code: https://github.com/mit-han-lab/radial-attention
☆ AnimaX: Animating the Inanimate in 3D with Joint Video-Pose Diffusion Models
We present AnimaX, a feed-forward 3D animation framework that bridges the motion priors of video diffusion models with the controllable structure of skeleton-based animation. Traditional motion synthesis methods are either restricted to fixed skeletal topologies or require costly optimization in high-dimensional deformation spaces. In contrast, AnimaX effectively transfers video-based motion knowledge to the 3D domain, supporting diverse articulated meshes with arbitrary skeletons. Our method represents 3D motion as multi-view, multi-frame 2D pose maps, and enables joint video-pose diffusion conditioned on template renderings and a textual motion prompt. We introduce shared positional encodings and modality-aware embeddings to ensure spatial-temporal alignment between video and pose sequences, effectively transferring video priors to motion generation task. The resulting multi-view pose sequences are triangulated into 3D joint positions and converted into mesh animation via inverse kinematics. Trained on a newly curated dataset of 160,000 rigged sequences, AnimaX achieves state-of-the-art results on VBench in generalization, motion fidelity, and efficiency, offering a scalable solution for category-agnostic 3D animation. Project page: \href{https://anima-x.github.io/}{https://anima-x.github.io/}.
comment: Project page: https://anima-x.github.io/
☆ Unified Vision-Language-Action Model
Vision-language-action models (VLAs) have garnered significant attention for their potential in advancing robotic manipulation. However, previous approaches predominantly rely on the general comprehension capabilities of vision-language models (VLMs) to generate action signals, often overlooking the rich temporal and causal structure embedded in visual observations. In this paper, we present UniVLA, a unified and native multimodal VLA model that autoregressively models vision, language, and action signals as discrete token sequences. This formulation enables flexible multimodal tasks learning, particularly from large-scale video data. By incorporating world modeling during post-training, UniVLA captures causal dynamics from videos, facilitating effective transfer to downstream policy learning--especially for long-horizon tasks. Our approach sets new state-of-the-art results across several widely used simulation benchmarks, including CALVIN, LIBERO, and Simplenv-Bridge, significantly surpassing previous methods. For example, UniVLA achieves 95.5% average success rate on LIBERO benchmark, surpassing pi0-FAST's 85.5%. We further demonstrate its broad applicability on real-world ALOHA manipulation and autonomous driving.
comment: technical report
☆ ScaleCap: Inference-Time Scalable Image Captioning via Dual-Modality Debiasing
This paper presents ScaleCap, an inference-time scalable image captioning strategy that generates comprehensive and detailed image captions. The key challenges of high-quality image captioning lie in the inherent biases of LVLMs: multimodal bias resulting in imbalanced descriptive granularity, offering detailed accounts of some elements while merely skimming over others; linguistic bias leading to hallucinated descriptions of non-existent objects. To address these issues, we propose a scalable debiased captioning strategy, which continuously enriches and calibrates the caption with increased inference budget. Specifically, we propose two novel components: heuristic question answering and contrastive sentence rating. The former generates content-specific questions based on the image and answers them to progressively inject relevant information into the caption. The latter employs sentence-level offline contrastive decoding to effectively identify and eliminate hallucinations caused by linguistic biases. With increased inference cost, more heuristic questions are raised by ScaleCap to progressively capture additional visual details, generating captions that are more accurate, balanced, and informative. Extensive modality alignment experiments demonstrate the effectiveness of ScaleCap. Annotating 450K images with ScaleCap and using them for LVLM pretraining leads to consistent performance gains across 11 widely used benchmarks. Furthermore, ScaleCap showcases superb richness and fidelity of generated captions with two additional tasks: replacing images with captions in VQA task, and reconstructing images from captions to assess semantic coverage. Code is available at https://github.com/Cooperx521/ScaleCap.
comment: Code is available at https://github.com/Cooperx521/ScaleCap
☆ Orthogonal Finetuning Made Scalable
Orthogonal finetuning (OFT) offers highly parameter-efficient adaptation while preventing catastrophic forgetting, but its high runtime and memory demands limit practical deployment. We identify the core computational bottleneck in OFT as its weight-centric implementation, which relies on costly matrix-matrix multiplications with cubic complexity. To overcome this, we propose OFTv2, an input-centric reformulation that instead uses matrix-vector multiplications (i.e., matrix-free computation), reducing the computational cost to quadratic. We further introduce the Cayley-Neumann parameterization, an efficient orthogonal parameterization that approximates the matrix inversion in Cayley transform via a truncated Neumann series. These modifications allow OFTv2 to achieve up to 10x faster training and 3x lower GPU memory usage without compromising performance. In addition, we extend OFTv2 to support finetuning quantized foundation models and show that it outperforms the popular QLoRA in training stability, efficiency, and memory usage.
comment: Technical report (17 pages, 7 figures, project page: https://spherelab.ai/oftv2/)
☆ A Comparative Study of NAFNet Baselines for Image Restoration
We study NAFNet (Nonlinear Activation Free Network), a simple and efficient deep learning baseline for image restoration. By using CIFAR10 images corrupted with noise and blur, we conduct an ablation study of NAFNet's core components. Our baseline model implements SimpleGate activation, Simplified Channel Activation (SCA), and LayerNormalization. We compare this baseline to different variants that replace or remove components. Quantitative results (PSNR, SSIM) and examples illustrate how each modification affects restoration performance. Our findings support the NAFNet design: the SimpleGate and simplified attention mechanisms yield better results than conventional activations and attention, while LayerNorm proves to be important for stable training. We conclude with recommendations for model design, discuss potential improvements, and future work.
☆ Active View Selector: Fast and Accurate Active View Selection with Cross Reference Image Quality Assessment
We tackle active view selection in novel view synthesis and 3D reconstruction. Existing methods like FisheRF and ActiveNeRF select the next best view by minimizing uncertainty or maximizing information gain in 3D, but they require specialized designs for different 3D representations and involve complex modelling in 3D space. Instead, we reframe this as a 2D image quality assessment (IQA) task, selecting views where current renderings have the lowest quality. Since ground-truth images for candidate views are unavailable, full-reference metrics like PSNR and SSIM are inapplicable, while no-reference metrics, such as MUSIQ and MANIQA, lack the essential multi-view context. Inspired by a recent cross-referencing quality framework CrossScore, we train a model to predict SSIM within a multi-view setup and use it to guide view selection. Our cross-reference IQA framework achieves substantial quantitative and qualitative improvements across standard benchmarks, while being agnostic to 3D representations, and runs 14-33 times faster than previous methods.
comment: Project page: https://avs.active.vision/
☆ GenHSI: Controllable Generation of Human-Scene Interaction Videos
Large-scale pre-trained video diffusion models have exhibited remarkable capabilities in diverse video generation. However, existing solutions face several challenges in using these models to generate long movie-like videos with rich human-object interactions that include unrealistic human-scene interaction, lack of subject identity preservation, and require expensive training. We propose GenHSI, a training-free method for controllable generation of long human-scene interaction videos (HSI). Taking inspiration from movie animation, our key insight is to overcome the limitations of previous work by subdividing the long video generation task into three stages: (1) script writing, (2) pre-visualization, and (3) animation. Given an image of a scene, a user description, and multiple images of a person, we use these three stages to generate long-videos that preserve human-identity and provide rich human-scene interactions. Script writing converts complex human tasks into simple atomic tasks that are used in the pre-visualization stage to generate 3D keyframes (storyboards). These 3D keyframes are rendered and animated by off-the-shelf video diffusion models for consistent long video generation with rich contacts in a 3D-aware manner. A key advantage of our work is that we alleviate the need for scanned, accurate scenes and create 3D keyframes from single-view images. We are the first to generate a long video sequence with a consistent camera pose that contains arbitrary numbers of character actions without training. Experiments demonstrate that our method can generate long videos that effectively preserve scene content and character identity with plausible human-scene interaction from a single image scene. Visit our project homepage https://kunkun0w0.github.io/project/GenHSI/ for more information.
☆ Improving Progressive Generation with Decomposable Flow Matching
Generating high-dimensional visual modalities is a computationally intensive task. A common solution is progressive generation, where the outputs are synthesized in a coarse-to-fine spectral autoregressive manner. While diffusion models benefit from the coarse-to-fine nature of denoising, explicit multi-stage architectures are rarely adopted. These architectures have increased the complexity of the overall approach, introducing the need for a custom diffusion formulation, decomposition-dependent stage transitions, add-hoc samplers, or a model cascade. Our contribution, Decomposable Flow Matching (DFM), is a simple and effective framework for the progressive generation of visual media. DFM applies Flow Matching independently at each level of a user-defined multi-scale representation (such as Laplacian pyramid). As shown by our experiments, our approach improves visual quality for both images and videos, featuring superior results compared to prior multistage frameworks. On Imagenet-1k 512px, DFM achieves 35.2% improvements in FDD scores over the base architecture and 26.4% over the best-performing baseline, under the same training compute. When applied to finetuning of large models, such as FLUX, DFM shows faster convergence speed to the training distribution. Crucially, all these advantages are achieved with a single model, architectural simplicity, and minimal modifications to existing training pipelines.
comment: Project Webpage: https://snap-research.github.io/dfm/
☆ SimpleGVR: A Simple Baseline for Latent-Cascaded Video Super-Resolution
Latent diffusion models have emerged as a leading paradigm for efficient video generation. However, as user expectations shift toward higher-resolution outputs, relying solely on latent computation becomes inadequate. A promising approach involves decoupling the process into two stages: semantic content generation and detail synthesis. The former employs a computationally intensive base model at lower resolutions, while the latter leverages a lightweight cascaded video super-resolution (VSR) model to achieve high-resolution output. In this work, we focus on studying key design principles for latter cascaded VSR models, which are underexplored currently. First, we propose two degradation strategies to generate training pairs that better mimic the output characteristics of the base model, ensuring alignment between the VSR model and its upstream generator. Second, we provide critical insights into VSR model behavior through systematic analysis of (1) timestep sampling strategies, (2) noise augmentation effects on low-resolution (LR) inputs. These findings directly inform our architectural and training innovations. Finally, we introduce interleaving temporal unit and sparse local attention to achieve efficient training and inference, drastically reducing computational overhead. Extensive experiments demonstrate the superiority of our framework over existing methods, with ablation studies confirming the efficacy of each design choice. Our work establishes a simple yet effective baseline for cascaded video super-resolution generation, offering practical insights to guide future advancements in efficient cascaded synthesis systems.
comment: Project webpage available at https://simplegvr.github.io/
☆ Bind-Your-Avatar: Multi-Talking-Character Video Generation with Dynamic 3D-mask-based Embedding Router
Recent years have witnessed remarkable advances in audio-driven talking head generation. However, existing approaches predominantly focus on single-character scenarios. While some methods can create separate conversation videos between two individuals, the critical challenge of generating unified conversation videos with multiple physically co-present characters sharing the same spatial environment remains largely unaddressed. This setting presents two key challenges: audio-to-character correspondence control and the lack of suitable datasets featuring multi-character talking videos within the same scene. To address these challenges, we introduce Bind-Your-Avatar, an MM-DiT-based model specifically designed for multi-talking-character video generation in the same scene. Specifically, we propose (1) A novel framework incorporating a fine-grained Embedding Router that binds `who' and `speak what' together to address the audio-to-character correspondence control. (2) Two methods for implementing a 3D-mask embedding router that enables frame-wise, fine-grained control of individual characters, with distinct loss functions based on observed geometric priors and a mask refinement strategy to enhance the accuracy and temporal smoothness of the predicted masks. (3) The first dataset, to the best of our knowledge, specifically constructed for multi-talking-character video generation, and accompanied by an open-source data processing pipeline, and (4) A benchmark for the dual-talking-characters video generation, with extensive experiments demonstrating superior performance over multiple state-of-the-art methods.
☆ Look to Locate: Vision-Based Multisensory Navigation with 3-D Digital Maps for GNSS-Challenged Environments
In Global Navigation Satellite System (GNSS)-denied environments such as indoor parking structures or dense urban canyons, achieving accurate and robust vehicle positioning remains a significant challenge. This paper proposes a cost-effective, vision-based multi-sensor navigation system that integrates monocular depth estimation, semantic filtering, and visual map registration (VMR) with 3-D digital maps. Extensive testing in real-world indoor and outdoor driving scenarios demonstrates the effectiveness of the proposed system, achieving sub-meter accuracy of 92% indoors and more than 80% outdoors, with consistent horizontal positioning and heading average root mean-square errors of approximately 0.98 m and 1.25 {\deg}, respectively. Compared to the baselines examined, the proposed solution significantly reduced drift and improved robustness under various conditions, achieving positioning accuracy improvements of approximately 88% on average. This work highlights the potential of cost-effective monocular vision systems combined with 3D maps for scalable, GNSS-independent navigation in land vehicles.
☆ CronusVLA: Transferring Latent Motion Across Time for Multi-Frame Prediction in Manipulation
Recent vision-language-action (VLA) models built on pretrained vision-language models (VLMs) have demonstrated strong generalization across manipulation tasks. However, they remain constrained by a single-frame observation paradigm and cannot fully benefit from the motion information offered by aggregated multi-frame historical observations, as the large vision-language backbone introduces substantial computational cost and inference latency. We propose CronusVLA, a unified framework that extends single-frame VLA models to the multi-frame paradigm through an efficient post-training stage. CronusVLA comprises three key components: (1) single-frame pretraining on large-scale embodied datasets with autoregressive action tokens prediction, which establishes an embodied vision-language foundation; (2) multi-frame encoding, adapting the prediction of vision-language backbones from discrete action tokens to motion features during post-training, and aggregating motion features from historical frames into a feature chunking; (3) cross-frame decoding, which maps the feature chunking to accurate actions via a shared decoder with cross-attention. By reducing redundant token computation and caching past motion features, CronusVLA achieves efficient inference. As an application of motion features, we further propose an action adaptation mechanism based on feature-action retrieval to improve model performance during finetuning. CronusVLA achieves state-of-the-art performance on SimplerEnv with 70.9% success rate, and 12.7% improvement over OpenVLA on LIBERO. Real-world Franka experiments also show the strong performance and robustness.
comment: 36 pages, 21 figures
☆ KnowRL: Exploring Knowledgeable Reinforcement Learning for Factuality
Large Language Models (LLMs), particularly slow-thinking models, often exhibit severe hallucination, outputting incorrect content due to an inability to accurately recognize knowledge boundaries during reasoning. While Reinforcement Learning (RL) can enhance complex reasoning abilities, its outcome-oriented reward mechanism often lacks factual supervision over the thinking process, further exacerbating the hallucination problem. To address the high hallucination in slow-thinking models, we propose Knowledge-enhanced RL, KnowRL. KnowRL guides models to perform fact-based slow thinking by integrating a factuality reward, based on knowledge verification, into the RL training process, helping them recognize their knowledge boundaries. KnowRL guides models to perform fact-based slow thinking by integrating a factuality reward, based on knowledge verification, into the RL training process, helping them recognize their knowledge boundaries. This targeted factual input during RL training enables the model to learn and internalize fact-based reasoning strategies. By directly rewarding adherence to facts within the reasoning steps, KnowRL fosters a more reliable thinking process. Experimental results on three hallucination evaluation datasets and two reasoning evaluation datasets demonstrate that KnowRL effectively mitigates hallucinations in slow-thinking models while maintaining their original strong reasoning capabilities. Our code is available at https://github.com/zjunlp/KnowRL.
comment: Work in progress
☆ CoCo4D: Comprehensive and Complex 4D Scene Generation
Existing 4D synthesis methods primarily focus on object-level generation or dynamic scene synthesis with limited novel views, restricting their ability to generate multi-view consistent and immersive dynamic 4D scenes. To address these constraints, we propose a framework (dubbed as CoCo4D) for generating detailed dynamic 4D scenes from text prompts, with the option to include images. Our method leverages the crucial observation that articulated motion typically characterizes foreground objects, whereas background alterations are less pronounced. Consequently, CoCo4D divides 4D scene synthesis into two responsibilities: modeling the dynamic foreground and creating the evolving background, both directed by a reference motion sequence. Given a text prompt and an optional reference image, CoCo4D first generates an initial motion sequence utilizing video diffusion models. This motion sequence then guides the synthesis of both the dynamic foreground object and the background using a novel progressive outpainting scheme. To ensure seamless integration of the moving foreground object within the dynamic background, CoCo4D optimizes a parametric trajectory for the foreground, resulting in realistic and coherent blending. Extensive experiments show that CoCo4D achieves comparable or superior performance in 4D scene generation compared to existing methods, demonstrating its effectiveness and efficiency. More results are presented on our website https://colezwhy.github.io/coco4d/.
comment: 16 pages,10 figures
☆ Systematic Review of Pituitary Gland and Pituitary Adenoma Automatic Segmentation Techniques in Magnetic Resonance Imaging
Purpose: Accurate segmentation of both the pituitary gland and adenomas from magnetic resonance imaging (MRI) is essential for diagnosis and treatment of pituitary adenomas. This systematic review evaluates automatic segmentation methods for improving the accuracy and efficiency of MRI-based segmentation of pituitary adenomas and the gland itself. Methods: We reviewed 34 studies that employed automatic and semi-automatic segmentation methods. We extracted and synthesized data on segmentation techniques and performance metrics (such as Dice overlap scores). Results: The majority of reviewed studies utilized deep learning approaches, with U-Net-based models being the most prevalent. Automatic methods yielded Dice scores of 0.19--89.00\% for pituitary gland and 4.60--96.41\% for adenoma segmentation. Semi-automatic methods reported 80.00--92.10\% for pituitary gland and 75.90--88.36\% for adenoma segmentation. Conclusion: Most studies did not report important metrics such as MR field strength, age and adenoma size. Automated segmentation techniques such as U-Net-based models show promise, especially for adenoma segmentation, but further improvements are needed to achieve consistently good performance in small structures like the normal pituitary gland. Continued innovation and larger, diverse datasets are likely critical to enhancing clinical applicability.
☆ Systematic Comparison of Projection Methods for Monocular 3D Human Pose Estimation on Fisheye Images
Fisheye cameras offer robots the ability to capture human movements across a wider field of view (FOV) than standard pinhole cameras, making them particularly useful for applications in human-robot interaction and automotive contexts. However, accurately detecting human poses in fisheye images is challenging due to the curved distortions inherent to fisheye optics. While various methods for undistorting fisheye images have been proposed, their effectiveness and limitations for poses that cover a wide FOV has not been systematically evaluated in the context of absolute human pose estimation from monocular fisheye images. To address this gap, we evaluate the impact of pinhole, equidistant and double sphere camera models, as well as cylindrical projection methods, on 3D human pose estimation accuracy. We find that in close-up scenarios, pinhole projection is inadequate, and the optimal projection method varies with the FOV covered by the human pose. The usage of advanced fisheye models like the double sphere model significantly enhances 3D human pose estimation accuracy. We propose a heuristic for selecting the appropriate projection model based on the detection bounding box to enhance prediction quality. Additionally, we introduce and evaluate on our novel dataset FISHnCHIPS, which features 3D human skeleton annotations in fisheye images, including images from unconventional angles, such as extreme close-ups, ground-mounted cameras, and wide-FOV poses, available at: https://www.vision.rwth-aachen.de/fishnchips
comment: Presented at IEEE International Conference on Robotics and Automation 2025
♻ ☆ Screen Them All: High-Throughput Pan-Cancer Genetic and Phenotypic Biomarker Screening from H&E Whole Slide Images
Molecular assays are standard of care for detecting genomic alterations in cancer prognosis and therapy selection but are costly, tissue-destructive and time-consuming. Artificial intelligence (AI) applied to routine hematoxylin and eosin (H&E)-stained whole slide images (WSIs) offers a fast and economical alternative for screening molecular biomarkers. We introduce OmniScreen, a high-throughput AI-based system leveraging Virchow2 embeddings extracted from 60,529 cancer patients with paired 489-gene MSK-IMPACT targeted biomarker panel and WSIs. Unlike conventional approaches that train separate models for each biomarker, OmniScreen employs a unified model to predict a broad range of clinically relevant biomarkers across cancers, including low-prevalence targets impractical to model individually. OmniScreen reliably identifies therapeutic targets and shared phenotypic features across common and rare tumors. We investigate the biomarker prediction probabilities and accuracies of OmniScreen in relation to tumor area, cohort size, histologic subtype alignment, and pathway-level morphological patterns. These findings underscore the potential of OmniScreen for routine clinical screening.
♻ ☆ DRO-Augment Framework: Robustness by Synergizing Wasserstein Distributionally Robust Optimization and Data Augmentation
In many real-world applications, ensuring the robustness and stability of deep neural networks (DNNs) is crucial, particularly for image classification tasks that encounter various input perturbations. While data augmentation techniques have been widely adopted to enhance the resilience of a trained model against such perturbations, there remains significant room for improvement in robustness against corrupted data and adversarial attacks simultaneously. To address this challenge, we introduce DRO-Augment, a novel framework that integrates Wasserstein Distributionally Robust Optimization (W-DRO) with various data augmentation strategies to improve the robustness of the models significantly across a broad spectrum of corruptions. Our method outperforms existing augmentation methods under severe data perturbations and adversarial attack scenarios while maintaining the accuracy on the clean datasets on a range of benchmark datasets, including but not limited to CIFAR-10-C, CIFAR-100-C, MNIST, and Fashion-MNIST. On the theoretical side, we establish novel generalization error bounds for neural networks trained using a computationally efficient, variation-regularized loss function closely related to the W-DRO problem.
comment: 26 pages,3 figures
♻ ☆ From Coarse to Continuous: Progressive Refinement Implicit Neural Representation for Motion-Robust Anisotropic MRI Reconstruction
In motion-robust magnetic resonance imaging (MRI), slice-to-volume reconstruction is critical for recovering anatomically consistent 3D brain volumes from 2D slices, especially under accelerated acquisitions or patient motion. However, this task remains challenging due to hierarchical structural disruptions. It includes local detail loss from k-space undersampling, global structural aliasing caused by motion, and volumetric anisotropy. Therefore, we propose a progressive refinement implicit neural representation (PR-INR) framework. Our PR-INR unifies motion correction, structural refinement, and volumetric synthesis within a geometry-aware coordinate space. Specifically, a motion-aware diffusion module is first employed to generate coarse volumetric reconstructions that suppress motion artifacts and preserve global anatomical structures. Then, we introduce an implicit detail restoration module that performs residual refinement by aligning spatial coordinates with visual features. It corrects local structures and enhances boundary precision. Further, a voxel continuous-aware representation module represents the image as a continuous function over 3D coordinates. It enables accurate inter-slice completion and high-frequency detail recovery. We evaluate PR-INR on five public MRI datasets under various motion conditions (3% and 5% displacement), undersampling rates (4x and 8x) and slice resolutions (scale = 5). Experimental results demonstrate that PR-INR outperforms state-of-the-art methods in both quantitative reconstruction metrics and visual quality. It further shows generalization and robustness across diverse unseen domains.
♻ ☆ WAFFLE: Finetuning Multi-Modal Model for Automated Front-End Development
Web development involves turning UI designs into functional webpages, which can be difficult for both beginners and experienced developers due to the complexity of HTML's hierarchical structures and styles. While Large Language Models (LLMs) have shown promise in generating source code, two major challenges persist in UI-to-HTML code generation: (1) effectively representing HTML's hierarchical structure for LLMs, and (2) bridging the gap between the visual nature of UI designs and the text-based format of HTML code. To tackle these challenges, we introduce Waffle, a new fine-tuning strategy that uses a structure-aware attention mechanism to improve LLMs' understanding of HTML's structure and a contrastive fine-tuning approach to align LLMs' understanding of UI images and HTML code. Models fine-tuned with Waffle show up to 9.00 pp (percentage point) higher HTML match, 0.0982 higher CW-SSIM, 32.99 higher CLIP, and 27.12 pp higher LLEM on our new benchmark WebSight-Test and an existing benchmark Design2Code, outperforming current fine-tuning methods.
♻ ☆ MaizeField3D: A Curated 3D Point Cloud and Procedural Model Dataset of Field-Grown Maize from a Diversity Panel
The development of artificial intelligence (AI) and machine learning (ML) based tools for 3D phenotyping, especially for maize, has been limited due to the lack of large and diverse 3D datasets. 2D image datasets fail to capture essential structural details such as leaf architecture, plant volume, and spatial arrangements that 3D data provide. To address this limitation, we present MaizeField3D (https://baskargroup.github.io/MaizeField3D/), a curated dataset of 3D point clouds of field-grown maize plants from a diverse genetic panel, designed to be AI-ready for advancing agricultural research. Our dataset includes 1,045 high-quality point clouds of field-grown maize collected using a terrestrial laser scanner (TLS). Point clouds of 520 plants from this dataset were segmented and annotated using a graph-based segmentation method to isolate individual leaves and stalks, ensuring consistent labeling across all samples. This labeled data was then used for fitting procedural models that provide a structured parametric representation of the maize plants. The leaves of the maize plants in the procedural models are represented using Non-Uniform Rational B-Spline (NURBS) surfaces that were generated using a two-step optimization process combining gradient-free and gradient-based methods. We conducted rigorous manual quality control on all datasets, correcting errors in segmentation, ensuring accurate leaf ordering, and validating metadata annotations. The dataset also includes metadata detailing plant morphology and quality, alongside multi-resolution subsampled point cloud data (100k, 50k, 10k points), which can be readily used for different downstream computational tasks. MaizeField3D will serve as a comprehensive foundational dataset for AI-driven phenotyping, plant structural analysis, and 3D applications in agricultural research.
comment: Elvis Kimara and Mozhgan Hadadi contributed equally to this work
♻ ☆ Temporal Differential Fields for 4D Motion Modeling via Image-to-Video Synthesis MICCAI
Temporal modeling on regular respiration-induced motions is crucial to image-guided clinical applications. Existing methods cannot simulate temporal motions unless high-dose imaging scans including starting and ending frames exist simultaneously. However, in the preoperative data acquisition stage, the slight movement of patients may result in dynamic backgrounds between the first and last frames in a respiratory period. This additional deviation can hardly be removed by image registration, thus affecting the temporal modeling. To address that limitation, we pioneeringly simulate the regular motion process via the image-to-video (I2V) synthesis framework, which animates with the first frame to forecast future frames of a given length. Besides, to promote the temporal consistency of animated videos, we devise the Temporal Differential Diffusion Model to generate temporal differential fields, which measure the relative differential representations between adjacent frames. The prompt attention layer is devised for fine-grained differential fields, and the field augmented layer is adopted to better interact these fields with the I2V framework, promoting more accurate temporal variation of synthesized videos. Extensive results on ACDC cardiac and 4D Lung datasets reveal that our approach simulates 4D videos along the intrinsic motion trajectory, rivaling other competitive methods on perceptual similarity and temporal consistency. Codes will be available soon.
comment: early accepted by MICCAI
♻ ☆ Exploring AI-based System Design for Pixel-level Protected Health Information Detection in Medical Images
De-identification of medical images is a critical step to ensure privacy during data sharing in research and clinical settings. The initial step in this process involves detecting Protected Health Information (PHI), which can be found in image metadata or imprinted within image pixels. Despite the importance of such systems, there has been limited evaluation of existing AI-based solutions, creating barriers to the development of reliable and robust tools. In this study, we present an AI-based pipeline for PHI detection, comprising three key modules: text detection, text extraction, and text analysis. We benchmark three models - YOLOv11, EasyOCR, and GPT-4o - across different setups corresponding to these modules, evaluating their performance on two different datasets encompassing multiple imaging modalities and PHI categories. Our findings indicate that the optimal setup involves utilizing dedicated vision and language models for each module, which achieves a commendable balance in performance, latency, and cost associated with the usage of Large Language Models (LLMs). Additionally, we show that the application of LLMs not only involves identifying PHI content but also enhances OCR tasks and facilitates an end-to-end PHI detection pipeline, showcasing promising outcomes through our analysis.
comment: In progress
♻ ☆ Shape and Texture Recognition in Large Vision-Language Models
Shapes and textures are the basic building blocks of visual perception. The ability to identify shapes regardless of orientation, texture, or context, and to recognize textures and materials independently of their associated objects, is essential for a general visual understanding of the world. This work introduces the Large Shape and Textures dataset (LAS&T), a giant collection of highly diverse shapes and textures, created by unsupervised extraction of patterns from natural images. This dataset is used to benchmark how effectively leading Large Vision-Language Models (LVLMs) understand shapes, textures, and materials in 2D and 3D scenes. For shape recognition, we test the models' ability to match images of identical shapes that differ in orientation, texture, color, or environment. Our results show that the shape recognition capabilities of the LVLMs remain significantly below human performance. LVLMs rely predominantly on high-level and semantic features and struggle with abstract shapes lacking clear class associations. For texture and material recognition, we evaluated the models' ability to identify images with identical textures and materials across different objects and environments. Interestingly, leading LVLMs approach human-level performance in recognizing materials in 3D scenes, yet substantially underperform humans when identifying simpler more abstract 2D textures. These results are consistent across a wide range of leading VLMs (GPT/Gemini/LLama/Qwen) and foundation vision models (DINO/CLIP), exposing major deficiencies in the ability of leading models to understand fundamental visual concepts. In contrast, simple nets trained directly for these tasks achieve high accuracy. The LAS&T dataset, featuring over 600,000 images for 2D/3D shape, texture, and material recognition and retrieval, is publicly available.
♻ ☆ GlyphPattern: An Abstract Pattern Recognition Benchmark for Vision-Language Models
Vision-Language Models (VLMs) building upon the foundation of powerful large language models have made rapid progress in reasoning across visual and textual data. While VLMs perform well on vision tasks that they are trained on, our results highlight key challenges in abstract pattern recognition. We present GlyphPattern, a 954 item dataset that pairs 318 human-written descriptions of visual patterns from 40 writing systems with three visual presentation styles. GlyphPattern evaluates abstract pattern recognition in VLMs, requiring models to understand and judge natural language descriptions of visual patterns. GlyphPattern patterns are drawn from a large-scale cognitive science investigation of human writing systems; as a result, they are rich in spatial reference and compositionality. Our experiments show that GlyphPattern is challenging for state-of-the-art VLMs (GPT-4o achieves only 55% accuracy), with marginal gains from few-shot prompting. Our detailed error analysis reveals challenges at multiple levels, including visual processing, natural language understanding, and pattern generalization.
♻ ☆ ObjCtrl-2.5D: Training-free Object Control with Camera Poses
This study aims to achieve more precise and versatile object control in image-to-video (I2V) generation. Current methods typically represent the spatial movement of target objects with 2D trajectories, which often fail to capture user intention and frequently produce unnatural results. To enhance control, we present ObjCtrl-2.5D, a training-free object control approach that uses a 3D trajectory, extended from a 2D trajectory with depth information, as a control signal. By modeling object movement as camera movement, ObjCtrl-2.5D represents the 3D trajectory as a sequence of camera poses, enabling object motion control using an existing camera motion control I2V generation model (CMC-I2V) without training. To adapt the CMC-I2V model originally designed for global motion control to handle local object motion, we introduce a module to isolate the target object from the background, enabling independent local control. In addition, we devise an effective way to achieve more accurate object control by sharing low-frequency warped latent within the object's region across frames. Extensive experiments demonstrate that ObjCtrl-2.5D significantly improves object control accuracy compared to training-free methods and offers more diverse control capabilities than training-based approaches using 2D trajectories, enabling complex effects like object rotation. Code and results are available at https://wzhouxiff.github.io/projects/ObjCtrl-2.5D/.
comment: Project Page: https://wzhouxiff.github.io/projects/ObjCtrl-2.5D/
♻ ☆ Two-Stream Spatial-Temporal Transformer Framework for Person Identification via Natural Conversational Keypoints
In the age of AI-driven generative technologies, traditional biometric recognition systems face unprecedented challenges, particularly from sophisticated deepfake and face reenactment techniques. In this study, we propose a Two-Stream Spatial-Temporal Transformer Framework for person identification using upper body keypoints visible during online conversations, which we term conversational keypoints. Our framework processes both spatial relationships between keypoints and their temporal evolution through two specialized branches: a Spatial Transformer (STR) that learns distinctive structural patterns in keypoint configurations, and a Temporal Transformer (TTR) that captures sequential motion patterns. Using the state-of-the-art Sapiens pose estimator, we extract 133 keypoints (based on COCO-WholeBody format) representing facial features, head pose, and hand positions. The framework was evaluated on a dataset of 114 individuals engaged in natural conversations, achieving recognition accuracies of 80.12% for the spatial stream, 63.61% for the temporal stream. We then explored two fusion strategies: a shared loss function approach achieving 82.22% accuracy, and a feature-level fusion method that concatenates feature maps from both streams, significantly improving performance to 94.86%. By jointly modeling both static anatomical relationships and dynamic movement patterns, our approach learns comprehensive identity signatures that are more robust to spoofing than traditional appearance-based methods.
comment: I would like to withdraw this submission due to the need for substantial revisions in the results and analysis. I plan to correct and improve the study and submit a more complete version in the near future
♻ ☆ 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 code and dataset are publicly available at https://github.com/bilibili/Index-anisora.
comment: 10 pages, 7 figures, 7 tables
♻ ☆ RA-NeRF: Robust Neural Radiance Field Reconstruction with Accurate Camera Pose Estimation under Complex Trajectories IROS 2025
Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have emerged as powerful tools for 3D reconstruction and SLAM tasks. However, their performance depends heavily on accurate camera pose priors. Existing approaches attempt to address this issue by introducing external constraints but fall short of achieving satisfactory accuracy, particularly when camera trajectories are complex. In this paper, we propose a novel method, RA-NeRF, capable of predicting highly accurate camera poses even with complex camera trajectories. Following the incremental pipeline, RA-NeRF reconstructs the scene using NeRF with photometric consistency and incorporates flow-driven pose regulation to enhance robustness during initialization and localization. Additionally, RA-NeRF employs an implicit pose filter to capture the camera movement pattern and eliminate the noise for pose estimation. To validate our method, we conduct extensive experiments on the Tanks\&Temple dataset for standard evaluation, as well as the NeRFBuster dataset, which presents challenging camera pose trajectories. On both datasets, RA-NeRF achieves state-of-the-art results in both camera pose estimation and visual quality, demonstrating its effectiveness and robustness in scene reconstruction under complex pose trajectories.
comment: IROS 2025
♻ ☆ Grounding Beyond Detection: Enhancing Contextual Understanding in Embodied 3D Grounding
Embodied 3D grounding aims to localize target objects described in human instructions from ego-centric viewpoint. Most methods typically follow a two-stage paradigm where a trained 3D detector's optimized backbone parameters are used to initialize a grounding model. In this study, we explore a fundamental question: Does embodied 3D grounding benefit enough from detection? To answer this question, we assess the grounding performance of detection models using predicted boxes filtered by the target category. Surprisingly, these detection models without any instruction-specific training outperform the grounding models explicitly trained with language instructions. This indicates that even category-level embodied 3D grounding may not be well resolved, let alone more fine-grained context-aware grounding. Motivated by this finding, we propose DEGround, which shares DETR queries as object representation for both DEtection and Grounding and enables the grounding to benefit from basic category classification and box detection. Based on this framework, we further introduce a regional activation grounding module that highlights instruction-related regions and a query-wise modulation module that incorporates sentence-level semantic into the query representation, strengthening the context-aware understanding of language instructions. Remarkably, DEGround outperforms state-of-the-art model BIP3D by 7.52% at overall accuracy on the EmbodiedScan validation set. The source code will be publicly available at https://github.com/zyn213/DEGround.
comment: 1st place on EmbodiedScan visual grounding
Information Retrieval 14
☆ Controlled Retrieval-augmented Context Evaluation for Long-form RAG
Retrieval-augmented generation (RAG) enhances large language models by incorporating context retrieved from external knowledge sources. While the effectiveness of the retrieval module is typically evaluated with relevance-based ranking metrics, such metrics may be insufficient to reflect the retrieval's impact on the final RAG result, especially in long-form generation scenarios. We argue that providing a comprehensive retrieval-augmented context is important for long-form RAG tasks like report generation and propose metrics for assessing the context independent of generation. We introduce CRUX, a \textbf{C}ontrolled \textbf{R}etrieval-a\textbf{U}gmented conte\textbf{X}t evaluation framework designed to directly assess retrieval-augmented contexts. This framework uses human-written summaries to control the information scope of knowledge, enabling us to measure how well the context covers information essential for long-form generation. CRUX uses question-based evaluation to assess RAG's retrieval in a fine-grained manner. Empirical results show that CRUX offers more reflective and diagnostic evaluation. Our findings also reveal substantial room for improvement in current retrieval methods, pointing to promising directions for advancing RAG's retrieval. Our data and code are publicly available to support and advance future research on retrieval.
☆ LSH-DynED: A Dynamic Ensemble Framework with LSH-Based Undersampling for Evolving Multi-Class Imbalanced Classification
The classification of imbalanced data streams, which have unequal class distributions, is a key difficulty in machine learning, especially when dealing with multiple classes. While binary imbalanced data stream classification tasks have received considerable attention, only a few studies have focused on multi-class imbalanced data streams. Effectively managing the dynamic imbalance ratio is a key challenge in this domain. This study introduces a novel, robust, and resilient approach to address these challenges by integrating Locality Sensitive Hashing with Random Hyperplane Projections (LSH-RHP) into the Dynamic Ensemble Diversification (DynED) framework. To the best of our knowledge, we present the first application of LSH-RHP for undersampling in the context of imbalanced non-stationary data streams. The proposed method undersamples the majority classes by utilizing LSH-RHP, provides a balanced training set, and improves the ensemble's prediction performance. We conduct comprehensive experiments on 23 real-world and ten semi-synthetic datasets and compare LSH-DynED with 15 state-of-the-art methods. The results reveal that LSH-DynED outperforms other approaches in terms of both Kappa and mG-Mean effectiveness measures, demonstrating its capability in dealing with multi-class imbalanced non-stationary data streams. Notably, LSH-DynED performs well in large-scale, high-dimensional datasets with considerable class imbalances and demonstrates adaptation and robustness in real-world circumstances. To motivate our design, we review existing methods for imbalanced data streams, outline key challenges, and offer guidance for future work. For the reproducibility of our results, we have made our implementation available on GitHub.
☆ CoVE: Compressed Vocabulary Expansion Makes Better LLM-based Recommender Systems ACL 2025
Recommender systems play a pivotal role in providing relevant content to users. With the rapid development of large language models (LLMs), researchers have begun utilizing LLMs to build more powerful recommender systems. However, existing approaches that focus on aligning LLMs with recommendation tasks do not fully leverage their sequential information processing capabilities, leading to suboptimal performance. In this paper, we propose a novel system called compressed vocabulary expansion (CoVE). In CoVE, each item is assigned a unique ID within the expanded vocabulary. Our framework effectively capitalizes on sequence understanding abilities of LLMs, significantly enhancing their performance on recommendation tasks. Additionally, we compress the embedding layer, making CoVE practical for large-scale industrial applications. The effectiveness and performance of CoVE are demonstrated through comprehensive experiments on multiple recommendation datasets and comparisons with prior works. Our code can be found at https://github.com/HaochenZhang717/CoVE-official-Repo.
comment: Accepted by ACL 2025 Findings
☆ KnowML: Improving Generalization of ML-NIDS with Attack Knowledge Graphs
Despite extensive research on Machine Learning-based Network Intrusion Detection Systems (ML-NIDS), their capability to detect diverse attack variants remains uncertain. Prior studies have largely relied on homogeneous datasets, which artificially inflate performance scores and offer a false sense of security. Designing systems that can effectively detect a wide range of attack variants remains a significant challenge. The progress of ML-NIDS continues to depend heavily on human expertise, which can embed subjective judgments of system designers into the model, potentially hindering its ability to generalize across diverse attack types. To address this gap, we propose KnowML, a framework for knowledge-guided machine learning that integrates attack knowledge into ML-NIDS. KnowML systematically explores the threat landscape by leveraging Large Language Models (LLMs) to perform automated analysis of attack implementations. It constructs a unified Knowledge Graph (KG) of attack strategies, on which it applies symbolic reasoning to generate KG-Augmented Input, embedding domain knowledge directly into the design process of ML-NIDS. We evaluate KnowML on 28 realistic attack variants, of which 10 are newly collected for this study. Our findings reveal that baseline ML-NIDS models fail to detect several variants entirely, achieving F1 scores as low as 0 %. In contrast, our knowledge-guided approach achieves up to 99 % F1 score while maintaining a False Positive Rate below 0.1 %.
☆ Why Do Open-Source LLMs Struggle with Data Analysis? A Systematic Empirical Study
Large Language Models (LLMs) hold promise in automating data analysis tasks, yet open-source models face significant limitations in these kinds of reasoning-intensive scenarios. In this work, we investigate strategies to enhance the data analysis capabilities of open-source LLMs. By curating a seed dataset of diverse, realistic scenarios, we evaluate models across three dimensions: data understanding, code generation, and strategic planning. Our analysis reveals three key findings: (1) Strategic planning quality serves as the primary determinant of model performance; (2) Interaction design and task complexity significantly influence reasoning capabilities; (3) Data quality demonstrates a greater impact than diversity in achieving optimal performance. We leverage these insights to develop a data synthesis methodology, demonstrating significant improvements in open-source LLMs' analytical reasoning capabilities.
comment: Work in progress
☆ Alleviating User-Sensitive bias with Fair Generative Sequential Recommendation Model
Recommendation fairness has recently attracted much attention. In the real world, recommendation systems are driven by user behavior, and since users with the same sensitive feature (e.g., gender and age) tend to have the same patterns, recommendation models can easily capture the strong correlation preference of sensitive features and thus cause recommendation unfairness. Diffusion model (DM) as a new generative model paradigm has achieved great success in recommendation systems. DM's ability to model uncertainty and represent diversity, and its modeling mechanism has a high degree of adaptability with the real-world recommendation process with bias. Therefore, we use DM to effectively model the fairness of recommendation and enhance the diversity. This paper proposes a FairGENerative sequential Recommendation model based on DM, FairGENRec. In the training phase, we inject random noise into the original distribution under the guidance of the sensitive feature recognition model, and a sequential denoise model is designed for the reverse reconstruction of items. Simultaneously, recommendation fairness modeling is completed by injecting multi-interests representational information that eliminates the bias of sensitive user features into the generated results. In the inference phase, the model obtains the noise in the form of noise addition by using the history interactions which is followed by reverse iteration to reconstruct the target item representation. Finally, our extensive experiments on three datasets demonstrate the dual enhancement effect of FairGENRec on accuracy and fairness, while the statistical analysis of the cases visualizes the degree of improvement on the fairness of the recommendation.
☆ NEAR$^2$: A Nested Embedding Approach to Efficient Product Retrieval and Ranking SIGIR
E-commerce information retrieval (IR) systems struggle to simultaneously achieve high accuracy in interpreting complex user queries and maintain efficient processing of vast product catalogs. The dual challenge lies in precisely matching user intent with relevant products while managing the computational demands of real-time search across massive inventories. In this paper, we propose a Nested Embedding Approach to product Retrieval and Ranking, called NEAR$^2$, which can achieve up to $12$ times efficiency in embedding size at inference time while introducing no extra cost in training and improving performance in accuracy for various encoder-based Transformer models. We validate our approach using different loss functions for the retrieval and ranking task, including multiple negative ranking loss and online contrastive loss, on four different test sets with various IR challenges such as short and implicit queries. Our approach achieves an improved performance over a smaller embedding dimension, compared to any existing models.
comment: This paper is accepted to the 2025 SIGIR Workshop on eCommerce
☆ Higher-Order Graph Databases
Recent advances in graph databases (GDBs) have been driving interest in large-scale analytics, yet current systems fail to support higher-order (HO) interactions beyond first-order (one-hop) relations, which are crucial for tasks such as subgraph counting, polyadic modeling, and HO graph learning. We address this by introducing a new class of systems, higher-order graph databases (HO-GDBs) that use lifting and lowering paradigms to seamlessly extend traditional GDBs with HO. We provide a theoretical analysis of OLTP and OLAP queries, ensuring correctness, scalability, and ACID compliance. We implement a lightweight, modular, and parallelizable HO-GDB prototype that offers native support for hypergraphs, node-tuples, subgraphs, and other HO structures under a unified API. The prototype scales to large HO OLTP & OLAP workloads and shows how HO improves analytical tasks, for example enhancing accuracy of graph neural networks within a GDB by 44%. Our work ensures low latency and high query throughput, and generalizes both ACID-compliant and eventually consistent systems.
☆ Health Sentinel: An AI Pipeline For Real-time Disease Outbreak Detection
Early detection of disease outbreaks is crucial to ensure timely intervention by the health authorities. Due to the challenges associated with traditional indicator-based surveillance, monitoring informal sources such as online media has become increasingly popular. However, owing to the number of online articles getting published everyday, manual screening of the articles is impractical. To address this, we propose Health Sentinel. It is a multi-stage information extraction pipeline that uses a combination of ML and non-ML methods to extract events-structured information concerning disease outbreaks or other unusual health events-from online articles. The extracted events are made available to the Media Scanning and Verification Cell (MSVC) at the National Centre for Disease Control (NCDC), Delhi for analysis, interpretation and further dissemination to local agencies for timely intervention. From April 2022 till date, Health Sentinel has processed over 300 million news articles and identified over 95,000 unique health events across India of which over 3,500 events were shortlisted by the public health experts at NCDC as potential outbreaks.
♻ ☆ Aug2Search: Enhancing Facebook Marketplace Search with LLM-Generated Synthetic Data Augmentation
Embedding-Based Retrieval (EBR) is an important technique in modern search engines, enabling semantic match between search queries and relevant results. However, search logging data on platforms like Facebook Marketplace lacks the diversity and details needed for effective EBR model training, limiting the models' ability to capture nuanced search patterns. To address this challenge, we propose Aug2Search, an EBR-based framework leveraging synthetic data generated by Generative AI (GenAI) models, in a multimodal and multitask approach to optimize query-product relevance. This paper investigates the capabilities of GenAI, particularly Large Language Models (LLMs), in generating high-quality synthetic data, and analyzing its impact on enhancing EBR models. We conducted experiments using eight Llama models and 100 million data points from Facebook Marketplace logs. Our synthetic data generation follows three strategies: (1) generate queries, (2) enhance product listings, and (3) generate queries from enhanced listings. We train EBR models on three different datasets: sampled engagement data or original data ((e.g., "Click" and "Listing Interactions")), synthetic data, and a mixture of both engagement and synthetic data to assess their performance across various training sets. Our findings underscore the robustness of Llama models in producing synthetic queries and listings with high coherence, relevance, and diversity, while maintaining low levels of hallucination. Aug2Search achieves an improvement of up to 4% in ROC_AUC with 100 million synthetic data samples, demonstrating the effectiveness of our approach. Moreover, our experiments reveal that with the same volume of training data, models trained exclusively on synthetic data often outperform those trained on original data only or a mixture of original and synthetic data.
♻ ☆ Answering Multimodal Exclusion Queries with Lightweight Sparse Disentangled Representations SIGIR
Multimodal representations that enable cross-modal retrieval are widely used. However, these often lack interpretability making it difficult to explain the retrieved results. Solutions such as learning sparse disentangled representations are typically guided by the text tokens in the data, making the dimensionality of the resulting embeddings very high. We propose an approach that generates smaller dimensionality fixed-size embeddings that are not only disentangled but also offer better control for retrieval tasks. We demonstrate their utility using challenging exclusion queries over MSCOCO and Conceptual Captions benchmarks. Our experiments show that our approach is superior to traditional dense models such as CLIP, BLIP and VISTA (gains up to 11% in AP@10), as well as sparse disentangled models like VDR (gains up to 21% in AP@10). We also present qualitative results to further underline the interpretability of disentangled representations.
comment: In Proceedings of the 2025 International ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval (ICTIR)
♻ ☆ Entropy and type-token ratio in gigaword corpora
There are different ways of measuring diversity in complex systems. In particular, in language, lexical diversity is characterized in terms of the type-token ratio and the word entropy. We here investigate both diversity metrics in six massive linguistic datasets in English, Spanish, and Turkish, consisting of books, news articles, and tweets. These gigaword corpora correspond to languages with distinct morphological features and differ in registers and genres, thus constituting a varied testbed for a quantitative approach to lexical diversity. We unveil an empirical functional relation between entropy and type-token ratio of texts of a given corpus and language, which is a consequence of the statistical laws observed in natural language. Further, in the limit of large text lengths we find an analytical expression for this relation relying on both Zipf and Heaps laws that agrees with our empirical findings.
comment: 15 pages, 10 figures, 8 tables
♻ ☆ jina-embeddings-v4: Universal Embeddings for Multimodal Multilingual Retrieval
We introduce jina-embeddings-v4, a 3.8 billion parameter multimodal embedding model that unifies text and image representations through a novel architecture supporting both single-vector and multi-vector embeddings in the late interaction style. The model incorporates task-specific Low-Rank Adaptation (LoRA) adapters to optimize performance across diverse retrieval scenarios, including query-document retrieval, semantic text similarity, and code search. Comprehensive evaluations demonstrate that jina-embeddings-v4 achieves state-of-the-art performance on both single-modal and cross-modal retrieval tasks, with particular strength in processing visually rich content such as tables, charts, diagrams, and mixed-media formats. To facilitate evaluation of this capability, we also introduce Jina-VDR, a novel benchmark specifically designed for visually rich image retrieval.
comment: 22 pages, 1-10 main, 14-22 experimental results, benchmark tables
♻ ☆ Talking to GDELT Through Knowledge Graphs
In this work we study various Retrieval Augmented Regeneration (RAG) approaches to gain an understanding of the strengths and weaknesses of each approach in a question-answering analysis. To gain this understanding we use a case-study subset of the Global Database of Events, Language, and Tone (GDELT) dataset as well as a corpus of raw text scraped from the online news articles. To retrieve information from the text corpus we implement a traditional vector store RAG as well as state-of-the-art large language model (LLM) based approaches for automatically constructing KGs and retrieving the relevant subgraphs. In addition to these corpus approaches, we develop a novel ontology-based framework for constructing knowledge graphs (KGs) from GDELT directly which leverages the underlying schema of GDELT to create structured representations of global events. For retrieving relevant information from the ontology-based KGs we implement both direct graph queries and state-of-the-art graph retrieval approaches. We compare the performance of each method in a question-answering task. We find that while our ontology-based KGs are valuable for question-answering, automated extraction of the relevant subgraphs is challenging. Conversely, LLM-generated KGs, while capturing event summaries, often lack consistency and interpretability. Our findings suggest benefits of a synergistic approach between ontology and LLM-based KG construction, with proposed avenues toward that end.
Robotics 55
☆ Low-Cost Infrastructure-Free 3D Relative Localization with Sub-Meter Accuracy in Near Field
Relative localization in the near-field scenario is critically important for unmanned vehicle (UxV) applications. Although related works addressing 2D relative localization problem have been widely studied for unmanned ground vehicles (UGVs), the problem in 3D scenarios for unmanned aerial vehicles (UAVs) involves more uncertainties and remains to be investigated. Inspired by the phenomenon that animals can achieve swarm behaviors solely based on individual perception of relative information, this study proposes an infrastructure-free 3D relative localization framework that relies exclusively on onboard ultra-wideband (UWB) sensors. Leveraging 2D relative positioning research, we conducted feasibility analysis, system modeling, simulations, performance evaluation, and field tests using UWB sensors. The key contributions of this work include: derivation of the Cram\'er-Rao lower bound (CRLB) and geometric dilution of precision (GDOP) for near-field scenarios; development of two localization algorithms -- one based on Euclidean distance matrix (EDM) and another employing maximum likelihood estimation (MLE); comprehensive performance comparison and computational complexity analysis against state-of-the-art methods; simulation studies and field experiments; a novel sensor deployment strategy inspired by animal behavior, enabling single-sensor implementation within the proposed framework for UxV applications. The theoretical, simulation, and experimental results demonstrate strong generalizability to other 3D near-field localization tasks, with significant potential for a cost-effective cross-platform UxV collaborative system.
☆ Situated Haptic Interaction: Exploring the Role of Context in Affective Perception of Robotic Touch
Affective interaction is not merely about recognizing emotions; it is an embodied, situated process shaped by context and co-created through interaction. In affective computing, the role of haptic feedback within dynamic emotional exchanges remains underexplored. This study investigates how situational emotional cues influence the perception and interpretation of haptic signals given by a robot. In a controlled experiment, 32 participants watched video scenarios in which a robot experienced either positive actions (such as being kissed), negative actions (such as being slapped) or neutral actions. After each video, the robot conveyed its emotional response through haptic communication, delivered via a wearable vibration sleeve worn by the participant. Participants rated the robot's emotional state-its valence (positive or negative) and arousal (intensity)-based on the video, the haptic feedback, and the combination of the two. The study reveals a dynamic interplay between visual context and touch. Participants' interpretation of haptic feedback was strongly shaped by the emotional context of the video, with visual context often overriding the perceived valence of the haptic signal. Negative haptic cues amplified the perceived valence of the interaction, while positive cues softened it. Furthermore, haptics override the participants' perception of arousal of the video. Together, these results offer insights into how situated haptic feedback can enrich affective human-robot interaction, pointing toward more nuanced and embodied approaches to emotional communication with machines.
☆ CUPID: Curating Data your Robot Loves with Influence Functions
In robot imitation learning, policy performance is tightly coupled with the quality and composition of the demonstration data. Yet, developing a precise understanding of how individual demonstrations contribute to downstream outcomes - such as closed-loop task success or failure - remains a persistent challenge. We propose CUPID, a robot data curation method based on a novel influence function-theoretic formulation for imitation learning policies. Given a set of evaluation rollouts, CUPID estimates the influence of each training demonstration on the policy's expected return. This enables ranking and selection of demonstrations according to their impact on the policy's closed-loop performance. We use CUPID to curate data by 1) filtering out training demonstrations that harm policy performance and 2) subselecting newly collected trajectories that will most improve the policy. Extensive simulated and hardware experiments show that our approach consistently identifies which data drives test-time performance. For example, training with less than 33% of curated data can yield state-of-the-art diffusion policies on the simulated RoboMimic benchmark, with similar gains observed in hardware. Furthermore, hardware experiments show that our method can identify robust strategies under distribution shift, isolate spurious correlations, and even enhance the post-training of generalist robot policies. Additional materials are made available at: https://cupid-curation.github.io.
comment: Project page: https://cupid-curation.github.io. 28 pages, 15 figures
☆ Analysis and experiments of the dissipative Twistcar: direction reversal and asymptotic approximations
Underactuated wheeled vehicles are commonly studied as nonholonomic systems with periodic actuation. Twistcar is a classical example inspired by a riding toy, which has been analyzed using a planar model of a dynamical system with nonholonomic constraints. Most of the previous analyses did not account for energy dissipation due to friction. In this work, we study a theoretical two-link model of the Twistcar while incorporating dissipation due to rolling resistance. We obtain asymptotic expressions for the system's small-amplitude steady-state periodic dynamics, which reveals the possibility of reversing the direction of motion upon varying the geometric and mass properties of the vehicle. Next, we design and construct a robotic prototype of the Twistcar whose center-of-mass position can be shifted by adding and removing a massive block, enabling demonstration of the Twistcar's direction reversal phenomenon. We also conduct parameter fitting for the frictional resistance in order to improve agreement with experiments.
☆ Multimodal Anomaly Detection with a Mixture-of-Experts IROS 2025
With a growing number of robots being deployed across diverse applications, robust multimodal anomaly detection becomes increasingly important. In robotic manipulation, failures typically arise from (1) robot-driven anomalies due to an insufficient task model or hardware limitations, and (2) environment-driven anomalies caused by dynamic environmental changes or external interferences. Conventional anomaly detection methods focus either on the first by low-level statistical modeling of proprioceptive signals or the second by deep learning-based visual environment observation, each with different computational and training data requirements. To effectively capture anomalies from both sources, we propose a mixture-of-experts framework that integrates the complementary detection mechanisms with a visual-language model for environment monitoring and a Gaussian-mixture regression-based detector for tracking deviations in interaction forces and robot motions. We introduce a confidence-based fusion mechanism that dynamically selects the most reliable detector for each situation. We evaluate our approach on both household and industrial tasks using two robotic systems, demonstrating a 60% reduction in detection delay while improving frame-wise anomaly detection performance compared to individual detectors.
comment: 8 pages, 5 figures, 1 table, the paper has been accepted for publication in the Proceedings of the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025)
☆ Faster Motion Planning via Restarts
Randomized methods such as PRM and RRT are widely used in motion planning. However, in some cases, their running-time suffers from inherent instability, leading to ``catastrophic'' performance even for relatively simple instances. We apply stochastic restart techniques, some of them new, for speeding up Las Vegas algorithms, that provide dramatic speedups in practice (a factor of $3$ [or larger] in many cases). Our experiments demonstrate that the new algorithms have faster runtimes, shorter paths, and greater gains from multi-threading (when compared with straightforward parallel implementation). We prove the optimality of the new variants. Our implementation is open source, available on github, and is easy to deploy and use.
comment: arXiv admin note: text overlap with arXiv:2503.04633
☆ MinD: Unified Visual Imagination and Control via Hierarchical World Models
Video generation models (VGMs) offer a promising pathway for unified world modeling in robotics by integrating simulation, prediction, and manipulation. However, their practical application remains limited due to (1) slowgeneration speed, which limits real-time interaction, and (2) poor consistency between imagined videos and executable actions. To address these challenges, we propose Manipulate in Dream (MinD), a hierarchical diffusion-based world model framework that employs a dual-system design for vision-language manipulation. MinD executes VGM at low frequencies to extract video prediction features, while leveraging a high-frequency diffusion policy for real-time interaction. This architecture enables low-latency, closed-loop control in manipulation with coherent visual guidance. To better coordinate the two systems, we introduce a video-action diffusion matching module (DiffMatcher), with a novel co-training strategy that uses separate schedulers for each diffusion model. Specifically, we introduce a diffusion-forcing mechanism to DiffMatcher that aligns their intermediate representations during training, helping the fast action model better understand video-based predictions. Beyond manipulation, MinD also functions as a world simulator, reliably predicting task success or failure in latent space before execution. Trustworthy analysis further shows that VGMs can preemptively evaluate task feasibility and mitigate risks. Extensive experiments across multiple benchmarks demonstrate that MinD achieves state-of-the-art manipulation (63%+) in RL-Bench, advancing the frontier of unified world modeling in robotics.
☆ GRAND-SLAM: Local Optimization for Globally Consistent Large-Scale Multi-Agent Gaussian SLAM
3D Gaussian splatting has emerged as an expressive scene representation for RGB-D visual SLAM, but its application to large-scale, multi-agent outdoor environments remains unexplored. Multi-agent Gaussian SLAM is a promising approach to rapid exploration and reconstruction of environments, offering scalable environment representations, but existing approaches are limited to small-scale, indoor environments. To that end, we propose Gaussian Reconstruction via Multi-Agent Dense SLAM, or GRAND-SLAM, a collaborative Gaussian splatting SLAM method that integrates i) an implicit tracking module based on local optimization over submaps and ii) an approach to inter- and intra-robot loop closure integrated into a pose-graph optimization framework. Experiments show that GRAND-SLAM provides state-of-the-art tracking performance and 28% higher PSNR than existing methods on the Replica indoor dataset, as well as 91% lower multi-agent tracking error and improved rendering over existing multi-agent methods on the large-scale, outdoor Kimera-Multi dataset.
☆ SViP: Sequencing Bimanual Visuomotor Policies with Object-Centric Motion Primitives
Imitation learning (IL), particularly when leveraging high-dimensional visual inputs for policy training, has proven intuitive and effective in complex bimanual manipulation tasks. Nonetheless, the generalization capability of visuomotor policies remains limited, especially when small demonstration datasets are available. Accumulated errors in visuomotor policies significantly hinder their ability to complete long-horizon tasks. To address these limitations, we propose SViP, a framework that seamlessly integrates visuomotor policies into task and motion planning (TAMP). SViP partitions human demonstrations into bimanual and unimanual operations using a semantic scene graph monitor. Continuous decision variables from the key scene graph are employed to train a switching condition generator. This generator produces parameterized scripted primitives that ensure reliable performance even when encountering out-of-the-distribution observations. Using only 20 real-world demonstrations, we show that SViP enables visuomotor policies to generalize across out-of-distribution initial conditions without requiring object pose estimators. For previously unseen tasks, SViP automatically discovers effective solutions to achieve the goal, leveraging constraint modeling in TAMP formulism. In real-world experiments, SViP outperforms state-of-the-art generative IL methods, indicating wider applicability for more complex tasks. Project website: https://sites.google.com/view/svip-bimanual
comment: Project website: https://sites.google.com/view/svip-bimanual
☆ Learning Physical Systems: Symplectification via Gauge Fixing in Dirac Structures
Physics-informed deep learning has achieved remarkable progress by embedding geometric priors, such as Hamiltonian symmetries and variational principles, into neural networks, enabling structure-preserving models that extrapolate with high accuracy. However, in systems with dissipation and holonomic constraints, ubiquitous in legged locomotion and multibody robotics, the canonical symplectic form becomes degenerate, undermining the very invariants that guarantee stability and long-term prediction. In this work, we tackle this foundational limitation by introducing Presymplectification Networks (PSNs), the first framework to learn the symplectification lift via Dirac structures, restoring a non-degenerate symplectic geometry by embedding constrained systems into a higher-dimensional manifold. Our architecture combines a recurrent encoder with a flow-matching objective to learn the augmented phase-space dynamics end-to-end. We then attach a lightweight Symplectic Network (SympNet) to forecast constrained trajectories while preserving energy, momentum, and constraint satisfaction. We demonstrate our method on the dynamics of the ANYmal quadruped robot, a challenging contact-rich, multibody system. To the best of our knowledge, this is the first framework that effectively bridges the gap between constrained, dissipative mechanical systems and symplectic learning, unlocking a whole new class of geometric machine learning models, grounded in first principles yet adaptable from data.
comment: Presented at Equivariant Systems: Theory and Applications in State Estimation, Artificial Intelligence and Control, Robotics: Science and Systems (RSS) 2025 Workshop, 6 Pages, 3 Figures
☆ OC-SOP: Enhancing Vision-Based 3D Semantic Occupancy Prediction by Object-Centric Awareness
Autonomous driving perception faces significant challenges due to occlusions and incomplete scene data in the environment. To overcome these issues, the task of semantic occupancy prediction (SOP) is proposed, which aims to jointly infer both the geometry and semantic labels of a scene from images. However, conventional camera-based methods typically treat all categories equally and primarily rely on local features, leading to suboptimal predictions, especially for dynamic foreground objects. To address this, we propose Object-Centric SOP (OC-SOP), a framework that integrates high-level object-centric cues extracted via a detection branch into the semantic occupancy prediction pipeline. This object-centric integration significantly enhances the prediction accuracy for foreground objects and achieves state-of-the-art performance among all categories on SemanticKITTI.
comment: under review
☆ SWA-SOP: Spatially-aware Window Attention for Semantic Occupancy Prediction in Autonomous Driving
Perception systems in autonomous driving rely on sensors such as LiDAR and cameras to perceive the 3D environment. However, due to occlusions and data sparsity, these sensors often fail to capture complete information. Semantic Occupancy Prediction (SOP) addresses this challenge by inferring both occupancy and semantics of unobserved regions. Existing transformer-based SOP methods lack explicit modeling of spatial structure in attention computation, resulting in limited geometric awareness and poor performance in sparse or occluded areas. To this end, we propose Spatially-aware Window Attention (SWA), a novel mechanism that incorporates local spatial context into attention. SWA significantly improves scene completion and achieves state-of-the-art results on LiDAR-based SOP benchmarks. We further validate its generality by integrating SWA into a camera-based SOP pipeline, where it also yields consistent gains across modalities.
comment: under reviewed
☆ DefFusionNet: Learning Multimodal Goal Shapes for Deformable Object Manipulation via a Diffusion-based Probabilistic Model
Deformable object manipulation is critical to many real-world robotic applications, ranging from surgical robotics and soft material handling in manufacturing to household tasks like laundry folding. At the core of this important robotic field is shape servoing, a task focused on controlling deformable objects into desired shapes. The shape servoing formulation requires the specification of a goal shape. However, most prior works in shape servoing rely on impractical goal shape acquisition methods, such as laborious domain-knowledge engineering or manual manipulation. DefGoalNet previously posed the current state-of-the-art solution to this problem, which learns deformable object goal shapes directly from a small number of human demonstrations. However, it significantly struggles in multi-modal settings, where multiple distinct goal shapes can all lead to successful task completion. As a deterministic model, DefGoalNet collapses these possibilities into a single averaged solution, often resulting in an unusable goal. In this paper, we address this problem by developing DefFusionNet, a novel neural network that leverages the diffusion probabilistic model to learn a distribution over all valid goal shapes rather than predicting a single deterministic outcome. This enables the generation of diverse goal shapes and avoids the averaging artifacts. We demonstrate our method's effectiveness on robotic tasks inspired by both manufacturing and surgical applications, both in simulation and on a physical robot. Our work is the first generative model capable of producing a diverse, multi-modal set of deformable object goals for real-world robotic applications.
☆ USVTrack: USV-Based 4D Radar-Camera Tracking Dataset for Autonomous Driving in Inland Waterways IROS
Object tracking in inland waterways plays a crucial role in safe and cost-effective applications, including waterborne transportation, sightseeing tours, environmental monitoring and surface rescue. Our Unmanned Surface Vehicle (USV), equipped with a 4D radar, a monocular camera, a GPS, and an IMU, delivers robust tracking capabilities in complex waterborne environments. By leveraging these sensors, our USV collected comprehensive object tracking data, which we present as USVTrack, the first 4D radar-camera tracking dataset tailored for autonomous driving in new generation waterborne transportation systems. Our USVTrack dataset presents rich scenarios, featuring diverse various waterways, varying times of day, and multiple weather and lighting conditions. Moreover, we present a simple but effective radar-camera matching method, termed RCM, which can be plugged into popular two-stage association trackers. Experimental results utilizing RCM demonstrate the effectiveness of the radar-camera matching in improving object tracking accuracy and reliability for autonomous driving in waterborne environments. The USVTrack dataset is public on https://usvtrack.github.io.
comment: Accepted by IROS
☆ TDACloud: Point Cloud Recognition Using Topological Data Analysis
Point cloud-based object/place recognition remains a problem of interest in applications such as autonomous driving, scene reconstruction, and localization. Extracting meaningful local descriptors from a query point cloud that can be matched with the descriptors of the collected point clouds is a challenging problem. Furthermore, when the query point cloud is noisy or has been transformed (e.g., rotated), it adds to the complexity. To this end, we propose a novel methodology, named TDACloud, using Topological Data Analysis (TDA) for local descriptor extraction from a point cloud, which does not need resource-intensive GPU-based machine learning training. More specifically, we used the ATOL vectorization method to generate vectors for point clouds. Unlike voxelization, our proposed technique can take raw point clouds as inputs and outputs a fixed-size TDA-descriptor vector. To test the quality of the proposed TDACloud technique, we have implemented it on multiple real-world (e.g., Oxford RobotCar, KITTI-360) and realistic (e.g., ShapeNet) point cloud datasets for object and place recognition. We have also tested TDACloud on noisy and transformed test cases where the query point cloud has been scaled, translated, or rotated. Our results demonstrate high recognition accuracies in noisy conditions and large-scale real-world place recognition while outperforming the baselines by up to approximately 14%.
☆ Including Semantic Information via Word Embeddings for Skeleton-based Action Recognition IJCNN
Effective human action recognition is widely used for cobots in Industry 4.0 to assist in assembly tasks. However, conventional skeleton-based methods often lose keypoint semantics, limiting their effectiveness in complex interactions. In this work, we introduce a novel approach to skeleton-based action recognition that enriches input representations by leveraging word embeddings to encode semantic information. Our method replaces one-hot encodings with semantic volumes, enabling the model to capture meaningful relationships between joints and objects. Through extensive experiments on multiple assembly datasets, we demonstrate that our approach significantly improves classification performance, and enhances generalization capabilities by simultaneously supporting different skeleton types and object classes. Our findings highlight the potential of incorporating semantic information to enhance skeleton-based action recognition in dynamic and diverse environments.
comment: IEEE International Joint Conference on Neural Networks (IJCNN) 2025
☆ Safety-Aware Optimal Scheduling for Autonomous Masonry Construction using Collaborative Heterogeneous Aerial Robots IROS 2025
This paper presents a novel high-level task planning and optimal coordination framework for autonomous masonry construction, using a team of heterogeneous aerial robotic workers, consisting of agents with separate skills for brick placement and mortar application. This introduces new challenges in scheduling and coordination, particularly due to the mortar curing deadline required for structural bonding and ensuring the safety constraints among UAVs operating in parallel. To address this, an automated pipeline generates the wall construction plan based on the available bricks while identifying static structural dependencies and potential conflicts for safe operation. The proposed framework optimizes UAV task allocation and execution timing by incorporating dynamically coupled precedence deadline constraints that account for the curing process and static structural dependency constraints, while enforcing spatio-temporal constraints to prevent collisions and ensure safety. The primary objective of the scheduler is to minimize the overall construction makespan while minimizing logistics, traveling time between tasks, and the curing time to maintain both adhesion quality and safe workspace separation. The effectiveness of the proposed method in achieving coordinated and time-efficient aerial masonry construction is extensively validated through Gazebo simulated missions. The results demonstrate the framework's capability to streamline UAV operations, ensuring both structural integrity and safety during the construction process.
comment: This paper has been accepted for publication at the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025)
☆ NOVA: Navigation via Object-Centric Visual Autonomy for High-Speed Target Tracking in Unstructured GPS-Denied Environments
Autonomous aerial target tracking in unstructured and GPS-denied environments remains a fundamental challenge in robotics. Many existing methods rely on motion capture systems, pre-mapped scenes, or feature-based localization to ensure safety and control, limiting their deployment in real-world conditions. We introduce NOVA, a fully onboard, object-centric framework that enables robust target tracking and collision-aware navigation using only a stereo camera and an IMU. Rather than constructing a global map or relying on absolute localization, NOVA formulates perception, estimation, and control entirely in the target's reference frame. A tightly integrated stack combines a lightweight object detector with stereo depth completion, followed by histogram-based filtering to infer robust target distances under occlusion and noise. These measurements feed a visual-inertial state estimator that recovers the full 6-DoF pose of the robot relative to the target. A nonlinear model predictive controller (NMPC) plans dynamically feasible trajectories in the target frame. To ensure safety, high-order control barrier functions are constructed online from a compact set of high-risk collision points extracted from depth, enabling real-time obstacle avoidance without maps or dense representations. We validate NOVA across challenging real-world scenarios, including urban mazes, forest trails, and repeated transitions through buildings with intermittent GPS loss and severe lighting changes that disrupt feature-based localization. Each experiment is repeated multiple times under similar conditions to assess resilience, showing consistent and reliable performance. NOVA achieves agile target following at speeds exceeding 50 km/h. These results show that high-speed vision-based tracking is possible in the wild using only onboard sensing, with no reliance on external localization or environment assumptions.
☆ MCN-SLAM: Multi-Agent Collaborative Neural SLAM with Hybrid Implicit Neural Scene Representation
Neural implicit scene representations have recently shown promising results in dense visual SLAM. However, existing implicit SLAM algorithms are constrained to single-agent scenarios, and fall difficulties in large-scale scenes and long sequences. Existing NeRF-based multi-agent SLAM frameworks cannot meet the constraints of communication bandwidth. To this end, we propose the first distributed multi-agent collaborative neural SLAM framework with hybrid scene representation, distributed camera tracking, intra-to-inter loop closure, and online distillation for multiple submap fusion. A novel triplane-grid joint scene representation method is proposed to improve scene reconstruction. A novel intra-to-inter loop closure method is designed to achieve local (single-agent) and global (multi-agent) consistency. We also design a novel online distillation method to fuse the information of different submaps to achieve global consistency. Furthermore, to the best of our knowledge, there is no real-world dataset for NeRF-based/GS-based SLAM that provides both continuous-time trajectories groundtruth and high-accuracy 3D meshes groundtruth. To this end, we propose the first real-world Dense slam (DES) dataset covering both single-agent and multi-agent scenarios, ranging from small rooms to large-scale outdoor scenes, with high-accuracy ground truth for both 3D mesh and continuous-time camera trajectory. This dataset can advance the development of the research in both SLAM, 3D reconstruction, and visual foundation model. Experiments on various datasets demonstrate the superiority of the proposed method in both mapping, tracking, and communication. The dataset and code will open-source on https://github.com/dtc111111/mcnslam.
☆ PG-LIO: Photometric-Geometric fusion for Robust LiDAR-Inertial Odometry
LiDAR-Inertial Odometry (LIO) is widely used for accurate state estimation and mapping which is an essential requirement for autonomous robots. Conventional LIO methods typically rely on formulating constraints from the geometric structure sampled by the LiDAR. Hence, in the lack of geometric structure, these tend to become ill-conditioned (degenerate) and fail. Robustness of LIO to such conditions is a necessity for its broader deployment. To address this, we propose PG-LIO, a real-time LIO method that fuses photometric and geometric information sampled by the LiDAR along with inertial constraints from an Inertial Measurement Unit (IMU). This multi-modal information is integrated into a factor graph optimized over a sliding window for real-time operation. We evaluate PG-LIO on multiple datasets that include both geometrically well-conditioned as well as self-similar scenarios. Our method achieves accuracy on par with state-of-the-art LIO in geometrically well-structured settings while significantly improving accuracy in degenerate cases including against methods that also fuse intensity. Notably, we demonstrate only 1 m drift over a 1 km manually piloted aerial trajectory through a geometrically self-similar tunnel at an average speed of 7.5m/s (max speed 10.8 m/s). For the benefit of the community, we shall also release our source code https://github.com/ntnu-arl/mimosa
comment: 8 pages, 6 figures
☆ Learning Point Correspondences In Radar 3D Point Clouds For Radar-Inertial Odometry
Using 3D point clouds in odometry estimation in robotics often requires finding a set of correspondences between points in subsequent scans. While there are established methods for point clouds of sufficient quality, state-of-the-art still struggles when this quality drops. Thus, this paper presents a novel learning-based framework for predicting robust point correspondences between pairs of noisy, sparse and unstructured 3D point clouds from a light-weight, low-power, inexpensive, consumer-grade System-on-Chip (SoC) Frequency Modulated Continuous Wave (FMCW) radar sensor. Our network is based on the transformer architecture which allows leveraging the attention mechanism to discover pairs of points in consecutive scans with the greatest mutual affinity. The proposed network is trained in a self-supervised way using set-based multi-label classification cross-entropy loss, where the ground-truth set of matches is found by solving the Linear Sum Assignment (LSA) optimization problem, which avoids tedious hand annotation of the training data. Additionally, posing the loss calculation as multi-label classification permits supervising on point correspondences directly instead of on odometry error, which is not feasible for sparse and noisy data from the SoC radar we use. We evaluate our method with an open-source state-of-the-art Radar-Inertial Odometry (RIO) framework in real-world Unmanned Aerial Vehicle (UAV) flights and with the widely used public Coloradar dataset. Evaluation shows that the proposed method improves the position estimation accuracy by over 14 % and 19 % on average, respectively. The open source code and datasets can be found here: https://github.com/aau-cns/radar_transformer.
☆ Design, fabrication and control of a cable-driven parallel robot
In cable driven parallel robots (CDPRs), the payload is suspended using a network of cables whose length can be controlled to maneuver the payload within the workspace. Compared to rigid link robots, CDPRs provide better maneuverability due to the flexibility of the cables and consume lesser power due to the high strength-to-weight ratio of the cables. However, amongst other things, the flexibility of the cables and the fact that they can only pull (and not push) render the dynamics of CDPRs complex. Hence advanced modelling paradigms and control algorithms must be developed to fully utilize the potential of CDPRs. Furthermore, given the complex dynamics of CDPRs, the models and control algorithms proposed for them must be validated on experimental setups to ascertain their efficacy in practice. We have recently developed an elaborate experimental setup for a CDPR with three cables and validated elementary open-loop motion planning algorithms on it. In this paper, we describe several aspects of the design and fabrication of our setup, including component selection and assembly, and present our experimental results. Our setup can reproduce complex phenomenon such as the transverse vibration of the cables seen in large CDPRs and will in the future be used to model and control such phenomenon and also to validate more sophisticated motion planning algorithms.
comment: 4 pages, 8 fugures
☆ Mirror Eyes: Explainable Human-Robot Interaction at a Glance
The gaze of a person tends to reflect their interest. This work explores what happens when this statement is taken literally and applied to robots. Here we present a robot system that employs a moving robot head with a screen-based eye model that can direct the robot's gaze to points in physical space and present a reflection-like mirror image of the attended region on top of each eye. We conducted a user study with 33 participants, who were asked to instruct the robot to perform pick-and-place tasks, monitor the robot's task execution, and interrupt it in case of erroneous actions. Despite a deliberate lack of instructions about the role of the eyes and a very brief system exposure, participants felt more aware about the robot's information processing, detected erroneous actions earlier, and rated the user experience higher when eye-based mirroring was enabled compared to non-reflective eyes. These results suggest a beneficial and intuitive utilization of the introduced method in cooperative human-robot interaction.
comment: Accepted to the 34th IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)
☆ A Motivational Architecture for Open-Ended Learning Challenges in Robots
Developing agents capable of autonomously interacting with complex and dynamic environments, where task structures may change over time and prior knowledge cannot be relied upon, is a key prerequisite for deploying artificial systems in real-world settings. The open-ended learning framework identifies the core challenges for creating such agents, including the ability to autonomously generate new goals, acquire the necessary skills (or curricula of skills) to achieve them, and adapt to non-stationary environments. While many existing works tackles various aspects of these challenges in isolation, few propose integrated solutions that address them simultaneously. In this paper, we introduce H-GRAIL, a hierarchical architecture that, through the use of different typologies of intrinsic motivations and interconnected learning mechanisms, autonomously discovers new goals, learns the required skills for their achievement, generates skill sequences for tackling interdependent tasks, and adapts to non-stationary environments. We tested H-GRAIL in a real robotic scenario, demonstrating how the proposed solutions effectively address the various challenges of open-ended learning.
comment: Accepted to RLDM 2025
☆ GraspMAS: Zero-Shot Language-driven Grasp Detection with Multi-Agent System IROS 2025
Language-driven grasp detection has the potential to revolutionize human-robot interaction by allowing robots to understand and execute grasping tasks based on natural language commands. However, existing approaches face two key challenges. First, they often struggle to interpret complex text instructions or operate ineffectively in densely cluttered environments. Second, most methods require a training or finetuning step to adapt to new domains, limiting their generation in real-world applications. In this paper, we introduce GraspMAS, a new multi-agent system framework for language-driven grasp detection. GraspMAS is designed to reason through ambiguities and improve decision-making in real-world scenarios. Our framework consists of three specialized agents: Planner, responsible for strategizing complex queries; Coder, which generates and executes source code; and Observer, which evaluates the outcomes and provides feedback. Intensive experiments on two large-scale datasets demonstrate that our GraspMAS significantly outperforms existing baselines. Additionally, robot experiments conducted in both simulation and real-world settings further validate the effectiveness of our approach.
comment: 8 pages, accepted to IROS 2025
☆ Radar and Event Camera Fusion for Agile Robot Ego-Motion Estimation
Achieving reliable ego motion estimation for agile robots, e.g., aerobatic aircraft, remains challenging because most robot sensors fail to respond timely and clearly to highly dynamic robot motions, often resulting in measurement blurring, distortion, and delays. In this paper, we propose an IMU-free and feature-association-free framework to achieve aggressive ego-motion velocity estimation of a robot platform in highly dynamic scenarios by combining two types of exteroceptive sensors, an event camera and a millimeter wave radar, First, we used instantaneous raw events and Doppler measurements to derive rotational and translational velocities directly. Without a sophisticated association process between measurement frames, the proposed method is more robust in texture-less and structureless environments and is more computationally efficient for edge computing devices. Then, in the back-end, we propose a continuous-time state-space model to fuse the hybrid time-based and event-based measurements to estimate the ego-motion velocity in a fixed-lagged smoother fashion. In the end, we validate our velometer framework extensively in self-collected experiment datasets. The results indicate that our IMU-free and association-free ego motion estimation framework can achieve reliable and efficient velocity output in challenging environments. The source code, illustrative video and dataset are available at https://github.com/ZzhYgwh/TwistEstimator.
☆ Integrating Maneuverable Planning and Adaptive Control for Robot Cart-Pushing under Disturbances
Precise and flexible cart-pushing is a challenging task for mobile robots. The motion constraints during cart-pushing and the robot's redundancy lead to complex motion planning problems, while variable payloads and disturbances present complicated dynamics. In this work, we propose a novel planning and control framework for flexible whole-body coordination and robust adaptive control. Our motion planning method employs a local coordinate representation and a novel kinematic model to solve a nonlinear optimization problem, thereby enhancing motion maneuverability by generating feasible and flexible push poses. Furthermore, we present a disturbance rejection control method to resist disturbances and reduce control errors for the complex control problem without requiring an accurate dynamic model. We validate our method through extensive experiments in simulation and real-world settings, demonstrating its superiority over existing approaches. To the best of our knowledge, this is the first work to systematically evaluate the flexibility and robustness of cart-pushing methods in experiments. The video supplement is available at https://sites.google.com/view/mpac-pushing/.
comment: 11 pages, 11 figures
☆ Robots and Children that Learn Together : Improving Knowledge Retention by Teaching Peer-Like Interactive Robots
Despite growing interest in Learning-by-Teaching (LbT), few studies have explored how this paradigm can be implemented with autonomous, peer-like social robots in real classrooms. Most prior work has relied on scripted or Wizard-of-Oz behaviors, limiting our understanding of how real-time, interactive learning can be supported by artificial agents. This study addresses this gap by introducing Interactive Reinforcement Learning (RL) as a cognitive model for teachable social robots. We conducted two between-subject experiments with 58 primary school children, who either taught a robot or practiced independently on a tablet while learning French vocabulary (memorization) and grammatical rules (inference). The robot, powered by Interactive RL, learned from the child's evaluative feedback. Children in the LbT condition achieved significantly higher retention gains compared to those in the self-practice condition, especially on the grammar task. Learners with lower prior knowledge benefited most from teaching the robot. Behavioural metrics revealed that children adapted their teaching strategies over time and engaged more deeply during inference tasks. This work makes two contributions: (1) it introduces Interactive RL as a pedagogically effective and scalable model for peer-robot learning, and (2) it demonstrates, for the first time, the feasibility of deploying multiple autonomous robots simultaneously in real classrooms. These findings extend theoretical understanding of LbT by showing that social robots can function not only as passive tutees but as adaptive partners that enhance meta-cognitive engagement and long-term learning outcomes.
☆ Robotic Manipulation of a Rotating Chain with Bottom End Fixed
This paper studies the problem of using a robot arm to manipulate a uniformly rotating chain with its bottom end fixed. Existing studies have investigated ideal rotational shapes for practical applications, yet they do not discuss how these shapes can be consistently achieved through manipulation planning. Our work presents a manipulation strategy for stable and consistent shape transitions. We find that the configuration space of such a chain is homeomorphic to a three-dimensional cube. Using this property, we suggest a strategy to manipulate the chain into different configurations, specifically from one rotation mode to another, while taking stability and feasibility into consideration. We demonstrate the effectiveness of our strategy in physical experiments by successfully transitioning from rest to the first two rotation modes. The concepts explored in our work has critical applications in ensuring safety and efficiency of drill string and yarn spinning operations.
comment: 6 pages, 5 figures
☆ TritonZ: A Remotely Operated Underwater Rover with Manipulator Arm for Exploration and Rescue Operations
The increasing demand for underwater exploration and rescue operations enforces the development of advanced wireless or semi-wireless underwater vessels equipped with manipulator arms. This paper presents the implementation of a semi-wireless underwater vehicle, "TritonZ" equipped with a manipulator arm, tailored for effective underwater exploration and rescue operations. The vehicle's compact design enables deployment in different submarine surroundings, addressing the need for wireless systems capable of navigating challenging underwater terrains. The manipulator arm can interact with the environment, allowing the robot to perform sophisticated tasks during exploration and rescue missions in emergency situations. TritonZ is equipped with various sensors such as Pi-Camera, Humidity, and Temperature sensors to send real-time environmental data. Our underwater vehicle controlled using a customized remote controller can navigate efficiently in the water where Pi-Camera enables live streaming of the surroundings. Motion control and video capture are performed simultaneously using this camera. The manipulator arm is designed to perform various tasks, similar to grasping, manipulating, and collecting underwater objects. Experimental results shows the efficacy of the proposed remotely operated vehicle in performing a variety of underwater exploration and rescue tasks. Additionally, the results show that TritonZ can maintain an average of 13.5cm/s with a minimal delay of 2-3 seconds. Furthermore, the vehicle can sustain waves underwater by maintaining its position as well as average velocity. The full project details and source code can be accessed at this link: https://github.com/kawser-ahmed-byte/TritonZ
comment: 6 pages, 5 figures
☆ Crowdsourcing Ubiquitous Indoor Localization with Non-Cooperative Wi-Fi Ranging
Indoor localization opens the path to potentially transformative applications. Although many indoor localization methods have been proposed over the years, they remain too impractical for widespread deployment in the real world. In this paper, we introduce PeepLoc, a deployable and scalable Wi-Fi-based solution for indoor localization that relies only on pre-existing devices and infrastructure. Specifically, PeepLoc works on any mobile device with an unmodified Wi-Fi transceiver and in any indoor environment with a sufficient number of Wi-Fi access points (APs) and pedestrian traffic. At the core of PeepLoc is (a) a mechanism which allows any Wi-Fi device to obtain non-cooperative time-of-flight (ToF) to any Wi-Fi AP and (b) a novel bootstrapping mechanism that relies on pedestrian dead reckoning (PDR) and crowdsourcing to opportunistically initialize pre-existing APs as anchor points within an environment. We implement PeepLoc using commodity hardware and evaluate it extensively across 4 campus buildings. We show PeepLoc leads to a mean and median positional error of 3.41 m and 3.06 m respectively, which is superior to existing deployed indoor localization systems and is competitive with commodity GPS in outdoor environments.
☆ Improvement on LiDAR-Camera Calibration Using Square Targets
Precise sensor calibration is critical for autonomous vehicles as a prerequisite for perception algorithms to function properly. Rotation error of one degree can translate to position error of meters in target object detection at large distance, leading to improper reaction of the system or even safety related issues. Many methods for multi-sensor calibration have been proposed. However, there are very few work that comprehensively consider the challenges of the calibration procedure when applied to factory manufacturing pipeline or after-sales service scenarios. In this work, we introduce a fully automatic LiDAR-camera extrinsic calibration algorithm based on targets that is fast, easy to deploy and robust to sensor noises such as missing data. The core of the method include: (1) an automatic multi-stage LiDAR board detection pipeline using only geometry information with no specific material requirement; (2) a fast coarse extrinsic parameter search mechanism that is robust to initial extrinsic errors; (3) a direct optimization algorithm that is robust to sensor noises. We validate the effectiveness of our methods through experiments on data captured in real world scenarios.
☆ Learning Approach to Efficient Vision-based Active Tracking of a Flying Target by an Unmanned Aerial Vehicle
Autonomous tracking of flying aerial objects has important civilian and defense applications, ranging from search and rescue to counter-unmanned aerial systems (counter-UAS). Ground based tracking requires setting up infrastructure, could be range limited, and may not be feasible in remote areas, crowded cities or in dense vegetation areas. Vision based active tracking of aerial objects from another airborne vehicle, e.g., a chaser unmanned aerial vehicle (UAV), promises to fill this important gap, along with serving aerial coordination use cases. Vision-based active tracking by a UAV entails solving two coupled problems: 1) compute-efficient and accurate (target) object detection and target state estimation; and 2) maneuver decisions to ensure that the target remains in the field of view in the future time-steps and favorably positioned for continued detection. As a solution to the first problem, this paper presents a novel integration of standard deep learning based architectures with Kernelized Correlation Filter (KCF) to achieve compute-efficient object detection without compromising accuracy, unlike standalone learning or filtering approaches. The proposed perception framework is validated using a lab-scale setup. For the second problem, to obviate the linearity assumptions and background variations limiting effectiveness of the traditional controllers, we present the use of reinforcement learning to train a neuro-controller for fast computation of velocity maneuvers. New state space, action space and reward formulations are developed for this purpose, and training is performed in simulation using AirSim. The trained model is also tested in AirSim with respect to complex target maneuvers, and is found to outperform a baseline PID control in terms of tracking up-time and average distance maintained (from the target) during tracking.
comment: AIAA Aviation 2025
☆ Robot Tactile Gesture Recognition Based on Full-body Modular E-skin
With the development of robot electronic skin technology, various tactile sensors, enhanced by AI, are unlocking a new dimension of perception for robots. In this work, we explore how robots equipped with electronic skin can recognize tactile gestures and interpret them as human commands. We developed a modular robot E-skin, composed of multiple irregularly shaped skin patches, which can be assembled to cover the robot's body while capturing real-time pressure and pose data from thousands of sensing points. To process this information, we propose an equivariant graph neural network-based recognizer that efficiently and accurately classifies diverse tactile gestures, including poke, grab, stroke, and double-pat. By mapping the recognized gestures to predefined robot actions, we enable intuitive human-robot interaction purely through tactile input.
☆ Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning
Large vision-language models (VLMs) for autonomous driving (AD) are evolving beyond perception and cognition tasks toward motion planning. However, we identify two critical challenges in this direction: (1) VLMs tend to learn shortcuts by relying heavily on history input information, achieving seemingly strong planning results without genuinely understanding the visual inputs; and (2) the chain-ofthought (COT) reasoning processes are always misaligned with the motion planning outcomes, and how to effectively leverage the complex reasoning capability to enhance planning remains largely underexplored. In this paper, we start from a small-scale domain-specific VLM and propose Drive-R1 designed to bridges the scenario reasoning and motion planning for AD. Drive-R1 first undergoes the supervised finetuning on a elaborate dataset containing both long and short COT data. Drive-R1 is encouraged to reason step-by-step from visual input to final planning decisions. Subsequently, Drive-R1 is trained within a reinforcement learning framework that incentivizes the discovery of reasoning paths that are more informative for planning, guided by rewards based on predicted trajectories and meta actions. Experimental evaluations on the nuScenes and DriveLM-nuScenes benchmarks demonstrate that Drive-R1 achieves superior performance compared to existing state-of-the-art VLMs. We believe that Drive-R1 presents a promising direction for bridging reasoning and planning in AD, offering methodological insights for future research and applications.
☆ Haptic-ACT -- Pseudo Oocyte Manipulation by a Robot Using Multimodal Information and Action Chunking with Transformers IROS2025
In this paper we introduce Haptic-ACT, an advanced robotic system for pseudo oocyte manipulation, integrating multimodal information and Action Chunking with Transformers (ACT). Traditional automation methods for oocyte transfer rely heavily on visual perception, often requiring human supervision due to biological variability and environmental disturbances. Haptic-ACT enhances ACT by incorporating haptic feedback, enabling real-time grasp failure detection and adaptive correction. Additionally, we introduce a 3D-printed TPU soft gripper to facilitate delicate manipulations. Experimental results demonstrate that Haptic-ACT improves the task success rate, robustness, and adaptability compared to conventional ACT, particularly in dynamic environments. These findings highlight the potential of multimodal learning in robotics for biomedical automation.
comment: Accepted at IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2025) Project website https://upedrou.github.io/haptic-act_IROS2025
♻ ☆ Stochastic Motion Planning as Gaussian Variational Inference: Theory and Algorithms
We present a novel formulation for motion planning under uncertainties based on variational inference where the optimal motion plan is modeled as a posterior distribution. We propose a Gaussian variational inference-based framework, termed Gaussian Variational Inference Motion Planning (GVI-MP), to approximate this posterior by a Gaussian distribution over the trajectories. We show that the GVI-MP framework is dual to a special class of stochastic control problems and brings robustness into the decision-making in motion planning. We develop two algorithms to numerically solve this variational inference and the equivalent control formulations for motion planning. The first algorithm uses a natural gradient paradigm to iteratively update a Gaussian proposal distribution on the sparse motion planning factor graph. We propose a second algorithm, the Proximal Covariance Steering Motion Planner (PCS-MP), to solve the same inference problem in its stochastic control form with an additional terminal constraint. We leverage a proximal gradient paradigm where, at each iteration, we quadratically approximate nonlinear state costs and solve a linear covariance steering problem in closed form. The efficacy of the proposed algorithms is demonstrated through extensive experiments on various robot models. An implementation is provided in https://github.com/hzyu17/VIMP.
comment: 20 pages
♻ ☆ Learning Realistic Joint Space Boundaries for Range of Motion Analysis of Healthy and Impaired Human Arms
A realistic human kinematic model that satisfies anatomical constraints is essential for human-robot interaction, biomechanics and robot-assisted rehabilitation. Modeling realistic joint constraints, however, is challenging as human arm motion is constrained by joint limits, inter- and intra-joint dependencies, self-collisions, individual capabilities and muscular or neurological constraints which are difficult to represent. Hence, physicians and researchers have relied on simple box-constraints, ignoring important anatomical factors. In this paper, we propose a data-driven method to learn realistic anatomically constrained upper-limb range of motion (RoM) boundaries from motion capture data. This is achieved by fitting a one-class support vector machine to a dataset of upper-limb joint space exploration motions with an efficient hyper-parameter tuning scheme. Our approach outperforms similar works focused on valid RoM learning. Further, we propose an impairment index (II) metric that offers a quantitative assessment of capability/impairment when comparing healthy and impaired arms. We validate the metric on healthy subjects physically constrained to emulate hemiplegia and different disability levels as stroke patients. [https://sites.google.com/seas.upenn.edu/learning-rom]
♻ ☆ Experimental Setup and Software Pipeline to Evaluate Optimization based Autonomous Multi-Robot Search Algorithms
Signal source localization has been a problem of interest in the multi-robot systems domain given its applications in search & rescue and hazard localization in various industrial and outdoor settings. A variety of multi-robot search algorithms exist that usually formulate and solve the associated autonomous motion planning problem as a heuristic model-free or belief model-based optimization process. Most of these algorithms however remains tested only in simulation, thereby losing the opportunity to generate knowledge about how such algorithms would compare/contrast in a real physical setting in terms of search performance and real-time computing performance. To address this gap, this paper presents a new lab-scale physical setup and associated open-source software pipeline to evaluate and benchmark multi-robot search algorithms. The presented physical setup innovatively uses an acoustic source (that is safe and inexpensive) and small ground robots (e-pucks) operating in a standard motion-capture environment. This setup can be easily recreated and used by most robotics researchers. The acoustic source also presents interesting uncertainty in terms of its noise-to-signal ratio, which is useful to assess sim-to-real gaps. The overall software pipeline is designed to readily interface with any multi-robot search algorithm with minimal effort and is executable in parallel asynchronous form. This pipeline includes a framework for distributed implementation of multi-robot or swarm search algorithms, integrated with a ROS (Robotics Operating System)-based software stack for motion capture supported localization. The utility of this novel setup is demonstrated by using it to evaluate two state-of-the-art multi-robot search algorithms, based on swarm optimization and batch-Bayesian Optimization (called Bayes-Swarm), as well as a random walk baseline.
comment: IDETC 2025
♻ ☆ cuVSLAM: CUDA accelerated visual odometry and mapping
Accurate and robust pose estimation is a key requirement for any autonomous robot. We present cuVSLAM, a state-of-the-art solution for visual simultaneous localization and mapping, which can operate with a variety of visual-inertial sensor suites, including multiple RGB and depth cameras, and inertial measurement units. cuVSLAM supports operation with as few as one RGB camera to as many as 32 cameras, in arbitrary geometric configurations, thus supporting a wide range of robotic setups. cuVSLAM is specifically optimized using CUDA to deploy in real-time applications with minimal computational overhead on edge-computing devices such as the NVIDIA Jetson. We present the design and implementation of cuVSLAM, example use cases, and empirical results on several state-of-the-art benchmarks demonstrating the best-in-class performance of cuVSLAM.
♻ ☆ Terrain-aware Low Altitude Path Planning
In this paper, we study the problem of generating low-altitude path plans for nap-of-the-earth (NOE) flight in real time with only RGB images from onboard cameras and the vehicle pose. We propose a novel training method that combines behavior cloning and self-supervised learning, where the self-supervision component allows the learned policy to refine the paths generated by the expert planner. Simulation studies show 24.7% reduction in average path elevation compared to the standard behavior cloning approach.
♻ ☆ Why Sample Space Matters: Keyframe Sampling Optimization for LiDAR-based Place Recognition
Recent advances in robotics are driving real-world autonomy for long-term and large-scale missions, where loop closures via place recognition are vital for mitigating pose estimation drift. However, achieving real-time performance remains challenging for resource-constrained mobile robots and multi-robot systems due to the computational burden of high-density sampling, which increases the complexity of comparing and verifying query samples against a growing map database. Conventional methods often retain redundant information or miss critical data by relying on fixed sampling intervals or operating in 3-D space instead of the descriptor feature space. To address these challenges, we introduce the concept of sample space and propose a novel keyframe sampling approach for LiDAR-based place recognition. Our method minimizes redundancy while preserving essential information in the hyper-dimensional descriptor space, supporting both learning-based and handcrafted descriptors. The proposed approach incorporates a sliding window optimization strategy to ensure efficient keyframe selection and real-time performance, enabling seamless integration into robotic pipelines. In sum, our approach demonstrates robust performance across diverse datasets, with the ability to adapt seamlessly from indoor to outdoor scenarios without parameter tuning, reducing loop closure detection times and memory requirements.
comment: The work is no longer intended for consideration in its current form. Readers are instead encouraged to refer to our related and more complete study, arXiv:2501.01791, which should be considered as a stand-alone contribution
♻ ☆ Employing Laban Shape for Generating Emotionally and Functionally Expressive Trajectories in Robotic Manipulators
Successful human-robot collaboration depends on cohesive communication and a precise understanding of the robot's abilities, goals, and constraints. While robotic manipulators offer high precision, versatility, and productivity, they exhibit expressionless and monotonous motions that conceal the robot's intention, resulting in a lack of efficiency and transparency with humans. In this work, we use Laban notation, a dance annotation language, to enable robotic manipulators to generate trajectories with functional expressivity, where the robot uses nonverbal cues to communicate its abilities and the likelihood of succeeding at its task. We achieve this by introducing two novel variants of Hesitant expressive motion (Spoke-Like and Arc-Like). We also enhance the emotional expressivity of four existing emotive trajectories (Happy, Sad, Shy, and Angry) by augmenting Laban Effort usage with Laban Shape. The functionally expressive motions are validated via a human-subjects study, where participants equate both variants of Hesitant motion with reduced robot competency. The enhanced emotive trajectories are shown to be viewed as distinct emotions using the Valence-Arousal-Dominance (VAD) spectrum, corroborating the usage of Laban Shape.
comment: Accepted for presentation at the 2025 IEEE RO-MAN Conference
♻ ☆ Agile, Autonomous Spacecraft Constellations with Disruption Tolerant Networking to Monitor Precipitation and Urban Floods
Fully re-orientable small spacecraft are now supported by commercial technologies, allowing them to point their instruments in any direction and capture images, with short notice. When combined with improved onboard processing, and implemented on a constellation of inter-communicable satellites, this intelligent agility can significantly increase responsiveness to transient or evolving phenomena. We demonstrate a ground-based and onboard algorithmic framework that combines orbital mechanics, attitude control, inter-satellite communication, intelligent prediction and planning to schedule the time-varying, re-orientation of agile, small satellites in a constellation. Planner intelligence is improved by updating the predictive value of future space-time observations based on shared observations of evolving episodic precipitation and urban flood forecasts. Reliable inter-satellite communication within a fast, dynamic constellation topology is modeled in the physical, access control and network layer. We apply the framework on a representative 24-satellite constellation observing 5 global regions. Results show appropriately low latency in information exchange (average within 1/3rd available time for implicit consensus), enabling the onboard scheduler to observe ~7% more flood magnitude than a ground-based implementation. Both onboard and offline versions performed ~98% better than constellations without agility.
♻ ☆ Learning to Insert for Constructive Neural Vehicle Routing Solver
Neural Combinatorial Optimisation (NCO) is a promising learning-based approach for solving Vehicle Routing Problems (VRPs) without extensive manual design. While existing constructive NCO methods typically follow an appending-based paradigm that sequentially adds unvisited nodes to partial solutions, this rigid approach often leads to suboptimal results. To overcome this limitation, we explore the idea of insertion-based paradigm and propose Learning to Construct with Insertion-based Paradigm (L2C-Insert), a novel learning-based method for constructive NCO. Unlike traditional approaches, L2C-Insert builds solutions by strategically inserting unvisited nodes at any valid position in the current partial solution, which can significantly enhance the flexibility and solution quality. The proposed framework introduces three key components: a novel model architecture for precise insertion position prediction, an efficient training scheme for model optimization, and an advanced inference technique that fully exploits the insertion paradigm's flexibility. Extensive experiments on both synthetic and real-world instances of the Travelling Salesman Problem (TSP) and Capacitated Vehicle Routing Problem (CVRP) demonstrate that L2C-Insert consistently achieves superior performance across various problem sizes.
♻ ☆ Context-Aware Human Behavior Prediction Using Multimodal Large Language Models: Challenges and Insights
Predicting human behavior in shared environments is crucial for safe and efficient human-robot interaction. Traditional data-driven methods to that end are pre-trained on domain-specific datasets, activity types, and prediction horizons. In contrast, the recent breakthroughs in Large Language Models (LLMs) promise open-ended cross-domain generalization to describe various human activities and make predictions in any context. In particular, Multimodal LLMs (MLLMs) are able to integrate information from various sources, achieving more contextual awareness and improved scene understanding. The difficulty in applying general-purpose MLLMs directly for prediction stems from their limited capacity for processing large input sequences, sensitivity to prompt design, and expensive fine-tuning. In this paper, we present a systematic analysis of applying pre-trained MLLMs for context-aware human behavior prediction. To this end, we introduce a modular multimodal human activity prediction framework that allows us to benchmark various MLLMs, input variations, In-Context Learning (ICL), and autoregressive techniques. Our evaluation indicates that the best-performing framework configuration is able to reach 92.8% semantic similarity and 66.1% exact label accuracy in predicting human behaviors in the target frame.
comment: Accepted at IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), 2025
♻ ☆ Shaken, Not Stirred: A Novel Dataset for Visual Understanding of Glasses in Human-Robot Bartending Tasks IROS
Datasets for object detection often do not account for enough variety of glasses, due to their transparent and reflective properties. Specifically, open-vocabulary object detectors, widely used in embodied robotic agents, fail to distinguish subclasses of glasses. This scientific gap poses an issue to robotic applications that suffer from accumulating errors between detection, planning, and action execution. The paper introduces a novel method for the acquisition of real-world data from RGB-D sensors that minimizes human effort. We propose an auto-labeling pipeline that generates labels for all the acquired frames based on the depth measurements. We provide a novel real-world glass object dataset that was collected on the Neuro-Inspired COLlaborator (NICOL), a humanoid robot platform. The data set consists of 7850 images recorded from five different cameras. We show that our trained baseline model outperforms state-of-the-art open-vocabulary approaches. In addition, we deploy our baseline model in an embodied agent approach to the NICOL platform, on which it achieves a success rate of 81% in a human-robot bartending scenario.
comment: Submitted and Accepted for Presentation at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025
♻ ☆ Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning
Long-horizon tasks in robotic manipulation present significant challenges in reinforcement learning (RL) due to the difficulty of designing dense reward functions and effectively exploring the expansive state-action space. However, despite a lack of dense rewards, these tasks often have a multi-stage structure, which can be leveraged to decompose the overall objective into manageable subgoals. In this work, we propose DEMO3, a framework that exploits this structure for efficient learning from visual inputs. Specifically, our approach incorporates multi-stage dense reward learning, a bi-phasic training scheme, and world model learning into a carefully designed demonstration-augmented RL framework that strongly mitigates the challenge of exploration in long-horizon tasks. Our evaluations demonstrate that our method improves data-efficiency by an average of 40% and by 70% on particularly difficult tasks compared to state-of-the-art approaches. We validate this across 16 sparse-reward tasks spanning four domains, including challenging humanoid visual control tasks using as few as five demonstrations.
comment: Project page can be found at https://adrialopezescoriza.github.io/demo3/
♻ ☆ Diffusion-based learning of contact plans for agile locomotion
Legged robots have become capable of performing highly dynamic maneuvers in the past few years. However, agile locomotion in highly constrained environments such as stepping stones is still a challenge. In this paper, we propose a combination of model-based control, search, and learning to design efficient control policies for agile locomotion on stepping stones. In our framework, we use nonlinear model predictive control (NMPC) to generate whole-body motions for a given contact plan. To efficiently search for an optimal contact plan, we propose to use Monte Carlo tree search (MCTS). While the combination of MCTS and NMPC can quickly find a feasible plan for a given environment (a few seconds), it is not yet suitable to be used as a reactive policy. Hence, we generate a dataset for optimal goal-conditioned policy for a given scene and learn it through supervised learning. In particular, we leverage the power of diffusion models in handling multi-modality in the dataset. We test our proposed framework on a scenario where our quadruped robot Solo12 successfully jumps to different goals in a highly constrained environment.
♻ ☆ HybridVLA: Collaborative Diffusion and Autoregression in a Unified Vision-Language-Action Model
A fundamental objective of manipulation policy design is to endow robots to comprehend human instructions, reason about scene cues, and execute generalized actions in dynamic environments. Recent autoregressive vision-language-action (VLA) methods inherit common-sense reasoning capabilities from vision-language models (VLMs) for next action-token prediction. However, these methods quantize actions into discrete bins, which disrupts the continuity required for precise control. In contrast, existing diffusion-based VLA methods incorporate an additional diffusion head to predict continuous actions solely conditioned on feature representations extracted by the VLM, without fully leveraging the VLM's pretrained reasoning capabilities through token-level generation. To address these limitations, we introduce HybridVLA, a unified framework that absorbs the continuous nature of diffusion-based actions and the contextual reasoning of autoregression within a single large language model. To mitigate interference between the two generation paradigms, we propose a collaborative training recipe that seamlessly incorporates diffusion denoising into the next-token prediction process. With this recipe, we find these two action prediction methods not only reinforce each other but also exhibit varying strength across different tasks. Therefore, we design a collaborative action ensemble mechanism that adaptively fuses both predictions, leading to more robust control. HybridVLA outperforms previous state-of-the-art VLA methods by 14\% and 19\% in mean success rate on simulation and real-world tasks, respectively, while demonstrating stable manipulation in unseen configurations.
♻ ☆ LoopSR: Looping Sim-and-Real for Lifelong Policy Adaptation of Legged Robots IROS 2025
Reinforcement Learning (RL) has shown its remarkable and generalizable capability in legged locomotion through sim-to-real transfer. However, while adaptive methods like domain randomization are expected to enhance policy robustness across diverse environments, they potentially compromise the policy's performance in any specific environment, leading to suboptimal real-world deployment due to the No Free Lunch theorem. To address this, we propose LoopSR, a lifelong policy adaptation framework that continuously refines RL policies in the post-deployment stage. LoopSR employs a transformer-based encoder to map real-world trajectories into a latent space and reconstruct a digital twin of the real world for further improvement. Autoencoder architecture and contrastive learning methods are adopted to enhance feature extraction of real-world dynamics. Simulation parameters for continual training are derived by combining predicted values from the decoder with retrieved parameters from a pre-collected simulation trajectory dataset. By leveraging simulated continual training, LoopSR achieves superior data efficiency compared with strong baselines, yielding eminent performance with limited data in both sim-to-sim and sim-to-real experiments.
comment: IROS 2025
♻ ☆ SALT: A Flexible Semi-Automatic Labeling Tool for General LiDAR Point Clouds with Cross-Scene Adaptability and 4D Consistency
We propose a flexible Semi-Automatic Labeling Tool (SALT) for general LiDAR point clouds with cross-scene adaptability and 4D consistency. Unlike recent approaches that rely on camera distillation, SALT operates directly on raw LiDAR data, automatically generating pre-segmentation results. To achieve this, we propose a novel zero-shot learning paradigm, termed data alignment, which transforms LiDAR data into pseudo-images by aligning with the training distribution of vision foundation models. Additionally, we design a 4D-consistent prompting strategy and 4D non-maximum suppression module to enhance SAM2, ensuring high-quality, temporally consistent presegmentation. SALT surpasses the latest zero-shot methods by 18.4% PQ on SemanticKITTI and achieves nearly 40-50% of human annotator performance on our newly collected low-resolution LiDAR data and on combined data from three LiDAR types, significantly boosting annotation efficiency. We anticipate that SALT's open-sourcing will catalyze substantial expansion of current LiDAR datasets and lay the groundwork for the future development of LiDAR foundation models. Code is available at https://github.com/Cavendish518/SALT.
♻ ☆ Accurate Simulation and Parameter Identification of Deformable Linear Objects using Discrete Elastic Rods in Generalized Coordinates
This paper presents a fast and accurate model of a deformable linear object (DLO) -- e.g., a rope, wire, or cable -- integrated into an established robot physics simulator, MuJoCo. Most accurate DLO models with low computational times exist in standalone numerical simulators, which are unable or require tedious work to handle external objects. Based on an existing state-of-the-art DLO model -- Discrete Elastic Rods (DER) -- our implementation provides an improvement in accuracy over MuJoCo's own native cable model. To minimize computational load, our model utilizes force-lever analysis to adapt the Cartesian stiffness forces of the DER into its generalized coordinates. As a key contribution, we introduce a novel parameter identification pipeline designed for both simplicity and accuracy, which we utilize to determine the bending and twisting stiffness of three distinct DLOs. We then evaluate the performance of each model by simulating the DLOs and comparing them to their real-world counterparts and against theoretically proven validation tests.
comment: 7 pages, 6 figures
♻ ☆ GAF: Gaussian Action Field as a Dynamic World Model for Robotic Manipulation
Accurate action inference is critical for vision-based robotic manipulation. Existing approaches typically follow either a Vision-to-Action (V-A) paradigm, predicting actions directly from visual inputs, or a Vision-to-3D-to-Action (V-3D-A) paradigm, leveraging intermediate 3D representations. However, these methods often struggle with action inaccuracies due to the complexity and dynamic nature of manipulation scenes. In this paper, we propose a Vision-to-4D-to-Action (V-4D-A) framework that enables direct action reasoning from motion-aware 4D representations via a Gaussian Action Field (GAF). GAF extends 3D Gaussian Splatting (3DGS) by incorporating learnable motion attributes, allowing simultaneous modeling of dynamic scenes and manipulation actions. To learn time-varying scene geometry and action-aware robot motion, GAF supports three key query types: reconstruction of the current scene, prediction of future frames, and estimation of initial action via robot motion. Furthermore, the high-quality current and future frames generated by GAF facilitate manipulation action refinement through a GAF-guided diffusion model. Extensive experiments demonstrate significant improvements, with GAF achieving +11.5385 dB PSNR and -0.5574 LPIPS improvements in reconstruction quality, while boosting the average success rate in robotic manipulation tasks by 10.33% over state-of-the-art methods. Project page: http://chaiying1.github.io/GAF.github.io/project_page/
comment: http://chaiying1.github.io/GAF.github.io/project_page/
♻ ☆ IDCAIS: Inter-Defender Collision-Aware Interception Strategy against Multiple Attackers
In the prior literature on multi-agent area defense games, the assignments of the defenders to the attackers are done based on a cost metric associated only with the interception of the attackers. In contrast to that, this paper presents an Inter-Defender Collision-Aware Interception Strategy (IDCAIS) for defenders to intercept attackers in order to defend a protected area, such that the defender-to-attacker assignment protocol not only takes into account an interception-related cost but also takes into account any possible future collisions among the defenders on their optimal interception trajectories. In particular, in this paper, the defenders are assigned to intercept attackers using a mixed-integer quadratic program (MIQP) that: 1) minimizes the sum of times taken by defenders to capture the attackers under time-optimal control, as well as 2) helps eliminate or delay possible future collisions among the defenders on the optimal trajectories. To prevent inevitable collisions on optimal trajectories or collisions arising due to time-sub-optimal behavior by the attackers, a minimally augmented control using exponential control barrier function (ECBF) is also provided. Simulations show the efficacy of the approach.
comment: 14 pages, 12 figures
Information Retrieval 15
☆ From Web Search towards Agentic Deep Research: Incentivizing Search with Reasoning Agents
Information retrieval is a cornerstone of modern knowledge acquisition, enabling billions of queries each day across diverse domains. However, traditional keyword-based search engines are increasingly inadequate for handling complex, multi-step information needs. Our position is that Large Language Models (LLMs), endowed with reasoning and agentic capabilities, are ushering in a new paradigm termed Agentic Deep Research. These systems transcend conventional information search techniques by tightly integrating autonomous reasoning, iterative retrieval, and information synthesis into a dynamic feedback loop. We trace the evolution from static web search to interactive, agent-based systems that plan, explore, and learn. We also introduce a test-time scaling law to formalize the impact of computational depth on reasoning and search. Supported by benchmark results and the rise of open-source implementations, we demonstrate that Agentic Deep Research not only significantly outperforms existing approaches, but is also poised to become the dominant paradigm for future information seeking. All the related resources, including industry products, research papers, benchmark datasets, and open-source implementations, are collected for the community in https://github.com/DavidZWZ/Awesome-Deep-Research.
☆ An Audio-centric Multi-task Learning Framework for Streaming Ads Targeting on Spotify KDD 2025
Spotify, a large-scale multimedia platform, attracts over 675 million monthly active users who collectively consume millions of hours of music, podcasts, audiobooks, and video content. This diverse content consumption pattern introduces unique challenges for computational advertising, which must effectively integrate a variety of ad modalities, including audio, video, and display, within a single user experience. Traditional ad recommendation models, primarily designed for foregrounded experiences, often struggle to reconcile the platform's inherent audio-centrality with the demands of optimizing ad performance across multiple formats and modalities. To overcome these challenges, we introduce Cross-modal Adaptive Mixture-of-Experts (CAMoE), a novel framework for optimizing click-through rate (CTR) prediction in both audio-centric and multi-modal settings. CAMoE enhances traditional mixture-of-experts models by incorporating modality-aware task grouping, adaptive loss masking, and deep-cross networks (DCN) to capture complex feature interactions within a multi-modal ad ecosystem. Through extensive ablation studies, we demonstrate that this approach achieves near Pareto-optimal performance across audio, video, and display ad formats, significantly improving AUC-PR compared to conventional single-task and content-based multi-task learning baselines. When deployed at scale on Spotify's ad serving platform, CAMoE delivered substantial gains, yielding a 14.5% increase in CTR for audio ads, a 1.3% increase for video ads, and a 4.8% reduction in expected cost-per-click (eCPC) for audio slots.
comment: Accepted at KDD 2025
☆ Harnessing the Power of Reinforcement Learning for Language-Model-Based Information Retriever via Query-Document Co-Augmentation
Recent studies have proposed leveraging Large Language Models (LLMs) as information retrievers through query rewriting. However, for challenging corpora, we argue that enhancing queries alone is insufficient for robust semantic matching; the LLM should also have sufficient understanding of the corpus by directly handling and augmenting the documents themselves. To this end, we present an LLM-based retriever empowered to augment both user queries and corpus documents, with its policy fully explored via reinforcement learning (RL) and minimal human inductive bias. Notably, we find that simply allowing the LLM to modify documents yields little benefit unless paired with our carefully designed bidirectional RL framework, which enables the LLM to simultaneously learn and collaborate on both query and document augmentation policies. A key technical challenge in realizing such a framework lies in jointly updating both policies during training, where the rewards for the two directions depend on each other, making their entangled reward intractable. Our approach addresses this by introducing a reward sampling strategy and a specifically designed RL algorithm that enables effective training with these sampled rewards. Experimental results demonstrate that our approach significantly enhances LLM-based retrieval performance in both sparse and dense settings, particularly in difficult retrieval domains, and achieves strong cross-benchmark generalization. Our code is released at https://github.com/liujm2001/CoAugRetriever.
☆ Rethinking Click Models in Light of Carousel Interfaces: Theory-Based Categorization and Design of Click Models ICTIR 2025
Click models are a well-established for modeling user interactions with web interfaces. Previous work has mainly focused on traditional single-list web search settings; this includes existing surveys that introduced categorizations based on the first generation of probabilistic graphical model (PGM) click models that have become standard. However, these categorizations have become outdated, as their conceptualizations are unable to meaningfully compare PGM with neural network (NN) click models nor generalize to newer interfaces, such as carousel interfaces. We argue that this outdated view fails to adequately explain the fundamentals of click model designs, thus hindering the development of novel click models. This work reconsiders what should be the fundamental concepts in click model design, grounding them - unlike previous approaches - in their mathematical properties. We propose three fundamental key-design choices that explain what statistical patterns a click model can capture, and thus indirectly, what user behaviors they can capture. Based on these choices, we create a novel click model taxonomy that allows a meaningful comparison of all existing click models; this is the first taxonomy of single-list, grid and carousel click models that includes PGMs and NNs. Finally, we show how our conceptualization provides a foundation for future click model design by an example derivation of a novel design for carousel interfaces.
comment: Accepted by ICTIR 2025
☆ When Fine-Tuning Fails: Lessons from MS MARCO Passage Ranking
This paper investigates the counterintuitive phenomenon where fine-tuning pre-trained transformer models degrades performance on the MS MARCO passage ranking task. Through comprehensive experiments involving five model variants-including full parameter fine-tuning and parameter efficient LoRA adaptations-we demonstrate that all fine-tuning approaches underperform the base sentence-transformers/all- MiniLM-L6-v2 model (MRR@10: 0.3026). Our analysis reveals that fine-tuning disrupts the optimal embedding space structure learned during the base model's extensive pre-training on 1 billion sentence pairs, including 9.1 million MS MARCO samples. UMAP visualizations show progressive embedding space flattening, while training dynamics analysis and computational efficiency metrics further support our findings. These results challenge conventional wisdom about transfer learning effectiveness on saturated benchmarks and suggest architectural innovations may be necessary for meaningful improvements.
☆ PERSCEN: Learning Personalized Interaction Pattern and Scenario Preference for Multi-Scenario Matching KDD 2025
With the expansion of business scales and scopes on online platforms, multi-scenario matching has become a mainstream solution to reduce maintenance costs and alleviate data sparsity. The key to effective multi-scenario recommendation lies in capturing both user preferences shared across all scenarios and scenario-aware preferences specific to each scenario. However, existing methods often overlook user-specific modeling, limiting the generation of personalized user representations. To address this, we propose PERSCEN, an innovative approach that incorporates user-specific modeling into multi-scenario matching. PERSCEN constructs a user-specific feature graph based on user characteristics and employs a lightweight graph neural network to capture higher-order interaction patterns, enabling personalized extraction of preferences shared across scenarios. Additionally, we leverage vector quantization techniques to distil scenario-aware preferences from users' behavior sequence within individual scenarios, facilitating user-specific and scenario-aware preference modeling. To enhance efficient and flexible information transfer, we introduce a progressive scenario-aware gated linear unit that allows fine-grained, low-latency fusion. Extensive experiments demonstrate that PERSCEN outperforms existing methods. Further efficiency analysis confirms that PERSCEN effectively balances performance with computational cost, ensuring its practicality for real-world industrial systems.
comment: Accepted by KDD 2025
☆ Bias vs Bias -- Dawn of Justice: A Fair Fight in Recommendation Systems
Recommendation systems play a crucial role in our daily lives by impacting user experience across various domains, including e-commerce, job advertisements, entertainment, etc. Given the vital role of such systems in our lives, practitioners must ensure they do not produce unfair and imbalanced recommendations. Previous work addressing bias in recommendations overlooked bias in certain item categories, potentially leaving some biases unaddressed. Additionally, most previous work on fair re-ranking focused on binary-sensitive attributes. In this paper, we address these issues by proposing a fairness-aware re-ranking approach that helps mitigate bias in different categories of items. This re-ranking approach leverages existing biases to correct disparities in recommendations across various demographic groups. We show how our approach can mitigate bias on multiple sensitive attributes, including gender, age, and occupation. We experimented on three real-world datasets to evaluate the effectiveness of our re-ranking scheme in mitigating bias in recommendations. Our results show how this approach helps mitigate social bias with little to no degradation in performance.
☆ Team LA at SCIDOCA shared task 2025: Citation Discovery via relation-based zero-shot retrieval SC
The Citation Discovery Shared Task focuses on predicting the correct citation from a given candidate pool for a given paragraph. The main challenges stem from the length of the abstract paragraphs and the high similarity among candidate abstracts, making it difficult to determine the exact paper to cite. To address this, we develop a system that first retrieves the top-k most similar abstracts based on extracted relational features from the given paragraph. From this subset, we leverage a Large Language Model (LLM) to accurately identify the most relevant citation. We evaluate our framework on the training dataset provided by the SCIDOCA 2025 organizers, demonstrating its effectiveness in citation prediction.
comment: In the Proceedings of SCIDOCA 2025
☆ Enhancing Document Retrieval in COVID-19 Research: Leveraging Large Language Models for Hidden Relation Extraction SC
In recent years, with the appearance of the COVID-19 pandemic, numerous publications relevant to this disease have been issued. Because of the massive volume of publications, an efficient retrieval system is necessary to provide researchers with useful information if an unexpected pandemic happens so suddenly, like COVID-19. In this work, we present a method to help the retrieval system, the Covrelex-SE system, to provide more high-quality search results. We exploited the power of the large language models (LLMs) to extract the hidden relationships inside the unlabeled publication that cannot be found by the current parsing tools that the system is using. Since then, help the system to have more useful information during retrieval progress.
comment: In the Proceedings of SCIDOCA 2024
☆ LettinGo: Explore User Profile Generation for Recommendation System
User profiling is pivotal for recommendation systems, as it transforms raw user interaction data into concise and structured representations that drive personalized recommendations. While traditional embedding-based profiles lack interpretability and adaptability, recent advances with large language models (LLMs) enable text-based profiles that are semantically richer and more transparent. However, existing methods often adhere to fixed formats that limit their ability to capture the full diversity of user behaviors. In this paper, we introduce LettinGo, a novel framework for generating diverse and adaptive user profiles. By leveraging the expressive power of LLMs and incorporating direct feedback from downstream recommendation tasks, our approach avoids the rigid constraints imposed by supervised fine-tuning (SFT). Instead, we employ Direct Preference Optimization (DPO) to align the profile generator with task-specific performance, ensuring that the profiles remain adaptive and effective. LettinGo operates in three stages: (1) exploring diverse user profiles via multiple LLMs, (2) evaluating profile quality based on their impact in recommendation systems, and (3) aligning the profile generation through pairwise preference data derived from task performance. Experimental results demonstrate that our framework significantly enhances recommendation accuracy, flexibility, and contextual awareness. This work enhances profile generation as a key innovation for next-generation recommendation systems.
comment: 11 pages, 3 figures
☆ Comparative Analysis of Lion and AdamW Optimizers for Cross-Encoder Reranking with MiniLM, GTE, and ModernBERT
Modern information retrieval systems often employ a two-stage pipeline: an efficient initial retrieval stage followed by a computationally intensive reranking stage. Cross-encoders have shown strong effectiveness for reranking due to their deep analysis of query-document pairs. This paper studies the impact of the Lion optimizer, a recent alternative to AdamW, during fine-tuning of cross-encoder rerankers. We fine-tune three transformer models-MiniLM, GTE, and ModernBERT-on the MS MARCO passage ranking dataset using both optimizers. GTE and ModernBERT support extended context lengths (up to 8192 tokens). We evaluate effectiveness using TREC 2019 Deep Learning Track and MS MARCO dev set (MRR@10). Experiments, run on the Modal cloud platform, reveal that ModernBERT with Lion achieves the best NDCG@10 (0.7225) and MAP (0.5121) on TREC DL 2019, while MiniLM with Lion ties ModernBERT for MRR@10 (0.5988) on MS MARCO dev. Lion also provides superior GPU efficiency, improving utilization by 2.67% to 10.33% across models. We analyze performance trends using standard IR metrics and discuss the optimizer's impact on training dynamics across architectures.
♻ ☆ C-SEO Bench: Does Conversational SEO Work?
Large Language Models (LLMs) are transforming search engines into Conversational Search Engines (CSE). Consequently, Search Engine Optimization (SEO) is being shifted into Conversational Search Engine Optimization (C-SEO). We are beginning to see dedicated C-SEO methods for modifying web documents to increase their visibility in CSE responses. However, they are often tested only for a limited breadth of application domains; we do not understand whether certain C-SEO methods would be effective for a broad range of domains. Moreover, existing evaluations consider only a single-actor scenario where only one web document adopts a C-SEO method; in reality, multiple players are likely to competitively adopt the cutting-edge C-SEO techniques, drawing an analogy from the dynamics we have seen in SEO. We present C-SEO Bench, the first benchmark designed to evaluate C-SEO methods across multiple tasks, domains, and number of actors. We consider two search tasks, question answering and product recommendation, with three domains each. We also formalize a new evaluation protocol with varying adoption rates among involved actors. Our experiments reveal that most current C-SEO methods are largely ineffective, contrary to reported results in the literature. Instead, traditional SEO strategies, those aiming to improve the ranking of the source in the LLM context, are significantly more effective. We also observe that as we increase the number of C-SEO adopters, the overall gains decrease, depicting a congested and zero-sum nature of the problem. Our code and data are available at https://github.com/parameterlab/c-seo-bench and https://huggingface.co/datasets/parameterlab/c-seo-bench.
♻ ☆ Affordable AI Assistants with Knowledge Graph of Thoughts
Large Language Models (LLMs) are revolutionizing the development of AI assistants capable of performing diverse tasks across domains. However, current state-of-the-art LLM-driven agents face significant challenges, including high operational costs and limited success rates on complex benchmarks like GAIA. To address these issues, we propose Knowledge Graph of Thoughts (KGoT), an innovative AI assistant architecture that integrates LLM reasoning with dynamically constructed knowledge graphs (KGs). KGoT extracts and structures task-relevant knowledge into a dynamic KG representation, iteratively enhanced through external tools such as math solvers, web crawlers, and Python scripts. Such structured representation of task-relevant knowledge enables low-cost models to solve complex tasks effectively while also minimizing bias and noise. For example, KGoT achieves a 29% improvement in task success rates on the GAIA benchmark compared to Hugging Face Agents with GPT-4o mini. Moreover, harnessing a smaller model dramatically reduces operational costs by over 36x compared to GPT-4o. Improvements for other models (e.g., Qwen2.5-32B and Deepseek-R1-70B) and benchmarks (e.g., SimpleQA) are similar. KGoT offers a scalable, affordable, versatile, and high-performing solution for AI assistants.
♻ ☆ AlzheimerRAG: Multimodal Retrieval Augmented Generation for Clinical Use Cases using PubMed articles
Recent advancements in generative AI have fostered the development of highly adept Large Language Models (LLMs) that integrate diverse data types to empower decision-making. Among these, multimodal retrieval-augmented generation (RAG) applications are promising because they combine the strengths of information retrieval and generative models, enhancing their utility across various domains, including clinical use cases. This paper introduces AlzheimerRAG, a Multimodal RAG application for clinical use cases, primarily focusing on Alzheimer's Disease case studies from PubMed articles. This application incorporates cross-modal attention fusion techniques to integrate textual and visual data processing by efficiently indexing and accessing vast amounts of biomedical literature. Our experimental results, compared to benchmarks such as BioASQ and PubMedQA, have yielded improved performance in the retrieval and synthesis of domain-specific information. We also present a case study using our multimodal RAG in various Alzheimer's clinical scenarios. We infer that AlzheimerRAG can generate responses with accuracy non-inferior to humans and with low rates of hallucination.
♻ ☆ Personalized News Recommendation with Multi-granularity Candidate-aware User Modeling
Matching candidate news with user interests is crucial for personalized news recommendations. Most existing methods can represent a user's reading interests through a single profile based on clicked news, which may not fully capture the diversity of user interests. Although some approaches incorporate candidate news or topic information, they remain insufficient because they neglect the multi-granularity relatedness between candidate news and user interests. To address this, this study proposed a multi-granularity candidate-aware user modeling framework that integrated user interest features across various levels of granularity. It consisted of two main components: candidate news encoding and user modeling. A news textual information extractor and a knowledge-enhanced entity information extractor can capture candidate news features, and word-level, entity-level, and news-level candidate-aware mechanisms can provide a comprehensive representation of user interests. Extensive experiments on a real-world dataset demonstrated that the proposed model could significantly outperform baseline models.
Robotics 4
☆ Integrating LLMs and Digital Twins for Adaptive Multi-Robot Task Allocation in Construction
Multi-robot systems are emerging as a promising solution to the growing demand for productivity, safety, and adaptability across industrial sectors. However, effectively coordinating multiple robots in dynamic and uncertain environments, such as construction sites, remains a challenge, particularly due to unpredictable factors like material delays, unexpected site conditions, and weather-induced disruptions. To address these challenges, this study proposes an adaptive task allocation framework that strategically leverages the synergistic potential of Digital Twins, Integer Programming (IP), and Large Language Models (LLMs). The multi-robot task allocation problem is formally defined and solved using an IP model that accounts for task dependencies, robot heterogeneity, scheduling constraints, and re-planning requirements. A mechanism for narrative-driven schedule adaptation is introduced, in which unstructured natural language inputs are interpreted by an LLM, and optimization constraints are autonomously updated, enabling human-in-the-loop flexibility without manual coding. A digital twin-based system has been developed to enable real-time synchronization between physical operations and their digital representations. This closed-loop feedback framework ensures that the system remains dynamic and responsive to ongoing changes on site. A case study demonstrates both the computational efficiency of the optimization algorithm and the reasoning performance of several LLMs, with top-performing models achieving over 97% accuracy in constraint and parameter extraction. The results confirm the practicality, adaptability, and cross-domain applicability of the proposed methods.
☆ Automated Plan Refinement for Improving Efficiency of Robotic Layup of Composite Sheets
The automation of composite sheet layup is essential to meet the increasing demand for composite materials in various industries. However, draping plans for the robotic layup of composite sheets are not robust. A plan that works well under a certain condition does not work well in a different condition. Changes in operating conditions due to either changes in material properties or working environment may lead a draping plan to exhibit suboptimal performance. In this paper, we present a comprehensive framework aimed at refining plans based on the observed execution performance. Our framework prioritizes the minimization of uncompacted regions while simultaneously improving time efficiency. To achieve this, we integrate human expertise with data-driven decision-making to refine expert-crafted plans for diverse production environments. We conduct experiments to validate the effectiveness of our approach, revealing significant reductions in the number of corrective paths required compared to initial expert-crafted plans. Through a combination of empirical data analysis, action-effectiveness modeling, and search-based refinement, our system achieves superior time efficiency in robotic layup. Experimental results demonstrate the efficacy of our approach in optimizing the layup process, thereby advancing the state-of-the-art in composite manufacturing automation.
♻ ☆ Structured Pneumatic Fingerpads for Actively Tunable Grip Friction
Grip surfaces with tunable friction can actively modify contact conditions, enabling transitions between higher- and lower-friction states for grasp adjustment. Friction can be increased to grip securely and then decreased to gently release (e.g., for handovers) or manipulate in-hand. Recent friction-tuning surface designs using soft pneumatic chambers show good control over grip friction; however, most require complex fabrication processes and/or custom gripper hardware. We present a practical structured fingerpad design for friction tuning that uses less than $1 USD of materials, takes only seconds to repair, and is easily adapted to existing grippers. Our design uses surface morphology changes to tune friction. The fingerpad is actuated by pressurizing its internal chambers, thereby deflecting its flexible grip surface out from or into these chambers. We characterize the friction-tuning capabilities of our design by measuring the shear force required to pull an object from a gripper equipped with two independently actuated fingerpads. Our results show that varying actuation pressure and timing changes the magnitude of friction forces on a gripped object by up to a factor of 2.8. We demonstrate additional features including macro-scale interlocking behaviour and pressure-based object detection.
comment: In Proceedings of the IEEE/RAS International Conference on Soft Robotics (RoboSoft'25), Lausanne, Switzerland, Apr. 22-26, 2025
♻ ☆ Learning to Adapt through Bio-Inspired Gait Strategies for Versatile Quadruped Locomotion
Legged robots must adapt their gait to navigate unpredictable environments, a challenge that animals master with ease. However, most deep reinforcement learning (DRL) approaches to quadruped locomotion rely on a fixed gait, limiting adaptability to changes in terrain and dynamic state. Here we show that integrating three core principles of animal locomotion-gait transition strategies, gait memory and real-time motion adjustments enables a DRL control framework to fluidly switch among multiple gaits and recover from instability, all without external sensing. Our framework is guided by biomechanics-inspired metrics that capture efficiency, stability and system limits, which are unified to inform optimal gait selection. The resulting framework achieves blind zero-shot deployment across diverse, real-world terrains and substantially significantly outperforms baseline controllers. By embedding biological principles into data-driven control, this work marks a step towards robust, efficient and versatile robotic locomotion, highlighting how animal motor intelligence can shape the next generation of adaptive machines.
comment: 19 pages, 8 figures, journal paper
Information Retrieval 8
☆ LLM-Enhanced Multimodal Fusion for Cross-Domain Sequential Recommendation
Cross-Domain Sequential Recommendation (CDSR) predicts user behavior by leveraging historical interactions across multiple domains, focusing on modeling cross-domain preferences and capturing both intra- and inter-sequence item relationships. We propose LLM-Enhanced Multimodal Fusion for Cross-Domain Sequential Recommendation (LLM-EMF), a novel and advanced approach that enhances textual information with Large Language Models (LLM) knowledge and significantly improves recommendation performance through the fusion of visual and textual data. Using the frozen CLIP model, we generate image and text embeddings, thereby enriching item representations with multimodal data. A multiple attention mechanism jointly learns both single-domain and cross-domain preferences, effectively capturing and understanding complex user interests across diverse domains. Evaluations conducted on four e-commerce datasets demonstrate that LLM-EMF consistently outperforms existing methods in modeling cross-domain user preferences, thereby highlighting the effectiveness of multimodal data integration and its advantages in enhancing sequential recommendation systems. Our source code will be released.
comment: arXiv admin note: substantial text overlap with arXiv:2504.15085
☆ A GenAI System for Improved FAIR Independent Biological Database Integration
Life sciences research increasingly requires identifying, accessing, and effectively processing data from an ever-evolving array of information sources on the Linked Open Data (LOD) network. This dynamic landscape places a significant burden on researchers, as the quality of query responses depends heavily on the selection and semantic integration of data sources --processes that are often labor-intensive, error-prone, and costly. While the adoption of FAIR (Findable, Accessible, Interoperable, and Reusable) data principles has aimed to address these challenges, barriers to efficient and accurate scientific data processing persist. In this paper, we introduce FAIRBridge, an experimental natural language-based query processing system designed to empower scientists to discover, access, and query biological databases, even when they are not FAIR-compliant. FAIRBridge harnesses the capabilities of AI to interpret query intents, map them to relevant databases described in scientific literature, and generate executable queries via intelligent resource access plans. The system also includes robust tools for mitigating low-quality query processing, ensuring high fidelity and responsiveness in the information delivered. FAIRBridge's autonomous query processing framework enables users to explore alternative data sources, make informed choices at every step, and leverage community-driven crowd curation when needed. By providing a user-friendly, automated hypothesis-testing platform in natural English, FAIRBridge significantly enhances the integration and processing of scientific data, offering researchers a powerful new tool for advancing their inquiries.
♻ ☆ BLAZE: Cross-Language and Cross-Project Bug Localization via Dynamic Chunking and Hard Example Learning
Software bugs require developers to exert significant effort to identify and resolve them, often consuming about one-third of their time. Bug localization, the process of pinpointing the exact source code files that need modification, is crucial in reducing this effort. Existing bug localization tools, typically reliant on deep learning techniques, face limitations in cross-project applicability and effectiveness in multi-language environments. Recent advancements with Large Language Models (LLMs) offer detailed representations for bug localization. However, they encounter challenges with limited context windows and mapping accuracy. To address these issues, we propose BLAZE, an approach that employs dynamic chunking and hard example learning. First, BLAZE dynamically segments source code to minimize continuity loss. Then, BLAZE fine-tunes a GPT-based model using challenging bug cases, in order to enhance cross-project and cross-language bug localization. To support the capability of BLAZE, we create the BEETLEBOX dataset, which comprises 26,321 bugs from 29 large and thriving open-source projects across five different programming languages (Java, C++, Python, Go, and JavaScript). Our evaluations of BLAZE on three benchmark datasets BEETLEBOX, SWE-Bench, and Ye et al. demonstrate substantial improvements compared to six state-of-the-art baselines. Specifically, BLAZE achieves up to an increase of 120% in Top 1 accuracy, 144% in Mean Average Precision (MAP), and 100% in Mean Reciprocal Rank (MRR). An extensive ablation study confirms the contributions of our pipeline components to the overall performance enhancement.
♻ ☆ DiscRec: Disentangled Semantic-Collaborative Modeling for Generative Recommendation
Generative recommendation is emerging as a powerful paradigm that directly generates item predictions, moving beyond traditional matching-based approaches. However, current methods face two key challenges: token-item misalignment, where uniform token-level modeling ignores item-level granularity that is critical for collaborative signal learning, and semantic-collaborative signal entanglement, where collaborative and semantic signals exhibit distinct distributions yet are fused in a unified embedding space, leading to conflicting optimization objectives that limit the recommendation performance. To address these issues, we propose DiscRec, a novel framework that enables Disentangled Semantic-Collaborative signal modeling with flexible fusion for generative Recommendation. First, DiscRec introduces item-level position embeddings, assigned based on indices within each semantic ID, enabling explicit modeling of item structure in input token sequences. Second, DiscRec employs a dual-branch module to disentangle the two signals at the embedding layer: a semantic branch encodes semantic signals using original token embeddings, while a collaborative branch applies localized attention restricted to tokens within the same item to effectively capture collaborative signals. A gating mechanism subsequently fuses both branches while preserving the model's ability to model sequential dependencies. Extensive experiments on four real-world datasets demonstrate that DiscRec effectively decouples these signals and consistently outperforms state-of-the-art baselines. Our codes are available on https://github.com/Ten-Mao/DiscRec.
comment: Fixed the indentation issue in the abstract that caused rendering errors on arXiv
♻ ☆ Leveraging Foundation Models for Content-Based Image Retrieval in Radiology
Content-based image retrieval (CBIR) has the potential to significantly improve diagnostic aid and medical research in radiology. However, current CBIR systems face limitations due to their specialization to certain pathologies, limiting their utility. On the other hand, several vision foundation models have been shown to produce general-purpose visual features. Therefore, in this work, we propose using vision foundation models as powerful and versatile off-the-shelf feature extractors for content-based image retrieval. Our contributions include: (1) benchmarking a diverse set of vision foundation models on an extensive dataset comprising 1.6 million 2D radiological images across four modalities and 161 pathologies; (2) identifying weakly-supervised models, particularly BiomedCLIP, as highly effective, achieving a achieving a P@1 of up to 0.594 (P@3: 0.590, P@5: 0.588, P@10: 0.583), comparable to specialized CBIR systems but without additional training; (3) conducting an in-depth analysis of the impact of index size on retrieval performance; (4) evaluating the quality of embedding spaces generated by different models; and (5) investigating specific challenges associated with retrieving anatomical versus pathological structures. Despite these challenges, our research underscores the vast potential of foundation models for CBIR in radiology, proposing a shift towards versatile, general-purpose medical image retrieval systems that do not require specific tuning. Our code, dataset splits and embeddings are publicly available under https://github.com/MIC-DKFZ/foundation-models-for-cbmir.
♻ ☆ GeAR: Graph-enhanced Agent for Retrieval-augmented Generation ACL 2025
Retrieval-augmented Generation (RAG) relies on effective retrieval capabilities, yet traditional sparse and dense retrievers inherently struggle with multi-hop retrieval scenarios. In this paper, we introduce GeAR, a system that advances RAG performance through two key innovations: (i) an efficient graph expansion mechanism that augments any conventional base retriever, such as BM25, and (ii) an agent framework that incorporates the resulting graph-based retrieval into a multi-step retrieval framework. Our evaluation demonstrates GeAR's superior retrieval capabilities across three multi-hop question answering datasets. Notably, our system achieves state-of-the-art results with improvements exceeding 10% on the challenging MuSiQue dataset, while consuming fewer tokens and requiring fewer iterations than existing multi-step retrieval systems. The project page is available at https://gear-rag.github.io.
comment: ACL 2025 Findings
♻ ☆ LightRetriever: A LLM-based Hybrid Retrieval Architecture with 1000x Faster Query Inference
Large Language Models (LLMs)-based hybrid retrieval uses LLMs to encode queries and documents into low-dimensional dense or high-dimensional sparse vectors. It retrieves documents relevant to search queries based on vector similarities. Documents are pre-encoded offline, while queries arrive in real-time, necessitating an efficient online query encoder. Although LLMs significantly enhance retrieval capabilities, serving deeply parameterized LLMs slows down query inference throughput and increases demands for online deployment resources. In this paper, we propose LightRetriever, a novel LLM-based hybrid retriever with extremely lightweight query encoders. Our method retains a full-sized LLM for document encoding, but reduces the workload of query encoding to no more than an embedding lookup. Compared to serving a full-sized LLM on an H800 GPU, our approach achieves over a 1000x speedup for query inference with GPU acceleration, and even a 20x speedup without GPU. Experiments on large-scale retrieval benchmarks demonstrate that our method generalizes well across diverse retrieval tasks, retaining an average of 95% full-sized performance.
♻ ☆ Text2Struct: A Machine Learning Pipeline for Mining Structured Data from Text
Many analysis and prediction tasks require the extraction of structured data from unstructured texts. However, an annotation scheme and a training dataset have not been available for training machine learning models to mine structured data from text without special templates and patterns. To solve it, this paper presents an end-to-end machine learning pipeline, Text2Struct, including a text annotation scheme, training data processing, and machine learning implementation. We formulated the mining problem as the extraction of metrics and units associated with numerals in the text. Text2Struct was trained and evaluated using an annotated text dataset collected from abstracts of medical publications regarding thrombectomy. In terms of prediction performance, a dice coefficient of 0.82 was achieved on the test dataset. By random sampling, most predicted relations between numerals and entities were well matched to the ground-truth annotations. These results show that Text2Struct is viable for the mining of structured data from text without special templates or patterns. It is anticipated to further improve the pipeline by expanding the dataset and investigating other machine learning models. A code demonstration can be found at: https://github.com/zcc861007/Text2Struct
Information Retrieval 7
☆ Expanding Relevance Judgments for Medical Case-based Retrieval Task with Multimodal LLMs SIGIR 2025
Evaluating Information Retrieval (IR) systems relies on high-quality manual relevance judgments (qrels), which are costly and time-consuming to obtain. While pooling reduces the annotation effort, it results in only partially labeled datasets. Large Language Models (LLMs) offer a promising alternative to reducing reliance on manual judgments, particularly in complex domains like medical case-based retrieval, where relevance assessment requires analyzing both textual and visual information. In this work, we explore using a Multimodal Large Language Model (MLLM) to expand relevance judgments, creating a new dataset of automated judgments. Specifically, we employ Gemini 1.5 Pro on the ImageCLEFmed 2013 case-based retrieval task, simulating human assessment through an iteratively refined, structured prompting strategy that integrates binary scoring, instruction-based evaluation, and few-shot learning. We systematically experimented with various prompt configurations to maximize agreement with human judgments. To evaluate agreement between the MLLM and human judgments, we use Cohen's Kappa, achieving a substantial agreement score of 0.6, comparable to inter-annotator agreement typically observed in multimodal retrieval tasks. Starting from the original 15,028 manual judgments (4.72% relevant) across 35 topics, our MLLM-based approach expanded the dataset by over 37x to 558,653 judgments, increasing relevant annotations to 5,950. On average, each medical case query received 15,398 new annotations, with approximately 99% being non-relevant, reflecting the high sparsity typical in this domain. Our results demonstrate the potential of MLLMs to scale relevance judgment collection, offering a promising direction for supporting retrieval evaluation in medical and multimodal IR tasks.
comment: To appear at the Third Workshop on Large Language Models for Evaluation in Information Retrieval (LLM4Eval 2025), co-located with SIGIR 2025. 9 pages, 2 figures, 5 tables
☆ CARTS: Collaborative Agents for Recommendation Textual Summarization
Current recommendation systems often require some form of textual data summarization, such as generating concise and coherent titles for product carousels or other grouped item displays. While large language models have shown promise in NLP domains for textual summarization, these approaches do not directly apply to recommendation systems, where explanations must be highly relevant to the core features of item sets, adhere to strict word limit constraints. In this paper, we propose CARTS (Collaborative Agents for Recommendation Textual Summarization), a multi-agent LLM framework designed for structured summarization in recommendation systems. CARTS decomposes the task into three stages-Generation Augmented Generation (GAG), refinement circle, and arbitration, where successive agent roles are responsible for extracting salient item features, iteratively refining candidate titles based on relevance and length feedback, and selecting the final title through a collaborative arbitration process. Experiments on large-scale e-commerce data and live A/B testing show that CARTS significantly outperforms single-pass and chain-of-thought LLM baselines, delivering higher title relevance and improved user engagement metrics.
☆ Reinforcing User Interest Evolution in Multi-Scenario Learning for recommender systems
In real-world recommendation systems, users would engage in variety scenarios, such as homepages, search pages, and related recommendation pages. Each of these scenarios would reflect different aspects users focus on. However, the user interests may be inconsistent in different scenarios, due to differences in decision-making processes and preference expression. This variability complicates unified modeling, making multi-scenario learning a significant challenge. To address this, we propose a novel reinforcement learning approach that models user preferences across scenarios by modeling user interest evolution across multiple scenarios. Our method employs Double Q-learning to enhance next-item prediction accuracy and optimizes contrastive learning loss using Q-value to make model performance better. Experimental results demonstrate that our approach surpasses state-of-the-art methods in multi-scenario recommendation tasks. Our work offers a fresh perspective on multi-scenario modeling and highlights promising directions for future research.
☆ A novel fast short-time root music method for vibration monitoring of high-speed spindles
Ultra-high-speed spindle bearings challenge traditional vibration monitoring due to broadband noise, non-stationarity, and limited time-frequency resolution. We present a fast Short-Time Root-MUSIC (fSTrM) algorithm that exploits FFT-accelerated Lanczos bidiagonalization to reduce computational complexity from $\mathcal{O}(N^3)$ to $SN\log_2N+S^2(N+S)+M^2(N+M)$ while preserving parametric super-resolution. The method constructs Hankel matrices from 16 ms signal frames and extracts fault frequencies through polynomial rooting on the unit circle. Experimental validation on the Politecnico di Torino bearing dataset demonstrates breakthrough micro-defect detection capabilities. The algorithm reliably identifies 150 $\mu$m defects -- previously undetectable by conventional methods -- providing 72+ hours additional warning time. Compared to STFT and wavelet methods, fSTrM achieves 1.2 Hz frequency resolution (vs. 12.5 Hz), 93\% detection rate at $-$5 dB SNR, and quantifies defect severity through harmonic content analysis. Critically, the algorithm processes each frame in 2.4 ms on embedded ARM Cortex-M7 hardware, enabling real-time deployment. This advancement transforms bearing monitoring from failure prevention to continuous degradation assessment, establishing a new paradigm for predictive maintenance in aerospace and precision machining.
☆ Context-Aware Scientific Knowledge Extraction on Linked Open Data using Large Language Models
The exponential growth of scientific literature challenges researchers extracting and synthesizing knowledge. Traditional search engines return many sources without direct, detailed answers, while general-purpose LLMs may offer concise responses that lack depth or omit current information. LLMs with search capabilities are also limited by context window, yielding short, incomplete answers. This paper introduces WISE (Workflow for Intelligent Scientific Knowledge Extraction), a system addressing these limits by using a structured workflow to extract, refine, and rank query-specific knowledge. WISE uses an LLM-powered, tree-based architecture to refine data, focusing on query-aligned, context-aware, and non-redundant information. Dynamic scoring and ranking prioritize unique contributions from each source, and adaptive stopping criteria minimize processing overhead. WISE delivers detailed, organized answers by systematically exploring and synthesizing knowledge from diverse sources. Experiments on HBB gene-associated diseases demonstrate WISE reduces processed text by over 80% while achieving significantly higher recall over baselines like search engines and other LLM-based approaches. ROUGE and BLEU metrics reveal WISE's output is more unique than other systems, and a novel level-based metric shows it provides more in-depth information. We also explore how the WISE workflow can be adapted for diverse domains like drug discovery, material science, and social science, enabling efficient knowledge extraction and synthesis from unstructured scientific papers and web sources.
♻ ☆ Benchmarking and Building Zero-Shot Hindi Retrieval Model with Hindi-BEIR and NLLB-E5
Given the large number of Hindi speakers worldwide, there is a pressing need for robust and efficient information retrieval systems for Hindi. Despite ongoing research, comprehensive benchmarks for evaluating retrieval models in Hindi are lacking. To address this gap, we introduce the Hindi-BEIR benchmark, comprising 15 datasets across seven distinct tasks. We evaluate state-of-the-art multilingual retrieval models on the Hindi-BEIR benchmark, identifying task and domain-specific challenges that impact Hindi retrieval performance. Building on the insights from these results, we introduce NLLB-E5, a multilingual retrieval model that leverages a zero-shot approach to support Hindi without the need for Hindi training data. We believe our contributions, which include the release of the Hindi-BEIR benchmark and the NLLB-E5 model, will prove to be a valuable resource for researchers and promote advancements in multilingual retrieval models.
comment: arXiv admin note: substantial text overlap with arXiv:2408.09437
♻ ☆ LaPuda: LLM-Enabled Policy-Based Query Optimizer for Multi-modal Data PAKDD 2025
Large language model (LLM) has marked a pivotal moment in the field of machine learning and deep learning. Recently its capability for query planning has been investigated, including both single-modal and multi-modal queries. However, there is no work on the query optimization capability of LLM. As a critical (or could even be the most important) step that significantly impacts the execution performance of the query plan, such analysis and attempts should not be missed. From another aspect, existing query optimizers are usually rule-based or rule-based + cost-based, i.e., they are dependent on manually created rules to complete the query plan rewrite/transformation. Given the fact that modern optimizers include hundreds to thousands of rules, designing a multi-modal query optimizer following a similar way is significantly time-consuming since we will have to enumerate as many multi-modal optimization rules as possible, which has not been well addressed today. In this paper, we investigate the query optimization ability of LLM and use LLM to design LaPuda, a novel LLM and Policy based multi-modal query optimizer. Instead of enumerating specific and detailed rules, LaPuda only needs a few abstract policies to guide LLM in the optimization, by which much time and human effort are saved. Furthermore, to prevent LLM from making mistakes or negative optimization, we borrow the idea of gradient descent and propose a guided cost descent (GCD) algorithm to perform the optimization, such that the optimization can be kept in the correct direction. In our evaluation, our methods consistently outperform the baselines in most cases. For example, the optimized plans generated by our methods result in 1~3x higher execution speed than those by the baselines.
comment: Yifan and Haodi contributed equally to the work, accepted by PAKDD 2025
Information Retrieval 18
☆ PreQRAG -- Classify and Rewrite for Enhanced RAG SIGIR 2025
This paper presents the submission of the UDInfo team to the SIGIR 2025 LiveRAG Challenge. We introduce PreQRAG, a Retrieval Augmented Generation (RAG) architecture designed to improve retrieval and generation quality through targeted question preprocessing. PreQRAG incorporates a pipeline that first classifies each input question as either single-document or multi-document type. For single-document questions, we employ question rewriting techniques to improve retrieval precision and generation relevance. For multi-document questions, we decompose complex queries into focused sub-questions that can be processed more effectively by downstream components. This classification and rewriting strategy improves the RAG performance. Experimental evaluation of the LiveRAG Challenge dataset demonstrates the effectiveness of our question-type-aware architecture, with PreQRAG achieving the preliminary second place in Session 2 of the LiveRAG challenge.
comment: 7 pages, SIGIR 2025 LiveRAG
☆ Towards AI Search Paradigm
In this paper, we introduce the AI Search Paradigm, a comprehensive blueprint for next-generation search systems capable of emulating human information processing and decision-making. The paradigm employs a modular architecture of four LLM-powered agents (Master, Planner, Executor and Writer) that dynamically adapt to the full spectrum of information needs, from simple factual queries to complex multi-stage reasoning tasks. These agents collaborate dynamically through coordinated workflows to evaluate query complexity, decompose problems into executable plans, and orchestrate tool usage, task execution, and content synthesis. We systematically present key methodologies for realizing this paradigm, including task planning and tool integration, execution strategies, aligned and robust retrieval-augmented generation, and efficient LLM inference, spanning both algorithmic techniques and infrastructure-level optimizations. By providing an in-depth guide to these foundational components, this work aims to inform the development of trustworthy, adaptive, and scalable AI search systems.
☆ From Drawings to Decisions: A Hybrid Vision-Language Framework for Parsing 2D Engineering Drawings into Structured Manufacturing Knowledge
Efficient and accurate extraction of key information from 2D engineering drawings is essential for advancing digital manufacturing workflows. Such information includes geometric dimensioning and tolerancing (GD&T), measures, material specifications, and textual annotations. Manual extraction is slow and labor-intensive, while generic OCR models often fail due to complex layouts, engineering symbols, and rotated text, leading to incomplete and unreliable outputs. These limitations result in incomplete and unreliable outputs. To address these challenges, we propose a hybrid vision-language framework that integrates a rotation-aware object detection model (YOLOv11-obb) with a transformer-based vision-language parser. Our structured pipeline applies YOLOv11-OBB to localize annotations and extract oriented bounding box (OBB) patches, which are then parsed into structured outputs using a fine-tuned, lightweight vision-language model (VLM). We curate a dataset of 1,367 2D mechanical drawings annotated across nine key categories. YOLOv11-OBB is trained on this dataset to detect OBBs and extract annotation patches. These are parsed using two open-source VLMs: Donut and Florence-2. Both models are lightweight and well-suited for specialized industrial tasks under limited computational overhead. Following fine-tuning of both models on the curated dataset of image patches paired with structured annotation labels, a comparative experiment is conducted to evaluate parsing performance across four key metrics. Donut outperforms Florence-2, achieving 88.5% precision, 99.2% recall, and a 93.5% F1-score, with a hallucination rate of 11.5%. Finally, a case study demonstrates how the extracted structured information supports downstream manufacturing tasks such as process and tool selection, showcasing the practical utility of the proposed framework in modernizing 2D drawing interpretation.
comment: Preprint submitted to Elsevier
☆ Universal Music Representations? Evaluating Foundation Models on World Music Corpora
Foundation models have revolutionized music information retrieval, but questions remain about their ability to generalize across diverse musical traditions. This paper presents a comprehensive evaluation of five state-of-the-art audio foundation models across six musical corpora spanning Western popular, Greek, Turkish, and Indian classical traditions. We employ three complementary methodologies to investigate these models' cross-cultural capabilities: probing to assess inherent representations, targeted supervised fine-tuning of 1-2 layers, and multi-label few-shot learning for low-resource scenarios. Our analysis shows varying cross-cultural generalization, with larger models typically outperforming on non-Western music, though results decline for culturally distant traditions. Notably, our approaches achieve state-of-the-art performance on five out of six evaluated datasets, demonstrating the effectiveness of foundation models for world music understanding. We also find that our targeted fine-tuning approach does not consistently outperform probing across all settings, suggesting foundation models already encode substantial musical knowledge. Our evaluation framework and benchmarking results contribute to understanding how far current models are from achieving universal music representations while establishing metrics for future progress.
comment: Accepted at ISMIR 2025
☆ PersonalAI: Towards digital twins in the graph form
The challenge of personalizing language models, specifically the ability to account for a user's history during interactions, is of significant interest. Despite recent advancements in large language models (LLMs) and Retrieval Augmented Generation that have enhanced the factual base of LLMs, the task of retaining extensive personal information and using it to generate personalized responses remains pertinent. To address this, we propose utilizing external memory in the form of knowledge graphs, which are constructed and updated by the LLM itself. We have expanded upon ideas of AriGraph architecture and for the first time introduced a combined graph featuring both standard edges and two types of hyperedges. Experiments conducted on the TriviaQA, HotpotQA and DiaASQ benchmarks indicates that this approach aids in making the process of graph construction and knowledge extraction unified and robust. Furthermore, we augmented the DiaASQ benchmark by incorporating parameters such as time into dialogues and introducing contradictory statements made by the same speaker at different times. Despite these modifications, the performance of the question-answering system remained robust, demonstrating the proposed architecture's ability to maintain and utilize temporal dependencies.
☆ RAGentA: Multi-Agent Retrieval-Augmented Generation for Attributed Question Answering SIGIR 2025
We present RAGentA, a multi-agent retrieval-augmented generation (RAG) framework for attributed question answering (QA). With the goal of trustworthy answer generation, RAGentA focuses on optimizing answer correctness, defined by coverage and relevance to the question and faithfulness, which measures the extent to which answers are grounded in retrieved documents. RAGentA uses a multi-agent architecture that iteratively filters retrieved documents, generates attributed answers with in-line citations, and verifies completeness through dynamic refinement. Central to the framework is a hybrid retrieval strategy that combines sparse and dense methods, improving Recall@20 by 12.5% compared to the best single retrieval model, resulting in more correct and well-supported answers. Evaluated on a synthetic QA dataset derived from the FineWeb index, RAGentA outperforms standard RAG baselines, achieving gains of 1.09% in correctness and 10.72% in faithfulness. These results demonstrate the effectiveness of the multi-agent architecture and hybrid retrieval in advancing trustworthy QA.
comment: Accepted at SIGIR 2025
☆ Pyramid Mixer: Multi-dimensional Multi-period Interest Modeling for Sequential Recommendation SIGIR'25
Sequential recommendation, a critical task in recommendation systems, predicts the next user action based on the understanding of the user's historical behaviors. Conventional studies mainly focus on cross-behavior modeling with self-attention based methods while neglecting comprehensive user interest modeling for more dimensions. In this study, we propose a novel sequential recommendation model, Pyramid Mixer, which leverages the MLP-Mixer architecture to achieve efficient and complete modeling of user interests. Our method learns comprehensive user interests via cross-behavior and cross-feature user sequence modeling. The mixer layers are stacked in a pyramid way for cross-period user temporal interest learning. Through extensive offline and online experiments, we demonstrate the effectiveness and efficiency of our method, and we obtain a +0.106% improvement in user stay duration and a +0.0113% increase in user active days in the online A/B test. The Pyramid Mixer has been successfully deployed on the industrial platform, demonstrating its scalability and impact in real-world applications.
comment: Accepted by SIGIR'25
☆ Multi-Objective Recommendation in the Era of Generative AI: A Survey of Recent Progress and Future Prospects
With the recent progress in generative artificial intelligence (Generative AI), particularly in the development of large language models, recommendation systems are evolving to become more versatile. Unlike traditional techniques, generative AI not only learns patterns and representations from complex data but also enables content generation, data synthesis, and personalized experiences. This generative capability plays a crucial role in the field of recommendation systems, helping to address the issue of data sparsity and improving the overall performance of recommendation systems. Numerous studies on generative AI have already emerged in the field of recommendation systems. Meanwhile, the current requirements for recommendation systems have surpassed the single utility of accuracy, leading to a proliferation of multi-objective research that considers various goals in recommendation systems. However, to the best of our knowledge, there remains a lack of comprehensive studies on multi-objective recommendation systems based on generative AI technologies, leaving a significant gap in the literature. Therefore, we investigate the existing research on multi-objective recommendation systems involving generative AI to bridge this gap. We compile current research on multi-objective recommendation systems based on generative techniques, categorizing them by objectives. Additionally, we summarize relevant evaluation metrics and commonly used datasets, concluding with an analysis of the challenges and future directions in this domain.
comment: 21 pages
☆ eSapiens: A Real-World NLP Framework for Multimodal Document Understanding and Enterprise Knowledge Processing
We introduce eSapiens, a unified question-answering system designed for enterprise settings, which bridges structured databases and unstructured textual corpora via a dual-module architecture. The system combines a Text-to-SQL planner with a hybrid Retrieval-Augmented Generation (RAG) pipeline, enabling natural language access to both relational data and free-form documents. To enhance answer faithfulness, the RAG module integrates dense and sparse retrieval, commercial reranking, and a citation verification loop that ensures grounding consistency. We evaluate eSapiens on the RAGTruth benchmark across five leading large language models (LLMs), analyzing performance across key dimensions such as completeness, hallucination, and context utilization. Results demonstrate that eSapiens outperforms a FAISS baseline in contextual relevance and generation quality, with optional strict-grounding controls for high-stakes scenarios. This work provides a deployable framework for robust, citation-aware question answering in real-world enterprise applications.
☆ A Simple Contrastive Framework Of Item Tokenization For Generative Recommendation
Generative retrieval-based recommendation has emerged as a promising paradigm aiming at directly generating the identifiers of the target candidates. However, in large-scale recommendation systems, this approach becomes increasingly cumbersome due to the redundancy and sheer scale of the token space. To overcome these limitations, recent research has explored the use of semantic tokens as an alternative to ID tokens, which typically leveraged reconstruction-based strategies, like RQ-VAE, to quantize content embeddings and significantly reduce the embedding size. However, reconstructive quantization aims for the precise reconstruction of each item embedding independently, which conflicts with the goal of generative retrieval tasks focusing more on differentiating among items. Moreover, multi-modal side information of items, such as descriptive text and images, geographical knowledge in location-based recommendation services, has been shown to be effective in improving recommendations by providing richer contexts for interactions. Nevertheless, effectively integrating such complementary knowledge into existing generative recommendation frameworks remains challenging. To overcome these challenges, we propose a novel unsupervised deep quantization exclusively based on contrastive learning, named SimCIT (a Simple Contrastive Item Tokenization framework). Specifically, different from existing reconstruction-based strategies, SimCIT propose to use a learnable residual quantization module to align with the signals from different modalities of the items, which combines multi-modal knowledge alignment and semantic tokenization in a mutually beneficial contrastive learning framework. Extensive experiments across public datasets and a large-scale industrial dataset from various domains demonstrate SimCIT's effectiveness in LLM-based generative recommendation.
comment: 12 pages,7 figures
♻ ☆ Learning Effective Representations for Retrieval Using Self-Distillation with Adaptive Relevance Margins ICTIR'25
Representation-based retrieval models, so-called bi-encoders, estimate the relevance of a document to a query by calculating the similarity of their respective embeddings. Current state-of-the-art bi-encoders are trained using an expensive training regime involving knowledge distillation from a teacher model and batch-sampling. Instead of relying on a teacher model, we contribute a novel parameter-free loss function for self-supervision that exploits the pre-trained language modeling capabilities of the encoder model as a training signal, eliminating the need for batch sampling by performing implicit hard negative mining. We investigate the capabilities of our proposed approach through extensive experiments, demonstrating that self-distillation can match the effectiveness of teacher distillation using only 13.5% of the data, while offering a speedup in training time between 3x and 15x compared to parametrized losses. All code and data is made openly available.
comment: 9 Pages, 5 Tables, 6 Figures; published at ICTIR'25
♻ ☆ ScholarSearch: Benchmarking Scholar Searching Ability of LLMs
Large Language Models (LLMs)' search capabilities have garnered significant attention. Existing benchmarks, such as OpenAI's BrowseComp, primarily focus on general search scenarios and fail to adequately address the specific demands of academic search. These demands include deeper literature tracing and organization, professional support for academic databases, the ability to navigate long-tail academic knowledge, and ensuring academic rigor. Here, we proposed ScholarSearch, the first dataset specifically designed to evaluate the complex information retrieval capabilities of Large Language Models (LLMs) in academic research. ScholarSearch possesses the following key characteristics: Academic Practicality, where question content closely mirrors real academic learning and research environments, avoiding deliberately misleading models; High Difficulty, with answers that are challenging for single models (e.g., Grok DeepSearch or Gemini Deep Research) to provide directly, often requiring at least three deep searches to derive; Concise Evaluation, where limiting conditions ensure answers are as unique as possible, accompanied by clear sources and brief solution explanations, greatly facilitating subsequent audit and verification, surpassing the current lack of analyzed search datasets both domestically and internationally; and Broad Coverage, as the dataset spans at least 15 different academic disciplines. Through ScholarSearch, we expect to more precisely measure and promote the performance improvement of LLMs in complex academic information retrieval tasks. The data is available at: https://huggingface.co/datasets/PKU-DS-LAB/ScholarSearch
♻ ☆ PromptDSI: Prompt-based Rehearsal-free Instance-wise Incremental Learning for Document Retrieval ECML
Differentiable Search Index (DSI) utilizes pre-trained language models to perform indexing and document retrieval via end-to-end learning without relying on external indexes. However, DSI requires full re-training to index new documents, causing significant computational inefficiencies. Continual learning (CL) offers a solution by enabling the model to incrementally update without full re-training. Existing CL solutions in document retrieval rely on memory buffers or generative models for rehearsal, which is infeasible when accessing previous training data is restricted due to privacy concerns. To this end, we introduce PromptDSI, a prompt-based, rehearsal-free continual learning approach for document retrieval. PromptDSI follows the Prompt-based Continual Learning (PCL) framework, using learnable prompts to efficiently index new documents without accessing previous documents or queries. To improve retrieval latency, we remove the initial forward pass of PCL, which otherwise greatly increases training and inference time, with a negligible trade-off in performance. Additionally, we introduce a novel topic-aware prompt pool that employs neural topic embeddings as fixed keys, eliminating the instability of prompt key optimization while maintaining competitive performance with existing PCL prompt pools. In a challenging rehearsal-free continual learning setup, we demonstrate that PromptDSI variants outperform rehearsal-based baselines, match the strong cache-based baseline in mitigating forgetting, and significantly improving retrieval performance on new corpora.
comment: ECML PKDD 2025 Research track. Camera-ready version. Code is available at https://github.com/LouisDo2108/PromptDSI
♻ ☆ Refining music sample identification with a self-supervised graph neural network
Automatic sample identification (ASID), the detection and identification of portions of audio recordings that have been reused in new musical works, is an essential but challenging task in the field of audio query-based retrieval. While a related task, audio fingerprinting, has made significant progress in accurately retrieving musical content under "real world" (noisy, reverberant) conditions, ASID systems struggle to identify samples that have undergone musical modifications. Thus, a system robust to common music production transformations such as time-stretching, pitch-shifting, effects processing, and underlying or overlaying music is an important open challenge. In this work, we propose a lightweight and scalable encoding architecture employing a Graph Neural Network within a contrastive learning framework. Our model uses only 9% of the trainable parameters compared to the current state-of-the-art system while achieving comparable performance, reaching a mean average precision (mAP) of 44.2%. To enhance retrieval quality, we introduce a two-stage approach consisting of an initial coarse similarity search for candidate selection, followed by a cross-attention classifier that rejects irrelevant matches and refines the ranking of retrieved candidates - an essential capability absent in prior models. In addition, because queries in real-world applications are often short in duration, we benchmark our system for short queries using new fine-grained annotations for the Sample100 dataset, which we publish as part of this work.
comment: Accepted at International Conference for Music Information Retrieval (ISMIR) 2025
♻ ☆ Alto: Orchestrating Distributed Compound AI Systems with Nested Ancestry
Compound AI applications chain together subcomponents such as generative language models, document retrievers, and embedding models. Applying traditional systems optimizations such as parallelism and pipelining in compound AI systems is difficult because each component has different constraints in terms of the granularity and type of data that it ingests. New data is often generated during intermediate computations, and text streams may be split into smaller, independent fragments (such as documents to sentences) which may then be re-aggregated at later parts of the computation. Due to this complexity, existing systems to serve compound AI queries do not fully take advantage of parallelism and pipelining opportunities. We present Alto, a framework that automatically optimizes execution of compound AI queries through streaming and parallelism. Bento introduces a new abstraction called nested ancestry, a metadata hierarchy that allows the system to correctly track partial outputs and aggregate data across the heterogeneous constraints of the components of compound AI applications. This metadata is automatically inferred from the programming model, allowing developers to express complex dataflow patterns without needing to reason manually about the details of routing and aggregation. Implementations of four applications in Alto outperform or match implementations in LangGraph, a popular existing AI programming framework. Alto implementations match or improve latency by between 10-30%.
♻ ☆ From Collapse to Stability: A Knowledge-Driven Ensemble Framework for Scaling Up Click-Through Rate Prediction Models
Click-through rate (CTR) prediction plays a crucial role in modern recommender systems. While many existing methods utilize ensemble networks to improve CTR model performance, they typically restrict the ensemble to only two or three sub-networks. Whether increasing the number of sub-networks consistently enhances CTR model performance to align with scaling laws remains unclear. In this paper, we investigate larger ensemble networks and find three inherent limitations in commonly used ensemble methods: (1) performance degradation as the number of sub-networks increases; (2) sharp declines and high variance in sub-network performance; and (3) significant discrepancies between sub-network and ensemble predictions. Meanwhile, we analyze the underlying causes of these limitations from the perspective of dimensional collapse: the collapse within sub-networks becomes increasingly severe as the number of sub-networks grows, leading to a lower knowledge abundance. In this paper, we employ knowledge transfer methods, such as Knowledge Distillation (KD) and Deep Mutual Learning (DML), to address the aforementioned limitations. We find that KD enables CTR models to better follow scaling laws, while DML reduces variance among sub-networks and minimizes discrepancies with ensemble predictions. Furthermore, by combining KD and DML, we propose a model-agnostic and hyperparameter-free Knowledge-Driven Ensemble Framework (KDEF) for CTR Prediction.
♻ ☆ MTGR: Industrial-Scale Generative Recommendation Framework in Meituan
Scaling law has been extensively validated in many domains such as natural language processing and computer vision. In the recommendation system, recent work has adopted generative recommendations to achieve scalability, but their generative approaches require abandoning the carefully constructed cross features of traditional recommendation models. We found that this approach significantly degrades model performance, and scaling up cannot compensate for it at all. In this paper, we propose MTGR (Meituan Generative Recommendation) to address this issue. MTGR is modeling based on the HSTU architecture and can retain the original deep learning recommendation model (DLRM) features, including cross features. Additionally, MTGR achieves training and inference acceleration through user-level compression to ensure efficient scaling. We also propose Group-Layer Normalization (GLN) to enhance the performance of encoding within different semantic spaces and the dynamic masking strategy to avoid information leakage. We further optimize the training frameworks, enabling support for our models with 10 to 100 times computational complexity compared to the DLRM, without significant cost increases. MTGR achieved 65x FLOPs for single-sample forward inference compared to the DLRM model, resulting in the largest gain in nearly two years both offline and online. This breakthrough was successfully deployed on Meituan, the world's largest food delivery platform, where it has been handling the main traffic.
♻ ☆ GenUP: Generative User Profilers as In-Context Learners for Next POI Recommender Systems
Traditional Point-of-Interest (POI) recommendation systems often lack transparency, interpretability, and scrutability due to their reliance on dense vector-based user embeddings. Furthermore, the cold-start problem -- where systems have insufficient data for new users -- limits their ability to generate accurate recommendations. Existing methods often address this by leveraging similar trajectories from other users, but this approach can be computationally expensive and increases the context length for LLM-based methods, making them difficult to scale. To address these limitations, we propose a method that generates natural language (NL) user profiles from large-scale, location-based social network (LBSN) check-ins, utilizing robust personality assessments and behavioral theories. These NL profiles capture user preferences, routines, and behaviors, improving POI prediction accuracy while offering enhanced transparency. By incorporating NL profiles as system prompts to LLMs, our approach reduces reliance on extensive historical data, while remaining flexible, easily updated, and computationally efficient. Our method is not only competitive with other LLM-based methods but is also more scalable for real-world POI recommender systems. Results demonstrate that our approach consistently outperforms baseline methods, offering a more interpretable and resource-efficient solution for POI recommendation systems. Our source code is available at: https://github.com/w11wo/GenUP/.
Information Retrieval 13
☆ Semantic Outlier Removal with Embedding Models and LLMs ACL 2025
Modern text processing pipelines demand robust methods to remove extraneous content while preserving a document's core message. Traditional approaches such as HTML boilerplate extraction or keyword filters often fail in multilingual settings and struggle with context-sensitive nuances, whereas Large Language Models (LLMs) offer improved quality at high computational cost. We introduce SORE (Semantic Outlier Removal), a cost-effective, transparent method that leverages multilingual sentence embeddings and approximate nearest-neighbor search to identify and excise unwanted text segments. By first identifying core content via metadata embedding and then flagging segments that either closely match predefined outlier groups or deviate significantly from the core, SORE achieves near-LLM extraction precision at a fraction of the cost. Experiments on HTML datasets demonstrate that SORE outperforms structural methods and yield high precision in diverse scenarios. Our system is currently deployed in production, processing millions of documents daily across multiple languages while maintaining both efficiency and accuracy. To facilitate reproducibility and further research, we release our implementation and evaluation datasets.
comment: Accepted to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025) Industry Track, 10 pages
☆ Revela: Dense Retriever Learning via Language Modeling
Dense retrievers play a vital role in accessing external and specialized knowledge to augment language models (LMs). Training dense retrievers typically requires annotated query-document pairs, which are costly and hard to obtain in specialized domains such as code-motivating growing interest in self-supervised retriever learning. Since LMs are trained to capture token-level dependencies through a self-supervised learning objective (i.e., next-token prediction), we can analogously cast retrieval as learning dependencies among chunks of tokens. This analogy naturally leads to the question: How can we adapt self-supervised learning objectives in the spirit of language modeling to train retrievers? To answer this question, we introduce Revela, a unified and scalable training framework for self-supervised retriever learning via language modeling. Revela models semantic dependencies among documents by conditioning next-token prediction on both local and cross-document context through an in-batch attention mechanism. This attention is weighted by retriever-computed similarity scores, enabling the retriever to be optimized as part of language modeling. We evaluate Revela on both general-domain (BEIR) and domain-specific (CoIR) benchmarks across various retriever backbones. At a comparable parameter scale, Revela outperforms the previous best method with absolute improvements of 5.2 % (18.3 % relative) and 5.6 % (14.4 % relative) on NDCG@10, respectively, underscoring its effectiveness. Performance increases with model size, highlighting both the scalability of our approach and its promise for self-supervised retriever learning.
☆ Agentic Personalisation of Cross-Channel Marketing Experiences
Consumer applications provide ample opportunities to surface and communicate various forms of content to users. From promotional campaigns for new features or subscriptions, to evergreen nudges for engagement, or personalised recommendations; across e-mails, push notifications, and in-app surfaces. The conventional approach to orchestration for communication relies heavily on labour-intensive manual marketer work, and inhibits effective personalisation of content, timing, frequency, and copy-writing. We formulate this task under a sequential decision-making framework, where we aim to optimise a modular decision-making policy that maximises incremental engagement for any funnel event. Our approach leverages a Difference-in-Differences design for Individual Treatment Effect estimation, and Thompson sampling to balance the explore-exploit trade-off. We present results from a multi-service application, where our methodology has resulted in significant increases to a variety of goal events across several product features, and is currently deployed across 150 million users.
☆ Analyzing the Influence of Knowledge Graph Information on Relation Extraction
We examine the impact of incorporating knowledge graph information on the performance of relation extraction models across a range of datasets. Our hypothesis is that the positions of entities within a knowledge graph provide important insights for relation extraction tasks. We conduct experiments on multiple datasets, each varying in the number of relations, training examples, and underlying knowledge graphs. Our results demonstrate that integrating knowledge graph information significantly enhances performance, especially when dealing with an imbalance in the number of training examples for each relation. We evaluate the contribution of knowledge graph-based features by combining established relation extraction methods with graph-aware Neural Bellman-Ford networks. These features are tested in both supervised and zero-shot settings, demonstrating consistent performance improvements across various datasets.
☆ Neural Prioritisation for Web Crawling ICTIR 2025
Given the vast scale of the Web, crawling prioritisation techniques based on link graph traversal, popularity, link analysis, and textual content are frequently applied to surface documents that are most likely to be valuable. While existing techniques are effective for keyword-based search, both retrieval methods and user search behaviours are shifting from keyword-based matching to natural language semantic matching. The remarkable success of applying semantic matching and quality signals during ranking leads us to hypothesize that crawling could be improved by prioritizing Web pages with high semantic quality. To investigate this, we propose a semantic quality-driven prioritisation technique to enhance the effectiveness of crawling and align the crawler behaviour with recent shift towards natural language search. We embed semantic understanding directly into the crawling process -- leveraging recent neural semantic quality estimators to prioritise the crawling frontier -- with the goal of surfacing content that is semantically rich and valuable for modern search needs. Our experiments on the English subset of ClueWeb22-B and the Researchy Questions query set show that, compared to existing crawling techniques, neural crawling policies significantly improve harvest rate, maxNDCG, and search effectiveness during the early stages of crawling. Meanwhile, crawlers based on our proposed neural policies maintain comparable search performance on keyword queries from the MS MARCO Web Search query set. While this work does not propose a definitive and complete solution, it presents a forward-looking perspective on Web crawling and opens the door to a new line of research on leveraging semantic analysis to effectively align crawlers with the ongoing shift toward natural language search.
comment: Published at ACM ICTIR 2025
☆ GFlowGR: Fine-tuning Generative Recommendation Frameworks with Generative Flow Networks
Generative recommendations (GR), which usually include item tokenizers and generative Large Language Models (LLMs), have demonstrated remarkable success across a wide range of scenarios. The majority of existing research efforts primarily concentrate on developing powerful item tokenizers or advancing LLM decoding strategies to attain superior performance. However, the critical fine-tuning step in GR frameworks, which is essential for adapting LLMs to recommendation data, remains largely unexplored. Current approaches predominantly rely on either the next-token prediction loss of supervised fine-tuning (SFT) or recommendationspecific direct preference optimization (DPO) strategies. Both methods ignore the exploration of possible positive unobserved samples, which is commonly referred to as the exposure bias problem. To mitigate this problem, this paper treats the GR as a multi-step generation task and constructs a GFlowNets-based fine-tuning framework (GFlowGR). The proposed framework integrates collaborative knowledge from traditional recommender systems to create an adaptive trajectory sampler and a comprehensive reward model. Leveraging the diverse generation property of GFlowNets, along with sampling and heuristic weighting techniques, GFlowGR emerges as a promising approach to mitigate the exposure bias problem. Extensive empirical results on two real-world datasets and with two different GR backbones highlight the effectiveness and robustness of GFlowGR.
☆ Vision-Guided Chunking Is All You Need: Enhancing RAG with Multimodal Document Understanding
Retrieval-Augmented Generation (RAG) systems have revolutionized information retrieval and question answering, but traditional text-based chunking methods struggle with complex document structures, multi-page tables, embedded figures, and contextual dependencies across page boundaries. We present a novel multimodal document chunking approach that leverages Large Multimodal Models (LMMs) to process PDF documents in batches while maintaining semantic coherence and structural integrity. Our method processes documents in configurable page batches with cross-batch context preservation, enabling accurate handling of tables spanning multiple pages, embedded visual elements, and procedural content. We evaluate our approach on a curated dataset of PDF documents with manually crafted queries, demonstrating improvements in chunk quality and downstream RAG performance. Our vision-guided approach achieves better accuracy compared to traditional vanilla RAG systems, with qualitative analysis showing superior preservation of document structure and semantic coherence.
comment: 11 pages, 1 Figure, 1 Table
☆ SEP-GCN: Leveraging Similar Edge Pairs with Temporal and Spatial Contexts for Location-Based Recommender Systems SIGIR
Recommender systems play a crucial role in enabling personalized content delivery amidst the challenges of information overload and human mobility. Although conventional methods often rely on interaction matrices or graph-based retrieval, recent approaches have sought to exploit contextual signals such as time and location. However, most existing models focus on node-level representation or isolated edge attributes, underutilizing the relational structure between interactions. We propose SEP-GCN, a novel graph-based recommendation framework that learns from pairs of contextually similar interaction edges, each representing a user-item check-in event. By identifying edge pairs that occur within similar temporal windows or geographic proximity, SEP-GCN augments the user-item graph with contextual similarity links. These links bridge distant but semantically related interactions, enabling improved long-range information propagation. The enriched graph is processed via an edge-aware convolutional mechanism that integrates contextual similarity into the message-passing process. This allows SEP-GCN to model user preferences more accurately and robustly, especially in sparse or dynamic environments. Experiments on benchmark data sets show that SEP-GCN consistently outperforms strong baselines in both predictive accuracy and robustness.
comment: Accepted for ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval (ICTIR) 2025, Padua, Itay
☆ Empowering Graph-based Approximate Nearest Neighbor Search with Adaptive Awareness Capabilities KDD2025
Approximate Nearest Neighbor Search (ANNS) in high-dimensional spaces finds extensive applications in databases, information retrieval, recommender systems, etc. While graph-based methods have emerged as the leading solution for ANNS due to their superior query performance, they still face several challenges, such as struggling with local optima and redundant computations. These issues arise because existing methods (i) fail to fully exploit the topological information underlying the proximity graph G, and (ii) suffer from severe distribution mismatches between the base data and queries in practice. To this end, this paper proposes GATE, high-tier proximity Graph with Adaptive Topology and Query AwarEness, as a lightweight and adaptive module atop the graph-based indexes to accelerate ANNS. Specifically, GATE formulates the critical problem to identify an optimal entry point in the proximity graph for a given query, facilitating faster online search. By leveraging the inherent clusterability of high-dimensional data, GATE first extracts a small set of hub nodes V as candidate entry points. Then, resorting to a contrastive learning-based two-tower model, GATE encodes both the structural semantics underlying G and the query-relevant features into the latent representations of these hub nodes V. A navigation graph index on V is further constructed to minimize the model inference overhead. Extensive experiments demonstrate that GATE achieves a 1.2-2.0X speed-up in query performance compared to state-of-the-art graph-based indexes.
comment: Accecpted by KDD2025
♻ ☆ Learning Multi-Branch Cooperation for Enhanced Click-Through Rate Prediction at Taobao
Existing click-through rate (CTR) prediction works have studied the role of feature interaction through a variety of techniques. Each interaction technique exhibits its own strength, and solely using one type usually constrains the model's capability to capture the complex feature relationships, especially for industrial data with enormous input feature fields. Recent research shows that effective CTR models often combine an MLP network with a dedicated feature interaction network in a two-parallel structure. However, the interplay and cooperative dynamics between different streams or branches remain under-researched. In this work, we introduce a novel Multi-Branch Cooperation Network (MBCnet) which enables multiple branch networks to collaborate with each other for better complex feature interaction modeling. Specifically, MBCnet consists of three branches: the Extensible Feature Grouping and Crossing (EFGC) branch that promotes the model's memorization ability of specific feature fields, the low rank Cross Net branch and Deep branch to enhance explicit and implicit feature crossing for improved generalization. Among these branches, a novel cooperation scheme is proposed based on two principles: Branch co-teaching and moderate differentiation. Branch co-teaching encourages well-learned branches to support poorly-learned ones on specific training samples. Moderate differentiation advocates branches to maintain a reasonable level of difference in their feature representations on the same inputs. This cooperation strategy improves learning through mutual knowledge sharing and boosts the discovery of diverse feature interactions across branches. Experiments on large-scale industrial datasets and online A/B test at Taobao app demonstrate MBCnet's superior performance, delivering a 0.09 point increase in CTR, 1.49% growth in deals, and 1.62% rise in GMV. Core codes are available online.
comment: 14 pages
♻ ☆ Hierarchical Multi-Positive Contrastive Learning for Patent Image Retrieval SIGIR 2025
Patent images are technical drawings that convey information about a patent's innovation. Patent image retrieval systems aim to search in vast collections and retrieve the most relevant images. Despite recent advances in information retrieval, patent images still pose significant challenges due to their technical intricacies and complex semantic information, requiring efficient fine-tuning for domain adaptation. Current methods neglect patents' hierarchical relationships, such as those defined by the Locarno International Classification (LIC) system, which groups broad categories (e.g., "furnishing") into subclasses (e.g., "seats" and "beds") and further into specific patent designs. In this work, we introduce a hierarchical multi-positive contrastive loss that leverages the LIC's taxonomy to induce such relations in the retrieval process. Our approach assigns multiple positive pairs to each patent image within a batch, with varying similarity scores based on the hierarchical taxonomy. Our experimental analysis with various vision and multimodal models on the DeepPatent2 dataset shows that the proposed method enhances the retrieval results. Notably, our method is effective with low-parameter models, which require fewer computational resources and can be deployed on environments with limited hardware.
comment: 5 pages, 3 figures, Accepted as a short paper at the 6th Workshop on Patent Text Mining and Semantic Technologies (PatentSemTech 2025), co-located with SIGIR 2025
♻ ☆ DrunkAgent: Stealthy Memory Corruption in LLM-Powered Recommender Agents
Large language model (LLM)-powered agents are increasingly used in recommender systems (RSs) to achieve personalized behavior modeling, where the memory mechanism plays a pivotal role in enabling the agents to autonomously explore, learn and self-evolve from real-world interactions. However, this very mechanism, serving as a contextual repository, inherently exposes an attack surface for potential adversarial manipulations. Despite its central role, the robustness of agentic RSs in the face of such threats remains largely underexplored. Previous works suffer from semantic mismatches or rely on static embeddings or pre-defined prompts, all of which hinder their applicability to systems with dynamic memory states. This challenge is exacerbated by the black-box nature of commercial RSs. To tackle the above problems, in this paper, we present the first systematic investigation of memory-based vulnerabilities in LLM-powered recommender agents, revealing their security limitations and guiding efforts to strengthen system resilience and trustworthiness. Specifically, we propose a novel black-box attack framework named DrunkAgent. DrunkAgent crafts semantically meaningful adversarial textual triggers for target item promotions and introduces a series of strategies to maximize the trigger effect by corrupting the memory updates during the interactions. The triggers and strategies are optimized on a surrogate model, enabling DrunkAgent transferable and stealthy. Extensive experiments on real-world datasets across diverse agentic RSs, including collaborative filtering, retrieval augmentation and sequential recommendations, demonstrate the generalizability, transferability and stealthiness of DrunkAgent.
♻ ☆ HSTU-BLaIR: Lightweight Contrastive Text Embedding for Generative Recommender KDD 2025
Recent advances in recommender systems have underscored the complementary strengths of generative modeling and pretrained language models. We propose HSTU-BLaIR, a hybrid framework that augments the Hierarchical Sequential Transduction Unit (HSTU)-based generative recommender with BLaIR, a lightweight 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 HSTU-BLaIR on two e-commerce datasets: three subsets from the Amazon Reviews 2023 dataset and the Steam dataset. We compare its performance against both the original HSTU-based recommender and a variant augmented with embeddings from OpenAI's state-of-the-art \texttt{text-embedding-3-large} model. Despite the latter being trained on a substantially larger corpus with significantly more parameters, our lightweight BLaIR-enhanced approach -- pretrained on domain-specific data -- achieves better performance in nearly all cases. Specifically, HSTU-BLaIR outperforms the OpenAI embedding-based variant on all but one metric, where it is marginally lower, and matches it on another. These findings highlight the effectiveness of contrastive text embeddings in compute-efficient recommendation settings.
comment: Accepted at the Workshop on Large Language Models for E-Commerce, KDD 2025. Code available at https://www.github.com/snapfinger/HSTU-BLaIR
Robotics 43
☆ Particle-Grid Neural Dynamics for Learning Deformable Object Models from RGB-D Videos
Modeling the dynamics of deformable objects is challenging due to their diverse physical properties and the difficulty of estimating states from limited visual information. We address these challenges with a neural dynamics framework that combines object particles and spatial grids in a hybrid representation. Our particle-grid model captures global shape and motion information while predicting dense particle movements, enabling the modeling of objects with varied shapes and materials. Particles represent object shapes, while the spatial grid discretizes the 3D space to ensure spatial continuity and enhance learning efficiency. Coupled with Gaussian Splattings for visual rendering, our framework achieves a fully learning-based digital twin of deformable objects and generates 3D action-conditioned videos. Through experiments, we demonstrate that our model learns the dynamics of diverse objects -- such as ropes, cloths, stuffed animals, and paper bags -- from sparse-view RGB-D recordings of robot-object interactions, while also generalizing at the category level to unseen instances. Our approach outperforms state-of-the-art learning-based and physics-based simulators, particularly in scenarios with limited camera views. Furthermore, we showcase the utility of our learned models in model-based planning, enabling goal-conditioned object manipulation across a range of tasks. The project page is available at https://kywind.github.io/pgnd .
comment: Project page: https://kywind.github.io/pgnd
☆ Embodied Web Agents: Bridging Physical-Digital Realms for Integrated Agent Intelligence
AI agents today are mostly siloed - they either retrieve and reason over vast amount of digital information and knowledge obtained online; or interact with the physical world through embodied perception, planning and action - but rarely both. This separation limits their ability to solve tasks that require integrated physical and digital intelligence, such as cooking from online recipes, navigating with dynamic map data, or interpreting real-world landmarks using web knowledge. We introduce Embodied Web Agents, a novel paradigm for AI agents that fluidly bridge embodiment and web-scale reasoning. To operationalize this concept, we first develop the Embodied Web Agents task environments, a unified simulation platform that tightly integrates realistic 3D indoor and outdoor environments with functional web interfaces. Building upon this platform, we construct and release the Embodied Web Agents Benchmark, which encompasses a diverse suite of tasks including cooking, navigation, shopping, tourism, and geolocation - all requiring coordinated reasoning across physical and digital realms for systematic assessment of cross-domain intelligence. Experimental results reveal significant performance gaps between state-of-the-art AI systems and human capabilities, establishing both challenges and opportunities at the intersection of embodied cognition and web-scale knowledge access. All datasets, codes and websites are publicly available at our project page https://embodied-web-agent.github.io/.
☆ Vision in Action: Learning Active Perception from Human Demonstrations
We present Vision in Action (ViA), an active perception system for bimanual robot manipulation. ViA learns task-relevant active perceptual strategies (e.g., searching, tracking, and focusing) directly from human demonstrations. On the hardware side, ViA employs a simple yet effective 6-DoF robotic neck to enable flexible, human-like head movements. To capture human active perception strategies, we design a VR-based teleoperation interface that creates a shared observation space between the robot and the human operator. To mitigate VR motion sickness caused by latency in the robot's physical movements, the interface uses an intermediate 3D scene representation, enabling real-time view rendering on the operator side while asynchronously updating the scene with the robot's latest observations. Together, these design elements enable the learning of robust visuomotor policies for three complex, multi-stage bimanual manipulation tasks involving visual occlusions, significantly outperforming baseline systems.
☆ FindingDory: A Benchmark to Evaluate Memory in Embodied Agents
Large vision-language models have recently demonstrated impressive performance in planning and control tasks, driving interest in their application to real-world robotics. However, deploying these models for reasoning in embodied contexts is limited by their ability to incorporate long-term experience collected across multiple days and represented by vast collections of images. Current VLMs typically struggle to process more than a few hundred images concurrently, highlighting the need for more efficient mechanisms to handle long-term memory in embodied settings. To effectively evaluate these models for long-horizon control, a benchmark must specifically target scenarios where memory is crucial for success. Existing long-video QA benchmarks overlook embodied challenges like object manipulation and navigation, which demand low-level skills and fine-grained reasoning over past interactions. Moreover, effective memory integration in embodied agents involves both recalling relevant historical information and executing actions based on that information, making it essential to study these aspects together rather than in isolation. In this work, we introduce a new benchmark for long-range embodied tasks in the Habitat simulator. This benchmark evaluates memory-based capabilities across 60 tasks requiring sustained engagement and contextual awareness in an environment. The tasks can also be procedurally extended to longer and more challenging versions, enabling scalable evaluation of memory and reasoning. We also present baselines that integrate state-of-the-art VLMs with low level navigation policies, assessing their performance on these memory-intensive tasks and highlight areas for improvement.
comment: Our dataset and code will be made available at: https://findingdory-benchmark.github.io/
☆ GRIM: Task-Oriented Grasping with Conditioning on Generative Examples
Task-Oriented Grasping (TOG) presents a significant challenge, requiring a nuanced understanding of task semantics, object affordances, and the functional constraints dictating how an object should be grasped for a specific task. To address these challenges, we introduce GRIM (Grasp Re-alignment via Iterative Matching), a novel training-free framework for task-oriented grasping. Initially, a coarse alignment strategy is developed using a combination of geometric cues and principal component analysis (PCA)-reduced DINO features for similarity scoring. Subsequently, the full grasp pose associated with the retrieved memory instance is transferred to the aligned scene object and further refined against a set of task-agnostic, geometrically stable grasps generated for the scene object, prioritizing task compatibility. In contrast to existing learning-based methods, GRIM demonstrates strong generalization capabilities, achieving robust performance with only a small number of conditioning examples.
☆ RaCalNet: Radar Calibration Network for Sparse-Supervised Metric Depth Estimation
Dense metric depth estimation using millimeter-wave radar typically requires dense LiDAR supervision, generated via multi-frame projection and interpolation, to guide the learning of accurate depth from sparse radar measurements and RGB images. However, this paradigm is both costly and data-intensive. To address this, we propose RaCalNet, a novel framework that eliminates the need for dense supervision by using sparse LiDAR to supervise the learning of refined radar measurements, resulting in a supervision density of merely around 1% compared to dense-supervised methods. Unlike previous approaches that associate radar points with broad image regions and rely heavily on dense labels, RaCalNet first recalibrates and refines sparse radar points to construct accurate depth priors. These priors then serve as reliable anchors to guide monocular depth prediction, enabling metric-scale estimation without resorting to dense supervision. This design improves structural consistency and preserves fine details. Despite relying solely on sparse supervision, RaCalNet surpasses state-of-the-art dense-supervised methods, producing depth maps with clear object contours and fine-grained textures. Extensive experiments on the ZJU-4DRadarCam dataset and real-world deployment scenarios demonstrate its effectiveness, reducing RMSE by 35.30% and 34.89%, respectively.
comment: 9 pages, 7 figures
☆ Aerial Grasping via Maximizing Delta-Arm Workspace Utilization
The workspace limits the operational capabilities and range of motion for the systems with robotic arms. Maximizing workspace utilization has the potential to provide more optimal solutions for aerial manipulation tasks, increasing the system's flexibility and operational efficiency. In this paper, we introduce a novel planning framework for aerial grasping that maximizes workspace utilization. We formulate an optimization problem to optimize the aerial manipulator's trajectory, incorporating task constraints to achieve efficient manipulation. To address the challenge of incorporating the delta arm's non-convex workspace into optimization constraints, we leverage a Multilayer Perceptron (MLP) to map position points to feasibility probabilities.Furthermore, we employ Reversible Residual Networks (RevNet) to approximate the complex forward kinematics of the delta arm, utilizing efficient model gradients to eliminate workspace constraints. We validate our methods in simulations and real-world experiments to demonstrate their effectiveness.
comment: 8 pages, 7 figures
☆ Real-Time Initialization of Unknown Anchors for UWB-aided Navigation
This paper presents a framework for the real-time initialization of unknown Ultra-Wideband (UWB) anchors in UWB-aided navigation systems. The method is designed for localization solutions where UWB modules act as supplementary sensors. Our approach enables the automatic detection and calibration of previously unknown anchors during operation, removing the need for manual setup. By combining an online Positional Dilution of Precision (PDOP) estimation, a lightweight outlier detection method, and an adaptive robust kernel for non-linear optimization, our approach significantly improves robustness and suitability for real-world applications compared to state-of-the-art. In particular, we show that our metric which triggers an initialization decision is more conservative than current ones commonly based on initial linear or non-linear initialization guesses. This allows for better initialization geometry and subsequently lower initialization errors. We demonstrate the proposed approach on two different mobile robots: an autonomous forklift and a quadcopter equipped with a UWB-aided Visual-Inertial Odometry (VIO) framework. The results highlight the effectiveness of the proposed method with robust initialization and low positioning error. We open-source our code in a C++ library including a ROS wrapper.
☆ SurfAAV: Design and Implementation of a Novel Multimodal Surfing Aquatic-Aerial Vehicle
Despite significant advancements in the research of aquatic-aerial robots, existing configurations struggle to efficiently perform underwater, surface, and aerial movement simultaneously. In this paper, we propose a novel multimodal surfing aquatic-aerial vehicle, SurfAAV, which efficiently integrates underwater navigation, surface gliding, and aerial flying capabilities. Thanks to the design of the novel differential thrust vectoring hydrofoil, SurfAAV can achieve efficient surface gliding and underwater navigation without the need for a buoyancy adjustment system. This design provides flexible operational capabilities for both surface and underwater tasks, enabling the robot to quickly carry out underwater monitoring activities. Additionally, when it is necessary to reach another water body, SurfAAV can switch to aerial mode through a gliding takeoff, flying to the target water area to perform corresponding tasks. The main contribution of this letter lies in proposing a new solution for underwater, surface, and aerial movement, designing a novel hybrid prototype concept, developing the required control laws, and validating the robot's ability to successfully perform surface gliding and gliding takeoff. SurfAAV achieves a maximum surface gliding speed of 7.96 m/s and a maximum underwater speed of 3.1 m/s. The prototype's surface gliding maneuverability and underwater cruising maneuverability both exceed those of existing aquatic-aerial vehicles.
☆ Model Predictive Path-Following Control for a Quadrotor
Automating drone-assisted processes is a complex task. Many solutions rely on trajectory generation and tracking, whereas in contrast, path-following control is a particularly promising approach, offering an intuitive and natural approach to automate tasks for drones and other vehicles. While different solutions to the path-following problem have been proposed, most of them lack the capability to explicitly handle state and input constraints, are formulated in a conservative two-stage approach, or are only applicable to linear systems. To address these challenges, the paper is built upon a Model Predictive Control-based path-following framework and extends its application to the Crazyflie quadrotor, which is investigated in hardware experiments. A cascaded control structure including an underlying attitude controller is included in the Model Predictive Path-Following Control formulation to meet the challenging real-time demands of quadrotor control. The effectiveness of the proposed method is demonstrated through real-world experiments, representing, to the best of the authors' knowledge, a novel application of this MPC-based path-following approach to the quadrotor. Additionally, as an extension to the original method, to allow for deviations of the path in cases where the precise following of the path might be overly restrictive, a corridor path-following approach is presented.
comment: 15 pages, 11 figures, submitted to PAMM 2025
☆ MCOO-SLAM: A Multi-Camera Omnidirectional Object SLAM System
Object-level SLAM offers structured and semantically meaningful environment representations, making it more interpretable and suitable for high-level robotic tasks. However, most existing approaches rely on RGB-D sensors or monocular views, which suffer from narrow fields of view, occlusion sensitivity, and limited depth perception-especially in large-scale or outdoor environments. These limitations often restrict the system to observing only partial views of objects from limited perspectives, leading to inaccurate object modeling and unreliable data association. In this work, we propose MCOO-SLAM, a novel Multi-Camera Omnidirectional Object SLAM system that fully leverages surround-view camera configurations to achieve robust, consistent, and semantically enriched mapping in complex outdoor scenarios. Our approach integrates point features and object-level landmarks enhanced with open-vocabulary semantics. A semantic-geometric-temporal fusion strategy is introduced for robust object association across multiple views, leading to improved consistency and accurate object modeling, and an omnidirectional loop closure module is designed to enable viewpoint-invariant place recognition using scene-level descriptors. Furthermore, the constructed map is abstracted into a hierarchical 3D scene graph to support downstream reasoning tasks. Extensive experiments in real-world demonstrate that MCOO-SLAM achieves accurate localization and scalable object-level mapping with improved robustness to occlusion, pose variation, and environmental complexity.
☆ Efficient Navigation Among Movable Obstacles using a Mobile Manipulator via Hierarchical Policy Learning IROS 2025
We propose a hierarchical reinforcement learning (HRL) framework for efficient Navigation Among Movable Obstacles (NAMO) using a mobile manipulator. Our approach combines interaction-based obstacle property estimation with structured pushing strategies, facilitating the dynamic manipulation of unforeseen obstacles while adhering to a pre-planned global path. The high-level policy generates pushing commands that consider environmental constraints and path-tracking objectives, while the low-level policy precisely and stably executes these commands through coordinated whole-body movements. Comprehensive simulation-based experiments demonstrate improvements in performing NAMO tasks, including higher success rates, shortened traversed path length, and reduced goal-reaching times, compared to baselines. Additionally, ablation studies assess the efficacy of each component, while a qualitative analysis further validates the accuracy and reliability of the real-time obstacle property estimation.
comment: 8 pages, 6 figures, Accepted to IROS 2025. Supplementary Video: https://youtu.be/sZ8_z7sYVP0
☆ Comparison of Innovative Strategies for the Coverage Problem: Path Planning, Search Optimization, and Applications in Underwater Robotics
In many applications, including underwater robotics, the coverage problem requires an autonomous vehicle to systematically explore a defined area while minimizing redundancy and avoiding obstacles. This paper investigates coverage path planning strategies to enhance the efficiency of underwater gliders, particularly in maximizing the probability of detecting a radioactive source while ensuring safe navigation. We evaluate three path-planning approaches: the Traveling Salesman Problem (TSP), Minimum Spanning Tree (MST), and Optimal Control Problem (OCP). Simulations were conducted in MATLAB, comparing processing time, uncovered areas, path length, and traversal time. Results indicate that OCP is preferable when traversal time is constrained, although it incurs significantly higher computational costs. Conversely, MST-based approaches provide faster but less optimal solutions. These findings offer insights into selecting appropriate algorithms based on mission priorities, balancing efficiency and computational feasibility.
☆ Offensive Robot Cybersecurity
Offensive Robot Cybersecurity introduces a groundbreaking approach by advocating for offensive security methods empowered by means of automation. It emphasizes the necessity of understanding attackers' tactics and identifying vulnerabilities in advance to develop effective defenses, thereby improving robots' security posture. This thesis leverages a decade of robotics experience, employing Machine Learning and Game Theory to streamline the vulnerability identification and exploitation process. Intrinsically, the thesis uncovers a profound connection between robotic architecture and cybersecurity, highlighting that the design and creation aspect of robotics deeply intertwines with its protection against attacks. This duality -- whereby the architecture that shapes robot behavior and capabilities also necessitates a defense mechanism through offensive and defensive cybersecurity strategies -- creates a unique equilibrium. Approaching cybersecurity with a dual perspective of defense and attack, rooted in an understanding of systems architecture, has been pivotal. Through comprehensive analysis, including ethical considerations, the development of security tools, and executing cyber attacks on robot software, hardware, and industry deployments, this thesis proposes a novel architecture for cybersecurity cognitive engines. These engines, powered by advanced game theory and machine learning, pave the way for autonomous offensive cybersecurity strategies for robots, marking a significant shift towards self-defending robotic systems. This research not only underscores the importance of offensive measures in enhancing robot cybersecurity but also sets the stage for future advancements where robots are not just resilient to cyber threats but are equipped to autonomously safeguard themselves.
comment: Doctoral thesis
☆ Designing Intent: A Multimodal Framework for Human-Robot Cooperation in Industrial Workspaces
As robots enter collaborative workspaces, ensuring mutual understanding between human workers and robotic systems becomes a prerequisite for trust, safety, and efficiency. In this position paper, we draw on the cooperation scenario of the AIMotive project in which a human and a cobot jointly perform assembly tasks to argue for a structured approach to intent communication. Building on the Situation Awareness-based Agent Transparency (SAT) framework and the notion of task abstraction levels, we propose a multidimensional design space that maps intent content (SAT1, SAT3), planning horizon (operational to strategic), and modality (visual, auditory, haptic). We illustrate how this space can guide the design of multimodal communication strategies tailored to dynamic collaborative work contexts. With this paper, we lay the conceptual foundation for a future design toolkit aimed at supporting transparent human-robot interaction in the workplace. We highlight key open questions and design challenges, and propose a shared agenda for multimodal, adaptive, and trustworthy robotic collaboration in hybrid work environments.
comment: 9 pages
☆ Minimizing Structural Vibrations via Guided Flow Matching Design Optimization
Structural vibrations are a source of unwanted noise in engineering systems like cars, trains or airplanes. Minimizing these vibrations is crucial for improving passenger comfort. This work presents a novel design optimization approach based on guided flow matching for reducing vibrations by placing beadings (indentations) in plate-like structures. Our method integrates a generative flow matching model and a surrogate model trained to predict structural vibrations. During the generation process, the flow matching model pushes towards manufacturability while the surrogate model pushes to low-vibration solutions. The flow matching model and its training data implicitly define the design space, enabling a broader exploration of potential solutions as no optimization of manually-defined design parameters is required. We apply our method to a range of differentiable optimization objectives, including direct optimization of specific eigenfrequencies through careful construction of the objective function. Results demonstrate that our method generates diverse and manufacturable plate designs with reduced structural vibrations compared to designs from random search, a criterion-based design heuristic and genetic optimization. The code and data are available from https://github.com/ecker-lab/Optimizing_Vibrating_Plates.
☆ Context-Aware Deep Lagrangian Networks for Model Predictive Control IROS
Controlling a robot based on physics-informed dynamic models, such as deep Lagrangian networks (DeLaN), can improve the generalizability and interpretability of the resulting behavior. However, in complex environments, the number of objects to potentially interact with is vast, and their physical properties are often uncertain. This complexity makes it infeasible to employ a single global model. Therefore, we need to resort to online system identification of context-aware models that capture only the currently relevant aspects of the environment. While physical principles such as the conservation of energy may not hold across varying contexts, ensuring physical plausibility for any individual context-aware model can still be highly desirable, particularly when using it for receding horizon control methods such as Model Predictive Control (MPC). Hence, in this work, we extend DeLaN to make it context-aware, combine it with a recurrent network for online system identification, and integrate it with a MPC for adaptive, physics-informed control. We also combine DeLaN with a residual dynamics model to leverage the fact that a nominal model of the robot is typically available. We evaluate our method on a 7-DOF robot arm for trajectory tracking under varying loads. Our method reduces the end-effector tracking error by 39%, compared to a 21% improvement achieved by a baseline that uses an extended Kalman filter.
comment: Accepted to IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025
☆ Multi-Agent Reinforcement Learning for Autonomous Multi-Satellite Earth Observation: A Realistic Case Study
The exponential growth of Low Earth Orbit (LEO) satellites has revolutionised Earth Observation (EO) missions, addressing challenges in climate monitoring, disaster management, and more. However, autonomous coordination in multi-satellite systems remains a fundamental challenge. Traditional optimisation approaches struggle to handle the real-time decision-making demands of dynamic EO missions, necessitating the use of Reinforcement Learning (RL) and Multi-Agent Reinforcement Learning (MARL). In this paper, we investigate RL-based autonomous EO mission planning by modelling single-satellite operations and extending to multi-satellite constellations using MARL frameworks. We address key challenges, including energy and data storage limitations, uncertainties in satellite observations, and the complexities of decentralised coordination under partial observability. By leveraging a near-realistic satellite simulation environment, we evaluate the training stability and performance of state-of-the-art MARL algorithms, including PPO, IPPO, MAPPO, and HAPPO. Our results demonstrate that MARL can effectively balance imaging and resource management while addressing non-stationarity and reward interdependency in multi-satellite coordination. The insights gained from this study provide a foundation for autonomous satellite operations, offering practical guidelines for improving policy learning in decentralised EO missions.
☆ SHeRLoc: Synchronized Heterogeneous Radar Place Recognition for Cross-Modal Localization
Despite the growing adoption of radar in robotics, the majority of research has been confined to homogeneous sensor types, overlooking the integration and cross-modality challenges inherent in heterogeneous radar technologies. This leads to significant difficulties in generalizing across diverse radar data types, with modality-aware approaches that could leverage the complementary strengths of heterogeneous radar remaining unexplored. To bridge these gaps, we propose SHeRLoc, the first deep network tailored for heterogeneous radar, which utilizes RCS polar matching to align multimodal radar data. Our hierarchical optimal transport-based feature aggregation method generates rotationally robust multi-scale descriptors. By employing FFT-similarity-based data mining and adaptive margin-based triplet loss, SHeRLoc enables FOV-aware metric learning. SHeRLoc achieves an order of magnitude improvement in heterogeneous radar place recognition, increasing recall@1 from below 0.1 to 0.9 on a public dataset and outperforming state of-the-art methods. Also applicable to LiDAR, SHeRLoc paves the way for cross-modal place recognition and heterogeneous sensor SLAM. The source code will be available upon acceptance.
comment: This work has been submitted to the IEEE for possible publication
☆ Robust Instant Policy: Leveraging Student's t-Regression Model for Robust In-context Imitation Learning of Robot Manipulation IROS
Imitation learning (IL) aims to enable robots to perform tasks autonomously by observing a few human demonstrations. Recently, a variant of IL, called In-Context IL, utilized off-the-shelf large language models (LLMs) as instant policies that understand the context from a few given demonstrations to perform a new task, rather than explicitly updating network models with large-scale demonstrations. However, its reliability in the robotics domain is undermined by hallucination issues such as LLM-based instant policy, which occasionally generates poor trajectories that deviate from the given demonstrations. To alleviate this problem, we propose a new robust in-context imitation learning algorithm called the robust instant policy (RIP), which utilizes a Student's t-regression model to be robust against the hallucinated trajectories of instant policies to allow reliable trajectory generation. Specifically, RIP generates several candidate robot trajectories to complete a given task from an LLM and aggregates them using the Student's t-distribution, which is beneficial for ignoring outliers (i.e., hallucinations); thereby, a robust trajectory against hallucinations is generated. Our experiments, conducted in both simulated and real-world environments, show that RIP significantly outperforms state-of-the-art IL methods, with at least $26\%$ improvement in task success rates, particularly in low-data scenarios for everyday tasks. Video results available at https://sites.google.com/view/robustinstantpolicy.
comment: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025 accepted
☆ Human Locomotion Implicit Modeling Based Real-Time Gait Phase Estimation
Gait phase estimation based on inertial measurement unit (IMU) signals facilitates precise adaptation of exoskeletons to individual gait variations. However, challenges remain in achieving high accuracy and robustness, particularly during periods of terrain changes. To address this, we develop a gait phase estimation neural network based on implicit modeling of human locomotion, which combines temporal convolution for feature extraction with transformer layers for multi-channel information fusion. A channel-wise masked reconstruction pre-training strategy is proposed, which first treats gait phase state vectors and IMU signals as joint observations of human locomotion, thus enhancing model generalization. Experimental results demonstrate that the proposed method outperforms existing baseline approaches, achieving a gait phase RMSE of $2.729 \pm 1.071%$ and phase rate MAE of $0.037 \pm 0.016%$ under stable terrain conditions with a look-back window of 2 seconds, and a phase RMSE of $3.215 \pm 1.303%$ and rate MAE of $0.050 \pm 0.023%$ under terrain transitions. Hardware validation on a hip exoskeleton further confirms that the algorithm can reliably identify gait cycles and key events, adapting to various continuous motion scenarios. This research paves the way for more intelligent and adaptive exoskeleton systems, enabling safer and more efficient human-robot interaction across diverse real-world environments.
☆ Probabilistic Trajectory GOSPA: A Metric for Uncertainty-Aware Multi-Object Tracking Performance Evaluation
This paper presents a generalization of the trajectory general optimal sub-pattern assignment (GOSPA) metric for evaluating multi-object tracking algorithms that provide trajectory estimates with track-level uncertainties. This metric builds on the recently introduced probabilistic GOSPA metric to account for both the existence and state estimation uncertainties of individual object states. Similar to trajectory GOSPA (TGOSPA), it can be formulated as a multidimensional assignment problem, and its linear programming relaxation--also a valid metric--is computable in polynomial time. Additionally, this metric retains the interpretability of TGOSPA, and we show that its decomposition yields intuitive costs terms associated to expected localization error and existence probability mismatch error for properly detected objects, expected missed and false detection error, and track switch error. The effectiveness of the proposed metric is demonstrated through a simulation study.
comment: 7 pages, 4 figures
☆ TACT: Humanoid Whole-body Contact Manipulation through Deep Imitation Learning with Tactile Modality
Manipulation with whole-body contact by humanoid robots offers distinct advantages, including enhanced stability and reduced load. On the other hand, we need to address challenges such as the increased computational cost of motion generation and the difficulty of measuring broad-area contact. We therefore have developed a humanoid control system that allows a humanoid robot equipped with tactile sensors on its upper body to learn a policy for whole-body manipulation through imitation learning based on human teleoperation data. This policy, named tactile-modality extended ACT (TACT), has a feature to take multiple sensor modalities as input, including joint position, vision, and tactile measurements. Furthermore, by integrating this policy with retargeting and locomotion control based on a biped model, we demonstrate that the life-size humanoid robot RHP7 Kaleido is capable of achieving whole-body contact manipulation while maintaining balance and walking. Through detailed experimental verification, we show that inputting both vision and tactile modalities into the policy contributes to improving the robustness of manipulation involving broad and delicate contact.
☆ Booster Gym: An End-to-End Reinforcement Learning Framework for Humanoid Robot Locomotion
Recent advancements in reinforcement learning (RL) have led to significant progress in humanoid robot locomotion, simplifying the design and training of motion policies in simulation. However, the numerous implementation details make transferring these policies to real-world robots a challenging task. To address this, we have developed a comprehensive code framework that covers the entire process from training to deployment, incorporating common RL training methods, domain randomization, reward function design, and solutions for handling parallel structures. This library is made available as a community resource, with detailed descriptions of its design and experimental results. We validate the framework on the Booster T1 robot, demonstrating that the trained policies seamlessly transfer to the physical platform, enabling capabilities such as omnidirectional walking, disturbance resistance, and terrain adaptability. We hope this work provides a convenient tool for the robotics community, accelerating the development of humanoid robots. The code can be found in https://github.com/BoosterRobotics/booster_gym.
☆ VIMS: A Visual-Inertial-Magnetic-Sonar SLAM System in Underwater Environments IROS 2025
In this study, we present a novel simultaneous localization and mapping (SLAM) system, VIMS, designed for underwater navigation. Conventional visual-inertial state estimators encounter significant practical challenges in perceptually degraded underwater environments, particularly in scale estimation and loop closing. To address these issues, we first propose leveraging a low-cost single-beam sonar to improve scale estimation. Then, VIMS integrates a high-sampling-rate magnetometer for place recognition by utilizing magnetic signatures generated by an economical magnetic field coil. Building on this, a hierarchical scheme is developed for visual-magnetic place recognition, enabling robust loop closure. Furthermore, VIMS achieves a balance between local feature tracking and descriptor-based loop closing, avoiding additional computational burden on the front end. Experimental results highlight the efficacy of the proposed VIMS, demonstrating significant improvements in both the robustness and accuracy of state estimation within underwater environments.
comment: This work has been accepted for publication at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025)
☆ I Know You're Listening: Adaptive Voice for HRI IROS 23
While the use of social robots for language teaching has been explored, there remains limited work on a task-specific synthesized voices for language teaching robots. Given that language is a verbal task, this gap may have severe consequences for the effectiveness of robots for language teaching tasks. We address this lack of L2 teaching robot voices through three contributions: 1. We address the need for a lightweight and expressive robot voice. Using a fine-tuned version of Matcha-TTS, we use emoji prompting to create an expressive voice that shows a range of expressivity over time. The voice can run in real time with limited compute resources. Through case studies, we found this voice more expressive, socially appropriate, and suitable for long periods of expressive speech, such as storytelling. 2. We explore how to adapt a robot's voice to physical and social ambient environments to deploy our voices in various locations. We found that increasing pitch and pitch rate in noisy and high-energy environments makes the robot's voice appear more appropriate and makes it seem more aware of its current environment. 3. We create an English TTS system with improved clarity for L2 listeners using known linguistic properties of vowels that are difficult for these listeners. We used a data-driven, perception-based approach to understand how L2 speakers use duration cues to interpret challenging words with minimal tense (long) and lax (short) vowels in English. We found that the duration of vowels strongly influences the perception for L2 listeners and created an "L2 clarity mode" for Matcha-TTS that applies a lengthening to tense vowels while leaving lax vowels unchanged. Our clarity mode was found to be more respectful, intelligible, and encouraging than base Matcha-TTS while reducing transcription errors in these challenging tense/lax minimal pairs.
comment: PhD Thesis Simon Fraser University https://summit.sfu.ca/item/39353 Read the Room: Adapting a Robot's Voice to Ambient and Social Contexts IROS 23 Mmm whatcha say? Uncovering distal and proximal context effects in first and second-language word perception using psychophysical reverse correlation INTERSPEECH 24 Emojivoice: Towards long-term controllable expressivity in robot speech RO-MAN 25
☆ DyNaVLM: Zero-Shot Vision-Language Navigation System with Dynamic Viewpoints and Self-Refining Graph Memory
We present DyNaVLM, an end-to-end vision-language navigation framework using Vision-Language Models (VLM). In contrast to prior methods constrained by fixed angular or distance intervals, our system empowers agents to freely select navigation targets via visual-language reasoning. At its core lies a self-refining graph memory that 1) stores object locations as executable topological relations, 2) enables cross-robot memory sharing through distributed graph updates, and 3) enhances VLM's decision-making via retrieval augmentation. Operating without task-specific training or fine-tuning, DyNaVLM demonstrates high performance on GOAT and ObjectNav benchmarks. Real-world tests further validate its robustness and generalization. The system's three innovations: dynamic action space formulation, collaborative graph memory, and training-free deployment, establish a new paradigm for scalable embodied robot, bridging the gap between discrete VLN tasks and continuous real-world navigation.
☆ 3D Vision-tactile Reconstruction from Infrared and Visible Images for Robotic Fine-grained Tactile Perception
To achieve human-like haptic perception in anthropomorphic grippers, the compliant sensing surfaces of vision tactile sensor (VTS) must evolve from conventional planar configurations to biomimetically curved topographies with continuous surface gradients. However, planar VTSs have challenges when extended to curved surfaces, including insufficient lighting of surfaces, blurring in reconstruction, and complex spatial boundary conditions for surface structures. With an end goal of constructing a human-like fingertip, our research (i) develops GelSplitter3D by expanding imaging channels with a prism and a near-infrared (NIR) camera, (ii) proposes a photometric stereo neural network with a CAD-based normal ground truth generation method to calibrate tactile geometry, and (iii) devises a normal integration method with boundary constraints of depth prior information to correcting the cumulative error of surface integrals. We demonstrate better tactile sensing performance, a 40$\%$ improvement in normal estimation accuracy, and the benefits of sensor shapes in grasping and manipulation tasks.
☆ EmojiVoice: Towards long-term controllable expressivity in robot speech
Humans vary their expressivity when speaking for extended periods to maintain engagement with their listener. Although social robots tend to be deployed with ``expressive'' joyful voices, they lack this long-term variation found in human speech. Foundation model text-to-speech systems are beginning to mimic the expressivity in human speech, but they are difficult to deploy offline on robots. We present EmojiVoice, a free, customizable text-to-speech (TTS) toolkit that allows social roboticists to build temporally variable, expressive speech on social robots. We introduce emoji-prompting to allow fine-grained control of expressivity on a phase level and use the lightweight Matcha-TTS backbone to generate speech in real-time. We explore three case studies: (1) a scripted conversation with a robot assistant, (2) a storytelling robot, and (3) an autonomous speech-to-speech interactive agent. We found that using varied emoji prompting improved the perception and expressivity of speech over a long period in a storytelling task, but expressive voice was not preferred in the assistant use case.
comment: Accepted to RO-MAN 2025, Demo at HRI 2025 : https://dl.acm.org/doi/10.5555/3721488.3721774
☆ HEAL: An Empirical Study on Hallucinations in Embodied Agents Driven by Large Language Models
Large language models (LLMs) are increasingly being adopted as the cognitive core of embodied agents. However, inherited hallucinations, which stem from failures to ground user instructions in the observed physical environment, can lead to navigation errors, such as searching for a refrigerator that does not exist. In this paper, we present the first systematic study of hallucinations in LLM-based embodied agents performing long-horizon tasks under scene-task inconsistencies. Our goal is to understand to what extent hallucinations occur, what types of inconsistencies trigger them, and how current models respond. To achieve these goals, we construct a hallucination probing set by building on an existing benchmark, capable of inducing hallucination rates up to 40x higher than base prompts. Evaluating 12 models across two simulation environments, we find that while models exhibit reasoning, they fail to resolve scene-task inconsistencies-highlighting fundamental limitations in handling infeasible tasks. We also provide actionable insights on ideal model behavior for each scenario, offering guidance for developing more robust and reliable planning strategies.
☆ Assigning Multi-Robot Tasks to Multitasking Robots
One simplifying assumption in existing and well-performing task allocation methods is that the robots are single-tasking: each robot operates on a single task at any given time. While this assumption is harmless to make in some situations, it can be inefficient or even infeasible in others. In this paper, we consider assigning multi-robot tasks to multitasking robots. The key contribution is a novel task allocation framework that incorporates the consideration of physical constraints introduced by multitasking. This is in contrast to the existing work where such constraints are largely ignored. After formulating the problem, we propose a compilation to weighted MAX-SAT, which allows us to leverage existing solvers for a solution. A more efficient greedy heuristic is then introduced. For evaluation, we first compare our methods with a modern baseline that is efficient for single-tasking robots to validate the benefits of multitasking in synthetic domains. Then, using a site-clearing scenario in simulation, we further illustrate the complex task interaction considered by the multitasking robots in our approach to demonstrate its performance. Finally, we demonstrate a physical experiment to show how multitasking enabled by our approach can benefit task efficiency in a realistic setting.
♻ ☆ An Advanced Framework for Ultra-Realistic Simulation and Digital Twinning for Autonomous Vehicles
Simulation is a fundamental tool in developing autonomous vehicles, enabling rigorous testing without the logistical and safety challenges associated with real-world trials. As autonomous vehicle technologies evolve and public safety demands increase, advanced, realistic simulation frameworks are critical. Current testing paradigms employ a mix of general-purpose and specialized simulators, such as CARLA and IVRESS, to achieve high-fidelity results. However, these tools often struggle with compatibility due to differing platform, hardware, and software requirements, severely hampering their combined effectiveness. This paper introduces BlueICE, an advanced framework for ultra-realistic simulation and digital twinning, to address these challenges. BlueICE's innovative architecture allows for the decoupling of computing platforms, hardware, and software dependencies while offering researchers customizable testing environments to meet diverse fidelity needs. Key features include containerization to ensure compatibility across different systems, a unified communication bridge for seamless integration of various simulation tools, and synchronized orchestration of input and output across simulators. This framework facilitates the development of sophisticated digital twins for autonomous vehicle testing and sets a new standard in simulation accuracy and flexibility. The paper further explores the application of BlueICE in two distinct case studies: the ICAT indoor testbed and the STAR campus outdoor testbed at the University of Delaware. These case studies demonstrate BlueICE's capability to create sophisticated digital twins for autonomous vehicle testing and underline its potential as a standardized testbed for future autonomous driving technologies.
comment: 6 Pages. 5 Figures, 1 Table
♻ ☆ Semantic-Geometric-Physical-Driven Robot Manipulation Skill Transfer via Skill Library and Tactile Representation
Developing general robotic systems capable of manipulating in unstructured environments is a significant challenge, particularly as the tasks involved are typically long-horizon and rich-contact, requiring efficient skill transfer across different task scenarios. To address these challenges, we propose knowledge graph-based skill library construction method. This method hierarchically organizes manipulation knowledge using "task graph" and "scene graph" to represent task-specific and scene-specific information, respectively. Additionally, we introduce "state graph" to facilitate the interaction between high-level task planning and low-level scene information. Building upon this foundation, we further propose a novel hierarchical skill transfer framework based on the skill library and tactile representation, which integrates high-level reasoning for skill transfer and low-level precision for execution. At the task level, we utilize large language models (LLMs) and combine contextual learning with a four-stage chain-of-thought prompting paradigm to achieve subtask sequence transfer. At the motion level, we develop an adaptive trajectory transfer method based on the skill library and the heuristic path planning algorithm. At the physical level, we propose an adaptive contour extraction and posture perception method based on tactile representation. This method dynamically acquires high-precision contour and posture information from visual-tactile images, adjusting parameters such as contact position and posture to ensure the effectiveness of transferred skills in new environments. Experiments demonstrate the skill transfer and adaptability capabilities of the proposed methods across different task scenarios. Project website: https://github.com/MingchaoQi/skill_transfer
♻ ☆ PP-Tac: Paper Picking Using Tactile Feedback in Dexterous Robotic Hands
Robots are increasingly envisioned as human companions, assisting with everyday tasks that often involve manipulating deformable objects. Although recent advances in robotic hardware and embodied AI have expanded their capabilities, current systems still struggle with handling thin, flat, and deformable objects such as paper and fabric. This limitation arises from the lack of suitable perception techniques for robust state estimation under diverse object appearances, as well as the absence of planning techniques for generating appropriate grasp motions. To bridge these gaps, this paper introduces PP-Tac, a robotic system for picking up paper-like objects. PP-Tac features a multi-fingered robotic hand with high-resolution omnidirectional tactile sensors \sensorname. This hardware configuration enables real-time slip detection and online frictional force control that mitigates such slips. Furthermore, grasp motion generation is achieved through a trajectory synthesis pipeline, which first constructs a dataset of finger's pinching motions. Based on this dataset, a diffusion-based policy is trained to control the hand-arm robotic system. Experiments demonstrate that PP-Tac can effectively grasp paper-like objects of varying material, thickness, and stiffness, achieving an overall success rate of 87.5\%. To our knowledge, this work is the first attempt to grasp paper-like deformable objects using a tactile dexterous hand. Our project webpage can be found at: https://peilin-666.github.io/projects/PP-Tac/
comment: accepted by Robotics: Science and Systems(RSS) 2025 url: https://peilin-666.github.io/projects/PP-Tac/
♻ ☆ Learning the Geometric Mechanics of Robot Motion Using Gaussian Mixtures
Data-driven models of robot motion constructed using principles from Geometric Mechanics have been shown to produce useful predictions of robot motion for a variety of robots. For robots with a useful number of DoF, these geometric mechanics models can only be constructed in the neighborhood of a gait. Here we show how Gaussian Mixture Models (GMM) can be used as a form of manifold learning that learns the structure of the Geometric Mechanics "motility map" and demonstrate: [i] a sizable improvement in prediction quality when compared to the previously published methods; [ii] a method that can be applied to any motion dataset and not only periodic gait data; [iii] a way to pre-process the data-set to facilitate extrapolation in places where the motility map is known to be linear. Our results can be applied anywhere a data-driven geometric motion model might be useful.
comment: 16 pages, 10 figures
♻ ☆ An Actionable Hierarchical Scene Representation Enhancing Autonomous Inspection Missions in Unknown Environments IROS 2025
In this article, we present the Layered Semantic Graphs (LSG), a novel actionable hierarchical scene graph, fully integrated with a multi-modal mission planner, the FLIE: A First-Look based Inspection and Exploration planner. The novelty of this work stems from aiming to address the task of maintaining an intuitive and multi-resolution scene representation, while simultaneously offering a tractable foundation for planning and scene understanding during an ongoing inspection mission of apriori unknown targets-of-interest in an unknown environment. The proposed LSG scheme is composed of locally nested hierarchical graphs, at multiple layers of abstraction, with the abstract concepts grounded on the functionality of the integrated FLIE planner. Furthermore, LSG encapsulates real-time semantic segmentation models that offer extraction and localization of desired semantic elements within the hierarchical representation. This extends the capability of the inspection planner, which can then leverage LSG to make an informed decision to inspect a particular semantic of interest. We also emphasize the hierarchical and semantic path-planning capabilities of LSG, which could extend inspection missions by improving situational awareness for human operators in an unknown environment. The validity of the proposed scheme is proven through extensive evaluations of the proposed architecture in simulations, as well as experimental field deployments on a Boston Dynamics Spot quadruped robot in urban outdoor environment settings.
comment: Accepted to IROS 2025
♻ ☆ LLM-as-BT-Planner: Leveraging LLMs for Behavior Tree Generation in Robot Task Planning ICRA 2025
Robotic assembly tasks remain an open challenge due to their long horizon nature and complex part relations. Behavior trees (BTs) are increasingly used in robot task planning for their modularity and flexibility, but creating them manually can be effort-intensive. Large language models (LLMs) have recently been applied to robotic task planning for generating action sequences, yet their ability to generate BTs has not been fully investigated. To this end, we propose LLM-as-BT-Planner, a novel framework that leverages LLMs for BT generation in robotic assembly task planning. Four in-context learning methods are introduced to utilize the natural language processing and inference capabilities of LLMs for producing task plans in BT format, reducing manual effort while ensuring robustness and comprehensibility. Additionally, we evaluate the performance of fine-tuned smaller LLMs on the same tasks. Experiments in both simulated and real-world settings demonstrate that our framework enhances LLMs' ability to generate BTs, improving success rate through in-context learning and supervised fine-tuning.
comment: 7 pages. presented in ICRA 2025
♻ ☆ A compact neuromorphic system for ultra-energy-efficient, on-device robot localization
Neuromorphic computing offers a transformative pathway to overcome the computational and energy challenges faced in deploying robotic localization and navigation systems at the edge. Visual place recognition, a critical component for navigation, is often hampered by the high resource demands of conventional systems, making them unsuitable for small-scale robotic platforms which still require accurate long-endurance localization. Although neuromorphic approaches offer potential for greater efficiency, real-time edge deployment remains constrained by the complexity of bio-realistic networks. In order to overcome this challenge, fusion of hardware and algorithms is critical to employ this specialized computing paradigm. Here, we demonstrate a neuromorphic localization system that performs competitive place recognition in up to 8 kilometers of traversal using models as small as 180 kilobytes with 44,000 parameters, while consuming less than 8% of the energy required by conventional methods. Our Locational Encoding with Neuromorphic Systems (LENS) integrates spiking neural networks, an event-based dynamic vision sensor, and a neuromorphic processor within a single SynSense Speck chip, enabling real-time, energy-efficient localization on a hexapod robot. When compared to a benchmark place recognition method, Sum-of-Absolute-Differences (SAD), LENS performs comparably in overall precision. LENS represents an accurate fully neuromorphic localization system capable of large-scale, on-device deployment for energy efficient robotic place recognition. Neuromorphic computing enables resource-constrained robots to perform energy efficient, accurate localization.
comment: 42 pages, 5 main figures, 8 supplementary figures, 2 supplementary tables, and 1 movie
♻ ☆ Map Space Belief Prediction for Manipulation-Enhanced Mapping
Searching for objects in cluttered environments requires selecting efficient viewpoints and manipulation actions to remove occlusions and reduce uncertainty in object locations, shapes, and categories. In this work, we address the problem of manipulation-enhanced semantic mapping, where a robot has to efficiently identify all objects in a cluttered shelf. Although Partially Observable Markov Decision Processes~(POMDPs) are standard for decision-making under uncertainty, representing unstructured interactive worlds remains challenging in this formalism. To tackle this, we define a POMDP whose belief is summarized by a metric-semantic grid map and propose a novel framework that uses neural networks to perform map-space belief updates to reason efficiently and simultaneously about object geometries, locations, categories, occlusions, and manipulation physics. Further, to enable accurate information gain analysis, the learned belief updates should maintain calibrated estimates of uncertainty. Therefore, we propose Calibrated Neural-Accelerated Belief Updates (CNABUs) to learn a belief propagation model that generalizes to novel scenarios and provides confidence-calibrated predictions for unknown areas. Our experiments show that our novel POMDP planner improves map completeness and accuracy over existing methods in challenging simulations and successfully transfers to real-world cluttered shelves in zero-shot fashion.
comment: 14 pages, 10 figures; Published at RSS 2025 - this version contains a small fix to figure 6 which was missing a plot in the original submission
♻ ☆ Closed-Loop Long-Horizon Robotic Planning via Equilibrium Sequence Modeling ICML 2025
In the endeavor to make autonomous robots take actions, task planning is a major challenge that requires translating high-level task descriptions to long-horizon action sequences. Despite recent advances in language model agents, they remain prone to planning errors and limited in their ability to plan ahead. To address these limitations in robotic planning, we advocate a self-refining scheme that iteratively refines a draft plan until an equilibrium is reached. Remarkably, this process can be optimized end-to-end from an analytical perspective without the need to curate additional verifiers or reward models, allowing us to train self-refining planners in a simple supervised learning fashion. Meanwhile, a nested equilibrium sequence modeling procedure is devised for efficient closed-loop planning that incorporates useful feedback from the environment (or an internal world model). Our method is evaluated on the VirtualHome-Env benchmark, showing advanced performance with improved scaling w.r.t. inference-time computation. Code is available at https://github.com/Singularity0104/equilibrium-planner.
comment: ICML 2025
♻ ☆ SurgSora: Object-Aware Diffusion Model for Controllable Surgical Video Generation
Surgical video generation can enhance medical education and research, but existing methods lack fine-grained motion control and realism. We introduce SurgSora, a framework that generates high-fidelity, motion-controllable surgical videos from a single input frame and user-specified motion cues. Unlike prior approaches that treat objects indiscriminately or rely on ground-truth segmentation masks, SurgSora leverages self-predicted object features and depth information to refine RGB appearance and optical flow for precise video synthesis. It consists of three key modules: (1) the Dual Semantic Injector, which extracts object-specific RGB-D features and segmentation cues to enhance spatial representations; (2) the Decoupled Flow Mapper, which fuses multi-scale optical flow with semantic features for realistic motion dynamics; and (3) the Trajectory Controller, which estimates sparse optical flow and enables user-guided object movement. By conditioning these enriched features within the Stable Video Diffusion, SurgSora achieves state-of-the-art visual authenticity and controllability in advancing surgical video synthesis, as demonstrated by extensive quantitative and qualitative comparisons. Our human evaluation in collaboration with expert surgeons further demonstrates the high realism of SurgSora-generated videos, highlighting the potential of our method for surgical training and education. Our project is available at https://surgsora.github.io/surgsora.github.io.
♻ ☆ Tailless Flapping-Wing Robot With Bio-Inspired Elastic Passive Legs for Multi-Modal Locomotion
Flapping-wing robots offer significant versatility; however, achieving efficient multi-modal locomotion remains challenging. This paper presents the design, modeling, and experimentation of a novel tailless flapping-wing robot with three independently actuated pairs of wings. Inspired by the leg morphology of juvenile water striders, the robot incorporates bio-inspired elastic passive legs that convert flapping-induced vibrations into directional ground movement, enabling locomotion without additional actuators. This vibration-driven mechanism facilitates lightweight, mechanically simplified multi-modal mobility. An SE(3)-based controller coordinates flight and mode transitions with minimal actuation. To validate the robot's feasibility, a functional prototype was developed, and experiments were conducted to evaluate its flight, ground locomotion, and mode-switching capabilities. Results show satisfactory performance under constrained actuation, highlighting the potential of multi-modal flapping-wing designs for future aerial-ground robotic applications. These findings provide a foundation for future studies on frequency-based terrestrial control and passive yaw stabilization in hybrid locomotion systems.
comment: 8 pages, 11 figures, accepted by IEEE Robotics and Automation Letters (RAL)
♻ ☆ Human-Robot Co-Transportation using Disturbance-Aware MPC with Pose Optimization
This paper proposes a new control algorithm for human-robot co-transportation using a robot manipulator equipped with a mobile base and a robotic arm. We integrate the regular Model Predictive Control (MPC) with a novel pose optimization mechanism to more efficiently mitigate disturbances (such as human behavioral uncertainties or robot actuation noise) during the task. The core of our methodology involves a two-step iterative design: At each planning horizon, we determine the optimal pose of the robotic arm (joint angle configuration) from a candidate set, aiming to achieve the lowest estimated control cost. This selection is based on solving a disturbance-aware Discrete Algebraic Ricatti Equation (DARE), which also determines the optimal inputs for the robot's whole body control (including both the mobile base and the robotic arm). To validate the effectiveness of the proposed approach, we provide theoretical derivation for the disturbance-aware DARE and perform simulated experiments and hardware demos using a Fetch robot under varying conditions, including different trajectories and different levels of disturbances. The results reveal that our proposed approach outperforms baseline algorithms.
comment: 8 pages, 6 figures
Artificial Intelligence 149
☆ Dense SAE Latents Are Features, Not Bugs
Sparse autoencoders (SAEs) are designed to extract interpretable features from language models by enforcing a sparsity constraint. Ideally, training an SAE would yield latents that are both sparse and semantically meaningful. However, many SAE latents activate frequently (i.e., are \emph{dense}), raising concerns that they may be undesirable artifacts of the training procedure. In this work, we systematically investigate the geometry, function, and origin of dense latents and show that they are not only persistent but often reflect meaningful model representations. We first demonstrate that dense latents tend to form antipodal pairs that reconstruct specific directions in the residual stream, and that ablating their subspace suppresses the emergence of new dense features in retrained SAEs -- suggesting that high density features are an intrinsic property of the residual space. We then introduce a taxonomy of dense latents, identifying classes tied to position tracking, context binding, entropy regulation, letter-specific output signals, part-of-speech, and principal component reconstruction. Finally, we analyze how these features evolve across layers, revealing a shift from structural features in early layers, to semantic features in mid layers, and finally to output-oriented signals in the last layers of the model. Our findings indicate that dense latents serve functional roles in language model computation and should not be dismissed as training noise.
☆ Embodied Web Agents: Bridging Physical-Digital Realms for Integrated Agent Intelligence
AI agents today are mostly siloed - they either retrieve and reason over vast amount of digital information and knowledge obtained online; or interact with the physical world through embodied perception, planning and action - but rarely both. This separation limits their ability to solve tasks that require integrated physical and digital intelligence, such as cooking from online recipes, navigating with dynamic map data, or interpreting real-world landmarks using web knowledge. We introduce Embodied Web Agents, a novel paradigm for AI agents that fluidly bridge embodiment and web-scale reasoning. To operationalize this concept, we first develop the Embodied Web Agents task environments, a unified simulation platform that tightly integrates realistic 3D indoor and outdoor environments with functional web interfaces. Building upon this platform, we construct and release the Embodied Web Agents Benchmark, which encompasses a diverse suite of tasks including cooking, navigation, shopping, tourism, and geolocation - all requiring coordinated reasoning across physical and digital realms for systematic assessment of cross-domain intelligence. Experimental results reveal significant performance gaps between state-of-the-art AI systems and human capabilities, establishing both challenges and opportunities at the intersection of embodied cognition and web-scale knowledge access. All datasets, codes and websites are publicly available at our project page https://embodied-web-agent.github.io/.
☆ Sekai: A Video Dataset towards World Exploration
Video generation techniques have made remarkable progress, promising to be the foundation of interactive world exploration. However, existing video generation datasets are not well-suited for world exploration training as they suffer from some limitations: limited locations, short duration, static scenes, and a lack of annotations about exploration and the world. In this paper, we introduce Sekai (meaning ``world'' in Japanese), a high-quality first-person view worldwide video dataset with rich annotations for world exploration. It consists of over 5,000 hours of walking or drone view (FPV and UVA) videos from over 100 countries and regions across 750 cities. We develop an efficient and effective toolbox to collect, pre-process and annotate videos with location, scene, weather, crowd density, captions, and camera trajectories. Experiments demonstrate the quality of the dataset. And, we use a subset to train an interactive video world exploration model, named YUME (meaning ``dream'' in Japanese). We believe Sekai will benefit the area of video generation and world exploration, and motivate valuable applications.
comment: 12 pages, 6 figures
☆ Leaky Thoughts: Large Reasoning Models Are Not Private Thinkers
We study privacy leakage in the reasoning traces of large reasoning models used as personal agents. Unlike final outputs, reasoning traces are often assumed to be internal and safe. We challenge this assumption by showing that reasoning traces frequently contain sensitive user data, which can be extracted via prompt injections or accidentally leak into outputs. Through probing and agentic evaluations, we demonstrate that test-time compute approaches, particularly increased reasoning steps, amplify such leakage. While increasing the budget of those test-time compute approaches makes models more cautious in their final answers, it also leads them to reason more verbosely and leak more in their own thinking. This reveals a core tension: reasoning improves utility but enlarges the privacy attack surface. We argue that safety efforts must extend to the model's internal thinking, not just its outputs.
☆ SwarmAgentic: Towards Fully Automated Agentic System Generation via Swarm Intelligence
The rapid progress of Large Language Models has advanced agentic systems in decision-making, coordination, and task execution. Yet, existing agentic system generation frameworks lack full autonomy, missing from-scratch agent generation, self-optimizing agent functionality, and collaboration, limiting adaptability and scalability. We propose SwarmAgentic, a framework for fully automated agentic system generation that constructs agentic systems from scratch and jointly optimizes agent functionality and collaboration as interdependent components through language-driven exploration. To enable efficient search over system-level structures, SwarmAgentic maintains a population of candidate systems and evolves them via feedback-guided updates, drawing inspiration from Particle Swarm Optimization (PSO). We evaluate our method on six real-world, open-ended, and exploratory tasks involving high-level planning, system-level coordination, and creative reasoning. Given only a task description and an objective function, SwarmAgentic outperforms all baselines, achieving a +261.8% relative improvement over ADAS on the TravelPlanner benchmark, highlighting the effectiveness of full automation in structurally unconstrained tasks. This framework marks a significant step toward scalable and autonomous agentic system design, bridging swarm intelligence with fully automated system multi-agent generation. Our code is publicly released at https://yaoz720.github.io/SwarmAgentic/.
comment: 41 pages
☆ AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning
Rule-based rewards offer a promising strategy for improving reinforcement learning from human feedback (RLHF), but current approaches often rely on manual rule engineering. We present AutoRule, a fully automated method for extracting rules from preference feedback and formulating them into rule-based rewards. AutoRule extraction operates in three stages: it leverages a reasoning model to interpret user preferences, identifies candidate rules from the reasoning chain of these interpretations, and synthesizes them into a unified rule set. Leveraging the finalized rule set, we employ language-model verifiers to compute the fraction of rules satisfied by each output, using this metric as an auxiliary reward alongside the learned reward model during policy optimization. Training a Llama-3-8B model with AutoRule results in a 28.6\% relative improvement in length-controlled win rate on AlpacaEval2.0, and a 6.1\% relative gain in second-turn performance on a held-out MT-Bench subset, compared to a GRPO baseline trained with the same learned reward model but without the rule-based auxiliary reward. Our analysis confirms that the extracted rules exhibit good agreement with dataset preference. We find that AutoRule demonstrates reduced reward hacking compared to a learned reward model when run over two episodes. Finally, our case study suggests that the extracted rules capture unique qualities valued in different datasets. The extracted rules are provided in the appendix, and the code is open-sourced at https://github.com/cxcscmu/AutoRule.
☆ Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement
Recent advancements in large reasoning models (LRMs) have significantly enhanced language models' capabilities in complex problem-solving by emulating human-like deliberative thinking. However, these models often exhibit overthinking (i.e., the generation of unnecessarily verbose and redundant content), which hinders efficiency and inflates inference cost. In this work, we explore the representational and behavioral origins of this inefficiency, revealing that LRMs inherently possess the capacity for more concise reasoning. Empirical analyses show that correct reasoning paths vary significantly in length, and the shortest correct responses often suffice, indicating untapped efficiency potential. Exploiting these findings, we propose two lightweight methods to enhance LRM efficiency. First, we introduce Efficiency Steering, a training-free activation steering technique that modulates reasoning behavior via a single direction in the model's representation space. Second, we develop Self-Rewarded Efficiency RL, a reinforcement learning framework that dynamically balances task accuracy and brevity by rewarding concise correct solutions. Extensive experiments on seven LRM backbones across multiple mathematical reasoning benchmarks demonstrate that our methods significantly reduce reasoning length while preserving or improving task performance. Our results highlight that reasoning efficiency can be improved by leveraging and guiding the intrinsic capabilities of existing models in a self-guided manner.
☆ Demystifying the Visual Quality Paradox in Multimodal Large Language Models
Recent Multimodal Large Language Models (MLLMs) excel on benchmark vision-language tasks, yet little is known about how input visual quality shapes their responses. Does higher perceptual quality of images already translate to better MLLM understanding? We conduct the first systematic study spanning leading MLLMs and a suite of vision-language benchmarks, applying controlled degradations and stylistic shifts to each image. Surprisingly, we uncover a visual-quality paradox: model, task, and even individual-instance performance can improve when images deviate from human-perceived fidelity. Off-the-shelf restoration pipelines fail to reconcile these idiosyncratic preferences. To close the gap, we introduce Visual-Quality Test-Time Tuning (VQ-TTT)-a lightweight adaptation module that: (1) inserts a learnable, low-rank kernel before the frozen vision encoder to modulate frequency content; and (2) fine-tunes only shallow vision-encoder layers via LoRA. VQ-TTT dynamically adjusts each input image in a single forward pass, aligning it with task-specific model preferences. Across the evaluated MLLMs and all datasets, VQ-TTT lifts significant average accuracy, with no external models, cached features, or extra training data. These findings redefine ``better'' visual inputs for MLLMs and highlight the need for adaptive, rather than universally ``clean'', imagery, in the new era of AI being the main data customer.
comment: 18 pages
☆ The AI Policy Module: Developing Computer Science Student Competency in AI Ethics and Policy
As artificial intelligence (AI) further embeds itself into many settings across personal and professional contexts, increasing attention must be paid not only to AI ethics, but also to the governance and regulation of AI technologies through AI policy. However, the prevailing post-secondary computing curriculum is currently ill-equipped to prepare future AI practitioners to confront increasing demands to implement abstract ethical principles and normative policy preferences into the design and development of AI systems. We believe that familiarity with the 'AI policy landscape' and the ability to translate ethical principles to practices will in the future constitute an important responsibility for even the most technically-focused AI engineers. Toward preparing current computer science (CS) students for these new expectations, we developed an AI Policy Module to introduce discussions of AI policy into the CS curriculum. Building on a successful pilot in fall 2024, in this innovative practice full paper we present an updated and expanded version of the module, including a technical assignment on "AI regulation". We present the findings from our pilot of the AI Policy Module 2.0, evaluating student attitudes towards AI ethics and policy through pre- and post-module surveys. Following the module, students reported increased concern about the ethical impacts of AI technologies while also expressing greater confidence in their abilities to engage in discussions about AI regulation. Finally, we highlight the AI Regulation Assignment as an effective and engaging tool for exploring the limits of AI alignment and emphasizing the role of 'policy' in addressing ethical challenges.
comment: Accepted at IEEE Frontiers in Education (FIE) 2025
☆ Revisiting Compositional Generalization Capability of Large Language Models Considering Instruction Following Ability ACL 2025
In generative commonsense reasoning tasks such as CommonGen, generative large language models (LLMs) compose sentences that include all given concepts. However, when focusing on instruction-following capabilities, if a prompt specifies a concept order, LLMs must generate sentences that adhere to the specified order. To address this, we propose Ordered CommonGen, a benchmark designed to evaluate the compositional generalization and instruction-following abilities of LLMs. This benchmark measures ordered coverage to assess whether concepts are generated in the specified order, enabling a simultaneous evaluation of both abilities. We conducted a comprehensive analysis using 36 LLMs and found that, while LLMs generally understand the intent of instructions, biases toward specific concept order patterns often lead to low-diversity outputs or identical results even when the concept order is altered. Moreover, even the most instruction-compliant LLM achieved only about 75% ordered coverage, highlighting the need for improvements in both instruction-following and compositional generalization capabilities.
comment: ACL 2025 Main
☆ Federated Learning for MRI-based BrainAGE: a multicenter study on post-stroke functional outcome prediction
$\textbf{Objective:}$ Brain-predicted age difference (BrainAGE) is a neuroimaging biomarker reflecting brain health. However, training robust BrainAGE models requires large datasets, often restricted by privacy concerns. This study evaluates the performance of federated learning (FL) for BrainAGE estimation in ischemic stroke patients treated with mechanical thrombectomy, and investigates its association with clinical phenotypes and functional outcomes. $\textbf{Methods:}$ We used FLAIR brain images from 1674 stroke patients across 16 hospital centers. We implemented standard machine learning and deep learning models for BrainAGE estimates under three data management strategies: centralized learning (pooled data), FL (local training at each site), and single-site learning. We reported prediction errors and examined associations between BrainAGE and vascular risk factors (e.g., diabetes mellitus, hypertension, smoking), as well as functional outcomes at three months post-stroke. Logistic regression evaluated BrainAGE's predictive value for these outcomes, adjusting for age, sex, vascular risk factors, stroke severity, time between MRI and arterial puncture, prior intravenous thrombolysis, and recanalisation outcome. $\textbf{Results:}$ While centralized learning yielded the most accurate predictions, FL consistently outperformed single-site models. BrainAGE was significantly higher in patients with diabetes mellitus across all models. Comparisons between patients with good and poor functional outcomes, and multivariate predictions of these outcomes showed the significance of the association between BrainAGE and post-stroke recovery. $\textbf{Conclusion:}$ FL enables accurate age predictions without data centralization. The strong association between BrainAGE, vascular risk factors, and post-stroke recovery highlights its potential for prognostic modeling in stroke care.
☆ The Effect of State Representation on LLM Agent Behavior in Dynamic Routing Games
Large Language Models (LLMs) have shown promise as decision-makers in dynamic settings, but their stateless nature necessitates creating a natural language representation of history. We present a unifying framework for systematically constructing natural language "state" representations for prompting LLM agents in repeated multi-agent games. Previous work on games with LLM agents has taken an ad hoc approach to encoding game history, which not only obscures the impact of state representation on agents' behavior, but also limits comparability between studies. Our framework addresses these gaps by characterizing methods of state representation along three axes: action informativeness (i.e., the extent to which the state representation captures actions played); reward informativeness (i.e., the extent to which the state representation describes rewards obtained); and prompting style (or natural language compression, i.e., the extent to which the full text history is summarized). We apply this framework to a dynamic selfish routing game, chosen because it admits a simple equilibrium both in theory and in human subject experiments \cite{rapoport_choice_2009}. Despite the game's relative simplicity, we find that there are key dependencies of LLM agent behavior on the natural language state representation. In particular, we observe that representations which provide agents with (1) summarized, rather than complete, natural language representations of past history; (2) information about regrets, rather than raw payoffs; and (3) limited information about others' actions lead to behavior that more closely matches game theoretic equilibrium predictions, and with more stable game play by the agents. By contrast, other representations can exhibit either large deviations from equilibrium, higher variation in dynamic game play over time, or both.
comment: 27 pages, 20 figures
☆ GFLC: Graph-based Fairness-aware Label Correction for Fair Classification
Fairness in machine learning (ML) has a critical importance for building trustworthy machine learning system as artificial intelligence (AI) systems increasingly impact various aspects of society, including healthcare decisions and legal judgments. Moreover, numerous studies demonstrate evidence of unfair outcomes in ML and the need for more robust fairness-aware methods. However, the data we use to train and develop debiasing techniques often contains biased and noisy labels. As a result, the label bias in the training data affects model performance and misrepresents the fairness of classifiers during testing. To tackle this problem, our paper presents Graph-based Fairness-aware Label Correction (GFLC), an efficient method for correcting label noise while preserving demographic parity in datasets. In particular, our approach combines three key components: prediction confidence measure, graph-based regularization through Ricci-flow-optimized graph Laplacians, and explicit demographic parity incentives. Our experimental findings show the effectiveness of our proposed approach and show significant improvements in the trade-off between performance and fairness metrics compared to the baseline.
comment: 25 pages, 6 figures
☆ The Compositional Architecture of Regret in Large Language Models
Regret in Large Language Models refers to their explicit regret expression when presented with evidence contradicting their previously generated misinformation. Studying the regret mechanism is crucial for enhancing model reliability and helps in revealing how cognition is coded in neural networks. To understand this mechanism, we need to first identify regret expressions in model outputs, then analyze their internal representation. This analysis requires examining the model's hidden states, where information processing occurs at the neuron level. However, this faces three key challenges: (1) the absence of specialized datasets capturing regret expressions, (2) the lack of metrics to find the optimal regret representation layer, and (3) the lack of metrics for identifying and analyzing regret neurons. Addressing these limitations, we propose: (1) a workflow for constructing a comprehensive regret dataset through strategically designed prompting scenarios, (2) the Supervised Compression-Decoupling Index (S-CDI) metric to identify optimal regret representation layers, and (3) the Regret Dominance Score (RDS) metric to identify regret neurons and the Group Impact Coefficient (GIC) to analyze activation patterns. Our experimental results successfully identified the optimal regret representation layer using the S-CDI metric, which significantly enhanced performance in probe classification experiments. Additionally, we discovered an M-shaped decoupling pattern across model layers, revealing how information processing alternates between coupling and decoupling phases. Through the RDS metric, we categorized neurons into three distinct functional groups: regret neurons, non-regret neurons, and dual neurons.
comment: 23 pages
☆ LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning
Large Language Models (LLMs) have become indispensable in real-world applications. However, their widespread adoption raises significant safety concerns, particularly in responding to socially harmful questions. Despite substantial efforts to improve model safety through alignment, aligned models can still have their safety protections undermined by subsequent fine-tuning - even when the additional training data appears benign. In this paper, we empirically demonstrate that this vulnerability stems from the sensitivity of safety-critical low-rank subspaces in LLM parameters to fine-tuning. Building on this insight, we propose a novel training-free method, termed Low-Rank Extrapolation (LoX), to enhance safety robustness by extrapolating the safety subspace of an aligned LLM. Our experimental results confirm the effectiveness of LoX, demonstrating significant improvements in robustness against both benign and malicious fine-tuning attacks while preserving the model's adaptability to new tasks. For instance, LoX leads to 11% to 54% absolute reductions in attack success rates (ASR) facing benign or malicious fine-tuning attacks. By investigating the ASR landscape of parameters, we attribute the success of LoX to that the extrapolation moves LLM parameters to a flatter zone, thereby less sensitive to perturbations. The code is available at github.com/VITA-Group/LoX.
☆ From Model to Classroom: Evaluating Generated MCQs for Portuguese with Narrative and Difficulty Concerns
While MCQs are valuable for learning and evaluation, manually creating them with varying difficulty levels and targeted reading skills remains a time-consuming and costly task. Recent advances in generative AI provide an opportunity to automate MCQ generation efficiently. However, assessing the actual quality and reliability of generated MCQs has received limited attention -- particularly regarding cases where generation fails. This aspect becomes particularly important when the generated MCQs are meant to be applied in real-world settings. Additionally, most MCQ generation studies focus on English, leaving other languages underexplored. This paper investigates the capabilities of current generative models in producing MCQs for reading comprehension in Portuguese, a morphologically rich language. Our study focuses on generating MCQs that align with curriculum-relevant narrative elements and span different difficulty levels. We evaluate these MCQs through expert review and by analyzing the psychometric properties extracted from student responses to assess their suitability for elementary school students. Our results show that current models can generate MCQs of comparable quality to human-authored ones. However, we identify issues related to semantic clarity and answerability. Also, challenges remain in generating distractors that engage students and meet established criteria for high-quality MCQ option design.
comment: This is a preprint version of the manuscript currently under review at an international journal
☆ WikiMixQA: A Multimodal Benchmark for Question Answering over Tables and Charts ACL 2025
Documents are fundamental to preserving and disseminating information, often incorporating complex layouts, tables, and charts that pose significant challenges for automatic document understanding (DU). While vision-language large models (VLLMs) have demonstrated improvements across various tasks, their effectiveness in processing long-context vision inputs remains unclear. This paper introduces WikiMixQA, a benchmark comprising 1,000 multiple-choice questions (MCQs) designed to evaluate cross-modal reasoning over tables and charts extracted from 4,000 Wikipedia pages spanning seven distinct topics. Unlike existing benchmarks, WikiMixQA emphasizes complex reasoning by requiring models to synthesize information from multiple modalities. We evaluate 12 state-of-the-art vision-language models, revealing that while proprietary models achieve ~70% accuracy when provided with direct context, their performance deteriorates significantly when retrieval from long documents is required. Among these, GPT-4-o is the only model exceeding 50% accuracy in this setting, whereas open-source models perform considerably worse, with a maximum accuracy of 27%. These findings underscore the challenges of long-context, multi-modal reasoning and establish WikiMixQA as a crucial benchmark for advancing document understanding research.
comment: ACL 2025 (Findings)
☆ One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution
It is a challenging problem to reproduce rich spatial details while maintaining temporal consistency in real-world video super-resolution (Real-VSR), especially when we leverage pre-trained generative models such as stable diffusion (SD) for realistic details synthesis. Existing SD-based Real-VSR methods often compromise spatial details for temporal coherence, resulting in suboptimal visual quality. We argue that the key lies in how to effectively extract the degradation-robust temporal consistency priors from the low-quality (LQ) input video and enhance the video details while maintaining the extracted consistency priors. To achieve this, we propose a Dual LoRA Learning (DLoRAL) paradigm to train an effective SD-based one-step diffusion model, achieving realistic frame details and temporal consistency simultaneously. Specifically, we introduce a Cross-Frame Retrieval (CFR) module to aggregate complementary information across frames, and train a Consistency-LoRA (C-LoRA) to learn robust temporal representations from degraded inputs. After consistency learning, we fix the CFR and C-LoRA modules and train a Detail-LoRA (D-LoRA) to enhance spatial details while aligning with the temporal space defined by C-LoRA to keep temporal coherence. The two phases alternate iteratively for optimization, collaboratively delivering consistent and detail-rich outputs. During inference, the two LoRA branches are merged into the SD model, allowing efficient and high-quality video restoration in a single diffusion step. Experiments show that DLoRAL achieves strong performance in both accuracy and speed. Code and models are available at https://github.com/yjsunnn/DLoRAL.
☆ Managing Complex Failure Analysis Workflows with LLM-based Reasoning and Acting Agents
Failure Analysis (FA) is a highly intricate and knowledge-intensive process. The integration of AI components within the computational infrastructure of FA labs has the potential to automate a variety of tasks, including the detection of non-conformities in images, the retrieval of analogous cases from diverse data sources, and the generation of reports from annotated images. However, as the number of deployed AI models increases, the challenge lies in orchestrating these components into cohesive and efficient workflows that seamlessly integrate with the FA process. This paper investigates the design and implementation of a Large Language Model (LLM)-based Planning Agent (LPA) to assist FA engineers in solving their analysis cases. The LPA integrates LLMs with advanced planning capabilities and external tool utilization, enabling autonomous processing of complex queries, retrieval of relevant data from external systems, and generation of human-readable responses. Evaluation results demonstrate the agent's operational effectiveness and reliability in supporting FA tasks.
☆ Towards Explainable Indoor Localization: Interpreting Neural Network Learning on Wi-Fi Fingerprints Using Logic Gates
Indoor localization using deep learning (DL) has demonstrated strong accuracy in mapping Wi-Fi RSS fingerprints to physical locations; however, most existing DL frameworks function as black-box models, offering limited insight into how predictions are made or how models respond to real-world noise over time. This lack of interpretability hampers our ability to understand the impact of temporal variations - caused by environmental dynamics - and to adapt models for long-term reliability. To address this, we introduce LogNet, a novel logic gate-based framework designed to interpret and enhance DL-based indoor localization. LogNet enables transparent reasoning by identifying which access points (APs) are most influential for each reference point (RP) and reveals how environmental noise disrupts DL-driven localization decisions. This interpretability allows us to trace and diagnose model failures and adapt DL systems for more stable long-term deployments. Evaluations across multiple real-world building floorplans and over two years of temporal variation show that LogNet not only interprets the internal behavior of DL models but also improves performance-achieving up to 1.1x to 2.8x lower localization error, 3.4x to 43.3x smaller model size, and 1.5x to 3.6x lower latency compared to prior DL-based models.
☆ DAILOC: Domain-Incremental Learning for Indoor Localization using Smartphones
Wi-Fi fingerprinting-based indoor localization faces significant challenges in real-world deployments due to domain shifts arising from device heterogeneity and temporal variations within indoor environments. Existing approaches often address these issues independently, resulting in poor generalization and susceptibility to catastrophic forgetting over time. In this work, we propose DAILOC, a novel domain-incremental learning framework that jointly addresses both temporal and device-induced domain shifts. DAILOC introduces a novel disentanglement strategy that separates domain shifts from location-relevant features using a multi-level variational autoencoder. Additionally, we introduce a novel memory-guided class latent alignment mechanism to address the effects of catastrophic forgetting over time. Experiments across multiple smartphones, buildings, and time instances demonstrate that DAILOC significantly outperforms state-of-the-art methods, achieving up to 2.74x lower average error and 4.6x lower worst-case error.
☆ CLAIM: Clinically-Guided LGE Augmentation for Realistic and Diverse Myocardial Scar Synthesis and Segmentation
Deep learning-based myocardial scar segmentation from late gadolinium enhancement (LGE) cardiac MRI has shown great potential for accurate and timely diagnosis and treatment planning for structural cardiac diseases. However, the limited availability and variability of LGE images with high-quality scar labels restrict the development of robust segmentation models. To address this, we introduce CLAIM: \textbf{C}linically-Guided \textbf{L}GE \textbf{A}ugmentation for Real\textbf{i}stic and Diverse \textbf{M}yocardial Scar Synthesis and Segmentation framework, a framework for anatomically grounded scar generation and segmentation. At its core is the SMILE module (Scar Mask generation guided by cLinical knowledgE), which conditions a diffusion-based generator on the clinically adopted AHA 17-segment model to synthesize images with anatomically consistent and spatially diverse scar patterns. In addition, CLAIM employs a joint training strategy in which the scar segmentation network is optimized alongside the generator, aiming to enhance both the realism of synthesized scars and the accuracy of the scar segmentation performance. Experimental results show that CLAIM produces anatomically coherent scar patterns and achieves higher Dice similarity with real scar distributions compared to baseline models. Our approach enables controllable and realistic myocardial scar synthesis and has demonstrated utility for downstream medical imaging task.
comment: 14 Pages
☆ Learning Algorithms in the Limit COLT 2025
This paper studies the problem of learning computable functions in the limit by extending Gold's inductive inference framework to incorporate \textit{computational observations} and \textit{restricted input sources}. Complimentary to the traditional Input-Output Observations, we introduce Time-Bound Observations, and Policy-Trajectory Observations to study the learnability of general recursive functions under more realistic constraints. While input-output observations do not suffice for learning the class of general recursive functions in the limit, we overcome this learning barrier by imposing computational complexity constraints or supplementing with approximate time-bound observations. Further, we build a formal framework around observations of \textit{computational agents} and show that learning computable functions from policy trajectories reduces to learning rational functions from input and output, thereby revealing interesting connections to finite-state transducer inference. On the negative side, we show that computable or polynomial-mass characteristic sets cannot exist for the class of linear-time computable functions even for policy-trajectory observations.
comment: Accepted at COLT 2025. This version matches the proceedings version
☆ Intrinsic and Extrinsic Organized Attention: Softmax Invariance and Network Sparsity
We examine the intrinsic (within the attention head) and extrinsic (amongst the attention heads) structure of the self-attention mechanism in transformers. Theoretical evidence for invariance of the self-attention mechanism to softmax activation is obtained by appealing to paradifferential calculus, (and is supported by computational examples), which relies on the intrinsic organization of the attention heads. Furthermore, we use an existing methodology for hierarchical organization of tensors to examine network structure by constructing hierarchal partition trees with respect to the query, key, and head axes of network 3-tensors. Such an organization is consequential since it allows one to profitably execute common signal processing tasks on a geometry where the organized network 3-tensors exhibit regularity. We exemplify this qualitatively, by visualizing the hierarchical organization of the tree comprised of attention heads and the diffusion map embeddings, and quantitatively by investigating network sparsity with the expansion coefficients of individual attention heads and the entire network with respect to the bi and tri-haar bases (respectively) on the space of queries, keys, and heads of the network. To showcase the utility of our theoretical and methodological findings, we provide computational examples using vision and language transformers. The ramifications of these findings are two-fold: (1) a subsequent step in interpretability analysis is theoretically admitted, and can be exploited empirically for downstream interpretability tasks (2) one can use the network 3-tensor organization for empirical network applications such as model pruning (by virtue of network sparsity) and network architecture comparison.
comment: 16 pages, 6 figures, 2 tables
☆ Capturing Polysemanticity with PRISM: A Multi-Concept Feature Description Framework
Automated interpretability research aims to identify concepts encoded in neural network features to enhance human understanding of model behavior. Current feature description methods face two critical challenges: limited robustness and the flawed assumption that each neuron encodes only a single concept (monosemanticity), despite growing evidence that neurons are often polysemantic. This assumption restricts the expressiveness of feature descriptions and limits their ability to capture the full range of behaviors encoded in model internals. To address this, we introduce Polysemantic FeatuRe Identification and Scoring Method (PRISM), a novel framework that captures the inherent complexity of neural network features. Unlike prior approaches that assign a single description per feature, PRISM provides more nuanced descriptions for both polysemantic and monosemantic features. We apply PRISM to language models and, through extensive benchmarking against existing methods, demonstrate that our approach produces more accurate and faithful feature descriptions, improving both overall description quality (via a description score) and the ability to capture distinct concepts when polysemanticity is present (via a polysemanticity score).
☆ RePCS: Diagnosing Data Memorization in LLM-Powered Retrieval-Augmented Generation
Retrieval-augmented generation (RAG) has become a common strategy for updating large language model (LLM) responses with current, external information. However, models may still rely on memorized training data, bypass the retrieved evidence, and produce contaminated outputs. We introduce Retrieval-Path Contamination Scoring (RePCS), a diagnostic method that detects such behavior without requiring model access or retraining. RePCS compares two inference paths: (i) a parametric path using only the query, and (ii) a retrieval-augmented path using both the query and retrieved context by computing the Kullback-Leibler (KL) divergence between their output distributions. A low divergence suggests that the retrieved context had minimal impact, indicating potential memorization. This procedure is model-agnostic, requires no gradient or internal state access, and adds only a single additional forward pass. We further derive PAC-style guarantees that link the KL threshold to user-defined false positive and false negative rates. On the Prompt-WNQA benchmark, RePCS achieves a ROC-AUC of 0.918. This result outperforms the strongest prior method by 6.5 percentage points while keeping latency overhead below 4.7% on an NVIDIA T4 GPU. RePCS offers a lightweight, black-box safeguard to verify whether a RAG system meaningfully leverages retrieval, making it especially valuable in safety-critical applications.
comment: 11 pages, 7 figures, 5 tables
☆ Optimizing Web-Based AI Query Retrieval with GPT Integration in LangChain A CoT-Enhanced Prompt Engineering Approach
Large Language Models have brought a radical change in the process of remote learning students, among other aspects of educative activities. Current retrieval of remote learning resources lacks depth in contextual meaning that provides comprehensive information on complex student queries. This work proposes a novel approach to enhancing remote learning retrieval by integrating GPT-based models within the LangChain framework. We achieve this system in a more intuitive and productive manner using CoT reasoning and prompt engineering. The framework we propose puts much emphasis on increasing the precision and relevance of the retrieval results to return comprehensive and contextually enriched explanations and resources that best suit each student's needs. We also assess the effectiveness of our approach against paradigmatic LLMs and report improvements in user satisfaction and learning outcomes.
☆ Over-squashing in Spatiotemporal Graph Neural Networks
Graph Neural Networks (GNNs) have achieved remarkable success across various domains. However, recent theoretical advances have identified fundamental limitations in their information propagation capabilities, such as over-squashing, where distant nodes fail to effectively exchange information. While extensively studied in static contexts, this issue remains unexplored in Spatiotemporal GNNs (STGNNs), which process sequences associated with graph nodes. Nonetheless, the temporal dimension amplifies this challenge by increasing the information that must be propagated. In this work, we formalize the spatiotemporal over-squashing problem and demonstrate its distinct characteristics compared to the static case. Our analysis reveals that counterintuitively, convolutional STGNNs favor information propagation from points temporally distant rather than close in time. Moreover, we prove that architectures that follow either time-and-space or time-then-space processing paradigms are equally affected by this phenomenon, providing theoretical justification for computationally efficient implementations. We validate our findings on synthetic and real-world datasets, providing deeper insights into their operational dynamics and principled guidance for more effective designs.
☆ Pixel-level Certified Explanations via Randomized Smoothing
Post-hoc attribution methods aim to explain deep learning predictions by highlighting influential input pixels. However, these explanations are highly non-robust: small, imperceptible input perturbations can drastically alter the attribution map while maintaining the same prediction. This vulnerability undermines their trustworthiness and calls for rigorous robustness guarantees of pixel-level attribution scores. We introduce the first certification framework that guarantees pixel-level robustness for any black-box attribution method using randomized smoothing. By sparsifying and smoothing attribution maps, we reformulate the task as a segmentation problem and certify each pixel's importance against $\ell_2$-bounded perturbations. We further propose three evaluation metrics to assess certified robustness, localization, and faithfulness. An extensive evaluation of 12 attribution methods across 5 ImageNet models shows that our certified attributions are robust, interpretable, and faithful, enabling reliable use in downstream tasks. Our code is at https://github.com/AlaaAnani/certified-attributions.
☆ SPARE: Single-Pass Annotation with Reference-Guided Evaluation for Automatic Process Supervision and Reward Modelling
Process or step-wise supervision has played a crucial role in advancing complex multi-step reasoning capabilities of Large Language Models (LLMs). However, efficient, high-quality automated process annotation remains a significant challenge. To address this, we introduce Single-Pass Annotation with Reference-Guided Evaluation (SPARE), a novel structured framework that enables single-pass, per-step annotation by aligning each solution step to one or multiple steps in a reference solution, accompanied by explicit reasoning for evaluation. We show that reference-guided step-level evaluation effectively facilitates process supervision on four datasets spanning three domains: mathematical reasoning, multi-hop compositional question answering, and spatial reasoning. We demonstrate that SPARE, when compared to baselines, improves reasoning performance when used for: (1) fine-tuning models in an offline RL setup for inference-time greedy-decoding, and (2) training reward models for ranking/aggregating multiple LLM-generated outputs. Additionally, SPARE achieves competitive performance on challenging mathematical datasets while offering 2.6 times greater efficiency, requiring only 38% of the runtime, compared to tree search-based automatic annotation. The codebase, along with a trained SPARE-PRM model, is publicly released to facilitate further research and reproducibility.
comment: 8 pages main content, 4 figures, 4 tables
☆ GenHOI: Generalizing Text-driven 4D Human-Object Interaction Synthesis for Unseen Objects
While diffusion models and large-scale motion datasets have advanced text-driven human motion synthesis, extending these advances to 4D human-object interaction (HOI) remains challenging, mainly due to the limited availability of large-scale 4D HOI datasets. In our study, we introduce GenHOI, a novel two-stage framework aimed at achieving two key objectives: 1) generalization to unseen objects and 2) the synthesis of high-fidelity 4D HOI sequences. In the initial stage of our framework, we employ an Object-AnchorNet to reconstruct sparse 3D HOI keyframes for unseen objects, learning solely from 3D HOI datasets, thereby mitigating the dependence on large-scale 4D HOI datasets. Subsequently, we introduce a Contact-Aware Diffusion Model (ContactDM) in the second stage to seamlessly interpolate sparse 3D HOI keyframes into densely temporally coherent 4D HOI sequences. To enhance the quality of generated 4D HOI sequences, we propose a novel Contact-Aware Encoder within ContactDM to extract human-object contact patterns and a novel Contact-Aware HOI Attention to effectively integrate the contact signals into diffusion models. Experimental results show that we achieve state-of-the-art results on the publicly available OMOMO and 3D-FUTURE datasets, demonstrating strong generalization abilities to unseen objects, while enabling high-fidelity 4D HOI generation.
☆ Context-Informed Grounding Supervision
Large language models (LLMs) are often supplemented with external knowledge to provide information not encoded in their parameters or to reduce hallucination. In such cases, we expect the model to generate responses by grounding its response in the provided external context. However, prior work has shown that simply appending context at inference time does not ensure grounded generation. To address this, we propose Context-INformed Grounding Supervision (CINGS), a post-training supervision in which the model is trained with relevant context prepended to the response, while computing the loss only over the response tokens and masking out the context. Our experiments demonstrate that models trained with CINGS exhibit stronger grounding in both textual and visual domains compared to standard instruction-tuned models. In the text domain, CINGS outperforms other training methods across 11 information-seeking datasets and is complementary to inference-time grounding techniques. In the vision-language domain, replacing a vision-language model's LLM backbone with a CINGS-trained model reduces hallucinations across four benchmarks and maintains factual consistency throughout the generated response. This improved grounding comes without degradation in general downstream performance. Finally, we analyze the mechanism underlying the enhanced grounding in CINGS and find that it induces a shift in the model's prior knowledge and behavior, implicitly encouraging greater reliance on the external context.
☆ Co-Creative Learning via Metropolis-Hastings Interaction between Humans and AI
We propose co-creative learning as a novel paradigm where humans and AI, i.e., biological and artificial agents, mutually integrate their partial perceptual information and knowledge to construct shared external representations, a process we interpret as symbol emergence. Unlike traditional AI teaching based on unilateral knowledge transfer, this addresses the challenge of integrating information from inherently different modalities. We empirically test this framework using a human-AI interaction model based on the Metropolis-Hastings naming game (MHNG), a decentralized Bayesian inference mechanism. In an online experiment, 69 participants played a joint attention naming game (JA-NG) with one of three computer agent types (MH-based, always-accept, or always-reject) under partial observability. Results show that human-AI pairs with an MH-based agent significantly improved categorization accuracy through interaction and achieved stronger convergence toward a shared sign system. Furthermore, human acceptance behavior aligned closely with the MH-derived acceptance probability. These findings provide the first empirical evidence for co-creative learning emerging in human-AI dyads via MHNG-based interaction. This suggests a promising path toward symbiotic AI systems that learn with humans, rather than from them, by dynamically aligning perceptual experiences, opening a new venue for symbiotic AI alignment.
☆ RE-IMAGINE: Symbolic Benchmark Synthesis for Reasoning Evaluation ICML 2025
Recent Large Language Models (LLMs) have reported high accuracy on reasoning benchmarks. However, it is still unclear whether the observed results arise from true reasoning or from statistical recall of the training set. Inspired by the ladder of causation (Pearl, 2009) and its three levels (associations, interventions and counterfactuals), this paper introduces RE-IMAGINE, a framework to characterize a hierarchy of reasoning ability in LLMs, alongside an automated pipeline to generate problem variations at different levels of the hierarchy. By altering problems in an intermediate symbolic representation, RE-IMAGINE generates arbitrarily many problems that are not solvable using memorization alone. Moreover, the framework is general and can work across reasoning domains, including math, code, and logic. We demonstrate our framework on four widely-used benchmarks to evaluate several families of LLMs, and observe reductions in performance when the models are queried with problem variations. These assessments indicate a degree of reliance on statistical recall for past performance, and open the door to further research targeting skills across the reasoning hierarchy.
comment: ICML 2025
☆ Uncovering Intention through LLM-Driven Code Snippet Description Generation
Documenting code snippets is essential to pinpoint key areas where both developers and users should pay attention. Examples include usage examples and other Application Programming Interfaces (APIs), which are especially important for third-party libraries. With the rise of Large Language Models (LLMs), the key goal is to investigate the kinds of description developers commonly use and evaluate how well an LLM, in this case Llama, can support description generation. We use NPM Code Snippets, consisting of 185,412 packages with 1,024,579 code snippets. From there, we use 400 code snippets (and their descriptions) as samples. First, our manual classification found that the majority of original descriptions (55.5%) highlight example-based usage. This finding emphasizes the importance of clear documentation, as some descriptions lacked sufficient detail to convey intent. Second, the LLM correctly identified the majority of original descriptions as "Example" (79.75%), which is identical to our manual finding, showing a propensity for generalization. Third, compared to the originals, the produced description had an average similarity score of 0.7173, suggesting relevance but room for improvement. Scores below 0.9 indicate some irrelevance. Our results show that depending on the task of the code snippet, the intention of the document may differ from being instructions for usage, installations, or descriptive learning examples for any user of a library.
comment: 6 pages, 3 figures, 4 tables, conference paper
☆ Warping and Matching Subsequences Between Time Series
Comparing time series is essential in various tasks such as clustering and classification. While elastic distance measures that allow warping provide a robust quantitative comparison, a qualitative comparison on top of them is missing. Traditional visualizations focus on point-to-point alignment and do not convey the broader structural relationships at the level of subsequences. This limitation makes it difficult to understand how and where one time series shifts, speeds up or slows down with respect to another. To address this, we propose a novel technique that simplifies the warping path to highlight, quantify and visualize key transformations (shift, compression, difference in amplitude). By offering a clearer representation of how subsequences match between time series, our method enhances interpretability in time series comparison.
☆ Zero-Shot Reinforcement Learning Under Partial Observability
Recent work has shown that, under certain assumptions, zero-shot reinforcement learning (RL) methods can generalise to any unseen task in an environment after reward-free pre-training. Access to Markov states is one such assumption, yet, in many real-world applications, the Markov state is only partially observable. Here, we explore how the performance of standard zero-shot RL methods degrades when subjected to partially observability, and show that, as in single-task RL, memory-based architectures are an effective remedy. We evaluate our memory-based zero-shot RL methods in domains where the states, rewards and a change in dynamics are partially observed, and show improved performance over memory-free baselines. Our code is open-sourced via: https://enjeeneer.io/projects/bfms-with-memory/.
comment: Reinforcement Learning Conference 2025
☆ Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material
3D AI-generated content (AIGC) is a passionate field that has significantly accelerated the creation of 3D models in gaming, film, and design. Despite the development of several groundbreaking models that have revolutionized 3D generation, the field remains largely accessible only to researchers, developers, and designers due to the complexities involved in collecting, processing, and training 3D models. To address these challenges, we introduce Hunyuan3D 2.1 as a case study in this tutorial. This tutorial offers a comprehensive, step-by-step guide on processing 3D data, training a 3D generative model, and evaluating its performance using Hunyuan3D 2.1, an advanced system for producing high-resolution, textured 3D assets. The system comprises two core components: the Hunyuan3D-DiT for shape generation and the Hunyuan3D-Paint for texture synthesis. We will explore the entire workflow, including data preparation, model architecture, training strategies, evaluation metrics, and deployment. By the conclusion of this tutorial, you will have the knowledge to finetune or develop a robust 3D generative model suitable for applications in gaming, virtual reality, and industrial design.
comment: Github link: https://github.com/Tencent-Hunyuan/Hunyuan3D-2.1
☆ Reward Models in Deep Reinforcement Learning: A Survey IJCAI 2025
In reinforcement learning (RL), agents continually interact with the environment and use the feedback to refine their behavior. To guide policy optimization, reward models are introduced as proxies of the desired objectives, such that when the agent maximizes the accumulated reward, it also fulfills the task designer's intentions. Recently, significant attention from both academic and industrial researchers has focused on developing reward models that not only align closely with the true objectives but also facilitate policy optimization. In this survey, we provide a comprehensive review of reward modeling techniques within the deep RL literature. We begin by outlining the background and preliminaries in reward modeling. Next, we present an overview of recent reward modeling approaches, categorizing them based on the source, the mechanism, and the learning paradigm. Building on this understanding, we discuss various applications of these reward modeling techniques and review methods for evaluating reward models. Finally, we conclude by highlighting promising research directions in reward modeling. Altogether, this survey includes both established and emerging methods, filling the vacancy of a systematic review of reward models in current literature.
comment: IJCAI 2025 Survey Track (To Appear)
☆ Unifying VXAI: A Systematic Review and Framework for the Evaluation of Explainable AI
Modern AI systems frequently rely on opaque black-box models, most notably Deep Neural Networks, whose performance stems from complex architectures with millions of learned parameters. While powerful, their complexity poses a major challenge to trustworthiness, particularly due to a lack of transparency. Explainable AI (XAI) addresses this issue by providing human-understandable explanations of model behavior. However, to ensure their usefulness and trustworthiness, such explanations must be rigorously evaluated. Despite the growing number of XAI methods, the field lacks standardized evaluation protocols and consensus on appropriate metrics. To address this gap, we conduct a systematic literature review following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and introduce a unified framework for the eValuation of XAI (VXAI). We identify 362 relevant publications and aggregate their contributions into 41 functionally similar metric groups. In addition, we propose a three-dimensional categorization scheme spanning explanation type, evaluation contextuality, and explanation quality desiderata. Our framework provides the most comprehensive and structured overview of VXAI to date. It supports systematic metric selection, promotes comparability across methods, and offers a flexible foundation for future extensions.
comment: Submitted to TMLR, under review
☆ MCOO-SLAM: A Multi-Camera Omnidirectional Object SLAM System
Object-level SLAM offers structured and semantically meaningful environment representations, making it more interpretable and suitable for high-level robotic tasks. However, most existing approaches rely on RGB-D sensors or monocular views, which suffer from narrow fields of view, occlusion sensitivity, and limited depth perception-especially in large-scale or outdoor environments. These limitations often restrict the system to observing only partial views of objects from limited perspectives, leading to inaccurate object modeling and unreliable data association. In this work, we propose MCOO-SLAM, a novel Multi-Camera Omnidirectional Object SLAM system that fully leverages surround-view camera configurations to achieve robust, consistent, and semantically enriched mapping in complex outdoor scenarios. Our approach integrates point features and object-level landmarks enhanced with open-vocabulary semantics. A semantic-geometric-temporal fusion strategy is introduced for robust object association across multiple views, leading to improved consistency and accurate object modeling, and an omnidirectional loop closure module is designed to enable viewpoint-invariant place recognition using scene-level descriptors. Furthermore, the constructed map is abstracted into a hierarchical 3D scene graph to support downstream reasoning tasks. Extensive experiments in real-world demonstrate that MCOO-SLAM achieves accurate localization and scalable object-level mapping with improved robustness to occlusion, pose variation, and environmental complexity.
☆ A Real-time Endoscopic Image Denoising System
Endoscopes featuring a miniaturized design have significantly enhanced operational flexibility, portability, and diagnostic capability while substantially reducing the invasiveness of medical procedures. Recently, single-use endoscopes equipped with an ultra-compact analogue image sensor measuring less than 1mm x 1mm bring revolutionary advancements to medical diagnosis. They reduce the structural redundancy and large capital expenditures associated with reusable devices, eliminate the risk of patient infections caused by inadequate disinfection, and alleviate patient suffering. However, the limited photosensitive area results in reduced photon capture per pixel, requiring higher photon sensitivity settings to maintain adequate brightness. In high-contrast medical imaging scenarios, the small-sized sensor exhibits a constrained dynamic range, making it difficult to simultaneously capture details in both highlights and shadows, and additional localized digital gain is required to compensate. Moreover, the simplified circuit design and analog signal transmission introduce additional noise sources. These factors collectively contribute to significant noise issues in processed endoscopic images. In this work, we developed a comprehensive noise model for analog image sensors in medical endoscopes, addressing three primary noise types: fixed-pattern noise, periodic banding noise, and mixed Poisson-Gaussian noise. Building on this analysis, we propose a hybrid denoising system that synergistically combines traditional image processing algorithms with advanced learning-based techniques for captured raw frames from sensors. Experiments demonstrate that our approach effectively reduces image noise without fine detail loss or color distortion, while achieving real-time performance on FPGA platforms and an average PSNR improvement from 21.16 to 33.05 on our test dataset.
☆ Evaluation Pipeline for systematically searching for Anomaly Detection Systems
Digitalization in the medical world provides major benefits while making it a target for attackers and thus hard to secure. To deal with network intruders we propose an anomaly detection system on hardware to detect malicious clients in real-time. We meet real-time and power restrictions using FPGAs. Overall system performance is achieved via the presented holistic system evaluation.
comment: Submitted to 18th HiPEAC Workshop on Reconfigurable Computing (WRC'2024)
☆ Efficient and Generalizable Environmental Understanding for Visual Navigation
Visual Navigation is a core task in Embodied AI, enabling agents to navigate complex environments toward given objectives. Across diverse settings within Navigation tasks, many necessitate the modelling of sequential data accumulated from preceding time steps. While existing methods perform well, they typically process all historical observations simultaneously, overlooking the internal association structure within the data, which may limit the potential for further improvements in task performance. We address this by examining the unique characteristics of Navigation tasks through the lens of causality, introducing a causal framework to highlight the limitations of conventional sequential methods. Leveraging this insight, we propose Causality-Aware Navigation (CAN), which incorporates a Causal Understanding Module to enhance the agent's environmental understanding capability. Empirical evaluations show that our approach consistently outperforms baselines across various tasks and simulation environments. Extensive ablations studies attribute these gains to the Causal Understanding Module, which generalizes effectively in both Reinforcement and Supervised Learning settings without computational overhead.
☆ Open-World Object Counting in Videos
We introduce a new task of open-world object counting in videos: given a text description, or an image example, that specifies the target object, the objective is to enumerate all the unique instances of the target objects in the video. This task is especially challenging in crowded scenes with occlusions and similar objects, where avoiding double counting and identifying reappearances is crucial. To this end, we make the following contributions: we introduce a model, CountVid, for this task. It leverages an image-based counting model, and a promptable video segmentation and tracking model to enable automated, open-world object counting across video frames. To evaluate its performance, we introduce VideoCount, a new dataset for our novel task built from the TAO and MOT20 tracking datasets, as well as from videos of penguins and metal alloy crystallization captured by x-rays. Using this dataset, we demonstrate that CountVid provides accurate object counts, and significantly outperforms strong baselines. The VideoCount dataset, the CountVid model, and all the code are available at https://github.com/niki-amini-naieni/CountVid/.
☆ When and How Unlabeled Data Provably Improve In-Context Learning
Recent research shows that in-context learning (ICL) can be effective even when demonstrations have missing or incorrect labels. To shed light on this capability, we examine a canonical setting where the demonstrations are drawn according to a binary Gaussian mixture model (GMM) and a certain fraction of the demonstrations have missing labels. We provide a comprehensive theoretical study to show that: (1) The loss landscape of one-layer linear attention models recover the optimal fully-supervised estimator but completely fail to exploit unlabeled data; (2) In contrast, multilayer or looped transformers can effectively leverage unlabeled data by implicitly constructing estimators of the form $\sum_{i\ge 0} a_i (X^\top X)^iX^\top y$ with $X$ and $y$ denoting features and partially-observed labels (with missing entries set to zero). We characterize the class of polynomials that can be expressed as a function of depth and draw connections to Expectation Maximization, an iterative pseudo-labeling algorithm commonly used in semi-supervised learning. Importantly, the leading polynomial power is exponential in depth, so mild amount of depth/looping suffices. As an application of theory, we propose looping off-the-shelf tabular foundation models to enhance their semi-supervision capabilities. Extensive evaluations on real-world datasets show that our method significantly improves the semisupervised tabular learning performance over the standard single pass inference.
☆ J3DAI: A tiny DNN-Based Edge AI Accelerator for 3D-Stacked CMOS Image Sensor
This paper presents J3DAI, a tiny deep neural network-based hardware accelerator for a 3-layer 3D-stacked CMOS image sensor featuring an artificial intelligence (AI) chip integrating a Deep Neural Network (DNN)-based accelerator. The DNN accelerator is designed to efficiently perform neural network tasks such as image classification and segmentation. This paper focuses on the digital system of J3DAI, highlighting its Performance-Power-Area (PPA) characteristics and showcasing advanced edge AI capabilities on a CMOS image sensor. To support hardware, we utilized the Aidge comprehensive software framework, which enables the programming of both the host processor and the DNN accelerator. Aidge supports post-training quantization, significantly reducing memory footprint and computational complexity, making it crucial for deploying models on resource-constrained hardware like J3DAI. Our experimental results demonstrate the versatility and efficiency of this innovative design in the field of edge AI, showcasing its potential to handle both simple and computationally intensive tasks. Future work will focus on further optimizing the architecture and exploring new applications to fully leverage the capabilities of J3DAI. As edge AI continues to grow in importance, innovations like J3DAI will play a crucial role in enabling real-time, low-latency, and energy-efficient AI processing at the edge.
comment: Preprint from ISLPED 2025. 979-8-3315-2710-5/25/$31.00 \c{opyright}2025 IEEE
☆ MapFM: Foundation Model-Driven HD Mapping with Multi-Task Contextual Learning
In autonomous driving, high-definition (HD) maps and semantic maps in bird's-eye view (BEV) are essential for accurate localization, planning, and decision-making. This paper introduces an enhanced End-to-End model named MapFM for online vectorized HD map generation. We show significantly boost feature representation quality by incorporating powerful foundation model for encoding camera images. To further enrich the model's understanding of the environment and improve prediction quality, we integrate auxiliary prediction heads for semantic segmentation in the BEV representation. This multi-task learning approach provides richer contextual supervision, leading to a more comprehensive scene representation and ultimately resulting in higher accuracy and improved quality of the predicted vectorized HD maps. The source code is available at https://github.com/LIvanoff/MapFM.
comment: Preprint. Submitted. 12 pages, 4 figures
☆ Active Learning-Guided Seq2Seq Variational Autoencoder for Multi-target Inhibitor Generation
Simultaneously optimizing molecules against multiple therapeutic targets remains a profound challenge in drug discovery, particularly due to sparse rewards and conflicting design constraints. We propose a structured active learning (AL) paradigm integrating a sequence-to-sequence (Seq2Seq) variational autoencoder (VAE) into iterative loops designed to balance chemical diversity, molecular quality, and multi-target affinity. Our method alternates between expanding chemically feasible regions of latent space and progressively constraining molecules based on increasingly stringent multi-target docking thresholds. In a proof-of-concept study targeting three related coronavirus main proteases (SARS-CoV-2, SARS-CoV, MERS-CoV), our approach efficiently generated a structurally diverse set of pan-inhibitor candidates. We demonstrate that careful timing and strategic placement of chemical filters within this active learning pipeline markedly enhance exploration of beneficial chemical space, transforming the sparse-reward, multi-objective drug design problem into an accessible computational task. Our framework thus provides a generalizable roadmap for efficiently navigating complex polypharmacological landscapes.
comment: 16 pages, 7 figures
☆ ConLID: Supervised Contrastive Learning for Low-Resource Language Identification EMNLP
Language identification (LID) is a critical step in curating multilingual LLM pretraining corpora from web crawls. While many studies on LID model training focus on collecting diverse training data to improve performance, low-resource languages -- often limited to single-domain data, such as the Bible -- continue to perform poorly. To resolve these class imbalance and bias issues, we propose a novel supervised contrastive learning (SCL) approach to learn domain-invariant representations for low-resource languages. Through an extensive analysis, we show that our approach improves LID performance on out-of-domain data for low-resource languages by 3.2%, demonstrating its effectiveness in enhancing LID models.
comment: Submitted to EMNLP
☆ Cohort Discovery: A Survey on LLM-Assisted Clinical Trial Recruitment
Recent advances in LLMs have greatly improved general-domain NLP tasks. Yet, their adoption in critical domains, such as clinical trial recruitment, remains limited. As trials are designed in natural language and patient data is represented as both structured and unstructured text, the task of matching trials and patients benefits from knowledge aggregation and reasoning abilities of LLMs. Classical approaches are trial-specific and LLMs with their ability to consolidate distributed knowledge hold the potential to build a more general solution. Yet recent applications of LLM-assisted methods rely on proprietary models and weak evaluation benchmarks. In this survey, we are the first to analyze the task of trial-patient matching and contextualize emerging LLM-based approaches in clinical trial recruitment. We critically examine existing benchmarks, approaches and evaluation frameworks, the challenges to adopting LLM technologies in clinical research and exciting future directions.
☆ Human Motion Capture from Loose and Sparse Inertial Sensors with Garment-aware Diffusion Models IJCAI 2025
Motion capture using sparse inertial sensors has shown great promise due to its portability and lack of occlusion issues compared to camera-based tracking. Existing approaches typically assume that IMU sensors are tightly attached to the human body. However, this assumption often does not hold in real-world scenarios. In this paper, we present a new task of full-body human pose estimation using sparse, loosely attached IMU sensors. To solve this task, we simulate IMU recordings from an existing garment-aware human motion dataset. We developed transformer-based diffusion models to synthesize loose IMU data and estimate human poses based on this challenging loose IMU data. In addition, we show that incorporating garment-related parameters while training the model on simulated loose data effectively maintains expressiveness and enhances the ability to capture variations introduced by looser or tighter garments. Experiments show that our proposed diffusion methods trained on simulated and synthetic data outperformed the state-of-the-art methods quantitatively and qualitatively, opening up a promising direction for future research.
comment: Accepted by IJCAI 2025
☆ Unlocking Post-hoc Dataset Inference with Synthetic Data ICML 2025
The remarkable capabilities of Large Language Models (LLMs) can be mainly attributed to their massive training datasets, which are often scraped from the internet without respecting data owners' intellectual property rights. Dataset Inference (DI) offers a potential remedy by identifying whether a suspect dataset was used in training, thereby enabling data owners to verify unauthorized use. However, existing DI methods require a private set-known to be absent from training-that closely matches the compromised dataset's distribution. Such in-distribution, held-out data is rarely available in practice, severely limiting the applicability of DI. In this work, we address this challenge by synthetically generating the required held-out set. Our approach tackles two key obstacles: (1) creating high-quality, diverse synthetic data that accurately reflects the original distribution, which we achieve via a data generator trained on a carefully designed suffix-based completion task, and (2) bridging likelihood gaps between real and synthetic data, which is realized through post-hoc calibration. Extensive experiments on diverse text datasets show that using our generated data as a held-out set enables DI to detect the original training sets with high confidence, while maintaining a low false positive rate. This result empowers copyright owners to make legitimate claims on data usage and demonstrates our method's reliability for real-world litigations. Our code is available at https://github.com/sprintml/PostHocDatasetInference.
comment: Accepted at ICML 2025
☆ Domain Adaptation for Image Classification of Defects in Semiconductor Manufacturing
In the semiconductor sector, due to high demand but also strong and increasing competition, time to market and quality are key factors in securing significant market share in various application areas. Thanks to the success of deep learning methods in recent years in the computer vision domain, Industry 4.0 and 5.0 applications, such as defect classification, have achieved remarkable success. In particular, Domain Adaptation (DA) has proven highly effective since it focuses on using the knowledge learned on a (source) domain to adapt and perform effectively on a different but related (target) domain. By improving robustness and scalability, DA minimizes the need for extensive manual re-labeling or re-training of models. This not only reduces computational and resource costs but also allows human experts to focus on high-value tasks. Therefore, we tested the efficacy of DA techniques in semi-supervised and unsupervised settings within the context of the semiconductor field. Moreover, we propose the DBACS approach, a CycleGAN-inspired model enhanced with additional loss terms to improve performance. All the approaches are studied and validated on real-world Electron Microscope images considering the unsupervised and semi-supervised settings, proving the usefulness of our method in advancing DA techniques for the semiconductor field.
☆ RAS-Eval: A Comprehensive Benchmark for Security Evaluation of LLM Agents in Real-World Environments
The rapid deployment of Large language model (LLM) agents in critical domains like healthcare and finance necessitates robust security frameworks. To address the absence of standardized evaluation benchmarks for these agents in dynamic environments, we introduce RAS-Eval, a comprehensive security benchmark supporting both simulated and real-world tool execution. RAS-Eval comprises 80 test cases and 3,802 attack tasks mapped to 11 Common Weakness Enumeration (CWE) categories, with tools implemented in JSON, LangGraph, and Model Context Protocol (MCP) formats. We evaluate 6 state-of-the-art LLMs across diverse scenarios, revealing significant vulnerabilities: attacks reduced agent task completion rates (TCR) by 36.78% on average and achieved an 85.65% success rate in academic settings. Notably, scaling laws held for security capabilities, with larger models outperforming smaller counterparts. Our findings expose critical risks in real-world agent deployments and provide a foundational framework for future security research. Code and data are available at https://github.com/lanzer-tree/RAS-Eval.
comment: 12 pages, 8 figures
☆ Singular Value Decomposition on Kronecker Adaptation for Large Language Model
Large pre-trained Transformer models achieve state-of-the-art results across diverse language and reasoning tasks, but full fine-tuning incurs substantial storage, memory, and computational overhead. Parameter-efficient fine-tuning (PEFT) methods mitigate these costs by learning only a small subset of task-specific parameters, yet existing approaches either introduce inference-time latency (adapter modules), suffer from suboptimal convergence (randomly initialized low-rank updates), or rely on fixed rank choices that may not match task complexity (Kronecker-based decompositions). We propose SoKA (SVD on Kronecker Adaptation), a novel PEFT strategy that combines Kronecker-product tensor factorization with SVD-driven initialization and spectrum-aware dynamic rank selection. Our Kronecker-Product SVD (KPSVD) procedure extracts principal components of the full weight update into compact Kronecker factors, while an adaptive rank selection algorithm uses energy-threshold and elbow-point criteria to prune negligible components. Empirical evaluation on LLaMA2-7B across arithmetic reasoning (GSM8K), formal mathematics (MATH), and code generation (MBPP) demonstrates that SoKA requires only 0.99M trainable parameters, 25% fewer than LoRA/PiSSA, while matching or exceeding baseline performance. Moreover, SoKA exhibits faster convergence and more stable gradients, highlighting its robustness and efficiency for large-scale model adaptation.
☆ Joint Computation Offloading and Resource Allocation for Uncertain Maritime MEC via Cooperation of UAVs and Vessels
The computation demands from the maritime Internet of Things (MIoT) increase rapidly in recent years, and the unmanned aerial vehicles (UAVs) and vessels based multi-access edge computing (MEC) can fulfill these MIoT requirements. However, the uncertain maritime tasks present significant challenges of inefficient computation offloading and resource allocation. In this paper, we focus on the maritime computation offloading and resource allocation through the cooperation of UAVs and vessels, with consideration of uncertain tasks. Specifically, we propose a cooperative MEC framework for computation offloading and resource allocation, including MIoT devices, UAVs and vessels. Then, we formulate the optimization problem to minimize the total execution time. As for the uncertain MIoT tasks, we leverage Lyapunov optimization to tackle the unpredictable task arrivals and varying computational resource availability. By converting the long-term constraints into short-term constraints, we obtain a set of small-scale optimization problems. Further, considering the heterogeneity of actions and resources of UAVs and vessels, we reformulate the small-scale optimization problem into a Markov game (MG). Moreover, a heterogeneous-agent soft actor-critic is proposed to sequentially update various neural networks and effectively solve the MG problem. Finally, simulations are conducted to verify the effectiveness in addressing computational offloading and resource allocation.
☆ A Comparative Study of Task Adaptation Techniques of Large Language Models for Identifying Sustainable Development Goals
In 2012, the United Nations introduced 17 Sustainable Development Goals (SDGs) aimed at creating a more sustainable and improved future by 2030. However, tracking progress toward these goals is difficult because of the extensive scale and complexity of the data involved. Text classification models have become vital tools in this area, automating the analysis of vast amounts of text from a variety of sources. Additionally, large language models (LLMs) have recently proven indispensable for many natural language processing tasks, including text classification, thanks to their ability to recognize complex linguistic patterns and semantics. This study analyzes various proprietary and open-source LLMs for a single-label, multi-class text classification task focused on the SDGs. Then, it also evaluates the effectiveness of task adaptation techniques (i.e., in-context learning approaches), namely Zero-Shot and Few-Shot Learning, as well as Fine-Tuning within this domain. The results reveal that smaller models, when optimized through prompt engineering, can perform on par with larger models like OpenAI's GPT (Generative Pre-trained Transformer).
comment: Submitted to IEEE Access
☆ Multi-Agent Reinforcement Learning for Autonomous Multi-Satellite Earth Observation: A Realistic Case Study
The exponential growth of Low Earth Orbit (LEO) satellites has revolutionised Earth Observation (EO) missions, addressing challenges in climate monitoring, disaster management, and more. However, autonomous coordination in multi-satellite systems remains a fundamental challenge. Traditional optimisation approaches struggle to handle the real-time decision-making demands of dynamic EO missions, necessitating the use of Reinforcement Learning (RL) and Multi-Agent Reinforcement Learning (MARL). In this paper, we investigate RL-based autonomous EO mission planning by modelling single-satellite operations and extending to multi-satellite constellations using MARL frameworks. We address key challenges, including energy and data storage limitations, uncertainties in satellite observations, and the complexities of decentralised coordination under partial observability. By leveraging a near-realistic satellite simulation environment, we evaluate the training stability and performance of state-of-the-art MARL algorithms, including PPO, IPPO, MAPPO, and HAPPO. Our results demonstrate that MARL can effectively balance imaging and resource management while addressing non-stationarity and reward interdependency in multi-satellite coordination. The insights gained from this study provide a foundation for autonomous satellite operations, offering practical guidelines for improving policy learning in decentralised EO missions.
☆ HeurAgenix: Leveraging LLMs for Solving Complex Combinatorial Optimization Challenges
Heuristic algorithms play a vital role in solving combinatorial optimization (CO) problems, yet traditional designs depend heavily on manual expertise and struggle to generalize across diverse instances. We introduce \textbf{HeurAgenix}, a two-stage hyper-heuristic framework powered by large language models (LLMs) that first evolves heuristics and then selects among them automatically. In the heuristic evolution phase, HeurAgenix leverages an LLM to compare seed heuristic solutions with higher-quality solutions and extract reusable evolution strategies. During problem solving, it dynamically picks the most promising heuristic for each problem state, guided by the LLM's perception ability. For flexibility, this selector can be either a state-of-the-art LLM or a fine-tuned lightweight model with lower inference cost. To mitigate the scarcity of reliable supervision caused by CO complexity, we fine-tune the lightweight heuristic selector with a dual-reward mechanism that jointly exploits singals from selection preferences and state perception, enabling robust selection under noisy annotations. Extensive experiments on canonical benchmarks show that HeurAgenix not only outperforms existing LLM-based hyper-heuristics but also matches or exceeds specialized solvers. Code is available at https://github.com/microsoft/HeurAgenix.
comment: 27 pages,9 figures
☆ Accessible Gesture-Driven Augmented Reality Interaction System
Augmented reality (AR) offers immersive interaction but remains inaccessible for users with motor impairments or limited dexterity due to reliance on precise input methods. This study proposes a gesture-based interaction system for AR environments, leveraging deep learning to recognize hand and body gestures from wearable sensors and cameras, adapting interfaces to user capabilities. The system employs vision transformers (ViTs), temporal convolutional networks (TCNs), and graph attention networks (GATs) for gesture processing, with federated learning ensuring privacy-preserving model training across diverse users. Reinforcement learning optimizes interface elements like menu layouts and interaction modes. Experiments demonstrate a 20% improvement in task completion efficiency and a 25% increase in user satisfaction for motor-impaired users compared to baseline AR systems. This approach enhances AR accessibility and scalability. Keywords: Deep learning, Federated learning, Gesture recognition, Augmented reality, Accessibility, Human-computer interaction
☆ Classification of Multi-Parametric Body MRI Series Using Deep Learning
Multi-parametric magnetic resonance imaging (mpMRI) exams have various series types acquired with different imaging protocols. The DICOM headers of these series often have incorrect information due to the sheer diversity of protocols and occasional technologist errors. To address this, we present a deep learning-based classification model to classify 8 different body mpMRI series types so that radiologists read the exams efficiently. Using mpMRI data from various institutions, multiple deep learning-based classifiers of ResNet, EfficientNet, and DenseNet are trained to classify 8 different MRI series, and their performance is compared. Then, the best-performing classifier is identified, and its classification capability under the setting of different training data quantities is studied. Also, the model is evaluated on the out-of-training-distribution datasets. Moreover, the model is trained using mpMRI exams obtained from different scanners in two training strategies, and its performance is tested. Experimental results show that the DenseNet-121 model achieves the highest F1-score and accuracy of 0.966 and 0.972 over the other classification models with p-value$<$0.05. The model shows greater than 0.95 accuracy when trained with over 729 studies of the training data, whose performance improves as the training data quantities grew larger. On the external data with the DLDS and CPTAC-UCEC datasets, the model yields 0.872 and 0.810 accuracy for each. These results indicate that in both the internal and external datasets, the DenseNet-121 model attains high accuracy for the task of classifying 8 body MRI series types.
☆ LLM Agent for Hyper-Parameter Optimization
Hyper-parameters are essential and critical for the performance of communication algorithms. However, current hyper-parameters tuning methods for warm-start particles swarm optimization with cross and mutation (WS-PSO-CM) algortihm for radio map-enabled unmanned aerial vehicle (UAV) trajectory and communication are primarily heuristic-based, exhibiting low levels of automation and unsatisfactory performance. In this paper, we design an large language model (LLM) agent for automatic hyper-parameters-tuning, where an iterative framework and model context protocol (MCP) are applied. In particular, the LLM agent is first setup via a profile, which specifies the mission, background, and output format. Then, the LLM agent is driven by the prompt requirement, and iteratively invokes WS-PSO-CM algorithm for exploration. Finally, the LLM agent autonomously terminates the loop and returns a set of hyper-parameters. Our experiment results show that the minimal sum-rate achieved by hyper-parameters generated via our LLM agent is significantly higher than those by both human heuristics and random generation methods. This indicates that an LLM agent with PSO knowledge and WS-PSO-CM algorithm background is useful in finding high-performance hyper-parameters.
comment: 6 pages, 6 figures
☆ SonicVerse: Multi-Task Learning for Music Feature-Informed Captioning
Detailed captions that accurately reflect the characteristics of a music piece can enrich music databases and drive forward research in music AI. This paper introduces a multi-task music captioning model, SonicVerse, that integrates caption generation with auxiliary music feature detection tasks such as key detection, vocals detection, and more, so as to directly capture both low-level acoustic details as well as high-level musical attributes. The key contribution is a projection-based architecture that transforms audio input into language tokens, while simultaneously detecting music features through dedicated auxiliary heads. The outputs of these heads are also projected into language tokens, to enhance the captioning input. This framework not only produces rich, descriptive captions for short music fragments but also directly enables the generation of detailed time-informed descriptions for longer music pieces, by chaining the outputs using a large-language model. To train the model, we extended the MusicBench dataset by annotating it with music features using MIRFLEX, a modular music feature extractor, resulting in paired audio, captions and music feature data. Experimental results show that incorporating features in this way improves the quality and detail of the generated captions.
comment: 14 pages, 2 figures, Accepted to AIMC 2025
☆ Thunder-Tok: Minimizing Tokens per Word in Tokenizing Korean Texts for Generative Language Models
This paper introduces Thunder-Tok, a new Korean tokenizer designed to reduce token fertility without compromising model performance. Our approach uses a rule-based pre-tokenization method that aligns with the linguistic structure of the Korean language. We also create a seed vocabulary containing tokens that resemble linguistic units and employ a branching entropy-based selection algorithm. These techniques increase the average token length, thus lowering fertility while preserving linguistic information. Experimental results indicate that Thunder-Tok reduces fertility by approximately 10% (i.e., reduces the number of tokens by 10%, improving the inference speed by 10%) compared to BPE without compromising performance across various downstream tasks. These findings demonstrate that our linguistically informed approach is effective and practical for designing efficient tokenizers for language models.
☆ Modeling the One-to-Many Property in Open-Domain Dialogue with LLMs
Open-domain Dialogue (OD) exhibits a one-to-many (o2m) property, whereby multiple appropriate responses exist for a single dialogue context. Despite prior research showing that modeling this property boosts response diversity, most modern LLM-based dialogue agents do not explicitly do so. In this work, we model the o2m property of OD in LLMs by decomposing OD generation into two key tasks: Multi-Response Generation (MRG) and Preference-based Selection (PS), which entail generating a set of n semantically and lexically diverse high-quality responses for a given dialogue context, followed by selecting a single response based on human preference, respectively. To facilitate MRG and PS, we introduce o2mDial, a dialogue corpus explicitly designed to capture the o2m property by featuring multiple plausible responses for each context. Leveraging o2mDial, we propose new in-context learning and instruction-tuning strategies, as well as novel evaluation metrics for MRG, alongside a model-based approach for PS. Empirical results demonstrate that applying the proposed two-stage framework to smaller LLMs for OD generation enhances overall response diversity while maintaining contextual coherence, improving response quality by up to 90%, bringing them closer to the performance of larger models.
☆ Advancing Loss Functions in Recommender Systems: A Comparative Study with a Rényi Divergence-Based Solution AAAI 2025
Loss functions play a pivotal role in optimizing recommendation models. Among various loss functions, Softmax Loss (SL) and Cosine Contrastive Loss (CCL) are particularly effective. Their theoretical connections and differences warrant in-depth exploration. This work conducts comprehensive analyses of these losses, yielding significant insights: 1) Common strengths -- both can be viewed as augmentations of traditional losses with Distributional Robust Optimization (DRO), enhancing robustness to distributional shifts; 2) Respective limitations -- stemming from their use of different distribution distance metrics in DRO optimization, SL exhibits high sensitivity to false negative instances, whereas CCL suffers from low data utilization. To address these limitations, this work proposes a new loss function, DrRL, which generalizes SL and CCL by leveraging R\'enyi-divergence in DRO optimization. DrRL incorporates the advantageous structures of both SL and CCL, and can be demonstrated to effectively mitigate their limitations. Extensive experiments have been conducted to validate the superiority of DrRL on both recommendation accuracy and robustness.
comment: AAAI 2025
☆ Transit for All: Mapping Equitable Bike2Subway Connection using Region Representation Learning
Ensuring equitable public transit access remains challenging, particularly in densely populated cities like New York City (NYC), where low-income and minority communities often face limited transit accessibility. Bike-sharing systems (BSS) can bridge these equity gaps by providing affordable first- and last-mile connections. However, strategically expanding BSS into underserved neighborhoods is difficult due to uncertain bike-sharing demand at newly planned ("cold-start") station locations and limitations in traditional accessibility metrics that may overlook realistic bike usage potential. We introduce Transit for All (TFA), a spatial computing framework designed to guide the equitable expansion of BSS through three components: (1) spatially-informed bike-sharing demand prediction at cold-start stations using region representation learning that integrates multimodal geospatial data, (2) comprehensive transit accessibility assessment leveraging our novel weighted Public Transport Accessibility Level (wPTAL) by combining predicted bike-sharing demand with conventional transit accessibility metrics, and (3) strategic recommendations for new bike station placements that consider potential ridership and equity enhancement. Using NYC as a case study, we identify transit accessibility gaps that disproportionately impact low-income and minority communities in historically underserved neighborhoods. Our results show that strategically placing new stations guided by wPTAL notably reduces disparities in transit access related to economic and demographic factors. From our study, we demonstrate that TFA provides practical guidance for urban planners to promote equitable transit and enhance the quality of life in underserved urban communities.
☆ Improving Dialogue Discourse Parsing through Discourse-aware Utterance Clarification ACL2025
Dialogue discourse parsing aims to identify and analyze discourse relations between the utterances within dialogues. However, linguistic features in dialogues, such as omission and idiom, frequently introduce ambiguities that obscure the intended discourse relations, posing significant challenges for parsers. To address this issue, we propose a Discourse-aware Clarification Module (DCM) to enhance the performance of the dialogue discourse parser. DCM employs two distinct reasoning processes: clarification type reasoning and discourse goal reasoning. The former analyzes linguistic features, while the latter distinguishes the intended relation from the ambiguous one. Furthermore, we introduce Contribution-aware Preference Optimization (CPO) to mitigate the risk of erroneous clarifications, thereby reducing cascading errors. CPO enables the parser to assess the contributions of the clarifications from DCM and provide feedback to optimize the DCM, enhancing its adaptability and alignment with the parser's requirements. Extensive experiments on the STAC and Molweni datasets demonstrate that our approach effectively resolves ambiguities and significantly outperforms the state-of-the-art (SOTA) baselines.
comment: Accepted by ACL2025(main conference)
☆ Sequential Policy Gradient for Adaptive Hyperparameter Optimization
Reinforcement learning is essential for neural architecture search and hyperparameter optimization, but the conventional approaches impede widespread use due to prohibitive time and computational costs. Inspired by DeepSeek-V3 multi-token prediction architecture, we propose Sequential Policy Gradient modeling (SPG), a novel trajectory generation paradigm for lightweight online hyperparameter optimization. In contrast to conventional policy gradient methods, SPG extends the base model with temporary modules, enabling it to generate state-action (padded) trajectories in a single forward pass. Our experiments demonstrate that models gain performance when retrained with SPG on their original datasets and also outperform standard transfer fine-tuning. We evaluate on five datasets spanning computer vision (ImageNet, COCO), natural language processing (GLUE, SQuAD), and audio (SUPERB) to assess the industrial applicability of SPG. The proposed method demonstrates consistent improvements across widely adopted models, achieving performance gains of $+0.2\sim7\%$, with significantly low computational costs. Fully reproducible code and pre-trained models: https://huggingface.co/UniversalAlgorithmic/SPG.
comment: 10 pages, 2 figures
☆ Truncated Proximal Policy Optimization
Recently, test-time scaling Large Language Models (LLMs) have demonstrated exceptional reasoning capabilities across scientific and professional tasks by generating long chains-of-thought (CoT). As a crucial component for developing these reasoning models, reinforcement learning (RL), exemplified by Proximal Policy Optimization (PPO) and its variants, allows models to learn through trial and error. However, PPO can be time-consuming due to its inherent on-policy nature, which is further exacerbated by increasing response lengths. In this work, we propose Truncated Proximal Policy Optimization (T-PPO), a novel extension to PPO that improves training efficiency by streamlining policy update and length-restricted response generation. T-PPO mitigates the issue of low hardware utilization, an inherent drawback of fully synchronized long-generation procedures, where resources often sit idle during the waiting periods for complete rollouts. Our contributions are two-folds. First, we propose Extended Generalized Advantage Estimation (EGAE) for advantage estimation derived from incomplete responses while maintaining the integrity of policy learning. Second, we devise a computationally optimized mechanism that allows for the independent optimization of the policy and value models. By selectively filtering prompt and truncated tokens, this mechanism reduces redundant computations and accelerates the training process without sacrificing convergence performance. We demonstrate the effectiveness and efficacy of T-PPO on AIME 2024 with a 32B base model. The experimental results show that T-PPO improves the training efficiency of reasoning LLMs by up to 2.5x and outperforms its existing competitors.
☆ Mapping Caregiver Needs to AI Chatbot Design: Strengths and Gaps in Mental Health Support for Alzheimer's and Dementia Caregivers
Family caregivers of individuals with Alzheimer's Disease and Related Dementia (AD/ADRD) face significant emotional and logistical challenges that place them at heightened risk for stress, anxiety, and depression. Although recent advances in generative AI -- particularly large language models (LLMs) -- offer new opportunities to support mental health, little is known about how caregivers perceive and engage with such technologies. To address this gap, we developed Carey, a GPT-4o-based chatbot designed to provide informational and emotional support to AD/ADRD caregivers. Using Carey as a technology probe, we conducted semi-structured interviews with 16 family caregivers following scenario-driven interactions grounded in common caregiving stressors. Through inductive coding and reflexive thematic analysis, we surface a systemic understanding of caregiver needs and expectations across six themes -- on-demand information access, emotional support, safe space for disclosure, crisis management, personalization, and data privacy. For each of these themes, we also identified the nuanced tensions in the caregivers' desires and concerns. We present a mapping of caregiver needs, AI chatbot's strengths, gaps, and design recommendations. Our findings offer theoretical and practical insights to inform the design of proactive, trustworthy, and caregiver-centered AI systems that better support the evolving mental health needs of AD/ADRD caregivers.
♻ ☆ Generalized Out-of-Distribution Detection and Beyond in Vision Language Model Era: A Survey
Detecting out-of-distribution (OOD) samples is crucial for ensuring the safety of machine learning systems and has shaped the field of OOD detection. Meanwhile, several other problems are closely related to OOD detection, including anomaly detection (AD), novelty detection (ND), open set recognition (OSR), and outlier detection (OD). To unify these problems, a generalized OOD detection framework was proposed, taxonomically categorizing these five problems. However, Vision Language Models (VLMs) such as CLIP have significantly changed the paradigm and blurred the boundaries between these fields, again confusing researchers. In this survey, we first present a generalized OOD detection v2, encapsulating the evolution of these fields in the VLM era. Our framework reveals that, with some field inactivity and integration, the demanding challenges have become OOD detection and AD. Then, we highlight the significant shift in the definition, problem settings, and benchmarks; we thus feature a comprehensive review of the methodology for OOD detection and related tasks to clarify their relationship to OOD detection. Finally, we explore the advancements in the emerging Large Vision Language Model (LVLM) era, such as GPT-4V. We conclude with open challenges and future directions. The resource is available at https://github.com/AtsuMiyai/Awesome-OOD-VLM.
comment: Accepted at TMLR2025. Survey paper. We welcome questions, issues, and paper requests via https://github.com/AtsuMiyai/Awesome-OOD-VLM
♻ ☆ J4R: Learning to Judge with Equivalent Initial State Group Relative Policy Optimization
To keep pace with the increasing pace of large language models (LLM) development, model output evaluation has transitioned away from time-consuming human evaluation to automatic evaluation, where LLMs themselves are tasked with assessing and critiquing other model outputs. LLM-as-judge models are a class of generative evaluators that excel in evaluating relatively simple domains, like chat quality, but struggle in reasoning intensive domains where model responses contain more substantive and challenging content. To remedy existing judge shortcomings, we explore training judges with reinforcement learning (RL). We make three key contributions: (1) We propose the Equivalent Initial State Group Relative Policy Optimization (EIS-GRPO) algorithm, which allows us to train our judge to be robust to positional biases that arise in more complex evaluation settings. (2) We introduce ReasoningJudgeBench, a benchmark that evaluates judges in diverse reasoning settings not covered by prior work. (3) We train Judge for Reasoning (J4R), a 7B judge trained with EIS-GRPO that outperforms GPT-4o and the next best small judge by 6.7% and 9%, matching or exceeding the performance of larger GRPO-trained judges on both JudgeBench and ReasoningJudgeBench.
comment: 25 pages, 4 figures, 6 tables. Updated with code and benchmark
♻ ☆ Router-R1: Teaching LLMs Multi-Round Routing and Aggregation via Reinforcement Learning
The rapid emergence of diverse large language models (LLMs) has spurred the development of LLM routers that assign user queries to the most suitable model. However, existing LLM routers typically perform a single-round, one-to-one mapping (\textit{i.e.}, assigning each query to a single model in isolation), which limits their capability to tackle complex tasks that demand the complementary strengths of multiple LLMs. In this paper, we present \textbf{Router-R1}, a reinforcement learning (RL)-based framework that formulates multi-LLM routing and aggregation as a sequential decision process. Router-R1 instantiates the router itself as a capable LLM, leveraging its reasoning ability to interleave "think" actions (internal deliberation) with "route" actions (dynamic model invocation), and integrates each response into its evolving context. To facilitate learning, we employ a lightweight rule-based reward comprising format rewards, final outcome rewards, and a novel cost reward for optimizing the balance between performance and cost, opening a pathway toward enhancing performance-cost trade-offs via RL. Router-R1 also conditions only on simple model descriptors such as pricing, latency, and example performance, enabling strong generalization to unseen model selection. Experiments on seven general and multi-hop QA benchmarks show that Router-R1 outperforms several strong baselines, achieving superior performance while maintaining robust generalization and cost management.
comment: Code is available at https://github.com/ulab-uiuc/Router-R1. Models and Datasets are available at https://huggingface.co/collections/ulab-ai/router-r1-6851bbe099c7a56914b5db03
♻ ☆ A Novel Perturb-ability Score to Mitigate Evasion Adversarial Attacks on Flow-Based ML-NIDS
As network security threats evolve, safeguarding flow-based Machine Learning (ML)-based Network Intrusion Detection Systems (NIDS) from evasion adversarial attacks is crucial. This paper introduces the notion of feature perturb-ability and presents a novel Perturb-ability Score (PS), which quantifies how susceptible NIDS features are to manipulation in the problem-space by an attacker. PS thereby identifies features structurally resistant to evasion attacks in flow-based ML-NIDS due to the semantics of network traffic fields, as these features are constrained by domain-specific limitations and correlations. Consequently, attempts to manipulate such features would likely either compromise the attack's malicious functionality, render the traffic invalid for processing, or potentially both outcomes simultaneously. We introduce and demonstrate the effectiveness of our PS-enabled defenses, PS-guided feature selection and PS-guided feature masking, in enhancing flow-based NIDS resilience. Experimental results across various ML-based NIDS models and public datasets show that discarding or masking highly manipulatable features (high-PS features) can maintain solid detection performance while significantly reducing vulnerability to evasion adversarial attacks. Our findings confirm that PS effectively identifies flow-based NIDS features susceptible to problem-space perturbations. This novel approach leverages problem-space NIDS domain constraints as lightweight universal defense mechanisms against evasion adversarial attacks targeting flow-based ML-NIDS.
♻ ☆ Refactoring Codebases through Library Design
Maintainable and general software allows developers to build robust applications efficiently, yet achieving these qualities often requires refactoring specialized solutions into reusable components. This challenge becomes particularly relevant as code agents become increasingly accurate at solving isolated programming problems. We investigate code agents' capacity to refactor code in ways supporting growth and reusability. We present both a method and a benchmark for refactoring: Librarian, a sample-and-rerank method for generating reusable libraries, and Minicode, a benchmark where code agents must minimize and refactor multiple independent solutions into a joint library. Compared to state-of-the-art code agents, Librarian achieves strong results on both compression and correctness on Minicode, obtaining compression rates 1.6-2x better than coding agents while also improving correctness. We open-source our code and benchmark at https://code-refactor.github.io/.
comment: 29 pages
♻ ☆ Vision Transformers Don't Need Trained Registers
We investigate the mechanism underlying a previously identified phenomenon in Vision Transformers -- the emergence of high-norm tokens that lead to noisy attention maps. We observe that in multiple models (e.g., CLIP, DINOv2), a sparse set of neurons is responsible for concentrating high-norm activations on outlier tokens, leading to irregular attention patterns and degrading downstream visual processing. While the existing solution for removing these outliers involves retraining models from scratch with additional learned register tokens, we use our findings to create a training-free approach to mitigate these artifacts. By shifting the high-norm activations from our discovered register neurons into an additional untrained token, we can mimic the effect of register tokens on a model already trained without registers. We demonstrate that our method produces cleaner attention and feature maps, enhances performance over base models across multiple downstream visual tasks, and achieves results comparable to models explicitly trained with register tokens. We then extend test-time registers to off-the-shelf vision-language models to improve their interpretability. Our results suggest that test-time registers effectively take on the role of register tokens at test-time, offering a training-free solution for any pre-trained model released without them.
comment: Project page and code: https://avdravid.github.io/test-time-registers
♻ ☆ "Generate" the Future of Work through AI: Empirical Evidence from Online Labor Markets
Large Language Model (LLM)-based generative AI systems, such as ChatGPT, demonstrate zero-shot learning capabilities across a wide range of downstream tasks. Owing to their general-purpose nature and potential to augment or even automate job functions, these systems are poised to reshape labor market dynamics. However, predicting their precise impact \textit{a priori} is challenging, given AI's simultaneous effects on both demand and supply, as well as the strategic responses of market participants. Leveraging an extensive dataset from a leading online labor platform, we document a pronounced displacement effect and an overall contraction in submarkets where required skills closely align with core LLM functionalities. Although demand and supply both decline, the reduction in supply is comparatively smaller, thereby intensifying competition among freelancers. Notably, further analysis shows that this heightened competition is especially pronounced in programming-intensive submarkets. This pattern is attributed to skill-transition effects: by lowering the human-capital barrier to programming, ChatGPT enables incumbent freelancers to enter programming tasks. Moreover, these transitions are not homogeneous, with high-skilled freelancers contributing disproportionately to the shift. Our findings illuminate the multifaceted impacts of general-purpose AI on labor markets, highlighting not only the displacement of certain occupations but also the inducement of skill transitions within the labor supply. These insights offer practical implications for policymakers, platform operators, and workers.
comment: 92 pages, 16 figures, 34 tables
♻ ☆ TARDIS STRIDE: A Spatio-Temporal Road Image Dataset and World Model for Autonomy
World models aim to simulate environments and enable effective agent behavior. However, modeling real-world environments presents unique challenges as they dynamically change across both space and, crucially, time. To capture these composed dynamics, we introduce a Spatio-Temporal Road Image Dataset for Exploration (STRIDE) permuting 360-degree panoramic imagery into rich interconnected observation, state and action nodes. Leveraging this structure, we can simultaneously model the relationship between egocentric views, positional coordinates, and movement commands across both space and time. We benchmark this dataset via TARDIS, a transformer-based generative world model that integrates spatial and temporal dynamics through a unified autoregressive framework trained on STRIDE. We demonstrate robust performance across a range of agentic tasks such as controllable photorealistic image synthesis, instruction following, autonomous self-control, and state-of-the-art georeferencing. These results suggest a promising direction towards sophisticated generalist agents--capable of understanding and manipulating the spatial and temporal aspects of their material environments--with enhanced embodied reasoning capabilities. Training code, datasets, and model checkpoints are made available at https://huggingface.co/datasets/Tera-AI/STRIDE.
comment: Computer Vision, Pattern Recognition, Early-Fusion, Dataset, Data Augmentation
♻ ☆ Fractured Chain-of-Thought Reasoning
Inference-time scaling techniques have significantly bolstered the reasoning capabilities of large language models (LLMs) by harnessing additional computational effort at inference without retraining. Similarly, Chain-of-Thought (CoT) prompting and its extension, Long CoT, improve accuracy by generating rich intermediate reasoning trajectories, but these approaches incur substantial token costs that impede their deployment in latency-sensitive settings. In this work, we first show that truncated CoT, which stops reasoning before completion and directly generates the final answer, often matches full CoT sampling while using dramatically fewer tokens. Building on this insight, we introduce Fractured Sampling, a unified inference-time strategy that interpolates between full CoT and solution-only sampling along three orthogonal axes: (1) the number of reasoning trajectories, (2) the number of final solutions per trajectory, and (3) the depth at which reasoning traces are truncated. Through extensive experiments on five diverse reasoning benchmarks and several model scales, we demonstrate that Fractured Sampling consistently achieves superior accuracy-cost trade-offs, yielding steep log-linear scaling gains in Pass@k versus token budget. Our analysis reveals how to allocate computation across these dimensions to maximize performance, paving the way for more efficient and scalable LLM reasoning. Code is available at https://github.com/BaohaoLiao/frac-cot.
♻ ☆ Exploring Personalized Federated Learning Architectures for Violence Detection in Surveillance Videos
The challenge of detecting violent incidents in urban surveillance systems is compounded by the voluminous and diverse nature of video data. This paper presents a targeted approach using Personalized Federated Learning (PFL) to address these issues, specifically employing the Federated Learning with Personalization Layers method within the Flower framework. Our methodology adapts learning models to the unique data characteristics of each surveillance node, effectively managing the heterogeneous and non-IID nature of surveillance video data. Through rigorous experiments conducted on balanced and imbalanced datasets, our PFL models demonstrated enhanced accuracy and efficiency, achieving up to 99.3% accuracy. This study underscores the potential of PFL to significantly improve the scalability and effectiveness of surveillance systems, offering a robust, privacy-preserving solution for violence detection in complex urban environments.
comment: 7 pages, 5 figures, 4 tables
♻ ☆ Pap2Pat: Benchmarking Outline-Guided Long-Text Patent Generation with Patent-Paper Pairs ACL 2025
Dealing with long and highly complex technical text is a challenge for Large Language Models (LLMs), which still have to unfold their potential in supporting expensive and timeintensive processes like patent drafting. Within patents, the description constitutes more than 90% of the document on average. Yet, its automatic generation remains understudied. When drafting patent applications, patent attorneys typically receive invention reports (IRs), which are usually confidential, hindering research on LLM-supported patent drafting. Often, prepublication research papers serve as IRs. We leverage this duality to build PAP2PAT, an open and realistic benchmark for patent drafting consisting of 1.8k patent-paper pairs describing the same inventions. To address the complex longdocument patent generation task, we propose chunk-based outline-guided generation using the research paper as invention specification. Our extensive evaluation using PAP2PAT and a human case study show that LLMs can effectively leverage information from the paper, but still struggle to provide the necessary level of detail. Fine-tuning leads to more patent-style language, but also to more hallucination. We release our data and code https://github.com/boschresearch/Pap2Pat.
comment: ACL 2025 Findings
♻ ☆ EgoBlind: Towards Egocentric Visual Assistance for the Blind
We present EgoBlind, the first egocentric VideoQA dataset collected from blind individuals to evaluate the assistive capabilities of contemporary multimodal large language models (MLLMs). EgoBlind comprises 1,392 videos that record the daily lives of real blind users from a first-person perspective. It also features 5,311 questions directly posed or generated and verified by blind individuals to reflect their in-situation needs for visual assistance under various scenarios. We provide each question with an average of 3 reference answers to alleviate subjective evaluation. Using EgoBlind, we comprehensively evaluate 16 advanced MLLMs and find that all models struggle, with the best performers achieving accuracy near 60\%, far behind human performance of 87.4\%. To guide future advancements, we identify and summarize major limitations of existing MLLMs in egocentric visual assistance for the blind and explore heuristic solutions for improvement. With these efforts, we hope EgoBlind can serve as a valuable foundation for developing more effective AI assistants to enhance the independence of the blind individuals' lives. Data and evaluation code are available at https://github.com/doc-doc/EgoBlind.
comment: We extend and resplit the dataset
♻ ☆ RadioRAG: Online Retrieval-augmented Generation for Radiology Question Answering
Large language models (LLMs) often generate outdated or inaccurate information based on static training datasets. Retrieval-augmented generation (RAG) mitigates this by integrating outside data sources. While previous RAG systems used pre-assembled, fixed databases with limited flexibility, we have developed Radiology RAG (RadioRAG), an end-to-end framework that retrieves data from authoritative radiologic online sources in real-time. We evaluate the diagnostic accuracy of various LLMs when answering radiology-specific questions with and without access to additional online information via RAG. Using 80 questions from the RSNA Case Collection across radiologic subspecialties and 24 additional expert-curated questions with reference standard answers, LLMs (GPT-3.5-turbo, GPT-4, Mistral-7B, Mixtral-8x7B, and Llama3 [8B and 70B]) were prompted with and without RadioRAG in a zero-shot inference scenario RadioRAG retrieved context-specific information from Radiopaedia in real-time. Accuracy was investigated. Statistical analyses were performed using bootstrapping. The results were further compared with human performance. RadioRAG improved diagnostic accuracy across most LLMs, with relative accuracy increases ranging up to 54% for different LLMs. It matched or exceeded non-RAG models and the human radiologist in question answering across radiologic subspecialties, particularly in breast imaging and emergency radiology. However, the degree of improvement varied among models; GPT-3.5-turbo and Mixtral-8x7B-instruct-v0.1 saw notable gains, while Mistral-7B-instruct-v0.2 showed no improvement, highlighting variability in RadioRAG's effectiveness. LLMs benefit when provided access to domain-specific data beyond their training data. RadioRAG shows potential to improve LLM accuracy and factuality in radiology question answering by integrating real-time domain-specific data.
comment: Published in Radiology: Artificial Intelligence
♻ ☆ M3-JEPA: Multimodal Alignment via Multi-gate MoE based on the Joint-Embedding Predictive Architecture ICML 2025
Current multimodal learning strategies primarily optimize in the original token space. Such a framework is easy to incorporate with the backbone of pretrained language model, but might result in modality collapse. To alleviate such issues, we leverage the Joint-Embedding Predictive Architecture (JEPA) on the multimodal tasks, which converts the input embedding into the output embedding space by a predictor and then conducts the cross-modal alignment on the latent space. We implement this predictor by a Multi-Gate Mixture of Experts (MMoE) and name the framework as M3-JEPA, accordingly. The gating function disentangles the modality-specific and shared information and derives information-theoretic optimality. The framework is implemented with both contrastive and regularization loss, and solved by alternative gradient descent (AGD) between different multimodal tasks. By thoroughly designed experiments, we show that M3-JEPA can obtain state-of-the-art performance on different modalities and tasks, generalize to unseen datasets and domains, and is computationally efficient in both training and inference. Our observation suggests that M3-JEPA might become a new basis to self-supervised learning in the open world.
comment: 16 pages, 5 figures. ICML 2025
♻ ☆ KANITE: Kolmogorov-Arnold Networks for ITE estimation
We introduce KANITE, a framework leveraging Kolmogorov-Arnold Networks (KANs) for Individual Treatment Effect (ITE) estimation under multiple treatments setting in causal inference. By utilizing KAN's unique abilities to learn univariate activation functions as opposed to learning linear weights by Multi-Layer Perceptrons (MLPs), we improve the estimates of ITEs. The KANITE framework comprises two key architectures: 1.Integral Probability Metric (IPM) architecture: This employs an IPM loss in a specialized manner to effectively align towards ITE estimation across multiple treatments. 2. Entropy Balancing (EB) architecture: This uses weights for samples that are learned by optimizing entropy subject to balancing the covariates across treatment groups. Extensive evaluations on benchmark datasets demonstrate that KANITE outperforms state-of-the-art algorithms in both $\epsilon_{\text{PEHE}}$ and $\epsilon_{\text{ATE}}$ metrics. Our experiments highlight the advantages of KANITE in achieving improved causal estimates, emphasizing the potential of KANs to advance causal inference methodologies across diverse application areas.
comment: 16 pages, 4 figures
♻ ☆ Adding Chocolate to Mint: Mitigating Metric Interference in Machine Translation
As automatic metrics become increasingly stronger and widely adopted, the risk of unintentionally "gaming the metric" during model development rises. This issue is caused by metric interference (MINT), i.e., the use of the same or related metrics for both model tuning and evaluation. MINT can misguide practitioners into being overoptimistic about the performance of their systems: as system outputs become a function of the interfering metric, their estimated quality loses correlation with human judgments. In this work, we analyze two common cases of MINT in machine translation-related tasks: filtering of training data, and decoding with quality signals. Importantly, we find that MINT strongly distorts instance-level metric scores, even when metrics are not directly optimized for-questioning the common strategy of leveraging a different, yet related metric for evaluation that is not used for tuning. To address this problem, we propose MINTADJUST, a method for more reliable evaluation under MINT. On the WMT24 MT shared task test set, MINTADJUST ranks translations and systems more accurately than state-of-the-art metrics across a majority of language pairs, especially for high-quality systems. Furthermore, MINTADJUST outperforms AUTORANK, the ensembling method used by the organizers.
♻ ☆ Position Paper: Rethinking Privacy in RL for Sequential Decision-making in the Age of LLMs IJCNN 2025
The rise of reinforcement learning (RL) in critical real-world applications demands a fundamental rethinking of privacy in AI systems. Traditional privacy frameworks, designed to protect isolated data points, fall short for sequential decision-making systems where sensitive information emerges from temporal patterns, behavioral strategies, and collaborative dynamics. Modern RL paradigms, such as federated RL (FedRL) and RL with human feedback (RLHF) in large language models (LLMs), exacerbate these challenges by introducing complex, interactive, and context-dependent learning environments that traditional methods do not address. In this position paper, we argue for a new privacy paradigm built on four core principles: multi-scale protection, behavioral pattern protection, collaborative privacy preservation, and context-aware adaptation. These principles expose inherent tensions between privacy, utility, and interpretability that must be navigated as RL systems become more pervasive in high-stakes domains like healthcare, autonomous vehicles, and decision support systems powered by LLMs. To tackle these challenges, we call for the development of new theoretical frameworks, practical mechanisms, and rigorous evaluation methodologies that collectively enable effective privacy protection in sequential decision-making systems.
comment: IJCNN 2025 Position Paper Track
♻ ☆ No-Regret Learning Under Adversarial Resource Constraints: A Spending Plan Is All You Need!
We study online decision making problems under resource constraints, where both reward and cost functions are drawn from distributions that may change adversarially over time. We focus on two canonical settings: $(i)$ online resource allocation where rewards and costs are observed before action selection, and $(ii)$ online learning with resource constraints where they are observed after action selection, under full feedback or bandit feedback. It is well known that achieving sublinear regret in these settings is impossible when reward and cost distributions may change arbitrarily over time. To address this challenge, we analyze a framework in which the learner is guided by a spending plan--a sequence prescribing expected resource usage across rounds. We design general (primal-)dual methods that achieve sublinear regret with respect to baselines that follow the spending plan. Crucially, the performance of our algorithms improves when the spending plan ensures a well-balanced distribution of the budget across rounds. We additionally provide a robust variant of our methods to handle worst-case scenarios where the spending plan is highly imbalanced. To conclude, we study the regret of our algorithms when competing against benchmarks that deviate from the prescribed spending plan.
♻ ☆ Breaking Bad Molecules: Are MLLMs Ready for Structure-Level Molecular Detoxification?
Toxicity remains a leading cause of early-stage drug development failure. Despite advances in molecular design and property prediction, the task of molecular toxicity repair - generating structurally valid molecular alternatives with reduced toxicity - has not yet been systematically defined or benchmarked. To fill this gap, we introduce ToxiMol, the first benchmark task for general-purpose Multimodal Large Language Models (MLLMs) focused on molecular toxicity repair. We construct a standardized dataset covering 11 primary tasks and 560 representative toxic molecules spanning diverse mechanisms and granularities. We design a prompt annotation pipeline with mechanism-aware and task-adaptive capabilities, informed by expert toxicological knowledge. In parallel, we propose an automated evaluation framework, ToxiEval, which integrates toxicity endpoint prediction, synthetic accessibility, drug-likeness, and structural similarity into a high-throughput evaluation chain for repair success. We systematically assess nearly 30 mainstream general-purpose MLLMs and design multiple ablation studies to analyze key factors such as evaluation criteria, candidate diversity, and failure attribution. Experimental results show that although current MLLMs still face significant challenges on this task, they begin to demonstrate promising capabilities in toxicity understanding, semantic constraint adherence, and structure-aware molecule editing.
♻ ☆ Deep Graph Anomaly Detection: A Survey and New Perspectives
Graph anomaly detection (GAD), which aims to identify unusual graph instances (nodes, edges, subgraphs, or graphs), has attracted increasing attention in recent years due to its significance in a wide range of applications. Deep learning approaches, graph neural networks (GNNs) in particular, have been emerging as a promising paradigm for GAD, owing to its strong capability in capturing complex structure and/or node attributes in graph data. Considering the large number of methods proposed for GNN-based GAD, it is of paramount importance to summarize the methodologies and findings in the existing GAD studies, so that we can pinpoint effective model designs for tackling open GAD problems. To this end, in this work we aim to present a comprehensive review of deep learning approaches for GAD. Existing GAD surveys are focused on task-specific discussions, making it difficult to understand the technical insights of existing methods and their limitations in addressing some unique challenges in GAD. To fill this gap, we first discuss the problem complexities and their resulting challenges in GAD, and then provide a systematic review of current deep GAD methods from three novel perspectives of methodology, including GNN backbone design, proxy task design for GAD, and graph anomaly measures. To deepen the discussions, we further propose a taxonomy of 13 fine-grained method categories under these three perspectives to provide more in-depth insights into the model designs and their capabilities. To facilitate the experiments and validation, we also summarize a collection of widely-used GAD datasets and empirical comparison. We further discuss multiple open problems to inspire more future high-quality research. A continuously updated repository for datasets, links to the codes of algorithms, and empirical comparison is available at https://github.com/mala-lab/Awesome-Deep-Graph-Anomaly-Detection.
comment: Accepted by TKDE
♻ ☆ OM4OV: Leveraging Ontology Matching for Ontology Versioning
Due to the dynamic nature of the Semantic Web, version control is necessary to capture time-varying information, particularly for widely used ontologies. Despite the long-standing recognition of ontology versioning (OV) as a crucial component for efficient ontology management, the growing size of ontologies and accumulating errors caused by manual labour overwhelm current OV approaches. In this paper, we propose yet another approach to performing OV using existing ontology matching (OM) techniques and systems. We introduce a unified OM4OV pipeline. From an OM perspective, we reconstruct a new task formulation and measurement for OV tasks. Building upon the prior alignment(s) from OM, we propose a pipeline optimisation method called the cross-reference (CR) mechanism to enhance overall OV performance. We experimentally validate the OM4OV pipeline and the cross-reference mechanism in the OV tested originating from the Ontology Alignment Evaluation Initiative (OAEI) datasets. We also discuss insights into OM used for OV tasks, where some false mappings detected by OV systems are not actually untrue.
comment: 15 pages, 8 figures, 1 table
♻ ☆ Probabilistic Aggregation and Targeted Embedding Optimization for Collective Moral Reasoning in Large Language Models ACL 2025
Large Language Models (LLMs) have shown impressive moral reasoning abilities. Yet they often diverge when confronted with complex, multi-factor moral dilemmas. To address these discrepancies, we propose a framework that synthesizes multiple LLMs' moral judgments into a collectively formulated moral judgment, realigning models that deviate significantly from this consensus. Our aggregation mechanism fuses continuous moral acceptability scores (beyond binary labels) into a collective probability, weighting contributions by model reliability. For misaligned models, a targeted embedding-optimization procedure fine-tunes token embeddings for moral philosophical theories, minimizing JS divergence to the consensus while preserving semantic integrity. Experiments on a large-scale social moral dilemma dataset show our approach builds robust consensus and improves individual model fidelity. These findings highlight the value of data-driven moral alignment across multiple models and its potential for safer, more consistent AI systems.
comment: Accepted to ACL 2025 (Findings)
♻ ☆ Semantic-Geometric-Physical-Driven Robot Manipulation Skill Transfer via Skill Library and Tactile Representation
Developing general robotic systems capable of manipulating in unstructured environments is a significant challenge, particularly as the tasks involved are typically long-horizon and rich-contact, requiring efficient skill transfer across different task scenarios. To address these challenges, we propose knowledge graph-based skill library construction method. This method hierarchically organizes manipulation knowledge using "task graph" and "scene graph" to represent task-specific and scene-specific information, respectively. Additionally, we introduce "state graph" to facilitate the interaction between high-level task planning and low-level scene information. Building upon this foundation, we further propose a novel hierarchical skill transfer framework based on the skill library and tactile representation, which integrates high-level reasoning for skill transfer and low-level precision for execution. At the task level, we utilize large language models (LLMs) and combine contextual learning with a four-stage chain-of-thought prompting paradigm to achieve subtask sequence transfer. At the motion level, we develop an adaptive trajectory transfer method based on the skill library and the heuristic path planning algorithm. At the physical level, we propose an adaptive contour extraction and posture perception method based on tactile representation. This method dynamically acquires high-precision contour and posture information from visual-tactile images, adjusting parameters such as contact position and posture to ensure the effectiveness of transferred skills in new environments. Experiments demonstrate the skill transfer and adaptability capabilities of the proposed methods across different task scenarios. Project website: https://github.com/MingchaoQi/skill_transfer
♻ ☆ FrontendBench: A Benchmark for Evaluating LLMs on Front-End Development via Automatic Evaluation
Large Language Models (LLMs) have made significant strides in front-end code generation. However, existing benchmarks exhibit several critical limitations: many tasks are overly simplistic, test cases often lack rigor, and end-to-end validation is absent. These issues hinder the accurate assessment of model performance. To address these challenges, we present FrontendBench, a benchmark co-developed by humans and LLMs. FrontendBench categorizes tasks based on code functionality and incorporates interactive test scenarios, enabling a more comprehensive and practical evaluation of front-end code generation capabilities. The benchmark comprises 148 meticulously crafted prompt-test case pairs spanning five levels of web components, from basic UI elements to complex interactive features. Each task reflects realistic front-end development challenges. Furthermore, we introduce an automatic evaluation framework that executes generated code within a sandbox environment and assesses outcomes using predefined test scripts. This framework achieves a 90.54% agreement rate with expert human evaluations, demonstrating high reliability. We benchmark several state-of-the-art LLMs on FrontendBench and observe substantial performance disparities in handling real-world front-end tasks. These results highlight FrontendBench as a reliable and scalable benchmark, supporting consistent multimodal evaluation and providing a robust foundation for future research in front-end code generation. Our data and code will be released soon.
♻ ☆ Machine Learners Should Acknowledge the Legal Implications of Large Language Models as Personal Data
Does GPT know you? The answer depends on your level of public recognition; however, if your information was available on a website, the answer could be yes. Most Large Language Models (LLMs) memorize training data to some extent. Thus, even when an LLM memorizes only a small amount of personal data, it typically falls within the scope of data protection laws. If a person is identified or identifiable, the implications are far-reaching. The LLM is subject to EU General Data Protection Regulation requirements even after the training phase is concluded. To back our arguments: (1.) We reiterate that LLMs output training data at inference time, be it verbatim or in generalized form. (2.) We show that some LLMs can thus be considered personal data on their own. This triggers a cascade of data protection implications such as data subject rights, including rights to access, rectification, or erasure. These rights extend to the information embedded within the AI model. (3.) This paper argues that machine learning researchers must acknowledge the legal implications of LLMs as personal data throughout the full ML development lifecycle, from data collection and curation to model provision on e.g., GitHub or Hugging Face. (4.) We propose different ways for the ML research community to deal with these legal implications. Our paper serves as a starting point for improving the alignment between data protection law and the technical capabilities of LLMs. Our findings underscore the need for more interaction between the legal domain and the ML community.
♻ ☆ Heterogeneous Relationships of Subjects and Shapelets for Semi-supervised Multivariate Series Classification
Multivariate time series (MTS) classification is widely applied in fields such as industry, healthcare, and finance, aiming to extract key features from complex time series data for accurate decision-making and prediction. However, existing methods for MTS often struggle due to the challenges of effectively modeling high-dimensional data and the lack of labeled data, resulting in poor classification performance. To address this issue, we propose a heterogeneous relationships of subjects and shapelets method for semi-supervised MTS classification. This method offers a novel perspective by integrating various types of additional information while capturing the relationships between them. Specifically, we first utilize a contrast temporal self-attention module to obtain sparse MTS representations, and then model the similarities between these representations using soft dynamic time warping to construct a similarity graph. Secondly, we learn the shapelets for different subject types, incorporating both the subject features and their shapelets as additional information to further refine the similarity graph, ultimately generating a heterogeneous graph. Finally, we use a dual level graph attention network to get prediction. Through this method, we successfully transform dataset into a heterogeneous graph, integrating multiple additional information and achieving precise semi-supervised node classification. Experiments on the Human Activity Recognition, sleep stage classification and University of East Anglia datasets demonstrate that our method outperforms current state-of-the-art methods in MTS classification tasks, validating its superiority.
comment: We would like to request the withdrawal of our manuscript due to logical errors in the paper
♻ ☆ Contrast Similarity-Aware Dual-Pathway Mamba for Multivariate Time Series Node Classification
Multivariate time series (MTS) data is generated through multiple sensors across various domains such as engineering application, health monitoring, and the internet of things, characterized by its temporal changes and high dimensional characteristics. Over the past few years, many studies have explored the long-range dependencies and similarities in MTS. However, long-range dependencies are difficult to model due to their temporal changes and high dimensionality makes it difficult to obtain similarities effectively and efficiently. Thus, to address these issues, we propose contrast similarity-aware dual-pathway Mamba for MTS node classification (CS-DPMamba). Firstly, to obtain the dynamic similarity of each sample, we initially use temporal contrast learning module to acquire MTS representations. And then we construct a similarity matrix between MTS representations using Fast Dynamic Time Warping (FastDTW). Secondly, we apply the DPMamba to consider the bidirectional nature of MTS, allowing us to better capture long-range and short-range dependencies within the data. Finally, we utilize the Kolmogorov-Arnold Network enhanced Graph Isomorphism Network to complete the information interaction in the matrix and MTS node classification task. By comprehensively considering the long-range dependencies and dynamic similarity features, we achieved precise MTS node classification. We conducted experiments on multiple University of East Anglia (UEA) MTS datasets, which encompass diverse application scenarios. Our results demonstrate the superiority of our method through both supervised and semi-supervised experiments on the MTS classification task.
comment: We would like to request the withdrawal of our manuscript due to logical errors in the paper
♻ ☆ A Bird Song Detector for improving bird identification through Deep Learning: a case study from Doñana
Passive Acoustic Monitoring is a key tool for biodiversity conservation, but the large volumes of unsupervised audio it generates present major challenges for extracting meaningful information. Deep Learning offers promising solutions. BirdNET, a widely used bird identification model, has shown success in many study systems but is limited at local scale due to biases in its training data, which focus on specific locations and target sounds rather than entire soundscapes. A key challenge in bird species identification is that many recordings either lack target species or contain overlapping vocalizations, complicating automatic identification. To address these problems, we developed a multi-stage pipeline for automatic bird vocalization identification in Do\~nana National Park (SW Spain), a wetland of high conservation concern. We deployed AudioMoth recorders in three main habitats across nine locations and manually annotated 461 minutes of audio, resulting in 3749 labeled segments spanning 34 classes. We first applied a Bird Song Detector to isolate bird vocalizations using spectrogram-based image processing. Then, species were classified using custom models trained at the local scale. Applying the Bird Song Detector before classification improved species identification, as all models performed better when analyzing only the segments where birds were detected. Specifically, the combination of detector and fine-tuned BirdNET outperformed the baseline without detection. This approach demonstrates the effectiveness of integrating a Bird Song Detector with local classification models. These findings highlight the need to adapt general-purpose tools to specific ecological challenges. Automatically detecting bird species helps track the health of this threatened ecosystem, given birds sensitivity to environmental change, and supports conservation planning to reduce biodiversity loss.
comment: 23 pages, 14 images, for associated dataset see https://huggingface.co/datasets/GrunCrow/BIRDeep_AudioAnnotations , for associated code see https://github.com/GrunCrow/BIRDeep_BirdSongDetector_NeuralNetworks and https://github.com/GrunCrow/Bird-Song-Detector
♻ ☆ PsychBench: A comprehensive and professional benchmark for evaluating the performance of LLM-assisted psychiatric clinical practice
The advent of Large Language Models (LLMs) offers potential solutions to address problems such as shortage of medical resources and low diagnostic consistency in psychiatric clinical practice. Despite this potential, a robust and comprehensive benchmarking framework to assess the efficacy of LLMs in authentic psychiatric clinical environments is absent. This has impeded the advancement of specialized LLMs tailored to psychiatric applications. In response to this gap, by incorporating clinical demands in psychiatry and clinical data, we proposed a benchmarking system, PsychBench, to evaluate the practical performance of LLMs in psychiatric clinical settings. We conducted a comprehensive quantitative evaluation of 16 LLMs using PsychBench, and investigated the impact of prompt design, chain-of-thought reasoning, input text length, and domain-specific knowledge fine-tuning on model performance. Through detailed error analysis, we identified strengths and potential limitations of the existing models and suggested directions for improvement. Subsequently, a clinical reader study involving 60 psychiatrists of varying seniority was conducted to further explore the practical benefits of existing LLMs as supportive tools for psychiatrists of varying seniority. Through the quantitative and reader evaluation, we show that while existing models demonstrate significant potential, they are not yet adequate as decision-making tools in psychiatric clinical practice. The reader study further indicates that, as an auxiliary tool, LLM could provide particularly notable support for junior psychiatrists, effectively enhancing their work efficiency and overall clinical quality. To promote research in this area, we will make the dataset and evaluation framework publicly available, with the hope of advancing the application of LLMs in psychiatric clinical settings.
♻ ☆ CORA: Coalitional Rational Advantage Decomposition for Multi-Agent Policy Gradients
This work focuses on the credit assignment problem in cooperative multi-agent reinforcement learning (MARL). Sharing the global advantage among agents often leads to suboptimal policy updates as it fails to account for the distinct contributions of agents. Although numerous methods consider global or individual contributions for credit assignment, a detailed analysis at the coalition level remains lacking in many approaches. This work analyzes the over-updating problem during multi-agent policy updates from a coalition-level perspective. To address this issue, we propose a credit assignment method called Coalitional Rational Advantage Decomposition (CORA). CORA evaluates coalitional advantages via marginal contributions from all possible coalitions and decomposes advantages using the core solution from cooperative game theory, ensuring coalitional rationality. To reduce computational overhead, CORA employs random coalition sampling. Experiments on matrix games, differential games, and multi-agent collaboration benchmarks demonstrate that CORA outperforms strong baselines, particularly in tasks with multiple local optima. These findings highlight the importance of coalition-aware credit assignment for improving MARL performance.
♻ ☆ Imagine Beyond! Distributionally Robust Auto-Encoding for State Space Coverage in Online Reinforcement Learning
Goal-Conditioned Reinforcement Learning (GCRL) enables agents to autonomously acquire diverse behaviors, but faces major challenges in visual environments due to high-dimensional, semantically sparse observations. In the online setting, where agents learn representations while exploring, the latent space evolves with the agent's policy, to capture newly discovered areas of the environment. However, without incentivization to maximize state coverage in the representation, classical approaches based on auto-encoders may converge to latent spaces that over-represent a restricted set of states frequently visited by the agent. This is exacerbated in an intrinsic motivation setting, where the agent uses the distribution encoded in the latent space to sample the goals it learns to master. To address this issue, we propose to progressively enforce distributional shifts towards a uniform distribution over the full state space, to ensure a full coverage of skills that can be learned in the environment. We introduce DRAG (Distributionally Robust Auto-Encoding for GCRL), a method that combines the $\beta$-VAE framework with Distributionally Robust Optimization. DRAG leverages an adversarial neural weighter of training states of the VAE, to account for the mismatch between the current data distribution and unseen parts of the environment. This allows the agent to construct semantically meaningful latent spaces beyond its immediate experience. Our approach improves state space coverage and downstream control performance on hard exploration environments such as mazes and robotic control involving walls to bypass, without pre-training nor prior environment knowledge.
♻ ☆ Behaviour Planning: A Toolkit for Diverse Planning
Diverse planning approaches are utilised in real-world applications like risk management, automated streamed data analysis, and malware detection. The current diverse planning formulations encode the diversity model as a distance function, which is computational inexpensive when comparing two plans. However, such modelling approach limits what can be encoded as measure of diversity, as well as the ability to explain why two plans are different. This paper introduces a novel approach to the diverse planning problem, allowing for more expressive modelling of diversity using a n-dimensional grid representation, where each dimension corresponds to a user-defined feature. Furthermore, we present a novel toolkit that generates diverse plans based on such customisable diversity models, called \emph{Behaviour Planning}. We provide an implementation for behaviour planning using planning-as-satisfiability. An empirical evaluation of our implementation shows that behaviour planning significantly outperforms the current diverse planning method in generating diverse plans measured on our new customisable diversity models. Our implementation is the first diverse planning approach to support planning categories beyond classical planning, such as over-subscription and numerical planning.
♻ ☆ Style-Preserving Lip Sync via Audio-Aware Style Reference
Audio-driven lip sync has recently drawn significant attention due to its widespread application in the multimedia domain. Individuals exhibit distinct lip shapes when speaking the same utterance, attributed to the unique speaking styles of individuals, posing a notable challenge for audio-driven lip sync. Earlier methods for such task often bypassed the modeling of personalized speaking styles, resulting in sub-optimal lip sync conforming to the general styles. Recent lip sync techniques attempt to guide the lip sync for arbitrary audio by aggregating information from a style reference video, yet they can not preserve the speaking styles well due to their inaccuracy in style aggregation. This work proposes an innovative audio-aware style reference scheme that effectively leverages the relationships between input audio and reference audio from style reference video to address the style-preserving audio-driven lip sync. Specifically, we first develop an advanced Transformer-based model adept at predicting lip motion corresponding to the input audio, augmented by the style information aggregated through cross-attention layers from style reference video. Afterwards, to better render the lip motion into realistic talking face video, we devise a conditional latent diffusion model, integrating lip motion through modulated convolutional layers and fusing reference facial images via spatial cross-attention layers. Extensive experiments validate the efficacy of the proposed approach in achieving precise lip sync, preserving speaking styles, and generating high-fidelity, realistic talking face videos.
comment: submitted to IEEE Transactions on Multimedia(TMM)
♻ ☆ Serving Large Language Models on Huawei CloudMatrix384
The rapid evolution of large language models (LLMs), driven by growing parameter scales, adoption of mixture-of-experts (MoE) architectures, and expanding context lengths, imposes unprecedented demands on AI infrastructure. Traditional AI clusters face limitations in compute intensity, memory bandwidth, inter-chip communication, and latency, compounded by variable workloads and strict service-level objectives. Addressing these issues requires fundamentally redesigned hardware-software integration. This paper introduces Huawei CloudMatrix, a next-generation AI datacenter architecture, realized in the production-grade CloudMatrix384 supernode. It integrates 384 Ascend 910C NPUs and 192 Kunpeng CPUs interconnected via an ultra-high-bandwidth Unified Bus (UB) network, enabling direct all-to-all communication and dynamic pooling of resources. These features optimize performance for communication-intensive operations, such as large-scale MoE expert parallelism and distributed key-value cache access. To fully leverage CloudMatrix384, we propose CloudMatrix-Infer, an advanced LLM serving solution incorporating three core innovations: a peer-to-peer serving architecture that independently scales prefill, decode, and caching; a large-scale expert parallelism strategy supporting EP320 via efficient UB-based token dispatch; and hardware-aware optimizations including specialized operators, microbatch-based pipelining, and INT8 quantization. Evaluation with the DeepSeek-R1 model shows CloudMatrix-Infer achieves state-of-the-art efficiency: prefill throughput of 6,688 tokens/s per NPU and decode throughput of 1,943 tokens/s per NPU (<50 ms TPOT). It effectively balances throughput and latency, sustaining 538 tokens/s per NPU even under stringent 15 ms latency constraints, while INT8 quantization maintains model accuracy across benchmarks.
comment: 59 pages, 24 figures
♻ ☆ SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models CVPR 2025
Typographic attacks exploit the interplay between text and visual content in multimodal foundation models, causing misclassifications when misleading text is embedded within images. However, existing datasets are limited in size and diversity, making it difficult to study such vulnerabilities. In this paper, we introduce SCAM, the largest and most diverse dataset of real-world typographic attack images to date, containing 1,162 images across hundreds of object categories and attack words. Through extensive benchmarking of Vision-Language Models (VLMs) on SCAM, we demonstrate that typographic attacks significantly degrade performance, and identify that training data and model architecture influence the susceptibility to these attacks. Our findings reveal that typographic attacks persist in state-of-the-art Large Vision-Language Models (LVLMs) due to the choice of their vision encoder, though larger Large Language Models (LLMs) backbones help mitigate their vulnerability. Additionally, we demonstrate that synthetic attacks closely resemble real-world (handwritten) attacks, validating their use in research. Our work provides a comprehensive resource and empirical insights to facilitate future research toward robust and trustworthy multimodal AI systems. We publicly release the datasets introduced in this paper along with the code for evaluations at www.bliss.berlin/research/scam.
comment: Accepted at CVPR 2025 Workshop EVAL-FoMo-2
♻ ☆ AIn't Nothing But a Survey? Using Large Language Models for Coding German Open-Ended Survey Responses on Survey Motivation
The recent development and wider accessibility of LLMs have spurred discussions about how they can be used in survey research, including classifying open-ended survey responses. Due to their linguistic capacities, it is possible that LLMs are an efficient alternative to time-consuming manual coding and the pre-training of supervised machine learning models. As most existing research on this topic has focused on English-language responses relating to non-complex topics or on single LLMs, it is unclear whether its findings generalize and how the quality of these classifications compares to established methods. In this study, we investigate to what extent different LLMs can be used to code open-ended survey responses in other contexts, using German data on reasons for survey participation as an example. We compare several state-of-the-art LLMs and several prompting approaches, and evaluate the LLMs' performance by using human expert codings. Overall performance differs greatly between LLMs, and only a fine-tuned LLM achieves satisfactory levels of predictive performance. Performance differences between prompting approaches are conditional on the LLM used. Finally, LLMs' unequal classification performance across different categories of reasons for survey participation results in different categorical distributions when not using fine-tuning. We discuss the implications of these findings, both for methodological research on coding open-ended responses and for their substantive analysis, and for practitioners processing or substantively analyzing such data. Finally, we highlight the many trade-offs researchers need to consider when choosing automated methods for open-ended response classification in the age of LLMs. In doing so, our study contributes to the growing body of research about the conditions under which LLMs can be efficiently, accurately, and reliably leveraged in survey research.
comment: to appear in Survey Research Methods
♻ ☆ HiURE: Hierarchical Exemplar Contrastive Learning for Unsupervised Relation Extraction NAACL 2022
Unsupervised relation extraction aims to extract the relationship between entities from natural language sentences without prior information on relational scope or distribution. Existing works either utilize self-supervised schemes to refine relational feature signals by iteratively leveraging adaptive clustering and classification that provoke gradual drift problems, or adopt instance-wise contrastive learning which unreasonably pushes apart those sentence pairs that are semantically similar. To overcome these defects, we propose a novel contrastive learning framework named HiURE, which has the capability to derive hierarchical signals from relational feature space using cross hierarchy attention and effectively optimize relation representation of sentences under exemplar-wise contrastive learning. Experimental results on two public datasets demonstrate the advanced effectiveness and robustness of HiURE on unsupervised relation extraction when compared with state-of-the-art models.
comment: In NAACL 2022 as a long paper. Code and data available at https://github.com/THU-BPM/HiURE
♻ ☆ VideoMAR: Autoregressive Video Generatio with Continuous Tokens
Masked-based autoregressive models have demonstrated promising image generation capability in continuous space. However, their potential for video generation remains under-explored. In this paper, we propose \textbf{VideoMAR}, a concise and efficient decoder-only autoregressive image-to-video model with continuous tokens, composing temporal frame-by-frame and spatial masked generation. We first identify temporal causality and spatial bi-directionality as the first principle of video AR models, and propose the next-frame diffusion loss for the integration of mask and video generation. Besides, the huge cost and difficulty of long sequence autoregressive modeling is a basic but crucial issue. To this end, we propose the temporal short-to-long curriculum learning and spatial progressive resolution training, and employ progressive temperature strategy at inference time to mitigate the accumulation error. Furthermore, VideoMAR replicates several unique capacities of language models to video generation. It inherently bears high efficiency due to simultaneous temporal-wise KV cache and spatial-wise parallel generation, and presents the capacity of spatial and temporal extrapolation via 3D rotary embeddings. On the VBench-I2V benchmark, VideoMAR surpasses the previous state-of-the-art (Cosmos I2V) while requiring significantly fewer parameters ($9.3\%$), training data ($0.5\%$), and GPU resources ($0.2\%$).
♻ ☆ Spatial Context-based Self-Supervised Learning for Handwritten Text Recognition
Handwritten Text Recognition (HTR) is a relevant problem in computer vision, and implies unique challenges owing to its inherent variability and the rich contextualization required for its interpretation. Despite the success of Self-Supervised Learning (SSL) in computer vision, its application to HTR has been rather scattered, leaving key SSL methodologies unexplored. This work focuses on one of them, namely Spatial Context-based SSL. We investigate how this family of approaches can be adapted and optimized for HTR and propose new workflows that leverage the unique features of handwritten text. Our experiments demonstrate that the methods considered lead to advancements in the state-of-the-art of SSL for HTR in a number of benchmark cases.
comment: Published at Pattern Recognition Letters (PRL)
♻ ☆ An Effective Incorporating Heterogeneous Knowledge Curriculum Learning for Sequence Labeling ACL 2025
Sequence labeling models often benefit from incorporating external knowledge. However, this practice introduces data heterogeneity and complicates the model with additional modules, leading to increased expenses for training a high-performing model. To address this challenge, we propose a two-stage curriculum learning (TCL) framework specifically designed for sequence labeling tasks. The TCL framework enhances training by gradually introducing data instances from easy to hard, aiming to improve both performance and training speed. Furthermore, we explore different metrics for assessing the difficulty levels of sequence labeling tasks. Through extensive experimentation on six Chinese word segmentation (CWS) and Part-of-speech tagging (POS) datasets, we demonstrate the effectiveness of our model in enhancing the performance of sequence labeling models. Additionally, our analysis indicates that TCL accelerates training and alleviates the slow training problem associated with complex models.
comment: 10 pages, 9 tables, 3 figures, Accepted by ACL 2025 (short paper)
♻ ☆ Trust Region Preference Approximation: A simple and stable reinforcement learning algorithm for LLM reasoning
Recently, Large Language Models (LLMs) have rapidly evolved, approaching Artificial General Intelligence (AGI) while benefiting from large-scale reinforcement learning to enhance Human Alignment (HA) and Reasoning. Recent reward-based optimization algorithms, such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO) have achieved significant performance on reasoning tasks, whereas preference-based optimization algorithms such as Direct Preference Optimization (DPO) significantly improve the performance of LLMs on human alignment. However, despite the strong performance of reward-based optimization methods in alignment tasks , they remain vulnerable to reward hacking. Furthermore, preference-based algorithms (such as Online DPO) haven't yet matched the performance of reward-based optimization algorithms (like PPO) on reasoning tasks, making their exploration in this specific area still a worthwhile pursuit. Motivated by these challenges, we propose the Trust Region Preference Approximation (TRPA) algorithm, which integrates rule-based optimization with preference-based optimization for reasoning tasks. As a preference-based algorithm, TRPA naturally eliminates the reward hacking issue. TRPA constructs preference levels using predefined rules, forms corresponding preference pairs, and leverages a novel optimization algorithm for RL training with a theoretical monotonic improvement guarantee. Experimental results demonstrate that TRPA not only achieves competitive performance on reasoning tasks but also exhibits robust stability. The code of this paper are released and updating on https://github.com/XueruiSu/Trust-Region-Preference-Approximation.git.
comment: 10pages
♻ ☆ Accurate and scalable exchange-correlation with deep learning
Density Functional Theory (DFT) is the most widely used electronic structure method for predicting the properties of molecules and materials. Although DFT is, in principle, an exact reformulation of the Schr\"odinger equation, practical applications rely on approximations to the unknown exchange-correlation (XC) functional. Most existing XC functionals are constructed using a limited set of increasingly complex, hand-crafted features that improve accuracy at the expense of computational efficiency. Yet, no current approximation achieves the accuracy and generality for predictive modeling of laboratory experiments at chemical accuracy -- typically defined as errors below 1 kcal/mol. In this work, we present Skala, a modern deep learning-based XC functional that bypasses expensive hand-designed features by learning representations directly from data. Skala achieves chemical accuracy for atomization energies of small molecules while retaining the computational efficiency typical of semi-local DFT. This performance is enabled by training on an unprecedented volume of high-accuracy reference data generated using computationally intensive wavefunction-based methods. Notably, Skala systematically improves with additional training data covering diverse chemistry. By incorporating a modest amount of additional high-accuracy data tailored to chemistry beyond atomization energies, Skala achieves accuracy competitive with the best-performing hybrid functionals across general main group chemistry, at the cost of semi-local DFT. As the training dataset continues to expand, Skala is poised to further enhance the predictive power of first-principles simulations.
comment: Main: 13 pages plus references, 11 figures and tables. Supplementary information: 19 pages, 12 figures and tables. v2 update: fix rendering of figure 1 and part of figure 5 in Safari PDF viewer
♻ ☆ Large Language Models for Automated Literature Review: An Evaluation of Reference Generation, Abstract Writing, and Review Composition
Large language models (LLMs) have emerged as a potential solution to automate the complex processes involved in writing literature reviews, such as literature collection, organization, and summarization. However, it is yet unclear how good LLMs are at automating comprehensive and reliable literature reviews. This study introduces a framework to automatically evaluate the performance of LLMs in three key tasks of literature writing: reference generation, literature summary, and literature review composition. We introduce multidimensional evaluation metrics that assess the hallucination rates in generated references and measure the semantic coverage and factual consistency of the literature summaries and compositions against human-written counterparts. The experimental results reveal that even the most advanced models still generate hallucinated references, despite recent progress. Moreover, we observe that the performance of different models varies across disciplines when it comes to writing literature reviews. These findings highlight the need for further research and development to improve the reliability of LLMs in automating academic literature reviews.
comment: 12 pages, 5 figures, 5 tables
♻ ☆ FLARE: Towards Universal Dataset Purification against Backdoor Attacks
Deep neural networks (DNNs) are susceptible to backdoor attacks, where adversaries poison datasets with adversary-specified triggers to implant hidden backdoors, enabling malicious manipulation of model predictions. Dataset purification serves as a proactive defense by removing malicious training samples to prevent backdoor injection at its source. We first reveal that the current advanced purification methods rely on a latent assumption that the backdoor connections between triggers and target labels in backdoor attacks are simpler to learn than the benign features. We demonstrate that this assumption, however, does not always hold, especially in all-to-all (A2A) and untargeted (UT) attacks. As a result, purification methods that analyze the separation between the poisoned and benign samples in the input-output space or the final hidden layer space are less effective. We observe that this separability is not confined to a single layer but varies across different hidden layers. Motivated by this understanding, we propose FLARE, a universal purification method to counter various backdoor attacks. FLARE aggregates abnormal activations from all hidden layers to construct representations for clustering. To enhance separation, FLARE develops an adaptive subspace selection algorithm to isolate the optimal space for dividing an entire dataset into two clusters. FLARE assesses the stability of each cluster and identifies the cluster with higher stability as poisoned. Extensive evaluations on benchmark datasets demonstrate the effectiveness of FLARE against 22 representative backdoor attacks, including all-to-one (A2O), all-to-all (A2A), and untargeted (UT) attacks, and its robustness to adaptive attacks. Codes are available at \href{https://github.com/THUYimingLi/BackdoorBox}{BackdoorBox} and \href{https://github.com/vtu81/backdoor-toolbox}{backdoor-toolbox}.
comment: 15 pages, This paper is accepted and will appear in TIFS (CCF-A)
♻ ☆ Influential Bandits: Pulling an Arm May Change the Environment
While classical formulations of multi-armed bandit problems assume that each arm's reward is independent and stationary, real-world applications often involve non-stationary environments and interdependencies between arms. In particular, selecting one arm may influence the future rewards of other arms, a scenario not adequately captured by existing models such as rotting bandits or restless bandits. To address this limitation, we propose the influential bandit problem, which models inter-arm interactions through an unknown, symmetric, positive semi-definite interaction matrix that governs the dynamics of arm losses. We formally define this problem and establish two regret lower bounds, including a superlinear $\Omega(T^2 / \log^2 T)$ bound for the standard LCB algorithm (loss minimization version of UCB) and an algorithm-independent $\Omega(T)$ bound, which highlight the inherent difficulty of the setting. We then introduce a new algorithm based on a lower confidence bound (LCB) estimator tailored to the structure of the loss dynamics. Under mild assumptions, our algorithm achieves a regret of $O(KT \log T)$, which is nearly optimal in terms of its dependence on the time horizon. The algorithm is simple to implement and computationally efficient. Empirical evaluations on both synthetic and real-world datasets demonstrate the presence of inter-arm influence and confirm the superior performance of our method compared to conventional bandit algorithms.
comment: TMLR
♻ ☆ Aligning AI Research with the Needs of Clinical Coding Workflows: Eight Recommendations Based on US Data Analysis and Critical Review ACL 2025
Clinical coding is crucial for healthcare billing and data analysis. Manual clinical coding is labour-intensive and error-prone, which has motivated research towards full automation of the process. However, our analysis, based on US English electronic health records and automated coding research using these records, shows that widely used evaluation methods are not aligned with real clinical contexts. For example, evaluations that focus on the top 50 most common codes are an oversimplification, as there are thousands of codes used in practice. This position paper aims to align AI coding research more closely with practical challenges of clinical coding. Based on our analysis, we offer eight specific recommendations, suggesting ways to improve current evaluation methods. Additionally, we propose new AI-based methods beyond automated coding, suggesting alternative approaches to assist clinical coders in their workflows.
comment: Accepted to the ACL 2025 Main Conference
♻ ☆ From Data-Driven to Purpose-Driven Artificial Intelligence: Systems Thinking for Data-Analytic Automation of Patient Care
In this work, we reflect on the data-driven modeling paradigm that is gaining ground in AI-driven automation of patient care. We argue that the repurposing of existing real-world patient datasets for machine learning may not always represent an optimal approach to model development as it could lead to undesirable outcomes in patient care. We reflect on the history of data analysis to explain how the data-driven paradigm rose to popularity, and we envision ways in which systems thinking and clinical domain theory could complement the existing model development approaches in reaching human-centric outcomes. We call for a purpose-driven machine learning paradigm that is grounded in clinical theory and the sociotechnical realities of real-world operational contexts. We argue that understanding the utility of existing patient datasets requires looking in two directions: upstream towards the data generation, and downstream towards the automation objectives. This purpose-driven perspective to AI system development opens up new methodological opportunities and holds promise for AI automation of patient care.
comment: The work is under review at ACM Health
♻ ☆ Advancing oncology with federated learning: transcending boundaries in breast, lung, and prostate cancer. A systematic review
Federated Learning (FL) has emerged as a promising solution to address the limitations of centralised machine learning (ML) in oncology, particularly in overcoming privacy concerns and harnessing the power of diverse, multi-center data. This systematic review synthesises current knowledge on the state-of-the-art FL in oncology, focusing on breast, lung, and prostate cancer. Distinct from previous surveys, our comprehensive review critically evaluates the real-world implementation and impact of FL on cancer care, demonstrating its effectiveness in enhancing ML generalisability, performance and data privacy in clinical settings and data. We evaluated state-of-the-art advances in FL, demonstrating its growing adoption amid tightening data privacy regulations. FL outperformed centralised ML in 15 out of the 25 studies reviewed, spanning diverse ML models and clinical applications, and facilitating integration of multi-modal information for precision medicine. Despite the current challenges identified in reproducibility, standardisation and methodology across studies, the demonstrable benefits of FL in harnessing real-world data and addressing clinical needs highlight its significant potential for advancing cancer research. We propose that future research should focus on addressing these limitations and investigating further advanced FL methods, to fully harness data diversity and realise the transformative power of cutting-edge FL in cancer care.
comment: 5 Figures, 3 Tables, 1 Supplementary Table
♻ ☆ BIS Reasoning 1.0: The First Large-Scale Japanese Benchmark for Belief-Inconsistent Syllogistic Reasoning
We present BIS Reasoning 1.0, the first large-scale Japanese dataset of syllogistic reasoning problems explicitly designed to evaluate belief-inconsistent reasoning in large language models (LLMs). Unlike prior datasets such as NeuBAROCO and JFLD, which focus on general or belief-aligned reasoning, BIS Reasoning 1.0 introduces logically valid yet belief-inconsistent syllogisms to uncover reasoning biases in LLMs trained on human-aligned corpora. We benchmark state-of-the-art models - including GPT models, Claude models, and leading Japanese LLMs - revealing significant variance in performance, with GPT-4o achieving 79.54% accuracy. Our analysis identifies critical weaknesses in current LLMs when handling logically valid but belief-conflicting inputs. These findings have important implications for deploying LLMs in high-stakes domains such as law, healthcare, and scientific literature, where truth must override intuitive belief to ensure integrity and safety.
comment: This version includes an updated literature review, added acknowledgements, and a revised author list
♻ ☆ A Systematic Survey of Natural Language Processing for the Greek Language
Comprehensive monolingual Natural Language Processing (NLP) surveys are essential for assessing language-specific challenges, resource availability, and research gaps. However, existing surveys often lack standardized methodologies, leading to selection bias and fragmented coverage of NLP tasks and resources. This study introduces a generalizable framework for systematic monolingual NLP surveys. Our approach integrates a structured search protocol to minimize bias, an NLP task taxonomy for classification, and language resource taxonomies to identify potential benchmarks and highlight opportunities for improving resource availability. We apply this framework to Greek NLP (2012-2023), providing an in-depth analysis of its current state, task-specific progress, and resource gaps. The survey results are publicly available (https://doi.org/10.5281/zenodo.15314882) and are regularly updated to provide an evergreen resource. This systematic survey of Greek NLP serves as a case study, demonstrating the effectiveness of our framework and its potential for broader application to other not so well-resourced languages as regards NLP.
comment: This version matches the paper published in Patterns (Cell Press). The title has been updated to reflect the published version
♻ ☆ Seewo's Submission to MLC-SLM: Lessons learned from Speech Reasoning Language Models
This paper presents Seewo's systems for both tracks of the Multilingual Conversational Speech Language Model Challenge (MLC-SLM), addressing automatic speech recognition (ASR) and speaker diarization with ASR (SD-ASR). We introduce a multi-stage training pipeline that explicitly enhances reasoning and self-correction in speech language models for ASR. Our approach combines curriculum learning for progressive capability acquisition, Chain-of-Thought data augmentation to foster intermediate reflection, and Reinforcement Learning with Verifiable Rewards (RLVR) to further refine self-correction through reward-driven optimization. This approach achieves substantial improvements over the official challenge baselines. On the evaluation set, our best system attains a WER/CER of 11.57% for Track 1 and a tcpWER/tcpCER of 17.67% for Track 2. Comprehensive ablation studies demonstrate the effectiveness of each component under challenge constraints.
♻ ☆ On Finding Small Hyper-Gradients in Bilevel Optimization: Hardness Results and Improved Analysis COLT 2024
Bilevel optimization reveals the inner structure of otherwise oblique optimization problems, such as hyperparameter tuning, neural architecture search, and meta-learning. A common goal in bilevel optimization is to minimize a hyper-objective that implicitly depends on the solution set of the lower-level function. Although this hyper-objective approach is widely used, its theoretical properties have not been thoroughly investigated in cases where the lower-level functions lack strong convexity. In this work, we first provide hardness results to show that the goal of finding stationary points of the hyper-objective for nonconvex-convex bilevel optimization can be intractable for zero-respecting algorithms. Then we study a class of tractable nonconvex-nonconvex bilevel problems when the lower-level function satisfies the Polyak-{\L}ojasiewicz (PL) condition. We show a simple first-order algorithm can achieve better complexity bounds of $\tilde{\mathcal{O}}(\epsilon^{-2})$, $\tilde{\mathcal{O}}(\epsilon^{-4})$ and $\tilde{\mathcal{O}}(\epsilon^{-6})$ in the deterministic, partially stochastic, and fully stochastic setting respectively. The complexities in the first two cases are optimal up to logarithmic factors.
comment: Published in COLT 2024. This arXiv version refines Assumption 4.1 (d); adds discussions on related works in Appendix A; and corrects the kappa dependency in the upper bounds
♻ ☆ Task-Aware Virtual Training: Enhancing Generalization in Meta-Reinforcement Learning for Out-of-Distribution Tasks ICML 2025
Meta reinforcement learning aims to develop policies that generalize to unseen tasks sampled from a task distribution. While context-based meta-RL methods improve task representation using task latents, they often struggle with out-of-distribution (OOD) tasks. To address this, we propose Task-Aware Virtual Training (TAVT), a novel algorithm that accurately captures task characteristics for both training and OOD scenarios using metric-based representation learning. Our method successfully preserves task characteristics in virtual tasks and employs a state regularization technique to mitigate overestimation errors in state-varying environments. Numerical results demonstrate that TAVT significantly enhances generalization to OOD tasks across various MuJoCo and MetaWorld environments. Our code is available at https://github.com/JM-Kim-94/tavt.git.
comment: 9 pages main paper, 20 pages appendices with reference. Accepted to ICML 2025
♻ ☆ Wolfpack Adversarial Attack for Robust Multi-Agent Reinforcement Learning ICML 2025
Traditional robust methods in multi-agent reinforcement learning (MARL) often struggle against coordinated adversarial attacks in cooperative scenarios. To address this limitation, we propose the Wolfpack Adversarial Attack framework, inspired by wolf hunting strategies, which targets an initial agent and its assisting agents to disrupt cooperation. Additionally, we introduce the Wolfpack-Adversarial Learning for MARL (WALL) framework, which trains robust MARL policies to defend against the proposed Wolfpack attack by fostering systemwide collaboration. Experimental results underscore the devastating impact of the Wolfpack attack and the significant robustness improvements achieved by WALL. Our code is available at https://github.com/sunwoolee0504/WALL.
comment: 9 pages main, 23 pages appendix with reference. Accepeted by ICML 2025
♻ ☆ LLäMmlein: Transparent, Compact and Competitive German-Only Language Models from Scratch ACL25
We create two German-only decoder models, LL\"aMmlein 120M and 1B, transparently from scratch and publish them, along with the training data, for the German NLP research community to use. The model training involved several key steps, including extensive data preprocessing, the creation of a custom German tokenizer, the training itself, as well as the evaluation of the final models on various benchmarks. Throughout the training process, multiple checkpoints were saved and analyzed using the SuperGLEBer benchmark to monitor the models' learning dynamics. Compared to state-of-the-art models on the SuperGLEBer benchmark, both LL\"aMmlein models performed competitively, consistently matching or surpassing models with similar parameter sizes. The results show that the models' quality scales with size as expected, but performance improvements on some tasks plateaued early, offering valuable insights into resource allocation for future model development.
comment: camera ready @ACL25; https://www.informatik.uni-wuerzburg.de/datascience/projects/nlp/llammlein/
♻ ☆ Efficient Long CoT Reasoning in Small Language Models
Recent large reasoning models such as DeepSeek-R1 exhibit strong complex problems solving abilities by generating long chain-of-thought (CoT) reasoning steps. It is challenging to directly train small language models (SLMs) to emerge long CoT. Thus, distillation becomes a practical method to enable SLMs for such reasoning ability. However, the long CoT often contains a lot of redundant contents (e.g., overthinking steps) which may make SLMs hard to learn considering their relatively poor capacity and generalization. To address this issue, we propose a simple-yet-effective method to prune unnecessary steps in long CoT, and then employ an on-policy method for the SLM itself to curate valid and useful long CoT training data. In this way, SLMs can effectively learn efficient long CoT reasoning and preserve competitive performance at the same time. Experimental results across a series of mathematical reasoning benchmarks demonstrate the effectiveness of the proposed method in distilling long CoT reasoning ability into SLMs which maintains the competitive performance but significantly reduces generating redundant reasoning steps.
♻ ☆ ALPS: Attention Localization and Pruning Strategy for Efficient Alignment of Large Language Models ACL25
Aligning general-purpose large language models (LLMs) to downstream tasks often incurs significant training adjustment costs. Prior research has explored various avenues to enhance alignment efficiency, primarily through minimal-data training or data-driven activations to identify key attention heads. However, these approaches inherently introduce data dependency, which hinders generalization and reusability. To address this issue and enhance model alignment efficiency, we propose the Attention Localization and Pruning Strategy (ALPS), an efficient algorithm that localizes the most task-sensitive attention heads and prunes by restricting attention training updates to these heads, thereby reducing alignment costs. Experimental results demonstrate that our method activates only 10% of attention parameters during fine-tuning while achieving a 2% performance improvement over baselines on three tasks. Moreover, the identified task-specific heads are transferable across datasets and mitigate knowledge forgetting. Our work and findings provide a novel perspective on efficient LLM alignment. The code is available at https://github.com/VoiceBeer/ALPS.
comment: Accepted@ACL25-findings, 17 pages, 8 figures, 14 tables
♻ ☆ TransXSSM: A Hybrid Transformer State Space Model with Unified Rotary Position Embedding
Transformers exhibit proficiency in capturing long-range dependencies, whereas State Space Models (SSMs) facilitate linear-time sequence modeling. Notwithstanding their synergistic potential, the integration of these architectures presents a significant challenge, primarily attributable to a fundamental incongr inuity their respective positional encoding mechanisms: Transformers rely on explicit Rotary Position Embeddings (RoPE), while SSMs leverage implicit positional representations via convolutions. This divergence often precipitates discontinuities and suboptimal performance.To address this impediment, we propose a unified rotary position embedding (Unified RoPE) methodology, thereby establishing a consistent positional encoding framework for both self-attention and state-space components. Using this Unified RoPE, we introduce TransXSSM, a hybrid architecture that coherently integrates the Transformer and SSM layers under this unified positional encoding scheme. At a 4 sequenceK length, TransXSSM exhibits training and inference speeds that are 42.3% and 29.5% faster, respectively, relative to standard Transformer models. It also delivers higher accuracy: under comparable settings, it surpasses a Transformer baseline by over 4% on language modeling benchmarks.TransXSSM furthermore scales more effectively: TransXSSM-1.3B gains 7.22% in average accuracy over its 320M version (versus about 6% gains for equivalent Transformers or SSMs). Our results show that unified positional encoding resolves positional incompatibility in hybrid models, enabling efficient, high-performance long-context modeling.
♻ ☆ Synthesizing Composite Hierarchical Structure from Symbolic Music Corpora IJCAI '25
Western music is an innately hierarchical system of interacting levels of structure, from fine-grained melody to high-level form. In order to analyze music compositions holistically and at multiple granularities, we propose a unified, hierarchical meta-representation of musical structure called the structural temporal graph (STG). For a single piece, the STG is a data structure that defines a hierarchy of progressively finer structural musical features and the temporal relationships between them. We use the STG to enable a novel approach for deriving a representative structural summary of a music corpus, which we formalize as a dually NP-hard combinatorial optimization problem extending the Generalized Median Graph problem. Our approach first applies simulated annealing to develop a measure of structural distance between two music pieces rooted in graph isomorphism. Our approach then combines the formal guarantees of SMT solvers with nested simulated annealing over structural distances to produce a structurally sound, representative centroid STG for an entire corpus of STGs from individual pieces. To evaluate our approach, we conduct experiments verifying that structural distance accurately differentiates between music pieces, and that derived centroids accurately structurally characterize their corpora.
comment: In Proceedings of the 34th International Joint Conference on Artificial Intelligence (IJCAI '25), Montreal, Canada, August 2025
♻ ☆ SurgSora: Object-Aware Diffusion Model for Controllable Surgical Video Generation
Surgical video generation can enhance medical education and research, but existing methods lack fine-grained motion control and realism. We introduce SurgSora, a framework that generates high-fidelity, motion-controllable surgical videos from a single input frame and user-specified motion cues. Unlike prior approaches that treat objects indiscriminately or rely on ground-truth segmentation masks, SurgSora leverages self-predicted object features and depth information to refine RGB appearance and optical flow for precise video synthesis. It consists of three key modules: (1) the Dual Semantic Injector, which extracts object-specific RGB-D features and segmentation cues to enhance spatial representations; (2) the Decoupled Flow Mapper, which fuses multi-scale optical flow with semantic features for realistic motion dynamics; and (3) the Trajectory Controller, which estimates sparse optical flow and enables user-guided object movement. By conditioning these enriched features within the Stable Video Diffusion, SurgSora achieves state-of-the-art visual authenticity and controllability in advancing surgical video synthesis, as demonstrated by extensive quantitative and qualitative comparisons. Our human evaluation in collaboration with expert surgeons further demonstrates the high realism of SurgSora-generated videos, highlighting the potential of our method for surgical training and education. Our project is available at https://surgsora.github.io/surgsora.github.io.
♻ ☆ GRAM: A Generative Foundation Reward Model for Reward Generalization ICML 2025
In aligning large language models (LLMs), reward models have played an important role, but are standardly trained as discriminative models and rely only on labeled human preference data. In this paper, we explore methods that train reward models using both unlabeled and labeled data. Building on the generative models in LLMs, we develop a generative reward model that is first trained via large-scale unsupervised learning and then fine-tuned via supervised learning. We also show that by using label smoothing, we are in fact optimizing a regularized pairwise ranking loss. This result, in turn, provides a new view of training reward models, which links generative models and discriminative models under the same class of training objectives. The outcome of these techniques is a foundation reward model, which can be applied to a wide range of tasks with little or no further fine-tuning effort. Extensive experiments show that this model generalizes well across several tasks, including response ranking, reinforcement learning from human feedback, and task adaptation with fine-tuning, achieving significant performance improvements over several strong baseline models.
comment: Accepted by ICML 2025
♻ ☆ REVOLVE: Optimizing AI Systems by Tracking Response Evolution in Textual Optimization ICML 2025
Recent advancements in large language models (LLMs) have significantly enhanced the ability of LLM-based systems to perform complex tasks through natural language processing and tool interaction. However, optimizing these LLM-based systems for specific tasks remains challenging, often requiring manual interventions like prompt engineering and hyperparameter tuning. Existing automatic optimization methods, such as textual feedback-based techniques (e.g., TextGrad), tend to focus on immediate feedback, analogous to using immediate derivatives in traditional numerical gradient descent. However, relying solely on such feedback can be limited when the adjustments made in response to this feedback are either too small or fluctuate irregularly, potentially slowing down or even stalling the optimization process. To overcome these challenges, more adaptive methods are needed, especially in situations where the system's response is evolving slowly or unpredictably. In this paper, we introduce REVOLVE, an optimization method that tracks how "R"esponses "EVOLVE" across iterations in LLM systems. By focusing on the evolution of responses over time, REVOLVE enables more stable and effective optimization by making thoughtful, progressive adjustments at each step. Experimental results demonstrate that REVOLVE outperforms competitive baselines, achieving a 7.8% improvement in prompt optimization, a 20.72% gain in solution refinement, and a 29.17% increase in code optimization. Additionally, REVOLVE converges in fewer iterations, resulting in significant computational savings. Beyond its practical contributions, REVOLVE highlights a promising direction, where the rich knowledge from established optimization principles can be leveraged to enhance LLM systems, which paves the way for further advancements in this hybrid domain.
comment: 20 pages, 2 figures, accepted by ICML 2025
♻ ☆ MSVIT: Improving Spiking Vision Transformer Using Multi-scale Attention Fusion IJCAI'25
The combination of Spiking Neural Networks (SNNs) with Vision Transformer architectures has garnered significant attention due to their potential for energy-efficient and high-performance computing paradigms. However, a substantial performance gap still exists between SNN-based and ANN-based transformer architectures. While existing methods propose spiking self-attention mechanisms that are successfully combined with SNNs, the overall architectures proposed by these methods suffer from a bottleneck in effectively extracting features from different image scales. In this paper, we address this issue and propose MSVIT. This novel spike-driven Transformer architecture firstly uses multi-scale spiking attention (MSSA) to enhance the capabilities of spiking attention blocks. We validate our approach across various main datasets. The experimental results show that MSVIT outperforms existing SNN-based models, positioning itself as a state-of-the-art solution among SNN-transformer architectures. The codes are available at https://github.com/Nanhu-AI-Lab/MSViT.
comment: 11pages, 2figures, accepted by IJCAI'25 (34th International Joint Conference on Artificial Intelligence)
♻ ☆ CODESYNC: Synchronizing Large Language Models with Dynamic Code Evolution at Scale
Large Language Models (LLMs) have exhibited exceptional performance in software engineering yet face challenges in adapting to continually evolving code knowledge, particularly regarding the frequent updates of third-party library APIs. This limitation, stemming from static pre-training datasets, often results in non-executable code or implementations with suboptimal safety and efficiency. To this end, this paper introduces CODESYNC, a data engine for identifying outdated code patterns and collecting real-time code knowledge updates from Python third-party libraries. Building upon CODESYNC, we develop CODESYNCBENCH, a comprehensive benchmark for assessing LLMs' ability to stay synchronized with code evolution, which covers real-world updates for 220 APIs from six Python libraries. Our benchmark offers 3,300 test cases across three evaluation tasks and an update-aware instruction tuning dataset consisting of 2,200 training samples. Extensive experiments on 14 state-of-the-art LLMs reveal that they struggle with dynamic code evolution, even with the support of advanced knowledge updating methods (e.g., DPO, ORPO, and SimPO). We believe that our benchmark can offer a strong foundation for the development of more effective methods for real-time code knowledge updating in the future. The experimental code and dataset are publicly available at: https://github.com/Lucky-voyage/Code-Sync.
♻ ☆ Beyond Propagation of Chaos: A Stochastic Algorithm for Mean Field Optimization
Gradient flow in the 2-Wasserstein space is widely used to optimize functionals over probability distributions and is typically implemented using an interacting particle system with $n$ particles. Analyzing these algorithms requires showing (a) that the finite-particle system converges and/or (b) that the resultant empirical distribution of the particles closely approximates the optimal distribution (i.e., propagation of chaos). However, establishing efficient sufficient conditions can be challenging, as the finite particle system may produce heavily dependent random variables. In this work, we study the virtual particle stochastic approximation, originally introduced for Stein Variational Gradient Descent. This method can be viewed as a form of stochastic gradient descent in the Wasserstein space and can be implemented efficiently. In popular settings, we demonstrate that our algorithm's output converges to the optimal distribution under conditions similar to those for the infinite particle limit, and it produces i.i.d. samples without the need to explicitly establish propagation of chaos bounds.
♻ ☆ SpokenWOZ: A Large-Scale Speech-Text Benchmark for Spoken Task-Oriented Dialogue Agents NeurIPS 2023
Task-oriented dialogue (TOD) models have made significant progress in recent years. However, previous studies primarily focus on datasets written by annotators, which has resulted in a gap between academic research and real-world spoken conversation scenarios. While several small-scale spoken TOD datasets are proposed to address robustness issues such as ASR errors, they ignore the unique challenges in spoken conversation. To tackle the limitations, we introduce SpokenWOZ, a large-scale speech-text dataset for spoken TOD, containing 8 domains, 203k turns, 5.7k dialogues and 249 hours of audios from human-to-human spoken conversations. SpokenWOZ further incorporates common spoken characteristics such as word-by-word processing and reasoning in spoken language. Based on these characteristics, we present cross-turn slot and reasoning slot detection as new challenges. We conduct experiments on various baselines, including text-modal models, newly proposed dual-modal models, and LLMs, e.g., ChatGPT. The results show that the current models still have substantial room for improvement in spoken conversation, where the most advanced dialogue state tracker only achieves 25.65% in joint goal accuracy and the SOTA end-to-end model only correctly completes the user request in 52.1% of dialogues. The dataset, code, and leaderboard are available: https://spokenwoz.github.io/.
comment: NeurIPS 2023
♻ ☆ PLD: A Choice-Theoretic List-Wise Knowledge Distillation
Knowledge distillation is a model compression technique in which a compact "student" network is trained to replicate the predictive behavior of a larger "teacher" network. In logit-based knowledge distillation it has become the de facto approach to augment cross-entropy with a distillation term. Typically this term is either a KL divergence-matching marginal probabilities or a correlation-based loss capturing intra- and inter-class relationships but in every case it sits as an add-on to cross-entropy with its own weight that must be carefully tuned. In this paper we adopt a choice-theoretic perspective and recast knowledge distillation under the Plackett-Luce model by interpreting teacher logits as "worth" scores. We introduce Plackett-Luce Distillation (PLD), a weighted list-wise ranking loss in which the teacher model transfers knowledge of its full ranking of classes, weighting each ranked choice by its own confidence. PLD directly optimizes a single teacher-optimal ranking of the true label first, followed by the remaining classes in descending teacher confidence, yielding a convex, translation-invariant surrogate that subsumes weighted cross-entropy. Empirically on standard image classification benchmarks, PLD improves Top-1 accuracy by an average of +0.42% over DIST (arXiv:2205.10536) and +1.04% over KD (arXiv:1503.02531) in homogeneous settings and by +0.48% and +1.09% over DIST and KD, respectively, in heterogeneous settings.
♻ ☆ Perspective Transition of Large Language Models for Solving Subjective Tasks ACL 2025
Large language models (LLMs) have revolutionized the field of natural language processing, enabling remarkable progress in various tasks. Different from objective tasks such as commonsense reasoning and arithmetic question-answering, the performance of LLMs on subjective tasks is still limited, where the perspective on the specific problem plays crucial roles for better interpreting the context and giving proper response. For example, in certain scenarios, LLMs may perform better when answering from an expert role perspective, potentially eliciting their relevant domain knowledge. In contrast, in some scenarios, LLMs may provide more accurate responses when answering from a third-person standpoint, enabling a more comprehensive understanding of the problem and potentially mitigating inherent biases. In this paper, we propose Reasoning through Perspective Transition (RPT), a method based on in-context learning that enables LLMs to dynamically select among direct, role, and third-person perspectives for the best way to solve corresponding subjective problem. Through extensive experiments on totally 12 subjective tasks by using both closed-source and open-source LLMs including GPT-4, GPT-3.5, Llama-3, and Qwen-2, our method outperforms widely used single fixed perspective based methods such as chain-of-thought prompting and expert prompting, highlights the intricate ways that LLMs can adapt their perspectives to provide nuanced and contextually appropriate responses for different problems.
comment: ACL 2025 Findings
♻ ☆ Learning Strategic Language Agents in the Werewolf Game with Iterative Latent Space Policy Optimization ICML 2025
Large language model (LLM) agents have recently demonstrated impressive capabilities in various domains like open-ended conversation and multi-step decision-making. However, it remains challenging for these agents to solve strategic language games, such as Werewolf, which demand both strategic decision-making and free-form language interactions. Existing LLM agents often suffer from intrinsic bias in their action distributions and limited exploration of the unbounded text action space, resulting in suboptimal performance. To address these challenges, we propose Latent Space Policy Optimization (LSPO), an iterative framework that combines game-theoretic methods with LLM fine-tuning to build strategic language agents. LSPO leverages the observation that while the language space is combinatorially large, the underlying strategy space is relatively compact. We first map free-form utterances into a finite latent strategy space, yielding an abstracted extensive-form game. Then we apply game-theoretic methods like Counterfactual Regret Minimization (CFR) to optimize the policy in the latent space. Finally, we fine-tune the LLM via Direct Preference Optimization (DPO) to align with the learned policy. By iteratively alternating between these steps, our LSPO agents progressively enhance both strategic reasoning and language communication. Experiment on the Werewolf game shows that our agents iteratively expand the strategy space with improving performance and outperform existing Werewolf agents, underscoring their effectiveness in free-form language games with strategic interactions.
comment: Published in ICML 2025
♻ ☆ Multiclass Post-Earthquake Building Assessment Integrating High-Resolution Optical and SAR Satellite Imagery, Ground Motion, and Soil Data with Transformers
Timely and accurate assessments of building damage are crucial for effective response and recovery in the aftermath of earthquakes. Conventional preliminary damage assessments (PDA) often rely on manual door-to-door inspections, which are not only time-consuming but also pose significant safety risks. To safely expedite the PDA process, researchers have studied the applicability of satellite imagery processed with heuristic and machine learning approaches. These approaches output binary or, more recently, multiclass damage states at the scale of a block or a single building. However, the current performance of such approaches limits practical applicability. To address this limitation, we introduce a metadata-enriched, transformer based framework that combines high-resolution post-earthquake satellite imagery with building-specific metadata relevant to the seismic performance of the structure. Our model achieves state-of-the-art performance in multiclass post-earthquake damage identification for buildings from the Turkey-Syria earthquake on February 6, 2023. Specifically, we demonstrate that incorporating metadata, such as seismic intensity indicators, soil properties, and SAR damage proxy maps not only enhances the model's accuracy and ability to distinguish between damage classes, but also improves its generalizability across various regions. Furthermore, we conducted a detailed, class-wise analysis of feature importance to understand the model's decision-making across different levels of building damage. This analysis reveals how individual metadata features uniquely contribute to predictions for each damage class. By leveraging both satellite imagery and metadata, our proposed framework enables faster and more accurate damage assessments for precise, multiclass, building-level evaluations that can improve disaster response and accelerate recovery efforts for affected communities.
comment: 28 Pages, 12 Figures
♻ ☆ ChemHAS: Hierarchical Agent Stacking for Enhancing Chemistry Tools
Large Language Model (LLM)-based agents have demonstrated the ability to improve performance in chemistry-related tasks by selecting appropriate tools. However, their effectiveness remains limited by the inherent prediction errors of chemistry tools. In this paper, we take a step further by exploring how LLMbased agents can, in turn, be leveraged to reduce prediction errors of the tools. To this end, we propose ChemHAS (Chemical Hierarchical Agent Stacking), a simple yet effective method that enhances chemistry tools through optimizing agent-stacking structures from limited data. ChemHAS achieves state-of-the-art performance across four fundamental chemistry tasks, demonstrating that our method can effectively compensate for prediction errors of the tools. Furthermore, we identify and characterize four distinct agent-stacking behaviors, potentially improving interpretability and revealing new possibilities for AI agent applications in scientific research. Our code and dataset are publicly available at https: //anonymous.4open.science/r/ChemHAS-01E4/README.md.
comment: 9 pages
♻ ☆ Ring-lite: Scalable Reasoning via C3PO-Stabilized Reinforcement Learning for LLMs
We present Ring-lite, a Mixture-of-Experts (MoE)-based large language model optimized via reinforcement learning (RL) to achieve efficient and robust reasoning capabilities. Built upon the publicly available Ling-lite model, a 16.8 billion parameter model with 2.75 billion activated parameters, our approach matches the performance of state-of-the-art (SOTA) small-scale reasoning models on challenging benchmarks (e.g., AIME, LiveCodeBench, GPQA-Diamond) while activating only one-third of the parameters required by comparable models. To accomplish this, we introduce a joint training pipeline integrating distillation with RL, revealing undocumented challenges in MoE RL training. First, we identify optimization instability during RL training, and we propose Constrained Contextual Computation Policy Optimization(C3PO), a novel approach that enhances training stability and improves computational throughput via algorithm-system co-design methodology. Second, we empirically demonstrate that selecting distillation checkpoints based on entropy loss for RL training, rather than validation metrics, yields superior performance-efficiency trade-offs in subsequent RL training. Finally, we develop a two-stage training paradigm to harmonize multi-domain data integration, addressing domain conflicts that arise in training with mixed dataset. We will release the model, dataset, and code.
comment: Technical Report
♻ ☆ LLMs can be Dangerous Reasoners: Analyzing-based Jailbreak Attack on Large Language Models
The rapid development of Large Language Models (LLMs) has brought impressive advancements across various tasks. However, despite these achievements, LLMs still pose inherent safety risks, especially in the context of jailbreak attacks. Most existing jailbreak methods follow an input-level manipulation paradigm to bypass safety mechanisms. Yet, as alignment techniques improve, such attacks are becoming increasingly detectable. In this work, we identify an underexplored threat vector: the model's internal reasoning process, which can be manipulated to elicit harmful outputs in a more stealthy way. To explore this overlooked attack surface, we propose a novel black-box jailbreak attack method, Analyzing-based Jailbreak (ABJ). ABJ comprises two independent attack paths: textual and visual reasoning attacks, which exploit the model's multimodal reasoning capabilities to bypass safety mechanisms, comprehensively exposing vulnerabilities in its reasoning chain. We conduct extensive experiments on ABJ across various open-source and closed-source LLMs, VLMs, and RLMs. In particular, ABJ achieves high attack success rate (ASR) (82.1% on GPT-4o-2024-11-20) with exceptional attack efficiency (AE) among all target models, showcasing its remarkable attack effectiveness, transferability, and efficiency. Our work reveals a new type of safety risk and highlights the urgent need to mitigate implicit vulnerabilities in the model's reasoning process.
♻ ☆ Aligning Evaluation with Clinical Priorities: Calibration, Label Shift, and Error Costs
Machine learning-based decision support systems are increasingly deployed in clinical settings, where probabilistic scoring functions are used to inform and prioritize patient management decisions. However, widely used scoring rules, such as accuracy and AUC-ROC, fail to adequately reflect key clinical priorities, including calibration, robustness to distributional shifts, and sensitivity to asymmetric error costs. In this work, we propose a principled yet practical evaluation framework for selecting calibrated thresholded classifiers that explicitly accounts for the uncertainty in class prevalences and domain-specific cost asymmetries often found in clinical settings. Building on the theory of proper scoring rules, particularly the Schervish representation, we derive an adjusted variant of cross-entropy (log score) that averages cost-weighted performance over clinically relevant ranges of class balance. The resulting evaluation is simple to apply, sensitive to clinical deployment conditions, and designed to prioritize models that are both calibrated and robust to real-world variations.
♻ ☆ Math Neurosurgery: Isolating Language Models' Math Reasoning Abilities Using Only Forward Passes ACL 2025
Math reasoning is an active area of Large Language Model (LLM) research because it is a hallmark of artificial intelligence and has implications in several domains, including math education. However, few works have explored how math reasoning is encoded within LLM parameters and if it is a skill that can be isolated within models. Doing so could allow targeted intervention to improve math performance without altering non-math behavior and foster understanding of how models encode math reasoning. We introduce Math Neurosurgery (MathNeuro), a computationally efficient method we use to isolate math-specific parameters in LLMs using only forward passes. MathNeuro builds on existing work by using weights and activations to calculate parameter importance, but isolates math-specific parameters by filtering out those important for general language tasks. Through pruning parameters MathNeuro identifies, we delete a LLM's math reasoning ability without significantly impacting its general language ability. Scaling the identified parameters by a small constant improves a pretrained or instruction-tuned LLM's performance by 4-17% on GSM8K and 5-35% on MATH while leaving non-math behavior unaltered. MathNeuro is also data efficient: most of its effectiveness holds when identifying math-specific parameters using a single sample. MathNeuro highlights the potential for future work to intervene on math-specific parameters.
comment: 38 pages, 54 figures, Accepted to ACL 2025 (Main)
♻ ☆ The NordDRG AI Benchmark for Large Language Models
Large language models (LLMs) are already being piloted for clinical coding and decision support. However, until now, no open benchmark has targeted the hospital funding layer where Diagnosis-Related Groups (DRG) determine reimbursement across many countries. We release NordDRG-AI-Benchmark, the first public test-bed that captures a complete DRG rule set and evaluates an LLM's ability to reason over multilingual diagnosis, procedure, and tariff logic. The benchmark bundles three classes of artefacts: (i) definition tables with 20 interlinked tables covering DRG logic, ICD and NCSP codes, age/sex splits, and country flags; (ii) expert manuals and changelog templates describing real governance workflows; and (iii) a prompt pack of 14 CaseMix tasks that span code lookup, cross-table inference, multilingual terminology, and quality-assurance audits. All artefacts are available at: https://github.com/longshoreforrest/norddrg-ai-benchmark A baseline demonstration shows that five state-of-the-art LLMs perform very differently on the nine automatically verifiable tasks: o3 (OpenAI) scores 9 out of 9, GPT-4o and o4-mini-high score 7 out of 9, while Gemini 2.5 Pro and Gemini 2.5 Flash solve only 5 out of 9 and 3 out of 9, respectively. These results confirm that NordDRG-AI-Benchmark highlights domain-specific strengths and weaknesses that remain hidden in generic LLM benchmarks, offering a reproducible baseline for research on trustworthy automation in hospital funding.
comment: 15 pages, 4 figures
♻ ☆ The Epochal Sawtooth Phenomenon: Unveiling Training Loss Oscillations in Adam and Other Optimizers
In this paper, we identify and analyze a recurring training loss pattern, which we term the \textit{Epochal Sawtooth Phenomenon (ESP)}, commonly observed during training with adaptive gradient-based optimizers, particularly Adam optimizer. This pattern is characterized by a sharp drop in loss at the beginning of each epoch, followed by a gradual increase, resulting in a sawtooth-shaped loss curve. Through empirical observations, we demonstrate that while this effect is most pronounced with Adam, it persists, although less severely, with other optimizers such as RMSProp. We empirically analyze the mechanisms underlying ESP, focusing on key factors such as Adam's $\beta$ parameters, batch size, data shuffling, and sample replacement. Our analysis shows that ESP arises from adaptive learning rate adjustments controlled by the second moment estimate. Additionally, we identify the ``immediate re-exposure to samples'' effect during data shuffling, which causes the model to learn or memorize more at the beginning of each epoch. We also find that smaller values of $\beta_2$ exacerbate ESP but can act as a form of regularization. While ESP is not necessarily indicative of overfitting, higher model capacity can amplify the phenomenon. To further support our analysis, we replicate ESP through a high-dimensional quadratic minimization task. We demonstrate that ESP can emerge even in simple optimization scenarios, reinforcing the generality of this pattern. The code for reproducing our experiments is available at https://github.com/qiliuchn/training-loss-pattern.
comment: 15 pages, 21 figures
Computer Vision and Pattern Recognition 111
☆ Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards
Generating high-quality and photorealistic 3D assets remains a longstanding challenge in 3D vision and computer graphics. Although state-of-the-art generative models, such as diffusion models, have made significant progress in 3D generation, they often fall short of human-designed content due to limited ability to follow instructions, align with human preferences, or produce realistic textures, geometries, and physical attributes. In this paper, we introduce Nabla-R2D3, a highly effective and sample-efficient reinforcement learning alignment framework for 3D-native diffusion models using 2D rewards. Built upon the recently proposed Nabla-GFlowNet method, which matches the score function to reward gradients in a principled manner for reward finetuning, our Nabla-R2D3 enables effective adaptation of 3D diffusion models using only 2D reward signals. Extensive experiments show that, unlike vanilla finetuning baselines which either struggle to converge or suffer from reward hacking, Nabla-R2D3 consistently achieves higher rewards and reduced prior forgetting within a few finetuning steps.
comment: Technical Report (21 pages, 21 figures)
☆ Evolutionary Caching to Accelerate Your Off-the-Shelf Diffusion Model
Diffusion-based image generation models excel at producing high-quality synthetic content, but suffer from slow and computationally expensive inference. Prior work has attempted to mitigate this by caching and reusing features within diffusion transformers across inference steps. These methods, however, often rely on rigid heuristics that result in limited acceleration or poor generalization across architectures. We propose Evolutionary Caching to Accelerate Diffusion models (ECAD), a genetic algorithm that learns efficient, per-model, caching schedules forming a Pareto frontier, using only a small set of calibration prompts. ECAD requires no modifications to network parameters or reference images. It offers significant inference speedups, enables fine-grained control over the quality-latency trade-off, and adapts seamlessly to different diffusion models. Notably, ECAD's learned schedules can generalize effectively to resolutions and model variants not seen during calibration. We evaluate ECAD on PixArt-alpha, PixArt-Sigma, and FLUX-1.dev using multiple metrics (FID, CLIP, Image Reward) across diverse benchmarks (COCO, MJHQ-30k, PartiPrompts), demonstrating consistent improvements over previous approaches. On PixArt-alpha, ECAD identifies a schedule that outperforms the previous state-of-the-art method by 4.47 COCO FID while increasing inference speedup from 2.35x to 2.58x. Our results establish ECAD as a scalable and generalizable approach for accelerating diffusion inference. Our project website is available at https://aniaggarwal.github.io/ecad and our code is available at https://github.com/aniaggarwal/ecad.
comment: 29 pages, 22 figures, 9 tables
☆ Particle-Grid Neural Dynamics for Learning Deformable Object Models from RGB-D Videos
Modeling the dynamics of deformable objects is challenging due to their diverse physical properties and the difficulty of estimating states from limited visual information. We address these challenges with a neural dynamics framework that combines object particles and spatial grids in a hybrid representation. Our particle-grid model captures global shape and motion information while predicting dense particle movements, enabling the modeling of objects with varied shapes and materials. Particles represent object shapes, while the spatial grid discretizes the 3D space to ensure spatial continuity and enhance learning efficiency. Coupled with Gaussian Splattings for visual rendering, our framework achieves a fully learning-based digital twin of deformable objects and generates 3D action-conditioned videos. Through experiments, we demonstrate that our model learns the dynamics of diverse objects -- such as ropes, cloths, stuffed animals, and paper bags -- from sparse-view RGB-D recordings of robot-object interactions, while also generalizing at the category level to unseen instances. Our approach outperforms state-of-the-art learning-based and physics-based simulators, particularly in scenarios with limited camera views. Furthermore, we showcase the utility of our learned models in model-based planning, enabling goal-conditioned object manipulation across a range of tasks. The project page is available at https://kywind.github.io/pgnd .
comment: Project page: https://kywind.github.io/pgnd
☆ Embodied Web Agents: Bridging Physical-Digital Realms for Integrated Agent Intelligence
AI agents today are mostly siloed - they either retrieve and reason over vast amount of digital information and knowledge obtained online; or interact with the physical world through embodied perception, planning and action - but rarely both. This separation limits their ability to solve tasks that require integrated physical and digital intelligence, such as cooking from online recipes, navigating with dynamic map data, or interpreting real-world landmarks using web knowledge. We introduce Embodied Web Agents, a novel paradigm for AI agents that fluidly bridge embodiment and web-scale reasoning. To operationalize this concept, we first develop the Embodied Web Agents task environments, a unified simulation platform that tightly integrates realistic 3D indoor and outdoor environments with functional web interfaces. Building upon this platform, we construct and release the Embodied Web Agents Benchmark, which encompasses a diverse suite of tasks including cooking, navigation, shopping, tourism, and geolocation - all requiring coordinated reasoning across physical and digital realms for systematic assessment of cross-domain intelligence. Experimental results reveal significant performance gaps between state-of-the-art AI systems and human capabilities, establishing both challenges and opportunities at the intersection of embodied cognition and web-scale knowledge access. All datasets, codes and websites are publicly available at our project page https://embodied-web-agent.github.io/.
☆ Sekai: A Video Dataset towards World Exploration
Video generation techniques have made remarkable progress, promising to be the foundation of interactive world exploration. However, existing video generation datasets are not well-suited for world exploration training as they suffer from some limitations: limited locations, short duration, static scenes, and a lack of annotations about exploration and the world. In this paper, we introduce Sekai (meaning ``world'' in Japanese), a high-quality first-person view worldwide video dataset with rich annotations for world exploration. It consists of over 5,000 hours of walking or drone view (FPV and UVA) videos from over 100 countries and regions across 750 cities. We develop an efficient and effective toolbox to collect, pre-process and annotate videos with location, scene, weather, crowd density, captions, and camera trajectories. Experiments demonstrate the quality of the dataset. And, we use a subset to train an interactive video world exploration model, named YUME (meaning ``dream'' in Japanese). We believe Sekai will benefit the area of video generation and world exploration, and motivate valuable applications.
comment: 12 pages, 6 figures
☆ UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting
We address the challenge of relighting a single image or video, a task that demands precise scene intrinsic understanding and high-quality light transport synthesis. Existing end-to-end relighting models are often limited by the scarcity of paired multi-illumination data, restricting their ability to generalize across diverse scenes. Conversely, two-stage pipelines that combine inverse and forward rendering can mitigate data requirements but are susceptible to error accumulation and often fail to produce realistic outputs under complex lighting conditions or with sophisticated materials. In this work, we introduce a general-purpose approach that jointly estimates albedo and synthesizes relit outputs in a single pass, harnessing the generative capabilities of video diffusion models. This joint formulation enhances implicit scene comprehension and facilitates the creation of realistic lighting effects and intricate material interactions, such as shadows, reflections, and transparency. Trained on synthetic multi-illumination data and extensive automatically labeled real-world videos, our model demonstrates strong generalization across diverse domains and surpasses previous methods in both visual fidelity and temporal consistency.
comment: Project page: https://research.nvidia.com/labs/toronto-ai/UniRelight/
☆ Dual-Stage Value-Guided Inference with Margin-Based Reward Adjustment for Fast and Faithful VLM Captioning
Despite significant advances in inference-time search for vision-language models (VLMs), existing approaches remain both computationally expensive and prone to unpenalized, low-confidence generations which often lead to persistent hallucinations. We introduce \textbf{Value-guided Inference with Margin-based Reward (ViMaR)}, a two-stage inference framework that improves both efficiency and output fidelity by combining a temporal-difference value model with a margin-aware reward adjustment. In the first stage, we perform a single pass to identify the highest-value caption among diverse candidates. In the second stage, we selectively refine only those segments that were overlooked or exhibit weak visual grounding, thereby eliminating frequently rewarded evaluations. A calibrated margin-based penalty discourages low-confidence continuations while preserving descriptive richness. Extensive experiments across multiple VLM architectures demonstrate that ViMaR generates captions that are significantly more reliable, factually accurate, detailed, and explanatory, while achieving over 4$\times$ speedup compared to existing value-guided methods. Specifically, we show that ViMaR trained solely on LLaVA Mistral-7B, \textit{generalizes effectively to guide decoding in a stronger unseen model}. To further validate this, we adapt the ViMaR to steer generation in LLaVA-OneVision-Qwen2-7B, leading to consistent improvements in caption quality and demonstrating robust cross-model guidance. This cross-model generalization highlights ViMaR's flexibility and modularity, positioning it as a scalable and transferable inference-time decoding strategy. Furthermore, when ViMaR-generated captions are used for self-training, the underlying models achieve substantial gains across a broad suite of visual comprehension benchmarks, underscoring the potential of fast, accurate, and self-improving VLM pipelines.
☆ Demystifying the Visual Quality Paradox in Multimodal Large Language Models
Recent Multimodal Large Language Models (MLLMs) excel on benchmark vision-language tasks, yet little is known about how input visual quality shapes their responses. Does higher perceptual quality of images already translate to better MLLM understanding? We conduct the first systematic study spanning leading MLLMs and a suite of vision-language benchmarks, applying controlled degradations and stylistic shifts to each image. Surprisingly, we uncover a visual-quality paradox: model, task, and even individual-instance performance can improve when images deviate from human-perceived fidelity. Off-the-shelf restoration pipelines fail to reconcile these idiosyncratic preferences. To close the gap, we introduce Visual-Quality Test-Time Tuning (VQ-TTT)-a lightweight adaptation module that: (1) inserts a learnable, low-rank kernel before the frozen vision encoder to modulate frequency content; and (2) fine-tunes only shallow vision-encoder layers via LoRA. VQ-TTT dynamically adjusts each input image in a single forward pass, aligning it with task-specific model preferences. Across the evaluated MLLMs and all datasets, VQ-TTT lifts significant average accuracy, with no external models, cached features, or extra training data. These findings redefine ``better'' visual inputs for MLLMs and highlight the need for adaptive, rather than universally ``clean'', imagery, in the new era of AI being the main data customer.
comment: 18 pages
☆ FindingDory: A Benchmark to Evaluate Memory in Embodied Agents
Large vision-language models have recently demonstrated impressive performance in planning and control tasks, driving interest in their application to real-world robotics. However, deploying these models for reasoning in embodied contexts is limited by their ability to incorporate long-term experience collected across multiple days and represented by vast collections of images. Current VLMs typically struggle to process more than a few hundred images concurrently, highlighting the need for more efficient mechanisms to handle long-term memory in embodied settings. To effectively evaluate these models for long-horizon control, a benchmark must specifically target scenarios where memory is crucial for success. Existing long-video QA benchmarks overlook embodied challenges like object manipulation and navigation, which demand low-level skills and fine-grained reasoning over past interactions. Moreover, effective memory integration in embodied agents involves both recalling relevant historical information and executing actions based on that information, making it essential to study these aspects together rather than in isolation. In this work, we introduce a new benchmark for long-range embodied tasks in the Habitat simulator. This benchmark evaluates memory-based capabilities across 60 tasks requiring sustained engagement and contextual awareness in an environment. The tasks can also be procedurally extended to longer and more challenging versions, enabling scalable evaluation of memory and reasoning. We also present baselines that integrate state-of-the-art VLMs with low level navigation policies, assessing their performance on these memory-intensive tasks and highlight areas for improvement.
comment: Our dataset and code will be made available at: https://findingdory-benchmark.github.io/
☆ HOIDiNi: Human-Object Interaction through Diffusion Noise Optimization
We present HOIDiNi, a text-driven diffusion framework for synthesizing realistic and plausible human-object interaction (HOI). HOI generation is extremely challenging since it induces strict contact accuracies alongside a diverse motion manifold. While current literature trades off between realism and physical correctness, HOIDiNi optimizes directly in the noise space of a pretrained diffusion model using Diffusion Noise Optimization (DNO), achieving both. This is made feasible thanks to our observation that the problem can be separated into two phases: an object-centric phase, primarily making discrete choices of hand-object contact locations, and a human-centric phase that refines the full-body motion to realize this blueprint. This structured approach allows for precise hand-object contact without compromising motion naturalness. Quantitative, qualitative, and subjective evaluations on the GRAB dataset alone clearly indicate HOIDiNi outperforms prior works and baselines in contact accuracy, physical validity, and overall quality. Our results demonstrate the ability to generate complex, controllable interactions, including grasping, placing, and full-body coordination, driven solely by textual prompts. https://hoidini.github.io.
comment: Project page: https://hoidini.github.io
☆ BoxFusion: Reconstruction-Free Open-Vocabulary 3D Object Detection via Real-Time Multi-View Box Fusion
Open-vocabulary 3D object detection has gained significant interest due to its critical applications in autonomous driving and embodied AI. Existing detection methods, whether offline or online, typically rely on dense point cloud reconstruction, which imposes substantial computational overhead and memory constraints, hindering real-time deployment in downstream tasks. To address this, we propose a novel reconstruction-free online framework tailored for memory-efficient and real-time 3D detection. Specifically, given streaming posed RGB-D video input, we leverage Cubify Anything as a pre-trained visual foundation model (VFM) for single-view 3D object detection by bounding boxes, coupled with CLIP to capture open-vocabulary semantics of detected objects. To fuse all detected bounding boxes across different views into a unified one, we employ an association module for correspondences of multi-views and an optimization module to fuse the 3D bounding boxes of the same instance predicted in multi-views. The association module utilizes 3D Non-Maximum Suppression (NMS) and a box correspondence matching module, while the optimization module uses an IoU-guided efficient random optimization technique based on particle filtering to enforce multi-view consistency of the 3D bounding boxes while minimizing computational complexity. Extensive experiments on ScanNetV2 and CA-1M datasets demonstrate that our method achieves state-of-the-art performance among online methods. Benefiting from this novel reconstruction-free paradigm for 3D object detection, our method exhibits great generalization abilities in various scenarios, enabling real-time perception even in environments exceeding 1000 square meters.
comment: 11 pages, 6 figures
☆ Mono-Modalizing Extremely Heterogeneous Multi-Modal Medical Image Registration MICCAI
In clinical practice, imaging modalities with functional characteristics, such as positron emission tomography (PET) and fractional anisotropy (FA), are often aligned with a structural reference (e.g., MRI, CT) for accurate interpretation or group analysis, necessitating multi-modal deformable image registration (DIR). However, due to the extreme heterogeneity of these modalities compared to standard structural scans, conventional unsupervised DIR methods struggle to learn reliable spatial mappings and often distort images. We find that the similarity metrics guiding these models fail to capture alignment between highly disparate modalities. To address this, we propose M2M-Reg (Multi-to-Mono Registration), a novel framework that trains multi-modal DIR models using only mono-modal similarity while preserving the established architectural paradigm for seamless integration into existing models. We also introduce GradCyCon, a regularizer that leverages M2M-Reg's cyclic training scheme to promote diffeomorphism. Furthermore, our framework naturally extends to a semi-supervised setting, integrating pre-aligned and unaligned pairs only, without requiring ground-truth transformations or segmentation masks. Experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that M2M-Reg achieves up to 2x higher DSC than prior methods for PET-MRI and FA-MRI registration, highlighting its effectiveness in handling highly heterogeneous multi-modal DIR. Our code is available at https://github.com/MICV-yonsei/M2M-Reg.
comment: 11 pages, 3 figures, 2 tables, Accepted at Medical Image Computing and Computer Assisted Intervention (MICCAI) 2025
☆ One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution
It is a challenging problem to reproduce rich spatial details while maintaining temporal consistency in real-world video super-resolution (Real-VSR), especially when we leverage pre-trained generative models such as stable diffusion (SD) for realistic details synthesis. Existing SD-based Real-VSR methods often compromise spatial details for temporal coherence, resulting in suboptimal visual quality. We argue that the key lies in how to effectively extract the degradation-robust temporal consistency priors from the low-quality (LQ) input video and enhance the video details while maintaining the extracted consistency priors. To achieve this, we propose a Dual LoRA Learning (DLoRAL) paradigm to train an effective SD-based one-step diffusion model, achieving realistic frame details and temporal consistency simultaneously. Specifically, we introduce a Cross-Frame Retrieval (CFR) module to aggregate complementary information across frames, and train a Consistency-LoRA (C-LoRA) to learn robust temporal representations from degraded inputs. After consistency learning, we fix the CFR and C-LoRA modules and train a Detail-LoRA (D-LoRA) to enhance spatial details while aligning with the temporal space defined by C-LoRA to keep temporal coherence. The two phases alternate iteratively for optimization, collaboratively delivering consistent and detail-rich outputs. During inference, the two LoRA branches are merged into the SD model, allowing efficient and high-quality video restoration in a single diffusion step. Experiments show that DLoRAL achieves strong performance in both accuracy and speed. Code and models are available at https://github.com/yjsunnn/DLoRAL.
☆ A Unified Graph-based Framework for Scalable 3D Tree Reconstruction and Non-Destructive Biomass Estimation from Point Clouds
Estimating forest above-ground biomass (AGB) is crucial for assessing carbon storage and supporting sustainable forest management. Quantitative Structural Model (QSM) offers a non-destructive approach to AGB estimation through 3D tree structural reconstruction. However, current QSM methods face significant limitations, as they are primarily designed for individual trees,depend on high-quality point cloud data from terrestrial laser scanning (TLS), and also require multiple pre-processing steps that hinder scalability and practical deployment. This study presents a novel unified framework that enables end-to-end processing of large-scale point clouds using an innovative graph-based pipeline. The proposed approach seamlessly integrates tree segmentation,leaf-wood separation and 3D skeletal reconstruction through dedicated graph operations including pathing and abstracting for tree topology reasoning. Comprehensive validation was conducted on datasets with varying leaf conditions (leaf-on and leaf-off), spatial scales (tree- and plot-level), and data sources (TLS and UAV-based laser scanning, ULS). Experimental results demonstrate strong performance under challenging conditions, particularly in leaf-on scenarios (~20% relative error) and low-density ULS datasets with partial coverage (~30% relative error). These findings indicate that the proposed framework provides a robust and scalable solution for large-scale, non-destructive AGB estimation. It significantly reduces dependency on specialized pre-processing tools and establishes ULS as a viable alternative to TLS. To our knowledge, this is the first method capable of enabling seamless, end-to-end 3D tree reconstruction at operational scales. This advancement substantially improves the feasibility of QSM-based AGB estimation, paving the way for broader applications in forest inventory and climate change research.
comment: 17 pages,19 figures
☆ Baltimore Atlas: FreqWeaver Adapter for Semi-supervised Ultra-high Spatial Resolution Land Cover Classification
Ultra-high Spatial Resolution Land Cover Classification is essential for fine-grained land cover analysis, yet it remains challenging due to the high cost of pixel-level annotations, significant scale variation, and the limited adaptability of large-scale vision models. Existing methods typically focus on 1-meter spatial resolution imagery and rely heavily on annotated data, whereas practical applications often require processing higher-resolution imagery under weak supervision. To address this, we propose a parameter-efficient semi-supervised segmentation framework for 0.3 m spatial resolution imagery, which leverages the knowledge of SAM2 and introduces a remote sensing-specific FreqWeaver Adapter to enhance fine-grained detail modeling while maintaining a lightweight design at only 5.96% of the total model parameters. By effectively leveraging unlabeled data and maintaining minimal parameter overhead, the proposed method delivers robust segmentation results with superior structural consistency, achieving a 1.78% improvement over existing parameter-efficient tuning strategies and a 3.44% gain compared to state-of-the-art high-resolution remote sensing segmentation approaches.
☆ Show-o2: Improved Native Unified Multimodal Models
This paper presents improved native unified multimodal models, \emph{i.e.,} Show-o2, that leverage autoregressive modeling and flow matching. Built upon a 3D causal variational autoencoder space, unified visual representations are constructed through a dual-path of spatial (-temporal) fusion, enabling scalability across image and video modalities while ensuring effective multimodal understanding and generation. Based on a language model, autoregressive modeling and flow matching are natively applied to the language head and flow head, respectively, to facilitate text token prediction and image/video generation. A two-stage training recipe is designed to effectively learn and scale to larger models. The resulting Show-o2 models demonstrate versatility in handling a wide range of multimodal understanding and generation tasks across diverse modalities, including text, images, and videos. Code and models are released at https://github.com/showlab/Show-o.
comment: Technical report
☆ Control and Realism: Best of Both Worlds in Layout-to-Image without Training ICML2025
Layout-to-Image generation aims to create complex scenes with precise control over the placement and arrangement of subjects. Existing works have demonstrated that pre-trained Text-to-Image diffusion models can achieve this goal without training on any specific data; however, they often face challenges with imprecise localization and unrealistic artifacts. Focusing on these drawbacks, we propose a novel training-free method, WinWinLay. At its core, WinWinLay presents two key strategies, Non-local Attention Energy Function and Adaptive Update, that collaboratively enhance control precision and realism. On one hand, we theoretically demonstrate that the commonly used attention energy function introduces inherent spatial distribution biases, hindering objects from being uniformly aligned with layout instructions. To overcome this issue, non-local attention prior is explored to redistribute attention scores, facilitating objects to better conform to the specified spatial conditions. On the other hand, we identify that the vanilla backpropagation update rule can cause deviations from the pre-trained domain, leading to out-of-distribution artifacts. We accordingly introduce a Langevin dynamics-based adaptive update scheme as a remedy that promotes in-domain updating while respecting layout constraints. Extensive experiments demonstrate that WinWinLay excels in controlling element placement and achieving photorealistic visual fidelity, outperforming the current state-of-the-art methods.
comment: Accepted by ICML2025
☆ Automated MRI Tumor Segmentation using hybrid U-Net with Transformer and Efficient Attention
Cancer is an abnormal growth with potential to invade locally and metastasize to distant organs. Accurate auto-segmentation of the tumor and surrounding normal tissues is required for radiotherapy treatment plan optimization. Recent AI-based segmentation models are generally trained on large public datasets, which lack the heterogeneity of local patient populations. While these studies advance AI-based medical image segmentation, research on local datasets is necessary to develop and integrate AI tumor segmentation models directly into hospital software for efficient and accurate oncology treatment planning and execution. This study enhances tumor segmentation using computationally efficient hybrid UNet-Transformer models on magnetic resonance imaging (MRI) datasets acquired from a local hospital under strict privacy protection. We developed a robust data pipeline for seamless DICOM extraction and preprocessing, followed by extensive image augmentation to ensure model generalization across diverse clinical settings, resulting in a total dataset of 6080 images for training. Our novel architecture integrates UNet-based convolutional neural networks with a transformer bottleneck and complementary attention modules, including efficient attention, Squeeze-and-Excitation (SE) blocks, Convolutional Block Attention Module (CBAM), and ResNeXt blocks. To accelerate convergence and reduce computational demands, we used a maximum batch size of 8 and initialized the encoder with pretrained ImageNet weights, training the model on dual NVIDIA T4 GPUs via checkpointing to overcome Kaggle's runtime limits. Quantitative evaluation on the local MRI dataset yielded a Dice similarity coefficient of 0.764 and an Intersection over Union (IoU) of 0.736, demonstrating competitive performance despite limited data and underscoring the importance of site-specific model development for clinical deployment.
comment: 16 pages, 5 figures
☆ RaCalNet: Radar Calibration Network for Sparse-Supervised Metric Depth Estimation
Dense metric depth estimation using millimeter-wave radar typically requires dense LiDAR supervision, generated via multi-frame projection and interpolation, to guide the learning of accurate depth from sparse radar measurements and RGB images. However, this paradigm is both costly and data-intensive. To address this, we propose RaCalNet, a novel framework that eliminates the need for dense supervision by using sparse LiDAR to supervise the learning of refined radar measurements, resulting in a supervision density of merely around 1% compared to dense-supervised methods. Unlike previous approaches that associate radar points with broad image regions and rely heavily on dense labels, RaCalNet first recalibrates and refines sparse radar points to construct accurate depth priors. These priors then serve as reliable anchors to guide monocular depth prediction, enabling metric-scale estimation without resorting to dense supervision. This design improves structural consistency and preserves fine details. Despite relying solely on sparse supervision, RaCalNet surpasses state-of-the-art dense-supervised methods, producing depth maps with clear object contours and fine-grained textures. Extensive experiments on the ZJU-4DRadarCam dataset and real-world deployment scenarios demonstrate its effectiveness, reducing RMSE by 35.30% and 34.89%, respectively.
comment: 9 pages, 7 figures
☆ CLAIM: Clinically-Guided LGE Augmentation for Realistic and Diverse Myocardial Scar Synthesis and Segmentation
Deep learning-based myocardial scar segmentation from late gadolinium enhancement (LGE) cardiac MRI has shown great potential for accurate and timely diagnosis and treatment planning for structural cardiac diseases. However, the limited availability and variability of LGE images with high-quality scar labels restrict the development of robust segmentation models. To address this, we introduce CLAIM: \textbf{C}linically-Guided \textbf{L}GE \textbf{A}ugmentation for Real\textbf{i}stic and Diverse \textbf{M}yocardial Scar Synthesis and Segmentation framework, a framework for anatomically grounded scar generation and segmentation. At its core is the SMILE module (Scar Mask generation guided by cLinical knowledgE), which conditions a diffusion-based generator on the clinically adopted AHA 17-segment model to synthesize images with anatomically consistent and spatially diverse scar patterns. In addition, CLAIM employs a joint training strategy in which the scar segmentation network is optimized alongside the generator, aiming to enhance both the realism of synthesized scars and the accuracy of the scar segmentation performance. Experimental results show that CLAIM produces anatomically coherent scar patterns and achieves higher Dice similarity with real scar distributions compared to baseline models. Our approach enables controllable and realistic myocardial scar synthesis and has demonstrated utility for downstream medical imaging task.
comment: 14 Pages
☆ NTIRE 2025 Image Shadow Removal Challenge Report
This work examines the findings of the NTIRE 2025 Shadow Removal Challenge. A total of 306 participants have registered, with 17 teams successfully submitting their solutions during the final evaluation phase. Following the last two editions, this challenge had two evaluation tracks: one focusing on reconstruction fidelity and the other on visual perception through a user study. Both tracks were evaluated with images from the WSRD+ dataset, simulating interactions between self- and cast-shadows with a large number of diverse objects, textures, and materials.
☆ Pixel-level Certified Explanations via Randomized Smoothing
Post-hoc attribution methods aim to explain deep learning predictions by highlighting influential input pixels. However, these explanations are highly non-robust: small, imperceptible input perturbations can drastically alter the attribution map while maintaining the same prediction. This vulnerability undermines their trustworthiness and calls for rigorous robustness guarantees of pixel-level attribution scores. We introduce the first certification framework that guarantees pixel-level robustness for any black-box attribution method using randomized smoothing. By sparsifying and smoothing attribution maps, we reformulate the task as a segmentation problem and certify each pixel's importance against $\ell_2$-bounded perturbations. We further propose three evaluation metrics to assess certified robustness, localization, and faithfulness. An extensive evaluation of 12 attribution methods across 5 ImageNet models shows that our certified attributions are robust, interpretable, and faithful, enabling reliable use in downstream tasks. Our code is at https://github.com/AlaaAnani/certified-attributions.
☆ Advanced cervical cancer classification: enhancing pap smear images with hybrid PMD Filter-CLAHE
Cervical cancer remains a significant health problem, especially in developing countries. Early detection is critical for effective treatment. Convolutional neural networks (CNN) have shown promise in automated cervical cancer screening, but their performance depends on Pap smear image quality. This study investigates the impact of various image preprocessing techniques on CNN performance for cervical cancer classification using the SIPaKMeD dataset. Three preprocessing techniques were evaluated: perona-malik diffusion (PMD) filter for noise reduction, contrast-limited adaptive histogram equalization (CLAHE) for image contrast enhancement, and the proposed hybrid PMD filter-CLAHE approach. The enhanced image datasets were evaluated on pretrained models, such as ResNet-34, ResNet-50, SqueezeNet-1.0, MobileNet-V2, EfficientNet-B0, EfficientNet-B1, DenseNet-121, and DenseNet-201. The results show that hybrid preprocessing PMD filter-CLAHE can improve the Pap smear image quality and CNN architecture performance compared to the original images. The maximum metric improvements are 13.62% for accuracy, 10.04% for precision, 13.08% for recall, and 14.34% for F1-score. The proposed hybrid PMD filter-CLAHE technique offers a new perspective in improving cervical cancer classification performance using CNN architectures.
☆ GenHOI: Generalizing Text-driven 4D Human-Object Interaction Synthesis for Unseen Objects
While diffusion models and large-scale motion datasets have advanced text-driven human motion synthesis, extending these advances to 4D human-object interaction (HOI) remains challenging, mainly due to the limited availability of large-scale 4D HOI datasets. In our study, we introduce GenHOI, a novel two-stage framework aimed at achieving two key objectives: 1) generalization to unseen objects and 2) the synthesis of high-fidelity 4D HOI sequences. In the initial stage of our framework, we employ an Object-AnchorNet to reconstruct sparse 3D HOI keyframes for unseen objects, learning solely from 3D HOI datasets, thereby mitigating the dependence on large-scale 4D HOI datasets. Subsequently, we introduce a Contact-Aware Diffusion Model (ContactDM) in the second stage to seamlessly interpolate sparse 3D HOI keyframes into densely temporally coherent 4D HOI sequences. To enhance the quality of generated 4D HOI sequences, we propose a novel Contact-Aware Encoder within ContactDM to extract human-object contact patterns and a novel Contact-Aware HOI Attention to effectively integrate the contact signals into diffusion models. Experimental results show that we achieve state-of-the-art results on the publicly available OMOMO and 3D-FUTURE datasets, demonstrating strong generalization abilities to unseen objects, while enabling high-fidelity 4D HOI generation.
☆ Multimodal Large Language Models for Medical Report Generation via Customized Prompt Tuning
Medical report generation from imaging data remains a challenging task in clinical practice. While large language models (LLMs) show great promise in addressing this challenge, their effective integration with medical imaging data still deserves in-depth exploration. In this paper, we present MRG-LLM, a novel multimodal large language model (MLLM) that combines a frozen LLM with a learnable visual encoder and introduces a dynamic prompt customization mechanism. Our key innovation lies in generating instance-specific prompts tailored to individual medical images through conditional affine transformations derived from visual features. We propose two implementations: prompt-wise and promptbook-wise customization, enabling precise and targeted report generation. Extensive experiments on IU X-ray and MIMIC-CXR datasets demonstrate that MRG-LLM achieves state-of-the-art performance in medical report generation. Our code will be made publicly available.
☆ Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material
3D AI-generated content (AIGC) is a passionate field that has significantly accelerated the creation of 3D models in gaming, film, and design. Despite the development of several groundbreaking models that have revolutionized 3D generation, the field remains largely accessible only to researchers, developers, and designers due to the complexities involved in collecting, processing, and training 3D models. To address these challenges, we introduce Hunyuan3D 2.1 as a case study in this tutorial. This tutorial offers a comprehensive, step-by-step guide on processing 3D data, training a 3D generative model, and evaluating its performance using Hunyuan3D 2.1, an advanced system for producing high-resolution, textured 3D assets. The system comprises two core components: the Hunyuan3D-DiT for shape generation and the Hunyuan3D-Paint for texture synthesis. We will explore the entire workflow, including data preparation, model architecture, training strategies, evaluation metrics, and deployment. By the conclusion of this tutorial, you will have the knowledge to finetune or develop a robust 3D generative model suitable for applications in gaming, virtual reality, and industrial design.
comment: Github link: https://github.com/Tencent-Hunyuan/Hunyuan3D-2.1
☆ NERO: Explainable Out-of-Distribution Detection with Neuron-level Relevance
Ensuring reliability is paramount in deep learning, particularly within the domain of medical imaging, where diagnostic decisions often hinge on model outputs. The capacity to separate out-of-distribution (OOD) samples has proven to be a valuable indicator of a model's reliability in research. In medical imaging, this is especially critical, as identifying OOD inputs can help flag potential anomalies that might otherwise go undetected. While many OOD detection methods rely on feature or logit space representations, recent works suggest these approaches may not fully capture OOD diversity. To address this, we propose a novel OOD scoring mechanism, called NERO, that leverages neuron-level relevance at the feature layer. Specifically, we cluster neuron-level relevance for each in-distribution (ID) class to form representative centroids and introduce a relevance distance metric to quantify a new sample's deviation from these centroids, enhancing OOD separability. Additionally, we refine performance by incorporating scaled relevance in the bias term and combining feature norms. Our framework also enables explainable OOD detection. We validate its effectiveness across multiple deep learning architectures on the gastrointestinal imaging benchmarks Kvasir and GastroVision, achieving improvements over state-of-the-art OOD detection methods.
☆ MCOO-SLAM: A Multi-Camera Omnidirectional Object SLAM System
Object-level SLAM offers structured and semantically meaningful environment representations, making it more interpretable and suitable for high-level robotic tasks. However, most existing approaches rely on RGB-D sensors or monocular views, which suffer from narrow fields of view, occlusion sensitivity, and limited depth perception-especially in large-scale or outdoor environments. These limitations often restrict the system to observing only partial views of objects from limited perspectives, leading to inaccurate object modeling and unreliable data association. In this work, we propose MCOO-SLAM, a novel Multi-Camera Omnidirectional Object SLAM system that fully leverages surround-view camera configurations to achieve robust, consistent, and semantically enriched mapping in complex outdoor scenarios. Our approach integrates point features and object-level landmarks enhanced with open-vocabulary semantics. A semantic-geometric-temporal fusion strategy is introduced for robust object association across multiple views, leading to improved consistency and accurate object modeling, and an omnidirectional loop closure module is designed to enable viewpoint-invariant place recognition using scene-level descriptors. Furthermore, the constructed map is abstracted into a hierarchical 3D scene graph to support downstream reasoning tasks. Extensive experiments in real-world demonstrate that MCOO-SLAM achieves accurate localization and scalable object-level mapping with improved robustness to occlusion, pose variation, and environmental complexity.
☆ A Real-time Endoscopic Image Denoising System
Endoscopes featuring a miniaturized design have significantly enhanced operational flexibility, portability, and diagnostic capability while substantially reducing the invasiveness of medical procedures. Recently, single-use endoscopes equipped with an ultra-compact analogue image sensor measuring less than 1mm x 1mm bring revolutionary advancements to medical diagnosis. They reduce the structural redundancy and large capital expenditures associated with reusable devices, eliminate the risk of patient infections caused by inadequate disinfection, and alleviate patient suffering. However, the limited photosensitive area results in reduced photon capture per pixel, requiring higher photon sensitivity settings to maintain adequate brightness. In high-contrast medical imaging scenarios, the small-sized sensor exhibits a constrained dynamic range, making it difficult to simultaneously capture details in both highlights and shadows, and additional localized digital gain is required to compensate. Moreover, the simplified circuit design and analog signal transmission introduce additional noise sources. These factors collectively contribute to significant noise issues in processed endoscopic images. In this work, we developed a comprehensive noise model for analog image sensors in medical endoscopes, addressing three primary noise types: fixed-pattern noise, periodic banding noise, and mixed Poisson-Gaussian noise. Building on this analysis, we propose a hybrid denoising system that synergistically combines traditional image processing algorithms with advanced learning-based techniques for captured raw frames from sensors. Experiments demonstrate that our approach effectively reduces image noise without fine detail loss or color distortion, while achieving real-time performance on FPGA platforms and an average PSNR improvement from 21.16 to 33.05 on our test dataset.
☆ When Model Knowledge meets Diffusion Model: Diffusion-assisted Data-free Image Synthesis with Alignment of Domain and Class ICML 2025
Open-source pre-trained models hold great potential for diverse applications, but their utility declines when their training data is unavailable. Data-Free Image Synthesis (DFIS) aims to generate images that approximate the learned data distribution of a pre-trained model without accessing the original data. However, existing DFIS meth ods produce samples that deviate from the training data distribution due to the lack of prior knowl edge about natural images. To overcome this limitation, we propose DDIS, the first Diffusion-assisted Data-free Image Synthesis method that leverages a text-to-image diffusion model as a powerful image prior, improving synthetic image quality. DDIS extracts knowledge about the learned distribution from the given model and uses it to guide the diffusion model, enabling the generation of images that accurately align with the training data distribution. To achieve this, we introduce Domain Alignment Guidance (DAG) that aligns the synthetic data domain with the training data domain during the diffusion sampling process. Furthermore, we optimize a single Class Alignment Token (CAT) embedding to effectively capture class-specific attributes in the training dataset. Experiments on PACS and Ima geNet demonstrate that DDIS outperforms prior DFIS methods by generating samples that better reflect the training data distribution, achieving SOTA performance in data-free applications.
comment: Published at ICML 2025
☆ Unsupervised Pelage Pattern Unwrapping for Animal Re-identification
Existing individual re-identification methods often struggle with the deformable nature of animal fur or skin patterns which undergo geometric distortions due to body movement and posture changes. In this paper, we propose a geometry-aware texture mapping approach that unwarps pelage patterns, the unique markings found on an animal's skin or fur, into a canonical UV space, enabling more robust feature matching. Our method uses surface normal estimation to guide the unwrapping process while preserving the geometric consistency between the 3D surface and the 2D texture space. We focus on two challenging species: Saimaa ringed seals (Pusa hispida saimensis) and leopards (Panthera pardus). Both species have distinctive yet highly deformable fur patterns. By integrating our pattern-preserving UV mapping with existing re-identification techniques, we demonstrate improved accuracy across diverse poses and viewing angles. Our framework does not require ground truth UV annotations and can be trained in a self-supervised manner. Experiments on seal and leopard datasets show up to a 5.4% improvement in re-identification accuracy.
☆ Open-World Object Counting in Videos
We introduce a new task of open-world object counting in videos: given a text description, or an image example, that specifies the target object, the objective is to enumerate all the unique instances of the target objects in the video. This task is especially challenging in crowded scenes with occlusions and similar objects, where avoiding double counting and identifying reappearances is crucial. To this end, we make the following contributions: we introduce a model, CountVid, for this task. It leverages an image-based counting model, and a promptable video segmentation and tracking model to enable automated, open-world object counting across video frames. To evaluate its performance, we introduce VideoCount, a new dataset for our novel task built from the TAO and MOT20 tracking datasets, as well as from videos of penguins and metal alloy crystallization captured by x-rays. Using this dataset, we demonstrate that CountVid provides accurate object counts, and significantly outperforms strong baselines. The VideoCount dataset, the CountVid model, and all the code are available at https://github.com/niki-amini-naieni/CountVid/.
☆ FedWSIDD: Federated Whole Slide Image Classification via Dataset Distillation MICCAI 2025
Federated learning (FL) has emerged as a promising approach for collaborative medical image analysis, enabling multiple institutions to build robust predictive models while preserving sensitive patient data. In the context of Whole Slide Image (WSI) classification, FL faces significant challenges, including heterogeneous computational resources across participating medical institutes and privacy concerns. To address these challenges, we propose FedWSIDD, a novel FL paradigm that leverages dataset distillation (DD) to learn and transmit synthetic slides. On the server side, FedWSIDD aggregates synthetic slides from participating centres and distributes them across all centres. On the client side, we introduce a novel DD algorithm tailored to histopathology datasets which incorporates stain normalisation into the distillation process to generate a compact set of highly informative synthetic slides. These synthetic slides, rather than model parameters, are transmitted to the server. After communication, the received synthetic slides are combined with original slides for local tasks. Extensive experiments on multiple WSI classification tasks, including CAMELYON16 and CAMELYON17, demonstrate that FedWSIDD offers flexibility for heterogeneous local models, enhances local WSI classification performance, and preserves patient privacy. This makes it a highly effective solution for complex WSI classification tasks. The code is available at FedWSIDD.
comment: MICCAI 2025
☆ OpenPath: Open-Set Active Learning for Pathology Image Classification via Pre-trained Vision-Language Models MICCAI 2025
Pathology image classification plays a crucial role in accurate medical diagnosis and treatment planning. Training high-performance models for this task typically requires large-scale annotated datasets, which are both expensive and time-consuming to acquire. Active Learning (AL) offers a solution by iteratively selecting the most informative samples for annotation, thereby reducing the labeling effort. However, most AL methods are designed under the assumption of a closed-set scenario, where all the unannotated images belong to target classes. In real-world clinical environments, the unlabeled pool often contains a substantial amount of Out-Of-Distribution (OOD) data, leading to low efficiency of annotation in traditional AL methods. Furthermore, most existing AL methods start with random selection in the first query round, leading to a significant waste of labeling costs in open-set scenarios. To address these challenges, we propose OpenPath, a novel open-set active learning approach for pathological image classification leveraging a pre-trained Vision-Language Model (VLM). In the first query, we propose task-specific prompts that combine target and relevant non-target class prompts to effectively select In-Distribution (ID) and informative samples from the unlabeled pool. In subsequent queries, Diverse Informative ID Sampling (DIS) that includes Prototype-based ID candidate Selection (PIS) and Entropy-Guided Stochastic Sampling (EGSS) is proposed to ensure both purity and informativeness in a query, avoiding the selection of OOD samples. Experiments on two public pathology image datasets show that OpenPath significantly enhances the model's performance due to its high purity of selected samples, and outperforms several state-of-the-art open-set AL methods. The code is available at \href{https://github.com/HiLab-git/OpenPath}{https://github.com/HiLab-git/OpenPath}..
comment: MICCAI 2025 early accept
☆ MapFM: Foundation Model-Driven HD Mapping with Multi-Task Contextual Learning
In autonomous driving, high-definition (HD) maps and semantic maps in bird's-eye view (BEV) are essential for accurate localization, planning, and decision-making. This paper introduces an enhanced End-to-End model named MapFM for online vectorized HD map generation. We show significantly boost feature representation quality by incorporating powerful foundation model for encoding camera images. To further enrich the model's understanding of the environment and improve prediction quality, we integrate auxiliary prediction heads for semantic segmentation in the BEV representation. This multi-task learning approach provides richer contextual supervision, leading to a more comprehensive scene representation and ultimately resulting in higher accuracy and improved quality of the predicted vectorized HD maps. The source code is available at https://github.com/LIvanoff/MapFM.
comment: Preprint. Submitted. 12 pages, 4 figures
☆ One-shot Face Sketch Synthesis in the Wild via Generative Diffusion Prior and Instruction Tuning
Face sketch synthesis is a technique aimed at converting face photos into sketches. Existing face sketch synthesis research mainly relies on training with numerous photo-sketch sample pairs from existing datasets. However, these large-scale discriminative learning methods will have to face problems such as data scarcity and high human labor costs. Once the training data becomes scarce, their generative performance significantly degrades. In this paper, we propose a one-shot face sketch synthesis method based on diffusion models. We optimize text instructions on a diffusion model using face photo-sketch image pairs. Then, the instructions derived through gradient-based optimization are used for inference. To simulate real-world scenarios more accurately and evaluate method effectiveness more comprehensively, we introduce a new benchmark named One-shot Face Sketch Dataset (OS-Sketch). The benchmark consists of 400 pairs of face photo-sketch images, including sketches with different styles and photos with different backgrounds, ages, sexes, expressions, illumination, etc. For a solid out-of-distribution evaluation, we select only one pair of images for training at each time, with the rest used for inference. Extensive experiments demonstrate that the proposed method can convert various photos into realistic and highly consistent sketches in a one-shot context. Compared to other methods, our approach offers greater convenience and broader applicability. The dataset will be available at: https://github.com/HanWu3125/OS-Sketch
comment: We propose a novel framework for face sketch synthesis, where merely a single pair of samples suffices to enable in-the-wild face sketch synthesis
☆ MEGC2025: Micro-Expression Grand Challenge on Spot Then Recognize and Visual Question Answering ACM MM 2025
Facial micro-expressions (MEs) are involuntary movements of the face that occur spontaneously when a person experiences an emotion but attempts to suppress or repress the facial expression, typically found in a high-stakes environment. In recent years, substantial advancements have been made in the areas of ME recognition, spotting, and generation. However, conventional approaches that treat spotting and recognition as separate tasks are suboptimal, particularly for analyzing long-duration videos in realistic settings. Concurrently, the emergence of multimodal large language models (MLLMs) and large vision-language models (LVLMs) offers promising new avenues for enhancing ME analysis through their powerful multimodal reasoning capabilities. The ME grand challenge (MEGC) 2025 introduces two tasks that reflect these evolving research directions: (1) ME spot-then-recognize (ME-STR), which integrates ME spotting and subsequent recognition in a unified sequential pipeline; and (2) ME visual question answering (ME-VQA), which explores ME understanding through visual question answering, leveraging MLLMs or LVLMs to address diverse question types related to MEs. All participating algorithms are required to run on this test set and submit their results on a leaderboard. More details are available at https://megc2025.github.io.
comment: Micro-Expression Grand Challenge (MEGC) at ACM MM 2025
☆ Human Motion Capture from Loose and Sparse Inertial Sensors with Garment-aware Diffusion Models IJCAI 2025
Motion capture using sparse inertial sensors has shown great promise due to its portability and lack of occlusion issues compared to camera-based tracking. Existing approaches typically assume that IMU sensors are tightly attached to the human body. However, this assumption often does not hold in real-world scenarios. In this paper, we present a new task of full-body human pose estimation using sparse, loosely attached IMU sensors. To solve this task, we simulate IMU recordings from an existing garment-aware human motion dataset. We developed transformer-based diffusion models to synthesize loose IMU data and estimate human poses based on this challenging loose IMU data. In addition, we show that incorporating garment-related parameters while training the model on simulated loose data effectively maintains expressiveness and enhances the ability to capture variations introduced by looser or tighter garments. Experiments show that our proposed diffusion methods trained on simulated and synthetic data outperformed the state-of-the-art methods quantitatively and qualitatively, opening up a promising direction for future research.
comment: Accepted by IJCAI 2025
☆ AI-driven visual monitoring of industrial assembly tasks
Visual monitoring of industrial assembly tasks is critical for preventing equipment damage due to procedural errors and ensuring worker safety. Although commercial solutions exist, they typically require rigid workspace setups or the application of visual markers to simplify the problem. We introduce ViMAT, a novel AI-driven system for real-time visual monitoring of assembly tasks that operates without these constraints. ViMAT combines a perception module that extracts visual observations from multi-view video streams with a reasoning module that infers the most likely action being performed based on the observed assembly state and prior task knowledge. We validate ViMAT on two assembly tasks, involving the replacement of LEGO components and the reconfiguration of hydraulic press molds, demonstrating its effectiveness through quantitative and qualitative analysis in challenging real-world scenarios characterized by partial and uncertain visual observations. Project page: https://tev-fbk.github.io/ViMAT
☆ BCRNet: Enhancing Landmark Detection in Laparoscopic Liver Surgery via Bezier Curve Refinement MICCAI 2025
Laparoscopic liver surgery, while minimally invasive, poses significant challenges in accurately identifying critical anatomical structures. Augmented reality (AR) systems, integrating MRI/CT with laparoscopic images based on 2D-3D registration, offer a promising solution for enhancing surgical navigation. A vital aspect of the registration progress is the precise detection of curvilinear anatomical landmarks in laparoscopic images. In this paper, we propose BCRNet (Bezier Curve Refinement Net), a novel framework that significantly enhances landmark detection in laparoscopic liver surgery primarily via the Bezier curve refinement strategy. The framework starts with a Multi-modal Feature Extraction (MFE) module designed to robustly capture semantic features. Then we propose Adaptive Curve Proposal Initialization (ACPI) to generate pixel-aligned Bezier curves and confidence scores for reliable initial proposals. Additionally, we design the Hierarchical Curve Refinement (HCR) mechanism to enhance these proposals iteratively through a multi-stage process, capturing fine-grained contextual details from multi-scale pixel-level features for precise Bezier curve adjustment. Extensive evaluations on the L3D and P2ILF datasets demonstrate that BCRNet outperforms state-of-the-art methods, achieving significant performance improvements. Code will be available.
comment: Accepted at MICCAI 2025, 11 pages, 2 figures
☆ MSNeRV: Neural Video Representation with Multi-Scale Feature Fusion
Implicit Neural representations (INRs) have emerged as a promising approach for video compression, and have achieved comparable performance to the state-of-the-art codecs such as H.266/VVC. However, existing INR-based methods struggle to effectively represent detail-intensive and fast-changing video content. This limitation mainly stems from the underutilization of internal network features and the absence of video-specific considerations in network design. To address these challenges, we propose a multi-scale feature fusion framework, MSNeRV, for neural video representation. In the encoding stage, we enhance temporal consistency by employing temporal windows, and divide the video into multiple Groups of Pictures (GoPs), where a GoP-level grid is used for background representation. Additionally, we design a multi-scale spatial decoder with a scale-adaptive loss function to integrate multi-resolution and multi-frequency information. To further improve feature extraction, we introduce a multi-scale feature block that fully leverages hidden features. We evaluate MSNeRV on HEVC ClassB and UVG datasets for video representation and compression. Experimental results demonstrate that our model exhibits superior representation capability among INR-based approaches and surpasses VTM-23.7 (Random Access) in dynamic scenarios in terms of compression efficiency.
☆ Domain Adaptation for Image Classification of Defects in Semiconductor Manufacturing
In the semiconductor sector, due to high demand but also strong and increasing competition, time to market and quality are key factors in securing significant market share in various application areas. Thanks to the success of deep learning methods in recent years in the computer vision domain, Industry 4.0 and 5.0 applications, such as defect classification, have achieved remarkable success. In particular, Domain Adaptation (DA) has proven highly effective since it focuses on using the knowledge learned on a (source) domain to adapt and perform effectively on a different but related (target) domain. By improving robustness and scalability, DA minimizes the need for extensive manual re-labeling or re-training of models. This not only reduces computational and resource costs but also allows human experts to focus on high-value tasks. Therefore, we tested the efficacy of DA techniques in semi-supervised and unsupervised settings within the context of the semiconductor field. Moreover, we propose the DBACS approach, a CycleGAN-inspired model enhanced with additional loss terms to improve performance. All the approaches are studied and validated on real-world Electron Microscope images considering the unsupervised and semi-supervised settings, proving the usefulness of our method in advancing DA techniques for the semiconductor field.
☆ Privacy-Preserving Chest X-ray Classification in Latent Space with Homomorphically Encrypted Neural Inference
Medical imaging data contain sensitive patient information requiring strong privacy protection. Many analytical setups require data to be sent to a server for inference purposes. Homomorphic encryption (HE) provides a solution by allowing computations to be performed on encrypted data without revealing the original information. However, HE inference is computationally expensive, particularly for large images (e.g., chest X-rays). In this study, we propose an HE inference framework for medical images that uses VQGAN to compress images into latent representations, thereby significantly reducing the computational burden while preserving image quality. We approximate the activation functions with lower-degree polynomials to balance the accuracy and efficiency in compliance with HE requirements. We observed that a downsampling factor of eight for compression achieved an optimal balance between performance and computational cost. We further adapted the squeeze and excitation module, which is known to improve traditional CNNs, to enhance the HE framework. Our method was tested on two chest X-ray datasets for multi-label classification tasks using vanilla CNN backbones. Although HE inference remains relatively slow and introduces minor performance differences compared with unencrypted inference, our approach shows strong potential for practical use in medical images
comment: 11 pages, 5 figures
☆ Retrospective Memory for Camouflaged Object Detection
Camouflaged object detection (COD) primarily focuses on learning subtle yet discriminative representations from complex scenes. Existing methods predominantly follow the parametric feedforward architecture based on static visual representation modeling. However, they lack explicit mechanisms for acquiring historical context, limiting their adaptation and effectiveness in handling challenging camouflage scenes. In this paper, we propose a recall-augmented COD architecture, namely RetroMem, which dynamically modulates camouflage pattern perception and inference by integrating relevant historical knowledge into the process. Specifically, RetroMem employs a two-stage training paradigm consisting of a learning stage and a recall stage to construct, update, and utilize memory representations effectively. During the learning stage, we design a dense multi-scale adapter (DMA) to improve the pretrained encoder's capability to capture rich multi-scale visual information with very few trainable parameters, thereby providing foundational inferences. In the recall stage, we propose a dynamic memory mechanism (DMM) and an inference pattern reconstruction (IPR). These components fully leverage the latent relationships between learned knowledge and current sample context to reconstruct the inference of camouflage patterns, thereby significantly improving the model's understanding of camouflage scenes. Extensive experiments on several widely used datasets demonstrate that our RetroMem significantly outperforms existing state-of-the-art methods.
☆ RA-NeRF: Robust Neural Radiance Field Reconstruction with Accurate Camera Pose Estimation under Complex Trajectories IROS 2025
Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have emerged as powerful tools for 3D reconstruction and SLAM tasks. However, their performance depends heavily on accurate camera pose priors. Existing approaches attempt to address this issue by introducing external constraints but fall short of achieving satisfactory accuracy, particularly when camera trajectories are complex. In this paper, we propose a novel method, RA-NeRF, capable of predicting highly accurate camera poses even with complex camera trajectories. Following the incremental pipeline, RA-NeRF reconstructs the scene using NeRF with photometric consistency and incorporates flow-driven pose regulation to enhance robustness during initialization and localization. Additionally, RA-NeRF employs an implicit pose filter to capture the camera movement pattern and eliminate the noise for pose estimation. To validate our method, we conduct extensive experiments on the Tanks\&Temple dataset for standard evaluation, as well as the NeRFBuster dataset, which presents challenging camera pose trajectories. On both datasets, RA-NeRF achieves state-of-the-art results in both camera pose estimation and visual quality, demonstrating its effectiveness and robustness in scene reconstruction under complex pose trajectories.
comment: IROS 2025
☆ Convolutional Feature Enhancement and Attention Fusion BiFPN for Ship Detection in SAR Images
Synthetic Aperture Radar (SAR) enables submeter-resolution imaging and all-weather monitoring via active microwave and advanced signal processing. Currently, SAR has found extensive applications in critical maritime domains such as ship detection. However, SAR ship detection faces several challenges, including significant scale variations among ships, the presence of small offshore vessels mixed with noise, and complex backgrounds for large nearshore ships. To address these issues, this paper proposes a novel feature enhancement and fusion framework named C-AFBiFPN. C-AFBiFPN constructs a Convolutional Feature Enhancement (CFE) module following the backbone network, aiming to enrich feature representation and enhance the ability to capture and represent local details and contextual information. Furthermore, C-AFBiFPN innovatively integrates BiFormer attention within the fusion strategy of BiFPN, creating the AFBiFPN network. AFBiFPN improves the global modeling capability of cross-scale feature fusion and can adaptively focus on critical feature regions. The experimental results on SAR Ship Detection Dataset (SSDD) indicate that the proposed approach substantially enhances detection accuracy for small targets, robustness against occlusions, and adaptability to multi-scale features.
comment: 5 pages, 4 figures, 2 tables. Code available at https://github.com/mlj666219/C-AFBiFPN/tree/master
☆ video-SALMONN 2: Captioning-Enhanced Audio-Visual Large Language Models
Videos contain a wealth of information, and generating detailed and accurate descriptions in natural language is a key aspect of video understanding. In this paper, we present video-SALMONN 2, an advanced audio-visual large language model (LLM) with low-rank adaptation (LoRA) designed for enhanced video (with paired audio) captioning through directed preference optimisation (DPO). We propose new metrics to evaluate the completeness and accuracy of video descriptions, which are optimised using DPO. To further improve training, we propose a novel multi-round DPO (MrDPO) approach, which involves periodically updating the DPO reference model, merging and re-initialising the LoRA module as a proxy for parameter updates after each training round (1,000 steps), and incorporating guidance from ground-truth video captions to stabilise the process. Experimental results show that MrDPO significantly enhances video-SALMONN 2's captioning accuracy, reducing the captioning error rates by 28\%. The final video-SALMONN 2 model, with just 7 billion parameters, surpasses leading models such as GPT-4o and Gemini-1.5-Pro in video captioning tasks, while maintaining highly competitive performance to the state-of-the-art on widely used video question-answering benchmarks among models of similar size. Codes are available at \href{https://github.com/bytedance/video-SALMONN-2}{https://github.com/bytedance/video-SALMONN-2}.
☆ DM-FNet: Unified multimodal medical image fusion via diffusion process-trained encoder-decoder
Multimodal medical image fusion (MMIF) extracts the most meaningful information from multiple source images, enabling a more comprehensive and accurate diagnosis. Achieving high-quality fusion results requires a careful balance of brightness, color, contrast, and detail; this ensures that the fused images effectively display relevant anatomical structures and reflect the functional status of the tissues. However, existing MMIF methods have limited capacity to capture detailed features during conventional training and suffer from insufficient cross-modal feature interaction, leading to suboptimal fused image quality. To address these issues, this study proposes a two-stage diffusion model-based fusion network (DM-FNet) to achieve unified MMIF. In Stage I, a diffusion process trains UNet for image reconstruction. UNet captures detailed information through progressive denoising and represents multilevel data, providing a rich set of feature representations for the subsequent fusion network. In Stage II, noisy images at various steps are input into the fusion network to enhance the model's feature recognition capability. Three key fusion modules are also integrated to process medical images from different modalities adaptively. Ultimately, the robust network structure and a hybrid loss function are integrated to harmonize the fused image's brightness, color, contrast, and detail, enhancing its quality and information density. The experimental results across various medical image types demonstrate that the proposed method performs exceptionally well regarding objective evaluation metrics. The fused image preserves appropriate brightness, a comprehensive distribution of radioactive tracers, rich textures, and clear edges. The code is available at https://github.com/HeDan-11/DM-FNet.
comment: This paper has been accepted by IEEE Transactions on Multimedia (TMM) in March 2025
☆ Privacy-Shielded Image Compression: Defending Against Exploitation from Vision-Language Pretrained Models ICML 2025
The improved semantic understanding of vision-language pretrained (VLP) models has made it increasingly difficult to protect publicly posted images from being exploited by search engines and other similar tools. In this context, this paper seeks to protect users' privacy by implementing defenses at the image compression stage to prevent exploitation. Specifically, we propose a flexible coding method, termed Privacy-Shielded Image Compression (PSIC), that can produce bitstreams with multiple decoding options. By default, the bitstream is decoded to preserve satisfactory perceptual quality while preventing interpretation by VLP models. Our method also retains the original image compression functionality. With a customizable input condition, the proposed scheme can reconstruct the image that preserves its full semantic information. A Conditional Latent Trigger Generation (CLTG) module is proposed to produce bias information based on customizable conditions to guide the decoding process into different reconstructed versions, and an Uncertainty-Aware Encryption-Oriented (UAEO) optimization function is designed to leverage the soft labels inferred from the target VLP model's uncertainty on the training data. This paper further incorporates an adaptive multi-objective optimization strategy to obtain improved encrypting performance and perceptual quality simultaneously within a unified training process. The proposed scheme is plug-and-play and can be seamlessly integrated into most existing Learned Image Compression (LIC) models. Extensive experiments across multiple downstream tasks have demonstrated the effectiveness of our design.
comment: 11 pages, 6 figures, publised to ICML 2025
☆ Conquering the Retina: Bringing Visual in-Context Learning to OCT
Recent advancements in medical image analysis have led to the development of highly specialized models tailored to specific clinical tasks. These models have demonstrated exceptional performance and remain a crucial research direction. Yet, their applicability is limited to predefined tasks, requiring expertise and extensive resources for development and adaptation. In contrast, generalist models offer a different form of utility: allowing medical practitioners to define tasks on the fly without the need for task-specific model development. In this work, we explore how to train generalist models for the domain of retinal optical coherence tomography using visual in-context learning (VICL), i.e., training models to generalize across tasks based on a few examples provided at inference time. To facilitate rigorous assessment, we propose a broad evaluation protocol tailored to VICL in OCT. We extensively evaluate a state-of-the-art medical VICL approach on multiple retinal OCT datasets, establishing a first baseline to highlight the potential and current limitations of in-context learning for OCT. To foster further research and practical adoption, we openly release our code.
☆ Classification of Multi-Parametric Body MRI Series Using Deep Learning
Multi-parametric magnetic resonance imaging (mpMRI) exams have various series types acquired with different imaging protocols. The DICOM headers of these series often have incorrect information due to the sheer diversity of protocols and occasional technologist errors. To address this, we present a deep learning-based classification model to classify 8 different body mpMRI series types so that radiologists read the exams efficiently. Using mpMRI data from various institutions, multiple deep learning-based classifiers of ResNet, EfficientNet, and DenseNet are trained to classify 8 different MRI series, and their performance is compared. Then, the best-performing classifier is identified, and its classification capability under the setting of different training data quantities is studied. Also, the model is evaluated on the out-of-training-distribution datasets. Moreover, the model is trained using mpMRI exams obtained from different scanners in two training strategies, and its performance is tested. Experimental results show that the DenseNet-121 model achieves the highest F1-score and accuracy of 0.966 and 0.972 over the other classification models with p-value$<$0.05. The model shows greater than 0.95 accuracy when trained with over 729 studies of the training data, whose performance improves as the training data quantities grew larger. On the external data with the DLDS and CPTAC-UCEC datasets, the model yields 0.872 and 0.810 accuracy for each. These results indicate that in both the internal and external datasets, the DenseNet-121 model attains high accuracy for the task of classifying 8 body MRI series types.
☆ ReSeDis: A Dataset for Referring-based Object Search across Large-Scale Image Collections
Large-scale visual search engines are expected to solve a dual problem at once: (i) locate every image that truly contains the object described by a sentence and (ii) identify the object's bounding box or exact pixels within each hit. Existing techniques address only one side of this challenge. Visual grounding yields tight boxes and masks but rests on the unrealistic assumption that the object is present in every test image, producing a flood of false alarms when applied to web-scale collections. Text-to-image retrieval excels at sifting through massive databases to rank relevant images, yet it stops at whole-image matches and offers no fine-grained localization. We introduce Referring Search and Discovery (ReSeDis), the first task that unifies corpus-level retrieval with pixel-level grounding. Given a free-form description, a ReSeDis model must decide whether the queried object appears in each image and, if so, where it is, returning bounding boxes or segmentation masks. To enable rigorous study, we curate a benchmark in which every description maps uniquely to object instances scattered across a large, diverse corpus, eliminating unintended matches. We further design a task-specific metric that jointly scores retrieval recall and localization precision. Finally, we provide a straightforward zero-shot baseline using a frozen vision-language model, revealing significant headroom for future study. ReSeDis offers a realistic, end-to-end testbed for building the next generation of robust and scalable multimodal search systems.
☆ Echo-DND: A dual noise diffusion model for robust and precise left ventricle segmentation in echocardiography
Recent advancements in diffusion probabilistic models (DPMs) have revolutionized image processing, demonstrating significant potential in medical applications. Accurate segmentation of the left ventricle (LV) in echocardiograms is crucial for diagnostic procedures and necessary treatments. However, ultrasound images are notoriously noisy with low contrast and ambiguous LV boundaries, thereby complicating the segmentation process. To address these challenges, this paper introduces Echo-DND, a novel dual-noise diffusion model specifically designed for this task. Echo-DND leverages a unique combination of Gaussian and Bernoulli noises. It also incorporates a multi-scale fusion conditioning module to improve segmentation precision. Furthermore, it utilizes spatial coherence calibration to maintain spatial integrity in segmentation masks. The model's performance was rigorously validated on the CAMUS and EchoNet-Dynamic datasets. Extensive evaluations demonstrate that the proposed framework outperforms existing SOTA models. It achieves high Dice scores of 0.962 and 0.939 on these datasets, respectively. The proposed Echo-DND model establishes a new standard in echocardiogram segmentation, and its architecture holds promise for broader applicability in other medical imaging tasks, potentially improving diagnostic accuracy across various medical domains. Project page: https://abdur75648.github.io/Echo-DND
comment: Version of record published in Discover Applied Sciences (Springer Nature). The definitive article is available at https://doi.org/10.1007/s42452-025-07055-5
☆ Enhancing point cloud analysis via neighbor aggregation correction based on cross-stage structure correlation
Point cloud analysis is the cornerstone of many downstream tasks, among which aggregating local structures is the basis for understanding point cloud data. While numerous works aggregate neighbor using three-dimensional relative coordinates, there are irrelevant point interference and feature hierarchy gap problems due to the limitation of local coordinates. Although some works address this limitation by refining spatial description though explicit modeling of cross-stage structure, these enhancement methods based on direct geometric structure encoding have problems of high computational overhead and noise sensitivity. To overcome these problems, we propose the Point Distribution Set Abstraction module (PDSA) that utilizes the correlation in the high-dimensional space to correct the feature distribution during aggregation, which improves the computational efficiency and robustness. PDSA distinguishes the point correlation based on a lightweight cross-stage structural descriptor, and enhances structural homogeneity by reducing the variance of the neighbor feature matrix and increasing classes separability though long-distance modeling. Additionally, we introducing a key point mechanism to optimize the computational overhead. The experimental result on semantic segmentation and classification tasks based on different baselines verify the generalization of the method we proposed, and achieve significant performance improvement with less parameter cost. The corresponding ablation and visualization results demonstrate the effectiveness and rationality of our method. The code and training weight is available at: https://github.com/AGENT9717/PointDistribution
comment: 17 papes, 7 figures
☆ Robust Instant Policy: Leveraging Student's t-Regression Model for Robust In-context Imitation Learning of Robot Manipulation IROS
Imitation learning (IL) aims to enable robots to perform tasks autonomously by observing a few human demonstrations. Recently, a variant of IL, called In-Context IL, utilized off-the-shelf large language models (LLMs) as instant policies that understand the context from a few given demonstrations to perform a new task, rather than explicitly updating network models with large-scale demonstrations. However, its reliability in the robotics domain is undermined by hallucination issues such as LLM-based instant policy, which occasionally generates poor trajectories that deviate from the given demonstrations. To alleviate this problem, we propose a new robust in-context imitation learning algorithm called the robust instant policy (RIP), which utilizes a Student's t-regression model to be robust against the hallucinated trajectories of instant policies to allow reliable trajectory generation. Specifically, RIP generates several candidate robot trajectories to complete a given task from an LLM and aggregates them using the Student's t-distribution, which is beneficial for ignoring outliers (i.e., hallucinations); thereby, a robust trajectory against hallucinations is generated. Our experiments, conducted in both simulated and real-world environments, show that RIP significantly outperforms state-of-the-art IL methods, with at least $26\%$ improvement in task success rates, particularly in low-data scenarios for everyday tasks. Video results available at https://sites.google.com/view/robustinstantpolicy.
comment: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025 accepted
☆ SynPo: Boosting Training-Free Few-Shot Medical Segmentation via High-Quality Negative Prompts
The advent of Large Vision Models (LVMs) offers new opportunities for few-shot medical image segmentation. However, existing training-free methods based on LVMs fail to effectively utilize negative prompts, leading to poor performance on low-contrast medical images. To address this issue, we propose SynPo, a training-free few-shot method based on LVMs (e.g., SAM), with the core insight: improving the quality of negative prompts. To select point prompts in a more reliable confidence map, we design a novel Confidence Map Synergy Module by combining the strengths of DINOv2 and SAM. Based on the confidence map, we select the top-k pixels as the positive points set and choose the negative points set using a Gaussian distribution, followed by independent K-means clustering for both sets. Then, these selected points are leveraged as high-quality prompts for SAM to get the segmentation results. Extensive experiments demonstrate that SynPo achieves performance comparable to state-of-the-art training-based few-shot methods.
☆ An Empirical Study of Bugs in Data Visualization Libraries
Data visualization (DataViz) libraries play a crucial role in presentation, data analysis, and application development, underscoring the importance of their accuracy in transforming data into visual representations. Incorrect visualizations can adversely impact user experience, distort information conveyance, and influence user perception and decision-making processes. Visual bugs in these libraries can be particularly insidious as they may not cause obvious errors like crashes, but instead mislead users of the underlying data graphically, resulting in wrong decision making. Consequently, a good understanding of the unique characteristics of bugs in DataViz libraries is essential for researchers and developers to detect and fix bugs in DataViz libraries. This study presents the first comprehensive analysis of bugs in DataViz libraries, examining 564 bugs collected from five widely-used libraries. Our study systematically analyzes their symptoms and root causes, and provides a detailed taxonomy. We found that incorrect/inaccurate plots are pervasive in DataViz libraries and incorrect graphic computation is the major root cause, which necessitates further automated testing methods for DataViz libraries. Moreover, we identified eight key steps to trigger such bugs and two test oracles specific to DataViz libraries, which may inspire future research in designing effective automated testing techniques. Furthermore, with the recent advancements in Vision Language Models (VLMs), we explored the feasibility of applying these models to detect incorrect/inaccurate plots. The results show that the effectiveness of VLMs in bug detection varies from 29% to 57%, depending on the prompts, and adding more information in prompts does not necessarily increase the effectiveness. More findings can be found in our manuscript.
comment: Proc. ACM Softw. Eng. 2, FSE
☆ Enhancing Vector Quantization with Distributional Matching: A Theoretical and Empirical Study
The success of autoregressive models largely depends on the effectiveness of vector quantization, a technique that discretizes continuous features by mapping them to the nearest code vectors within a learnable codebook. Two critical issues in existing vector quantization methods are training instability and codebook collapse. Training instability arises from the gradient discrepancy introduced by the straight-through estimator, especially in the presence of significant quantization errors, while codebook collapse occurs when only a small subset of code vectors are utilized during training. A closer examination of these issues reveals that they are primarily driven by a mismatch between the distributions of the features and code vectors, leading to unrepresentative code vectors and significant data information loss during compression. To address this, we employ the Wasserstein distance to align these two distributions, achieving near 100\% codebook utilization and significantly reducing the quantization error. Both empirical and theoretical analyses validate the effectiveness of the proposed approach.
☆ Break Stylistic Sophon: Are We Really Meant to Confine the Imagination in Style Transfer?
In this pioneering study, we introduce StyleWallfacer, a groundbreaking unified training and inference framework, which not only addresses various issues encountered in the style transfer process of traditional methods but also unifies the framework for different tasks. This framework is designed to revolutionize the field by enabling artist level style transfer and text driven stylization. First, we propose a semantic-based style injection method that uses BLIP to generate text descriptions strictly aligned with the semantics of the style image in CLIP space. By leveraging a large language model to remove style-related descriptions from these descriptions, we create a semantic gap. This gap is then used to fine-tune the model, enabling efficient and drift-free injection of style knowledge. Second, we propose a data augmentation strategy based on human feedback, incorporating high-quality samples generated early in the fine-tuning process into the training set to facilitate progressive learning and significantly reduce its overfitting. Finally, we design a training-free triple diffusion process using the fine-tuned model, which manipulates the features of self-attention layers in a manner similar to the cross-attention mechanism. Specifically, in the generation process, the key and value of the content-related process are replaced with those of the style-related process to inject style while maintaining text control over the model. We also introduce query preservation to mitigate disruptions to the original content. Under such a design, we have achieved high-quality image-driven style transfer and text-driven stylization, delivering artist-level style transfer results while preserving the original image content. Moreover, we achieve image color editing during the style transfer process for the first time.
☆ An accurate and revised version of optical character recognition-based speech synthesis using LabVIEW
Knowledge extraction through sound is a distinctive property. Visually impaired individuals often rely solely on Braille books and audio recordings provided by NGOs. Due to limitations in these approaches, blind individuals often cannot access books of their choice. Speech is a more effective mode of communication than text for blind and visually impaired persons, as they can easily respond to sounds. This paper presents the development of an accurate, reliable, cost-effective, and user-friendly optical character recognition (OCR)-based speech synthesis system. The OCR-based system has been implemented using Laboratory Virtual Instrument Engineering Workbench (LabVIEW).
comment: 9 pages, 9 figures
♻ ☆ Cosmos-Drive-Dreams: Scalable Synthetic Driving Data Generation with World Foundation Models
Collecting and annotating real-world data for safety-critical physical AI systems, such as Autonomous Vehicle (AV), is time-consuming and costly. It is especially challenging to capture rare edge cases, which play a critical role in training and testing of an AV system. To address this challenge, we introduce the Cosmos-Drive-Dreams - a synthetic data generation (SDG) pipeline that aims to generate challenging scenarios to facilitate downstream tasks such as perception and driving policy training. Powering this pipeline is Cosmos-Drive, a suite of models specialized from NVIDIA Cosmos world foundation model for the driving domain and are capable of controllable, high-fidelity, multi-view, and spatiotemporally consistent driving video generation. We showcase the utility of these models by applying Cosmos-Drive-Dreams to scale the quantity and diversity of driving datasets with high-fidelity and challenging scenarios. Experimentally, we demonstrate that our generated data helps in mitigating long-tail distribution problems and enhances generalization in downstream tasks such as 3D lane detection, 3D object detection and driving policy learning. We open source our pipeline toolkit, dataset and model weights through the NVIDIA's Cosmos platform. Project page: https://research.nvidia.com/labs/toronto-ai/cosmos_drive_dreams
comment: Only the core contributors are listed. The full list of contributors can be found in Appendix A of this paper
♻ ☆ Generalized Out-of-Distribution Detection and Beyond in Vision Language Model Era: A Survey
Detecting out-of-distribution (OOD) samples is crucial for ensuring the safety of machine learning systems and has shaped the field of OOD detection. Meanwhile, several other problems are closely related to OOD detection, including anomaly detection (AD), novelty detection (ND), open set recognition (OSR), and outlier detection (OD). To unify these problems, a generalized OOD detection framework was proposed, taxonomically categorizing these five problems. However, Vision Language Models (VLMs) such as CLIP have significantly changed the paradigm and blurred the boundaries between these fields, again confusing researchers. In this survey, we first present a generalized OOD detection v2, encapsulating the evolution of these fields in the VLM era. Our framework reveals that, with some field inactivity and integration, the demanding challenges have become OOD detection and AD. Then, we highlight the significant shift in the definition, problem settings, and benchmarks; we thus feature a comprehensive review of the methodology for OOD detection and related tasks to clarify their relationship to OOD detection. Finally, we explore the advancements in the emerging Large Vision Language Model (LVLM) era, such as GPT-4V. We conclude with open challenges and future directions. The resource is available at https://github.com/AtsuMiyai/Awesome-OOD-VLM.
comment: Accepted at TMLR2025. Survey paper. We welcome questions, issues, and paper requests via https://github.com/AtsuMiyai/Awesome-OOD-VLM
♻ ☆ Vision Transformers Don't Need Trained Registers
We investigate the mechanism underlying a previously identified phenomenon in Vision Transformers -- the emergence of high-norm tokens that lead to noisy attention maps. We observe that in multiple models (e.g., CLIP, DINOv2), a sparse set of neurons is responsible for concentrating high-norm activations on outlier tokens, leading to irregular attention patterns and degrading downstream visual processing. While the existing solution for removing these outliers involves retraining models from scratch with additional learned register tokens, we use our findings to create a training-free approach to mitigate these artifacts. By shifting the high-norm activations from our discovered register neurons into an additional untrained token, we can mimic the effect of register tokens on a model already trained without registers. We demonstrate that our method produces cleaner attention and feature maps, enhances performance over base models across multiple downstream visual tasks, and achieves results comparable to models explicitly trained with register tokens. We then extend test-time registers to off-the-shelf vision-language models to improve their interpretability. Our results suggest that test-time registers effectively take on the role of register tokens at test-time, offering a training-free solution for any pre-trained model released without them.
comment: Project page and code: https://avdravid.github.io/test-time-registers
♻ ☆ I2I-Mamba: Multi-modal medical image synthesis via selective state space modeling
Multi-modal medical image synthesis involves nonlinear transformation of tissue signals between source and target modalities, where tissues exhibit contextual interactions across diverse spatial distances. As such, the utility of a network architecture in synthesis depends on its ability to express these contextual features. Convolutional neural networks (CNNs) offer high local precision at the expense of poor sensitivity to long-range context. While transformers promise to alleviate this issue, they suffer from an unfavorable trade-off between sensitivity to long- versus short-range context due to the intrinsic complexity of attention filters. To effectively capture contextual features while avoiding the complexity-driven trade-offs, here we introduce a novel multi-modal synthesis method, I2I-Mamba, based on the state space modeling (SSM) framework. Focusing on semantic representations across a hybrid residual architecture, I2I-Mamba leverages novel dual-domain Mamba (ddMamba) blocks for complementary contextual modeling in image and Fourier domains, while maintaining spatial precision with convolutional layers. Diverting from conventional raster-scan trajectories, ddMamba leverages novel SSM operators based on a spiral-scan trajectory to learn context with enhanced radial coverage and angular isotropy, and a channel-mixing layer to aggregate context across the channel dimension. Comprehensive demonstrations on multi-contrast MRI and MRI-CT protocols indicate that I2I-Mamba offers superior performance against state-of-the-art CNNs, transformers and SSMs.
comment: 14 pages, 6 figures
♻ ☆ A dataset of high-resolution plantar pressures for gait analysis across varying footwear and walking speeds
Gait refers to the patterns of limb movement generated during walking, which are unique to each individual due to both physical and behavioral traits. Walking patterns have been widely studied in biometrics, biomechanics, sports, and rehabilitation. While traditional methods rely on video and motion capture, advances in plantar pressure sensing technology now offer deeper insights into gait. However, underfoot pressures during walking remain underexplored due to the lack of large, publicly accessible datasets. To address this, we introduce the UNB StepUP-P150 dataset: a footStep database for gait analysis and recognition using Underfoot Pressure, including data from 150 individuals. This dataset comprises high-resolution plantar pressure data (4 sensors per cm-squared) collected using a 1.2m by 3.6m pressure-sensing walkway. It contains over 200,000 footsteps from participants walking with various speeds (preferred, slow-to-stop, fast, and slow) and footwear conditions (barefoot, standard shoes, and two personal shoes), supporting advancements in biometric gait recognition and presenting new research opportunities in biomechanics and deep learning. UNB StepUP-P150 establishes a new benchmark for plantar pressure-based gait analysis and recognition.
♻ ☆ VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding
Synthetic video generation has gained significant attention for its realism and broad applications, but remains prone to violations of common sense and physical laws. This highlights the need for reliable abnormality detectors that understand such principles and are robust to hallucinations. To address this, we introduce VideoHallu, a benchmark of over 3,000 video QA pairs built from synthetic videos generated by models like Veo2, Sora, and Kling, paired with expert-crafted counterintuitive QA to evaluate the critical thinking abilities of Multi-modal Large Language Models (MLLMs) on abnormalities that are perceptually obvious to humans but often hallucinated due to language priors. VideoHallu evaluates MLLMs' abnormality detection abilities with examples across alignment, consistency, commonsense, and physics. We benchmark SOTA MLLMs, including GPT-4o, Gemini-2.5-Pro, Qwen2.5-VL, Video-R1, and VideoChat-R1. We observe that these models perform well on many real-world benchmarks like MVBench and MovieChat, but still struggle with basic physics-based and commonsense reasoning in synthetic videos. We further show that post-training with Group Relative Policy Optimization (GRPO), using curriculum learning on datasets combining video QA with counterintuitive commonsense and physics reasoning over real and synthetic videos, improves MLLMs' abnormality detection and critical thinking, demonstrating the value of targeted training for improving their understanding of commonsense and physical laws. Our code is available at https://github.com/zli12321/VideoHallu.git.
♻ ☆ RDD: Robust Feature Detector and Descriptor using Deformable Transformer
As a core step in structure-from-motion and SLAM, robust feature detection and description under challenging scenarios such as significant viewpoint changes remain unresolved despite their ubiquity. While recent works have identified the importance of local features in modeling geometric transformations, these methods fail to learn the visual cues present in long-range relationships. We present Robust Deformable Detector (RDD), a novel and robust keypoint detector/descriptor leveraging the deformable transformer, which captures global context and geometric invariance through deformable self-attention mechanisms. Specifically, we observed that deformable attention focuses on key locations, effectively reducing the search space complexity and modeling the geometric invariance. Furthermore, we collected an Air-to-Ground dataset for training in addition to the standard MegaDepth dataset. Our proposed method outperforms all state-of-the-art keypoint detection/description methods in sparse matching tasks and is also capable of semi-dense matching. To ensure comprehensive evaluation, we introduce two challenging benchmarks: one emphasizing large viewpoint and scale variations, and the other being an Air-to-Ground benchmark -- an evaluation setting that has recently gaining popularity for 3D reconstruction across different altitudes.
♻ ☆ TARDIS STRIDE: A Spatio-Temporal Road Image Dataset and World Model for Autonomy
World models aim to simulate environments and enable effective agent behavior. However, modeling real-world environments presents unique challenges as they dynamically change across both space and, crucially, time. To capture these composed dynamics, we introduce a Spatio-Temporal Road Image Dataset for Exploration (STRIDE) permuting 360-degree panoramic imagery into rich interconnected observation, state and action nodes. Leveraging this structure, we can simultaneously model the relationship between egocentric views, positional coordinates, and movement commands across both space and time. We benchmark this dataset via TARDIS, a transformer-based generative world model that integrates spatial and temporal dynamics through a unified autoregressive framework trained on STRIDE. We demonstrate robust performance across a range of agentic tasks such as controllable photorealistic image synthesis, instruction following, autonomous self-control, and state-of-the-art georeferencing. These results suggest a promising direction towards sophisticated generalist agents--capable of understanding and manipulating the spatial and temporal aspects of their material environments--with enhanced embodied reasoning capabilities. Training code, datasets, and model checkpoints are made available at https://huggingface.co/datasets/Tera-AI/STRIDE.
comment: Computer Vision, Pattern Recognition, Early-Fusion, Dataset, Data Augmentation
♻ ☆ LaViDa: A Large Diffusion Language Model for Multimodal Understanding
Modern Vision-Language Models (VLMs) can solve a wide range of tasks requiring visual reasoning. In real-world scenarios, desirable properties for VLMs include fast inference and controllable generation (e.g., constraining outputs to adhere to a desired format). However, existing autoregressive (AR) VLMs like LLaVA struggle in these aspects. Discrete diffusion models (DMs) offer a promising alternative, enabling parallel decoding for faster inference and bidirectional context for controllable generation through text-infilling. While effective in language-only settings, DMs' potential for multimodal tasks is underexplored. We introduce LaViDa, a family of VLMs built on DMs. We build LaViDa by equipping DMs with a vision encoder and jointly fine-tune the combined parts for multimodal instruction following. To address challenges encountered, LaViDa incorporates novel techniques such as complementary masking for effective training, prefix KV cache for efficient inference, and timestep shifting for high-quality sampling. Experiments show that LaViDa achieves competitive or superior performance to AR VLMs on multi-modal benchmarks such as MMMU, while offering unique advantages of DMs, including flexible speed-quality tradeoff, controllability, and bidirectional reasoning. On COCO captioning, LaViDa surpasses Open-LLaVa-Next-8B by +4.1 CIDEr with 1.92x speedup. On bidirectional tasks, it achieves +59% improvement on Constrained Poem Completion. These results demonstrate LaViDa as a strong alternative to AR VLMs. Code and models will be released in the camera-ready version.
comment: 26 pages, 8 figures
♻ ☆ Exploring Personalized Federated Learning Architectures for Violence Detection in Surveillance Videos
The challenge of detecting violent incidents in urban surveillance systems is compounded by the voluminous and diverse nature of video data. This paper presents a targeted approach using Personalized Federated Learning (PFL) to address these issues, specifically employing the Federated Learning with Personalization Layers method within the Flower framework. Our methodology adapts learning models to the unique data characteristics of each surveillance node, effectively managing the heterogeneous and non-IID nature of surveillance video data. Through rigorous experiments conducted on balanced and imbalanced datasets, our PFL models demonstrated enhanced accuracy and efficiency, achieving up to 99.3% accuracy. This study underscores the potential of PFL to significantly improve the scalability and effectiveness of surveillance systems, offering a robust, privacy-preserving solution for violence detection in complex urban environments.
comment: 7 pages, 5 figures, 4 tables
♻ ☆ A Comprehensive Survey on Continual Learning in Generative Models
The rapid advancement of generative models has enabled modern AI systems to comprehend and produce highly sophisticated content, even achieving human-level performance in specific domains. However, these models remain fundamentally constrained by catastrophic forgetting - a persistent challenge where adapting to new tasks typically leads to significant degradation in performance on previously learned tasks. To address this practical limitation, numerous approaches have been proposed to enhance the adaptability and scalability of generative models in real-world applications. In this work, we present a comprehensive survey of continual learning methods for mainstream generative models, including large language models, multimodal large language models, vision language action models, and diffusion models. Drawing inspiration from the memory mechanisms of the human brain, we systematically categorize these approaches into three paradigms: architecture-based, regularization-based, and replay-based methods, while elucidating their underlying methodologies and motivations. We further analyze continual learning setups for different generative models, including training objectives, benchmarks, and core backbones, offering deeper insights into the field. The project page of this paper is available at https://github.com/Ghy0501/Awesome-Continual-Learning-in-Generative-Models.
comment: Preprint
♻ ☆ EgoBlind: Towards Egocentric Visual Assistance for the Blind
We present EgoBlind, the first egocentric VideoQA dataset collected from blind individuals to evaluate the assistive capabilities of contemporary multimodal large language models (MLLMs). EgoBlind comprises 1,392 videos that record the daily lives of real blind users from a first-person perspective. It also features 5,311 questions directly posed or generated and verified by blind individuals to reflect their in-situation needs for visual assistance under various scenarios. We provide each question with an average of 3 reference answers to alleviate subjective evaluation. Using EgoBlind, we comprehensively evaluate 16 advanced MLLMs and find that all models struggle, with the best performers achieving accuracy near 60\%, far behind human performance of 87.4\%. To guide future advancements, we identify and summarize major limitations of existing MLLMs in egocentric visual assistance for the blind and explore heuristic solutions for improvement. With these efforts, we hope EgoBlind can serve as a valuable foundation for developing more effective AI assistants to enhance the independence of the blind individuals' lives. Data and evaluation code are available at https://github.com/doc-doc/EgoBlind.
comment: We extend and resplit the dataset
♻ ☆ Translation-Equivariance of Normalization Layers and Aliasing in Convolutional Neural Networks COLT 2025
The design of convolutional neural architectures that are exactly equivariant to continuous translations is an active field of research. It promises to benefit scientific computing, notably by making existing imaging systems more physically accurate. Most efforts focus on the design of downsampling/pooling layers, upsampling layers and activation functions, but little attention is dedicated to normalization layers. In this work, we present a novel theoretical framework for understanding the equivariance of normalization layers to discrete shifts and continuous translations. We also determine necessary and sufficient conditions for normalization layers to be equivariant in terms of the dimensions they operate on. Using real feature maps from ResNet-18 and ImageNet, we test those theoretical results empirically and find that they are consistent with our predictions.
comment: Accepted at the Workshop on the Theory of AI for Scientific Computing (COLT 2025)
♻ ☆ Unsourced Adversarial CAPTCHA: A Bi-Phase Adversarial CAPTCHA Framework
With the rapid advancements in deep learning, traditional CAPTCHA schemes are increasingly vulnerable to automated attacks powered by deep neural networks (DNNs). Existing adversarial attack methods often rely on original image characteristics, resulting in distortions that hinder human interpretation and limit applicability in scenarios lacking initial input images. To address these challenges, we propose the Unsourced Adversarial CAPTCHA (UAC), a novel framework generating high-fidelity adversarial examples guided by attacker-specified text prompts. Leveraging a Large Language Model (LLM), UAC enhances CAPTCHA diversity and supports both targeted and untargeted attacks. For targeted attacks, the EDICT method optimizes dual latent variables in a diffusion model for superior image quality. In untargeted attacks, especially for black-box scenarios, we introduce bi-path unsourced adversarial CAPTCHA (BP-UAC), a two-step optimization strategy employing multimodal gradients and bi-path optimization for efficient misclassification. Experiments show BP-UAC achieves high attack success rates across diverse systems, generating natural CAPTCHAs indistinguishable to humans and DNNs.
♻ ☆ Leveraging Depth and Language for Open-Vocabulary Domain-Generalized Semantic Segmentation
Open-Vocabulary semantic segmentation (OVSS) and domain generalization in semantic segmentation (DGSS) highlight a subtle complementarity that motivates Open-Vocabulary Domain-Generalized Semantic Segmentation (OV-DGSS). OV-DGSS aims to generate pixel-level masks for unseen categories while maintaining robustness across unseen domains, a critical capability for real-world scenarios such as autonomous driving in adverse conditions. We introduce Vireo, a novel single-stage framework for OV-DGSS that unifies the strengths of OVSS and DGSS for the first time. Vireo builds upon the frozen Visual Foundation Models (VFMs) and incorporates scene geometry via Depth VFMs to extract domain-invariant structural features. To bridge the gap between visual and textual modalities under domain shift, we propose three key components: (1) GeoText Prompts, which align geometric features with language cues and progressively refine VFM encoder representations; (2) Coarse Mask Prior Embedding (CMPE) for enhancing gradient flow for faster convergence and stronger textual influence; and (3) the Domain-Open-Vocabulary Vector Embedding Head (DOV-VEH), which fuses refined structural and semantic features for robust prediction. Comprehensive evaluation on these components demonstrates the effectiveness of our designs. Our proposed Vireo achieves the state-of-the-art performance and surpasses existing methods by a large margin in both domain generalization and open-vocabulary recognition, offering a unified and scalable solution for robust visual understanding in diverse and dynamic environments. Code is available at https://github.com/anonymouse-9c53tp182bvz/Vireo.
♻ ☆ YOLOv11-RGBT: Towards a Comprehensive Single-Stage Multispectral Object Detection Framework
Multispectral object detection, which integrates information from multiple bands, can enhance detection accuracy and environmental adaptability, holding great application potential across various fields. Although existing methods have made progress in cross-modal interaction, low-light conditions, and model lightweight, there are still challenges like the lack of a unified single-stage framework, difficulty in balancing performance and fusion strategy, and unreasonable modality weight allocation. To address these, based on the YOLOv11 framework, we present YOLOv11-RGBT, a new comprehensive multimodal object detection framework. We designed six multispectral fusion modes and successfully applied them to models from YOLOv3 to YOLOv12 and RT-DETR. After reevaluating the importance of the two modalities, we proposed a P3 mid-fusion strategy and multispectral controllable fine-tuning (MCF) strategy for multispectral models. These improvements optimize feature fusion, reduce redundancy and mismatches, and boost overall model performance. Experiments show our framework excels on three major open-source multispectral object detection datasets, like LLVIP and FLIR. Particularly, the multispectral controllable fine-tuning strategy significantly enhanced model adaptability and robustness. On the FLIR dataset, it consistently improved YOLOv11 models' mAP by 3.41%-5.65%, reaching a maximum of 47.61%, verifying the framework and strategies' effectiveness. The code is available at: https://github.com/wandahangFY/YOLOv11-RGBT.
comment: 29 pages, 8 figures . The errors in the first version have been corrected, and no new version will be submitted in the near future. The next version will include more experiments
♻ ☆ Instance-Adaptive Keypoint Learning with Local-to-Global Geometric Aggregation for Category-Level Object Pose Estimation
Category-level object pose estimation aims to predict the 6D pose and size of previously unseen instances from predefined categories, requiring strong generalization across diverse object instances. Although many previous methods attempt to mitigate intra-class variations, they often struggle with instances exhibiting complex geometries or significant deviations from canonical shapes. To address this issue, we propose INKL-Pose, a novel category-level object pose estimation framework that enables INstance-adaptive Keypoint Learning with local-to-global geometric aggregation. Specifically, our method first predicts semantically consistent and geometrically informative keypoints using an Instance-Adaptive Keypoint Detector, then refines them: (1) a Local Keypoint Feature Aggregator capturing fine-grained geometries, and (2) a Global Keypoint Feature Aggregator using bidirectional Mamba for structural consistency. To enable bidirectional modeling in Mamba, we introduce a simple yet effective Feature Sequence Flipping strategy that preserves spatial coherence while constructing backward feature sequence. Additionally, we design a surface loss and a separation loss to encourage uniform coverage and spatial diversity in keypoint distribution. The resulting keypoints are mapped to a canonical space for 6D pose and size regression. Extensive experiments on CAMERA25, REAL275, and HouseCat6D show that INKL-Pose achieves state-of-the-art performance with 16.7M parameters and runs at 36 FPS on an NVIDIA RTX 4090D GPU.
♻ ☆ RefChartQA: Grounding Visual Answer on Chart Images through Instruction Tuning ICDAR 2025
Recently, Vision Language Models (VLMs) have increasingly emphasized document visual grounding to achieve better human-computer interaction, accessibility, and detailed understanding. However, its application to visualizations such as charts remains under-explored due to the inherent complexity of interleaved visual-numerical relationships in chart images. Existing chart understanding methods primarily focus on answering questions without explicitly identifying the visual elements that support their predictions. To bridge this gap, we introduce RefChartQA, a novel benchmark that integrates Chart Question Answering (ChartQA) with visual grounding, enabling models to refer elements at multiple granularities within chart images. Furthermore, we conduct a comprehensive evaluation by instruction-tuning 5 state-of-the-art VLMs across different categories. Our experiments demonstrate that incorporating spatial awareness via grounding improves response accuracy by over 15%, reducing hallucinations, and improving model reliability. Additionally, we identify key factors influencing text-spatial alignment, such as architectural improvements in TinyChart, which leverages a token-merging module for enhanced feature fusion. Our dataset is open-sourced for community development and further advancements. All models and code will be publicly available at https://github.com/moured/RefChartQA.
comment: Accepted by ICDAR 2025. All models and code will be publicly available at https://github.com/moured/RefChartQA
♻ ☆ Rasterizing Wireless Radiance Field via Deformable 2D Gaussian Splatting
Modeling the wireless radiance field (WRF) is fundamental to modern communication systems, enabling key tasks such as localization, sensing, and channel estimation. Traditional approaches, which rely on empirical formulas or physical simulations, often suffer from limited accuracy or require strong scene priors. Recent neural radiance field (NeRF-based) methods improve reconstruction fidelity through differentiable volumetric rendering, but their reliance on computationally expensive multilayer perceptron (MLP) queries hinders real-time deployment. To overcome these challenges, we introduce Gaussian splatting (GS) to the wireless domain, leveraging its efficiency in modeling optical radiance fields to enable compact and accurate WRF reconstruction. Specifically, we propose SwiftWRF, a deformable 2D Gaussian splatting framework that synthesizes WRF spectra at arbitrary positions under single-sided transceiver mobility. SwiftWRF employs CUDA-accelerated rasterization to render spectra at over 100000 fps and uses a lightweight MLP to model the deformation of 2D Gaussians, effectively capturing mobility-induced WRF variations. In addition to novel spectrum synthesis, the efficacy of SwiftWRF is further underscored in its applications in angle-of-arrival (AoA) and received signal strength indicator (RSSI) prediction. Experiments conducted on both real-world and synthetic indoor scenes demonstrate that SwiftWRF can reconstruct WRF spectra up to 500x faster than existing state-of-the-art methods, while significantly enhancing its signal quality. The project page is https://evan-sudo.github.io/swiftwrf/.
♻ ☆ A Bird Song Detector for improving bird identification through Deep Learning: a case study from Doñana
Passive Acoustic Monitoring is a key tool for biodiversity conservation, but the large volumes of unsupervised audio it generates present major challenges for extracting meaningful information. Deep Learning offers promising solutions. BirdNET, a widely used bird identification model, has shown success in many study systems but is limited at local scale due to biases in its training data, which focus on specific locations and target sounds rather than entire soundscapes. A key challenge in bird species identification is that many recordings either lack target species or contain overlapping vocalizations, complicating automatic identification. To address these problems, we developed a multi-stage pipeline for automatic bird vocalization identification in Do\~nana National Park (SW Spain), a wetland of high conservation concern. We deployed AudioMoth recorders in three main habitats across nine locations and manually annotated 461 minutes of audio, resulting in 3749 labeled segments spanning 34 classes. We first applied a Bird Song Detector to isolate bird vocalizations using spectrogram-based image processing. Then, species were classified using custom models trained at the local scale. Applying the Bird Song Detector before classification improved species identification, as all models performed better when analyzing only the segments where birds were detected. Specifically, the combination of detector and fine-tuned BirdNET outperformed the baseline without detection. This approach demonstrates the effectiveness of integrating a Bird Song Detector with local classification models. These findings highlight the need to adapt general-purpose tools to specific ecological challenges. Automatically detecting bird species helps track the health of this threatened ecosystem, given birds sensitivity to environmental change, and supports conservation planning to reduce biodiversity loss.
comment: 23 pages, 14 images, for associated dataset see https://huggingface.co/datasets/GrunCrow/BIRDeep_AudioAnnotations , for associated code see https://github.com/GrunCrow/BIRDeep_BirdSongDetector_NeuralNetworks and https://github.com/GrunCrow/Bird-Song-Detector
♻ ☆ Incorporating Pre-training Data Matters in Unsupervised Domain Adaptation
In deep learning, initializing models with pre-trained weights has become the de facto practice for various downstream tasks. Many unsupervised domain adaptation (UDA) methods typically adopt a backbone pre-trained on ImageNet, and focus on reducing the source-target domain discrepancy. However, the impact of pre-training on adaptation received little attention. In this study, we delve into UDA from the novel perspective of pre-training. We first demonstrate the impact of pre-training by analyzing the dynamic distribution discrepancies between pre-training data domain and the source/ target domain during adaptation. Then, we reveal that the target error also stems from the pre-training in the following two factors: 1) empirically, target error arises from the gradually degenerative pre-trained knowledge during adaptation; 2) theoretically, the error bound depends on difference between the gradient of loss function, \ie, on the target domain and pre-training data domain. To address these two issues, we redefine UDA as a three-domain problem, \ie, source domain, target domain, and pre-training data domain; then we propose a novel framework, named TriDA. We maintain the pre-trained knowledge and improve the error bound by incorporating pre-training data into adaptation for both vanilla UDA and source-free UDA scenarios. For efficiency, we introduce a selection strategy for pre-training data, and offer a solution with synthesized images when pre-training data is unavailable during adaptation. Notably, TriDA is effective even with a small amount of pre-training or synthesized images, and seamlessly complements the two scenario UDA methods, demonstrating state-of-the-art performance across multiple benchmarks. We hope our work provides new insights for better understanding and application of domain adaptation.
♻ ☆ Ophora: A Large-Scale Data-Driven Text-Guided Ophthalmic Surgical Video Generation Model MICCAI25
In ophthalmic surgery, developing an AI system capable of interpreting surgical videos and predicting subsequent operations requires numerous ophthalmic surgical videos with high-quality annotations, which are difficult to collect due to privacy concerns and labor consumption. Text-guided video generation (T2V) emerges as a promising solution to overcome this issue by generating ophthalmic surgical videos based on surgeon instructions. In this paper, we present Ophora, a pioneering model that can generate ophthalmic surgical videos following natural language instructions. To construct Ophora, we first propose a Comprehensive Data Curation pipeline to convert narrative ophthalmic surgical videos into a large-scale, high-quality dataset comprising over 160K video-instruction pairs, Ophora-160K. Then, we propose a Progressive Video-Instruction Tuning scheme to transfer rich spatial-temporal knowledge from a T2V model pre-trained on natural video-text datasets for privacy-preserved ophthalmic surgical video generation based on Ophora-160K. Experiments on video quality evaluation via quantitative analysis and ophthalmologist feedback demonstrate that Ophora can generate realistic and reliable ophthalmic surgical videos based on surgeon instructions. We also validate the capability of Ophora for empowering downstream tasks of ophthalmic surgical workflow understanding. Code is available at https://github.com/mar-cry/Ophora.
comment: Early accepted in MICCAI25
♻ ☆ SFDLA: Source-Free Document Layout Analysis ICDAR 2025
Document Layout Analysis (DLA) is a fundamental task in document understanding. However, existing DLA and adaptation methods often require access to large-scale source data and target labels. This requirements severely limiting their real-world applicability, particularly in privacy-sensitive and resource-constrained domains, such as financial statements, medical records, and proprietary business documents. According to our observation, directly transferring source-domain fine-tuned models on target domains often results in a significant performance drop (Avg. -32.64%). In this work, we introduce Source-Free Document Layout Analysis (SFDLA), aiming for adapting a pre-trained source DLA models to an unlabeled target domain, without access to any source data. To address this challenge, we establish the first SFDLA benchmark, covering three major DLA datasets for geometric- and content-aware adaptation. Furthermore, we propose Document Layout Analysis Adapter (DLAdapter), a novel framework that is designed to improve source-free adaptation across document domains. Our method achieves a +4.21% improvement over the source-only baseline and a +2.26% gain over existing source-free methods from PubLayNet to DocLayNet. We believe this work will inspire the DLA community to further investigate source-free document understanding. To support future research of the community, the benchmark, models, and code will be publicly available at https://github.com/s3setewe/sfdla-DLAdapter.
comment: Accepted by ICDAR 2025. The benchmark, models, and code will be publicly available at https://github.com/s3setewe/sfdla-DLAdapter
♻ ☆ Style-Preserving Lip Sync via Audio-Aware Style Reference
Audio-driven lip sync has recently drawn significant attention due to its widespread application in the multimedia domain. Individuals exhibit distinct lip shapes when speaking the same utterance, attributed to the unique speaking styles of individuals, posing a notable challenge for audio-driven lip sync. Earlier methods for such task often bypassed the modeling of personalized speaking styles, resulting in sub-optimal lip sync conforming to the general styles. Recent lip sync techniques attempt to guide the lip sync for arbitrary audio by aggregating information from a style reference video, yet they can not preserve the speaking styles well due to their inaccuracy in style aggregation. This work proposes an innovative audio-aware style reference scheme that effectively leverages the relationships between input audio and reference audio from style reference video to address the style-preserving audio-driven lip sync. Specifically, we first develop an advanced Transformer-based model adept at predicting lip motion corresponding to the input audio, augmented by the style information aggregated through cross-attention layers from style reference video. Afterwards, to better render the lip motion into realistic talking face video, we devise a conditional latent diffusion model, integrating lip motion through modulated convolutional layers and fusing reference facial images via spatial cross-attention layers. Extensive experiments validate the efficacy of the proposed approach in achieving precise lip sync, preserving speaking styles, and generating high-fidelity, realistic talking face videos.
comment: submitted to IEEE Transactions on Multimedia(TMM)
♻ ☆ The OCR Quest for Generalization: Learning to recognize low-resource alphabets with model editing
Achieving robustness in recognition systems across diverse domains is crucial for their practical utility. While ample data availability is usually assumed, low-resource languages, such as ancient manuscripts and non-western languages, tend to be kept out of the equations of massive pretraining and foundational techniques due to an under representation. In this work, we aim for building models which can generalize to new distributions of data, such as alphabets, faster than centralized fine-tune strategies. For doing so, we take advantage of the recent advancements in model editing to enhance the incorporation of unseen scripts (low-resource learning). In contrast to state-of-the-art meta-learning, we showcase the effectiveness of domain merging in sparse distributions of data, with agnosticity of its relation to the overall distribution or any other prototyping necessity. Even when using the same exact training data, our experiments showcase significant performance boosts in \textbf{transfer learning} to new alphabets and \textbf{out-of-domain evaluation} in challenging domain shifts, including historical ciphered texts and non-Latin scripts. This research contributes a novel approach into building models that can easily adopt under-represented alphabets and, therefore, enable document recognition to a wider set of contexts and cultures.
comment: Preprint (under review) For Journal
♻ ☆ SCAM: A Real-World Typographic Robustness Evaluation for Multimodal Foundation Models CVPR 2025
Typographic attacks exploit the interplay between text and visual content in multimodal foundation models, causing misclassifications when misleading text is embedded within images. However, existing datasets are limited in size and diversity, making it difficult to study such vulnerabilities. In this paper, we introduce SCAM, the largest and most diverse dataset of real-world typographic attack images to date, containing 1,162 images across hundreds of object categories and attack words. Through extensive benchmarking of Vision-Language Models (VLMs) on SCAM, we demonstrate that typographic attacks significantly degrade performance, and identify that training data and model architecture influence the susceptibility to these attacks. Our findings reveal that typographic attacks persist in state-of-the-art Large Vision-Language Models (LVLMs) due to the choice of their vision encoder, though larger Large Language Models (LLMs) backbones help mitigate their vulnerability. Additionally, we demonstrate that synthetic attacks closely resemble real-world (handwritten) attacks, validating their use in research. Our work provides a comprehensive resource and empirical insights to facilitate future research toward robust and trustworthy multimodal AI systems. We publicly release the datasets introduced in this paper along with the code for evaluations at www.bliss.berlin/research/scam.
comment: Accepted at CVPR 2025 Workshop EVAL-FoMo-2
♻ ☆ VideoMAR: Autoregressive Video Generatio with Continuous Tokens
Masked-based autoregressive models have demonstrated promising image generation capability in continuous space. However, their potential for video generation remains under-explored. In this paper, we propose \textbf{VideoMAR}, a concise and efficient decoder-only autoregressive image-to-video model with continuous tokens, composing temporal frame-by-frame and spatial masked generation. We first identify temporal causality and spatial bi-directionality as the first principle of video AR models, and propose the next-frame diffusion loss for the integration of mask and video generation. Besides, the huge cost and difficulty of long sequence autoregressive modeling is a basic but crucial issue. To this end, we propose the temporal short-to-long curriculum learning and spatial progressive resolution training, and employ progressive temperature strategy at inference time to mitigate the accumulation error. Furthermore, VideoMAR replicates several unique capacities of language models to video generation. It inherently bears high efficiency due to simultaneous temporal-wise KV cache and spatial-wise parallel generation, and presents the capacity of spatial and temporal extrapolation via 3D rotary embeddings. On the VBench-I2V benchmark, VideoMAR surpasses the previous state-of-the-art (Cosmos I2V) while requiring significantly fewer parameters ($9.3\%$), training data ($0.5\%$), and GPU resources ($0.2\%$).
♻ ☆ Pro-AD: Learning Comprehensive Prototypes with Prototype-based Constraint for Multi-class Unsupervised Anomaly Detection
Prototype-based reconstruction methods for unsupervised anomaly detection utilize a limited set of learnable prototypes which only aggregates insufficient normal information, resulting in undesirable reconstruction. However, increasing the number of prototypes may lead to anomalies being well reconstructed through the attention mechanism, which we refer to as the "Soft Identity Mapping" problem. In this paper, we propose Pro-AD to address these issues and fully utilize the prototypes to boost the performance of anomaly detection. Specifically, we first introduce an expanded set of learnable prototypes to provide sufficient capacity for semantic information. Then we employ a Dynamic Bidirectional Decoder which integrates the process of the normal information aggregation and the target feature reconstruction via prototypes, with the aim of allowing the prototypes to aggregate more comprehensive normal semantic information from different levels of the image features and the target feature reconstruction to not only utilize its contextual information but also dynamically leverage the learned comprehensive prototypes. Additionally, to prevent the anomalies from being well reconstructed using sufficient semantic information through the attention mechanism, Pro-AD introduces a Prototype-based Constraint that applied within the target feature reconstruction process of the decoder, which further improves the performance of our approach. Extensive experiments on multiple challenging benchmarks demonstrate that our Pro-AD achieve state-of-the-art performance, highlighting its superior robustness and practical effectiveness for Multi-class Unsupervised Anomaly Detection task.
♻ ☆ DRL-Based Resource Allocation for Motion Blur Resistant Federated Self-Supervised Learning in IoV
In the Internet of Vehicles (IoV), Federated Learning (FL) provides a privacy-preserving solution by aggregating local models without sharing data. Traditional supervised learning requires image data with labels, but data labeling involves significant manual effort. Federated Self-Supervised Learning (FSSL) utilizes Self-Supervised Learning (SSL) for local training in FL, eliminating the need for labels while protecting privacy. Compared to other SSL methods, Momentum Contrast (MoCo) reduces the demand for computing resources and storage space by creating a dictionary. However, using MoCo in FSSL requires uploading the local dictionary from vehicles to Base Station (BS), which poses a risk of privacy leakage. Simplified Contrast (SimCo) addresses the privacy leakage issue in MoCo-based FSSL by using dual temperature instead of a dictionary to control sample distribution. Additionally, considering the negative impact of motion blur on model aggregation, and based on SimCo, we propose a motion blur-resistant FSSL method, referred to as BFSSL. Furthermore, we address energy consumption and delay in the BFSSL process by proposing a Deep Reinforcement Learning (DRL)-based resource allocation scheme, called DRL-BFSSL. In this scheme, BS allocates the Central Processing Unit (CPU) frequency and transmission power of vehicles to minimize energy consumption and latency, while aggregating received models based on the motion blur level. Simulation results validate the effectiveness of our proposed aggregation and resource allocation methods.
comment: This paper has been accepted by IEEE Internet of Things Journal. The source code has been released at: https://github.com/qiongwu86/DRL-BFSSL
♻ ☆ FLARE: Towards Universal Dataset Purification against Backdoor Attacks
Deep neural networks (DNNs) are susceptible to backdoor attacks, where adversaries poison datasets with adversary-specified triggers to implant hidden backdoors, enabling malicious manipulation of model predictions. Dataset purification serves as a proactive defense by removing malicious training samples to prevent backdoor injection at its source. We first reveal that the current advanced purification methods rely on a latent assumption that the backdoor connections between triggers and target labels in backdoor attacks are simpler to learn than the benign features. We demonstrate that this assumption, however, does not always hold, especially in all-to-all (A2A) and untargeted (UT) attacks. As a result, purification methods that analyze the separation between the poisoned and benign samples in the input-output space or the final hidden layer space are less effective. We observe that this separability is not confined to a single layer but varies across different hidden layers. Motivated by this understanding, we propose FLARE, a universal purification method to counter various backdoor attacks. FLARE aggregates abnormal activations from all hidden layers to construct representations for clustering. To enhance separation, FLARE develops an adaptive subspace selection algorithm to isolate the optimal space for dividing an entire dataset into two clusters. FLARE assesses the stability of each cluster and identifies the cluster with higher stability as poisoned. Extensive evaluations on benchmark datasets demonstrate the effectiveness of FLARE against 22 representative backdoor attacks, including all-to-one (A2O), all-to-all (A2A), and untargeted (UT) attacks, and its robustness to adaptive attacks. Codes are available at \href{https://github.com/THUYimingLi/BackdoorBox}{BackdoorBox} and \href{https://github.com/vtu81/backdoor-toolbox}{backdoor-toolbox}.
comment: 15 pages, This paper is accepted and will appear in TIFS (CCF-A)
♻ ☆ EmoEdit: Evoking Emotions through Image Manipulation
Affective Image Manipulation (AIM) seeks to modify user-provided images to evoke specific emotional responses. This task is inherently complex due to its twofold objective: significantly evoking the intended emotion, while preserving the original image composition. Existing AIM methods primarily adjust color and style, often failing to elicit precise and profound emotional shifts. Drawing on psychological insights, we introduce EmoEdit, which extends AIM by incorporating content modifications to enhance emotional impact. Specifically, we first construct EmoEditSet, a large-scale AIM dataset comprising 40,120 paired data through emotion attribution and data construction. To make existing generative models emotion-aware, we design the Emotion adapter and train it using EmoEditSet. We further propose an instruction loss to capture the semantic variations in data pairs. Our method is evaluated both qualitatively and quantitatively, demonstrating superior performance compared to existing state-of-the-art techniques. Additionally, we showcase the portability of our Emotion adapter to other diffusion-based models, enhancing their emotion knowledge with diverse semantics.
♻ ☆ PanopticNeRF-360: Panoramic 3D-to-2D Label Transfer in Urban Scenes
Training perception systems for self-driving cars requires substantial 2D annotations that are labor-intensive to manual label. While existing datasets provide rich annotations on pre-recorded sequences, they fall short in labeling rarely encountered viewpoints, potentially hampering the generalization ability for perception models. In this paper, we present PanopticNeRF-360, a novel approach that combines coarse 3D annotations with noisy 2D semantic cues to generate high-quality panoptic labels and images from any viewpoint. Our key insight lies in exploiting the complementarity of 3D and 2D priors to mutually enhance geometry and semantics. Specifically, we propose to leverage coarse 3D bounding primitives and noisy 2D semantic and instance predictions to guide geometry optimization, by encouraging predicted labels to match panoptic pseudo ground truth. Simultaneously, the improved geometry assists in filtering 3D&2D annotation noise by fusing semantics in 3D space via a learned semantic field. To further enhance appearance, we combine MLP and hash grids to yield hybrid scene features, striking a balance between high-frequency appearance and contiguous semantics. Our experiments demonstrate PanopticNeRF-360's state-of-the-art performance over label transfer methods on the challenging urban scenes of the KITTI-360 dataset. Moreover, PanopticNeRF-360 enables omnidirectional rendering of high-fidelity, multi-view and spatiotemporally consistent appearance, semantic and instance labels. We make our code and data available at https://github.com/fuxiao0719/PanopticNeRF
comment: TPAMI 2025. Project page: http://fuxiao0719.github.io/projects/panopticnerf360/ Code: https://github.com/fuxiao0719/PanopticNeRF/tree/panopticnerf360
♻ ☆ Improving LLM Video Understanding with 16 Frames Per Second
Human vision is dynamic and continuous. However, in video understanding with multimodal large language models (LLMs), existing methods primarily rely on static features extracted from images sampled at a fixed low frame rate of frame-per-second (FPS) $\leqslant$2, leading to critical visual information loss. In this paper, we introduce F-16, the first multimodal LLM designed for high-frame-rate video understanding. By increasing the frame rate to 16 FPS and compressing visual tokens within each 1-second clip, F-16 efficiently captures dynamic visual features while preserving key semantic information. Experimental results demonstrate that higher frame rates considerably enhance video understanding across multiple benchmarks, providing a new approach to improving video LLMs beyond scaling model size or training data. F-16 achieves state-of-the-art performance among 7-billion-parameter video LLMs on both general and fine-grained video understanding benchmarks, such as Video-MME and TemporalBench. Furthermore, F-16 excels in complex spatiotemporal tasks, including high-speed sports analysis (\textit{e.g.}, basketball, football, gymnastics, and diving), outperforming SOTA proprietary visual models like GPT-4o and Gemini-1.5-pro. Additionally, we introduce a novel decoding method for F-16 that enables highly efficient low-frame-rate inference without requiring model retraining. We will release the source code, model checkpoints, and data at \href{https://github.com/bytedance/F-16}{https://github.com/bytedance/F-16}.
♻ ☆ PRO: Projection Domain Synthesis for CT Imaging
Synthesizing high quality CT projection data remains a significant challenge due to the limited availability of annotated data and the complex nature of CT imaging. In this work, we present PRO, a projection domain synthesis foundation model for CT imaging. To the best of our knowledge, this is the first study that performs CT synthesis in the projection domain. Unlike previous approaches that operate in the image domain, PRO learns rich structural representations from raw projection data and leverages anatomical text prompts for controllable synthesis. This projection domain strategy enables more faithful modeling of underlying imaging physics and anatomical structures. Moreover, PRO functions as a foundation model, capable of generalizing across diverse downstream tasks by adjusting its generative behavior via prompt inputs. Experimental results demonstrated that incorporating our synthesized data significantly improves performance across multiple downstream tasks, including low-dose and sparse-view reconstruction. These findings underscore the versatility and scalability of PRO in data generation for various CT applications. These results highlight the potential of projection domain synthesis as a powerful tool for data augmentation and robust CT imaging. Our source code is publicly available at: https://github.com/yqx7150/PRO.
♻ ☆ Jailbreak Large Vision-Language Models Through Multi-Modal Linkage
With the significant advancement of Large Vision-Language Models (VLMs), concerns about their potential misuse and abuse have grown rapidly. Previous studies have highlighted VLMs' vulnerability to jailbreak attacks, where carefully crafted inputs can lead the model to produce content that violates ethical and legal standards. However, existing methods struggle against state-of-the-art VLMs like GPT-4o, due to the over-exposure of harmful content and lack of stealthy malicious guidance. In this work, we propose a novel jailbreak attack framework: Multi-Modal Linkage (MML) Attack. Drawing inspiration from cryptography, MML utilizes an encryption-decryption process across text and image modalities to mitigate over-exposure of malicious information. To align the model's output with malicious intent covertly, MML employs a technique called "evil alignment", framing the attack within a video game production scenario. Comprehensive experiments demonstrate MML's effectiveness. Specifically, MML jailbreaks GPT-4o with attack success rates of 97.80% on SafeBench, 98.81% on MM-SafeBench and 99.07% on HADES-Dataset. Our code is available at https://github.com/wangyu-ovo/MML.
♻ ☆ ImmerseGen: Agent-Guided Immersive World Generation with Alpha-Textured Proxies
Automatic creation of 3D scenes for immersive VR presence has been a significant research focus for decades. However, existing methods often rely on either high-poly mesh modeling with post-hoc simplification or massive 3D Gaussians, resulting in a complex pipeline or limited visual realism. In this paper, we demonstrate that such exhaustive modeling is unnecessary for achieving compelling immersive experience. We introduce ImmerseGen, a novel agent-guided framework for compact and photorealistic world modeling. ImmerseGen represents scenes as hierarchical compositions of lightweight geometric proxies, i.e., simplified terrain and billboard meshes, and generates photorealistic appearance by synthesizing RGBA textures onto these proxies. Specifically, we propose terrain-conditioned texturing for user-centric base world synthesis, and RGBA asset texturing for midground and foreground scenery. This reformulation offers several advantages: (i) it simplifies modeling by enabling agents to guide generative models in producing coherent textures that integrate seamlessly with the scene; (ii) it bypasses complex geometry creation and decimation by directly synthesizing photorealistic textures on proxies, preserving visual quality without degradation; (iii) it enables compact representations suitable for real-time rendering on mobile VR headsets. To automate scene creation from text prompts, we introduce VLM-based modeling agents enhanced with semantic grid-based analysis for improved spatial reasoning and accurate asset placement. ImmerseGen further enriches scenes with dynamic effects and ambient audio to support multisensory immersion. Experiments on scene generation and live VR showcases demonstrate that ImmerseGen achieves superior photorealism, spatial coherence and rendering efficiency compared to prior methods. Project webpage: https://immersegen.github.io.
comment: Project webpage: https://immersegen.github.io
♻ ☆ SUEDE:Shared Unified Experts for Physical-Digital Face Attack Detection Enhancement ICME 2025
Face recognition systems are vulnerable to physical attacks (e.g., printed photos) and digital threats (e.g., DeepFake), which are currently being studied as independent visual tasks, such as Face Anti-Spoofing and Forgery Detection. The inherent differences among various attack types present significant challenges in identifying a common feature space, making it difficult to develop a unified framework for detecting data from both attack modalities simultaneously. Inspired by the efficacy of Mixture-of-Experts (MoE) in learning across diverse domains, we explore utilizing multiple experts to learn the distinct features of various attack types. However, the feature distributions of physical and digital attacks overlap and differ. This suggests that relying solely on distinct experts to learn the unique features of each attack type may overlook shared knowledge between them. To address these issues, we propose SUEDE, the Shared Unified Experts for Physical-Digital Face Attack Detection Enhancement. SUEDE combines a shared expert (always activated) to capture common features for both attack types and multiple routed experts (selectively activated) for specific attack types. Further, we integrate CLIP as the base network to ensure the shared expert benefits from prior visual knowledge and align visual-text representations in a unified space. Extensive results demonstrate SUEDE achieves superior performance compared to state-of-the-art unified detection methods.
comment: Accepted in ICME 2025 (Oral)
♻ ☆ MOS: Model Surgery for Pre-Trained Model-Based Class-Incremental Learning AAAI 2025
Class-Incremental Learning (CIL) requires models to continually acquire knowledge of new classes without forgetting old ones. Despite Pre-trained Models (PTMs) have shown excellent performance in CIL, catastrophic forgetting still occurs as the model learns new concepts. Existing work seeks to utilize lightweight components to adjust the PTM, while the forgetting phenomenon still comes from {\em parameter and retrieval} levels. Specifically, iterative updates of the model result in parameter drift, while mistakenly retrieving irrelevant modules leads to the mismatch during inference. To this end, we propose MOdel Surgery (MOS) to rescue the model from forgetting previous knowledge. By training task-specific adapters, we continually adjust the PTM to downstream tasks. To mitigate parameter-level forgetting, we present an adapter merging approach to learn task-specific adapters, which aims to bridge the gap between different components while reserve task-specific information. Besides, to address retrieval-level forgetting, we introduce a training-free self-refined adapter retrieval mechanism during inference, which leverages the model's inherent ability for better adapter retrieval. By jointly rectifying the model with those steps, MOS can robustly resist catastrophic forgetting in the learning process. Extensive experiments on seven benchmark datasets validate MOS's state-of-the-art performance. Code is available at: https://github.com/sun-hailong/AAAI25-MOS
comment: Accepted to AAAI 2025. Code is available at: https://github.com/sun-hailong/AAAI25-MOS
♻ ☆ A Curated and Re-annotated Peripheral Blood Cell Dataset Integrating Four Public Resources
We present TXL-PBC, a curated and re-annotated peripheral blood cell dataset constructed by integrating four publicly available resources: Blood Cell Count and Detection (BCCD), Blood Cell Detection Dataset (BCDD), Peripheral Blood Cells (PBC), and Raabin White Blood Cell (Raabin-WBC). Through rigorous sample selection, semi-automatic annotation using the YOLOv8n model, and comprehensive manual review, we ensured high annotation accuracy and consistency. The final dataset contains 1,260 images and 18,143 bounding box annotations for three major blood cell types: white blood cells (WBC), red blood cells (RBC), and platelets. We provide detailed visual analyses of the data distribution, demonstrating the diversity and balance of the dataset. To further validate the quality and utility of TXL-PBC, we trained several mainstream object detection models, including YOLOv5s, YOLOv8s, YOLOv11s, SSD300, Faster R-CNN, and RetinaNet, and report their baseline performance. The TXL-PBC dataset is openly available on Figshare and GitHub, offering a valuable resource for the development and benchmarking of blood cell detection models and related machine learning research.
♻ ☆ Data Augmentation Through Random Style Replacement
In this paper, we introduce a novel data augmentation technique that combines the advantages of style augmentation and random erasing by selectively replacing image subregions with style-transferred patches. Our approach first applies a random style transfer to training images, then randomly substitutes selected areas of these images with patches derived from the style-transferred versions. This method is able to seamlessly accommodate a wide range of existing style transfer algorithms and can be readily integrated into diverse data augmentation pipelines. By incorporating our strategy, the training process becomes more robust and less prone to overfitting. Comparative experiments demonstrate that, relative to previous style augmentation methods, our technique achieves superior performance and faster convergence.
comment: Accepted by 2025 6th International Conference on Computer Vision, Image and Deep Learning
♻ ☆ SurgSora: Object-Aware Diffusion Model for Controllable Surgical Video Generation
Surgical video generation can enhance medical education and research, but existing methods lack fine-grained motion control and realism. We introduce SurgSora, a framework that generates high-fidelity, motion-controllable surgical videos from a single input frame and user-specified motion cues. Unlike prior approaches that treat objects indiscriminately or rely on ground-truth segmentation masks, SurgSora leverages self-predicted object features and depth information to refine RGB appearance and optical flow for precise video synthesis. It consists of three key modules: (1) the Dual Semantic Injector, which extracts object-specific RGB-D features and segmentation cues to enhance spatial representations; (2) the Decoupled Flow Mapper, which fuses multi-scale optical flow with semantic features for realistic motion dynamics; and (3) the Trajectory Controller, which estimates sparse optical flow and enables user-guided object movement. By conditioning these enriched features within the Stable Video Diffusion, SurgSora achieves state-of-the-art visual authenticity and controllability in advancing surgical video synthesis, as demonstrated by extensive quantitative and qualitative comparisons. Our human evaluation in collaboration with expert surgeons further demonstrates the high realism of SurgSora-generated videos, highlighting the potential of our method for surgical training and education. Our project is available at https://surgsora.github.io/surgsora.github.io.
♻ ☆ Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think ICLR 2025
Recent studies have shown that the denoising process in (generative) diffusion models can induce meaningful (discriminative) representations inside the model, though the quality of these representations still lags behind those learned through recent self-supervised learning methods. We argue that one main bottleneck in training large-scale diffusion models for generation lies in effectively learning these representations. Moreover, training can be made easier by incorporating high-quality external visual representations, rather than relying solely on the diffusion models to learn them independently. We study this by introducing a straightforward regularization called REPresentation Alignment (REPA), which aligns the projections of noisy input hidden states in denoising networks with clean image representations obtained from external, pretrained visual encoders. The results are striking: our simple strategy yields significant improvements in both training efficiency and generation quality when applied to popular diffusion and flow-based transformers, such as DiTs and SiTs. For instance, our method can speed up SiT training by over 17.5$\times$, matching the performance (without classifier-free guidance) of a SiT-XL model trained for 7M steps in less than 400K steps. In terms of final generation quality, our approach achieves state-of-the-art results of FID=1.42 using classifier-free guidance with the guidance interval.
comment: ICLR 2025 (Oral). Project page: https://sihyun.me/REPA
♻ ☆ Generative diffusion model surrogates for mechanistic agent-based biological models
Mechanistic, multicellular, agent-based models are commonly used to investigate tissue, organ, and organism-scale biology at single-cell resolution. The Cellular-Potts Model (CPM) is a powerful and popular framework for developing and interrogating these models. CPMs become computationally expensive at large space- and time- scales making application and investigation of developed models difficult. Surrogate models may allow for the accelerated evaluation of CPMs of complex biological systems. However, the stochastic nature of these models means each set of parameters may give rise to different model configurations, complicating surrogate model development. In this work, we leverage denoising diffusion probabilistic models to train a generative AI surrogate of a CPM used to investigate in vitro vasculogenesis. We describe the use of an image classifier to learn the characteristics that define unique areas of a 2-dimensional parameter space. We then apply this classifier to aid in surrogate model selection and verification. Our CPM model surrogate generates model configurations 20,000 timesteps ahead of a reference configuration and demonstrates approximately a 22x reduction in computational time as compared to native code execution. Our work represents a step towards the implementation of DDPMs to develop digital twins of stochastic biological systems.
♻ ☆ MSVIT: Improving Spiking Vision Transformer Using Multi-scale Attention Fusion IJCAI'25
The combination of Spiking Neural Networks (SNNs) with Vision Transformer architectures has garnered significant attention due to their potential for energy-efficient and high-performance computing paradigms. However, a substantial performance gap still exists between SNN-based and ANN-based transformer architectures. While existing methods propose spiking self-attention mechanisms that are successfully combined with SNNs, the overall architectures proposed by these methods suffer from a bottleneck in effectively extracting features from different image scales. In this paper, we address this issue and propose MSVIT. This novel spike-driven Transformer architecture firstly uses multi-scale spiking attention (MSSA) to enhance the capabilities of spiking attention blocks. We validate our approach across various main datasets. The experimental results show that MSVIT outperforms existing SNN-based models, positioning itself as a state-of-the-art solution among SNN-transformer architectures. The codes are available at https://github.com/Nanhu-AI-Lab/MSViT.
comment: 11pages, 2figures, accepted by IJCAI'25 (34th International Joint Conference on Artificial Intelligence)
♻ ☆ PLD: A Choice-Theoretic List-Wise Knowledge Distillation
Knowledge distillation is a model compression technique in which a compact "student" network is trained to replicate the predictive behavior of a larger "teacher" network. In logit-based knowledge distillation it has become the de facto approach to augment cross-entropy with a distillation term. Typically this term is either a KL divergence-matching marginal probabilities or a correlation-based loss capturing intra- and inter-class relationships but in every case it sits as an add-on to cross-entropy with its own weight that must be carefully tuned. In this paper we adopt a choice-theoretic perspective and recast knowledge distillation under the Plackett-Luce model by interpreting teacher logits as "worth" scores. We introduce Plackett-Luce Distillation (PLD), a weighted list-wise ranking loss in which the teacher model transfers knowledge of its full ranking of classes, weighting each ranked choice by its own confidence. PLD directly optimizes a single teacher-optimal ranking of the true label first, followed by the remaining classes in descending teacher confidence, yielding a convex, translation-invariant surrogate that subsumes weighted cross-entropy. Empirically on standard image classification benchmarks, PLD improves Top-1 accuracy by an average of +0.42% over DIST (arXiv:2205.10536) and +1.04% over KD (arXiv:1503.02531) in homogeneous settings and by +0.48% and +1.09% over DIST and KD, respectively, in heterogeneous settings.
♻ ☆ Bi-VLDoc: Bidirectional Vision-Language Modeling for Visually-Rich Document Understanding
Multi-modal document pre-trained models have proven to be very effective in a variety of visually-rich document understanding (VrDU) tasks. Though existing document pre-trained models have achieved excellent performance on standard benchmarks for VrDU, the way they model and exploit the interactions between vision and language on documents has hindered them from better generalization ability and higher accuracy. In this work, we investigate the problem of vision-language joint representation learning for VrDU mainly from the perspective of supervisory signals. Specifically, a pre-training paradigm called Bi-VLDoc is proposed, in which a bidirectional vision-language supervision strategy and a vision-language hybrid-attention mechanism are devised to fully explore and utilize the interactions between these two modalities, to learn stronger cross-modal document representations with richer semantics. Benefiting from the learned informative cross-modal document representations, Bi-VLDoc significantly advances the state-of-the-art performance on three widely-used document understanding benchmarks, including Form Understanding (from 85.14% to 93.44%), Receipt Information Extraction (from 96.01% to 97.84%), and Document Classification (from 96.08% to 97.12%). On Document Visual QA, Bi-VLDoc achieves the state-of-the-art performance compared to previous single model methods.
comment: IJDAR 2025
♻ ☆ Multiclass Post-Earthquake Building Assessment Integrating High-Resolution Optical and SAR Satellite Imagery, Ground Motion, and Soil Data with Transformers
Timely and accurate assessments of building damage are crucial for effective response and recovery in the aftermath of earthquakes. Conventional preliminary damage assessments (PDA) often rely on manual door-to-door inspections, which are not only time-consuming but also pose significant safety risks. To safely expedite the PDA process, researchers have studied the applicability of satellite imagery processed with heuristic and machine learning approaches. These approaches output binary or, more recently, multiclass damage states at the scale of a block or a single building. However, the current performance of such approaches limits practical applicability. To address this limitation, we introduce a metadata-enriched, transformer based framework that combines high-resolution post-earthquake satellite imagery with building-specific metadata relevant to the seismic performance of the structure. Our model achieves state-of-the-art performance in multiclass post-earthquake damage identification for buildings from the Turkey-Syria earthquake on February 6, 2023. Specifically, we demonstrate that incorporating metadata, such as seismic intensity indicators, soil properties, and SAR damage proxy maps not only enhances the model's accuracy and ability to distinguish between damage classes, but also improves its generalizability across various regions. Furthermore, we conducted a detailed, class-wise analysis of feature importance to understand the model's decision-making across different levels of building damage. This analysis reveals how individual metadata features uniquely contribute to predictions for each damage class. By leveraging both satellite imagery and metadata, our proposed framework enables faster and more accurate damage assessments for precise, multiclass, building-level evaluations that can improve disaster response and accelerate recovery efforts for affected communities.
comment: 28 Pages, 12 Figures
♻ ☆ AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding
Multimodal Large Language Models (MLLMs) struggle with long videos due to fixed context windows and weak long-term dependency modeling. Existing Retrieval-Augmented Generation (RAG) methods for videos use static retrieval strategies, leading to inefficiencies for simple queries and information loss for complex tasks. To address this, we propose AdaVideoRAG, a novel framework that dynamically adapts retrieval granularity based on query complexity using a lightweight intent classifier. Our framework employs an Omni-Knowledge Indexing module to build hierarchical databases from text (captions, ASR, OCR), visual features, and semantic graphs, enabling optimal resource allocation across tasks. We also introduce the HiVU benchmark for comprehensive evaluation. Experiments demonstrate improved efficiency and accuracy for long-video understanding, with seamless integration into existing MLLMs. AdaVideoRAG establishes a new paradigm for adaptive retrieval in video analysis. Codes will be open-sourced at https://github.com/xzc-zju/AdaVideoRAG.
♻ ☆ Click-Calib: A Robust Extrinsic Calibration Method for Surround-View Systems
Surround-View System (SVS) is an essential component in Advanced Driver Assistance System (ADAS) and requires precise calibrations. However, conventional offline extrinsic calibration methods are cumbersome and time-consuming as they rely heavily on physical patterns. Additionally, these methods primarily focus on short-range areas surrounding the vehicle, resulting in lower calibration quality in more distant zones. To address these limitations, we propose Click-Calib, a pattern-free approach for offline SVS extrinsic calibration. Without requiring any special setup, the user only needs to click a few keypoints on the ground in natural scenes. Unlike other offline calibration approaches, Click-Calib optimizes camera poses over a wide range by minimizing reprojection distance errors of keypoints, thereby achieving accurate calibrations at both short and long distances. Furthermore, Click-Calib supports both single-frame and multiple-frame modes, with the latter offering even better results. Evaluations on our in-house dataset and the public WoodScape dataset demonstrate its superior accuracy and robustness compared to baseline methods. Code is available at https://github.com/lwangvaleo/click_calib.
♻ ☆ SemVink: Advancing VLMs' Semantic Understanding of Optical Illusions via Visual Global Thinking
Vision-language models (VLMs) excel in semantic tasks but falter at a core human capability: detecting hidden content in optical illusions or AI-generated images through perceptual adjustments like zooming. We introduce HC-Bench, a benchmark of 112 images with hidden text, objects, and illusions, revealing that leading VLMs achieve near-zero accuracy (0-5.36%)-even with explicit prompting. Humans resolve such ambiguities instinctively, yet VLMs fail due to an overreliance on high-level semantics. Strikingly, we propose SemVink (Semantic Visual Thinking) by simply scaling images to low resolutions (32-128 pixels), which unlocks >99% accuracy by eliminating redundant visual noise. This exposes a critical architectural flaw: VLMs prioritize abstract reasoning over low-level visual operations crucial for real-world robustness. Our work urges a shift toward hybrid models integrating multi-scale processing, bridging the gap between computational vision and human cognition for applications in medical imaging, security, and beyond.
♻ ☆ Towards Cross-Subject EMG Pattern Recognition via Dual-Branch Adversarial Feature Disentanglement
Cross-subject electromyography (EMG) pattern recognition faces significant challenges due to inter-subject variability in muscle anatomy, electrode placement, and signal characteristics. Traditional methods rely on subject-specific calibration data to adapt models to new users, an approach that is both time-consuming and impractical for large-scale, real-world deployment. This paper presents an approach to eliminate calibration requirements through feature disentanglement, enabling effective cross-subject generalization. We propose an end-to-end dual-branch adversarial neural network that simultaneously performs pattern recognition and individual identification by disentangling EMG features into pattern-specific and subject-specific components. The pattern-specific components facilitate robust pattern recognition for new users without model calibration, while the subject-specific components enable downstream applications such as task-invariant biometric identification. Experimental results demonstrate that the proposed model achieves robust performance on data from unseen users, outperforming various baseline methods in cross-subject scenarios. Overall, this study offers a new perspective for cross-subject EMG pattern recognition without model calibration and highlights the proposed model's potential for broader applications, such as task-independent biometric systems.
comment: 6 pages, 3 figures. This work has been accepted for presentation at the IEEE Engineering in Medicine and Biology Conference (EMBC) 2025. New version corrects numerical errors in Table 1. Conclusions are unaffected
Information Retrieval 13
☆ DiscRec: Disentangled Semantic-Collaborative Modeling for Generative Recommendation
Generative recommendation is emerging as a powerful paradigm that directly generates item predictions, moving beyond traditional matching-based approaches. However, current methods face two key challenges: token-item misalignment, where uniform token-level modeling ignores item-level granularity that is critical for collaborative signal learning, and semantic-collaborative signal entanglement, where collaborative and semantic signals exhibit distinct distributions yet are fused in a unified embedding space, leading to conflicting optimization objectives that limit the recommendation performance. To address these issues, we propose DiscRec, a novel framework that enables Disentangled Semantic-Collaborative signal modeling with flexible fusion for generative Recommendation.First, DiscRec introduces item-level position embeddings, assigned based on indices within each semantic ID, enabling explicit modeling of item structure in input token sequences.Second, DiscRec employs a dual-branch module to disentangle the two signals at the embedding layer: a semantic branch encodes semantic signals using original token embeddings, while a collaborative branch applies localized attention restricted to tokens within the same item to effectively capture collaborative signals. A gating mechanism subsequently fuses both branches while preserving the model's ability to model sequential dependencies. Extensive experiments on four real-world datasets demonstrate that DiscRec effectively decouples these signals and consistently outperforms state-of-the-art baselines. Our codes are available on https://github.com/Ten-Mao/DiscRec.
☆ Multi-Interest Recommendation: A Survey
Existing recommendation methods often struggle to model users' multifaceted preferences due to the diversity and volatility of user behavior, as well as the inherent uncertainty and ambiguity of item attributes in practical scenarios. Multi-interest recommendation addresses this challenge by extracting multiple interest representations from users' historical interactions, enabling fine-grained preference modeling and more accurate recommendations. It has drawn broad interest in recommendation research. However, current recommendation surveys have either specialized in frontier recommendation methods or delved into specific tasks and downstream applications. In this work, we systematically review the progress, solutions, challenges, and future directions of multi-interest recommendation by answering the following three questions: (1) Why is multi-interest modeling significantly important for recommendation? (2) What aspects are focused on by multi-interest modeling in recommendation? and (3) How can multi-interest modeling be applied, along with the technical details of the representative modules? We hope that this survey establishes a fundamental framework and delivers a preliminary overview for researchers interested in this field and committed to further exploration. The implementation of multi-interest recommendation summarized in this survey is maintained at https://github.com/WHUIR/Multi-Interest-Recommendation-A-Survey.
☆ Next-User Retrieval: Enhancing Cold-Start Recommendations via Generative Next-User Modeling
The item cold-start problem is critical for online recommendation systems, as the success of this phase determines whether high-quality new items can transition to popular ones, receive essential feedback to inspire creators, and thus lead to the long-term retention of creators. However, modern recommendation systems still struggle to address item cold-start challenges due to the heavy reliance on item and historical interactions, which are non-trivial for cold-start items lacking sufficient exposure and feedback. Lookalike algorithms provide a promising solution by extending feedback for new items based on lookalike users. Traditional lookalike algorithms face such limitations: (1) failing to effectively model the lookalike users and further improve recommendations with the existing rule- or model-based methods; and (2) struggling to utilize the interaction signals and incorporate diverse features in modern recommendation systems. Inspired by lookalike algorithms, we propose Next-User Retrieval, a novel framework for enhancing cold-start recommendations via generative next-user modeling. Specifically, we employ a transformer-based model to capture the unidirectional relationships among recently interacted users and utilize these sequences to generate the next potential user who is most likely to interact with the item. The additional item features are also integrated as prefix prompt embeddings to assist the next-user generation. The effectiveness of Next-User Retrieval is evaluated through both offline experiments and online A/B tests. Our method achieves significant improvements with increases of 0.0142% in daily active users and +0.1144% in publications in Douyin, showcasing its practical applicability and scalability.
☆ Advancing Loss Functions in Recommender Systems: A Comparative Study with a Rényi Divergence-Based Solution AAAI 2025
Loss functions play a pivotal role in optimizing recommendation models. Among various loss functions, Softmax Loss (SL) and Cosine Contrastive Loss (CCL) are particularly effective. Their theoretical connections and differences warrant in-depth exploration. This work conducts comprehensive analyses of these losses, yielding significant insights: 1) Common strengths -- both can be viewed as augmentations of traditional losses with Distributional Robust Optimization (DRO), enhancing robustness to distributional shifts; 2) Respective limitations -- stemming from their use of different distribution distance metrics in DRO optimization, SL exhibits high sensitivity to false negative instances, whereas CCL suffers from low data utilization. To address these limitations, this work proposes a new loss function, DrRL, which generalizes SL and CCL by leveraging R\'enyi-divergence in DRO optimization. DrRL incorporates the advantageous structures of both SL and CCL, and can be demonstrated to effectively mitigate their limitations. Extensive experiments have been conducted to validate the superiority of DrRL on both recommendation accuracy and robustness.
comment: AAAI 2025
☆ MoR: Better Handling Diverse Queries with a Mixture of Sparse, Dense, and Human Retrievers
Retrieval-augmented Generation (RAG) is powerful, but its effectiveness hinges on which retrievers we use and how. Different retrievers offer distinct, often complementary signals: BM25 captures lexical matches; dense retrievers, semantic similarity. Yet in practice, we typically fix a single retriever based on heuristics, which fails to generalize across diverse information needs. Can we dynamically select and integrate multiple retrievers for each individual query, without the need for manual selection? In our work, we validate this intuition with quantitative analysis and introduce mixture of retrievers: a zero-shot, weighted combination of heterogeneous retrievers. Extensive experiments show that such mixtures are effective and efficient: Despite totaling just 0.8B parameters, this mixture outperforms every individual retriever and even larger 7B models by +10.8% and +3.9% on average, respectively. Further analysis also shows that this mixture framework can help incorporate specialized non-oracle human information sources as retrievers to achieve good collaboration, with a 58.9% relative performance improvement over simulated humans alone.
comment: 19 pages, 3 figures
☆ MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents
Modern language agents must operate over long-horizon, multi-turn interactions, where they retrieve external information, adapt to observations, and answer interdependent queries. Yet, most LLM systems rely on full-context prompting, appending all past turns regardless of their relevance. This leads to unbounded memory growth, increased computational costs, and degraded reasoning performance on out-of-distribution input lengths. We introduce MEM1, an end-to-end reinforcement learning framework that enables agents to operate with constant memory across long multi-turn tasks. At each turn, MEM1 updates a compact shared internal state that jointly supports memory consolidation and reasoning. This state integrates prior memory with new observations from the environment while strategically discarding irrelevant or redundant information. To support training in more realistic and compositional settings, we propose a simple yet effective and scalable approach to constructing multi-turn environments by composing existing datasets into arbitrarily complex task sequences. Experiments across three domains, including internal retrieval QA, open-domain web QA, and multi-turn web shopping, show that MEM1-7B improves performance by 3.5x while reducing memory usage by 3.7x compared to Qwen2.5-14B-Instruct on a 16-objective multi-hop QA task, and generalizes beyond the training horizon. Our results demonstrate the promise of reasoning-driven memory consolidation as a scalable alternative to existing solutions for training long-horizon interactive agents, where both efficiency and performance are optimized.
☆ Architecture is All You Need: Improving LLM Recommenders by Dropping the Text
In recent years, there has been an explosion of interest in the applications of large pre-trained language models (PLMs) to recommender systems, with many studies showing strong performance of PLMs on common benchmark datasets. PLM-based recommender models benefit from flexible and customizable prompting, an unlimited vocabulary of recommendable items, and general ``world knowledge'' acquired through pre-training on massive text corpora. While PLM-based recommenders show promise in settings where data is limited, they are hard to implement in practice due to their large size and computational cost. Additionally, fine-tuning PLMs to improve performance on collaborative signals may degrade the model's capacity for world knowledge and generalizability. We propose a recommender model that uses the architecture of large language models (LLMs) while reducing layer count and dimensions and replacing the text-based subword tokenization of a typical LLM with discrete tokens that uniquely represent individual content items. We find that this simplified approach substantially outperforms both traditional sequential recommender models and PLM-based recommender models at a tiny fraction of the size and computational complexity of PLM-based models. Our results suggest that the principal benefit of LLMs in recommender systems is their architecture, rather than the world knowledge acquired during extensive pre-training.
comment: 7 pages, 1 figure
☆ cAST: Enhancing Code Retrieval-Augmented Generation with Structural Chunking via Abstract Syntax Tree
Retrieval-Augmented Generation (RAG) has become essential for large-scale code generation, grounding predictions in external code corpora to improve actuality. However, a critical yet underexplored aspect of RAG pipelines is chunking -- the process of dividing documents into retrievable units. Existing line-based chunking heuristics often break semantic structures, splitting functions or merging unrelated code, which can degrade generation quality. We propose chunking via Abstract Syntax Trees (\ourwork), a structure-aware method that recursively breaks large AST nodes into smaller chunks and merges sibling nodes while respecting size limits. This approach generates self-contained, semantically coherent units across programming languages and tasks, improving performance on diverse code generation tasks, e.g., boosting Recall@5 by 4.3 points on RepoEval retrieval and Pass@1 by 2.67 points on SWE-bench generation. Our work highlights the importance of structure-aware chunking for scaling retrieval-enhanced code intelligence.
♻ ☆ Aug2Search: Enhancing Facebook Marketplace Search with LLM-Generated Synthetic Data Augmentation
Embedding-Based Retrieval (EBR) is an important technique in modern search engines, enabling semantic match between search queries and relevant results. However, search logging data on platforms like Facebook Marketplace lacks the diversity and details needed for effective EBR model training, limiting the models' ability to capture nuanced search patterns. To address this challenge, we propose Aug2Search, an EBR-based framework leveraging synthetic data generated by Generative AI (GenAI) models, in a multimodal and multitask approach to optimize query-product relevance. This paper investigates the capabilities of GenAI, particularly Large Language Models (LLMs), in generating high-quality synthetic data, and analyzing its impact on enhancing EBR models. We conducted experiments using eight Llama models and 100 million data points from Facebook Marketplace logs. Our synthetic data generation follows three strategies: (1) generate queries, (2) enhance product listings, and (3) generate queries from enhanced listings. We train EBR models on three different datasets: sampled engagement data or original data ((e.g., "Click" and "Listing Interactions")), synthetic data, and a mixture of both engagement and synthetic data to assess their performance across various training sets. Our findings underscore the robustness of Llama models in producing synthetic queries and listings with high coherence, relevance, and diversity, while maintaining low levels of hallucination. Aug2Search achieves an improvement of up to 4% in ROC_AUC with 100 million synthetic data samples, demonstrating the effectiveness of our approach. Moreover, our experiments reveal that with the same volume of training data, models trained exclusively on synthetic data often outperform those trained on original data only or a mixture of original and synthetic data.
♻ ☆ Contributions to Representation Learning with Graph Autoencoders and Applications to Music Recommendation
Graph autoencoders (GAE) and variational graph autoencoders (VGAE) emerged as two powerful groups of unsupervised node embedding methods, with various applications to graph-based machine learning problems such as link prediction and community detection. Nonetheless, at the beginning of this Ph.D. project, GAE and VGAE models were also suffering from key limitations, preventing them from being adopted in the industry. In this thesis, we present several contributions to improve these models, with the general aim of facilitating their use to address industrial-level problems involving graph representations. Firstly, we propose two strategies to overcome the scalability issues of previous GAE and VGAE models, permitting to effectively train these models on large graphs with millions of nodes and edges. These strategies leverage graph degeneracy and stochastic subgraph decoding techniques, respectively. Besides, we introduce Gravity-Inspired GAE and VGAE, providing the first extensions of these models for directed graphs, that are ubiquitous in industrial applications. We also consider extensions of GAE and VGAE models for dynamic graphs. Furthermore, we argue that GAE and VGAE models are often unnecessarily complex, and we propose to simplify them by leveraging linear encoders. Lastly, we introduce Modularity-Aware GAE and VGAE to improve community detection on graphs, while jointly preserving good performances on link prediction. In the last part of this thesis, we evaluate our methods on several graphs extracted from the music streaming service Deezer. We put the emphasis on graph-based music recommendation problems. In particular, we show that our methods can improve the detection of communities of similar musical items to recommend to users, that they can effectively rank similar artists in a cold start setting, and that they permit modeling the music genre perception across cultures.
comment: Ph.D. thesis defended at \'Ecole Polytechnique (IPP) in March 2022. As mentioned in this thesis, several chapters present results also published in scientific articles written with co-authors
♻ ☆ OM4OV: Leveraging Ontology Matching for Ontology Versioning
Due to the dynamic nature of the Semantic Web, version control is necessary to capture time-varying information, particularly for widely used ontologies. Despite the long-standing recognition of ontology versioning (OV) as a crucial component for efficient ontology management, the growing size of ontologies and accumulating errors caused by manual labour overwhelm current OV approaches. In this paper, we propose yet another approach to performing OV using existing ontology matching (OM) techniques and systems. We introduce a unified OM4OV pipeline. From an OM perspective, we reconstruct a new task formulation and measurement for OV tasks. Building upon the prior alignment(s) from OM, we propose a pipeline optimisation method called the cross-reference (CR) mechanism to enhance overall OV performance. We experimentally validate the OM4OV pipeline and the cross-reference mechanism in the OV tested originating from the Ontology Alignment Evaluation Initiative (OAEI) datasets. We also discuss insights into OM used for OV tasks, where some false mappings detected by OV systems are not actually untrue.
comment: 15 pages, 8 figures, 1 table
♻ ☆ Generative Next POI Recommendation with Semantic ID KDD 2025
Point-of-interest (POI) recommendation systems aim to predict the next destinations of user based on their preferences and historical check-ins. Existing generative POI recommendation methods usually employ random numeric IDs for POIs, limiting the ability to model semantic relationships between similar locations. In this paper, we propose Generative Next POI Recommendation with Semantic ID (GNPR-SID), an LLM-based POI recommendation model with a novel semantic POI ID (SID) representation method that enhances the semantic understanding of POI modeling. There are two key components in our GNPR-SID: (1) a Semantic ID Construction module that generates semantically rich POI IDs based on semantic and collaborative features, and (2) a Generative POI Recommendation module that fine-tunes LLMs to predict the next POI using these semantic IDs. By incorporating user interaction patterns and POI semantic features into the semantic ID generation, our method improves the recommendation accuracy and generalization of the model. To construct semantically related SIDs, we propose a POI quantization method based on residual quantized variational autoencoder, which maps POIs into a discrete semantic space. We also propose a diversity loss to ensure that SIDs are uniformly distributed across the semantic space. Extensive experiments on three benchmark datasets demonstrate that GNPR-SID substantially outperforms state-of-the-art methods, achieving up to 16% improvement in recommendation accuracy.
comment: 11 pages, 4 figures, the paper has been accepted by KDD 2025
♻ ☆ MERGE -- A Bimodal Audio-Lyrics Dataset for Static Music Emotion Recognition
The Music Emotion Recognition (MER) field has seen steady developments in recent years, with contributions from feature engineering, machine learning, and deep learning. The landscape has also shifted from audio-centric systems to bimodal ensembles that combine audio and lyrics. However, a lack of public, sizable and quality-controlled bimodal databases has hampered the development and improvement of bimodal audio-lyrics systems. This article proposes three new audio, lyrics, and bimodal MER research datasets, collectively referred to as MERGE, which were created using a semi-automatic approach. To comprehensively assess the proposed datasets and establish a baseline for benchmarking, we conducted several experiments for each modality, using feature engineering, machine learning, and deep learning methodologies. Additionally, we propose and validate fixed train-validation-test splits. The obtained results confirm the viability of the proposed datasets, achieving the best overall result of 81.74\% F1-score for bimodal classification.
comment: 18 pages, 2 figures, 8 tables, submitted to IEEE Transactions on Affective Computing