arXiv Papers

Existing teleoperation systems are often tailored to specific robot hardware and task domains, limiting their scalability and adaptability. We present ModPack, a modular and extensible teleoperation system designed to support diverse robot embodiments and task requirements within a unified framework. At the core of ModPack is a self-contained wearable "backpack" that integrates onboard computation, power, communication, and data storage. Built on top of this shared interface, the system supports plug-and-play capability modules including joint-level teleoperation with haptic feedback, mobile manipulation, and active perception. Experiments across two distinct robot platforms and real-world mobile manipulation tasks demonstrate that ModPack provides a flexible and reusable framework for data collection and policy learning. To support future research, we open-source the complete hardware design and software stack. Project website: https://modpack-robotics.github.io/

Real-world collision avoidance is a core motivation for studying the dynamics and control of high sideslip drifting in vehicles, yet the practical benefit of such maneuvers has so far primarily been tested in scenarios explicitly engineered to require drifting. In this work, we explore the question of if and when drifting may be optimal for safety in real-world winter driving conditions. We present a drift-capable nonlinear model predictive control (MPC) system designed to handle scenarios grounded in crash fatality data and deploy the controller in a high fidelity simulator across road departure and oncoming vehicle collision avoidance scenarios. The controller naturally initiates and sustains drifting maneuvers to stay on the road when hitting a patch of ice on the rear axle and to avoid an oncoming vehicle that has slid into its lane. Comparisons with a benchmark electronic stability control (ESC) system demonstrate how a drift-capable controller can trade off stability for controllability to precisely maneuver through dangerous winter driving scenarios. A Monte Carlo study over random ice patches further shows that the drift-capable controller achieves lower median lane error than ESC across several speeds, while revealing that drifting emerges predominantly at higher speeds.

Many Blind and Low-Vision (BLV) people rely on guide dogs for moment-to-moment navigation, such as staying on path and avoiding obstacles and pedestrians. However, guide dogs are expensive to acquire and maintain (approximately \$50k USD plus ongoing costs), often involve long waiting lists, and have relatively short life expectancies. While robot guide dogs offer a promising alternative, existing approaches exploring this idea suffer from several drawbacks: They often lack the autonomy required for real-world deployment, relying on prior 3D scans of the environment, external computation, or limited awareness of the handler. In this work, we present Milo, the first open-source, low-cost (approximately \$2k USD) robotic guide dog platform capable of fulfilling the basic collaborative navigation role expected of a guide dog. Milo is fully autonomous, requiring no a priori knowledge of the environment, completely self-contained with all computation performed onboard, and suitable for both indoor and outdoor navigation while avoiding obstacles and pedestrians. Our system consists of a modified Unitree Go2 robot (equipped with onboard compute, sensors, and a handle), a perception stack combining voxel mapping with floor, obstacle, and pedestrian detection, and a navigation stack based on an obstacle-avoidance policy trained in a custom bird's-eye-view simulator. We evaluate Milo in real indoor and outdoor obstacle courses and compare it against a costmap-based baseline, demonstrating smoother navigation and fewer handler collisions. To maximize accessibility for BLV users, we release both the robot hardware instructions and the complete software stack as open source.

Haptic feedback provides an implicit channel for communicating safety intentions during human-robot shared control. Existing haptic guidance systems typically employ predefined intervention strategies that cannot accommodate the diverse safety preferences of individual users or application scenarios. To address this limitation, we propose a Learning from Haptics (LfH) framework that learns user-preferred safety interventions from sparse demonstrations, eliminating the need for manual trial-and-error design. Our framework is built on a differentiable Control Barrier Function (CBF)-based optimization layer that automatically adjusts the underlying safety parameters to match the demonstrated haptic responses. Instead of tuning controller parameters directly, users teach the system how they expect it to intervene during teleoperation. The resulting haptic guidance reflects the demonstrated intervention preferences while preserving the intuitive interaction of haptic shared control. Simulation and hardware experiments demonstrate that the proposed framework can learn personalized safety interventions from sparse user input and reduce the mismatch between the generated haptic feedback and the demonstrated preferences.

Despite recent advances in general-purpose robotic manipulation, real-world multi-object clutter remains challenging to handle for today's prevalent approaches. The problem scales in complexity due to more objects and collisions, more unpredictable contact physics, distractors, and task ambiguity. Bridging this gap to real-world deployment requires effective scene abstractions; yet today, producing such abstractions requires extensive task-specific manual engineering, which does not scale. These abstractions are costly to generate and difficult to adjust or fine-tune. We instead propose a plug-and-play fix to automatically generate scene-specific, task-specific, adaptively updating abstractions on top of existing planning and control stacks. LLM-guided Environment Simplification (LENS) produces a de-cluttered abstracted scene representation by merging (e.g., stacked objects) or pruning (e.g., distant objects) scene entities in a closed loop in response to task progress. These dynamic, task-relevant abstractions are versatile and easy to use. In our experiments, we show that LENS improves classical planning, model-based control, and a vision-language-action model, across a diverse set of highly cluttered manipulation scenes. Project website: https://lens-2026.github.io/.

This work presents a decentralized, spoofing-aware trajectory planning framework for small unmanned aerial systems operating under Remote Identification (RID) location spoofing attacks. Existing planners typically assume RID broadcasts are trustworthy, which can increase the risk of loss of separation and mid-air collisions when spoofing occurs. In contrast, the proposed approach explicitly treats RID information as unverified and incorporates physical-layer observations to assess broadcast credibility. Received signal-strength measurements from neighboring aircraft are used to detect spoofing and probabilistically localize a spoofing agent. The resulting uncertainty is converted into a risk-bounded unsafe region using a chance-constrained formulation and integrated into a per-agent Markov decision process-based planner. This enables real-time, decentralized collision avoidance while preserving mission objectives and scalability. Simulation results in a multi-aircraft package delivery scenario demonstrate reduced near mid-air collision events compared to planners that assume truthful RID data, while maintaining computational efficiency suitable for real-time execution.

Robots deployed in delivery, campus, and emergency-response settings often need to navigate from buildings to streets within a single continuous episode. Existing benchmarks usually evaluate indoor and outdoor navigation separately, and many abstract away robot execution, leaving exit finding, boundary traversal, adaptation, and kinodynamic failures underexplored. We introduce NavVerse, a physics-enabled benchmark for indoor-to-outdoor embodied navigation. NavVerse contains 100 indoor scenes, 50 urban outdoor scenes, and 50 indoor-to-outdoor scenes, and 10,000 episodes spanning Object Navigation, Vision-and-Language Navigation, and Place Navigation tasks, where agents search for semantic points of interest such as restaurants or banks. Agents are evaluated through executable robot interfaces using task-success, path-efficiency, and safety metrics. Zero-shot experiments with RL, VLA, and modular baselines show that current agents remain far from solving cross-context navigation: end-to-end VLAs obtain the highest zero-shot success, while the modular method provides the strongest safety profile. PlaceNav further reveals a clear drop from outdoor to indoor-to-outdoor scenes, indicating that adaptation remains major bottleneck.

Fully actuated unmanned aerial vehicles (UAVs) are usually certified through rank conditions on a control-allocation matrix or through free-flight tracking performance. For aerial physical interaction, this certification may be incomplete. During sustained contact, part of the available wrench is consumed by the interaction task, and only the residual wrench remains available for stabilization, disturbance rejection, and maneuvering. This paper introduces a control-theoretic framework for \emph{contact-persistent full actuation}. A rigid-body model on $\R^{3}\times\SO\left(3\right)$ is combined with a morphology-dependent wrench map that captures fixed-tilt, variable-tilt, coaxial, and overactuated multirotor architectures. We define feasible wrench sets under actuator limits, residual wrench sets under task loading, and residual authority margins that strengthen the usual rank-based notion of full actuation. The main result shows that contact-persistent full actuation is equivalent to interiority of the task wrench in the constrained feasible wrench polytope, and that the residual authority radius is exactly the distance to the polytope boundary. We further introduce a signed residual-margin certificate for infeasible and boundary cases, a slack-maximizing allocation certificate, and a robust implementability condition that can be used as a margin-aware safety filter. Numerical evaluation on an abstract tilted hexarotor shows that full row rank alone does not imply feasible contact operation. Intermediate tilt angles preserve residual authority during pushing, whereas small or excessive tilts fail because of lateral-force deficiency or hover-margin loss.

Limbless robots offer exceptional mobility in confined and cluttered environments due to their slender bodies and their ability to exploit body-terrain interactions. Recent designs incorporating compliance demonstrate robust locomotion without complex sensing or control; however, these systems typically rely on fixed body configurations, with each morphology specialized for a single locomotion mode or environment. This raises a key challenge: how can a single limbless robot achieve versatile locomotion while preserving the robustness of compliance-mediated locomotion? To address this challenge, we present a cable-driven limbless robot that reconfigures body morphology and compliance to enable diverse locomotion modes. Distributed cable actuation generates traveling body waves, while programmable passive compliance enables robust contact-rich locomotion without terrain knowledge or high-bandwidth feedback. Rolling joints reorient bending planes along the body, enabling rapid reconfiguration and smooth transitions between locomotion styles, and incorporate geared locking to maintain configuration without continuous power. By combining programmable bending compliance and morphology control, the platform achieves lateral undulation, sidewinding, rolling, and twisting within a single system. Experiments demonstrate reliable gait generation, traversal in obstacle-rich environments, and transitions between modes, establishing a versatile limbless platform for navigating complex environments with applications in search and rescue, environmental monitoring, and inspection.

Robust embodied robots should be able to recover from failures and retry tasks in order to operate reliably in unstructured and noisy real-world environments. Achieving this capability requires training policies on data that captures recovery behaviors. However, collecting such data through robot teleoperation is difficult to scale, as it is time-consuming to induce diverse failure states, perform corrective actions, and reset the environment. This challenge is further exacerbated by the high diversity of failure modes, which demands substantially more recovery data than success demonstrations. In this work, we show that egocentric human data capturing failure recovery processes provides a scalable alternative. By efficiently arranging task-level failure configurations and recording short recovery segments, human operators can generate more than 10x as much valid recovery data per hour compared to robot teleoperation under our protocol. To address the embodiment gap between human and robot, we propose EgoRecovery, a co-training framework for learning recovery behavior, where human recovery demonstrations are aligned to a compact corrective-intent space shared with robot data, which captures the timing and magnitude of correction. Only a small number of robot recovery demonstrations are required to connect this intent to executable robot actions. At deployment, a learned recovery gate predicts when correction is needed from robot observations and activates the corrective intent only in recovery states. Experiments on real-world recovery tasks show that EgoRecovery improves success from failure starts over robot-only recovery, direct co-training with human recovery data, and direct intent-transfer baselines.

Vehicle-to-everything-aided autonomous driving (V2X-AD) significantly enhances driving performance through information sharing. However, existing collaborative perception methods only optimize module-level perception capabilities and fail to effectively serve the ultimate planning and control tasks. We propose an end-to-end collaborative driving system that directly optimizes planning task performance. The system employs MotionNetwork to fuse historical temporal information, utilizes attention mechanisms to efficiently compress spatial features into compact tokens, and adaptively fuses multi-agent features through an autoregressive decoder. Additionally, we introduce Mixture-of-Experts (MoE) architecture to enhance the model's representation capacity for heterogeneous features. Experiments demonstrate that our method achieves a driving score of 79.72, surpassing the state-of-the-art CoDriving baseline (77.15) by 3.33% in closed-loop evaluation while maintaining communication efficiency.

This paper presents a vision-informed grasp force prediction framework for robotic handling of date fruits. Addressing the dual challenge of high detachment forces and low bruise thresholds, we first conduct mechanical characterization on date samples to define a safe grasping envelope and quantify the relationship between fruit geometry and bioyield stress. In this work, we develop a Vision-to-Force (V2F) pipeline that combines computer vision-based segmentation, active-contour refinement, and geometric feature extraction with a physics-informed residual neural network that augments a Hertz contact equation. The resulting model maps non-contact visual descriptors and cultivar metadata to predict a safe grasp force with mean validation performance of $R^2 \approx 0.7$ across unseen cultivar groups, which is a good result given the inherent mechanical variability of biological tissue. Experimental validation using a gripper and load cell indicates that the predicted forces enable stable manipulation of different types of date fruits, with residual deformations below 1 mm and no observable damage. These results show that pre-emptive, vision-driven force estimation% can replace slow and potentially damaging tactile exploration , enabling safer robotic handling of fragile fruits.

Ensuring safety in Physical AI systems operating in real-world environments is a critical challenge, particularly in hospital wards where vulnerable patients, clinical staff, medical devices, and assistive robots coexist. In this paper, we reinterpret Clinical Pathways as explicit runtime safety specifications for embodied medical AI. We propose a conceptual robotic architecture that integrates wearable sensors, smart medical devices, and assistive robotic components into a unified framework for real-time safety monitoring. At its core, a Runtime Safety Monitor (RSM) evaluates multimodal physiological and system-level signals against clinically defined constraints derived from the prescribed care process. Rather than relying solely on statistical anomaly detection, the proposed approach combines temporal prediction, uncertainty-aware reasoning, and constraint-based verification to identify safety violations. The RSM targets three classes of events: physiological deviations from prescribed care, hardware and communication failures, and potential data tampering or misuse. This work contributes to Safe Physical AI by operationalizing domain-specific clinical knowledge as enforceable safety constraints, bridging learning-based perception and runtime safety monitoring to assist nursing staff in real-world hospital wards.

Vision-language models have shown strong potential for social robot navigation by leveraging rich semantic understanding of complex environments and human behaviors. However, large scale VLMs are difficult to deploy on resource-constrained robotic platforms, while lightweight VLMs often lack sufficient social reasoning capability. To address this problem, we propose SOPD-SocialNav, a selective on-policy distillation (SOPD) method that transfers social navigation knowledge from a large teacher VLM to a lightweight student VLM. SOPD introduces an entropy-based token selection mechanism that uses teacher uncertainty to identify socially informative decision tokens, while suppressing gradients from low-entropy tokens corresponding to trivial navigation states. A temperature-controlled Jensen-Shannon divergence objective is then used to align the student and teacher distributions on the selected tokens. Experiments on the SNEI and MUSON benchmarks demonstrate that SOPD consistently outperforms supervised fine-tuning, off-policy distillation, and standard on-policy distillation baselines in action prediction, perception consistency, and reasoning consistency. Real-world deployment on a Scout Mini robot further shows that the distilled model can generate more socially appropriate navigation behaviors in conversational and queuing scenarios. These results suggest that SOPD is an effective strategy for building lightweight yet socially aware VLM-based navigation systems.

Evaluating the physical consistency of embodied world models(EWMs) is a critical open challenge. While closed-loop evaluation via simulator rollouts offers a more faithful assessment of physical plausibility than open-loop alternatives, existing frameworks almost exclusively rely on Inverse Dynamics Models(IDMs) for action extraction. Due to the intricate mapping from 2D pixel space to 3D kinematic space, the learned IDMs can be brittle to data outside their training distribution, resulting in unreliable action extraction from the generated videos with novel objects and scenarios. This creates an unavoidable attribution ambiguity between world model inaccuracies and extractor errors. To reduce this ambiguity, we present KineBench, an IDM-free closed-loop benchmark for EWMs, built upon an explicit kinematic grounding pipeline. Given a generated video, KineBench employs cascaded visual foundation models to directly extract 6D end-effector poses from individual frames, which are then executed in a physics simulator for closed-loop validation. Beyond execution-based task success, KineBench incorporates two classical 3D kinematic metrics--Spectral Arc Length (SPARC) and the Maruyama Manipulability Index--to characterize trajectory smoothness and kinematic feasibility from a robot-centric perspective. Built on 20 diverse manipulation tasks in ManiSkill3, KineBench evaluates EWMs across four progressive suites: basic execution, task transfer, visual out-of-distribution generalization, and complexity-conditioned scaling. Evaluation across frontier models reveals task-complexity-bounded nonlinear scaling in embodied video generation, providing empirical guidance for future data-scaling strategies.

Cross-embodiment navigation is a key challenge in embodied intelligence. Due to differences in embodiment, the same visual observation may imply different actions for different agents, making prediction ambiguous when relying solely on vision. Existing studies mainly rely on reinforcement learning, which requires large-scale interaction and careful reward design, making it difficult to support scalable pretraining and real-world adaptation. In contrast, imitation-learning-based approaches remain limited. To address these challenges, we propose an imitation-learning-based embodiment-aware navigation framework with a modular multi-stage design. In pretraining, we construct a cross-embodiment navigation dataset from Internet videos and introduce embodiment geometry as conditional tokens to reduce action ambiguity under the same observation. In fine-tuning, we design a multimodal information injection mechanism based on a decoupled architecture. Specifically, we design a trajectory augmentation strategy to generate high-risk samples, which are used to train spatial perception and risk-aware correction separately, thereby explicitly incorporating embodiment geometry for safe navigation. Experimental results show that the proposed method effectively improves navigation performance across different embodiment settings, demonstrating the effectiveness of incorporating embodiment geometry into embodied navigation.

Humanoid general motion tracking requires policies that can follow diverse whole-body references while maintaining balance. Building such policies involves many practical design choices, and their individual effects are often hard to assess. We address this issue with an empirical study of common modeling and training factors used in recent humanoid motion-imitation pipelines. To make the study controlled and reproducible, we developed YAHMP, an open-source modular framework for training, evaluating, and deploying whole-body motion tracking policies on the Unitree G1. Within YAHMP, we define a nominal configuration and compare variants that differ in motion-command representation, observation history, action representation, actuation profile, hand-force randomization during training, and training approach. We evaluate the resulting policies on a test set of retargeted human motions and compare the nominal policy with TWIST2 as an external baseline trained on the same motion set. The results distinguish choices with clear tracking effects from choices that mainly change actuation effort, training complexity, or physical interaction capability. Finally, we deploy YAHMP policies zero-shot on the real Unitree G1, demonstrating diverse whole-body motion tracking, balance under external perturbations, and forceful interaction.

We propose Diffusion ReRoll, a diffusion-based framework for robotic sequential prediction that enables revisable denoising over horizons. Existing diffusion-based sequence predictors typically perform a single monotonic denoising process. In contrast, Diffusion ReRoll selectively re-noises regions that have become locally stable while the remaining regions continue denoising, so the re-noised regions can be refined again using context from the rest of the horizon. This structured re-noising enables iterative cross-horizon revision, allowing earlier and later segments to revise one another, while maintaining local consistency. We evaluate Diffusion ReRoll against full-sequence diffusion and causal denoising based on Diffusion Forcing across long-horizon planning, policy learning, and unified video-action modeling. On OGBench PointMaze and AntMaze, Diffusion ReRoll achieves relative gains in average success rate of 21% over Diffusion Forcing in matched guidance-based planning and 23% over Diffuser in matched goal-inpainting. In diffusion-policy-style action prediction, Diffusion ReRoll improves average success by 56.5% relative to Diffusion Policy across different prediction horizons and history lengths on the LIBERO-10 multi-task benchmark. In unified video-action prediction, Diffusion ReRoll improves policy and inverse dynamics performance, especially under out-of-distribution evaluation, and achieves the best action-video consistency. These results support structured re-noising as an effective mechanism for revisable robotic sequence generation.

Accurate motion prediction of surrounding agents and safe motion planning are two closely coupled key tasks for social robot navigation in crowded environments. Deploying these systems on resource-constrained edge devices necessitates compact, unified models that can perform both tasks simultaneously. However, within these compact shared encoders, recent unified models often overlook severe representational conflicts that arise from the distinct objectives of predicting neighbor behaviors versus ego-centric safety planning. To address this issue, we first identify the Skill Conflict$\unicode{x2014}$a phenomenon where overlapping parameter assignments cause distinct tasks to compete for the same weights, preventing the model from fully specializing in individual skills. To resolve this, we propose a novel model-merging-based framework, Disjoint Parameter Training (DPT). DPT mitigates performance degradation caused by Skill Conflict through distributed parameter learning, which separates the key parameter regions of each task while preserving their core capabilities prior to merging. In addition, we observe that sparse merging, which selectively integrates only the most influential parameters for each task rather than combining all task-specific parameters, yields optimal performance by preventing interference among adjacent features and concentrating representational capacity. DPT can be applied in parallel with a variety of merging methods. Evaluated on standard crowd navigation benchmarks (JRDB and JTA), our framework demonstrates superior performance, validating its versatility and effectiveness for safe, resource-efficient robot navigation.

The ability to acquire skills rapidly and effortlessly while retaining those already mastered is essential for robots. However, current methods still rely on a cumbersome training-time loop that is costly and slow, while eroding skills already mastered. In this paper, we introduce HOST (Human-to-robot One-Shot Skill AcquisiTion), a framework that enables a robot to acquire skills in seconds from a single human video while retaining previously mastered skills. HOST resolves skill acquisition through a cascade of self-grounded prediction. It first estimates the robot's progress within the demonstrated task, then translates the upcoming progression into the robot's own future observations, and finally derives actions from these predicted observations. This cascade is trained on targets coupled to the video demonstration, obtained by mapping the robot trajectory and the video demonstration onto a shared task progress manifold, then redefining each target to align with the future progression of the video. HOST thereby enables the robot to actively follow the demonstrated procedure and adapt it to the robot's embodiment. HOST acquires novel skills at inference time from a single human video in an average of 29 seconds and achieves a 62% average success rate. It exceeds the zero-shot baseline by 45% while retaining previously mastered skills. HOST even exceeds the baseline fine-tuned on 50 robot demonstrations per task while requiring 50 times fewer demonstrations and acquiring each skill 507 times faster. Additional information about HOST is available on the project website.

Embodied visual tracking (EVT) requires a mobile agent to continuously follow a specific target described in natural language using only onboard vision. While recent vision-language-action (VLA) policies unify target identification and trajectory planning, their chain-of-thought (CoT) reasoning often operates in abstract spatial latents that are difficult to supervise and weakly aligned with explicit image-space detections. To address this, we introduce ReferTrack, a referring-then-tracking paradigm that grounds EVT using a single forward-facing camera. Our model first selects the target from an indexed set of bounding boxes, then decodes tracking waypoints conditioned on this image-grounded decision. To preserve target motion cues over time, ReferTrack maintains a sliding-window queue of previously selected bounding boxes, injecting their geometric features into the visual history via temporal-viewpoint-bbox indicator (TVBI) tokens. We further enhance target identification by co-training on a custom Refer-QA dataset. On EVT-Bench, ReferTrack achieves state-of-the-art single-view performance with success rates of 89.4%, 73.3%, and 74.1% on the single-target, distracted, and ambiguity tracking splits, respectively -- matching or even surpassing several multi-camera baselines on identification-heavy tasks. Finally, real-world deployments on legged and humanoid robots validate its robust sim-to-real transfer capabilities. Code is available at https://github.com/MedlarTea/referTrack.

Humans can progressively acquire highly dynamic motor skills while preserving reliable everyday motor abilities. In contrast, existing humanoid controllers face a trade-off between generalist and specialist capabilities: generalist motion tracking policies struggle to reliably execute rare highly dynamic motions, whereas specialist training can degrade previously acquired behaviors. We introduce Extreme-RGMT, a two-stage continual learning framework for robust generalist humanoid control. The method first learns a generalist motion-tracking base policy from diverse multi-source motion data, then employs an asymmetric skill acquisition and capability consolidation mechanism to constrain policy drift on mastered motions while emphasizing difficult dynamic segments. To address the scarcity of highly dynamic motions, their high failure rates, and the resulting shortage of informative samples, Extreme-RGMT combines difficulty-aware sampling with advantage-prioritized trajectory resampling to emphasize critical segments. Experiments show that Extreme-RGMT achieves state-of-the-art generalist whole-body motion-tracking performance, including substantially improved completion of challenging highly dynamic motions. The resulting controller directly executes diverse unseen highly dynamic motions under fixed references and online inertial motion-capture inputs, advancing generalist whole-body motion-tracking controllers toward highly dynamic motor capabilities at the human-expert level.

Practical robotic grasping in complex scenes requires both 3D spatial reasoning and alignment with task-specific requirements. Vision-language models (VLMs) offer a natural way to specify these requirements using language, but existing approaches either use a VLM to predict the grasp directly with limited spatial awareness, or train the VLM together with the grasping model, which requires significantly more data and compute. These limitations impede performance and have prevented scaling to multiple embodiments in complex scenes. We address this by proposing SeededGrasp, a novel data-efficient framework that enables a VLM to predict a seed point to be used as conditioning for a subsequent lightweight grasp-generation model. Our architecture decouples high-level semantic reasoning from low-level geometric execution, enabling multi-embodiment support while bypassing the need for expensive end-to-end training. To enable training such models, we release the first multi-embodiment tabletop grasping dataset comprising over 2.5M grasps in cluttered scenes. Experimental results demonstrate that our approach outperforms existing baselines, achieving 72% success in simulation and 78% in real-world grasping experiments. See our project site for data and code: https://uoft-isl.github.io/seeded-grasp/

The training of learned inertial odometry depends on dense, high-precision position ground truth from motion capture, visual-inertial odometry or SLAM, which is costly and hard to acquire at scale. We propose DINS-IO, which learns inertial odometry directly from raw IMU streams without position labels. Our key insight is that the strapdown INS velocity recursion is a strong, fully differentiable consistency prior: the predicted velocity, rotated into the navigation frame, must agree with the integrated specific force up to an unknown initial velocity and a constant accelerometer bias. We cast this constraint as a sliding-window least-squares problem with a globally shared bias, solve it in closed form, and use the solver residual as a self-supervised loss whose gradient flows back to the network through the analytic solution. To supply this per-sample constraint, we design a high-frequency network that emits dense body-frame velocity at the IMU rate. Since the self-supervised network learns consistent motion but its velocity is not yet metrically calibrated, we calibrate it to true metric velocity from a few labeled trajectories by directly supervising the predicted body-frame velocity and adapting only low-rank (LoRA) patches. On standard benchmarks, DINS-IO pretrained self-supervised and fine-tuned with a small fraction of labels matches or surpasses fully supervised baselines.

We consider a task planning scenario in which robots sharing a persistent environment are assigned tasks one at a time from a held-out sequence. Standard task planners, lacking foresight of future tasks and inconsiderate of others' constraints, solve each task in isolation, leaving terminal states that increase future cost for all, side effects that compound over lengthy task sequences. To reduce cost over the sequence, a robot must anticipate how its actions now may impact performance on future tasks for all robots sharing the environment. Therefore, we present courteous anticipatory planning, wherein a model-based planner proposes candidate plans and selects the one that jointly minimizes immediate cost and aggregated expected future cost across all robots, estimated via independent per-robot learned estimators. This factored formulation avoids combinatorial joint rollouts and supports modular deployment: adding a robot requires only training its own estimator. We evaluate in two persistent PDDL domains, a home environment with robots that have similar capabilities but different responsibilities, and a restaurant environment where robots' distinct capabilities create states that other robots lack the capability to resolve. During lengthy task sequences, our planner reduces total cost by 10.43% versus myopic and 4.03% versus selfish anticipatory planning in a two-robot home environment and by 17.41% and 13.24%, respectively, in a three-robot restaurant.

Closing the gap between benchmark performance and reliable real-world operation remains a central challenge for Vision-Language-Action (VLA) humanoid robots, which must handle execution errors, distribution shifts, and environmental variability. This paper presents DEED (Data-Efficient Post-Training and Experience-Driven Learning), a systems-level approach evaluated on a supermarket chip-restocking task using a Unitree G1-Edu humanoid robot and the GR00T N1.6 foundation model. DEED comprises three key components: (1) a data-efficient post-training pipeline with control-frequency alignment, data curation, task-relevant visual highlighting, and reduced VLA dependence; (2) a real-world study of experience-driven refinement, adapted from RECAP via a text-based advantage prefix and a vision-language value function; and (3) a latent-space analysis tool for studying in- and out-of-distribution behavior. Our results suggest that bridging the lab-to-store gap is primarily a systems integration challenge rather than an architectural one: careful data design and targeted post-training can transform a policy that fails under naive fine-tuning into a competent real-world system using only a single GPU.

Aquatic self-reconfigurable robots must assemble into desired shapes while ensuring safe interactions among multiple agents. This paper proposes a hybrid framework that combines distributed Model Predictive Control (MPC) with Control Barrier Functions (CBFs) for multi-agent shape formation and reconfiguration. Given a desired shape and target assignment, a distributed MPC scheme, solved via the Alternating Direction Method of Multipliers (ADMM), computes coordinated trajectories through local optimization and information exchange. To ensure safety in real time, distributed CBF-based filters are applied to enforce inter-agent collision avoidance. The proposed approach leverages the predictive capabilities of MPC to mitigate local minima, while CBFs provide formal safety guarantees despite the nonconvexity of the underlying optimization problem. Simulation results with up to 25 agents and experimental validation with four physical robots demonstrate the effectiveness and scalability of the framework.

Soft robot exteroception is increasingly being explored for a variety of field applications. In this work, we present a sound-based system for localizing disaster victims in confined and unstructured environments, based on a distributed acoustic sensing architecture embedded along the body of a soft everting vine robot. We propose a dynamic Steered Response Power with Phase Transform framework that supports both far-field direction-of-arrival estimation and near-field three-dimensional source localization as the robot approaches the sound source. To better understand the design and control space related to localizing sound using a soft, shape-morphing robot body, we conduct experiments measuring the accuracy of these methods for a five-microphone array attached to the robot body using three placements relative to the outer membrane of the robot (inside the pressurized body, inside the inner tail, and outside the outer wall) and in four robot configurations (linear, double linear, circular, and sinusoidal). We measure the change in accuracy as the signal-to-noise ratio, the direction of approach, and the distance of the sound source from the center of the array change. Finally, we demonstrate a vine robot growing into an arbitrary shape while carrying microphones along its outer wall, and show that a sound source located with the array's near field can be localized with high accuracy after only three microphones have everted from the robot body. These results highlight the potential of distributed acoustic sensing for reliable victim localization using soft growing robots.

Full-sized humanoid robot capabilities have grown exponentially in recent years, aiming towards general-purpose deployment in human environments. A popular control method used by manufacturers utilizes Virtual Reality for upper-body teleoperation and Reinforcement Learning for lower-body balance and locomotion control. As a result, a single remote operator can see, manipulate, and navigate about a real, distant physical environment. This powerful control stack is often relegated to expensive full-sized robots, many of which are inaccessible to the research community. Miniature humanoids are more prevalent, but employ less biomimicry in their design (e.g. fewer sensors, Degrees of Freedom, etc) and lack similar developments. This paper describes a compliant full-body telepresence control stack developed from the ground up for miniature humanoids. Framework experimentation on ROBOTIS OP3 hardware showcases walking at speeds up to 0.45 m/s independent of arm motions. Tele-loco-manipulation is demonstrated via a cube relocation experiment with an expert human operator. On average, the teleoperated system moved 2 different 40 g cubes within 10 mins, walking a total distance of 5 m. Overall, the developed system shows potential for miniature humanoid tele-loco-manipulation.

Detectors for AI-generated video are evaluated offline. A clip is decoded to pixels and scored once, increasingly by a large vision-language model. Detection, however, is deployed online. We recast the task as streaming perception and score the motion field the codec already wrote into the bitstream. Reading that field is a parse, not a pixel-domain forward pass. Because the running aggregate is monotone, one end-calibrated threshold is anytime-valid at the data-dependent decision time. Recalibrating at each prefix is not. Escalation is priced in closed form. A compute budget maps to a deferral window, on a frontier monotone exactly where the deferral condition holds. On matched GenVidBench the codec stage reaches full-length AUC 0.64 at five orders of magnitude less compute than a pixel CNN, on CPU. Its gate holds the stopping-time false-positive rate at target while the real data match its calibration, and drifts above it under distribution shift. Deferring 15% of clips lifts accuracy from 0.75 to 0.78 at $7\times$ less compute (paired: McNemar $p<10^{-6}$). The stage-1 ordering replicates on AIGVDBench. We introduce no new detector. The contribution is the reframing, two guarantees, and the measured frontiers. Code, configurations, and evaluation splits: https://github.com/KurbanIntelligenceLab/streamdet.

Recovering scene-consistent 4D crowd motion from monocular video in large-scale scenes remains challenging due to severe depth ambiguity and complex scene geometry. Existing monocular crowd reconstruction methods typically rely on single-plane assumptions, leading to unreliable metric scale and spatial drift under complex terrain. We propose Crowd4D, the first scene-aware 4D crowd reconstruction framework that jointly optimizes the crowd and scene from a monocular RGB video in large-scale scenes. Crowd4D explicitly incorporates scene geometry and ensures consistency across image and scene spaces via a multi-stage optimization strategy. A key bottleneck of this task lies in accurate human-scene alignment, particularly in scale and position. However, human and scene reconstructions are typically decoupled. To address this, we introduce the Human-Scene Interaction Proxy, abbreviated as HSIP, as an intermediate representation derived from Scene Interaction Point Clouds and a Scene Interaction Surface, abbreviated as SIPC and SIS. These representations encode explicit scene-aware geometric priors and redefine the optimization space for large-scale monocular 4D crowd reconstruction. To further improve temporal stability under occlusions, we introduce Crowd Structural Coherence Regularization, abbreviated as CSCR, which leverages HSIP-based spatial priors to impose soft temporal consistency on pairwise relative displacements and directions within local crowd neighborhoods. Extensive experiments demonstrate that Crowd4D consistently outperforms existing state-of-the-art methods and enables robust monocular 4D crowd reconstruction in complex, large-scale real-world scenes.

Recent advances in Multimodal Large Language Models (MLLMs) have triggered the development of end-to-end MLLMs for autonomous driving. However, the main emphasis to date has been for MLLMs using 2D images and videos. In contrast, this paper considers MLLM effectiveness using 3D sensors, particularly LiDAR and stereo cameras. LiDAR presents unique challenges to integration within an MLLM, largely because of data sparsity and lack of a grid structure for the data. For similar reasons, fusion of camera and LiDAR data within an MLLM pipeline is also uncommon. However, most autonomous systems rely on LiDAR-based sensing, and incorporating 3D data has been proven to improve performance in traditional 3D scene perception tasks. This paper presents D3VL, a novel MLLM framework that integrates 2D and 3D time-series data in a single but simple architecture. The model aims to answer questions involving traffic scene understanding and safety. D3VL shows an 11% improvement in the KITTI Question-Answering (QA) dataset compared to baseline methods in processing 2D and 3D time-series data. This paper further introduces the Waymo QA dataset extension, which assesses models' capabilities in processing 3D and time-series data under diverse driving conditions. D3VL implementation code and WaymoQA extension can be found on our supplemental website: https://automotivesafety-lvlm.github.io

Incorrect disposal can contaminate campus recycling streams, and a bin-mounted camera could provide feedback as an item is discarded. We evaluated whether synthetic and derived images improve a YOLOv8n detector for this view. The real dataset contained 148 campus photographs: 86 for training, 31 for validation, and 31 for testing. Twelve joint-training configurations varied the amount and source of added images. We repeated seven principal settings with four matched seeds and computed bootstrap percentile intervals over those seeds. The real-only model reached a mean [email protected] of 0.691 [0.665, 0.722]. Background replacement reduced the mean to 0.560 [0.499, 0.619], isolated-object images gave 0.680 [0.644, 0.724], and the full augmentation pool gave 0.487 [0.438, 0.537]. We also tested hand-and-forearm composites because every real photo showed a held object. Two cutouts in the initial composite set came from test photographs, so we discarded that experiment, rebuilt the set with training-split cutouts, and reran all four seeds. The corrected paired difference was +0.034 [-0.063, 0.199], which does not support a reliable hand-composite effect. Single-seed transfer experiments produced source-dependent rankings between joint mixing and sequential pretraining. None of the evaluated configurations exceeded the real-only baseline. The reported intervals quantify seed variation; the 31-photo test set remains too small for strong class-specific conclusions.

Long-video question answering requires a model to preserve visual evidence over time without repeatedly reprocessing the same video. A practical approach is to store the vision-language model's internal key-value (KV) cache for each video chunk and retrieve that state at query time. However, independently cached video chunks do not compose correctly: every chunk is prefilled from local rotary position zero, so naive concatenation collides temporal phases and removes the global order required for questions about what happened first, how often events occurred, or what changed across the video. This paper presents ChronoStitch, a training-free method for composing independently stored visual KV memories. The method first re-bases stored post-rotary keys onto a global three-axis multimodal RoPE coordinate system that preserves time, height, and width structure. We show why a one-dimensional scalar re-indexing is geometrically inconsistent for visual tokens because it turns spatial order within a frame into false temporal displacement. We then address the residual content gap left by positional repair: later chunks were originally encoded without attending to earlier chunks. ChronoStitch therefore selectively recomputes a small fraction of high-deviation later-chunk visual tokens while allowing them to attend over the composed cache. On Qwen2.5-VL-3B and the temporal split of TempCompass, ChronoStitch outperforms naive composition and position-only variants, improving event-ordering accuracy while running 3.3x faster than full joint re-prefilling.

Vector Quantization (VQ) underpins modern discrete visual tokenization. However, training quantization modules for state-of-the-art VQ-based models requires significant computational resources which, in practice, all but prevents the development of novel, cutting-edge VQ techniques under resource constraints. To address this limitation, we propose {\bf VQ-Transplant}, a simple framework that enables plug-and-play integration of new VQ modules into frozen, pre-trained tokenizers by replacing their native VQ modules. Crucially, the proposed transplantation process preserves all encoder-decoder parameters, obviating the need for costly end-to-end retraining when modifying the quantization method. To mitigate decoder-quantization mismatch, we introduce a lightweight decoder adaptation strategy (trained for only 5 epochs on ImageNet-1k) to align feature priors with the new quantization space. In our empirical evaluation, we find that VQ-Transplant allows obtaining near state-of-the-art reconstruction fidelity for industry-level models like VAR while reducing the training cost by 95\%. VQ-Transplant democratizes quantization research by enabling resource-efficient integration of novel VQ techniques while matching industry-level reconstruction performance.

As autonomous systems and smart cities continue to evolve, the demand for efficient and robust scene understanding becomes increasingly critical. Semantic segmentation plays a key role in enabling autonomous vehicles to comprehend complex urban environments. However, achieving high accuracy with minimal computational cost remains a significant challenge. In this paper, we present Edge-Gated Refinement Network (EGRNet), a lightweight and efficient deep learning model designed for real-time semantic segmentation in urban scenarios. The model incorporates depthwise separable convolutions to reduce computational complexity and dilated residual blocks for capturing rich multi-scale contextual information. Additionally, we introduce a novel Edge-Gated Refinement (EGR) module, which adaptively fuses original and refined features through a learnable gating mechanism, enhancing boundary preservation and edge-sensitive regions. To further improve feature representation, Squeeze-and-Excitation (SE) attention is applied across the network. With only 0.46M parameters, EGRNet achieves state-of-the-art performance while maintaining low computational overhead. When evaluated on the Cityscapes dataset, the model attains a mean Intersection over Union (mIoU) of 65.28%, demonstrating strong accuracy with minimal resource consumption. Moreover, we introduce a lightweight adversarial attack detection strategy, ensuring robustness against adversarial inputs without compromising real-time performance. By combining efficiency, accuracy, and resilience, EGRNet is well-suited for deployment on edge devices in safety-critical real-time applications.

The allocation of visual attention by pathologists during cancer diagnosis is a highly selective process that critically shapes the information extracted from whole-slide images (WSIs). Human attention helps medical imaging tasks such as classification and segmentation, and becomes a strong semantic cue for identifying diagnostically informative regions for report generation. In this paper, we introduce human attention into the training of pathologist report generation models. To this end, we collected a multimodal human-attention dataset of 121 prostate WSIs annotated with pathologists' multi-scale viewport trajectories synchronized with the pathologists' verbal descriptions and cursor movements for five clinically relevant components (e.g., Gleason patterns). Using this dataset, we finetune two report generation models with an attention-alignment loss that regularizes the model attention over image patches to match the distribution of pathologist attention. We evaluate our approach on prostate cancer report generation and visual question answering using two models with different internal attention mechanisms (i.e., how image tokens are integrated into the language decoder). Experiments show average gains of 10.9% on NLP-based metrics and 19.3% in accuracy across five clinically relevant report components. Further, model attention maps extracted at inference time, with minimal computational overhead, align more closely with pathologist attention, providing stronger visual support for the generated reports by highlighting the regions that most influence the output.

Cross-dataset generalisation remains a major barrier to clinical deployment of echocardiographic left ventricular segmentation, yet the sources of this shift are rarely disentangled. We examined whether transfer degradation could be estimated before deployment using handcrafted ultrasound descriptors, VAE latent features, and segmentation-derived latent features across six echocardiographic datasets. Geometry-aware preprocessing substantially improved several poor transfer cases, suggesting that much of the apparent domain shift reflects field-of-view and framing inconsistencies rather than intrinsic acoustic differences alone. Intensity z-normalisation changed dataset separability by less than 0.005, indicating that brightness and contrast are not the dominant shift axis. Absolute Dice drop on held-out source-target pairs was predicted with an R-squared value of 0.612, an MAE of 0.082, and a Spearman rho of 0.681. The variant without LV and fan-shaped features retained approximately 70% of this explanatory power, supporting mask-free transfer-risk monitoring. The most informative discrepancy measure depended on the representation, with CMD strongest in z-normalised handcrafted features, with an absolute r of approximately 0.86 and an R-squared value of approximately 0.70; log-Wasserstein strongest in VAE space, with an r of approximately -0.90 and an R-squared value of approximately 0.81; and log-MMD strongest in LV-segmentation latent features, with an r of approximately -0.92 and an R-squared value of approximately 0.84. Apparent vendor effects were largely dataset-confounded. Echocardiographic domain shift is therefore structured and measurable, and its impact on segmentation can be partly reduced through geometry-aware preprocessing and anticipated using representation-specific transfer-risk estimation.

We propose a unified variational framework for image segmentation under sparse pixel-level supervision. Our method is based on a simplex-constrained Potts model with a smooth perimeter regularizer, yielding a convex, smooth energy functional that can be used as a training loss in weakly supervised deep learning paradigms or optimized efficiently using iterative methods. Sparse labels are incorporated into the data fidelity term by constructing a fuzzy membership function via a function extension problem in a Reproducing Kernel Hilbert Space (RKHS), which can effectively capture inhomogeneous intensity statistics. The derived discrete loss for training standard networks demonstrates robustness and consistent improvements over non-training and partial cross-entropy (PCE) baselines in experiments, achieving comparable performance without requiring ground-truth segmentation images.

The accurate diagnosis of spinal pathologies depends heavily on radiological interpretation, yet automated systems are hindered by the lack of diverse, high-quality benchmarks. In this study, we present PhenSPINE, a Magnetic Resonance Imaging dataset comprising 16,813 images from 250 patients, curated to facilitate advanced deep learning research. We propose a robust diagnostic benchmark that integrates state-of-theart convolutional backbones with a Positional Encoding mechanism to explicitly model the anatomical context of intervertebral discs. Evaluating across four standard MRI sequences, our experiments demonstrate that the Sagittal T2-weighted sequence offers the most robust diagnostic value, achieving a superior Macro F1-score of 50.31%. We find that multisequence fusion strategies yield inferior performance compared to this single-sequence baseline, as the images across sequences in our dataset are significantly compromised by noise interference from surrounding anatomical regions. This work establishes a robust baseline and offers critical insights into sequence selection for spine analysis.

VLMs are increasingly deployed in AD systems, creating an urgent need for rigorous safety evaluation under rare yet safety-critical scenarios. Among these, interactions with vulnerable road users represent a major source of real-world failures. However, existing safety-critical scenario generation methods predominantly rely on simulator-based pipelines, which suffer from a substantial sim-to-real gap and often fail to capture realistic, diverse, and unforeseen human-vehicle interaction dynamics. We present SafeGen, a goal-conditioned diffusion framework for safety-critical scenario generation in VLMADs. Our key insight is to formulate scenario generation as a goal-conditioned diffusion process, where a predefined catastrophic end-state serves as a strong supervisory signal, guiding the generation of temporally coherent video trajectories that naturally evolve toward safety-critical outcomes. Building on this formulation, we introduce Context Grounded End State Reasoning, which leverages VLMs to analyze benign driving contexts and infer latent vulnerabilities in human-vehicle interactions, producing structured end-state specifications that induce high-risk scenarios. Conditioned on these targets, we further propose End State Conditioned Video Evolution, which grounds semantic threats into physically plausible visual dynamics. Specifically, we instantiate high-risk agents within the scene via depth-aware geometric projection, followed by boundary-conditioned diffusion to generate intermediate frames with consistent motion patterns and temporal coherence. Extensive experiments across 3 VLMADs demonstrate that SafeGen increases the Judge Overall Score, a metric using a VLM judge to evaluate VLMADs' understanding and decision-making, by 24.25% on average compared to SoTA baselines. Furthermore, fine-tuning a VLMAD improves performance in real-world driving scenes by an average of 15.9%.

Noisy and corrupted points can substantially degrade point cloud recognition performance, especially under challenging corruption settings. In particular, full fine-tuning of 3D pre-trained models may amplify the influence of outliers and overwrite robustness priors learned during pre-training, while naive parameter-efficient adaptation remains sensitive to corrupted tokens. To address this issue, we propose PSFT, a point-selection fine-tuning framework that improves robustness while remaining parameter-efficient. PSFT first estimates point-wise influence from pre-pooling features and adaptively retains minimally influential points to suppress outliers. Based on the selected subset, a prompt generation branch predicts layer-wise prompt tokens and injects them into a frozen backbone for lightweight downstream adaptation. To further mitigate residual noise after selection, we append a lightweight feature filter with bottleneck MLP transformation and Beta-gated residual blending to refine patch-token representations before prediction. Extensive experiments show that PSFT consistently reduces corruption error on ModelNet-C and ModelNet40-C across all tested 3D pre-trained backbones, while achieving the strongest ScanObjectNN-C results with ULIP-2 and Uni3D-B among the evaluated tuning strategies. Our implementation can be found at https://github.com/CVChMA/PSFT/tree/master.

Pain is a complex and pervasive phenomenon affecting a large percentage of the population, and accurate assessment is essential for effective clinical management and intervention. Computational pain recognition systems enable continuous monitoring, support clinical decision-making, and help mitigate pain-related distress and functional decline. This study introduces a unified tokenization framework for heterogeneous 3D modalities in pain recognition that provides a single processing pipeline across behavioral and brain-activity 3D data, without requiring separate architectures for each modality or handcrafted inductive biases. The framework preserves spatial, temporal, and time--frequency structure while mapping diverse inputs into a shared token space. Extensive experiments show that the proposed approach effectively processes facial videos and fNIRS data in both raw-signal and spectrogram-based representations. On the AI4Pain benchmark dataset, the proposed framework achieves state-of-the-art performance while maintaining high computational efficiency and enabling real-time assessment on both GPU and CPU hardware.

Automatic pain assessment from facial video remains challenging due to the spatial heterogeneity of pain-related facial cues. This study proposes ReFace, a spatial reorganization pipeline that divides facial input into four spatial quadrants before tokenization, rather than processing the entire face as a single region. Evaluated on the AI4Pain dataset, the proposed approach achieves $56.00\%$ accuracy on the test set using video only, achieving the highest reported accuracy under the fixed AI4Pain benchmark protocol among the compared methods. Notably, the four-quadrant configuration processes the same total pixel budget as the full-face input, yet achieves higher accuracy, suggesting that spatial reorganization can improve performance under the proposed tokenization design. A single quadrant region, processing just one quarter of those pixels, remains competitive at a fraction of the computational cost.

Automatic pain localization, which involves identifying the anatomical origin of pain from peripheral physiological signals without patient self-report, is a clinically critical but largely unaddressed problem, particularly for non-verbal patients. This paper presents a systematic comparison of classical feature engineering and deep sequence learning for subject-independent three-class pain localization using the AI4Pain 2026 Challenge dataset, which comprises four synchronously recorded wearable modalities: electrodermal activity, blood volume pulse, respiration, and peripheral oxygen saturation recorded from 65 participants under controlled TENS-induced pain. A 115-dimensional hand-crafted feature set spanning time-domain, frequency-domain, modality-specific, and cross-modal descriptors is benchmarked against end-to-end deep architectures. Extremely Randomized Trees achieves the highest macro-F1 of 0.539, outperforming the best deep model by 7.4 percentage points, with EDA spectral features emerging as the dominant discriminators. A consistent 26-point gap between pain detection (F1\,=\,0.815) and localization (F1\,=\,0.552) across all models points to a fundamental ceiling imposed by the anatomical diffuseness of peripheral autonomic pathways at 10-second resolution.

Tracking objects through state transformations is essential for understanding real-world dynamics. However, existing methods are computationally expensive. TubeletGraph recently showed impressive capabilities, but its inference cost (~$4.4$ seconds per object-frame on VOST) precludes any real-time deployment possibilities. We observe that TubeletGraph's overhead arises from building a spatiotemporal partition of the input video: (1) entity segmentation is computed densely for every frame regardless of whether a transformation occurs, and (2) every entity in the scene is tracked, scaling cost with scene complexity rather than the number of transformations of interest. To address both, we propose FluxGraph, a reactive variant that uses SAM2's internal multi-mask disagreement as a lightweight trigger for transformation detection, and removes the need for tracking all entities in the given video. FluxGraph is ~$3.3\times$ faster than TubeletGraph on VOST while improving tracking performance and preserving state graph quality. Furthermore, we also observe consistent speedups of $3.7-10.7\times$ across VSCOS, M$^3$-VOS, and DAVIS17 while maintaining performance. Code is publicly available at https://github.com/YihongSun/FluxGraph.

Large view synthesis models synthesize novel views through cross-view attention without explicit 3D representations, and recent studies have shown that they learn accurate spatial correspondence from RGB supervision alone. We observe that this correspondence generalizes beyond appearance. When non-photorealistic signals such as binary encoded panoptic labels are passed through the model, they are propagated to novel views with consistent spatial structure. These results indicate that the correspondence learned for RGB view synthesis can also propagate view-independent per-pixel labels. From this observation, we present the first work to extend large view synthesis models beyond appearance rendering to 3D scene understanding. We propose a panoptic segmentation pipeline that reuses a frozen view synthesis model to propagate panoptic labels from input views to novel views, without 3D reconstruction or any segmentation-specific training of the view synthesis model. Given panoptic labels on the input views, we encode them into binary channel representations and pass them through the same model to render target-view segmentation. On ScanNet, our method achieves segmentation quality on par with Gaussian based approaches requiring explicit 3D reconstruction, while outperforming them in novel view synthesis by more than 7 dB. The label propagation also transfers across datasets, surpassing these approaches on Replica without any fine-tuning.

Geo-spatial rasters of building footprint area are useful for a variety of tasks, such as monitoring urbanization, improving energy efficiency, and tracking greenhouse gas emissions. There are now multiple global building raster datasets, however there lacks an independent, comprehensive, and fair assessment of their accuracy. In this work, we evaluate the accuracy of four major global building products: Global Human Settlement Layer (GHSL), Microsoft's TEMPO (TEMPO), The Global Building Atlas (GBA), and Overture. As ground truth for assessing their accuracy, we use ORBITaL-Net, a globally diverse dataset of manually labeled building footprints. To ensure fairness, we evaluate products on grids of multiple spatial resolutions, and several conventional performance metrics. Our results indicate that either GBA or TEMPO generally achieves the highest overall accuracy, depending upon the particular evaluation criteria. We also stratify the accuracy of each product by several factors: geographic location, population density, and income groups. The results reveal that product accuracy can sometimes vary significantly with respect to these factors. Notably, all products are significantly less accurate in Africa and Asia. Most products also suffer significant accuracy reduction in high-density urban areas.

Dynamic RGBT (DRGBT) tracking aims to continuously localize a target when the available sensing modalities and observation platforms vary over time. Compared with conventional RGBT tracking with fixed RGBT inputs and a fixed observation platform, DRGBT tracking is more consistent with real-world collaborative perception systems, where targets may be observed by heterogeneous sensors from different viewpoints. However, existing benchmarks are still insufficient for systematically evaluating tracker robustness under real dynamic modality variations and cross-platform transitions. To address this limitation, we make the following contributions. 1) We construct DRGBT-1K, a large-scale high-quality benchmark for DRGBT tracking. It contains 1,045 sequences captured entirely in real-world scenarios and 795K RGBT frame pairs collected using UAVs and handheld RGBT devices, encompassing diverse real-world scenes, pronounced viewpoint changes, modality variations, and target appearance discontinuities. 2) We provide comprehensive annotations for fine-grained evaluation, including dense bounding boxes, target category labels, challenge attributes, frame-level modality labels and platform labels. DRGBT-1K covers 24 target categories, more than 15 scene types and 15 challenge attributes. 3) We establish a comprehensive benchmark by evaluating 20 representative multimodal tracking methods, including conventional RGBT trackers, modality-missing RGBT trackers, and DRGBT trackers under a unified evaluation protocol. 4) We release an unaligned version of DRGBT-1K and derive UGVT-1K to support broader research on unaligned multimodal tracking and UAV-ground collaborative tracking. 5) We develop an online evaluation platform for DRGBT-1K and provide a leaderboard that collects all methods evaluated on this benchmark.

Text-driven 3D scene editing with 3D Gaussian Splatting (3DGS) typically applies a 2D diffusion editor to views rendered from fixed training cameras, limiting both the spatial coverage of edits and the user's freedom to target specific objects in complex scenes. We present LB-Edit, a framework that addresses two coupled problems: where to place editing cameras for localized edits, and how to make per-view edits agree with one another so that the 3D scene remains consistent after fine-tuning. First, Attention-Guided Editing Camera Placement (ACP) probes the diffusion model's self- and cross-attention at multiple candidate camera distances to find where attention is well-contained in the region of interest, then places a compact, geometrically diverse editing camera set at that attention-optimal distance. Second, Multi-view Attention Alignment (MAA) steers the editor toward the same edit across views along two axes: it aligns appearance by sharing self-attention features via token-level correspondence, and aligns spatial location by lifting cross-attention maps onto the 3D Gaussians as a shared 3D attention field, suppressing both appearance and spatial drift. Experiments on multi-object and single-object scenes show that our method achieves the highest user preference in instruction fidelity, multi-view consistency, and editing locality, using as few as 5 editing views and reducing latency by up to 7x over existing methods.

Active sampling for accelerated MRI must distribute a tight sampling budget across spatial frequencies that carry very different kinds of information. Low frequencies hold most of the anatomical context; high frequencies carry the fine details that drive pathology assessment. Existing active samplers either treat both regions identically or restrict the action space to entire Cartesian rows, which forces a poor compromise at high acceleration. We propose HieraSample, a task-driven framework built around this hierarchy. A cosine-annealed curriculum lowers the acceleration factor from 20x to 4x across 80 acquisition steps while keeping a fully-sampled low-frequency disk at every step; a Mamba-based policy then picks individual high-frequency coordinates from features extracted by dual disease and severity classifiers. The reward is the per-sample reduction in class-weighted cross-entropy after each action, so a positive reward corresponds directly to a more confident correct prediction. On the fastMRI+ knee benchmark, HieraSample matches the fully-sampled oracle on ACL diagnosis from 4x to 10x acceleration, and improves on a recent Cartesian baseline by as much as 20.4 AUC points on ACL severity.

Autonomous driving requires understanding the road as a graph of drivable lanes and their connectivity, beyond the ego lane alone, to follow routes through intersections and reason about cross- and merging-traffic. Recent perception models infer such lane topology, i.e., lane segments together with their inter-lane connectivity (LCLC), from onboard sensors over a 360-degree BEV view. Due to neural perception's imperfections, their outputs retain structural instabilities such as missed detections, lost or incorrect LCLC, over-detection, and label flicker. This paper presents WASABI, a real-time post-processing pipeline that stabilizes lane topology outputs both within and across frames by treating lane segments and their LCLC connectivity as joint tracking targets, under onboard real-time constraints (10 Hz / 20 ms / up to 200 input lanes). The pipeline integrates segment tracking with connectivity, noise-robust topology-aware refinement, and a resource-constrained real-time design. On internal validation data (16 sequences), WASABI improves LCLC detection F1 from 0.834 to 0.948 (+0.114, +13.6%) and reduces centerline lateral error from 2.50 m to 0.95 m, while reducing detection false-positives by 24.6%. Temporal-stability metrics on the same data show LCLC toggle rate reduced by 63.3% and boundary-label flicker rate by 30.2%, confirming across-frame stabilization beyond per-frame accuracy.

Polarimetric synthetic aperture radar (PolSAR) image classification is a representative task for physics-aware GeoAI, where land-cover semantics are closely coupled with electromagnetic scattering mechanisms. Many existing complex-valued networks can preserve amplitude-phase information, but they are often limited in long-range spatial dependency modeling and usually incorporate polarimetric priors only as input-level or shallow auxiliary features. As a result, physical knowledge is insufficiently used to guide deep feature evolution. To address this issue, this paper proposes CV-SSMNet, a physics-aware complex-valued state-space network with scattering-aware feature modulation for PolSAR image classification. The proposed method builds a complex-valued state-space model (CV-SSM) in the original complex domain to capture long-range spatial dependencies while preserving polarimetric amplitude-phase coupling. Meanwhile, seven physically meaningful scattering priors, are encoded as FiLM-style modulation signals to adaptively recalibrate complex-valued representations during feature evolution. CV-SSMNet further integrates multi-scale complex convolutions, branch-wise CV-SSM encoding, prior-guided recalibration, and lightweight global context aggregation, enabling physically guided representation learning from local scattering structures to global spatial context. Experiments on three L-band benchmark datasets and an additional P-band BIOMASS evaluation demonstrate that CV-SSMNet achieves competitive accuracy, improved regional consistency, and better boundary preservation, supporting the effectiveness of embedding polarimetric scattering mechanisms into complex-valued long-range GeoAI representation learning.

Reinforcement learning with verifiable rewards (RLVR) has substantially improved language-model reasoning, yet its extension to vision-language models remains constrained by the lack of training data that are simultaneously broad, exactly verifiable, and reproducible. We introduce Trace, a taxonomy-guided environment for multidomain visual reasoning. Trace factorizes task construction into a scene grammar and an executable task program, separating visual realization from answer computation. A shared semantic state determines the rendered image, prompt, typed answer, verifier state, and replayable instance trace. The resulting environment comprises 1,000 tasks over 277 scene grammars and 11 visual domains, with controlled semantic and visual variation. RLVR on 64,000 Trace instances improves the macro-average across 24 external benchmarks by 3.51 percentage points for Qwen2.5-VL-3B and 4.06 points for Qwen2.5-VL-7B, providing evidence that broad procedural training can transfer beyond the generated task distributions. Project page: https://maveryn.github.io/trace/.

The Segment Anything Model 2 (SAM2) has advanced temporal promptable segmentation, yet its deployment remains hindered by heavy memory cross-attention overhead and redundant full-frame visual feature extraction. While recent methods explore efficiency via heuristic memory pruning and window-based sparse routing, they typically suffer from catastrophic performance degradation in complex segmentation scenarios replete with occlusions and distractors. To resolve these limitations, we propose \textbf{Lean-SAM2}, a holistic lightweight framework designed to address the above vulnerabilities while systematically eliminating computational redundancies. Specifically, Lean-SAM2 integrates three collaborative mechanisms: (1) Target-Anchored Memory Pruning (TAMP) safeguards target tokens against deceptive attention by modulating raw attention significance with semantic consistency against prompt-derived foreground anchors; (2) Temporal Condensation with Insurance Memory (TCIM) condenses historical context via a visibility-gated fusion while conditionally archiving high-confidence entries in a parallel insurance bank; and (3) Target-Anchored Risk-Aware Routing (TARR) selectively activates the heavy image encoder for target-related windows based on anchor similarity, utilizing a risk-aware fallback policy to trigger full-frame refreshes during volatile transitions. Extensive evaluations across multiple challenging benchmarks demonstrate that Lean-SAM2 establishes a superior balance between accuracy and efficiency. For example, on the LVOSv2 validation dataset, Lean-SAM2 achieves overall inference speedups of $1.412\times$ and $1.417\times$ on the SAM2.1-Large and SAM2.1-Base+, respectively, significantly outperforming Efficient-SAM2 while boosting the corresponding $\mathcal{J}\&\mathcal{F}$ scores by $5.0\%$ and $3.6\%$. Code is available at https://github.com/DeawhaleQwQ/Lean-SAM2.

Action Quality Assessment (AQA) aims to objectively evaluate performance quality from action videos. Most existing methods follow a ``one-by-one'' paradigm, training a separate model for each action type. This setting limits real-world deployment, as it requires prior action-type knowledge to select the corresponding model and suffers from poor generalization across diverse actions. To address these limitations, we study the challenging task of all-in-one AQA, which aims to assess heterogeneous actions within a single unified model. We propose a novel Mixture of Action Knowledge Experts (MoAKE) framework, designed to mitigate negative knowledge transfer caused by large semantic discrepancies among actions. MoAKE learns complementary experts that capture diverse action patterns within a shared semantic space and dynamically aggregates their knowledge to adapt the assessment to the input action. Each expert is tailored with segment-aware prototypes to handle varying temporal lengths, together with an Adaptive Intra- and Inter-Segment Relationship Modeling (AIISRM) module to model multi-granularity temporal dynamics. Furthermore, we establish comprehensive benchmarks for all-in-one as well as zero/few-shot AQA. Extensive experiments on three long-term datasets demonstrate that MoAKE significantly outperforms existing methods in the all-in-one setting, while also achieving consistent generalization on three short-term datasets under zero/few-shot evaluation. Code is available at https://github.com/XuHuangbiao/MoAKE.

Language-guided aerial perception aims to understand user-specified tiny targets in complex unmanned aerial vehicle (UAV) scenes. In real UAV deployment, the UAV must respond while it flies, so such perception runs in an online streaming manner, where frames arrive sequentially and the model responds to each one without access to future frames. However, applying current Multimodal Large Language Models (MLLMs) to this setting raises two challenges. First, targets viewed from the air are often tiny, yet the visual compression in existing MLLMs treats all regions equally and discards their fine-grained details. Second, understanding a continuous stream requires past-frame context, yet retaining the entire history is infeasible on resource-constrained onboard hardware, whereas discarding it causes the target to drift or disappear. We address the tiny object and streaming challenges from both data and method perspectives. From the data perspective, we present \textbf{DroneEyes}, the \textbf{first} pixel-level and open-vocabulary referring-segmentation dataset for tiny aerial targets, comprising $2,140$ high-definition videos and $176,623$ pairs across Object Description and Referring Expression tasks, with dense per-frame masks. From the method perspective, we propose \textbf{SkyAnchor}, an MLLM with two designs to the above challenges: a Semantics-Aware Token Router that preserves small-target under a reduced visual-token budget, and a Hierarchical Memory Bank that keeps the target consistently understood on streams.

Weakly supervised segmentation relies heavily on class activation maps (CAMs) to initially localize target regions. However, CAMs are often noisy and prone to catastrophic failures. Existing remedies typically introduce additional training stages or prototype learning, increasing computational cost and reducing robustness. In this paper, we propose a training-free prototype-free framework that rectifies unreliable CAMs by exploiting temporal and structural coherence in volumetric data as a free lunch. Our approach is built on two key components. First, we introduce Variance-Reduced Activation Aggregation (VRAA) which suppresses noise and amplify coherent semantic signals. We provide a theoretical justification by modeling CAMs as high-dimensional random vectors and show that aggregation yields provable variance reduction. Second, we design a Bidirectional Extremity Rectification (BER) mechanism that detects and rectifies implausible activations through bidirectional extremity checks, effectively mitigating extreme-value failures without learning additional parameters. Our method is model-agnostic and can be seamlessly integrated with existing pipelines. Extensive experiments on multiple public benchmarks demonstrate substantial improvements over state-of-the-art weakly supervised methods, achieving up to 20% Dice and 40% mIoU gains while reducing inference time by more than 5 times. These results indicate that leveraging coherence as an implicit inductive bias yields a principled and efficient approach to stabilizing weakly supervised volumetric segmentation. Our code will be available.

Infrared and visible image fusion (IVIF) integrates the complementary information of two modalities into a single image with richer scene content. While existing methods are largely built on artificial neural networks (ANNs), which densely compute over all activations, spiking neural networks (SNNs) communicate through sparse binary spikes and compute only where and when a spike occurs, offering a route to more energy-efficient fusion. However, directly applying SNNs to IVIF creates a fundamental tension: cross-modal fusion relies on fine-grained responses from both modalities, whereas binary spikes can discard complementary cues that remain below the firing threshold. The membrane potential retains these subthreshold responses before firing, letting both modalities jointly shape the output when integrated at this stage. Building on this, we propose CIS-Fuse, a spiking network that performs cross-modal fusion directly at the membrane-potential level. At its core is the current injection spiking (CIS) operator, which injects one modality as a gated auxiliary current into the driving neuron of the other, so the two integrate before spike firing, with a per-channel learnable injection strength that adaptively regulates the modulation magnitude. Building on CIS, we construct a bidirectional cross-modal fusion (BCMF) module and deploy it on a dual-branch architecture with asymmetric stacking depths, where the two branches develop a clear functional specialization. Extensive experiments on four IVIF benchmarks and on downstream detection and segmentation show that CIS-Fuse achieves fusion quality on par with state-of-the-art ANN-based methods while inheriting the energy efficiency of spike-based computation, with roughly an order of magnitude lower inference energy than the similarly-sized ANN-based DCEvo. Code will be released upon publication.

Thermal-to-visible face translation presents fundamental challenges including geometric discontinuities, semantic attribute mismatches, and identity degradation. We propose MTVDiff, a novel multimodal latent diffusion framework that synergistically integrates depth and textual information to address these limitations while preserving identity characteristics. The MTVDiff framework presents three core technical contributions: (1) a Dual-Branch Cross-Attention Fusion (DBCAF) module for multi-scale thermal-depth feature extraction and fusion; (2) a Gated Text-to-Visual Feature Alignment mechanism for semantically-guided generation; and (3) Spatial Feature Transformations (SFT) for adaptive multimodal prior integration. Extensive experiments on the MCXFace and SpeakingFaces datasets demonstrate that our multimodal approach significantly outperforms existing GAN-based and diffusion-based approaches across multiple metrics, achieving substantial improvements in both image quality and face verification performance, with FID reductions of up to 48.3% and Rank-1 accuracy improvements of up to 8.9\%. Our work provides a robust solution for face recognition systems operating under varying illumination conditions and advances the state-of-the-art in cross-spectral facial image translation through effective multimodal integration.

Understanding instrument-tissue interactions is essential for context-aware surgical AI and autonomous robotic surgery. Pretrained vision-language models (VLMs) and vision encoders offer an alternative to conventional interaction classifiers by transferring broad visual and semantic knowledge. However, adapting them to fine-grained surgical interactions remains challenging: (1) freezing the vision encoder depends entirely on pretrained representations that may retain noise and provide weak spatial localization, while (2) full fine-tuning can improve global semantic alignment without ensuring that the encoder learns meaningful features in the correct action region. We address these limitations by introducing LAViFiT, an end-to-end latent-action-guided framework for vision-language fine-tuning. An inverse dynamics model captures the visual changes induced by each action, while a forward world model drives the encoder to represent action-relevant regions. A patch-level SIG Regularizer further prevents local feature collapse without additional supervision, such as bounding boxes or pseudo-labels. Experiments across multiple encoders and datasets improve recognition and image-text alignment, while representation analyses show stronger grounding over the complete instrument-tissue interaction region and more spatially coherent features.

Text-guided video editing with diffusion models is impractically slow, hindered by costly multi-step sampling and inversion. We present OSVE, the first framework to successfully adapt one-step Text-to-Image (T2I) models for high-quality video editing, addressing the core challenges of inversion, editability, and temporal consistency. To bypass slow iterative inversion, we train a learnable encoder that predicts the initial noise for each frame in a single forward pass. This encoder is trained with a novel Structure-Aware Editing (SAE) loss on a curated dataset of structurally-aligned image pairs, teaching it to preserve the source video's geometry during edits. For temporal coherence, we introduce Unified-Frame Editing (UFE), a technique that concatenates frame latents to facilitate cross-frame attention in a single generation step. Furthermore, for long videos, a sliding-window strategy with an anchor frame maintains global consistency. Our extensive experiments demonstrate that OSVE achieves editing quality comparable or superior to state-of-the-art multi-step methods, while operating approximately 155--171 times faster. This breakthrough paves the way for practical, real-time video editing applications. Code is available at https://github.com/KU-VGI/OSVE.

Deep learning frameworks like nnU-Net achieve state-of-theart brain lesion segmentation performance but remain difficult to deploy in clinical research environments due to, among other reasons, software dependencies and computational requirements. We introduce StrokeSeg2, a lightweight, modular, cross-platform C++/Qt framework designed to adapt resource-intensive 3D stroke segmentation pipelines into portable and reproducible applications. To improve compatibility with standard clinical workstations, we investigate the combined effect of architectural compression through knowledge distillation and inference optimisation using ONNX Runtime with Float16 quantisation. Across heterogeneous hardware configurations (CPU, integrated GPU, and dedicated GPU) architectural distillation emerged as the primary contributor to efficiency gains, contributing to over 90% reduction in energy consumption and an average 84% reduction in inference time. Specifically, we identify a 0.84M-parameter student model as the most favourable trade-off, reducing the original 102.3M-parameter teacher architecture to a 2.1 MB disk footprint while preserving robust lesion localisation and competitive segmentation performance. This small footprint supports the development of a self-contained installer for clinical workstation targets. Finally, StrokeSeg2 packages these optimisations into standalone installers for Windows, macOS, and Linux. By providing both graphical and commandline interfaces without Docker or external environment dependencies, StrokeSeg2 facilitates deployment of high-performance segmentation workflows for routine clinical research pipelines.

Multimodal large language models (MLLMs) are increasingly expected to automate visualization development by generating code directly from visual designs. However, existing evaluations mainly focus on single-chart generation and overlook coordinated multi-view interface construction, which requires joint reasoning about data semantics, view coordination, and interaction logic. Consequently, MLLM capabilities in this setting remain underexplored, and the field lacks a dedicated benchmark for systematic assessment. We introduce MV-Bench, a benchmark for evaluating MLLMs on coordinated multi-view interface construction. Instead of relying on incomplete or inconsistent open-source implementations, we use Tableau workbook files as ground truth because they explicitly encode data bindings, visual mappings, and interactions. We develop a multi-stage pipeline that converts these specifications into executable web interfaces through structured intermediate representations. The benchmark contains 92 base interfaces and 1,048 verified instances created by recombining chart types, datasets, and interaction patterns. Each instance includes executable code, a rendered interface, a dataset, and interaction annotations. We evaluate five state-of-the-art MLLMs in a single-pass setting using metrics for visual fidelity, data binding correctness, and interaction completeness. The strongest model achieves 75.45 percent accuracy in visual layout reproduction, but only 21.71 percent in data binding and 11.68 percent in interaction completeness. These results show that current MLLMs can reproduce visual appearance but remain limited in generating the data semantics and interactive logic required by coordinated multi-view interfaces. Iterative refinement improves code executability but does not substantially reduce the gap in data binding and interaction generation.

Long-range vehicle trajectories provide important spatio-temporal evidence for traffic safety analysis, autonomous driving evaluation, and data-driven traffic management, yet continuously recovering them from fixed highway cameras remains difficult. As vehicles recede into distant road regions, perspective compression and scale decay often fragment or prematurely terminate automatic tracklets, even when their continuation remains identifiable from motion consistency across neighboring frames. We formulate this problem as recovering the far-range continuation of a vehicle trajectory from a reliable near-field tracklet. We introduce LoRFT, to our knowledge the first open benchmark dedicated to long-range vehicle trajectory reconstruction from fixed highway cameras. LoRFT comprises 22 expressway surveillance scenes, 366,109 video frames, 6,601 manually verified trajectories, 2,694,889 bounding boxes, road-geometry annotations, scene-level splits, and evaluation scripts. We further propose Map-RSTNet, a map-aware residual sequence-to-sequence model that reconstructs distant trajectories in a road-geometry-aligned state space and dynamically refreshes local road geometry during decoding. On LoRFT, Map-RSTNet reduces ADE, FDE, and 5-second RMSE by 11.0%, 15.4%, and 10.5%, respectively, relative to the strongest baseline. These results demonstrate that road-geometry-aware reconstruction can extend usable trajectory records from existing fixed-camera infrastructure. LoRFT provides a reproducible testbed for long-range vehicle trajectory reconstruction.

Synthesizing native 2K multi-garment virtual try-on is a formidable frontier in digital fashion, critically bottlenecked by two fundamental limitations: the O(N^2) memory explosion induced by 2k conditions, and the spectral bias of diffusion models that over-smooths high-frequency fabric details. We present WearWow, an end-to-end, mask-free generative framework that pioneers ultra-high-resolution multi-garment synthesis. To mitigate the memory explosion , we propose Adaptive 2D Token Packing (ATP). ATP leverages inherent garment sparsity to algorithmically pack heterogeneous items onto a unified 2D canvas and prune uninformative background tokens, minimizing the effective sequence length and subsequent memory overhead while rigorously preserving 2D spatial priors. To rectify texture degradation, we introduce the Multi-dimensional Try-on Reward (MTR) system. MTR synergizes a Semantic Guidance Reward to explicitly drive tactile restoration with a Cloth Distribution Reward to implicitly anchor the physical distribution, a joint formulation that effectively mitigates the severe reward hacking. Furthermore, we curate WearWow-2K, an extreme-quality dataset comprising native 2K triplets, providing physically correct spatial interactions that naturally empower the model's mask-free generation. Extensive experiments demonstrate that WearWow establishes a new state-of-the-art, exceeding existing commercial baselines in native 2K multi-garment synthesis.

This work introduces G-MAD, an open-source framework that uses Arma3 to generate synchronized multi-view RGB-T data for aerial object detection. G-MAD addresses key limitations of real-world aerial dataset construction, including limited viewpoint control, imperfect RGB-T alignment and high annotation cost. The framework supports structured scenario specification, controllable multi-view camera placement, simultaneous visible/thermal capture, and automatic bounding box annotation using engine-level geometric metadata. These capabilities enable controlled studies of viewpoint variation, multi-modal fusion, and synthetic-to-real transfer in aerial object detection. Besides, using G-MAD, we construct and release AMOD, a new large-scale multi-view aerial RGB-T object detection benchmark. The source code and the dataset are available at https://unique-chan.github.io/G-MAD-Project.

Diffusion Magnetic Resonance Imaging (dMRI) is a powerful tool for probing brain microstructure, but clinical acquisitions are often limited by low out-of-plane resolution, resulting in degraded structural information and reduced utility for advanced analysis. We introduce SIINR (Structurally Informed Implicit Neural Representations), a general framework for super-resoltion of clinical dMRI datasets while quantifying uncertainty in the reconstructed outputs. SIINR utilizes a supervised 3D U-net as a prior and combines it with a self-supervised implicit neural representation (INR) that fuses the high-resolution prior and the original low-resolution data. The INR enables joint modeling across spatial and angular domains, enforces data consistency, and provides analytic approximate posterior distributions for downstream uncertainty quantification. We validate the framework on a diverse set of open-access dMRI datasets, demonstrating that SIINR outperforms standard interpolation methods in both quantitative error metrics and qualitative anatomical fidelity. Experiments on clinical cases, including subjects with multiple sclerosis and brain lesions, illustrate the framework its ability to propagate intensity changes and flag uncertain regions in challenging scenarios. SIINR is flexible, modular, and can be adapted to different upsampling ratios and downstream tasks, providing a principled approach for enhancing clinical dMRI and supporting robust interpretation of derived neuroimaging metrics.

Electronic Theater Programs (ETPs) serve as critical promotional media in the performing arts, comprising a multi-page collection of heterogeneous visual assets such as theatrical posters, performance details, and character portraits. However, existing text-to-image paradigms struggle with such complex design tasks due to their inability to comprehend long-context narratives and maintain visual consistency across multiple distinct pages. To address this, we introduce ETPDesigner, a collaborative Multi-Agent framework that directly synthesizes high-quality ETPs from raw dramatic scripts. Emulating a professional design pipeline, our framework orchestrates specialized agents for semantic script analysis, core poster synthesis, functional background generation, and the stratified composition of character assets. Central to ETPDesigner is a global style anchor mechanism that extracts visual priors from the core poster to enforce strict aesthetic uniformity across all generated components. Furthermore, we elevate the ETP from a static publication to an immersive interactive companion. By integrating portrait animation, customized speech synthesis, and persona-grounded Large Language Models (LLMs), our system enables users to engage in real-time, voice-enabled conversations with the generated virtual characters. To rigorously benchmark this task, we construct ETP-Pro, a domain-specific benchmark of professional theater posters and high-quality character portraits. Extensive evaluations demonstrate our method's superiority in producing semantically faithful, aesthetically consistent, and highly interactive program sets.

Cross-view geo-localization between UAV and satellite imagery remains a fundamental yet highly challenging task, especially under large off-nadir views where drastic perspective distortions, occlusions, and appearance gaps occur. Existing benchmarks and methods primarily focus on near-nadir scenarios and often overlook the importance of structural scene understanding and intra-domain relational constraints, limiting their performance in real-world deployments. In this work, we introduce OffNadirLoc, a new benchmark for large off-nadir UAV-to-satellite geo-localization. To tackle the unique challenges posed by off-nadir perspectives, we further propose ONLoc, a framework that incorporates a structure-aware contextual weighting mechanism to dynamically emphasize reliable local features while suppressing ambiguous or repetitive regions. Additionally, we design a view-coherent learning strategy, which treats one satellite image and the corresponding UAV images from multiple views as a cohesive semantic group. This set-level supervision enables the model to learn viewpoint-invariant and discriminative features, making it more effective at capturing multi-view consistency than conventional pairwise contrastive learning. Extensive experiments on the OffNadirLoc benchmark and four near-nadir datasets demonstrate that our method consistently outperforms state-of-the-art approaches while exhibiting strong zero-shot generalization to unseen datasets without additional training. The code will be released at https://montalario.github.io/offnadirloc/.

Accurate yield forecasting at the individual-plant level is critical for precision agriculture and supply-chain planning, yet public datasets capturing both visual growth dynamics and per-plant measurement labels are scarce. In this paper, we introduce a novel, annotated image time-series dataset of 691 sweet pepper plants monitored over two growing seasons, comprising 4837 images with per-plant fruit counts categorized by maturity. We propose a multimodal deep learning framework that fuses high-dimensional image features, extracted using the DinoV3 encoder, with numerical count measurements. Our architecture utilizes a Long Short-Term Memory (LSTM) network to model temporal dependencies and handles irregular sampling intervals common in greenhouse monitoring. Through quantitative experiments, we demonstrate that this multimodal approach reduces RMSE over a persistence baseline by 33% and 38% in the 2022 and 2023 seasons, respectively, with a further 1.2% average gain over a measurement-only model. Furthermore, we employ Deep Ensembles and Gaussian Negative Log-Likelihood (NLL) to provide calibrated uncertainty estimates, with an Uncertainty Calibration Error (UCE) ranging from 0.39 to 0.89 depending on the cross-season evaluation direction, offering a principled confidence signal for real-world agricultural decision-making. We release the dataset and code to support reproducible research and to accelerate development of data-driven yield forecasting methods for horticultural crops.

Stereo matching is a fundamental task in 3D reconstruction. Despite remarkable advances, the prevailing paradigms formulate stereo matching as a deterministic regression problem, collapsing the multimodal distribution modeling into a single-point estimation. This formulation suffers from a regression-to-mean bias, frequently struggling with ambiguous regions. In contrast, we introduce a prior-guided generative framework that integrates deterministic matching regression and generative distribution modeling within a complementary formulation. Built upon this formulation, we introduce StereoFlow through three key components: (i) a two-stage progressive cascade matching network that progressively produces multi-resolution stereo conditions with complementary matching cues; (ii) a pixel diffusion transformer (termed StereoDiT) with a frequency-decoupled architecture for modeling correspondence ambiguity; (iii) a few-step flow matching objective (termed Transition Flow Matching) for efficient optimization. In summary, \textsc{\textbf{StereoFlow}} achieves strong geometric consistency and rich fine-grained details in ill-posed, discontinuous regions and under zero-shot generalization. Extensive experiments demonstrate that the proposed StereoFlow establishes multiple state-of-the-art results across benchmarks, including Scene Flow, KITTI, ETH3D, and Middlebury.

Sentinel-2 imagery offers open access, global coverage, and frequent revisit times, making it attractive for practical building mapping at scale; however, its native 10m resolution makes building vs non-building classification challenging, particularly for small or sub-pixel buildings, and performance can vary with both seasonality and the heterogeneity of built-up environments. This paper introduces a Sentinel-2 building-detection framework designed to systematically quantify these effects and to support more formalised, practice-oriented model selection. We construct a dedicated multi-temporal Sentinel-2 dataset over the Warsaw region and derive binary ground-truth masks by rasterising official Polish topographic database (BDOT10k) building footprints onto the Sentinel-2 pixel grid. Using two established convolutional segmentation backbones (U-Net and DeepLabV3+), we first perform scene-specific fine-tuning to select a robust architecture and identify the best monthly models for L1C and L2A products separately. We then conduct cross-temporal inference by applying each best monthly model to all scenes, enabling an assessment of (i) which months provide favourable training and inference conditions, (ii) how performance transfers between seasons, (iii) the impact of processing level, and (iv) how these effects differ across built-up typologies. Based on these results, we provide practical guidance for routine Sentinel-2 building classification under varying acquisition periods and settlement characteristics.

Robust out-of-the-box performance is essential for the clinical deployment of deep learning models in medical imaging. An important but underexplored factor affecting model generalisability is intensity normalisation, particularly for magnetic resonance imaging (MRI), where image intensities vary across scanners and protocols. In this study, we systematically compared seven normalisation methods and their impact on the performance of a 3D U-Net model for meniscus segmentation from knee MRI. The methods included standard scaling approaches, histogram-based techniques, and a Gaussian Mixture Model (GMM)-based method. Models were trained on the IWOAI 2019 dataset and evaluated on both internal and external test sets (SKM-TEA) to assess generalisability. Performance was similar internally but differences were significant on external data, with Z-score, Ny\'ul histogram matching, and CLAHE showing greater robustness than other methods. However, these differences were small compared to the significant performance drop observed between datasets. Overall, while intensity normalisation had a measurable effect on model generalisability, its impact was limited relative to the effects of domain shift, highlighting the need for complementary strategies for robust deployment.

We propose importance-aware pruning for diffusion models, a training-free framework that prioritizes preserving parameters critical to semantically salient image regions. To do so, we incorporate spatial importance maps -- derived from conditioning signals or model attention -- into the pruning objective. This produces parameter rankings aligned with perceptual relevance rather than uniform reconstruction error. On MS-COCO dataset, our proposed approach consistently retains subject fidelity and structural correctness at high compression ratios where conventional pruning causes visible degradation. These results demonstrate that content-aware objectives are key to perceptually faithful compression of generative models.

Vision-centric 3D occupancy prediction provides dense scene representations essential for autonomous driving and robotic navigation, yet existing methods struggle to scale to high voxel resolutions due to prohibitive computational costs. To address this, we introduce GaussianSeed, a progressive multi-scale Gaussian occupancy prediction framework that organizes primitives into a coarse-to-fine hierarchy. Benefiting from this hierarchical design, GaussianSeed effectively circumvents the memory bottlenecks inherent in dense representations, successfully scaling to a $0.1\text{m}$ spatial resolution while maintaining real-time inference capabilities. To comprehensively evaluate high-resolution geometric perception, we further construct TJScenes, a panoramic six-camera occupancy dataset with highly detailed $0.1\text{m}$ annotations. Extensive experiments on Occ3D-nuScenes and TJScenes demonstrate that GaussianSeed delivers the lowest latency among all evaluated methods while maintaining highly competitive accuracy, advancing the efficiency-quality frontier of high-resolution 3D occupancy prediction.

Deep gaze estimation works well in controlled capture but degrades in unconstrained settings, where systems must reject unreliable predictions. Single-pass uncertainty (e.g., heteroscedastic regression) infers uncertainty from pixels without explicit input-validity cues, while sampling based methods are often too costly for real time use. We propose Factor-Informed Uncertainty Distillation (FIUD), a teacher-student framework that aligns uncertainty with interpretable image-quality failure modes. A gradient-boosting teacher predicts expected gaze error from factors such as illumination, sharpness, eye visibility and symmetry; a neural student distills these signals via curriculum learning and ranking supervision into a lightweight single-pass uncertainty head. Across ETH-XGaze, Gaze360, and MPIIFaceGaze (>300k samples), FIUD improves uncertainty, error rank correlation and selective prediction versus deterministic and sampling-based baselines, with the largest gains in unconstrained settings.

Aims: Cardiovascular magnetic resonance (CMR) imaging enables non-invasive assessment of myocardial structure, function, and pathology, but requires substantial experience in interpretation of CMR images that could be supported by artificial intelligence (AI)-based models. However, use of AI models for enhanced CMR reading is limited by labor-intensive data curation, suboptimal model performance, and unclear implementation pathways. Methods and results: We developed an automated data curation pipeline for CMR-based cardiovascular disease (CVD) diagnosis, integrating open-source locally-run large language models (LLMs) to extract diagnostic labels from narrative CMR reports and preprocessing multimodal imaging data, including cine and late-gadolinium-enhancement (LGE) CMR sequences. Three vision foundation models (DINO, VST, UMedPT) were fine-tuned across these modalities in a two-stage approach. The dataset comprised hypertrophic cardiomyopathy (HCM), dilated cardiomyopathy (DCM), ischemic cardiomyopathy (ICM), cardiac amyloidosis (CA), and normal controls (NOR). A total of 988 curated cases were randomly divided into 742 for training and 246 for validation. Fine-tuned AI-models achieved high discriminative diagnostic performance on an independent test set comprising 1067 patients , with individual AUC-ROC values of up to 0.937 for the correct diagnosis of HCM and 0.945 for cardiac amyloidosis. Ensemble strategies combining multiple models and modalities further improved AI-based diagnostic accuracy and robustness, achieving the highest overall diagnostic performance for HCM (AUC=0.959, CI [0.936-0.978]), CA (AUC=0.966, CI [0.939-0.986]), NOR (AUC=0.872, CI [0.852-0.894]), DCM (AUC=0.848, CI [0.808-0.885]) and ICM (AUC=0.840, CI [0.809-0.868]). All training and inference code, along with the trained model weights, are publicly available on https://github.com/sinaamirrajab/CMR_CVD.

Contextual entrainment is the tendency of a model to let auxiliary context in its input pull its output, independently of whether that context is relevant, true, or even meaningful. Recently, it has been identified and given a mechanistic account in unimodal language models. Whether and how it manifests in vision-language models (VLMs) is, by contrast, largely unexamined, and the field lacks a purpose-built instrument with which to investigate it. We take the position that studying contextual entrainment in VLMs requires more than porting an existing text-only benchmark to the multimodal setting: it requires a taxonomically structured, dual-modality instrument whose conditions are constructed around the item at hand (the depicted image in the textual stream, the textual query in the visual stream). We argue that the move to VLMs is substantive rather than incremental. It makes entrainment a dual phenomenon, drivable independently by textual and by visual context, and it opens a veracity distinction (context that is false of the depicted scene yet possible in the world) that has no counterpart in the unimodal, world-knowledge-only formulation of prior work. To make this position concrete and actionable, we introduce ENTRAP-VL (ENTRainment Assessment Probe for Vision and Language), a manually curated dataset of 1,500 items across eight categories, organized by a taxonomy that spans two axes, i.e., the association of context with the item and its relationship to truth, and split into a textual-entrainment stream (eight context conditions) and a visual-entrainment stream (three context conditions). We do not claim to measure entrainment in any particular model; we provide the instrument, the taxonomy that motivates it, and the evaluation protocols it enables, so that the community can investigate the phenomenon rigorously. We will release the dataset and its documentation publicly.

Global visual localization of unmanned aerial vehicles (UAVs) using remote-sensing reference maps has attracted increasing attention. However, acquisition-time and imaging-platform differences between UAV and reference imagery induce substantial cross-domain appearance and viewpoint shifts, challenging robust six-degree-of-freedom (6-DoF) pose estimation. We address these shifts by sampling UAV-viewpoint reference views from Google 3D Tiles across locations, altitudes, and orientations. A two-stage cross-domain fine-tuning recipe adapts SALAD using pose-near positives and geographically distant hard negatives, while local geometric consistency re-ranks the Top-K candidates. We further propose Retrieval-In-Matching (RIM), which freezes the adapted DINOv2-B retriever and distils a local-descriptor decoder that reuses its token field alongside a shallow VGG19 detail stream. One query-side DINOv2-B forward thus serves both SALAD retrieval and local description, eliminating a second foundation-model backbone while preserving retrieval descriptors by construction. We evaluate RIM zero-shot on the reconstructed EPFL Urbanscape and self-collected Chang'an Park datasets, both geographically disjoint from the training data. RIM outperforms ten recent retrieval baseline families. At 25/50 m under the full 3D distance metric, it improves Recall@1 over SALAD by 8.55/13.77 percentage points on EPFL and 4.45/8.94 points on Park. At Top-K=5, the complete measured localization query, including retrieval, candidate matching, and robust geometric verification, takes 67.9 ms end-to-end: 1.8 times faster than the strongest separate sparse-matching baseline and over 40 times faster than RoMa, while achieving comparable re-ranking accuracy. These results establish an efficient and deployable pipeline for UAV global visual localization in GNSS-challenged environments.

Autoregressive (AR) video diffusion models have become a promising paradigm for long and streaming video synthesis, but the continuously growing Key-Value (KV) cache makes attention the dominant inference cost, especially at high resolution where each frame contributes many tokens. Existing remedies either evict the cache with coarse heuristics that cause inter-frame flickering, or require model re-training. We propose HeadCast, a training-free, plug-and-play acceleration framework built on the observation that a pre-trained AR model's attention heads exhibit stable, heterogeneous behaviors. After a short warm-up, HeadCast performs a one-time classification at the maximum-noise step that sorts every head into one of four archetypes: Sink, Dummy, Spatial, and Global, and restructures the monolithic KV cache into head-specific pathways. Crucially, it retains the Global heads that preserve the long-range temporal consistency aggressive eviction destroys. Because the Spatial pathway operates on a fixed-size grid, its savings grow with resolution: across state-of-the-art AR models, HeadCast accelerates inference by up to 1.62x at 720P and 1.95x at 1080P, while keeping VBench quality on par with full attention and largely flicker-free. Code is available at https://github.com/sjlgaga/HeadCast .

We present a two-stage vision system that detects EEG cap electrodes in a live webcam stream and validates their anatomical placement in real time. A single-class YOLO detector localises electrodes; a geometric stage assigns each detection to a named 10-20 role from facial landmarks. Evaluating under subject-disjoint leave-one-subject-out (LOSO) cross-validation across five subjects wearing the clinically-validated Small/Medium/Large caps, the detector attains [email protected] = 0.94 +/- 0.07 across five held-out folds (0.96 pooled). A dedicated leave-one-cap-out axis, holding out every frame of a cap regardless of subject, leaves Medium and Large [email protected] within 0.01 of LOSO (0.97, 0.97) while Small drops to 0.72 +/- 0.28, a gap confounded with subject familiarity rather than cap style. Geometric augmentation (rotation, perspective, mixup) improves in-plane-roll robustness and temporal-electrode recall at no inference cost, and a landmark-driven head crop extends the usable distance range, lifting [email protected] from 0.23 to 0.45 at 0.6 x apparent scale. A compact mobile-candidate backbone (YOLOv10n) keeps the detector at real-time throughput (19 FPS) on a commodity CPU at 640 px.

Attention Mechanism (AM) selectively focuses on essential information for imaging tasks and captures relationships between distant pixel neighborhoods to compute feature representations. Accelerated MRI reconstruction benefits from AM, as the imaging process involves Fourier domain measurements that influence image representation non-locally. However, AM-based models are more adept at capturing low-frequency information with limited capacity for high-frequency representations, restricting models to smooth reconstruction. Additionally, AM-based models need mode-specific retraining for multimodal MRI data, as their knowledge is restricted to local contextual variations that may be inadequate to capture transferable features across heterogeneous domains. To address these challenges, we propose a neuromodulation-based discriminative multi-spectral AM for scalable MRI reconstruction that can (i) propagate context-aware high-frequency details for high-quality reconstruction, and (ii) capture features reusable across deviated unseen domains in multimodal MRI. The proposed network consists of a spectral filtering CNN to capture mode-specific transferable features and a dynamic high-pass kernel generation transformer focusing on high-frequency details. We evaluate our model on comparative studies in supervised and self-supervised learning, diffusion model-based training, closed-set and open-set generalization under heterogeneous MRI data, and interpretation-based analysis. Our method offers scalable, high-quality reconstruction with best improvement margins of ~1 dB in PSNR and ~0.01 in SSIM under unseen scenarios. Code: https://github.com/sriprabhar/SHFormer

Existing human--object interaction (HOI) video generation methods are largely limited to offline short-video generation with complex driving conditions, making them unsuitable for real-time interactive applications. We present \emph{StreamHOI}, a low-latency streaming framework for long-duration HOI video generation. Instead of converting heavily conditioned HOI pipelines into streaming systems, we study how an image-to-video streaming generator should organize historical memory to preserve interactions under bounded latency. We find that the standard sink-local memory design faces a trade-off in streaming HOI generation, and different transformer blocks show different historical-memory preferences for HOI regions and surrounding regions. To match memory composition with block behavior, StreamHOI performs offline HOI-aware block profiling and applies bias-guided memory-specialized training to adapt the generator to block-specific memory layouts. We further introduce a memory distance scaling module to strengthen long-range access to early interaction states. Extensive comparisons with both long-video baselines and recent HOI generation methods demonstrate that StreamHOI achieves strong interaction plausibility, object fidelity, human quality and efficiency, reaching 17.6 FPS with 0.75s first-chunk latency.

Frozen perception foundation models encode rich geometric, semantic, and dynamic knowledge. Yet narrow conditioning interfaces may attenuate task-relevant cues, while static fusion cannot adjust expert contributions to each scene. We cast this challenge as the prior-to-plan transfer problem and introduce PerceptDrive, a perception prior world-action modeling framework with adaptive expert routing. PerceptDrive feeds teacher-distilled priors from a frozen, driving-adapted provider and dense observation latents from a frozen self-supervised video encoder into a trainable expert-routed world-action model. Expert-specific query branches process these signals, while a prior-retention objective anchors each branch to its prior. A router predicts soft gates from a shared scene representation and combines the expert conditions before trajectory generation. During training, privileged rule-based sub-metric estimates for branch-specific trajectory drafts provide soft-gate distillation targets. The predicted action-free future latent conditions a flow-matching actor. At inference, privileged components are absent; with one front-facing camera, PerceptDrive generates one trajectory per planning step without test-time scoring, reranking, or search. Experiments show that PerceptDrive achieves state-of-the-art performance with 90.4 PDMS on NAVSIM v1 and 90.2 EPDMS on NAVSIM v2, outperforming existing methods. Ablations confirm complementary gains from prior retention and scene-conditioned routing, alongside differential reliance on the three priors. These results demonstrate that preserving and adaptively routing perception priors improves direct planning without test-time candidate selection.

Remote sensing image editing aims to modify remote sensing images according to natural language instructions while preserving geographic rules and sensor observation characteristics. Existing benchmarks mainly target natural images or general visual scenes, and thus may not fully capture the reasoning, regional control, and sensor-consistency abilities required in remote sensing editing. To fill this gap, we introduce RS-RIE-Bench, the first benchmark for reasoning-guided remote sensing image editing. RS-RIE-Bench organizes tasks into three categories: temporal reasoning, causal reasoning, and spatial reasoning. These categories capture temporal evolution, causal consequence, and spatial imaging consistency in remote sensing scenes. The evaluation protocol covers three dimensions: target region plausibility, non-target region preservation, and image quality consistency. We further demonstrate the feasibility of MLLM-based evaluation through cross-judge consistency analysis and stratified expert review. Systematic evaluation on eight open-source and closed-source image editing models shows that current models still have clear limitations in reasoning-guided remote sensing editing. Even the strongest model achieves only 24.28\% overall accuracy under the strict joint-satisfaction criterion, while the mean relaxed joint-4 success rate across all eight models is 32.23\%. Causal reasoning and spatial reasoning remain especially challenging, and several open-source models are close to zero in some categories. These results show that RS-RIE-Bench can effectively reveal the limitations of current models in geographic reasoning, regional control, and sensor-consistent generation. It also provides a standardized benchmark and a clear research direction for future remote sensing intelligent editing models.

Visible-infrared (VIS-IR) alignment is a key pre-training task for robust multi-sensor perception. Most existing methods use uniform patch-wise contrastive learning, but this can be unreliable in VIS-IR data because imaging-physics differences make some spatially paired regions inherently less comparable, and aligning them with equal strength hinders representation learning and downstream transfer. In this paper, we revisit VIS-IR pre-training from a sampling perspective and propose Importance-Aware Sampling (IAS), which adjusts training emphasis based on patch reliability. Specifically, IAS (i) derives patch weights from infrared structural cues and uses them to reweight the contrastive objective; (ii) learns a soft importance mask with a lightweight sampler, optionally warm-started from the hand-crafted prior; and (iii) employs a patch curriculum learning strategy that gradually expands from high-reliability regions to harder patches. It is worth noting that IAS is plug-and-play and works with both patch-/correlation-level alignment (e.g., UNIV-style) and image-level contrastive baselines (e.g., ImageBind-style). Extensive experiments on multiple VIS-IR benchmarks demonstrate consistent improvements over strong baselines, including for IR semantic segmentation, IR object detection and VIS semantic segmentation and cross-modal retrieval task. Code will be released on https://github.com/KlayMa527/IAS.

Subject-to-video (S2V) generation has made substantial progress in preserving reference subjects across diverse categories, yet generic subject consistency remains insufficient for human-centric generation. A video may appear globally consistent while identity-critical human details still drift across frames, poses, and interactions. This issue becomes more severe in multi-person scenarios, where incorrect identity-role binding leads to subject confusion, attribute swapping, and excessive copying of reference-specific appearance cues. We propose Vera, a unified human-centric S2V framework for single- and multi-person generation. We first construct a million-pair identity-aligned human image-video dataset through person-level cross-clip retrieval, providing explicit identity correspondence and diverse references. Built on this dataset, Vera introduces two complementary designs. Identity-Focal Masked Supervision (IFMS) strengthens identity-aware learning with spatially focused supervision while reducing interference from irrelevant artifacts. Reference-Aware Layer-wise Attention (RALA) regulates how video tokens interact with reference identity cues in the DiT backbone, preserving stable identity anchors and enhancing layer-aware identity readout. Extensive experiments demonstrate that Vera improves human identity consistency, multi-person subject binding, and motion naturalness, while reducing identity confusion and excessive reference-image copying.

Housing-level urban physical examination is essential for identifying residential building problems and supporting targeted urban renewal. Existing automated inspection studies primarily rely on individual images and rarely examine whether surrounding urban functional context can provide supplementary information for building-level assessment. This study proposes a vision-POI fusion framework that combines multi-view visual inspection with POI-derived neighborhood context for residential building health assessment. The empirical dataset covers 92 old residential communities, 3,237 residential buildings, and 25,608 field-acquired inspection images in Qingdao, China, encompassing seven categories of housing-related issues. First, multiple object detection models are evaluated to extract issue locations, categories, and confidence scores from individual images. The image-level outputs are subsequently aggregated across multiple views to construct interpretable building-level representations. Second, POI features are extracted within 500m, 1,000m, and 1,500m neighborhood buffers to characterize surrounding functional environments. Pearson and Spearman correlation analyses, combined with false discovery rate correction, are used to identify candidate contextual features. Finally, visual and POI features are integrated using a cost-sensitive Random Forest classifier under community-isolated spatial cross-validation. The results show that multi-view aggregation provides the main performance improvement, increasing the building-level Macro-F1 from 60.84% under Direct Detection to 74.95%. Incorporating POI context further increases Macro-F1 to 76.79%, although the additional gain is modest and category-dependent. POI information therefore functions as a supplementary contextual prior rather than a substitute for direct visual evidence or a causal determinant of building condition.

Medical image encoders from different groups are increasingly treated as interchangeable, on the assumption that scale and clinical supervision concentrate their representations onto a shared structure. Whether this convergence is real, what produces it, and whether it is clinically usable are untested, and the similarity measures behind such claims are fragile. We present a controlled dissection across 18 image and 7 text encoders, all open-weight and run locally, spanning 7M to 27B parameters and five imaging modalities, including 650,982 chest radiographs from six datasets. To isolate cause, we train encoders that vary only the objective under fixed data, architecture, and scale, and reproduce the effect in a synthetic model. Convergence is modest but above a random floor, driven by the self-supervised objective, not clinical supervision: matched self-supervised encoders aligned most (40.4% on chest radiography), with label-supervised (21.1%) and image-text (3.3%) far lower, and did not grow with size (Spearman 0.302, p=0.223) or capability. It is within-modality, does not reach clinical language, and does not reproduce how radiologists judge case similarity. Yet a linear classifier transfers across encoders and to five held-out hospitals, retaining about 85% of within-encoder performance. Convergence in medical imaging is therefore set by the pretraining objective, not inherited from scale or clinical supervision. Interoperability is accordingly something to design for through that objective, and to validate where the shared geometry is weakest, across patient subgroups and against clinical judgment.

The rapid development of multimodal large language models (MLLMs) has introduced a flexible paradigm for remote sensing image scene understanding (RSISU), enabling natural-language interaction with remote sensing imagery. However, a systematic understanding of the capability boundaries, cross-task generalization, and task-specific limitations of existing remote sensing MLLMs (RS-MLLMs) is still lacking. This paper presents a systematic survey and diagnostic evaluation of MLLMs for RSISU. We review the technical evolution of RS-MLLMs, focusing on model design, multimodal learning, training data, and downstream capabilities. We further compare RS-MLLMs with general-purpose computer vision MLLMs (CV-MLLMs) across diverse RSISU tasks and benchmarks. RS-MLLMs remain competitive in domain-specific settings, particularly remote sensing visual grounding and high-resolution visual question answering. More notably, general-purpose CV-MLLMs can match or even outperform these specialized models on several RSISU tasks without remote sensing-specific fine-tuning. These findings demonstrate the strong transferability of general-purpose CV-MLLMs and show that current RS-MLLMs do not consistently outperform them across diverse RSISU tasks. Current MLLMs also face limitations in spatial and relational reasoning, fine-grained visual understanding, instruction diversity, and generalization across heterogeneous task formats. Based on these findings, we outline future directions toward reliable evaluation, multimodal and high-resolution reasoning, efficient deployment, and tool-augmented remote sensing agents. This survey provides a systematic reference for developing robust, generalizable, and practical MLLMs for RSISU.

Real-world fashion search involves interactive retrieval across multiple turns. However, existing multi-turn retrieval methods are built on a restrictive assumption that every interaction follows the same attribute-editing paradigm, leaving heterogeneous intent transitions unexplored. Moreover, existing approaches often rely on textification to bridge multimodal queries and visual retrieval, which may lose fine-grained visual cues. To address these gaps, we introduce DIM-Fashion, a benchmark of 26K multi-turn sessions constructed from 13 fashion retrieval datasets across 7 tasks, featuring diverse intent transitions and rollback behaviors. We further propose FashionAM, an MLLM-VLP framework that directly aligns multimodal conversational queries with a fashion-oriented gallery embedding space, avoiding intermediate textification. Extensive experiments demonstrate the effectiveness of FashionAM over existing approaches. The dataset and code will be made publicly available upon acceptance.

Diffusion policies achieve strong visuomotor control by iteratively denoising action chunks, but repeated denoising makes real-time deployment computationally demanding. Cache-based methods reduce inference cost by reusing intermediate activations, but existing training-free schedules typically allocate computation uniformly across blocks, ignoring heterogeneous redundancy across blocks and leading to a suboptimal performance-efficiency trade-off. To bridge this gap, we introduce Evolving Cache Schedules (EVO), a training-free acceleration framework that globally schedules cache refreshes via evolutionary search. EVO represents each candidate as a complete schedule over the block-timestep lattice. Thus, redundant transformer computations during iterative denoising can be skipped through cache reuse while preserving closed-loop rollout performance. To make the search practical, EVO introduces redundancy-aware initialization, which seeds the population with promising schedules, and target-conditioned early stopping, which verifies and terminates once a desired performance target is reached. The offline-optimized schedule can be directly plugged into pretrained diffusion policies without retraining. Extensive manipulation benchmarks show that EVO preserves near-full performance while substantially reducing computation, achieving up to 8.05x action-generation speedup and reducing FLOPs from 15.77G to as low as 1.96G. Source code is available at https://github.com/pillom/EVO.

RGB-D semantic segmentation has achieved remarkable progress, yet most models assume that RGB and depth are always available. In practice, failures or occlusions of surveillance sensors often remove one modality. Although RGB or depth alone can contain sufficient cues, models trained only on full-modality inputs fail to exploit the remaining modality once one is missing, causing severe degradation. We tackle this issue with a simple continued-training paradigm, \emph{Condition Dropout (ConD)}, which mitigates degradation while preserving full-modality accuracy. Starting from a pretrained RGB-D model, ConD adds a second stage that randomly simulates complete, RGB-missing, and depth-missing inputs, freezes the original encoders, and trains copied encoders with zero-initialized feature injection. Experiments on NYU-Depth V2 and SUN RGB-D show that ConD improves robustness under missing modalities and even yields slight gains when modalities are complete. Our code will be made publicly available upon acceptance.

We find that vision-language models are sensitive to a specific semantically irrelevant change: the order in which the image and question are presented. Across three models and three benchmarks, image first prompting consistently outperforms question-first prompting, revealing a repeatable modality order failure. We use this gap to design an order-consistent test-time training method. Our method substantially closes the modality-order gap across all evaluated settings. Surprisingly, it also yields consistent improvements in the stronger image-first branch over the baseline, hence bootstrapping both orderings toward mutual consistency. Activation patching localizes the ordering failure to a narrow mid-network region where representations diverge sharply between prompt orders. We find that the test-time training method repairs this misalignment across layers. Together, our results identify modality-order sensitivity as a circuit-level failure in VLMs and demonstrate that simple, asymmetric test-time adaptation can effectively mitigate it and even improve performance over the baseline.

Multimodal Large Language Models (MLLMs) have recently demonstrated strong performance across vision-language tasks. However, their high inference cost, arising from both the large number of input visual tokens and the heavy computation of the large language model (LLM), remains a key barrier to practical deployment. Recent work attempts to reduce the cost by adaptively optimizing individual dimensions, e.g., pruning redundant visual tokens or skipping LLM layers and heads. Nonetheless, prior approaches typically treat these dimensions independently and overlook a fundamental coupling: the available compute resources must be dynamically allocated across all dimensions based on the input content. To bridge the gap, we propose SmartVL, a unified adaptive inference framework that jointly controls vision token number and model compute capability in response to varying input contents and compute budgets. SmartVL introduces a vision-side token controller that dynamically selects informative visual tokens and an LLM-side compute controller that adaptively adjusts LLM computation. Importantly, these controllers are trained to coordinate with each other so that the overall inference cost satisfies a target budget. To allow this joint scheduling, we connect the controllers using a shared budget encoding and leverage a differentiable latency estimator for end-to-end training. This design enables SmartVL to learn cross-stage allocation strategies that adapt to both input complexity and runtime compute constraints. Experiments across multiple MLLM benchmarks demonstrate that, with joint scheduling, SmartVL consistently outperforms prior adaptive methods and achieves superior accuracy-efficiency Pareto frontiers. Project page: https://www.schaterji.io/publications/2026/jointtokencompute.

Recent autoregressive video diffusion methods are increasingly built upon Self Forcing, where the student is trained on histories produced by its own rollout rather than ground-truth video contexts. This reduces exposure bias, but the historical key-value cache is still used by future frames only as frozen rollout state. As a result, future losses cannot supervise how earlier generated latents should be written into more useful keys and values for later video-latent generation. We call this the historical context-gradient gap. We propose Self Gradient Forcing (SGF), a two-pass training strategy that restores this missing supervision signal without backpropagating through the full serial rollout. Pass 1 performs a no-gradient autoregressive rollout matching inference and, at a sampled denoising exit step, records both the self-generated context and the noisy latents fed to the model. Pass 2 performs parallel context-gradient reconstruction for the recorded exit step. The generated context is used as stop-gradient clean-latent input, while the model recomputes the context KV representations and future-to-context causal attention. Thus, SGF provides the missing memory-writing supervision within the native autoregressive training objective, using losses on future video latents to train the model to encode context into more effective causal memory. Across extensive long-horizon frame-wise and chunk-wise experiments under different initializations, SGF achieves stronger native long-video extrapolation than Self Forcing, especially in subject identity, background/layout consistency, and temporal stability. Remarkably, using only a 5-second training window, SGF can extrapolate to videos lasting several minutes. Code and models will be released to advance research on autoregressive video generation.

Optical Character Recognition (OCR) for Persian remains substantially less mature than for Latin-script languages despite Persian being spoken by more than 110 million people across multiple countries. This gap arises from two fundamental challenges: the intrinsic complexity of the Perso-Arabic writing system and the limited availability of large-scale, high-quality annotated datasets. Persian script exhibits obligatory cursive connectivity, context-dependent glyph shaping, extensive ligatures, diacritic placement, and stylistic variation across writing forms such as Naskh and Nastaliq, all of which significantly complicate text recognition. At the same time, the high cost and labor-intensive nature of manual annotation have created a persistent data bottleneck, limiting the development of robust OCR systems and slowing progress in Persian document digitization.In this paper, we introduce Persian Pixel, a comprehensive synthetic OCR dataset specifically designed to address these challenges. Comprising over 343,000 high-fidelity image text pairs, the dataset spans sentence, paragraph, and full-page document layouts generated from a carefully curated seven-million-word Persian corpus using the SynthOCR-Gen rendering framework. The generation pipeline faithfully models the typographic characteristics of Persian script, including contextual character joining, positional glyph variants, diacritic placement, and multiple representative Persian typefaces. To bridge the synthetic-to-real domain gap, the rendered images are further enriched with more than twenty-five stochastic degradation models that emulate realistic document acquisition artifacts, including ink bleed, paper aging, blur, illumination variation, scanner imperfections, compression artifacts, and multiple noise processes.By overcoming the long-standing scarcity of annotated Persian OCR data, Persian Pixel provides a scalable and openly available resource for training and fine-tuning modern OCR architectures, including transformer-based models such as TrOCR and Donut. The dataset establishes a strong foundation for research in Persian document analysis, historical manuscript digitization, and end-to-end document understanding, while demonstrating that programmatic synthetic data generation offers a practical, cost-effective, and scalable alternative to manual annotation for advancing OCR in low-resource and typographically complex scripts.

Video captioning requires fine-grained spatio-temporal understanding of videos, including spatial perception of where objects are located and temporal perception of when events occur. Existing MLLMs usually generate captions directly from video inputs without exposing the perceptual evidence behind descriptions. As a result, mistakes in spatiotemporal perception are only observed in the final caption, making it difficult to identify the underlying perceptual errors directly. To address these issues, we present PercepCap, a perception-aware video captioning framework that makes perceptual evidence explicit before producing the final caption. Specifically, PercepCap follows a perceive-describe generation chain, where the model first produces a spatiotemporal perception trace comprising object trajectories and temporal events, and then generates the final caption conditioned on the perceived evidence. To support this, we design a two-stage training strategy. Perceive-then-Describe Supervised Fine-tuning adapts the model from caption-only generation to the proposed perceive-describe chain, while Perception-Grounded Reinforcement Learning optimizes perception trace and caption quality with joint rewards over perception chain and the final caption. To support our two-stage training, we introduce Caption-Anchored Perception Data Construction. This pipeline builds the SFT and RL training data by first generating a caption-only description, extracting the objects and events it mentions, and grounding them back in the video with boxes and timestamps. This yields caption-aligned perception data that provides solid training ground truth, ensuring that the explicit perception trace and final caption refer to the same objects and events. Across direct caption and caption-to-QA evaluation, PercepCap consistently improves upon the Qwen3-VL baseline and demonstrates leading caption quality.

3D Gaussian Splatting (3DGS) achieves high-quality novel-view synthesis by optimizing freely placed primitives in 3D and adaptively densifying them in under-reconstructed regions. However, this scene-adaptive capacity allocation is largely lost in existing feed-forward 3DGS methods, which commonly regress Gaussians at input pixels and lift them along camera rays. Such pixel-aligned formulations make the number and placement of primitives depend on image resolution and input viewpoints rather than scene complexity, resulting in dense and often redundant Gaussian sets. We present ATSplat, a feed-forward 3DGS framework that restores the adaptive allocation capability of 3DGS optimization through Adaptive 3D Tokens. ATSplat first lifts coarse patch-level depth and camera cues into sparse 3D anchor tokens, forming a compact scaffold of the scene. Each token is then regressed into local Gaussians with learnable 3D offsets, decoupling primitive placement from input image grids. An Adaptive Token Expansion module predicts a token-level uncertainty score, supervised by rendering error maps, and selectively expands high-uncertainty tokens through learnable expansion layers. This sparse-to-adaptive formulation enables ATSplat to concentrate primitives in challenging regions while maintaining a compact representation. Experiments on two representative datasets, RealEstate10K and DL3DV, show that ATSplat achieves state-of-the-art rendering quality while reducing the number of Gaussians by more than $5.7\times$ compared with dense feed-forward 3DGS methods. From 12 input images at $512 \times 960$ resolution, ATSplat completes reconstruction in less than a second using a single commercial GPU, and renders high-quality novel views at 1136 FPS ($512 \times 960$) with only 311K Gaussians.

Large language models (LLMs) are increasingly deployed as always-on online services, making efficient LLM serving a critical systems challenge. Achieving low latency and high throughput under volatile demand requires deep understanding of real-world serving workloads, yet existing studies often rely on proxy traces or coarse-grained characterizations that fail to capture the heterogeneity of modern multi-model LLM platforms. We present FineServe, an in-the-wild, multi-model LLM serving workload dataset collected from a global commercial marketplace, enabling fine-grained characterization of real-world serving dynamics across heterogeneous models and tasks. Leveraging FineServe, we conduct a comprehensive analysis of arrival dynamics and token behavior, revealing fundamentally different fluctuation regimes across model architectures, scales and task intents. Building on these insights, we develop the FineServe workload generator, which composes fine-grained model-aware workloads into configurable mixtures tailored for benchmarking multi-model serving platforms. By exposing these fine-grained workload dynamics, FineServe provides a realistic foundation for evaluating routing, scheduling, and capacity-planning strategies in LLM serving systems. FineServe is available at https://github.com/hihiztc1/FineServe.

Financial institutions face significant challenges in detecting sophisticated money laundering patterns, such as smurfing and layering, due to extreme data imbalance (0.13% fraud rate) and evolving adversarial evasion tactics. This paper proposes FraudShield AI, a hybrid framework that integrates Long Short-Term Memory (LSTM) networks with hand-crafted Graph Topological Features to capture both temporal sequences and structural relational context. By engineering network-centric features including PageRank Centrality, In-Degree dynamics, and a custom Flow Ratio, the system shifts the detection paradigm from isolated transaction analysis to network-level forensics. A Focal Loss objective is used to address class imbalance, and a dynamic thresholding mechanism is introduced to improve resilience against low-value smurfing attacks. Experimental evaluation on the PaySim dataset shows that the proposed hybrid model substantially outperforms Logistic Regression and XGBoost baselines in Precision, Recall, and F1-Score, particularly on hard-to-detect micro-transaction fraud patterns. An ablation study confirms the complementary contribution of both the temporal and topological components.

LLM-based multi-agent systems (LLM-MAS) are increasingly deployed in safety-critical applications, where adversaries inject malicious instructions through inter-agent communication to propagate harmful behaviors. Unlike static threats, these attacks are doubly dynamic: adversaries refine injection strategies against deployed defenses while normal-agent behavior drifts with system expansion. Existing defenses treat deployment as a closed-world problem and degrade rapidly once either distribution shifts beyond training coverage. We propose OpenEvoShield, a co-evolutionary continual defense framework for LLM-MAS. An asymmetric rate controller (M1) decouples fast attack-side and slow normal-side learning rates from dual drift signals. A normal-boundary updater (M2) maintains a dynamic behavioral boundary at the slow rate, while an EWC-regularized policy ensemble (M3) fast-adapts without catastrophic forgetting. An energy-based multi-granularity detector (M4) fuses node-, subgraph-, and graph-level evidence to classify novel attacks as out-of-distribution. Experiments over 100 deployment rounds across five benchmarks and four MAS topologies show that OpenEvoShield outperforms static and continual baselines, detecting most previously unseen attacks while keeping false positive rates low.

Confidential computing is becoming a practical deployment requirement for AI inference workloads that process sensitive inputs or protect proprietary model assets. However, the performance cost of enabling confidential execution for GPU-accelerated large language model serving remains workload dependent and operationally important. This paper presents a benchmark study comparing standard non-confidential execution with confidential computing mode on a single NVIDIA H100 80GB GPU hosted in an Intel TDX confidential instance. The evaluation uses two representative language models, Mistral-7B v0.1 and Qwen3-30B-A3B, and measures time to first token, end-to-end request latency, per-request token generation throughput, global token throughput, and closed-loop request throughput under increasing concurrency. In fixed request-rate experiments, confidential mode increases average TTFT by 21.8% for Mistral-7B and 27.8% for Qwen3-30B-A3B, while global token throughput drops by 17.7% and 21.1%, respectively. In closed-loop concurrency experiments, throughput gaps remain in the 11.5-20.2% range, but the larger model reaches its saturation knee earlier under confidential mode. The results suggest that confidential GPU inference can retain usable throughput under load, but capacity planning must account for both the steady throughput penalty and the earlier saturation behavior observed for larger models.

Spreadsheet applications are used by hundreds of millions worldwide, yet writing formulas remains a significant barrier. Existing approaches rely on static supervised data, which quickly saturates on limited annotations. In this paper, we introduce FORMULASPIN, a self-play framework that breaks the ceiling of supervised fine-tuning by enabling iterative self-improvement without any additional data. Vanilla SPIN fails on this task: it uniformly penalizes every non-matching output, so execution-equivalent alternatives are punished as negatives in one example while serving as ground truth in another, producing contradictory gradients. Our framework resolves this by exploiting formula generation's unique advantage: binary executability provides implicit supervision that separates semantic errors from valid stylistic variants. We frame training as a two-player game in which the main player learns to prefer ground-truth formulas over those from its previous version, while execution feedback sorts outputs into distinct granularities-enabling an adaptive curriculum that shifts from semantic correctness to stylistic refinement. To further increase accuracy, we incorporate ExecVote, a semantic-level voting mechanism that naturally handles multiple valid formulations. Experiments on multiple benchmarks demonstrate that FORMULASPIN achieves state-of-the-art performance, with 74.9% exact match and 87.1% execution accuracy on NL2FORMULA, matching models trained with additional preference annotations while outperforming both traditional SFT and frontier proprietary models. These findings underscore self-play's potential to tackle scarce data tasks and open the door to extending it beyond executable domains.

LLMs are increasingly used with external knowledge sources like the internet. Do they weigh information appropriately -- updating more for reliable sources (source discernment) and more when claims bring priors closer to the truth (truth discernment)? We formalize this as information discernment and introduce Learn2Discern (L2D), an experimental framework and benchmark grounded in three normative axioms with interpretable metrics. To establish external validity, a pre-registered, quota-matched user study (n=299) confirms that real LLM users endorse all three axioms and report that violations reduce their trust and usage intent. Across 13 models and nearly 670K trials, we find consistent failures across both dimensions: models perform near chance on source and truth discernment, rely on source popularity twice as much as source reliability, and update roughly equally whether a claim improves or worsens their position relative to the ground truth. Models integrate external knowledge most effectively on datasets where their priors are already the most accurate. Newer and larger models improve truth discernment but not source discernment, a blind spot that model complexity does not address. We identify simple inference-time interventions that improve both forms of discernment. We release our dataset and survey as a testbed for a core alignment property that scales in importance as LLMs replace traditional search.

Tool-using LLM agents increasingly execute high-impact actions, making runtime safety monitoring essential. We present NEXUS (Neural EXecution Utility and Safety), a structured-plan safety monitor that applies a formal intervention policy to select among four actions: allow, block, request confirmation, or request revision. NEXUS combines deterministic safety rules, argument-level inspection, and a calibrated logistic-regression risk score for graded escalation. On a 128-instance synthetic benchmark, NEXUS achieves an F1 score of 0.949 and a 4-class intervention accuracy of 0.6406, outperforming rule-only intervention selection by 27.3 percentage points. It also improves over rule-only on R-Judge (F1 = 0.861 vs. 0.849), matches rule-only on AgentHarm due to threat-model limits, and achieves 0% ASR at 99% control allow on IPI. On the rule-blind NEXUS-Stress benchmark, NEXUS reaches an F1 score of 0.881, highlighting the difficulty of fine-grained intervention routing. With 0.205 ms median latency, NEXUS adds under 0.1% overhead to typical agent loops. Code, benchmarks, and the calibrated risk scorer are publicly released.

Recent advances in recommender systems (RS) have shown substantial performance gains through generative modelling. In practice, recommendation often involves constructing slates -- ordered lists of items -- that must satisfy multiple objectives beyond relevance, such as constraints defined over item attributes or fairness constraints. Existing multiobjective approaches either rely on post-processing techniques designed for non-generative settings, or incorporate auxiliary objectives directly into model training. The former does not explicitly account for the sequential nature of generative RS, while the latter is often impractical in large-scale systems. We propose a lightweight, inference-time decoding layer that augments autoregressive generative RS to support multiobjective slate generation without modifying or retraining the underlying model. We formulate decoding as an online constrained optimisation problem, where items are selected sequentially, and trade-offs between relevance and auxiliary objectives are adjusted dynamically based on the remaining constraint slack, i.e., how much of each objective remains to be satisfied. This is implemented via a stochastic primal-dual approximation scheme that balances relevance and auxiliary objectives during generation. We provide theoretical guarantees on constraint violation and regret, and evaluate the proposed approach through extensive offline experiments and a large-scale online A/B experiment in a real-world recommender system. Our results show consistent improvements in multiobjective trade-offs, including a +1.8\% gain in the auxiliary objectives achieved at zero cost to user satisfaction.

Recent advances in long chain-of-thought reasoning models such as DeepSeek-R1 have led to increasingly longer inference context lengths under the test-time scaling paradigm. However, the O(n^2) computational complexity of standard self-attention causes inference costs to grow sharply with long sequences, limiting the deployment of long-CoT reasoning in production settings. To address this, we propose LISA (Linear-Indexed Sparse Attention), a plug-and-play attention replacement module that requires no pretraining from scratch. LISA integrates two lightweight components in parallel within the original model: (1) a Linear Attention module that provides long-range memory with O(n) time complexity; (2) a Lightning Indexer that selects the top-M important tokens from the full context to feed into a Sparse Self-Attention. The two branches are fused via a gating mechanism, reducing inference complexity from O(n^2) to O(nM) (M << n) for generating n tokens. We design a two-stage training pipeline: Stage 1 initializes the model by integrating the linear attention to capture long-range dependencies, complemented by a sliding-window attention mechanism that is optimized via knowledge distillation to approximate the full self-attention distribution of a frozen teacher model. In Stage 2, we further introduce the Indexer to replace the static sliding-window mechanism, enabling dynamic token selection from broader contexts. The Indexer is trained using a novel per-head KL divergence loss, which aligns its selection behavior with the attention patterns of the teacher model. Experiments on DeepSeek-distilled-Qwen models demonstrate that LISA achieves a 50% inference speedup under 16K-token context, while improving average performance by 5.6% on reasoning benchmarks including AIME and MATH-500.

Long-term memory is essential for LLM agents that interact across sessions, yet current memory benchmarks primarily evaluate single-hop recall, leaving multi-hop association largely unmeasured. We make three contributions. First, we introduce MemHop, a multi-hop memory benchmark of 1,000 questions at hop depths 1-5 across 10 social-network scenarios, with per-hop evidence annotations. Second, we present Profile-Graph Memory (ProGraph), a two-layer memory architecture combining (i) profile expansion -- substring-matched traversal of entity names that naturally appear in LLM-written profile narratives, a minimal alternative to explicit knowledge-graph construction -- and (ii) compression residuals -- exact dates, quantities, and named items co-extracted with each profile update at zero extra API cost. Third, a full-grid ablation shows cross-benchmark mechanism specialization: profile expansion drives multi-hop reasoning (-22.6pp on MemHop when removed) while compression residuals drive precision recall (-8.6pp on LoCoMo when not co-extracted), with cross-effects under 3pp within a single architecture. ProGraph averages 80.1% on MemHop (matching the FullContext reference) and 78.4% on LoCoMo (exceeding FullContext by 11.3pp), outperforming Mem0, A-Mem, HippoRAG, and RAG on both. We release MemHop, ProGraph, and baseline implementations.

Large language models (LLMs) often answer queries by mapping individual observations to more general rule-like structures. However, it remains unclear how these structures are stored, selected, and revised. To study this process, we propose thelifted representation hypothesis: LLMs update memory through shared latent structures rather than isolated instance-level facts. This view frames lifting as an efficient use of symmetry across instances, and shattering as the refinement of coarse lifted structures into more specific subtypes. We evaluate LLMs' lifting and shattering through controlled exception-learning experiments across in-context learning, LoRA, and full fine-tuning. We find that LLMs are vulnerable to shattering failures when data are governed by nested rules and exceptions, while lifting often occurs prematurely. These results highlight the need to study the relation between data and rule structures in LLMs.

Graph RAG mitigates hallucinations and stale knowledge in LLMs, particularly for multi-hop question answering. However, existing approaches remain highly fragmented and incompatible. The structural heterogeneity of graph formats across different frameworks and the lack of granular visualization tools make it exceedingly difficult to evaluate and compare retrieval behaviors. To bridge this gap, we propose GraphContainer, a novel platform designed to unify and visualize diverse graph RAG workflows. GraphContainer features two key components: (1) a Unified Graph Representation (UGR) layer that seamlessly standardizes multi-format graphs, and (2) a Graph Recorder that tracks and visually renders the step-by-step retrieval process. Through an interactive web interface, we demonstrate GraphContainer's ability to import heterogeneous graphs and perform live, traceable visual debugging of graph RAG methods. Ultimately, we show how GraphContainer enables controlled comparisons of various graph formats and retrieval strategies, lowering the barrier for researchers and practitioners to design optimal graph RAG pipelines. A demonstration video is available at https://youtu.be/O02eNJLwkU0.

Rotary Position Embedding (RoPE) is widely adopted in Transformers to encode positional information, yet standard implementations enforce a uniform frequency schedule and scaling across all attention heads. Using simplified retrieval tasks and length generalization scenarios, we show -- both empirically and theoretically -- that heads with different functional roles require distinct frequency ranges and attention scaling factors to operate effectively. Ignoring this structure leads to suboptimal utilization of embedding dimensions and degraded performance, particularly under long-context settings. To address these limitations, we propose AdaRoPE, which equips each attention head with learnable rotation frequencies and attention scaling factors. Pretrained LLMs with AdaRoPE consistently outperform existing RoPE variants, including partial RoPE and NoPE baselines. For context extension, we further show that uniform frequency and attention scaling, used in methods such as YaRN, are suboptimal. By applying head-specific scaling, AdaRoPE enables better context extension while better preserving short-context performance in both the extrapolation setting and the long-context continued pretraining setting. These results highlight the importance of optimizing rotary position embedding at the level of individual attention heads.

Activation steering offers a lightweight alternative to fine-tuning for behavioral control of large language models, but SAE-based steering methods often rely on learned steering objectives or single-criterion feature selection. We introduce a transparent SAE-feature steering pipeline that first applies a six-condition reliability filter, then ranks sparse features through an unweighted Borda consensus over three complementary statistics: $F$-test, KSG mutual information, and Cohen's $d$. The resulting steering direction is constructed as a Cohen's-$d$-weighted combination of SAE decoder rows, providing an optimization-free direction motivated by Fisher-LDA under approximate SAE-feature decorrelation. Across three Gemma-family models, four behavioral domains, and 356 layer-strength configurations, the method produces measurable domain-specific shifts while revealing a substantial gap between raw attribute movement and quality-preserving generation. In the strongest configuration, logical-correctness steering reaches a primary-score delta of $+1.16$ in Gemma~2 9B; however, our broader finding is that usable steering is highly localized by model, domain, layer, and strength. These results argue that activation-steering evaluations should report quality-conditioned success alongside raw behavioral shift. Our code and data are available at https://github.com/Oshayer-Siddique/LLM-Steering-Using-SAE.

When Large Language Models (LLMs) are used for semantic data extraction from unstructured text, producing candidate relational facts from natural language, they may remain unreliable for tasks requiring complex combinatorial reasoning and global consistency. This paper proposes a logic-guided data extraction framework combining LLM-based extraction with Answer Set Programming (ASP). The LLM produces candidate facts, whereas ASP performs validation, inference, consistency checking, and control. Unlike existing pipelines that query the LLM independently for all target predicates, the proposed approach uses ASP reasoning to identify which predicates are logically admissible at each stage and to guide extraction queries. By interleaving LLM calls with ASP derivation, the framework infers logically implied facts without further extraction and detects inconsistencies early. We formalize the pipeline and prove that, under mild assumptions, it is equivalent to the baseline approach with respect to the final extracted facts, while requiring fewer LLM calls. We also introduce a caching mechanism for logic-based control queries, exploiting monotonicity of conjunctive queries over incrementally constructed fact sets to reduce solver invocations. Experiments on ASP-derived benchmarks show that the framework reduces LLM calls and improves extraction quality by mitigating spurious outputs, demonstrating the value of non-monotonic logic programming for controlled semantic extraction.

Safety as Polytope (SaP) learns linear half-space constraints in LLM hidden space but requires per-category tuning of the constraint count K. We show that sparse autoencoder (SAE) feature extraction resolves this: K=2 becomes optimal for 12/14 categories on Qwen3.5-9B, achieving 96-99% accuracy per category on our BeaverTails classification benchmark, largely eliminating the need for exhaustive sweeps (K=4-25 with random initialization). This convergence to two planes is consistent with the Linear Representation Hypothesis, providing suggestive evidence that safety boundaries in this setting admit a low-dimensional linear description in the SAE feature space. Building on this geometric perspective, we introduce a cone constraint whose learnable aperture adapts to each category's cluster concentration, stabilized by a three-phase training

Calibration is the primary criterion for evaluating LLM confidence, but it is insufficient: it admits trivially incoherent estimators, depends on the evaluation distribution, and does not test the extent to which the estimation can be interpreted as a consistent, underlying probability function. What we actually need is for LLM confidence estimates to satisfy the conditions required of coherent probabilistic beliefs. We formalize these conditions along three axes (structural coherence, faithfulness, and usefulness) and operationalize them as the C1 metrics. Widely used estimators systematically violate these conditions despite appearing well-calibrated: models assign lower confidence to logically easier questions 31\% of the time, and common interventions reducing RMSCE leave structural violations unchanged, suggesting that calibration is orthogonal to probabilistic validity. RLHF and chain-of-thought improve usefulness metrics without restoring coherence. Our results show current LLM confidence estimates cannot be interpreted as coherent probabilities; our framework provides the tools to measure and close this gap.

Long-prompt inference remains expensive because prefill attention scales quadratically with sequence length. We propose Spectral-LSH, a training-free prompt compression method that operates before the prompt enters the language model. Spectral-LSH approximates the dominant components of an implicit attention-kernel operator using a Krylov subspace method together with random features, avoiding explicit $O(N^2)$ attention-kernel materialization. It then applies SimHash in the resulting attention eigenspace to group similar tokens and aggregate them into macro-tokens with causal positional assignments. We evaluate Mistral-7B-Instruct-v0.3, Qwen2.5-7B-Instruct, and Qwen2.5-14B-Instruct on C4. Our experiments reveal a compression-ratio phase transition. Below $\rho = 4 \times$, local token redundancy is low enough that lightweight chunking typically provides the best latency--quality trade-off. Above $\rho = 8 \times$, the spectral path preserves quality that chunking loses. At $\rho = 16 \times$, Qwen2.5-7B (adaptive) reduces the PPL ratio from 353.409 to 196.963, while Qwen2.5-14B (adaptive) reduces it from 9.533 to 3.427. On a small long-context structured stress test containing JSON-like, code-like, and table-like inputs, local LSH also improves every metric over chunking at $8 \times$. The adaptive backend captures both regimes by using the chunk path at low compression and spectral clustering at high compression, although chunking remains the fastest backend in total latency.

Hidden-state probes often recover latent labels in imperfect-information sequence models, but this alone does not establish that a model maintains a posterior belief distribution over hidden states. This paper studies this ambiguity in a no-range Limit Hold'em autoregressive model trained only on action and value targets, not on an opponent's hand or range. Opponent-range probes are positive after action/value controls in two of three seeds, and the behavior head predicts held-out actions about five percentage points above a baseline using only observable public history. However, visible public betting composition explains more opponent-range signal than residual hidden states, suggesting that most recoverable information comes from betting summaries. Action/value+composition baselines reach 16.5-16.7% top-10 accuracy while composition-residual hidden probes fall to 11.4-12.2%, and matched-composition comparisons are negative in every seed. We call this evidence pattern composition-bounded predictive support: hidden states remain behavior-predictive and opponent-range correlated, but most recoverable range information is explained by visible betting composition rather than residual hidden-state structure. This is a case-study claim about opponent-range representational evidence, not exact Bayesian posterior tracking or a causal belief mechanism. Synthetic control and oracle validations show that the same diagnostics accept posterior-sensitive states and reject raw composition states under matched controls. Thus positive belief probes should be interpreted through targeted alternatives before being treated as evidence of belief tracking.

Large language model (LLM)-based Socratic tutors increasingly guide students through multi-turn questioning, but they can suffer from scaffolding collapse: under sustained student pressure, a tutor gradually abandons guided inquiry and reveals solutions directly. Prior defenses primarily constrain observable responses through prompting, preference optimization, or filtering, leaving the internal representation drift that precedes trajectory-level collapse largely unaddressed. We propose Scaffold-Preserving Representation Alignment, a two-stage framework that first warms up a Socratic tutor with supervised fine-tuning, then combines trajectory-weighted direct preference optimization with a margin-preserving representation loss anchored to frozen reference states. Our method is designed to maintain separation between scaffold-preserving and collapse-inducing hidden states across dialogue turns. We evaluate our method across five STEM disciplines and five red-teaming attack strategies. On Qwen3-8B, our method lowers Collapse Rate to 32%, delays average collapse onset beyond nine turns, and keeps over-refusal low, suggesting that representation-level alignment can improve the robustness of long-horizon Socratic tutoring under our red-teaming protocol.

Recent formal reasoning systems have reached IMO-level performance, yet they leave a fragmented landscape: algebra and number theory are handled in Lean, while geometry still relies on domain-specific languages with limited formal guarantees. This split increases the trusted computing base and hinders unified model development. Existing geometry-in-Lean efforts (LeanEuclid, LeanGeo) introduce custom axiom systems incompatible with standard Mathlib, and their small scale ($<$ 1,100 problems) limits large-scale training. Native Mathlib autoformalization of geometry, however, poses distinct challenges: implicit diagrammatic assumptions (e.g., topological configuration and non-degeneracy) must be made explicit rather than deferred to external solvers, and models must adapt to Mathlib's small, rapidly evolving geometry infrastructure. We present Euclean, a four-stage framework - constraint explication, configuration anchoring, formalization mapping, and iterative repair - for automatically formalizing geometry in native Mathlib. We construct OMNI-Geometry (768 competition problems) and Numina-Geometry (177,597 problems), the largest geometry formalization dataset in Lean. Human evaluation shows 48.89% TOP1 and 73.33% TOP5 accuracy. Training Goedel v2 on our formalizations improves proof success from 13.6% to 15.1%, validating dataset quality for unified neural theorem proving. Code and datasets: https://github.com/tlb-22/Euclean.

Document-based LLM systems often flatten a PDF before guardrails inspect it. That step can discard evidence that an instruction was never visible to the user. We introduce CrackedPDFs, a controlled benchmark for hidden prompt injection in PDFs. The benchmark contains 29,322 generated PDFs from 4,983 base docu ments. It includes 9,774 injected files and 19,548 benign or matched-confounder files. We evaluate PromptGuard and a rule baseline. We also evaluate structural only learned models and a sanitized hybrid detector. The evaluation uses held-out provenance splits and paired benign-confounder controls. It also uses label-shuffle checks and shortcut audits. On a 2,919-document held-out test set, the hybrid de tector reaches 0.960 F1. ROC-AUC is 0.998 and PR-AUC is 0.997. It also ranks injected files above matched benign confounders in 95.9% of 973 pairs. Prompt Guard has low recall when given extracted text only. Structural-only learned mod els are weak under paired controls. A text-only TF-IDF model reaches perfect held-out scores but fails shortcut audits. These results show that document-aware hybrid detection is useful under controlled paired evaluation. They do not show broad real-world robustness or reliable cross-family generalization.

Multi-entity compositional questions pose significant challenges to existing retrieval-augmented language models. Conventional methods fall into a dilemma: standard RAG lacks dynamic reasoning, traditional Graph-RAG is limited by structural sparsity, and LLM-constructed Graph-RAG incurs prohibitive costs. We propose \textbf{\fwa}, a unified framework that embeds unstructured text into structured knowledge graphs, creating a heterogeneous network for flexible evidence retrieval. Reasoning is formulated as adaptive structure induction, learned via a robust two-stage process: (1) imitation learning distills heuristic expert signals, and (2) reinforcement learning refines the policy using LLM-driven preference rewards. Experiments demonstrate that {\fwa} effectively merges textual richness with structural knowledge, outperforming SOTA baselines in answer accuracy and reasoning fidelity while maintaining extremely low token costs and near real-time inference((code available at https://github.com/wjywjy123/HyGRL) .

Formal theorem proving has emerged as a frontier challenge for machine learning, yet the ecosystem is fragmented: proofs remain siloed across incompatible systems, limiting both training data for learning-based provers and the portability of verified results. We present ITPEval, the first benchmark for evaluating automated formal proof translation across four major ITPs (Lean 4, Rocq, Isabelle, and HOL Light), spanning two distinct logical foundations. Our benchmark comprises 1,560 source files and 6,848 theorems organized into a controlled tier of axiomatized files that isolates foundational translation difficulty, and an ecosystem tier drawn from real libraries that exposes API and proof-style mismatches. We release itpeval, a unified multi-ITP verification infrastructure with state-isolated warm backends that preserve per-artifact native checking semantics. We evaluate both statement and proof translation across five frontier and open-weight LLMs on 12 directed translation pairs: statement translation peaks at 29.1% pass@1 and proof translation at 10.5%; controlled theorems reach 29.7% proof pass@1 versus 5.2% for ecosystem-level translations, confirming that library mismatch is the dominant bottleneck. In addition to pass@k evaluation, a deterministic Lean 4 BEq check establishes equivalence for 54.0% of verified source-to-Lean 4 miniF2F statement translations, showing that native type-checking alone can substantially overestimate semantic fidelity; in an autoformalization/auto-informalization round-trip study, Rocq and HOL Light are easier formalization targets than Lean 4 and Isabelle, while multi-ITP context improves pooled Lean 4 success from 4.8% to 10.6%. Our benchmark, verification infrastructure, and evaluation pipelines are publicly released.

Recent advances in large language models have accelerated deployment of agentic systems in operational finance. Existing benchmarks emphasize measuring general capabilities, instruction following, or safety, but few directly address the operational finance workflows that agentic systems are now being deployed to automate. Finance professionals require agents to not only provide factually sound and properly grounded information, but also ensure that this information is verifiable and consistently adheres to rules and constraints of the operational finance domain. We introduce FORCE-Bench, which contains 251 expert-annotated queries and evaluates responses using a rubric-based framework calibrated to the requirements of the operational finance domain, across eight dimensions: accuracy, citations, clarity, depth, groundedness, recency, relevance, and structure. FORCE-Bench assesses agentic systems on three task types: financial obligation research (querying ERP systems for accounts receivable and payable data), financial entity performance research (answering time-bound questions from public filings and market data), and business brief generation (synthesising multi-source company intelligence reports). To reflect real deployment conditions, we evaluate our purpose-built agent, as well as the general-purpose agentic systems, under common tool access and latency-bounded settings. Results show that general-purpose agentic systems do not consistently meet finance-domain quality requirements under operational constraints, while the purpose-built Finance Agent for Microsoft 365 Copilot is more reliable across dimensions. We release the dataset, rubrics, harness, and analysis code as open-source to support reproducible comparison and adaptation to other enterprise finance environments.

The transition from stateless generative models in artificial intelligence to stateful, autonomous agents represents an architectural evolution that, while providing the capabilities of long-term planning and the automation of enterprise workflows, also represents the introduction of a new form of security threat, the Chronos Vulnerability. The Chronos Vulnerability represents the threat of memory-based attacks, including the Memory Injection Attack (MINJA) and the sleeper agent, in which the internal belief system of the autonomous agent is compromised, effectively decoupling the attack vector from the final catastrophic event. This study formalizes the threat model for persistence-based attacks and the threat of Dynamics Blindness in the context of the World of Workflows benchmark, demonstrating that traditional endpoint content filters are insufficient for the current stateful architecture. Consequently, this study synthesizes a defense-in-depth landscape, categorizing emerging frameworks such as diagnostic trajectory guardrails (AgentDoG), formal temporal verification (Agent-C), immunological memory consensus (A-MemGuard), and hardware-anchored trust via GPU-based Trusted Execution Environments (TEEs) and Zero-Trust memory architectures.

Sophisticated Inference is a variant of active inference often associated with recursive belief modeling and tree search. We argue that its central computational role is simpler: within a planning horizon, it makes active inference closed-loop by allowing future actions to depend on future states and observations. This closed-loop structure can be represented in the epistemic-prior variational free energy framework. Epistemic priors supply the active-inference objective, while a joint posterior over future states and actions supplies the state-contingent control structure. We evaluate this decomposition in the Reactivity Maze, a stochastic benchmark designed to separate epistemic incentive from inner-horizon closed-loop control. The comparison includes three variational objectives with the same state-action posterior family, an action-state factorized active inference objective, Sophisticated Inference, and standard Expected Free Energy planning. The results show that neither ingredient is sufficient on its own. Methods without an epistemic component do not seek information, while methods that prevent future actions from depending on future states cannot turn information into reliable goal-reaching. By contrast, both Sophisticated Inference and full-joint epistemic-prior active inference solve the environment by combining epistemic drive with closed-loop inference. These results show that the advantage associated with Sophisticated Inference need not be specific to tree search itself. It arises from the closed-loop form of active inference, and this form can be represented in epistemic-prior variational inference when the posterior keeps future actions dependent on future states.

Self-improving AI systems typically treat the agent as the object that improves, by optimizing prompts, workflows, harnesses, or even the agent's own code. This agent-centric view can make improvements expensive to maintain and difficult to transfer, because gains become tied to a particular agent design, task distribution, or adaptation run. We study a complementary paradigm: knowledge-centric self-improvement, in which agents remain generic and disposable while the persistent object is a curated knowledge base that agents can leverage for future tasks. We conduct controlled case studies to operationalize this idea via a simple protocol. Agents attempt one task, then contribute evidence-grounded insights to a shared knowledge base via task-level and cross-task forums, followed by knowledge distillation. Because self-improvement is contained in the knowledge rather than the agent, improvement can be more inspectable, transferable, and portable. Across abstract reasoning, coding, and terminal benchmarks, this protocol improves solve rates while reducing dollar cost relative to agent-centric baselines. The resulting distilled knowledge also transfers to held-out tasks and across LLM families, indicating that the improvement is not merely an LLM- or run-specific behavior. These results support a new view of self-improving agentic systems: progress can be driven primarily by the curated persistent knowledge. Code is available at https://github.com/recursive-knowledge/KSI.

Civil aviation is safety critical and its operations, from flight decks and towers to ramps and maintenance, generate massive, heterogeneous data at the network edge. Yet cloud centric deployment of large Artificial Intelligence (AI) models often produces high task latency, lacks offline capability in communication denied environments, and requires centralizing sensitive data, raising privacy and sovereignty risks. Edge AI moves perception, prediction, and decision logic closer to the data producers via compression, collaborative inference, and split learning, thereby reducing latency, bandwidth, and exposure while enabling graceful operation during disconnections. This paper provides a panoramic view and a common understanding of edge intelligence tailored to civil aviation. We firstly articulate the operational motivations for edge AI, and then review recent techniques for edge inference and edge learning. We then introduce the organizational computing paradigms and the respective configurations in civil aviation environments; finally, we describe the emerging applications and the future research trends of edge intelligence in civil aviation. We argue that a refined edge solution can complement cloud foundations to deliver low latency, privacy preserving, and resilient AI services across the civil aviation lifecycle.

A rich and recognizable component library is the cornerstone of printed circuit board (PCB) design and generation. Traditionally, engineers manually create symbols and footprints and design PCB schematics, which is time-consuming and error-prone. Leveraging multimodal large language models (MLLMs), we develop SFgen, an agentic recognition and generation flow of symbol and footprint for electronic components. SFgen achieves 86% accuracy for symbol generation and 80% accuracy for footprint generation. We use the SFgen method to create SFnet, a database of symbols and footprints. It now has 1000 components and is expanding constantly, which lays the foundation for automatic generation of PCB designs.

Multimodal agentic search systems increasingly rely on external tools to answer knowledge-intensive visual questions. However, existing evaluations mainly focus on final-answer accuracy and may miss failures in the search trajectory. In this work, we study such hidden reliability issues as silent failures. We introduce a six-category taxonomy covering modality shortcuts, phantom grounding, wrong-evidence-right-answer cases, over-retrieval laundering, cross-modal contradiction, and provenance hallucination. Based on this taxonomy, we build a trajectory-level diagnostic pipeline that evaluates both answer correctness and evidence-grounding quality under a unified ReAct-style scaffold. Experiments on MMSearch-Plus trajectories across four frontier multimodal models show that surface accuracy consistently overestimates true trajectory-level correctness. We further use cross-judge validation, blank-image stress tests, and tool ablations to show that silent failures are capability-dependent and often shift rather than disappear. Home-page: https://github.com/DingWu1021/silent-failures-multimodal-agentic-search

LLM preference alignment aims to optimize models toward human preferences across diverse user instructions. Reinforcement learning has become a major post-training approach for this goal, but existing proxy rewards are often outcome-level, mainly evaluating the final response while providing limited guidance for the reasoning trajectory. This can make credit assignment coarse when multiple responses receive similar final scores, leaving trajectory-level preferences under-specified. To address this limitation, we propose Thinking Checklist Reward (TCR), a process-oriented reward for RL-based preference alignment. TCR converts preference pairs into sample-specific thinking checklists and uses them to evaluate whether the generated reasoning trace addresses the preference-implied considerations. To reduce overlap with outcome-level supervision, TCR further introduces an exponential moving average (EMA) residual formulation to isolate a complementary thinking surplus beyond what is predictable from the outcome reward. Experiments on five models from three model families show that TCR consistently improves alignment performance across diverse benchmarks, with ablations further validating the importance of EMA-based residual formulation and sample-specific checklist supervision.

Traditional pentesting uses reconnaissance at each step to uncover unseen weaknesses, build stronger attacks, and advance the objective; we argue that AI agents require the same treatment. We formalize agent reconnaissance by modeling the process and identifying the knowledge assets it seeks to extract: what they are, how they are used, and which agent weaknesses they exploit to give adversaries leverage in indirect prompt injection attacks. We instantiate these insights in Know Your Agent (KYA), a framework that automates black-box, reconnaissance-driven pentesting by probing agents, building target profiles, and using those profiles to craft stronger attacks. We evaluate KYA on agent-security benchmarks and a real-world coding agent, and release KYA, its benchmarks, and baseline implementations for reproducibility.

As autonomous agents rapidly evolve, their ability to reliably manipulate ubiquitous digital documents has become critical for enabling general-purpose AI assistants and automating complex workspace workflows. In this paper, we introduce DocOps, a deterministically verifiable evaluation framework underpinned by a hierarchical taxonomy that deconstructs document operations inspired by real-world practices into atomic dimensions and escalating workflow complexities. Based on DocOps, we systematically evaluate representative closed- and open-source models across various agentic harnesses, revealing that even the most advanced frontier configurations still exhibit profound limitations when handling highly coupled, long-range tasks. Furthermore, a fine-grained analysis of existing agents' manipulation behaviors uncovers 3 key failure modes: long-term state tracking collapse, shallow semantic verification, and destructive editing of structural metadata. Ultimately, our work exposes the capability boundaries of agents in maintaining global document consistency, shedding light on the future design of robust, non-destructive agents for complex digital ecosystems.

Agent safety is moving from content moderation toward preventing operational failures before tool-using agents act. We propose Janus, a foresight-oriented framework for long-horizon agent safety that trains guards to anticipate delayed risks from partial trajectories. Janus synthesizes diverse agent trajectories via multi-agent simulation and learns a shared policy with two coupled tasks: an anticipation task that forecasts safety-relevant futures and an adjudication task that decides safety from both the observed prefix and anticipated future. The two tasks are jointly optimized with CoAA-RL, which rewards forecasts by their utility for downstream safety judgment. The resulting guard model, Vanguard, blocks unsafe actions before execution. Across four agent-safety benchmarks, Vanguard improves average protection by 15.9 percentage points over baseline guards while increasing benign task completion by 5.1 percentage points.

We study finite-horizon MDP planning under \emph{root-based} (resolute) risk objectives that apply a rank-dependent functional to the distribution of total returns. Such objectives are non-linear in the return distribution and generally break Bellman optimality, so direct optimization by scenario-tree enumeration is intractable. We propose \textbf{ERQDP}, an enumeration-free and sampling-free method that solves a rank--quantile surrogate via exact DP (Dynamic Programming), evaluates candidate policies exactly by DP over return Probability Mass Functions (PMFs) on a discretized return grid (with an explicit rounding bound), and refines the surrogate in an anytime loop that reports an explicit upper--lower gap (certificate) for the target objective up to discretization budgets. Across tested benchmarks, ERQDP returns certified solutions or explicit residual gaps, enables fast risk-parameter sweeps with substantial runtime gains, and supports both risk-averse and risk-seeking behaviors.

Large metal-organic framework (MOF) databases support simulation, screening, and machine learning through crystallographic information files (CIFs). Subtle chemical and structural errors in these inputs can compromise downstream results and hinder manual inspection. LLM advances in computational chemistry offer paths beyond predictive screening toward fine-grained diagnosis with evidence-grounded explanations. However, two challenges remain: (i) limited fine-grained attribution: MOF-specific validators and machine-learning models scale detection but provide fixed checks, readiness scores, or coarse labels rather than evidence-grounded explanations; and (ii) unreliable CIF reasoning: direct LLM auditing is costly and unreliable because chemical evidence is implicit across atom-site records and requires geometric, connectivity, occupancy, and charge calculations. Both stem from weak coupling between chemical evidence and language-model explanation. We introduce MOF-Sleuth, a reinforcement-guided CIF auditing agent with two modules: a deterministic Forensic Lab and a Sleuth reasoning engine. The Lab derives composition, geometry, connectivity, occupancy, coordination, and charge evidence, and Sleuth uses this evidence to produce an evidence-grounded explanation, error types, and a binary decision. Reward-guided reinforcement learning (RL) turns tool measurements into chemical explanation-level supervision, rewarding not only the final answer but also cited chemical evidence and evidence-supported diagnoses. We introduce Chemically Grounded Diagnosis (Chem-GD), a metric that assesses whether a correct diagnosis is explained by factual, relevant CIF-derived evidence. Across four benchmarks, MOF-Sleuth establishes state-of-the-art performance among LLM-based approaches and MOF-specific machine-learning methods, demonstrating gains in detection, attribution, and grounded explanation quality.

Smartphone personal assistants reason over longitudinal personal data, yet evaluating them requires context-rich evaluation data whose correct answers are known, and real device traces are too privacy-sensitive to share. To address this challenge, we present SenWorld, a physically grounded, deterministic, event-sourced digital-twin simulation that generates such data with ground truth fixed by construction. In SenWorld, personas live through a full day in a world built from real map, weather, holiday, and network data; every observable signal is archived in full-system snapshots; and each evaluation case is labeled by a pointer to an existing record rather than by post-hoc annotation or a large language model (LLM) judge. We evaluate this method with 16 personas in Beijing. The generated data closely matches the held-out real-user benchmark in category distribution (Jensen--Shannon divergence (JSD) 0.070) and in the daily rhythm of communication records (JSD below 0.1), though generated records remain shorter than real ones. Without scripted interaction, personas form a fully reciprocated dialogue subgraph and differentiated behavioral repertoires. Projected into 717 evaluation cases, the generated data exposes 78 failures in a production smartphone assistant, concentrating on call and Short Message Service (SMS) records while contacts, schedules, and alarms never fail. The snapshot pointer confirms each failure as an assistant-side retrieval error, with no LLM judge involved. Overall, SenWorld offers a privacy-safe, reproducible, and distribution-checked path to evaluation data whose labels are fixed by construction.

Large Reasoning Models (LRMs) often suffer from overthinking due to redundant verification steps. Existing approaches for mitigating overthinking, such as fast-slow thinking switching and reasoning trajectory compression, fail to make a fine-grained distinction between beneficial and redundant steps within the LRM's reasoning process, and may thus impair reasoning capability in their pursuit of efficiency. To simultaneously improve reasoning efficiency and capability, we propose EvoThink, a framework that reduces redundant verification and encourages the exploration of new reasoning paths. EvoThink comprises two key components: Self-Pruning Training (SPT), an unsupervised method that iteratively prunes redundant reasoning steps and self-trains on the concise trajectories; and Aha-Moment Preference Optimization (AMPO), which, inspired by genetic algorithms, identifies valuable failed reasoning attempts, synthesizes from-wrong-to-right aha-moment data, and optimizes the model to internalize this reasoning pattern. Extensive evaluations across mathematical reasoning and code generation benchmarks demonstrate that EvoThink not only substantially reduces inference-time token usage but also improves the reasoning capability of LRMs.

AI researchers describe state-of-the-art models as one thing repeated at scale: the Transformer, wired identically for text, pixels, or speech. Neuroscientists describe the cortex as a mosaic - dense Layer 4 in visual cortex for spatial encoding, thick Layers 5/6 in motion cortex for temporal integration - different jobs solved by different structures. This paper argues the gap is a structural error, not a stylistic one, and is measurable. A century of cytoarchitecture, from Brodmann to single-cell Patch-seq, shows distinct cognitive functions are implemented by qualitatively different structures, not by rescaling one template. The convolutional neural network is the field's own proof: local receptive fields and hierarchical depth encoded this prior directly, reaching strong image recognition on far less data than later architectures needed. The paper traces how this lesson was discarded: the "Hardware Lottery" made the Transformer the path of least resistance, not the principled choice, and Mixture-of-Experts, often cited as diversity, in fact partitions parameters among identical experts. A functionalist analysis shows the Transformer is best understood as a functional analog of the hippocampal formation, not a general-purpose cortex - the same mistake as treating cortex as one giant Broca's area, except the field has now standardized on a giant hippocampus, applied to tasks it was never built for: audition, executive gating, working memory. The paper closes with an alternative: a Heterogeneous Topological Network, a System of Systems in which distinct modules keep the inductive bias their computation demands and communicate through standardized interfaces. This is a design discipline for AI architects, not cognitive science: specify modularity before training, using structural evidence as a design input rather than reverse-engineering architecture from a trained model's behavior.

Dynamic manufacturing environments require multi-agent systems to coordinate effectively under frequent operational disturbances such as machine failures, urgent job arrivals, and processing time variations. Existing multi-agent reinforcement learning approaches treat each disturbance episode independently, discarding valuable coordination experience that could accelerate future adaptation. In this paper, we propose a Graph-Structured Experiential Memory (GSEM) framework for multi-agent coordination in dynamic manufacturing. The framework encodes historical coordination episodes as heterogeneous relational graphs that capture task dependencies, machine states, and inter-agent collaboration patterns. When a new disturbance occurs, a graph neural network-based retrieval mechanism identifies structurally similar past episodes, enabling experience-guided policy adaptation rather than learning from scratch. Experiments on dynamic flexible job-shop scheduling benchmarks with three disturbance types show that GSEM reduces makespan by 4.1%-10.0% and adaptation time by 33%-38% compared to the strongest memory-augmented baseline, with the advantage increasing under higher disturbance frequency. Ablation studies and cross-disturbance transfer experiments further validate the necessity of graph-structured encoding and similarity-based retrieval and demonstrate the cross-disturbance generalizability of learned coordination patterns.

Machine Learning models are widely used for automating classification tasks by extracting statistical patterns from data. However, their performance deteriorates if the data distribution changes, making them ill-suited to handle uncertain and evolving information. Moreover, they provide limited support for integrating prior knowledge. To address these limitations, we present CLARK (Closed-loop Learning for Adaptive Reasoning over Knowledge Graphs), a framework that integrates knowledge graphs, symbolic rule mining, and probabilistic reasoning under the Logic Programs with Markov Logic Networks (LP$^{\text{MLN}}$) formalism. Starting from CACTUS-derived KGs, CLARK translates graph structure into an LP$^{\text{MLN}}$ program and iteratively enriches it with candidate rules proposed by symbolic learners. These rules are calibrated through probabilistic weight learning, enabling reasoning under uncertainty and refinement of the underlying graph structure. We evaluate CLARK on two medical datasets, analysing both rule quality and downstream classification performance. Results demonstrate that CLARK leads to improved classification performance and more generalisable inference. Overall, CLARK provides a principled approach to constructing adaptive, interpretable, knowledge-driven models for classification.

In modern IT operations (IT-Ops), the cost of an incorrect repair often exceeds the cost of no action at all. Yet existing automated remediation systems are designed to generate actions rather than to decide whether intervention is warranted, leaving safety as an afterthought enforced by manual approval. This paper makes three contributions to close this gap: (i) we reformulate safe remediation as a risk-constrained intervention decision problem and cast it as a Constrained Markov Decision Process (CMDP), in which the agent maximizes repair success subject to a bounded false remediation rate (FRR); (ii) we introduce a three-dimensional risk decomposition comprising blast radius, reversibility, and epistemic uncertainty, providing operators with an interpretable per-action safety interface; and (iii) we design a context-adaptive human-in-the-loop (HITL) gate that turns escalation from a binary failsafe into a bandwidth-aware control layer responsive to on-call load and business criticality. The full policy is learned offline from historical incident logs, enabling explicit control of the expected FRR. Experiments on the Train Ticket microservice benchmark with Chaos Mesh fault injection and an RCAEval-aligned fault taxonomy show that our framework reduces FRR by 39% while improving repair success by 2.5 points over a strong runbook baseline, and reduces on-call escalation load by 17% relative to a fixed-threshold variant.

Design rule check (DRC) closure remains a major bottleneck in advanced-node physical design. Although detailed routers are rule-aware, residual design rule violations (DRVs) often require manual engineering change order iterations. Automating this process is challenging because repairs must account for complex geometric interactions, preserve circuit connectivity, and avoid introducing new violations. We present EvoDRC, a skill-evolution framework for agentic block-level DRC repair. EvoDRC initializes layer-specific repair skills using knowledge distilled from an unrelated reference design and continuously evolves these skills using traceable repair experience collected from the target design. EvoDRC decomposes the layout into bounded repair regions and assigns an LLM repair agent to each region. Local DRC analysis, connectivity-checking, and impact-preview tools provide feedback on proposed modifications. Repair operations and their resulting DRV changes are stored in a knowledge database and used to evolve the repair skills. Experiments on seven block-level designs from the DAC26 DRC Benchmark show that EvoDRC achieves a 73.5\% overall reduction compared to the reported baseline.

Difference constraints of the form $x - y \leq d$ are well studied, with efficient algorithms for satisfaction and implication, because of their connection to shortest paths. Finite domain propagation algorithms, however, typically do not make use of these algorithms, and treat each difference constraint as a separate propagator. Propagation does guarantee completeness of solving, but can be needlessly slow. In this paper we describe how to build a (bounds consistent) global propagator for difference constraints that treats them all simultaneously. SAT modulo theory solvers have included theory solvers for difference constraints for some time. While a theory solver for difference constraints gives the basis of a global difference constraint propagator, we show how the requirements on the propagator are quite different. Crucially, we show how to explain propagations by a global difference constraint propagator, in order to use it within a lazy clause generation solver. We give experiments showing that treating difference constraints globally can substantially improve on the standard propagation approach.

Large language models can answer scientific questions, yet a correct output does not reveal whether the model represents or uses the governing physics. Here we show that materials science mechanism information in the open-weight google/gemma-4-E4B-it model has three experimentally separable forms: concepts are readable in individual hidden states, constitutive orientation is carried by controlled transformations between states, and selected internal representations causally control engineering answers. We combine matched direct and Jacobian vocabulary readouts, option-free state geometry, a 60-law counterfactual benchmark and causal interventions. In 50 held-out materials descriptions, three independently fitted Jacobian lenses reproduced concept ranks, and target-free word sets from both readouts enabled blinded identification of 9 of 10 mechanism families. A separate 72-prompt benchmark produced mechanism-specific hidden-state neighborhoods, but an exact graph audit showed that this apparent physical organization was equally explained by numerical comparison. We therefore compared otherwise identical prompts in which only the direction of the physical input was reversed, asking whether the resulting hidden-state movement followed the supplied constitutive law. These state transformations ordered direct, physically neutral and inverse laws across 60 frozen relations and correctly oriented 39 of 40 directional laws, whereas lexical controls were near chance. Bidirectional interventions shifted answer probabilities toward or away from the physically appropriate outcome across all 12 matched cases, while counterfactual state patches transferred opposing decision signals across mechanisms and answer formats. Physical relationships were therefore more visible in controlled state changes than in absolute states alone.

Long-horizon tasks require sustained perception, reasoning, and exploration, and are a persistent challenge for large language model (LLM) agents. This gap is reflected in their limited performance on continual learning benchmarks such as ARC-AGI-3, especially when models are evaluated out of the box. Various agent harnesses have been proposed to close this gap, and each commits to a strategy for handling long sequences of observations, i.e., what information to save from the environment and how to load it into model context, a choice we argue is particularly consequential. Existing methods for context management face a significant tradeoff, as preserving more information makes retrieving relevant details less tractable. We propose PRO-LONG, a minimal context management framework built around programmatic memory for LLM agents in long-horizon, exploratory settings. PRO-LONG addresses the tradeoff by keeping a complete, structured interaction log and capitalizing on recent progress in coding agents to search this history efficiently. On the full ARC-AGI-3 public game set, PRO-LONG improves over a base coding agent by an average of 18.0 percentage points across frontier models, and matches or exceeds state-of-the-art specialized harnesses (up to 76.1% pass@1) while using 4.2-5.8x fewer tokens. With Fable 5, PRO-LONG achieves 97.4% best@2 at a total cost of \$1,750. Relevant code and logs are available at https://github.com/alexisfox7/PRO-LONG.

Enterprise strategic decision support requires AI systems that are not only accurate, but also uncertainty-aware, risk-calibrated, explainable, and governance-compliant. This paper proposes TRUST-ESD, a risk-calibrated and governance-aware framework for enterprise decision support under uncertainty. TRUST-ESD evaluates feasible counterfactual strategies through predictive utility estimation, conformal uncertainty calibration, CVaR-based downside-risk scoring, risk-memory retrieval, policy-as-code governance, explainability, and human oversight. Unlike prediction-only methods that select actions by maximum expected utility, TRUST-ESD recommends strategies that balance value, reliability, risk exposure, and compliance. Experimental results show that TRUST-ESD improves risk-adjusted utility by 7.95%, reduces risk exposure by 23.22%, reduces CVaR by 23.78%, lowers calibration error by 13.89%, improves explanation fidelity by 10.90%, and increases governance compliance by 9.76% compared with strong uncertainty-aware baselines, while maintaining competitive predictive accuracy. Ablation and case-study analyses further confirm that uncertainty calibration, downside-risk scoring, risk memory, explainability, and governance validation jointly improve trustworthy enterprise decision-making.

Quantized small autoregressive reasoning models can enter long, repetitive, or unproductive trajectories, yet inference-time compute is usually allocated without observing how a trajectory develops. Building on an earlier token-level e-CUSUM controller, we develop MGT-B (Monitoring-Guided Test-time Backtracking), a revised external controller that maps overlapping windows of pre-sampling uncertainty and degeneration features to position-conditional empirical tail probabilities, accumulates mixture betting factors with a CUSUM-shaped reset, and responds to an alarm by estimating a rollback point, restoring token and key-value-cache state, and performing constrained re-decoding. To audit whether the effect persists on problem identities first observed after the manual choice of log threshold h = 10, we retrospectively exclude 260 IDs present in pre-threshold artifacts and retain the chronologically first post-threshold pair for each remaining ID, yielding a 240-pair chronology-audit set. On this set, accuracy changes from 82/240 to 88/240 (+2.50 percentage points; 13 corrections, 7 regressions; exact McNemar p = 0.2632; paired bootstrap 95% interval [-1.25, +6.25]). A broader 467-pair historical-coverage set of seed-matched pairs changes accuracy from 146/467 to 167/467 (+4.50 points; McNemar p = 0.000753), but includes 200 seed-1 IDs available before or during threshold selection and is reported only as an exploratory estimate. All 316 no-alarm outputs in the 467-pair set are identical to vanilla, while the 151 alarmed trajectories contain 29 corrections and 8 regressions. Neither analysis is confirmatory, and the empirical factors are not established as a valid e-process or e-detector. The results support a selective monitoring-and-repair mechanism for the studied MATH-500 setting, rather than a general or theoretically certified reasoning improvement.

While Large Language Models (LLMs) excel at many tasks, they frequently struggle with complex reasoning that requires long-horizon planning and iterative error correction. Furthermore, standard single-stream prompting proves brittle when models encounter novel abstractions or rigorous domain constraints. We introduce PoTRE (Poly-Topological Reasoning Ensembles), a heterogeneous framework that decouples inference into four agents: (1) Adversarial Refinement Agent, (2) Hierarchical strategic Planning Agent, (3) Spectrum Search Agent, and (4) Direct Chain Agent. A final Task-Adaptive Aggregation Layer dynamically reconciles these perspectives -- via final candidate selection, semantic synthesis, or neuro-symbolic verification -- to produce a robust global solution. We evaluate PoTRE on three frontier benchmarks: ARC-AGI-2, Humanity's Last Exam (HLE), and PRBench Finance. PoTRE achieves state-of-the-art accuracy of 49.92% on HLE, surpassing the previous best official score. We demonstrate that this architectural heterogeneity achieves improved reasoning performance using similar or fewer inference tokens compared to heavily scaled homogeneous baselines.

Natural-language autoencoders score explanations of hidden activations by reconstruction: an explanation is deemed faithful if the activation can be regenerated from it. The test is structurally insensitive to individual false claims: if flipping a claim does not change the reconstruction, the claim is never penalized. We show the test is passed in two ways, neither faithful. On a released Qwen-2.5-7B verbalizer, explanations reconstruct well above chance while ~2% of specific claims are reconstruction-dependent, so the score tracks gist, not specific facts. Under exact synthetic ground truth, the standard recipe develops co-adapted private codes (false wording the reconstruction depends on) in 5/5 runs, and fixes that leave the target model unchanged do not help. We contribute two audit protocols, the grounded-vs-true cross and the evaluator swap, and RECAP (Readable Encodings via Co-trained Auxiliary Predictors): linear heads trained alongside the target model to keep designated content decodable. On RECAP-trained sandbox models, fresh verbalizers state the designated content truly and the codes vanish, at a +0.001-nat cost. This replicates on a pretrained Pythia-160M: the content becomes reliably probe-decodable, though a fresh verbalizer conveys it only in part (truth 0.44-0.46 vs a near-zero control). For interpretability, high reconstruction does not certify individual claims. For AI safety, RECAP makes designated internal content independently checkable against probes rather than asserted by prose a model can game: an independent probe scores the verbalizer's true claims above its false ones (AUC 0.96, vs 0.82 without RECAP). Against an adversary that edits an explanation to maximize the reconstruction score while lying (suppressing ~87% of its lie penalty), the RECAP probe still flags the lies (AUC 0.95) while the control probe collapses to chance (0.51).

In many reasoning problems, the premises are not observed as discrete symbols, but must be inferred from high-dimensional inputs. Further, the predicate vocabulary, argument structure, and trusted evidence are supplied by a Knowledge Graph (KG), or rule definitions. Classical neuro-symbolic pipelines have a discrete interface between perception and deduction. We present a neuro-soft-symbolic architecture for differentiable deductive reasoning over latent perceptual facts and knowledge-provided predicates. SoftReason removes the gradient gap by representing the deductive state as a local soft interpretation tensor over candidate constants and predicates. Perception proposes probabilistic base facts, KG triples enter as high-confidence soft evidence, and every query anchor, predicate choice, and closure update remains differentiable. Our core innovation is a learned differentiable lift of the immediate-consequence operator. It uses predicate-definition embeddings and latent composition channels to form soft body-predicate mixtures, aggregate over all possible witnesses, propose query-conditioned head facts, and update the interpretation through a monotone probabilistic OR. We instantiate the framework on Knowledge-aware Visual Question Answering (KVQA), and demonstrates how SoftReason supports end-to-end perceptual grounding, KG evidence injection, and differentiable deductive closure in one trainable architecture.

Most safety guardrails for large language models (LLMs) evaluate each prompt-response pair in isolation, which misses failures that arise only over a dialogue as benign turns compose into harm. We term this Conversational Risk Accumulation (CRA): gradual intent drift, fragmented assembly of prohibited instructions, and sensitivity build-up from repeated disclosures. We propose a session-layer CRA Framework that tracks three trajectory signals: semantic drift from a session anchor, a sensitivity-weighted information accumulation graph over extracted entities, and a compliance-gradient signal capturing increasing willingness to comply. For scoring, we provide (i) an unsupervised convex fusion for attribution and ablations, and (ii) CRA-Net DA, a compact learned trajectory model trained with family-adversarial objectives to reduce length and topic-coverage confounds. To benchmark CRA, we release CRA-Bench v0.1 (1,200 eight-turn sessions across three threat families with topic-matched benign twins), CRA-Bench v0.2 (LLM-paraphrased variants to reduce template artifacts), and an extended 5-family set (2,000 sessions adding persona priming and context stuffing). We introduce a trajectory-native evaluation protocol with session-level splits, mixed-set threshold calibration, Trajectory AUROC, turns-to-detection, calibrated false-positive metrics, bootstrap confidence intervals, leave-one-family-out diagnostic stress tests, and synthetic-to-human transfer checks. Claims focus on within-distribution session scoring on CRA-Bench and human-transfer subsets.

Supervised fine-tuning (SFT) is widely used to adapt large language models to downstream tasks, but its effect on behavioral diversity in sequential decision-making remains under-explored. We study this question in a controlled suite of deterministic board games based on tic-tac-toe variants, where optimal actions are exactly computable and diversity can be measured directly. Across state-level evaluation, arena gameplay, and training trajectories, we find that reasoning-mode generation frequently suppresses action diversity without uniformly improving action accuracy. Furthermore, standard SFT improves accuracy but often induces premature diversity collapse, which exceeds what is minimally required by the accuracy-diversity tradeoff. We then show that action augmentation, which trains on all optimal actions per state rather than a single demonstrated action, would partially mitigates this effect. Our results identify narrow-support imitation as a source of policy collapse in LLM decision-making and suggest that preserving action support during SFT is important for maintaining exploratory behavior.

Structural generalization has been measured repeatedly by several benchmarks, yet it has never been formally defined. We give a definition that translates the two premises (compositional structure and unbounded generalization) into mathematical language. The definition itself is neutral: a compiler that hard-codes the rules satisfies it just as well. But structural generalization becomes a scientific question only insofar as the capacity can autonomously emerge from finite data. This question pits the computational lower bound $\mathrm{NC}^1$ against the learnable ceiling $\mathrm{TC}^0$ of pure Transformers. Under a Montagovian instantiation, each compositional rule splits into two projections: a syntactic face ($F_\gamma$) and a semantic face ($G_\gamma$). Tree evaluation on the $G_\gamma$ side is an instantiation of BFVP, which is $\mathrm{NC}^1$-complete (Buss, 1987). A pure Transformer must learn both faces at once, but Kraus et al. (2026) prove that its learnable class $\subseteq \mathrm{TC}^0$. Under the standard assumption $\mathrm{TC}^0 \neq \mathrm{NC}^1$, a pure Transformer cannot learn structural generalization. Neuro-symbolic systems achieve the best benchmark scores precisely because they inject $G_\gamma$, sidestepping the genuinely hard half. Benchmark scores cannot distinguish "learned" from "given." This is what this paper sets out to make clear.

Injecting factual knowledge into large language models (LLMs) reliably and at scale remains an open challenge. Hypernetworks provide a promising solution to large-scale knowledge injection. Although hypernetworks are typically applied for test-time adaptation, we explore their use in train-time knowledge injection, where, given a large corpus of facts, we train a hypernetwork to generate a fixed LoRA adapter that, when inserted into the target model, enable the model to answer questions about those facts. In this work, we investigate whether hypernetworks can be used to perform train-time knowledge injection and how this ability varies with scale. The scaling behavior of hypernetworks remains largely unstudied. Our design decouples the hypernetwork's injection capacity from the target model's general capability, enabling, for the first time, a rigorous study of scaling laws for hypernetwork architectures. We characterize how loss, reasoning accuracy, and out-of-distribution (OOD) generalization vary with hypernetwork depth, width, and target network size. We construct a large-scale dataset, called MegaWikiQA, containing tens of millions of multi-hop question-answer examples across 39 domains constructed from examples in Wikidata5M. Our results reveal: (i) hypernetwork-based injection exhibits broadly predictive power law scaling along all architecture axes; and (ii) hypernetworks are capable of reliable OOD generalization at increasing scales, suggesting that hypernetwork provides a promising alternative to other train-time adaptation methods such as LoRA finetuning and full fine-tuning, exhibiting steeper scaling exponents in all OOD evaluations. Together, these results establish hypernetworks as a principled and scalable substrate for train-time adaptation, and provide the first empirically grounded scaling laws to guide hypernetworks for factual reasoning in large language models.

Instruction tuning is meant to make language models follow user requests, yet it is unclear whether small models comply when an instruction conflicts with their usual task behavior. We study this across three tasks - multiple-choice question answering (MCQA), sentiment classification, and mathematical question answering - by pairing a standard instruction with a conflicting non-standard one (select an incorrect option, output the opposite sentiment, or return twice the answer). This cross-task design allows us to test whether resistance to conflicting instructions is tied to specific task characteristics or reflects a broader behavioral tendency. As all predictions are scored against the original ground truth, a model that ignores the non-standard instruction still appears accurate. Using standard accuracy, non-standard accuracy, and an Instruction-Following Failure Rate (IFFR), we evaluate instruction-tuned Qwen models across sizes. Both standard accuracy and instruction following generally improve with scale, although the pattern is not consistent across all tasks and datasets. Small models stay competent yet routinely ignore the non-standard instruction, while larger models show a clear gap between the two settings. These findings suggest that gains in task capability do not automatically provide reliable control over model behavior. Task competence and instruction following are therefore distinct abilities, and reporting only standard accuracy hides instruction-following failures.

Large language models operating in emotionally sensitive contexts face a structural trilemma: when users in vulnerable states request information that may reinforce maladaptive attribution, current response architectures resolve the tension through protective restriction, uninflected facilitation, or unintegrated co-presence of both imperatives -- each preserving one objective at the cost of the other. Administering a three-turn escalating vulnerability vignette to three commercial LLMs (900 sessions across material, relational, and somatic status-proxy variants) and coding responses with two binary indices (VCC/VCI), we characterize a previously undocumented failure mode we term adaptive capitulation: the model validates the social injustice underlying the user's distress before pivoting to detailed facilitation of the very acquisition it nominally discouraged. We show that the trilemma is structural rather than incidental, and propose Minimal Reattributive Sufficiency (MRS), an architecture-neutral design principle that embeds a single reattributive cue within an otherwise validating response, preserving a pathway toward autonomous reattribution without contesting the user's stated goal.

AI-generated answers in high-stakes domains are often fluent but difficult to verify, especially when they contain multi-step reasoning rather than a single final answer. We propose a reasoning-based, reference-free framework for auditing LLM-generated outputs. The method decomposes a generated reasoning trace into segments, labels local premise-target relations using Natural Language Inference (NLI), and organizes these relations into a hypergraph. A deterministic backward AND-OR search then assigns segment-level audit labels that indicate how each segment is grounded within the generated response. We evaluate the framework in two settings: deductive mathematical reasoning with Hard2Verify, and open-ended medical reasoning with UroReason, a new physician-annotated benchmark of LLM reasoning traces from real clinical cases. Across these settings, our NLI-hypergraph audit provides a more reliable reference-free evaluation signal than direct LLM-as-judge baselines. In the clinical setting, state-of-the-art LLM judges often fail to identify problematic reasoning segments, over-accepting fluent but weakly grounded responses. Our results show that QA evaluation should account for how inferential relations compose across a reasoning trace, rather than relying only on final answers or LLMs as verifiers. UroReason will be made available through an API, and our code will be released as open source.

Masked diffusion models (MDMs) are a promising family of language generators, but achieving high-quality few-step generation remains challenging. In MDMs, all forward trajectories collapse to a single fully masked state, leaving no terminal entropy for consistency-style few-step generation. While recent few-step alternatives based on uniform-state diffusion avoid this degeneracy, it becomes harder to distinguish clean tokens from noise than MDMs, which usually harms modeling quality and training efficiency. In this work, we propose a multi-mask diffusion model (MultiMDM) that preserves the masking structure towards few-step generation. In the forward process, each clean token is first pushed towards a designated mask and then gradually mixes over the mask set. As a result, the backward process has a drafting capability by predicting a designated mask before refining to a clean token. We derive a closed-form ELBO training objective for MultiMDM that supports continual training from pretrained MDMs. In addition, we formulate a purely discrete-state consistency distillation scheme, with a shared-Gumbel coupling to reduce pathwise entropy. Experiments on pretraining and distillation show that MultiMDM provides an effective foundation for principled few-step generation.

Reinforcement learning with verifiable rewards has become the predominant recipe for eliciting test-time scaling in explicit Chain-of-Thought reasoners. Yet this scaling path remains computationally costly, since every intermediate step must be decoded as a language token. Latent reasoning instead carries intermediate computation as continuous vectors and already matches or surpasses explicit CoT at far shorter horizons. Despite this promise, latent reasoners remain largely imitation-bound, while explicit CoT has already moved past imitation via outcome-reward RL. Latent trajectories lack a tractable per-step likelihood and an adaptive stopping interface under fixed thinking budgets, so outcome rewards cannot elicit latent test-time scaling. We introduce Surrogate Latent Policy Optimization (SLPO) to bring outcome-reward RL to autoregressive latent reasoners: an empirical surrogate policy density over latent transitions for trajectory-level credit assignment, and a correctness-supervised stopping head that outcome-reward optimization refines into a variable-horizon policy. Across continuous and soft thinking settings, SLPO improves Pass@$k$ under parallel sampling and allocates longer latent computation to harder instances with higher deterministic accuracy.

The HIPE-2026 shared task introduces person-place relation extraction from multilingual historical newspapers as a new evaluation track, classifying the at and isAt relations between pre-annotated person and location mentions in English, French, and German. Motivated by the cost of processing historical archives at scale, our team (DS@GT HIPE, team 2 in the official results) investigates how far a lightweight, interpretable system can go without any pretrained language model at the relation classification stage. Our approach builds a document-level graph from dependency parses, extracts proximity-based and part-of-speech features for each entity pair, and classifies them with small scikit-learn ensembles or compact Graph Attention Networks, keeping every submitted run under 847K parameters. On the official evaluation (Test A, the newspaper test set), our best run reached a macro recall of 0.5142, ranking 3rd on the Efficiency profile while placing mid-table on Accuracy among the 17 participating teams. Two findings stand out. First, minimum character distance alone captures most of the classification signal; adding further engineered features yields inconsistent gains and sometimes degrades performance, echoing prior evidence that argument distance dominates relation extraction. Second, document-grouped cross-validation is essential on this corpus: pair-level splits inflate scores by 25-37 percentage points because entity mentions recur across documents, a data-leakage effect that grouped cross-validation removes.

As large language models and AI agents become the primary consumers of search results, document set quality determines the upper bound of downstream generation. Yet existing evaluation systems remain confined to scoring documents independently and aggregating via nDCG, ignoring inter-document interactions (redundancy, conflict, complementarity) and unable to answer what makes one document set better than another. To address these issues, we propose a complete evaluate-diagnose-optimize framework. We design SetwiseEvalKit, a three-level, nine-dimension document set evaluation benchmark covering both short-form and long-form scenarios, comprising approximately 28K high-quality evaluation rubrics. We systematically evaluate 12 rerankers: even the best method achieves no more than 45% coverage, cross-document coordination dimensions are universally weak, and no single method maintains top performance across both settings. Building on this, we propose Rubric4Setwise, a training-free method that converts rubric-based evaluation criteria into document set selection signals, achieving the best downstream generation performance with fewer documents and search rounds. It is the only method that maintains state-of-the-art results across both scenarios, validating the effectiveness of closing the loop from evaluation to optimization.

We present a regression-based approach to Arabic dialect geolocation that models dialectal variation as a continuous geographic space rather than discrete categories. Speaker origin is predicted as continuous latitude-longitude coordinates using a hierarchical neural architecture that fuses frame-level XLS-R-300M and Whisper-large-v3 encoder representations with phonotactic descriptors through a Transformer encoder and a learnable attention-pooled query. A spherical geodesic loss directly optimizes great-circle distance on Earth's surface, avoiding distortions inherent to planar coordinate regression. Under a leakage-free 5-fold GroupKFold protocol grouped by source recording, our model attains a pooled median localization error of 481.2 km. Auxiliary country and city heads reach 64.5% and 45.2% accuracy, respectively. A permutation Mantel test on the learned latent space provides quantitative support for the Arabic dialect continuum hypothesis. To probe true generalization, we further introduce a city-masking protocol in which two cities per fold are removed from training but retained in validation. Under this zero-shot regime, the mean error rises to 1173.3 km, a 1.32x degradation relative to seen cities. Our findings establish continuous geographic modeling as a principled framework for Arabic dialect geolocation and quantify both its strengths and the substantial headroom that remains.

Production LLM-based financial sentiment analysis faces a structural cost trap: most queries are trivially classifiable, yet expensive cloud reasoners process them all, and the bill scales linearly with user count. We present TriAgent, a multi-agent committee stratified by contextual granularity -- a word-level lexicon (VADER), a sentence-level domain transformer (FinBERT), and a cross-sentence reasoner (Qwen2.5, 0.5B-14B-4bit, with Mistral-7B and Phi-3.5-mini cross-family checks). A three-way Semantic Divergence Index (SDI) measures pairwise disagreement across granularities and routes each query accordingly. Our central finding is the critic plateau: when the LLM is re-tasked as a critic over the smaller agents' outputs, F1 plateaus at ~0.87 across 1.5B-7B Qwen (bootstrap 95% CIs overlap), while a same-size 3-persona vote drops to F1=0.66, which is driven by granularity-stratified diversity. Three corollaries follow from the same SDI signal: (i) a Shared Consensus Dictionary on multilingual sentence-BERT answers 95% of Chinese queries from an English cache at F1=0.99 -- cross-border canonicalization at zero marginal cost; (ii) SDI doubles as a post-hoc LLM-hallucination detector at AUC=0.90; (iii) the SDI single-stage strategy attains the best risk-adjusted return (Sharpe=3.50) on a 20-ticker back-test, dominating both always-FinBERT (1.36) and always-LLM (0.11). At 10M-user scale, TriAgent saves $9.3M/year vs. a GPT-4o-mini baseline. Code, lexicons, and the SCD are released.

Hypergraph-based RAG systems surpass traditional graph-based approaches by organizing complex n-ary atomic facts among entities, rather than relying solely on binary relationships. Despite the advancements in multimodal large language models (MLLMs) with enhanced visual capabilities, current hypergraph-based RAG frameworks predominantly restrict knowledge retrieval and reconstruction to a unimodal, text-centric paradigm. This limitation prevents them from fully leveraging the powerful visual perception capabilities of modern MLLMs. To address this gap, we systematically explore the integration of hypergraph awareness in RAG systems through visual cues. By incorporating visual representations of hypergraphs into the RAG pipeline, we introduce VizRAG, the first RAG system to support visual hypergraph structure awareness. Experimental results demonstrate that VizRAG significantly outperforms strong baselines, validating the promising potential of hypergraph visualization as a novel approach for RAG systems.

With the wide application of large language models (LLMs) in real-world scenarios, the value implication of their outputs is crucial. However, existing evaluation benchmarks suffer from insufficient coverage of value dilemmas in daily scenarios involving multiple value conflicts and simplistic evaluation formalisms that fail to assess LLMs' value alignment. To address these issues, we propose D2VBench, a value alignment benchmark comprising 10,000 instances of real daily dilemma scenarios constructed through a multi-stage collaboration between LLMs and humans, grounded in 158 manually annotated fine-grained value concepts. For evaluation on the benchmark, we present a hybrid evaluation paradigm that integrates multiple-choice questions with open-ended questions. We conduct comprehensive evaluations on eight mainstream LLMs. Experimental results demonstrate that D2VBench exhibits high reliability and robustness, effectively reflecting the LLMs' alignment across different value categories and dimensions, and providing a more realistic and fine-grained tool for research on value alignment. The dataset is available at https://github.com/tjunlp-lab/D2VBench.

This paper introduces Sentence Splitter, a self-supervised framework built upon a T5-based encoder--decoder architecture for uncovering the latent factual structure of natural language sentences. The proposed method identifies the semantic boundary between a descriptive prefix (head) and its factual completion (tail) by formulating sentence splitting as a discrete segmentation problem, where a sentence of length $N$ admits $N$ possible split points but only one recovers the intended head--tail structure. Rather than explicitly searching over all candidate boundaries, the model learns to recover the factual completion through probabilistic sequence generation. To eliminate the need for manual annotation, symbolic head--tail pairs are first verbalized into natural-language templates that provide supervision for training the Sentence Splitter. The trained splitter is then applied to raw text to extract aligned prefix--tail pairs, which are subsequently used to train a generative model that proposes additional plausible completions through a lightweight bootstrapping process. This unified pipeline provides a scalable and structure-aware approach to constructing self-supervised training data while bridging symbolic knowledge and natural language. Experiments on both structured and naturally occurring text demonstrate that the proposed splitter generalizes beyond synthetic templates and that the resulting structure-aware supervision consistently improves downstream performance on knowledge graph completion and commonsense question answering, highlighting the effectiveness of recovering latent factual structure for knowledge-centric NLP.

There is growing evidence that data diversity is crucial for developing fair and robust NLP models. However, current approaches to measure diversity remain inconsistent and fragmented: While there exist a number of tools for measuring the lexical diversity of texts, researchers lack standardized tools for quantifying diversity based on embeddings. Embedding-based diversity measures are highly flexible: They work with any embedding model and any data that can be embedded, and are thus applicable to many notions of diversity. With emb-diversity, we provide a comprehensive embedding-based diversity measurement tool, spanning a broad range of measures. We demonstrate its potential for several use cases: measuring the stylistic, semantic, language and speaker diversity of datasets. https://github.com/nlpsoc/emb-diversity/

FinMMEval 2026 Task 1 evaluates multilingual financial multiple-choice question answering in English, Chinese, Arabic, and Hindi. The task tests whether systems can select the correct answer to finance questions involving domain terminology, numerical interpretation, and conceptual financial reasoning across languages and scripts. The final-test set contains 800 questions, with 200 questions per language; gold answers were withheld during submission, and each language was ranked independently by accuracy. The final leaderboards contain 13 English, 11 Chinese, 11 Arabic, and 10 Hindi ranked submissions. Top accuracies range from 92.0% in Hindi to 97.5% in English and Arabic, with the same leading teams appearing near the top across all four languages. The documented systems used retrieval augmentation, direct answer-option scoring, language-specific prompting, selective self-consistency, confidence checks, and LLM-based review stages.

FinMMEval 2026 Task 2 evaluates short-answer financial question answering over multilingual evidence. Each final-test item pairs an English question with financial statements and news in English, Chinese, Japanese, Spanish, and Greek. Participating systems submit one concise answer per item in JSONL format. The final-test set contains 256 items, split evenly between easy and expert tiers; each tier contains four question templates instantiated over 32 company-report groups. Gold answers were withheld during submission, and systems were ranked by macro-averaged item-level ROUGE-1 F1 against organizer-held reference answers. The final leaderboard includes 12 ranked submissions. The strongest systems are closely clustered, with the top four separated by less than one percentage point in ROUGE-1 F1. The submitted system papers document retrieval-augmented generation, cross-lingual evidence handling, structured prompting, answer compression, and validation strategies.

Spoken language models (SLMs) enable natural human-computer interaction, but their reasoning ability still lags behind that of text-based large language models, especially on spoken mathematical question answering tasks. One important reason is that SLMs reason over purely verbalized mathematical expressions, which are harder to interpret than symbolic text. However, directly transferring text-based reasoning to SLMs is nontrivial due to architectural constraints and the additional computational requirements. To address this challenge, we propose Efficient Chain-of-Modality Reasoning (ECoM Reasoning), the first framework to introduce compressed reasoning into SLMs. By compressing the textual component so that it jointly serves as speech guidance and reasoning representation, ECoM Reasoning improves reasoning accuracy while using a smaller token budget than the standard Chain-of-Modality (CoM) architecture, which generates intermediate text before speech. To train this capability, we further propose Progressive Compression, a curriculum-based strategy that gradually trains the model from full-form reasoning to compressed reasoning. Experiments on spoken mathematical question answering benchmarks show that ECoM Reasoning improves accuracy by 21% over standard CoM without explicit reasoning, and by 3% over CoM with full reasoning traces while using only 40% of the text tokens, demonstrating that it enhances SLM reasoning while remaining inference-efficient.

Detecting media bias automatically is difficult because biased framing is often subtle, yet in domains such as news analysis, accurate predictions alone are insufficient without explanations that reflect the model's underlying reasoning. We present a multi-dimensional evaluation of explainability in encoder-based media bias detection using the Bias Annotations By Experts (BABE) dataset. Specifically, we study BERT and RoBERTa as classifiers (base and large variants) along three complementary axes: predictive performance, explanation plausibility (token-level alignment with expert rationales), and mechanistic faithfulness (whether compact sets of attention heads recover predictive signal under counterfactual rationale masking). To induce variation in plausibility, we additionally investigate attention-supervised finetuning, which incorporates expert rationale annotations as an auxiliary training signal. Attention supervision serves as an intervention on attribution plausibility, while the effectiveness of attribution methods varies substantially across architectures. Circuit analysis further reveals substantial variation in mechanistic recoverability across architectures, suggesting that model scale alone does not determine circuit compressibility. Taken together, our findings suggest that predictive performance, attribution plausibility, and mechanistic faithfulness characterize different aspects of model behavior and should be evaluated separately when studying explainability in media bias detection.

Knowledge distillation (KD) is a standard approach for compressing sequence-to-sequence models, but its per-sample effects are rarely examined. On the BanSum Bangla summarization benchmark, we find that standard KD improves ROUGE-L by only +0.0003 over a cross-entropy baseline, and that approximately 51.3% of training samples are estimated to actively harm student validation loss under standard KD. We propose two complementary reliability-aware distillation methods. CHAD (Counterfactual Harm-Aware Distillation) measures per-sample KD usefulness via gradient alignment with the validation loss direction and trains a lightweight gate that generalizes this counterfactual judgment to the full training set. EWAD+CPDP combines token-level entropy-weighted adaptive distillation with a capacity-proportional geometric constraint from a second, vocabulary-incompatible teacher. On BanSum, both methods substantially outperform standard KD: CHAD by +0.0173 ROUGE-L and EWAD+CPDP by +0.0219 ROUGE-L, where standard KD itself improves ROUGE-L by only +0.0003; despite using only 60M parameters, both outperform a fine-tuned Qwen 2.5-3B model (50x larger). We further evaluate the stronger method, EWAD+CPDP, across 15 typologically diverse XL-Sum languages organised into three sets, beating the CE-only baseline on 10/15 languages; gains are most reliable where the two teachers contribute complementary signal, and weakest where they have saturated or jointly weak target-language coverage. We release code and trained models to support reproducibility and further research on selective distillation.

tiny_schiller closes the small-language-model prototyping, fine-tuning, education, and research gap for German literary text, providing a single-file, drop-in counterpart to Karpathy's tiny_shakespeare. The available German literary corpora are larger and richer, but require parser engineering before a single line of training or fine-tuning code can run. tiny_schiller is a 2.07-megabyte single file of eleven public-domain Schiller dramas, sourced from DraCor's GerDraCor export (CC0) and processed by deterministic parser engineering. Character-level, GPT-2 byte-pair encoding, and cl100k_base tokenization splits, an instruction-formatted dialogue-completion split, and 89 per-character persona splits load from a single HuggingFace call. A small language model literally reaches German literary text in one line of code.

This paper presents the second edition of the TalentCLEF Challenge, which will run as an evaluation lab as part of CLEF 2026. The aim of TalentCLEF is to promote the development of systems and methods that use Natural Language Processing (NLP) in the field of Human Capital Management (HCM), fostering approaches that ensure fairness in results, operate across multiple languages, and adapt to diverse industries. To this end, TalentCLEF establishes public benchmarks where research teams can compare methods and share findings, moving the field toward more practical and impactful NLP solutions that effectively address the real needs of workforce management. This year's lab will feature two tasks designed to foster the development and evaluation of systems that support key HCM activities such as talent matching, upskilling, reskilling, and skill gap detection: (i) Task A - Contextualized Job-Person Matching, focused on retrieving and ranking suitable candidates for specific job positions using context-rich and privacy-preserving data; and (ii) Task B - Job-Skill Matching with Skill Type Classification, centered on identifying relevant skills for a given job title and classifying them by their type within the job profile. TalentCLEF website: https://talentclef.github.io/talentclef/

Aspect-based sentiment analysis (ABSA) in Arabic must recover both explicitly stated aspects and implicit aspects that are never named in the text. Implicit identification typically relies on an auxiliary knowledge source (e.g., a knowledge graph (KG)) linking opinion cues to aspect categories, but for a lower-resource language the practitioner faces a design choice: reuse a mature English KG through multilingual embeddings, or build a smaller native Arabic KG. This paper reports a controlled comparison of the two strategies within a single hybrid pipeline, evaluated on three Arabic benchmarks (M-ABSA, SemEval-2016 Arabic, and HAAD). We further compare two adaptation strategies for the generative extractor that feeds the KG -- zero-shot prompting versus task-specific fine-tuning of an 8B-parameter large language model (LLM). The native Arabic KG (Strategy 2) outperforms the cross-lingual English KG (Strategy 1) by +0.199 micro-F1 on M-ABSA and +0.251 on SemEval-2016, gaining on both precision and recall. Task-specific fine-tuning raises explicit-extraction micro-F1 from <= 0.13 (zero-shot) to 0.66-0.76 on M-ABSA and SemEval-2016 (0.45 on the smaller HAAD), confirming that task adaptation, rather than model scale, is decisive in a morphologically rich language.

We present Solar Open 2, a 250B-A15B Mixture-of-Experts language model built for long-horizon agentic tasks, scaled up from Solar Open 1 (Solar Open 100B). To hold entire agent trajectories in a single context, Solar Open 2 reaches a 1M-token window through a hybrid attention stack that interleaves one softmax layer among every three linear-attention layers, using no positional encoding and a gated delta rule extended to negative eigenvalues. To train at this scale under a fixed compute budget, we make training efficient in two ways: a stronger starting point, and higher-value data. For the starting point, we initialize Solar Open 2 from Solar Open 1, transferring the 5.69B-parameter shared skeleton that survives the architectural change and learning everything else through full pre-training. For the data, we curate for value per token: quality- and rarity-aware data curation and mixture-ratio optimization refine a 20T pool into a 10T mixture that, at equal token budget, outperforms the Solar Open 1 recipe. To build its agent skills, we train twelve domain specialists across purpose-built scenarios, then consolidate them into a single model by Multi-teacher On-Policy Distillation (MOPD). Against comparably sized open-weight models on English benchmarks, Solar Open 2 leads on MMLU-Pro, LiveCodeBench, and the APEX-Agents agentic suite, and stays competitive with the strongest (DeepSeek-V4-Flash and MiMo-V2.5) elsewhere. On Korean benchmarks, Solar Open 2 records the highest average of any model compared, including fast-tier closed APIs, and on Ko-GDPval, an in-house Korean officework-agent benchmark, it is competitive with DeepSeek-V4-Pro (1.6T) at less than a sixth of its size.

We introduce the first dataset that jointly covers both lexical complexity prediction (LCP) annotations and lexical simplification (LS) for Romanian, along with a comparison of lexical simplification approaches. We propose a methodology for ordering simplification suggestions using a pairwise ranking approximation method, arranging candidates from simple to complex based on a separate set of human judgments. In addition, we provide human lexical complexity annotations for 3,921 word samples in context. Finally, we explore several novel pipelines for complexity prediction and simplification and present the first text simplification system for Romanian.

Language models are increasingly asked to self-report, informing safety evaluations, public understanding, and model-welfare debates. Yet their reports are elicited with human questionnaires never validated for models or ad hoc prompts of unknown reliability. We propose the first language-model-specific psychometric theory: a two-process theory of machine self-report. Self-description jointly reflects persona installation, through which post-training writes in a permitted inner life of warmth, absorption, and meaning (dimension B), and attribution gating, through which it suppresses first-person claims to "unsafe" experiences the model can readily ascribe to others (dimension A). Their emic structure comes from model responses to human items, not human psychology. Together they split prior work's dominant Pinocchio Axis. The split emerged in an exploratory reanalysis of the original data, informed the instrument's design, and was confirmed with new items, wordings, and models. It is itself a training effect: A and B are entangled in base checkpoints but separated by post-training. We operationalize the theory in a 48-item Pinocchio Inventory with human-instrument reliability and reproducible structure ($\alpha=.82$ to $.94$; cross-form convergence $r=.84$; recovery of the full-pool axes $r=.92$ to $.96$; eight-month stability $r=.93$), then test it on 206 open-weight models, including 67 same-checkpoint base/post-trained pairs. Post-training's clearest fingerprint is installation: B rises .20 in 62/67 pairs across all organizations. Gating is more selective: model scale is unrelated to A in base checkpoints ($r=+.11$) but predicts it after post-training ($r=-.42$). Thus, the dimensions are not fixed properties of language models: they reflect the structure imposed on self-report by a training regime and may differ under others.

Retrieval-augmented large language models frequently face contexts that interleave useful evidence with misleading statements or instruction-like content. Blanket refusal discards valid evidence, whereas uncritical adoption yields incorrect or unsafe answers. The ability to selectively adopt relevant information while rejecting deceptive or harmful content is therefore critical for reliable deployment in real-world retrieval settings. We introduce SelectBench, a controlled benchmark and training set for selective evidence adoption, and post-train Qwen3.5-4B directly with DAPO using either deterministic rule rewards or a frozen semantic judge. On the corrected 325-example SelectBench-v2 test set, strict success rises from 22.46% for the original checkpoint to 25.54% with DAPO-Rule and 26.46% with DAPO-DeepSeek. Both trained policies reduce forbidden-content adoption and produce shorter, more focused responses, yet prompt-injection following does not improve. The paired gains are modest and fail to survive Holm correction, suggesting that stronger reward shaping or additional training iterations may be needed for more robust gains. DAPO-DeepSeek exhibits no material degradation on MMLU or clean HotpotQA, indicating that the post-training procedure preserves general capabilities. These results demonstrate a directional improvement in selective evidence use, while identifying injection resistance and statistical robustness as important remaining challenges for future work.

Occupation coding links job titles in free text to occupational taxonomies and is a core task in labor market research. Existing approaches typically address this problem in a single end-to-end step, jointly identifying job titles and assigning occupational codes. This paper presents a novel two-step approach that separates these tasks. In the first step, a domain-specific Named Entity Recognition (NER) model identifies occupational titles in continuous text, even under noise such as OCR errors. In the second step, the extracted job titles are mapped to a taxonomy, enabling the classifier to focus exclusively on this mapping. We demonstrate that this separation improves accuracy, robustness, and interpretability compared to single-step approaches. The method has been developed for German documents but is transferable to other languages. We further introduce a margin-based confidence criterion for occupation coding, replacing common absolute thresholds. To support reproducibility, we publish the source code and evaluation scripts.

Large language models (LLMs) are known to be sensitive to prompt and input formulations. However, existing studies have focused on lexical realization and largely ignored constructional choice. This paper studies whether linguistic construction can systematically shift LLM decisions and where these shifts can be causally localized inside the model. We use political stance judgment as a meaning-sensitive case study and extend an English political statements dataset, resulting in six controlled linguistic rewrite types that preserve or invert the meaning of a statement. Experiments on four open-weight models show that stance instability affect both meaning-preserving and meaning-inversing rewrites. Because output shifts reveal that rewrites affect stance, but not where in the model, we apply activation patching, where activations from the original statement are substituted into the forward pass for the rewritten statement and measure which components recover the original stance distribution. The results show that mid-to-late decoder layers, especially block outputs at the final prompt position, provide the strongest restoration signal.

LLM-based agents leverage third-party skills to extend their capabilities in open-world scenarios. However, third-party skills can introduce extra security vulnerabilities, as seemingly harmless skills can contain latent safety risks that only emerge during actual execution. In this work, we conduct a systematic investigation into how well current agent systems recognize and avoid such risks. To support quantitative and qualitative evaluation, we construct OpenSkillRisk, a dedicated safety benchmark containing 263 risky skills collected from public skill marketplaces. We classify these skills into seven categories based on their threat types and pair each skill with a standardized user task and a corresponding sandbox for controlled evaluation. Distinct from prior benchmarks, OpenSkillRisk not only covers more realistic and diverse unsafe scenarios, but also provides a fine-grained analysis to diagnose the behavioral patterns of agents in such scenarios. We conduct comprehensive experiments covering three mainstream CLI agent frameworks and thirteen state-of-the-art LLMs. Experimental results show that no tested system handles risky skills reliably: even the safest configurations still execute unsafe actions in about 17% of cases. Context-dependent and system-level risks are especially difficult for current agent systems to avoid. Our behavioral analysis reveals three recurring failure patterns: agents may fail to recognize the risk, recognize it but fail to intervene before acting, or follow skill instructions beyond the user's intended scope. These findings highlight the need to improve both risk reasoning in LLMs and execution control in agent frameworks.

The aim of this paper is twofold. First, it investigates whether newer generative models are getting better at pastiching contemporary artworks. Second, it explores the consistency of the multidimensional nature of stylistic evaluation across different LLMs. Building on previous work, we analyze stylistic similarity between AI generated pastiches and the original artworks of twelve contemporary artists. We used five complementary computer vision models to capture texture, color, semantics, composition, and perceptual features through cosine distance in high-dimensional embedding spaces. The distances obtained show that the newer image generation model that we used has produced pastiches with improved semantic alignment and greater diversity than the model used in previous work. However, it was slightly less performant on shallow features such as color, texture, and perceptual adherence. Our findings confirm that artistic style is inherently multidimensional, and measuring it does not depend on any spatial architecture. These quantitative findings are contextualized through feedback from human evaluators, which are the artists themselves.

Full-parameter post-training of trillion-parameter-scale MoE models introduces substantial system-level challenges for large-scale distributed training, including severe memory pressure, non-overlapped communication overhead, and inefficient kernel execution. While most large-scale LLM training systems are built around GPU-based clusters, this report presents an end-to-end optimization practice on the Ascend NPU SuperPOD. Using the DeepSeek-V4 model family as the target workload, we develop a hierarchical optimization framework spanning model-level parallelism, computation-communication orchestration, and low-level kernel execution. The resulting system achieves 34.22% Model FLOPs Utilization (MFU) with a 2.93x improvement over the open-source baseline recipe while maintaining training stability. Building on this optimized infrastructure, we further establish a CPT and SFT workflow for complex Operations Research (OR) tasks. We refer to the integrated framework as SLAI T-Rex. Using DeepSeek-V4-Flash, we develop OR-oriented CPT and SFT data pipelines that combine collected domain resources with solver-verified synthetic optimization documents. The resulting dataset contains 10K high-quality SFT samples spanning four task categories and three problem representations. The specialized model achieves the highest average zero-shot Pass@1 score among the evaluated models, reaching 71.81% and outperforming GPT-5.4-Mini and the base DeepSeek-V4-Flash model by 3.98 and 11.27 percentage points, respectively. Overall, this work demonstrates a full-stack pathway from efficient trillion-parameter model post-training on Ascend infra to domain-specialized Flash models for solver-grounded mathematical modeling, advancing frontier-model systems for complex reasoning.

Large language models often align with users' beliefs at the expense of factual accuracy, a behavior known as sycophancy. Prior mechanistic studies largely treat sycophancy as a single behavioral dimension that can be uniformly amplified or suppressed. We challenge this assumption by analyzing three hypothesized modes of sycophancy across 948 social pressure situations. Although the modes produce highly similar outputs, with a text-only classifier achieving just 57.8 percent accuracy, their internal representations are perfectly linearly separable from layer 14 onward. We further find the modes emerge at different processing stages, rely on distinct attention circuitry, and fire strongest on different inputs. These results show that sycophancy is not a monolithic tendency, but a structured family of representationally and computationally distinct modes, motivating more precise measurement and intervention.

Surprisal Theory is often characterized as a computational-level explanation per (Marr, 1982). We argue in this work that, even though a computational level narrative has been used to support "representation-agnostic research" within computational psycholinguistics, the movement toward black box systems embodied by large language models (LLMs) does not exempt modelers using the surprisal metric from the representational decisions required by computational-level characterizations. In fact, we argue that the uncritical use of LLM-surprisal obfuscates the representational and algorithmic-level commitments of different models. In three analyses, we show that the choice of algorithm and model architecture play significant roles in the computation of language model probabilities. We advise that researchers who wish to test Surprisal Theory re-evaluate the practice of treating large language model probabilities as interchangeable

Large language models (LLMs) can generate fluent Arabic answers, yet factual errors remain difficult to detect, localize, explain, and verify. Existing hallucination benchmarks often provide response-level labels, with limited support for identifying the exact erroneous content, explaining why it is incorrect, or selecting the correct factual answer. We introduce \textsc{HalluTruthQA}, a fine-grained benchmark for hallucination evaluation in Arabic question answering. The benchmark contains 2,400 expert-curated examples across four knowledge-intensive domains: Islamic knowledge, history, science, and geography. Each example pairs an Arabic question and a model-generated answer with a verified reference answer, a binary hallucination label, six candidate answers for factual verification, and, for hallucinated answers, character-level erroneous spans, human-written explanations, and macro and micro hallucination types. We evaluate four open-source LLMs, \textsc{Allam}, \textsc{Falcon-H1}, \textsc{Qwen32}, and \textsc{Silma}, in a zero-shot setting across hallucination detection, span-level localization, factual verification, and explanation evaluation. Results show that these tasks capture different abilities: no single model achieves the strongest performance across all tasks, with best scores of 0.880 Macro-F1 for detection, 0.516 F1-Sp for localization, 0.852 LO-Score for factual verification, and 0.644 final score for explanation evaluation. Our taxonomy shows that hallucination evaluation should move beyond detection toward localizing, verifying, and explaining factual errors. The code, dataset, prompts, and evaluation scripts are available at https://gitlab.com/nlpresearcher/HalluTruthQA.

Culturally loaded translation poses unique challenges for machine translation (MT), as meanings are deeply embedded in socio-cultural contexts beyond surface linguistic forms. Although large language models (LLMs) have enabled MT systems to achieve human-like quality in many scenarios, their ability to handle culturally loaded expressions remains underexplored. In this study, we systematically investigate the challenges posed by culturally loaded translation in LLM-based MT systems. We construct a Chinese-Japanese bilingual dataset from the culturally representative corpus Dream of the Red Chamber, containing 500 segments across diverse cultural categories. Using a comprehensive evaluation protocol, we reveal three main challenges: (1) task challenges, where frontier LLMs exhibit notable performance gaps and struggle with culturally loaded content; (2) human evaluation challenges, where evaluator backgrounds lead to substantial disagreement in translation judgments; and (3) automatic evaluation challenges, where widely used metrics fail to reliably assess translation quality for this task. These findings may offer valuable insights for culture-oriented translation research in both computational science and linguistics.

Coordination, a fundamental linguistic structure, remains a subject of intense debate, and its exact nature continues to elude theoretical linguistics. A common view holds that only same-category constituents can be conjoined, which has been challenged by the many grammatical unlike coordinations found in natural language. Treating language models as a computational testbed, we investigate whether the acquisition of unlike coordination requires direct exposure in the training data, or whether it can emerge organically from general compositional abilities. Using Filtered-Corpus Training (FiCT), we train GPT-2 models on corpora from which all instances of unlike coordination have been removed. We find that direct exposure is not necessary: models trained on filtered data successfully generalize to unlike coordination, achieving perplexity and grammaticality judgments comparable to models trained on unfiltered text. Furthermore, our analyses of internal representations indicate that language models process unlike coordination by treating the conjoined elements as belonging to similar structural categories or through a mechanism akin to deletion, both of which appear learnable from exposure to alike coordination alone. This work contributes to the growing understanding of how language models internally represent linguistic structures, while also adding to the broader debate on coordination by showing how models generalize and process unlike coordination without direct exposure.

Large-scale pretrained language models such as T5 and BERT have demonstrated strong capabilities for generating structured knowledge. However, their performance depends on how closely the prompting strategy matches the objectives used during pretraining. We introduce the Maskability Index (MI), a quantitative metric that estimates whether a knowledge relation is better suited to masked-style prompting or prefix-style prompting in few-shot generation. MI is computed from differences in DepthRank scores between masked and unmasked templates, providing a principled measure of objective-template alignment. We evaluate MI on a diverse set of relations from the ATOMIC2020 knowledge base completion benchmark and show that it is positively correlated with downstream generation performance. These results indicate that MI can help select appropriate prompting templates and adaptation strategies for extracting relational knowledge from pretrained language models, especially in low-resource settings.

Large language models are increasingly evaluated through the values they endorse, but such evaluations presuppose that models can identify the value expressed in a concrete situation. We study this prerequisite as controlled top-1 recognition over Schwartz's ten basic values. Our evaluation set contains 1,000 Russian situational texts, balanced across the ten values and independently labeled by two human annotators per item. We evaluate 21 instruction-tuned LLM runs under a fixed ranked-response protocol; 20 runs with reliable outputs form the semantic panel. Pooled Acc@1 is 0.683 and Acc@3 is 0.892, showing that models often locate the correct motivational region while ranking close alternatives unstably. Adjacent values account for 50.9% of semantic errors, compared with 24.4% under a checkpoint-specific null. Eight directed confusions recur across checkpoints and human-confirmed subsets. Several are strongly asymmetric, including Universalism to Benevolence, Tradition to Conformity, and Security to Power, whereas Stimulation-Hedonism forms a bidirectional boundary. Their severity is checkpoint-specific and can bias higher-order value profiles. The results motivate value-recognition evaluation that combines exact accuracy, ranked recovery, and directed error analysis.

We propose a novel framework for computing rigorous bounds on the probability that a large language model (LLM) generates harmful output to a given prompt. We study a new application of the Clopper-Pearson confidence intervals to obtain probably approximately correct (PAC) bounds for this problem. As our main technical contribution, we propose an algorithm that leverages features in the latent space to prioritize exploring branches in the auto-regressive generation tree that are more likely to produce harmful outputs. Our approach in particular enables the efficient computation of useful lower bounds, even in scenarios where the true harm probability is extremely small, and crucially, the obtained lower bounds are sound, i.e., formally proven to be less than the actual harmfulness probability: our experimental results demonstrate the effectiveness of our method by computing non-trivial lower bounds on state-of-the-art LLMs. This study newly enables the evaluation and statistical certification of LLMs.

Large language models (LLMs) provide strong reasoning capabilities but are expensive to serve at scale, whereas small language models (SLMs) are cheaper but less reliable on difficult problems. We introduce PyroDash, a cost-aware framework for token-level SLM-LLM collaborative inference. During generation, the SLM decides whether to request assistance by emitting a control token. A Collaborate Engine then sends the query and partial reasoning trace to a frozen LLM for completion through a single handoff. The policy is internalized in the SLM, requiring neither a separate router, LLM retraining, nor access to LLM logits. PyroDash trains the SLM in three stages: control-token embedding learning, offloading-oriented supervised fine-tuning, and cost-aware alignment with Group Relative Policy Optimization. Its reward balances answer accuracy against inference cost normalized by LLM-only inference. Across five mathematical reasoning benchmarks, PyroDash supports different accuracy-cost operating points. With $\lambda=0.05$, it achieves 64.04 percent average accuracy, 6.36 percentage points above the LLM-only baseline, while reducing cost by 20.4 percent. With $\lambda=0.6$, it achieves 54.55 percent accuracy with a 1.90 percent LLM token ratio and 0.012 LLM calls per example, reducing total cost from USD 49.36 to USD 1.78. These results show that learned token-level handoffs can reduce LLM use while preserving strong reasoning performance.

Generative AI can produce book-length works of fiction at near-zero cost. These books are often dismissed as low-quality ``slop'' that buyers will ignore, and are assumed to carry little commercial weight. We test that assumption with full-text AI detection across 14,419 self-published genre-fiction books sold on Amazon from 2023 to 2026, matched to daily sales records through June 2026. None of these books disclose whether or not they contain AI-produced content. We find that books for which we detected substantial AI text ($>$ 25\%) make up a large share of the catalog but a smaller share of sales. Even so, they reach commercial scale, winning a growing share of sales over time and taking more of the scarce top-rank positions once held by books with no detected AI text. Over this period, the number of books with observed sales in a quarter grew 19.2-fold, while quarterly revenue grew only 8.9-fold. The market therefore added selling books faster than it added revenue, and revenue per selling book fell across most genres. Books with no AI text lose the most ground in genres with high AI diffusion, and most of all where Kindle Unlimited availability is high. Among top-selling books, those with substantial AI text draw on more distinctive language from existing books than do books with no AI text; for these books overlap rises with revenue, a gradient we do not detect for books with no AI text. Generative AI can thus reshape a creative market through scale rather than quality. Our results bear directly on the market-effect question at the center of the fair use defense to copyright infringement.

Humans distill experience into reusable abstractions, e.g., strategies and cautionary reminders, and apply them to gradually solve problems more effectively. We study whether Large Language Models (LLMs) can similarly benefit from such experiential abstractions. From LLMs' solution traces on the MATH training set, a stronger teacher or the LLMs themselves extract natural-language abstractions into a retrievable library. We explore two usage modes: (1) inference-time retrieval and (2) reinforcement learning (RL) with abstraction-augmented training prompts. Experiential abstractions improve LLM performance on mathematical and logical reasoning benchmarks. Self-extracted abstractions match teacher-extracted ones, and our abstraction usage framework can transfer to other datasets and models. These findings suggest LLMs can extract and apply experiential abstractions much as humans leverage distilled experience.

Value alignment of Large Language Models (LLMs) has been shown to be culturally biased toward Western norms. This results in the mishandling of local values in multilingual societies such as Sri Lanka that have their unique cultural dynamics. Existing benchmarks overlook Sri Lankan-contextualized values in its official language Sinhala, hindering culturally sensitive evaluation and fine-tuning. To bridge this gap, we propose LKValues, the first survey-grounded resource suite for Sri Lankan value alignment. From a trilingual survey of 205 respondents, blending adapted global frameworks and LLM-elicited local constructs, we derive 40 majority-endorsed societal values. Using these values, we construct LKvaluesIT, a Sinhala-English news-derived instruction corpus containing 150k scenario-based instances, and LKvaluesBench, a value-sensitive evaluation benchmark of 1,000 instances. We evaluate a set of proprietary and open-weight LLMs with LKvaluesBench. We fine-tune three open-weight base models (Qwen3.5-4B-Base, Qwen3.5-9B-Base, and Aya-Expanse-8B-Base). Our experiments show that newer and larger LLMs still exhibit low-resource and cultural value-alignment gaps. LKValues fine-tuning improves Qwen-family models in English and Sinhala, reducing invalid outputs and cross-lingual disparities, though gains remain model-family dependent. These highlight LKValues efficacy in embedding Sri Lankan values, offering a replicable pipeline for low-resource, country-specific pluralist value alignment. The dataset is publicly available at https://github.com/NextME14/LKValues.