arXiv Papers

The human-like morphology of humanoid robots grants them exceptional potential for agile and versatile motor capabilities, but it also introduces significant challenges in acquiring complex skills. Traditional Learning-from-Demonstrations methods are often constrained by the high cost of collecting real-world data, the difficulty of capturing motion-specific behaviors, and the limited diversity of demonstrations across individuals. Moreover, even for the same task, humans may execute the motion in multiple distinct ways. In this paper, we propose a new framework that leverages the power of Generative AI to convert textual prompts into realistic and diverse sequences of human body movements, enabling the robot to observe multiple variations of how a single task can be performed. These synthetic demonstrations are then used as a training resource, allowing the robot to learn a broad range of task-execution styles without requiring direct human intervention. We evaluate the proposed method across four simulation scenarios. Experimental results show that the robot not only completes the tasks successfully but also demonstrates strong adaptability to complex variations in motion.

Robotic learning takes place in dynamic environments with large behavior spaces. A terminal success signal only tells the robot whether the task is completed. It does not explain whether the current behavior is making progress, remaining unchanged, or undoing earlier progress. For this reason, recent studies have increasingly explored progress rewards that provide feedback during task execution. However, the current literature lacks a shared framework. Existing methods use different observations, goal specifications, output signals, supervision sources, and evaluation protocols. This makes it difficult to compare them and understand what their results actually validate. In this survey, we provide a unified view of progress reward modeling for robotic learning. We organize the field in three connected steps. We first study the interface of a progress model. This defines the problem from the outside by asking what information the model receives and what form of progress signal it produces. We then move inside the model and study the methods used to construct this signal. This reveals the different assumptions and mechanisms behind progress estimation and reward generation. Finally, we examine the data and benchmarks that support these methods. This shows how progress supervision is obtained and what different evaluations actually measure. Together, these three perspectives connect what a progress model is, how it is built, and how its quality is validated. We further summarize the main limitations of current approaches and discuss future research directions.

Diffusion policies generate multimodal robot action sequences from demonstrations, but steering them toward deployment-time constraints typically relies on differentiable guidance costs. This excludes many practical safety constraints, such as binary collision checks, joint limits, and black-box rollout costs that are nondifferentiable. We propose Gradient-free Robot Action generation via Combined diffusion-MPPI posterior mean Estimation (GRACE), which guides a pretrained diffusion policy with Model Predictive Path Integral (MPPI) control using only forward cost evaluations. Building on the common score-ascent structure of diffusion and MPPI, GRACE constructs a cost-conditioned guidance posterior at each reverse step and estimates its mean with a single MPPI update centered at the diffusion reverse mean. For differentiable costs, GRACE recovers conventional gradient guidance under a first-order, matched-covariance approximation. GRACE attains higher success rates than diffusion-based and sampling-based baselines in simulation. On a real 7-DoF manipulator, GRACE avoids a deployment-time obstacle that the unguided prior collides with in every trial. Code and experiment videos are available at https://anonymous.4open.science/w/grace-70BB/.

Action tokenization maps continuous robot action chunks to discrete tokens and has become an important interface for modern visuomotor policies. Existing approaches either rely on analytical discretization methods that produce prohibitively long token sequences or learned latent tokenizers that lack structure, limiting their compatibility with downstream policies. In this work, we identify three desiderata for action tokenization - high compression, total decodability, and an ordered token space - and introduce Ordered Action Tokenization (OAT), a learned action tokenizer that satisfies all three. OAT discretizes action chunks into an ordered sequence of tokens using a transformer with registers, finite scalar quantization, and ordering-inducing training mechanisms. By training each token prefix to decode into a valid action chunk, OAT places coarse control information in early tokens and uses later tokens to refine residual detail, yielding an anytime tradeoff between inference cost and action fidelity. We validate OAT in two prevailing uses of action tokens: autoregressive policies that generate tokens for control, and token co-training policies that use token losses to shape the vision-language model context consumed by a flow-based action expert. Across three policy backbones and more than 60 tasks spanning five simulation benchmarks and real-world settings, OAT consistently delivers strong policy performance while offering significantly greater flexibility at inference time.

General-purpose robots need to reason about their actions, combining perception, world knowledge, planning, success detection, recovery, and low-level control. Today's state-of-the-art models attempt to combine all these capabilities into the learned policy via large-scale pre-training. Instead, we show that these capabilities can be decomposed into a general language-conditioned policy/control agent and a high-level agent manager/orchestrator. Rather than training policies to reason via pre-training, we build a closed-loop physical agent orchestrator that can do high-level planning, decompose the goal into achievable subgoals, command low-level motor commands, track and verify the outcome from low-level observations, and recover from failures. Our Physical Agency orchestrator (Pigey) can control existing vision-language-action (VLA) policies as well as parametrized skills to solve complex reasoning tasks in the real world, without any additional data collection or post-training. We evaluate Pigey extensively across simulation benchmarks and challenging real-world robotic manipulation tasks, and demonstrate significant performance improvements over existing generalist policies. On LIBERO-PRO, Pigey advances the state-of-the-art by over 4x (12.8% -> 53.3%) with no task-specific fine-tuning. On a real robot, Pigey lifts the frozen policy from near-zero to over 90% on reasoning-limited tasks. We call the difference between what frozen motor skills achieve alone and inside the agentic loop the orchestration gap.

Social appropriateness in human-robot interaction (HRI) is not universal: different people can judge the same robot action differently in the same situation. To capture this inter-subject variability, we reformulate socially appropriate action generation as a preference modelling problem inspired by recommender systems, treating annotators as users, contexts/scenes as items, and appropriateness scores over a set of candidate robot actions as targets. We propose StARS, a novel model-agnostic framework that integrates collaborative filtering with learnable scene representations to generate user-specific appropriateness scores over candidate robot actions. StARS is model-agnostic: it can be integrated with various scene encoders and backbones, enabling personalisation without redesigning the underlying model. We evaluate StARS on two socially aware robotics datasets, MannersDB+ and SocNav1, and analyse robustness under sparse preference feedback. Across datasets and backbones, StARS consistently improves performance and agreement with annotators, supporting personalised action selection aligned with user norms. Our code is publicly available at https://github.com/Cambridge-AFAR/StARS.git.

We present an action-conditioned world model framework for goal plane probe guidance in robotic ultrasound, with a focus on neck ultrasound scanning. Autonomous ultrasound tasks often require large numbers of probe-motion trajectories for training, but collecting high-quality demonstrations is labor-intensive and explicit simulators are difficult to build because ultrasound appearance depends on contact, tissue deformation, and view-dependent acoustic artifacts. We address this problem with a two-stage model-based learning pipeline. First, a latent conditional diffusion world model predicts future ultrasound observations from recent context frames, probe motions and temporal offset. Second, a goal-conditioned temporal transformer predicts ordered probe motions and is fine-tuned using rewards from the frozen world model. Experiments on the self-collected dataset show that the world model preserves action-dependent anatomical structure on target-directed scans. In real-world closed loop experiments, the framework achieves success rates of 70.0\% for carotid guidance and 65.0\% for thyroid guidance. These results demonstrate the potential of learned ultrasound dynamics for training goal-directed robotic probe navigation.

This paper demonstrates how deep reinforcement learning (DRL) enables adaptive locomotion of snake-like robots in dynamically changing viscous environments, overcoming the inherent performance limitations of classical predefined control methods. The lack of direct onboard sensors for fluid properties necessitates formulating this task as a partially observable Markov decision process. By employing an asymmetric actor-critic framework, a teacher policy trained using privileged information available only in the physics simulator distills its knowledge into a student policy that relies solely on proprioceptive sensor information. Simulation results across a wide range of dynamic viscosity changes ($10^{-7}$ to $10^{-2} m^2/s$) reveal that the DRL agent autonomously acquires non-sinusoidal adaptive gaits. These gaits improve propulsion velocity and transport efficiency, breaking the inherent limits of conventional sinusoidal and kinematic control. The findings establish that implicit environment inference via privileged information distillation is an effective approach to bypass the constraints of classical models under unpredictable fluid dynamics.

Understanding how robots and humans move in shared spaces is essential for designing effective social robot navigation policies and predicting human behavior. However, existing datasets often lack the diversity needed to capture differences in culture, geography, and human-robot interaction-factors that strongly shape appropriate social behavior. To address this gap, we introduce ACME: A Cross-cultural, Multi-Embodiment dataset for social navigation. A large-scale data collection effort across 8 sites in 5 countries, using 7 robot embodiments, ACME is a large and diverse multi-modal dataset aimed at advancing social navigation research, providing 29.35 hours of onboard robot data and 43.5 hours of overhead pedestrian tracking data. Unlike prior datasets, it focuses on capturing goal-driven social navigation behavior in complex social scenarios with explicit robot-crowd interaction through robot speech. To facilitate learning navigation policies and predicting pedestrian trajectories, ACME provides 3D and 2D scene features, odometry, interaction information, and human-annotated pedestrian trajectory labels. We make ACME easy to use by providing both human-readable data for each sensor modality as well as raw binary data. Our qualitative and quantitative analyses show that our dataset captures more challenging scenarios and a broader distribution of pedestrian behavior than previous datasets.

Geomagnetic sensing offers an infrastructure-free, absolute orientation reference that is robust to GNSS denial and visual degradation, yet no large-scale outdoor robotics dataset supports its systematic study in SLAM. Existing magnetic datasets are confined to small-scale indoor environments and lack the synchronized multi-modal sensing, repeated-traversal structure, and high-precision 6-DoF ground truth required for geomagnetic SLAM research. We present Mag4D-SLAM, the first large-scale outdoor geomagnetic SLAM dataset. It comprises 14 sequences totaling over 18 km of synchronized LiDAR, camera, IMU, tri-axis magnetometer, and GNSS measurements with SE(3) ground-truth poses, collected along structured campus trajectories under paired day/night conditions in both forward and reverse directions. Through repeated-traversal experiments, we analyze three core properties: magnetic field repeatability across different recording sessions (daytime and nighttime), drift-free global heading estimation, and location-discriminative magnetic signatures for cross-session place recognition. Mag4D-SLAM is designed to support research on yaw drift mitigation, magnetic loop closure, and long-term localization and to open new research questions on how geomagnetic sensing can complement visual and LiDAR modalities or provide a fallback cue under illumination changes, structural repetition, and GNSS-denied long-term operation.

Learned manipulation policies increasingly predict motions for abstract "hands" and are attractive in practice because they rely on easily collected demonstrations and transfer across robot platforms. Executing these trajectories on multi-arm robots, however, is not trivial. Multi-hand policy outputs must be assigned to physical arms, each arm must realize a configuration-space motion that tracks its prescribed end-effector trajectory, and all arms must respect kinematic limits and avoid collisions. In the absence of algorithms that directly address this problem, practitioners typically extend single-arm inverse-kinematics (IK) pipelines in an ad hoc way, with no guarantees of feasibility or safety. In this work, we close this execution gap with a search-based framework that is theoretically complete for grounding policy-generated multi-hand trajectories onto physical multi-arm systems. Building on Conflict-Based Search, our method explicitly searches over both the discrete assignment of trajectories to arms and the continuous Jacobian null spaces of redundant manipulators, using redundancy to avoid inter-arm collisions while tracking the prescribed motions. This unified treatment of assignment and null-space motion yields a practically efficient planner that safely realizes coordinated manipulation-policy outputs on multi-arm robots. See omcbsa.github.io for more.

Ship-borne manipulators operating in maritime environments are subject to stochastic wave-induced base motions that introduce kinematic disturbances and dynamic coupling, degrading trajectory tracking accuracy and complicating safe, contact-rich manipulation. This paper proposes a torque-level optimization-based control framework that integrates high-precision trajectory tracking with task-space impedance for ship-borne manipulators. The controller is formulated using task-space inverse dynamics (TSID) and solved via quadratic programming to explicitly compensate for the dynamic coupling introduced by base motion. To enable accurate feedforward compensation, an error-state Kalman filter (ESKF) is developed to estimate the base state by fusing inertial measurements with end-effector pose feedback. The framework is validated in simulation and real-world experiments using a 7-DOF manipulator mounted on a 6-DOF Stewart platform. The proposed method reduces real-world end-effector position tracking error by over 25.7% compared with the best baseline. Furthermore, the controller enables dynamic peg-in-hole insertion with 1~mm clearance under base motion, increasing the success rate while reducing average contact forces by 45%, demonstrating precise and compliant manipulation in contact-rich environments.

Multi-stream robot manipulation policies achieve unparalleled sample efficiency and generalization by modeling actions relative to environmental reference frames. However, existing approaches typically assume these frames to be strictly exogenous. This causal assumption collapses in dynamic settings, such as when a single robot arm manipulates a moving object or when two arms coordinate, where each arm effectively becomes part of the dynamic environment of the other. We propose DynaMAC, a lightweight, policy-agnostic framework that resolves this causal limitation while preserving the sample efficiency, computational speed, and flexibility of multi-stream policies, DynaMAC treats the opposite arm as a dynamic task parameter, thereby providing a unified formulation for dynamic manipulation and bimanual coordination without requiring an explicit leader-follower relationship. To rigorously evaluate these capabilities, we introduce DynaBench, a novel benchmark for robot manipulation in dynamic environments. Across both dynamic environments and bimanual manipulation tasks, DynaMAC outperforms leading probabilistic and generative baselines by over 35 percentage points while requiring 20 times fewer samples. Crucially, DynaMAC generalizes zero-shot from static demonstrations to dynamic environments, substantially simplifying data collection and establishing an elegant bridge toward human-robot collaboration.

Visual inertial odometry (VIO) is essential for accurate 6-DoF motion estimation in mobile robotic systems. Recent learning-based VIO methods have shown promising progress, but they often rely on unified visual--inertial representations and a single temporal model for full-pose estimation, limiting their ability to capture the heterogeneous dynamics of rotation and translation. Moreover, monocular visual features often lack explicit geometric structure, while raw inertial encoding leaves the underlying rotational kinematics implicit, weakening the rotation-related cues in IMU features. To address these issues, we propose DB-VIO, a dual-branch visual inertial odometry framework with enhanced visual--inertial representation. DB-VIO incorporates depth cues to improve monocular visual perception, injects an explicit integrated-attitude prior to strengthen rotation-aware inertial representation, and decouples pose estimation into dedicated rotational and translational branches for motion-specific temporal modeling. Experiments on autonomous driving and aerial robot benchmarks show that DB-VIO achieves state-of-the-art performance, improving the corresponding baselines by 20\% on KITTI and 33\% on EuRoC. Notably, under the more agile motion patterns of EuRoC, DB-VIO improves the rotational metric by 65.7\% over prior methods. These results demonstrate the effectiveness and generalization of DB-VIO across different platforms and motion scenarios.

Open-source LiDAR-inertial odometry (LIO) systems have achieved remarkable benchmark accuracy, yet current state-of-the-art implementations are primarily optimized for evaluation performance rather than the requirements of real-time closed-loop aerial control. When deployed onboard UAVs, this can introduce limitations that degrade flight performance. In this work, we identify five architectural deficiencies in a representative tightly coupled IESKF-based LIO implementation: odometry publishing tied to the LiDAR rate (10 Hz instead of the IMU's 200 Hz), missing velocity outputs, execution bottlenecks that block IMU processing, mutex contention, and synchronization race conditions. We introduce corresponding modifications including IMU-rate forward propagation, direct body-frame velocity publishing, SLERP-based smoothing, dual-executor isolation, and explicit synchronization protection. The resulting system increases odometry output from ~10 Hz to a stable 200 Hz, provides a complete Twist state at every IMU sample, and preserves continuity during transient LiDAR loss. Experiments on a Livox Mid-360 / Pixhawk 4 Mini autonomous UAV with motion-capture ground truth validate the approach. Since the underlying estimator (IESKF + ikd-Tree) remains unchanged, the proposed improvements can be directly applied to FAST-LIO2-derived implementations.

Path planning under partial observability remains challenging because an agent must make long-horizon navigation decisions from only locally bounded observations. Nevertheless, historical trajectories contain reusable experience-guided directional preferences. Classical planners, however, typically solve each instance from scratch and lack an explicit mechanism to exploit such transferable decision knowledge, often leading to redundant node expansions and locally myopic search behaviors. Motivated by this limitation, this paper proposes ImiPath, a prior-guided learning framework that distills reusable spatiotemporal decision priors from demonstration trajectories and uses them as experience-informed directional guidance to bias planners toward reliable and promising search directions under partial observability. Specifically, ImiPath first constructs a local spatiotemporal observation representation, which encodes the spatial information of the local environment and the temporal information of historical trajectories. The SpatioTemporal-Attention Policy Network (STAPNet) then transforms this representation into dicision priors. These priors are further incorporated into heterogeneous planners as directional guidance, biasing the search toward locally promising regions. Extensive experiments demonstrate that ImiPath achieves competitive path quality and improves search efficiency by reducing redundant node expansions under local observability. Additional physical experiments on a magnetic microrobot platform further validate the adaptability and practical deployment potential of the proposed framework.

Safe autonomous navigation in shared environments requires the ability to anticipate and react to the latent behaviors of surrounding robots. In this paper, we propose SOWL-MPC, a safe learning-based predictive control strategy for a novel scenario, which we name ego-world robotic framework. In this setting, the control policy of the world robot is unknown and the ego exploits data to learn it and perform safe maneuvers. The proposed architecture combines an online learning mechanism based on Sparse Variational Gaussian Processes (SVGPs) with a receding-horizon control scheme. Relying solely on noisy state measurements, our approach infers a posterior distribution over the latent world policy, which is updated on streaming data via Online Variational Conditioning (OVC). The learned policy is propagated through the nonlinear world dynamics using an approximate moment propagation scheme, and fed to an uncertainty-aware Model Predictive Control (MPC), thus enabling safe maneuvering of the ego robot. The real-time feasibility and safety guarantees of SOWL-MPC are demonstrated through extensive Monte Carlo virtual experiments in ROS 2, and validated on real-world robotic hardware in an indoor arena.

Foundation-model-based vision-language navigation (VLN) has advanced autonomous robot navigation by enabling robots to interpret natural-language instructions, identify semantic goals, and follow user-specified behavioral rules. However, existing VLN systems rely heavily on cloud-hosted foundation models for language understanding and semantic grounding, limiting their applicability where network connectivity is unavailable and reliable metric goal localization is required. Although recent small language models (SLMs) enable fully onboard inference, their suitability for navigation instruction decomposition has not been systematically evaluated. This paper makes three contributions toward fully onboard VLN for outdoor environments. First, we present the first systematic benchmark of 17 edge-deployable SLMs against 4 online APIs for robotic navigation instruction decomposition, evaluating accuracy and latency on human-annotated instructions across three computing platforms and providing practical guidance for selecting onboard language models. Second, we propose a lightweight hybrid semantic-geometric goal localization framework that combines open-vocabulary object detection, prompted segmentation, and LiDAR geometry to estimate metric goals, while maintaining visual bearing guidance when reliable geometric observations are unavailable. Third, we integrate these advances into Edge-BehAV, a fully onboard extension of the BehAV architecture that enables cloud-independent behavior-guided navigation. Experimental results show that the best offline SLM matches the instruction decomposition performance of the strongest cloud API while running approximately 9x faster and without network connectivity. The proposed goal localization framework reduces mean goal-distance error from 2.05 m to 0.20 m at lower computational cost, and the complete system succeeds in 31 of 32 closed-loop outdoor trials.

Nociception is a protective biological mechanism that links harmful stimulation to a reaction. This paper investigates artificial nociception for a robotic arm with whole-body tactile sensing. We present a complete pipeline that maps pressure changes from sensitive skin on a robot manipulator to bio-inspired withdrawal motions. The system first converts skin pressure into a scalar pain gain using a nonlinear continuous model. We compare three reflexes: (i) uniform reflex moves four robot joints by a fixed amount, whereby the withdrawal is approximated by a movement of the arm "toward the base", independent of where the robot was touched; (ii) biologically motivated location-dependent joint-space withdrawal derived from human withdrawal reflex characteristics; (iii) Cartesian space withdrawal along the surface normal of the contacted skin pad. All behaviors are integrated in a reflex controller that interrupts the task, executes the withdrawal, and returns to a pre-contact pose. A user study with 15 participants compared the strategies using Godspeed questionnaire subscales, custom perceived-naturalness and safety items, forced-choice comparisons, and qualitative feedback. Interestingly, participants rated more highly the uniform reflex behavior over one or both competitors on the anthropomorphism, animacy, and likeability Godspeed subscales and on the Naturalness and Realism custom scale. When asked to compare the conditions, the uniform reflex was scored best in "felt safest", "most human-like", and "most natural". This suggests that predictability of the robot behavior is key for user acceptance. The Cartesian reflex was judged the most appropriate reaction to touch. The bio-inspired reflex did not lead any evaluated measure. This may be partly attributed to the embodiment gap between the robot arm and human arm and participants having different expectations from a robot manipulator.

The passive adaptability inherent in soft robotic hands affords them advantages in applications that require safe and compliant interaction. However, existing soft robotic hands often struggle to simultaneously achieve adequate output performance and easy manufacturing due to their complicated structures. In this paper, we introduce the asymmetric origami bending (AOB) pattern for generating bending motion and the asymmetric dual-chamber (ADC) design for obtaining multifunction capability. The AOB single (AOB-S) chamber and AOB dual-chamber (AOB-D) units are designed and constitute the finger and palm actuators of the proposed Origami-inspired SOft Robotic (OSOR) hand. The OSOR hand achieves bio-inspired fingers-palm motions and adequate output performance within a monolithic structure that significantly simplifies the manufacturing process. By defining the asymmetric ratio to characterize the geometric asymmetry of the unit, the analytical models of the AOB and ADC structures are proposed. The Finite Element Analysis tool for the design of AOB actuators is obtained by geometric analysis. The asymmetric origami design grants the integrated manufacturing of the OSOR hand through a Selective Laser Sintering printing process with a single thermoplastic polyurethane material. The model and simulations are validated by experimental results. Experiments show the finger and palm maximum bending motion range of 203{\deg} and 40{\deg}, respectively, with output forces of 6.3 N and 16 N. The OSOR hand is capable of pinching a piece of tissue, stably grasping water bottles with two fingers, palm-only grasping, and completing the power grasps in the taxonomy of manufacturing grasps. The compactness, performance, and easy manufacturing of the proposed hand benefit the development of the soft robotic hand with new possibilities.

Generative social robots (GSRs) powered by large language models offer new possibilities for personalized tutoring in higher education, but also introduce risks related to misinformation, missing transparency, or reinforcing incorrect student responses. Prior work identified knowledge-based design (KBD) requirements that define the informational prerequisites for GSRs to manifest responsible and effective tutoring behavior in higher education. In this paper, we operationalized selected KBD requirements in the Reachy Mini robot platform through system prompting, retrieval-augmented generation, and stateful prompt orchestration. As a result, we present Teachy Mini, a GSR tutoring system that was developed using KBD. To test the system, we conducted a preliminary evaluation study. Participants (N = 24) completed a robot-guided learning session about research methodologies. They learned either with Teachy Mini or with a control version that did not follow KBD principles. Teachy Mini was perceived as significantly more aligned with responsible tutoring behavior than the control robot. Moreover, a manipulation check illustrated that Teachy Mini used personalization, slide-grounded explanations, Socratic questioning, affective support, and learner-anchored feedback more consistently than the control robot. No significant between-condition differences were found in system acceptance, intrinsic motivation, or learning effectiveness, although exploratory analyses suggested a positive effect of KBD on objective learning gains when accounting for learner preferences. Overall, the study offered an initial implementation and preliminary evaluation of KBD for GSR tutoring, indicating that KBD can shape responsible robot behavior and potentially increase learning effectiveness in robot-supported learning.

Motion planning algorithms compute control sequences that drive autonomous robots to goal regions while avoiding unsafe states. Existing methods, from sampling-based planning to deep reinforcement learning, typically provide task-completion guarantees only with respect to a nominal model or simulator, which may be invalidated when the true dynamics are unknown or difficult to model accurately. This letter addresses this limitation for systems with unknown dynamics and an available approximate nominal model, contributing a planner-agnostic constraint-tightening procedure that equips existing planners with a probabilistic task-completion guarantee on the true system. We leverage conformal prediction to provide a probabilistic bound on the nominal-to-true trajectory deviation over a distribution of planning problems. We tighten the planning constraints using that bound, and show that solving the tightened problem under the nominal model is a sufficient condition for solving the original problem on the true system with a prescribed probability. We validate the theoretical guarantees empirically and demonstrate substantially improved task completion relative to nominal-model planning.

Humans routinely communicate through abstractions of their bodies, including shadows, silhouettes, and reflections. Yet robots remain largely confined to expressing themselves through their physical morphology. Enabling robots to communicate through such projected visual abstractions requires reasoning not only about bodily motion but also about how that motion is transformed into an external representation perceived by an observer. Among these abstractions, shadows provide a particularly compelling example because they emerge directly from the robot's embodiment while remaining visually distinct from the body itself. Here, we present a robotic system capable of dynamic shadow expression using a 21-degree-of-freedom dexterous hand with compliant soft skin and a learned shadow self-model. The soft-skinned embodiment reduces light leakage to produce visually continuous silhouettes, while the differentiable self-model learns the mapping between hand configurations and projected shadow appearance through task-agnostic self-exploration. Given a target shadow image or video, the robot optimizes its hand configurations through gradient-based search over 1 the learned self-model and refines the solution through collision-aware simulation to obtain physically feasible motions. For dynamic shadow performance, we further introduce expressive-region objectives, temporal smoothness regularization, and keyframe-based optimization to preserve visually important motion cues while reducing optimization complexity. We demonstrate robotic shadow expression across sign-language gestures, hand-shadow puppetry, and animal motion imitation in both simulation and physical experiments. These results establish a framework for enabling robots to manipulate projected visual abstractions of themselves for communication and visual storytelling.

The paper proposes a robot-agnostic compliant-control framework that extends the ROS control ecosystem with standardized joint and Cartesian command interfaces. It addresses a key limitation of existing control software: no reusable infrastructure for implementing compliant-control algorithms across different manipulators while preserving a common interface to higher-level applications. A plugin-based architecture separates controller infrastructure from control-law implementation. Generic wrappers use existing hardware abstractions to interface with different manipulators, while runtime-loaded plugins implement only the control law. Command interfaces support joint- and Cartesian-space references, stiffness and damping gains, nullspace targets, and feedforward terms, enabling variable impedance and diverse compliant-control formulations. Robot kinematics and dynamics are computed from URDF models using Pinocchio. The architecture facilitates the development of compliant-control strategies and enables the same implementation to be deployed across platforms unchanged. The complete framework, including reference controllers, high-level task interfaces, and example configurations for various manipulators, is open-sourced. The reference Cartesian impedance controller supports task-dependent compliance by rotating translational and rotational stiffness and damping, allowing the principal compliance directions to be updated online according to local task geometry rather than remaining fixed in the robot base or TCP frame. This is particularly important in contact-rich manipulation, where the desired directions of motion, constraints, and compliance directions may vary throughout task execution. Real-robot experiments demonstrate task-dependent compliance in contact-rich manipulation, while simulations show portability across manipulators with distinct kinematic and dynamic characteristics.

Contact-rich robot manipulation requires physical interaction cues that are often invisible to cameras, making tactile sensing essential for robust control. However, scaling visuo-tactile robot learning remains difficult because real tactile interaction data are expensive to collect, hardware-dependent, and limited in task and scene diversity. We present ViTacWorld, an action-conditioned visuo-tactile world model for scalable contact-rich robot manipulation. ViTacWorld leverages public real tactile datasets and a constructed simulation environment to scale visuo-tactile-action data, exploiting the fact that tactile signals are directly grounded in physical contact and can exhibit a smaller simulation-to-real gap than purely visual observations. The model is first pretrained with large-scale real and simulated visuo-tactile trajectories, and then finetuned with real-world policy rollouts to better match downstream manipulation behaviors. Given robot actions, ViTacWorld predicts temporally aligned visual observations and tactile feedback, enabling visuo-tactile-action rollout generation. To the best of our knowledge, ViTacWorld is the first framework that uses a world model for robot visuo-tactile-action trajectory generation and policy evaluation. It serves two roles: synthesizing rollouts to improve downstream tactile policies, and evaluating policies by predicting action-conditioned visuo-tactile outcomes under controlled action sequences. Experiments on contact-rich manipulation tasks show that ViTacWorld generates physically meaningful rollouts, improves policy performance through scalable data augmentation, and enables action-conditioned policy evaluation. Project page: https://vitacworld.github.io/

Action-conditioned video world models predict future observations from an initial observation and an action signal. In robotics, actions influence future observations through two distinct processes: they are first realized into robot motion by the robot body and controller, and the scene then responds through contact and object motion. Conditioning directly on action commands asks the world model to learn the realization process itself, while conditioning on logged future states leaks the interaction outcomes it is meant to predict. We propose robot-factored world models, which move two robot-specific factors outside the world model. First, action realization: each command is rolled through the robot's own controller and kinematics into a deployment-available nominal trajectory, a middle signal that avoids both action-realization learning and future-state leakage. Second, robot rendering: this nominal trajectory is rendered through the robot URDF, factoring the robot's geometry, kinematics, and appearance out of the model and into explicit rendered robot geometry. To resolve depth ambiguity, we pair end-effector depth with scene depth, giving geometric cues for contact and occlusion beyond image-plane overlap. Together, camera-aware static RGB/depth context and rendered robot geometry form a shared visual world-model interface that stays consistent across viewpoints and robot embodiments, so the model sees the action only as visible robot geometry and learns how objects respond to it. Our experiments show that the rendered interface outperforms vector-conditioned baselines and generalizes to unseen robot embodiments at inference. We further demonstrate that our model generates robot manipulation videos from human demonstrations by retargeting and rendering the hand motion as robot geometry.

We present Oxygen-TryOn, a unified foundation model for any-item virtual try-on. Rather than repurposing a general-purpose image editor, Oxygen-TryOn is fashion-native, built for try-on through a dedicated data engine and try-on-specific training. Given one or more reference items (clean product shots or in-the-wild worn-on photos) and a single target subject image, it synthesizes a photorealistic image of the subject wearing the items across virtually any fashion category. Prior systems handle a single garment category in a studio setting, and recent multi-reference methods remain garment-centric; in contrast, Oxygen-TryOn supports diverse items and scenarios, including full- and half-body views, a variable number of references, and free multi-item composition, while faithfully preserving both subject identity and item appearance. Instead of mask-based inpainting, we reformulate try-on as a multi-reference, understanding-driven generation task. We build a data engine that collects, manufactures, annotates, and filters high-quality try-on data at scale, and design a three-stage recipe of continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL). The RL stage uses a hybrid reward combining an in-house try-on reward model with a proprietary, rubric-guided general-purpose model, jointly supervising fine-grained consistency and instruction-level quality. It also follows general editing instructions (e.g., pose changes) in the same pass. Across public benchmarks and our in-house Oxygen-TryOn Bench, it achieves state-of-the-art consistency and realism on single-item try-on and leads on multi-item try-on, matching or surpassing both leading proprietary systems (Nano Banana Pro, GPT-Image-2, Seedream5 Lite) and open-source models (FLUX.2).

While Large Vision-Language Models (LVLMs) exhibit strong perceptual capabilities, they remain vulnerable in visual reasoning tasks. Existing benchmarks largely focus on symbolic mathematical or scientific problems and simple vision-centric tasks, offering limited assessment of complex visual reasoning and logical consistency, a critical requirement for reliable reasoning systems. We introduce ConVBench, a complex vision-centric reasoning benchmark in which each image is paired with two logically equivalent questions across six categories: action and state, complex counting, spatial reasoning, causal and intent understanding, commonsense reasoning, and temporal perception. To complement this benchmark, we define two evaluation metrics, logical consistency and robust accuracy, that jointly assess both the correctness and consistency of model responses. We further present ConVLM, which improves LVLM reasoning through Group Relative Policy Optimization (GRPO)-based reinforcement learning with a novel consistency reward. This method leverages automatically generated logically equivalent question-answer pairs and a dual-reward design combining accuracy- and consistency-based signals, encouraging agreement between paired responses. The framework functions effectively with or without strict answer supervision.

Medical ultrasound (US) image segmentation faces significant challenges due to speckle noise, low-contrast boundaries, acoustic shadowing, and acquisition variation across operators and clinical centers. Although encoder-decoder and transformer-based networks have achieved strong performance, many methods recover boundary details through dense decoders or larger backbones, which may still produce over-smoothed contours or unstable predictions under external distribution shifts. In this article, we propose Risk-routed Implicit Boundary Refinement (RIBR), a compact segmentation framework that uses implicit neural representation as a risk-routed residual correction rather than an unconstrained full-mask predictor. RIBR combines boundary-refinement implicit residuals, risk-routed residual control, and geometry- and speckle-aware boundary regularization to refine uncertain contours while suppressing non-boundary oscillations. Evaluation on nine US datasets covering lymph nodes, breast lesions, thyroid nodules, and prostate shows that RIBR achieves the best overall macro-average and consistently reduces boundary error across grouped and organ-specific comparisons under a compact parameter budget. These findings suggest that controlled implicit residual learning is a practical strategy for resource-constrained and boundary-sensitive US segmentation. Source code is available at https://github.com/jinggqu/ribr.

One-to-many facial identification is commonly used to match a probe image from surveillance video against a gallery of driver's licenses and/or booking photos. The algorithm's rank-one image from the gallery, or a human examiner's selection from the algorithm's top-ranked images, may then be placed in a photo lineup shown to a witness. Witness selection of the gallery image in the photo lineup may then lead directly to the person in the gallery image being arrested. This facial identification process is involved in at least 9 wrongful arrests. This work specifically examines whether the probability of a witness making an incorrect identification increases with the size of the gallery searched. We compare photo lineup accuracy when the "suspect" image is drawn from galleries of 500, 5,000, and 24,000 images. We find that larger galleries increase both the likelihood of a witness making an incorrect identification and their confidence in that (incorrect) identification. These results raise questions of whether an image resulting from such a facial identification process should be used in photo lineups and of whether results of a photo lineup alone should constitute probable cause for arrest.

Learning high-fidelity point-cloud features in the 3D space poses significant challenges, including permutation invariance, lack of local context, difficulty in fine-grained surface reconstruction, and high computational cost. In this article, we propose Trans-Unet, a novel framework that addresses these issues by first tansforming 3D point-cloud data into a 2D grid domain and then employing a U-shaped hybrid model that integrates Convolutional Neural Networks, and self-attention mechanisms. The proposed Trans-Unet effectively learns and reconstructs precise features from high-resolution 3D point-cloud data (with 40,401 points in surface and 2,382 points in fiber) derived from a predefined finite element brain patch growth model, enabling accurate prediction of brain folding patterns. By combining multiple techniques, Trans-Unet leverages the complementary strengths: the 3D-to-2D transformation preserves fine-grained structural information while significantly reducing computational cost and the curse of dimensionality; convolutional blocks capture hierarchical, low-level local representations; and the self-attention mechanism models global, high-level semantics and long-range dependencies. The dataset consists of 3D point-clouds containing both brain surface patches and fiber information generated by a large-scale finite element model. Trans-Unet is applied to predict brain surface folding from the initial state (state 0 or states 0-2) to the final state (state 3). Experimental results demonstrate that Trans-Unet achieves high-resolution predictions of brain patch growth, surpassing existing methods in both fidelity and accuracy.

Recent conditional video generation models have shown promising potentials to transform 3D engine renderings, such as depth maps and untextured geometry, into photorealistic videos for gaming and immersive content creation. These applications require long-horizon auto-regressive generation that continuously synthesizes new frames while preserving a persistent 3D world. Auto-regressive generators synthesize video chunk by chunk with a bounded KV cache, so when the camera revisits a location after its context has been evicted, the model often regenerates inconsistent appearance, even though the conditioning renderings (e.g., depth) remain perfectly aligned with the underlying geometry.We address this revisit inconsistency without any post-training by exploiting correspondences the 3D engine already provides: temporal correspondence retrieves pose-matched historical latent chunks into the KV cache as loop-closure memory, while spatial correspondence from camera pose and depth reprojection biases token-level attention toward geometrically corresponding regions of the retrieved chunks. We demonstrate our method on loop-closure trajectories mined from TartanAir and TartanGround dataset to mirror complicate real-world application scenarios, where it outperforms existing training-free baselines on revisit consistency without losing overall video quality. Project Page: https://wenchao-m.github.io/ClosetheLoop.github.io/

As semiconductor manufacturing advances toward sub-2nm nodes, local place-and-route (P&R) design-rule violation (DRV) fixing is increasingly limited by complex rule interactions, dense multi-layer routing geometries, and foundry-specific constraints. While Large Language Models (LLMs) have recently demonstrated strong capabilities in EDA scripting and documentation, their application to visual layout understanding remains largely unexplored: diagnosing DRC violations from layout imagery demands precise geometric reasoning and foundry-specific rule knowledge absent from general-purpose VLM training. We propose SCALE, a framework with a self-supervised layout-generation stage for local DRV fixing at advanced nodes. Multi-layer layout geometry is serialized into structured text, and a fine-tuned language model learns to reconstruct randomly masked polygons from surrounding BEOL context alone without violation labels. At inference, natural-language rule constraints and high-temperature sampling steer generation toward diverse, violation-prone layout variants validated by an industrial signoff DRC checker, producing DRC-annotated layout--violation pairs used to fine-tune a domain-adapted DRC-VLM. This VLM provides rule-aware geometric guidance for local DRV repair, boosting state-of-the-art agents' solve rates by +12--25% (up to 97%) on 100 real sub-2nm cases spanning enclosure, spacing, width, and color-spacing violations.

Agricultural field maps are often proprietary, incomplete, or outdated, yet they provide the spatial framework for crop monitoring, production accounting, and land-conversion analysis. This study presents a reproducible workflow for mapping farmland extent and visible boundaries from 1 m NAIP RGB imagery. Thirty-seven scenes spanning open cropland, peri-urban interfaces, semi-arid irrigation geometries, and fragmented mosaics were annotated in CVAT and converted to binary masks. Non-overlapping 256 x 256 patches yielded 5,698 samples, split by source scene into 3,850 training, 770 validation, and 1,078 test patches. A residual U-Net (ResUNet) trained with a Dice-dominant loss, L = 2.5(1 - Dice) + BCE, achieved test accuracy 0.8808, IoU 0.8605, Dice 0.9234, precision 0.8766, and recall 0.9794. A frozen SAM 3 branch prompted with "agricultural farmland field" was fused with ResUNet by logical OR. On selected difficult patches, Dice improved from 0.858 to 0.955 (orchard rows) and from 0.804 to 0.903 (fragmented parcels). Sliding-window stitching produced coherent regional masks (example tile Dice 0.898 and 0.919). The product is a semantic farmland-extent layer, not a cadastral parcel map, and supports agricultural monitoring where current field layers are unavailable.

Semantic 3D Gaussians provide a compact representation for 3D semantic occupancy prediction by rendering semantic primitives into a voxel volume under voxel-wise supervision. Recent methods have improved the modeling ability and efficiency of this representation through more flexible primitive shapes, geometry-guided initialization, and progressive densification. However, these advances mainly determine how primitives are represented, initialized, or added, and do not explicitly address how to select the most useful Gaussians when their total number must be limited to control memory and computation. This imbalance creates an allocation bottleneck: redundant Gaussians remain in simple regions, while difficult regions receive insufficient semantic support. We propose the Semantic Gaussian Allocation Transformer (SAGFormer), which uses Gaussian attributes and local geometric-semantic features to score candidates and select a fixed final Gaussian set. Experiments on nuScenes-SurroundOcc and SSCBench-KITTI-360 show that SAGFormer improves occupancy prediction under the evaluated protocols and yields more semantically consistent and better-utilized Gaussian representations. Under similar final counts and raw coverage, it reduces semantic mixing, strengthens class-consistent voxel support, and produces fewer unused Gaussians. The results indicate that explicit capacity allocation is a useful complement to Gaussian refinement for semantic occupancy prediction.

The rapid advancement of generative models has spurred the critical need to evaluate the worst-case robustness of deepfake detectors. In this paper, we reveal a fundamental blind spot in current forensic paradigms: while existing detectors excel at capturing digital synthesis artifacts, their effectiveness drops drastically when AI-generated content is cloaked in authentic physical imaging characteristics. We posit that genuine photographs inherently possess hardware-intrinsic statistical signatures, which are imperceptible footprints imprinted by optical sensors and Image Signal Processing (ISP) pipelines, and are fundamentally absent in purely data-driven generative models. Driven by this insight, we propose ISPCloak, a novel optimization-free adversarial attack framework that explicitly weaponizes the ISP pipeline to mislead the judgment of deepfake detectors. Rather than relying on computationally expensive gradient perturbations, our method first employs an Invertible ISP network to project images into the RAW domain. Then, we seamlessly imprint the complex statistical priors of real cameras onto AI-generated images by injecting realistic Poisson-Gaussian sensor noise and conducting forward ISP reconstruction. Synergized with generative artifact suppression and adaptive masking, this streamlined physical simulation enables ultra-fast generation of adversarial examples. Extensive experiments show that embedding authentic physical perturbations fundamentally disrupts a broad range of current detection mechanisms, yielding universally evasive adversarial examples with imperceptible visual alterations.

Image inpainting aims to reconstruct missing or corrupted regions of an image while preserving as much as possible, visual and semantic consistency. In medical imaging, this task is particularly important because artifacts, missing information, and pathological alterations can compromise diagnostic reliability and downstream clinical applications. Recently, diffusion models have emerged as state-of-the-art generative approaches for medical image inpainting due to their ability to generate anatomically consistent reconstructions. This survey presents a systematic review of diffusion-based methods for medical image inpainting, covering the main architectures, applications, datasets, and evaluation strategies reported across 60 studies. In addition, we propose a taxonomy for diffusion-based approaches. The analysis reveals a rapid growth of research interest in diffusion-based medical image inpainting, with denoising diffusion probabilistic models and latent diffusion models emerging as the dominant architectures. The reviewed studies mainly focus on artifact removal, data augmentation, pseudo-healthy tissue reconstruction, and anomaly detection, particularly in magnetic resonance imaging and computed tomography imaging. Overall, diffusion models demonstrate strong performance in producing anatomically plausible reconstructions and aiding downstream clinical tasks. However, the review also highlights important challenges, including the lack of standardized benchmarks, limited dataset diversity, and restricted validation procedures across diverse clinical applications and imaging scenarios.

Unmanned aerial vehicle (UAV) communication is expected to support a wide range of low-altitude applications in 6G mobile networks. However, traditional statistical channel models provide limited accuracy in specific environments, while deterministic methods such as ray tracing usually rely on accurate three-dimensional environment models and involve high computational complexity. Existing multimodal channel prediction approaches mainly focus on large-scale metrics such as path loss, and remain insufficient for modeling small-scale parameters. To address these limitations, this paper proposes PanoLAMP, a Panoramic perception and vision-language model-based Low-Altitude Multipath Prediction framework. It adopts a pretrained vision-language model as the backbone and captures the propagation environment features through panoramic RGB-D observations collected at both the transmitter and receiver to predict the delay, power, azimuth angle, and zenith angle offset relative to the line-of-sight path. Experiments are conducted on a synthetic dataset containing 18,949 UAV-vehicle links across seven UAV altitudes. Experimental results show that the proposed method consistently outperforms representative baselines in both multipath parameters and statistical metrics, and demonstrates stronger generalization across different flight heights.

Automating radiology report generation is important for improving reporting consistency and clinical workflows . While Contrastive Language--Image Pretraining (CLIP) has advanced medical vision language modeling, existing CLIP-style approaches may still provide insufficient fine-grained semantic supervision for complex report generation. Standard CLIP primarily optimizes cross-modal alignment, without explicitly structuring the textual embedding space that guides visual representation learning. To address this limitation, we propose TextSLIP, a general medical vision-language pretraining framework that augments CLIP with intra-modal text contrastive learning. By improving textual embedding discriminability through self-supervised augmented text pairs, TextSLIP is designed to provide finer-grained linguistic supervision to the visual encoder. As an initial validation, we pretrain TextSLIP on a curated dataset of 7 million brain MRI image-text pairs and fine-tune the pretrained visual encoder within a report generation architecture. In controlled comparisons with CLIP-style baselines, TextSLIP shows consistent improvements on report generation metrics. Ablation studies further suggest that text-side self-supervision contributes to the observed gains. These results indicate that text-level contrastive learning is a promising direction for improving medical visual-textual alignment, while broader validation across additional medical domains remains an important next step.

Vision foundation models are increasingly reused as frozen backbones for downstream visual recognition, making parameter-efficient adaptation a central problem. Prompt-based adaptation, including Visual Prompt Tuning (VPT), provides a lightweight way to specialize these models, but its layer-wise behavior remains poorly understood: performance is sensitive to prompt depth, placement, and task distribution, and gains on standard in-domain benchmarks do not always translate into robust generalization. We argue that this limitation is not solely an optimization issue, but a layer-wise information allocation issue: existing prompt-based methods lack principled control over what prompt-conditioned representations should preserve, suppress, and propagate across depth. Inspired by the Information Bottleneck principle, we introduce Prompted Information Bottlenecks (PIB), a framework that regularizes layer-wise compression-sufficiency trade-offs and promotes a more coherent cross-layer information path. The key idea is that effective adaptation should be minimal yet sufficient, retaining task-relevant local evidence in earlier layers while progressively discarding nuisance factors and redundant details in deeper layers. Extensive experiments show that PIB achieves strong performance across 34 datasets, reaching 92.1% on FGVC, 93.01% on HTA, and 77.33% on VTAB-1k, while tuning only 0.35% parameters on average across the main settings. Beyond benchmark accuracy, PIB helps explain the non-monotonic behavior of prompt capacity scaling, reduces shortcut reliance, and improves robustness under distribution shift and fine-grained recognition settings. These results position PIB as both a practical method and an information-allocation perspective for adapting frozen vision foundation models. Our code is available at https://github.com/itsnotacie/MM-26-PIB

This paper introduces a new benchmark test, Medical-Checklist, for assessing medical multimodal models. The recent advancements in multimodal models have demonstrated significant potential in the field of medical vision-language tasks. However, it is becoming increasingly clear that evaluating these models' performance, whether they are applied to natural or medical images, is challenging. The critical question is whether the models can accurately understand an input image while associating it with relevant input text. To address this, Medical-Checklist imposes a binary test on the models: they are given an image and two captions, where one is correct and the other incorrect, and the model must select the correct one. The incorrect caption contains a single medical concept (word or phrase) that is inaccurately substituted from the correct caption. Although the task is simple, this simplicity enables the unified assessment of diverse multimodal models designed and learned on different principles. It also enables us to verify whether models correctly understand a wide range of medical concepts across various medical sub-domains. Medical-Checklist is designed to reduce potential biases in data and to enable evaluation of the models' ability to handle out-of-distribution inputs, which were difficult in existing datasets. When evaluating four state-of-the-art medical multimodal models with Medical-Checklist, it was revealed that despite their excellent performance in specific tasks such as Med-VQA, they may not correctly understand images, suggesting a long journey ahead for clinical application. The dataset and code will be made public upon acceptance.

Multimodal chain-of-thought (CoT) reasoning integrates visual and textual cues through step-by-step inference. In small models with limited token budgets, modality-interaction fusion often suppresses tiny cross-modal differences. In particular, multimodal CoT often struggles when different images pair with identical text or different texts pair with an identical image, making such inputs nearly indistinguishable after fusion. This study proposes Visual Saliency Steering Distillation (VSSD). VSSD leverages the attention maps of multimodal large language models to generate perturbed images that capture task-sensitive feature directions, and then applies singular value decomposition to extract dominant steering vectors to guide inter-layer distillation. Experiments on ScienceQA and M$^3$CoT demonstrate that VSSD improves rationale generation and answer inference. The code is available at https://github.com/BGWH123/VSSD.

MEMEs are widely used on the internet and often carry strong elements of sarcasm or irony. Understanding their hidden meanings typically requires a joint interpretation of text and vision. Existing methods focus on the dual-stream vision-language model to extract the visual and text simultaneously, which lacks background information and prior knowledge about the comprehensive explanation of MEME. One feasible option is to adopt chain-of-thought (CoT). However, the simple CoT approach lacks multi-perspective thinking, which may compromise the reliability of the resulting answers. Moreover, it often relies on shallow feature fusion, lacking the fusion of local details and fine-grained visual-prompt text alignment. This limitation prevents a deeper understanding of the intricate connections between the visual and the text. Herein, an enhanced vision-language multi-CoT (EVL-MCoT) approach is proposed to address these limitations. By promoting multi-CoT, EVL-MCoT enhances consistency and reduces bias in the decision-making process. Additionally, we design a prototype-guided and context-guided decoding framework, which incorporates visual prototypes to guide the fusion process and enables the model to align textual and visual information more precisely. We achieve promising results on the HatefulMemes and MultiOff datasets. The source code has been publicly released and is available at https://github.com/BGWH123/EVL-MCoT.

Vision-language models (VLMs) are increasingly deployed on consumer hardware where input images are degraded by compression, camera shake, and poor lighting. In such settings, a reliable uncertainty signal matters more than raw accuracy, because it determines when a system should defer rather than answer. We evaluate two small open-weight VLMs -- Qwen2-VL-2B-Instruct and SmolVLM-Instruct -- across six realistic photographic degradations at three severity levels, comparing two confidence signals: the confidence the model states in natural language, and the model's own mean token probability over its generated answer. Across 3,800 predictions, we find a large and consistent gap. Verbalized confidence in Qwen2-VL is almost constant (mean 0.87-0.90 across all conditions) and detects its own errors at chance level (AUROC 0.39-0.75, typically ~0.50), while internal token probability from the same model separates correct from incorrect answers with AUROC 0.92-0.99. In SmolVLM, verbalized confidence proved largely unobtainable: across three prompt templates, only one of five pilot attempts produced a parseable confidence value, while internal probability again yielded above-chance error detection (AUROC 0.54-0.92). Both models fail in the same place: under severe underexposure, accuracy collapses (0.99->0.22 for Qwen2-VL, 0.97->0.42 for SmolVLM) while both confidence signals barely move, and internal error-detection falls to chance. We conclude that small VLMs encode usable self-knowledge that their verbalized output does not express, that internal probability is therefore the better deferral signal in constrained deployment, and that neither signal should be trusted under severe low-light conditions.

Segmentation of adjacent structures with similar intensity distributions remains a challenging problem in image analysis, particularly when object boundaries are weak or ambiguous. Under such conditions, classical variational models may suffer from degenerated image-driven forces, leading to boundary leakage or undesired merging of neighboring regions. To address these limitations, we propose a smooth phase-separation variational model based on the Cahn--Hilliard equation for weak-boundary segmentation of homogeneous-appearance structures. The proposed framework integrates softmax-based region fitting with Cahn--Hilliard phase-field regularization to maintain interface discrimination under weak image-driven forces. We further introduce a mixed $L^2-H^{-1}$ gradient flow, which preserves higher-order interfacial regularization while allowing adaptive changes of phase masses, establish the continuous energy dissipation law, and prove the existence and uniqueness of weak solutions in the natural solution class. For numerical computation, we develop a stabilized scalar auxiliary variable (SAV) scheme that is linear, FFT-based, and satisfies a modified discrete energy dissipation law. Numerical experiments on synthetic and medical images demonstrate that the proposed method effectively separates adjacent homogeneous structures across weak boundaries and achieves competitive segmentation accuracy and improved boundary localization compared with representative variational, phase-field, and deep learning methods.

Multi-branch architectures and CNN-Transformer fusion have long been regarded as effective ways to improve vehicle re-identification (Re-ID) by combining complementary representations. In this work, we revisit this assumption in the foundation-model era through a comprehensive empirical study. A single DINOv3-pretrained ConvNeXt trained with a tuned recipe achieves 88.19 mAP on VeRi-Wild Small and 77.47 mAP on VeRi-Wild Large using visual cues alone, matching the strongest protocol-verified metadata-dependent multi-branch baseline. Applying training-free re-ranking further improves performance to 92.38 and 83.68 mAP, respectively. Using this strong baseline together with retrieval-level branch diagnostics, we evaluate whether increasing representational diversity still provides measurable gains. Across both benchmarks, concatenating multiple branches built on a shared backbone changes the best single-branch performance by less than one mAP point while increasing the embedding dimension by 4x, and the resulting representation has an effective rank close to the original feature dimension. We further study cross-backbone fusion using an asymmetric frozen-anchor strategy to combine ConvNeXt and Vision Transformer representations. Despite these favorable conditions, Transformer branches consistently remain 13-15 mAP below the ConvNeXt backbone, and paired per-query bootstrap analysis estimates the largest observed fusion gain to be only +0.11 mAP (95% confidence interval). Our results suggest that, under the evaluated setting, improving a single strong foundation-model backbone together with retrieval-stage re-ranking is more effective than increasing architectural complexity through additional branches or heterogeneous backbones. We restrict our conclusions to single-seed training and one family of foundation models and discuss conditions under which these observations may not hold.

With the development of precision livestock farming and the advances in computer vision, visual animal biometrics has gained attention. Using biometric technologies that have been proven effective for humans to identify livestock can increase animal welfare as well as production efficiency. However, challenges such as complex scenarios, similar appearances, occlusions, and non-cooperative behavior, as well as the limited amount of publicly available labeled datasets, remain. In this work, we contribute a novel, publicly available cow face benchmark dataset that has been collected in a realistic automatic scenario with 6,838 images of 161 different cows at a dairy farm. In addition to the public dataset, we define two verification and four identification evaluation protocols to foster comparable research in the cow recognition research field. Further, we provide evaluation results on our dataset of six benchmark models, which include models trained on limited data, cross-species fine-tuned models, and zero-shot foundation model approaches.

Few-shot class-incremental learning (FSCIL) aims to incrementally learn novel classes with only a few samples while avoiding forgetting base classes. However, current methods show a tendency to misclassify novel-class samples into base classes, which we find to be caused by the excessive focus on base-class-discriminative regions on novel-class samples. In this work, we aim to explore the underlying mechanism for an interpretation and solution. We first provide a compositional view to analyze the transferred and reused spatial patterns on novel-class samples. Then, through extensive experiments and theoretical analysis, we identify both empirically and theoretically that a shortcut exists in the model's base-class training, which naturally forms the excessive focus on only the most discriminative regions (primitives), which we term as the regional shortcut. Finally, based on this interpretation, to address this problem, we propose a compositional-learning-based method to learn two primitive sets (a common set and a discriminative set), which alleviates the regional shortcut by constraining the model to learn and utilize the common primitive set for base- and novel-class recognition. Extensive experiments on standard FSCIL benchmarks demonstrate the effectiveness of our approach, yielding consistent improvements over existing state-of-the-art methods in both accuracy and interpretability.

As autonomous driving systems move toward real-world deployment, interpretable, behavior-level decision-making is essential for safety, trust, and regulation. We introduce CommandLM, a multimodal large language model that generates concise, human-readable behavior descriptions for ego vehicles from fused multi-sensor data. Our model processes temporally fused bird's-eye view representations from LiDAR and multi-camera inputs via a Q-Former adapter connected to a quantized, LoRA-fine-tuned large language model. Trained on our CommandLM-nuScenes dataset, CommandLM produces intent-aware, interpretable captions suitable for planner supervision and safety auditing. Experiments demonstrate strong linguistic and behavioral alignment, achieving CIDEr 0.67, and BERT-F1 0.88, substantially outperforming the BLIP-2 baseline (CIDEr 0.52, BERT-F1 0.86). In human evaluation, 58% of the generated descriptions were rated accurate, efficient and rule-compliant, confirming their real-world plausibility. While the remaining descriptions may not always select the most efficient, goal-oriented behavior, CommandLM's interpretable outputs enable downstream validation systems to identify and correct such cases, making it an effective tool for transparent behavior auditing. These results show that integrating multimodal fusion with language reasoning yields efficient and transparent behavior-level understanding for autonomous driving. We release our code and dataset at: https://github.com/b-tok/CommandLM

Generalization remains a critical bottleneck in AI-generated image detection. Because many modern generators are proprietary or adversarially modified, existing detectors overfit to the low-level textural patterns of accessible training data, resulting in severe failures on unseen domains. Conventional regularization techniques (e.g., $L_1$/$L_2$ norms, Dropout) apply indiscriminate parametric constraints and fail to provide the domain-invariant structure necessary for cross-generator robustness. To address this, we propose Feature-Augmented Implicit Regularization (FAIR). FAIR introduces an orthogonal, macro-structural prior, specifically, Scene Composition Structure (SCS), during training to geometrically constrain the model's optimization trajectory. By augmenting the primary feature space with domain-invariant SCS features, FAIR explicitly penalizes texture-biased shortcut learning. Crucially, this structural prior is entirely discarded at inference, yielding a smoothed, generalized decision boundary with zero architectural or computational overhead. Extensive evaluations across five massive benchmarks demonstrate that integrating FAIR into state-of-the-art detectors significantly improves cross-generator generalization, boosting accuracy by up to 8.04% and establishing new state-of-the-art robustness in zero-shot transfer scenarios.

Diffusion models typically suffer from error accumulation during iterative sampling, commonly referred to as exposure bias. We reveal systematic frequency-dependent discrepancies between training and inference, which can be interpreted as frequency-dependent SNR error. Crucially, the direction of this mismatch varies across models and timesteps, indicating that fixed correction rules do not generalize. We propose Spectral Alignment (SPA), a lightweight, guidance-based method that calibrates the power spectrum of intermediate predictions to a pre-computed prior. Our approach consists of two stages: (1) offline fitting of a parametric spectrum model from training data, and (2) inference-time guidance via efficient FFT-based gradient computation. SPA introduces minimal computational overhead (3-4\%) and is complementary to Classifier-Free Guidance (CFG). We demonstrate consistent improvements across diverse architectures, from pixel-space models (DDPM, ADM) to latent diffusion models (SD2.0, SDXL) and flow-matching models (SD3.5, FLUX). Our implementation is available at https://github.com/SonyResearch/SPA.

Diffusion models have recently achieved remarkable success in high-fidelity image synthesis, yet their application to visual text generation and editing remains relatively underexplored. Unlike general image generation, visual text tasks demand precise structural regularity and legibility, which may pose additional challenges for small-scale text and non-Latin scripts such as Chinese. Existing UNet-based models often struggle to produce clear and coherent text, while DiT-based models, though more expressive, are typically limited to a single task, which may lead to redundant training pipelines, inconsistent visual styles, and reduced cross-task generalization. To address these challenges, we propose InnoText, a unified DiT-based framework capable of performing both text generation and editing within a single model. We introduce a Font Size-Aware Modulation (FSAM) module to enhance representations across font scales, a Small-Character Aware Augmentation strategy to improve fine-grained fidelity, and a Task-Specific Region Weighted Loss for adaptive optimization. To support training and evaluation, we also construct a high-quality bilingual (English-Chinese) visual text dataset covering diverse fonts, sizes, and backgrounds. Experimental results demonstrate that our method achieves superior generation accuracy and editing quality, producing visually appealing and realistic text images.

Domain Generalization (DG) aims to learn representations robust to distribution shifts. Recent geometric alignment methods, such as CPCANet, extract domain-invariant structures through batch-wise Common Principal Component Analysis (CPCA). However, CPCANet suffers from rank-deficient covariance estimation due to the small-sample-size issue in mini-batch training. To address this limitation, we propose Projection Pursuit CPCANet (PP-CPCANet), a covariance-free framework that learns a global orthogonal basis on the Stiefel manifold and jointly optimizes it with network parameters via the Cayley transform. We further introduce a symmetry-breaking detached-median PP dispersion objective to extract common principal components (CPCs) with dense and robust optimization signals. Experiments on four DG benchmarks show that PP-CPCANet achieves SOTA performance while maintaining stable training.

Accurate tree-level forest monitoring using laser scanning data requires reliable tree delineation, consistent tree correspondence across multitemporal point clouds, and accurate estimation of tree attributes and their change. Reconstructing tree growth in boreal forests is challenging due to the scarcity of historical stem-level data, propagation of errors from older sensors into change estimation, and growth rates with a magnitude of measurement uncertainty. This study investigates a framework for estimating individual tree diameter at breast height (DBH) and stem volume growth using 136 point clouds acquired between 2014--2025 with 11 scanners on airborne (ALS), mobile (MLS), and terrestrial laser scanning (TLS) platforms across boreal forest test sites. Trees were delineated from an MLS point cloud using deep learning-based segmentation which was transferred to the remaining point clouds, resulting in reliable multitemporal tree correspondence. Stem curves were derived from MLS/TLS data, with ALS data used for height estimation, enabling DBH and volume estimation and time series. A height growth-based scaling model was used to reconstruct stem attributes across time and estimate growth. Results showed that modeled growth achieved higher agreement with manual growth estimates than differencing independently estimated attributes from point clouds. The modeled-manual 5- and 10-year growth RMSEs were 55--111\% and 26--67\% for DBH, and 31--87\% and 21--67\% for volume, respectively, depending on plot difficulty. The scaling model was temporally robust, with errors remaining stable or stabilizing after 5--6 years, reaching maximum RMSEs of 8--12\% for DBH and 12--23\% for volume after 12 years. Combining MLS/TLS-derived stem measurements with multitemporal ALS-derived heights provided a robust framework for individual tree growth estimation without requiring multiple under-canopy scans.

Existing BraTS-GLI datasets provide a widely used benchmark for adult glioma MRI segmentation, but their task definition focuses on tumor subregions and does not systematically represent coexisting white matter hyperintensities (WMH). In joint segmentation settings, such unlabeled abnormalities introduce task-specific label noise by treating pathological regions as normal tissue. To address this limitation, we introduce BraTS-GLI Anatomy-Lesion, a controlled-access, labels-only derived resource built from the BraTS 2023-GLI training cohort. The resource provides 1,251 unified eight-class anatomy-lesion label sets aligned with the original four-modal MRI cases, including image-repair labels for 116 cases requiring repaired imaging inputs. The cohort is organized into a 394-case purified subset and an 857-case extended subset, with case-level metadata covering label source, image-repair requirements, quality-control status, access conditions, checksums, and release boundaries. Compared with the original BraTS-GLI annotations, the resource substantially expands foreground supervision by incorporating healthy brain tissues and previously unlabeled coexisting abnormalities within a unified label space. A validation study using MedNeXt and T1/FLAIR inputs suggests that WMH-aware supervision preserves healthy-tissue segmentation performance across both in-domain GLI and external WMH datasets, while improving sensitivity to coexisting lesions relative to noisy-control training. The resource is intended for scientific research and supports joint anatomy-lesion supervision, label-noise analysis, and reproducible evaluation. Data are available at https://www.synapse.org/Synapse:syn75210889/wiki/, and code is available at https://github.com/xyx200/brats-gli-anatomy-lesion-code. The data resource DOI is https://doi.org/10.7303/SYN75210889.

Accurate pixel-level classification of coronary angiograms is critical for cardiovascular disease assessment, yet the field lacks standardized evaluation protocols. In this work we demonstrate a new benchmark for the assessment of deep learning models which densely classify pixels of coronary angiograms to one of SYNTAX classes (or background). The evaluation covers 24 distinct architectures starting with classic convnets to recent state-space-based vision algorithms. We release CARDIAG - a multi-center, multi-label dataset which we carefully split to reliably compute metrics, accounting for diameter error, overlap, centerline quality and calibration. The data contains SYNTAX labels, binary, uncertainty and segmentation masks as well as intermediate frames together with the selected non-sensitive DICOM metadata. From the multitude of algorithms, we nominate ConvNeXt V2 encoder with DeepLab V3 Plus decoder as the best performing, achieving macro $F_1=0.456$, which we then ensemble with Mamba U-Net and Feature Pyramid Network, for an increased $F_1=0.479$. We demonstrate all the architectures to be well calibrated and determine the generalization of the top 5 methods, together with the data efficiency of these architectures. We highlight the importance of both high-resolution and low-resolution features in encoding. We also demonstrate the model correctness in the context of patient demographic, vessel sides and projection angle configurations. Overall the released benchmark allows for future studies to robustly and rigorously assess the proposals, not only for SYNTAX segmentation, but lesion detection and many more.

Visual localization becomes extremely challenging in planetary-like terrains characterized by low texture, perceptual aliasing, harsh illumination, and sparse, weakly overlapping viewpoints induced by forward rover motion and unconstrained driving directions. Under these conditions, state-of-the-art image-to-image and image-to-map matching pipelines suffer significant performance degradation. In this work, we propose a visual relocalization method that departs from classical correspondence-based pipelines by directly estimating camera poses against a differentiable map representation built with 3D Gaussian Splatting (3DGS). Our key contribution is a geometry-aware training strategy that combines photometric and geometric losses, where the geometric supervision is provided for the first time by combining multi-view stereo (MVS) and LiDAR depths. We show that this joint optimization produces a 3DGS model that better fits the underlying scene geometry, leading to improved photometric and geometric consistency and more robust, accurate single-image 6-DoF pose estimation. Extensive experiments on data acquired in planetary-analog environments validate the effectiveness of our approach, showing substantial gains in relocalization accuracy under challenging conditions. Code is available at https://github.com/DLR-RM/multimodal-gsplat-relocalization.

In this work, we aim to discretize the high-dimensional visual representations to bridge the gap with language models - a non-trivial challenge, as existing quantization methods suffer from codebook collapse, failing to scale while preserving semantic coherence. We identify the root cause as metric mismatch: standard Euclidean codebook objectives are fundamentally misaligned with the anisotropic geometry of representation space, leading to codebook embeddings with high-variance magnitude scales and uneven angular distributions that hinder scalability. To address this, we propose Hyper-Spherical Quantization (HSQ), which decouples semantic content from feature magnitude via angular routing, preventing code assignment from being dominated by scale rather than meaning. The resulting discrete Representation Autoencoder (dRAE) achieves high-fidelity reconstruction while preserving semantic integrity and supporting scalable codebook budget. Extensive experiments demonstrate consistent performance gains as the vocabulary size scales to 131{,}072, along with 100\% codebook utilization, simplified training pipeline, and strong performance across understanding and generation tasks.

Accurate yet low-latency depth is essential for radar-camera perception in autonomous systems. Cameras provide rich appearance but lack metric scale, whereas automotive radar offers metric range but is sparse and noisy. Many pipelines are multi-stage or depend on auxiliary annotations, increasing latency and limiting portability. We introduce JustDepth, a single-stage radar-camera depth estimator trained only with radar, camera, and single-scan LiDAR. All radar returns are aggregated into a fixed-width 1D representation, decoupling runtime from point count. A Height Fusion Block fuses modalities, a lightweight GNN propagates depth globally, and a training-only confidence decoder stabilizes learning with zero test-time cost. We mitigate stripe artifacts via simple augmentations and quantify them using the Vertical-Horizontal Gradient Ratio (VHGR). On nuScenes, compared to recent state-of-the-art methods, JustDepth maintains accuracy while reducing inference time by 39.7x and stripe artifacts by 66% as measured by VHGR.

Bowel obstruction is a common and potentially life-threatening gastrointestinal condition. In the face of rising diagnostic workloads, the automated diagnosis of bowel obstruction on CT scans supports radiologists by accelerating detection and improving patient outcomes. In this work, we propose a deep learning framework with a multi-task objective that jointly detects bowel obstruction and localizes its transition zone. Additionally, we extend the method with an inherently interpretable classification method that locates the suspected transition point within a slice. It does so by learning a probabilistic selection mask that faithfully bases the classifier's prediction solely on a small image region. The proposed method is evaluated on an internal dataset comprising 1,427 abdominal CTs. Here, the model achieves an obstruction detection test accuracy of 93% and a Hit@10 transition zone localization of 95%. As the first method to reliably localize the transition zone, this marks a significant step towards the automated identification of this critical clinical landmark.

End-to-end OCR systems based on vision-language models have achieved strong performance in complex document OCR, but their efficiency is limited by the large number of visual tokens produced from document images. Many of these tokens correspond to blank margins or visually redundant regions, yet directly applying generic visual token compression methods may remove OCR-critical fine-grained details. In this paper, we propose LayoutLite, a lightweight plug-and-play module for efficient document OCR. Instead of relying on explicit document layout detection, LayoutLite performs implicit layout analysis at the token level between the vision encoder and the language decoder. It aggregates multi-layer visual representations from the vision encoder, and predicts an importance score for each visual token with a lightweight scoring network. Low-information tokens are then removed before entering the language decoder while preserving the original spatial positional information of retained tokens. To train LayoutLite without human annotations, we cast token selection as a reinforcement learning problem and optimize it with a group-relative policy optimization objective driven by OCR output consistency, together with an auxiliary layout supervision signal to stabilize training. Experiments on OmniDocBench demonstrate that LayoutLite can substantially reduce visual token length and inference cost with negligible degradation in recognition quality. We further evaluate LayoutLite on two OCR-specialized VLMs, FireRed-OCR and Logics-Parsing-V2. Under up to 50% token compression, LayoutLite preserves almost the same score on both models while reducing prefill latency, FLOPs, and KV cache memory by over 40%, with only a small additional inference overhead. These results show that token-level implicit layout analysis is an effective and practical approach for accelerating VLM-based OCR systems.

Remote sensing multimodal large language models (RS-MLLMs) have improved general aerial-image understanding. However, Earth observation applications require fine-grained scenario specialization, constrained by scarce high-quality scenario data and incomplete capability coverage. We formulate this adaptation as a capability-gap-driven post-training problem and propose filling before advancing (FBA). Rather than relying on single-stage supervised fine-tuning (SFT) over target-domain samples, FBA first fills prerequisite capability gaps before advancing toward scenario specialization. We instantiate FBA for coastal harbor understanding, a representative multi-source scenario, by constructing CPRS (Coastal-Port Remote Sensing), a three-layer supervision dataset coupled with three ordered stages: (1) RS semantic anchoring for overhead-view visual-language alignment; (2) domain-bridge convergence for shared RS priors across target and bridging scenarios under different modalities; and (3) evidence-grounded scenario tuning for downstream performance. We construct HarborEval, an eight-track diagnostic benchmark covering perception, spatial understanding, robustness, and generation. Under comparable training budgets, HarborEval increases from 57.95 with Direct-SFT to 70.29 with FBA on LLaVA-v1.5, and from 81.09 to 83.37 on Qwen3-VL. FBA also outperforms Collapsed-SFT and leads on harbor-related VRSBench/RSVQA subsets and OpenEval. Stage-wise and role-replacement analyses validate progressive gap filling and stage-specific roles. Public examples and release updates for CPRS, HarborEval, code, and trained weights are available at https://github.com/Z0ngL1ng/filling-before-advancing.

Visual anomaly detection requires adaptive representations and reliable decision boundaries, particularly when anomalous training samples are scarce and class distributions are highly imbalanced. Classical kernel-based methods yield principled geometric decision regions but typically operate on fixed features, while deep detectors learn task-specific representations but often fail to provide an explicit margin-aware kernel boundary. In this study, we propose DLM-SVDD, a deep large-margin novelty-detection framework that jointly learns convolutional features and an explicit kernel-based decision boundary. By drawing on the large-margin $\ell_p$-Support Vector Data Description ($\ell_p$-SVDD) approach, the proposed method performs explicit margin maximization and nonlinear slack penalization while adapting the representation to the target task. To train the proposed model, we present an optimization scheme that alternates between a Frank--Wolfe--based update of the convex dual boundary and a CNN update step operating on a smooth margin-violation loss induced by the recovered boundary. To improve scalability, we analyze the efficiency--accuracy trade-offs for different kernel approximation strategies, deriving practical propositions for large-scale anomaly detection. Extensive experiments on multiple standard benchmarks show consistent performance improvements over the baseline and strong overall performance compared with state-of-the-art methods while illustrating that the proposed joint representation--boundary learning scheme remains effective under severe imbalanced class distributions.

Video super-resolution (VSR) using large-scale Diffusion Transformer (DiT) priors achieves exceptional perceptual quality but is often impractical due to the quadratic computational cost of processing dense spatio-temporal token sequences. Existing efficiency-oriented methods risk irreversible detail loss and temporal flickering, a vulnerability especially pronounced in one-step diffusion models. To address this, we propose TRaM-VSR, a Token Routing and Merging framework for adaptive token allocation, leveraging both context-aware video priors and network-level priors. First, token importance is estimated by fusing motion-sensitive temporal cues with semantic text similarity, isolating dynamic objects and structural boundaries. Next, this importance is further calibrated and adjusted by an offline planner to guide routing across optimally grouped network blocks. Technically, within each routed group, structurally critical tokens are processed in a high-fidelity local stream, while less informative tokens are aggregated into a compact global stream, both modulated by network depth and aligned with the multigranular nature of diffusion models. Extensive experiments show that TRaM-VSR accelerates inference significantly while preserving state-of-the-art reconstruction quality and robust temporal consistency. The code is available at https://github.com/Ree1s/TRaM-VSR.

Recent advances in video diffusion models have spurred interest in human-object interaction (HOI) video generation, which demands fine-grained control over interaction logic beyond single-subject animation. However, existing HOI methods rely heavily on explicit motion control, limiting scalability and generalization across diverse objects and interactions. In this study, we propose AgentHOI, a text-driven HOI video generation following a thinking-before-generation framework that bridges the gap between high-level textual intent and physical execution through multi-agent reasoning over perception, interaction, and motion planning. Building upon the generated interaction plans, we further strengthen text-driven motion understanding. We introduce an implicit text-motion alignment strategy that distills text-to-motion priors into the video diffusion model, enabling robust HOI synthesis without explicit motion inputs at inference. Experiments show that AgentHOI significantly improves interaction naturalness, object appearance preservation, and adherence to complex textual instructions across challenging object-centric scenarios such as wearing and riding. The code is available at https://github.com/bone-11/agenthoi.

The segmentation of multiple degradations has been a challenging problem in the field of image segmentation. Existing level set approaches commonly adopt a length regularization term to constrain the geometric shape of the segmentation contour. However, the introduction of the length term often results in numerical instability and high computational cost. In this paper, we show that the length term is not essential under certain smoothness constraints, and theoretically prove that the presence of the length term affects the property of $|\nabla \phi|=1$. Based on the finding, we define a class of smooth images, construct the grayscale level set, and propose a fast segmentation framework for degraded images, such as heavily noisy images and intensity inhomogeneous images. The framework transforms PDE evolution into one-dimensional threshold search, which has significant advantages in computational speed, especially on large-scale images. Experiments validate the segmentation performance of the proposed framework on various degraded images.

Medical multimodal large language models (MLLMs) are increasingly expected to perform complex image understanding tasks, yet their reliability is often compromised by frequent errors in visual interpretation. To systematically trace these failures, we traverse the hierarchy from high-level clinical tasks down to fundamental visual perception. We therefore introduce Perception-Bench, a large-scale benchmark comprising 1.13 million samples that assesses medical MLLMs across six dimensions: attribute judgment, spatial grounding, spatial understanding, disease prediction, anomaly detection, and report generation, spanning both 2D and 3D radiology images. Our analysis on Perception-Bench reveals that existing MLLMs lack the ability to capture even the most basic lesion attributes, such as location, size, and density. This inability to ground clinical outputs in primary visual evidence reveals that the models' diagnostic unreliability is rooted in a critical but overlooked bottleneck in low-level visual perception. Motivated by this, we propose RadSight, a perception-driven MLLM built upon a dual 2D/3D encoder architecture that preserves native imaging spatial structures. RadSight formulates medical image understanding as a four-stage progressive process: visual-language alignment, fine-grained visual perception, clinical diagnosis, and diagnostic interpretation. The model is trained on an 8.37 million perception-oriented corpus using progressive curriculum learning. On Perception-Bench, RadSight consistently outperforms existing MLLMs across all six evaluation dimensions, with particularly strong gains in spatial grounding and clinical diagnosis. It also achieves consistent improvements on public 2D and 3D medical benchmarks, further demonstrating that robust low-level visual perception is a critical foundation for reliable clinical understanding. Code and model will be publicly available.

Reconstructing dynamic human faces from brain activity provides a powerful way to study how the mind perceives identity, expression, and facial motion. However, progress in fMRI-based face decoding has been limited by scarce controlled, high-resolution neural datasets and by methods that struggle to recover both identity-specific appearance and time-varying facial dynamics. We present fMRI-Face, the first fMRI dataset paired with controllable full-HD digital human facial videos rendered at 1920$\times$1080 resolution. During scanning, participants watched photorealistic, background-free facial videos with controlled identity, expression, and head pose, while fMRI activity was recorded. The resulting dataset contains 62,856 paired fMRI-video samples, providing a structured resource for studying dynamic face perception and reconstruction. Building on this dataset, we propose fMRI2Face, a geometry-guided neural video decoding framework for reconstructing facial videos from fMRI signals. fMRI2Face derives two complementary neural controls from brain activity: Brain-derived Appearance Context, which captures global identity-related visual attributes, and Morphable 3D Facial Control, which provides explicit geometry-aware guidance for pose, expression, and non-rigid facial dynamics. These controls are integrated through Neural-Controlled Video Diffusion with auxiliary latent completion, enabling high-fidelity facial video reconstruction directly from brain activity. Experiments show that fMRI2Face consistently improves reconstruction fidelity, identity preservation, facial geometry, and motion consistency over representative neural decoding baselines. Together, fMRI-Face and fMRI2Face establish a controlled platform for studying dynamic face perception and provide a new benchmark for fMRI-based digital human reconstruction.

In production processes for consumer products, assembly instructions are essential not only for planning but also for executing the production process. Likewise in robotics, it is crucial for an assembly robot to understand how components fit together and can be assembled. To facilitate these tasks, we contribute a method for constructing scene graphs to represent and characterize assembly relationships between components. Our approach does not rely on semantic data and is capable of handling a very small dataset. To realize this, the output of a Faster R-CNN model is used to create geometric representations, which are then processed by a transformer architecture to generate an adjacency matrix. This matrix serves as input to a Siamese network that uses message passing based on an attentional graph convolutional network (aGCN) architecture to characterize the connections between the components. We validate our method on a study dataset of toy model components which can be assembled into transportation vehicles.

Interactive deep image segmentation enables efficient medical image annotation by iteratively refining predictions from user prompts, such as positive and negative clicks. Recent patch-based methods, including nnInteractive, achieve strong segmentation performance but remain limited in annotation workflows by high interaction latency, limited responsiveness to successive interactions, and the lack of support for reversible prompting. Furthermore, evaluation relies predominantly on simulated rather than controlled real-user interaction studies. We present SLIP, an end-to-end trainable framework for interactive 3D medical image segmentation that decouples image encoding from prompt-guided refinement. Image features are computed once and reused, while a lightweight patch memory bank maintains an interaction-aware segmentation state shared across patches. This representation enables prediction updates by propagating interaction context throughout the image, supports reversible prompting without recomputing image features, and substantially reduces interaction latency. By separating image representation from interactive reasoning, SLIP remains compatible with a wide range of image encoders. We train a single SLIP model for general interactive segmentation across diverse anatomical structures and imaging modalities. Beyond standard simulated evaluation, we conduct a controlled prospective user study comparing manual segmentation, nnInteractive, and SLIP across three clinical annotation tasks, six expert participants, and subjective usability measures, addressing the limited human validation of interactive segmentation methods. SLIP achieves SOTA interactive segmentation performance across 13 public datasets while providing lower interaction latency, greater responsiveness, support for reversible prompting, and higher user preference than existing approaches.

We introduce time-reversed imaging, a new paradigm that infers what just happened in a scene from fading multimodal traces. Instead of extrapolating or interpolating video frames, our goal is to infer past human-environment interactions from residual physical imprints observable in thermal, ultraviolet, and visible spectra. To study this problem, we present TRACE-HEI, the first proof-of-concept dataset for time-reversed imaging, containing synchronized tri-modal video sequences of actions such as sitting, touching, moving objects, and liquid spills, captured across diverse materials and recorded up to three minutes after contact. To establish the benchmark, we propose a multimodal inference approach that extracts structured textual descriptions of detected traces and uses them to constrain a vision-language-guided diffusion model for reconstructing plausible past frames. Experiments show that inferring recent events from fading traces is challenging but feasible when complementary modalities reduce solution ambiguity. This work defines the first computational and experimental foundation for time-reversed imaging, bridging vision, physics, and generative reasoning, and opening new directions for scene understanding beyond instantaneous observation.

Inferring physical properties such as mass, stiffness, and elasticity from a single image is essential for simulation and embodied AI, yet most existing approaches rely on multi-view reconstruction or physics-based supervision. We introduce SiPhy, a unified framework for single-image physical property reasoning that aligns 3D-aware visual cues, depth with language-based material knowledge. From one RGB image, SiPhy samples pseudo-voxel points, extracts CLIP features, and grounds them to material candidates proposed by a VLM. A part-based contrastive aggregator enforces region consistency, while a heaviness-aware refinement improves thickness and volume estimation for dense objects. Across ABO-500, MVImgNet-100, and PhysXNet-100, SiPhy achieves state-of-the-art single-image performance, surpassing multi-view reconstruction methods by improving mass MnRE by up to 93% (vs. PUGS), reducing density MAE by 35.5% (vs. NeRF2Physics), and lowering Young's modulus error by 23.5%. We further validate SiPhy on real hand-object interaction datasets, demonstrating its potential as a data annotation engine for physical understanding from single-view imagery.

Few-shot learning enables medical image segmentation models to adapt to new tasks using only a small number of labelled examples. However, adaptation performance depends strongly on which examples are selected for the support set. Effective support sets should capture relevant variation within the target domain and be informative for adaptation, with constituent samples providing complementary information. Despite this, existing active data selection approaches largely prioritise samples individually and do not explicitly account for interactions between examples. In this work, we propose a reinforcement learning framework for support-set selection in few-shot medical image segmentation, enabling support sets to be optimised jointly rather than through independent sample scoring. Given a pool of unlabelled candidate images, an agent directly predicts a support set that maximises downstream segmentation performance. Experiments on a cross-institutional pelvic MRI dataset demonstrate improvements over random selection and current state-of-the-art methods. Our findings highlight the importance of support-set complementarity for effective adaptation and demonstrate the potential of reinforcement learning for optimising adaptation sets.

Efficient processing is becoming increasingly important in infrared remote sensing, where satellite constellations produce large volumes of observations under constrained detector resolution, power, and downlink bandwidth. Multi-frame super-resolution (MFSR) offers a software-based route to spatial enhancement, but its evaluation in infrared sensing remains fragmented across private datasets and ad-hoc protocols. Existing benchmarks do not explicitly capture the thermal contrast, sensor noise, weak texture, and platform-induced frame-to-frame variation that characterize infrared video. We introduce IR275K, a curated benchmark containing 594 infrared video sequences and 275,196 frames. It provides sequence-level train/validation/test splits and a reproducible X4 evaluation protocol. As an initial architectural probe, we further evaluate CGMamba, a lightweight state-space model with 10.90M parameters and 112.14G FLOPs. CGMamba combines 2D rotary position encoding (2D~RoPE) with center-guided cross-Mamba (CGCM) fusion for implicit multi-frame reconstruction. It achieves 33.19dB PSNR, outperforming infrared single-image super-resolution references by 0.35--0.52~dB at substantially lower computational cost. Ablation results show that removing 2D~RoPE from CGCM causes a 1.53dB drop and severe grid-like artifacts. This indicates that explicit spatial anchoring is critical for stabilizing SSM-based cross-frame gating under infrared conditions. IR275K provides a reproducible foundation for accuracy--efficiency evaluation of infrared MFSR methods, while the architectural analysis offers a concrete starting point for spatially aware SSM design under resource-constrained infrared sensing. Dataset and evaluation resources are available at: https://github.com/InfraRecon7/IR275K.

Latent domain watermarking for diffusion models embeds watermarks directly into the latent prior, enjoying non-intrusiveness to model parameters and seamless integration with the generation process. However, due to the violation of latent Gaussianity or sensitivity to normal and malicious perturbations during latent inversion, existing methods are prone to watermark detection or removal attacks. A further overlooked problem is the violation of the i.i.d. latent condition after watermarking, which leads to latent correlation degradation and generation fidelity loss. Although this has been externally measured by FID, the internal correlation structure has yet to be rigorously characterized. To address the above issues, and motivated by the rotation-invariant property of isotropic Gaussian, we propose \textit{Latent Angular Watermarking (LAW)}, which encodes watermark bits as antipodal angles ($\pm\pi/2$ relative to a reference pair) between disjoint pairs of latent elements while preserving the Gaussianity. The antipodal ($\pi$-separation) encoding maximizes geometric separation between bit values, and we prove that the decoding angular-error variance is proportional to the norm of the latent pair, i.e., $\operatorname{var}(\Delta\phi) \propto 1/\rho^2$. We further propose a magnitude-driven variant, LAW-M, which anchors watermark bits in the most geometrically stable latent dimensions, yielding additional robustness gains. Theoretically, we provide a rigorous characterization of the induced correlation degradation, deriving in closed form the autocorrelation structure of the watermarked latent and proving that correlations are confined to a sparse, structured set of off-diagonal elements with fixed $\pm\pi/4$ values.

Recent radiance field methods represent scenes with 2D primitives that offer surface alignment and efficient rasterization, from Gaussian disks to triangles, yet all rely on convex boundaries: curved and concave structures demand excessive primitives. We introduce Deformable Triangle Splatting, which augments each triangle with $K$ control points per edge, each parameterized by a single learnable scalar displacement that shifts the boundary inward or outward, enabling non-convex shape representation while preserving the three base vertices that define the 3D plane. To render these non-convex primitives differentiably, we design a rasterization pipeline in the triangle's barycentric coordinate space, ensuring view-consistent rendering. A winding number test determines whether each pixel lies inside the deformed primitive, and a window function controlled by two learnable parameters, sharpness and corner smoothness, together with a per-primitive scalar opacity, produces the smooth opacity transition from interior to boundary. Validation is done in a variety of real-world scenes, outperforming recent works based on non-volumetric primitives in terms of visual quality and versatility while still achieving competitive rendering efficiency.

Control valve stiction is a common cause of unwanted oscillations and poor control-loop performance in industrial processes. Data-driven methods can automatically detect stiction, but models trained purely on simulated data often struggle to generalize to real industrial control loops due to domain shift. To bridge this gap, this work propose a novel stiction detection methodology that combines optimal transport (OT) imaging technique with deep correlation alignment (Deep CORAL) algorithm. Closed loop signals: controller output and process variable are converted into two-dimensional OT images. These images capture the dynamic behaviour of control loops. The proposed methodology includes a convolutional neural network (CNN) encoder (or feature extractor) trained to learn domain-invariant representations by optimizing a combined objective: a cross-entropy loss on labeled simulation data and a Deep CORAL (covariance-alignment) loss between simulation data and unlabeled target-domain industrial data. Downstream classifiers trained on the domain-invariant target features were evaluated on an independent test set of 20 benchmark loops from industrial stiction data benchmark. The proposed methodology successfully diagnosed 18 out of the 20 loops and achieved100% recall across all 13 stiction cases, an accuracy of 90.00% and an F1-score of 92.86%. Compared to standard baseline approach (hand-crafted features-based method), the proposed methodology significantly mitigates domain shift, providing robust, highly reliable stiction detection for real-world industrial control loops.

Unified multimodal models seek a shared visual token space that supports both multimodal understanding and image generation. Discrete methods unify the interface via a shared codebook, whereas continuous pipelines often rely on two disparate representations -- semantic features (e.g., ViT) for understanding and low-level latents (e.g., VAE) for synthesis -- resulting in mismatched latent spaces. We propose Twins, a unified continuous token space formed by channel-wise concatenating ViT and VAE features on the same token grid, so the sequence length is unchanged and attention cost does not increase. However, jointly modeling Twins in a Diffusion Transformer exposes a severe optimization imbalance: the model fits the ViT component well but struggles to match the VAE latent distribution. We trace this imbalance to three sources of heterogeneity: frequency bias, intrinsic dimensionality, and condition-aligned vs condition-independent uncertainty. To address it, we adapt a focal regression objective for flow matching that upweights large-error VAE dimensions, better balancing optimization across the ViT and VAE components. On ImageNet, this yields up to 10.57 gFID gain over naive MSE loss without classifier-free guidance. Twins also performs competitively on multimodal understanding benchmarks and improves reconstruction fidelity, narrowing the gap between understanding- and generation-oriented representations.

Geometry Foundation Models (GFMs) have substantially advanced monocular 3D reconstruction, yet extending this capability to 4D dynamic understanding remains a fundamental challenge. Most existing motion perception methods (e.g., sparse tracking, dense point-wise flow) treat motion as independent point-wise displacements, ignoring the structured nature of physical motion. However, real-world objects usually obey rigid-body kinematics, and points thus usually move collectively, not in isolation. Motion itself possesses geometric structure: physical objects undergo a set of rigid-body transformations governed by SE(3), rather than unstructured point-wise displacements. Building on this insight, we propose SM4RT, a Structured Motion 4D Reconstruction Transformer for end-to-end 3D reconstruction and structured motion perception. SM4RT introduces Structure-of-Motion to represent scene dynamics, where scene motion is decomposed into a compact set of motion bases, each represented as a temporal sequence of 6D twists in SE(3). Dense scene motion is then recovered by sparse, time-shared per-pixel assignment weights over these bases, ensuring points on the same object share a common rigid-body motion trajectory. SM4RT introduces a parallel motion geometry encoder and decoder that jointly infer 3D geometry, world-coordinate motion, and scene kinematic structure in a single forward pass from monocular RGB video. SM4RT achieves strong motion reconstruction performance while preserving the geometric structure of scene motion.

Large language model agents increasingly solve complex tasks by constructing inference-time workflows that combine reasoning, tool use, and code execution. While such workflows enable flexible problem solving, the useful procedures discovered during execution are often transient: they help solve the current task but are not retained in a form that can systematically benefit future tasks. We present FlowEvo, a training-free framework that compiles successful traces into reusable skill records. Each record pairs a callable artifact with auxiliary structured guidance, and admission applies interface, replay, and safety checks where feasible. These skill records persist in a skill bank at inference time. FlowEvo is organized around three coupled mechanisms: (1)~workflow-to-skill compilation, which extracts reusable executable artifacts from successful traces; (2)~skill-to-workflow feedback, which retrieves accumulated skills to support future problem solving through either direct execution or structured context injection; and (3)~skill curation, which monitors downstream utility and suppresses skills that cause negative transfer. Through this workflow--skill--workflow feedback loop, FlowEvo enables agents to accumulate and refine task-solving capability over time without updating model parameters. Experiments on benchmarks spanning interactive environments (ALFWorld) and code/math generation (HumanEval, GSM8K) show that FlowEvo achieves the best accuracy-cost tradeoff among the evaluated baselines under our implementation settings. On ALFWorld, FlowEvo achieves an 82.8\% success rate, 23.6 percentage points above the strongest baseline, while its average token usage per episode is less than half that of the most efficient baseline. Controlled ablations confirm that each mechanism contributes to the overall result. The code is public at https://github.com/DEFENSE-SEU/FlowEvo.

Evaluating wildfire risk systems using standard machine-learning metrics such as F1-score or IoU is fundamentally flawed: these metrics assess event prediction accuracy, not the operational coherence of a continuous risk signal. This work proposes a novel monotonic evaluation framework that measures whether increases in a predicted risk score consistently correspond to increases in observed operational load, such as number of fires, intervention time, and deployed resources. Moreover, we compare three structurally different approaches on the French Alpes-Maritimes department: the expert-based DFE index, GRU- based predictive models, and FARS, a hybrid multi-agent system combining predictive AI with LLM-based reasoning. Experimental results reveal that the DFE, despite poor classification metrics, exhibits the most balanced monotonic behavior across the full risk scale. GRU models achieve strong local monotonicity but fail to produce well-distributed risk levels. FARS inherits and reveals the structural limitations of upstream signals rather than correcting them. The central finding is a paradigm shift: a good risk model does not predict fires accurately, but one whose ordinal scale meaningfully explains operational dynamics, as proved in this paper. Code of the monotonic framework is available on github.

Multimodal large language models introduce attack surfaces absent in unimodal systems: adversaries can distribute malicious intent across modalities to evade unimodal safeguards. This motivates using cross-modal consistency as a detection signal rather than inspecting each modality in isolation. Our key observation is that benign inputs induce compatible predictive behavior from text-only and vision-only reasoning that stabilizes when fused, whereas adversarial manipulation disrupts this consistency, causing abnormal multimodal behavior. Existing defenses that examine raw inputs or outputs overlook this internal fusion process, rendering them brittle and computationally expensive. We propose FlowGuard, a lightweight inference-time framework that detects harmful inputs by monitoring internal multimodal consistency. Unlike approaches that rely on scalar confidence metrics, FlowGuard derives FlowVectors inspired by Partial Information Decomposition that quantify cross-modal redundancy, synergy, and modality-specific dominance, capturing whether fused multimodal predictions remain aligned with unimodal semantic evidence. In a one-class classification problem trained solely on benign data, FlowGuard reduces Attack Success Rates from >90% to <15% on unseen attacks, with <3% utility loss and up to a 6 times latency reduction. Our results demonstrate that monitoring cross-modal consistency offers an efficient and effective defense for multimodal reasoning.

Public-space gesture interaction is often evaluated as a frame-level recognition problem, but deployed systems expose a different failure boundary. In scenic kiosks, exhibition halls, and service terminals, users experience whether an intended action becomes a stable interaction event, not whether individual hand-landmark frames are correct. We call this the recognition-to-interaction gap. This paper analyzes 8 engineering repair records from a scenic-area interactive kiosk project, covering 4 gesture tasks: two-hand bowing, single-hand fist shaking, two-hand catching control, and knowledge-graph node hovering. From these traces, we extract 20 failure instances and organize them into six non-exclusive working failure classes: model-output degeneration, temporal mismatch, geometric-scale instability, coordinate-rendering mismatch, runtime lifecycle failure, and feedback synchronization and recovery failure. We further organize recurring repair mechanisms into an event-level runtime abstraction between the hand-landmark model and the interaction task. The contribution is deliberately bounded: a deployment-grounded failure taxonomy, an event-confirmation runtime abstraction, and case-study findings. We do not claim a new recognition model, large-scale user evaluation, or quantified accuracy gains.

Accurate latency prediction is critical for deploying large language models (LLMs) on heterogeneous edge devices, where inference latency is affected by model architecture, prompt behavior, runtime backend, hardware utilization, dynamic voltage and frequency scaling (DVFS), and thermal variation. This paper presents a runtime-aware latency prediction framework for deployment-oriented LLM selection. The framework represents each inference request as a hardware-runtime-model-prompt configuration, separates inference into prefill and decode phases, and adaptively fuses static descriptors with dynamic hardware telemetry through a gated prediction model. We evaluate the framework using Pixel mobile devices and validate the profiling pipeline on Jetson Nano, Orange Pi 5 Pro, and an RTX 3090-class GPU platform. On Pixel 8, the full predictor improves total-latency R-squared from 0.953 to 0.960 and decode-latency R-squared from 0.957 to 0.973 over a static-only baseline. On Pixel 8 Pro, it improves prefill-latency R-squared from -1.383 to 0.966. For cross-device transfer, calibration improves Pixel 8 Pro to Pixel 8 total-latency R-squared from -0.974 to 0.940 and decode-latency R-squared from -1.085 to 0.927. Heterogeneous profiling further shows that latency is highly device- and runtime-dependent: the same SmolLM2 model family reaches 8.42 tokens/s on Orange Pi 5 Pro but 64.38 tokens/s on an RTX 3090-class GPU. These results demonstrate that runtime-aware prediction with lightweight calibration can reduce profiling cost and support latency-aware LLM deployment across heterogeneous edge platforms.

Memory-augmented LLM agents maintain context across hundreds of interactions through agentic memory systems that actively curate retrieved content with LLM-generated metadata such as summaries, keywords, and tags. From an inference cost standpoint, every retrieval triggers a full re-encoding of these structured memory units into Key-Value (KV) states, which dominates prefill latency. Existing training-free KV reuse methods mitigate this by selectively recomputing a small fraction of tokens, but were designed for RAG-style raw passages and degrade on structured agentic memories. We present AgentKVShift, a training-free, probe-guided KV residual correction method that operates per retrieved memory unit. One of the crucial insights we demonstrate is that the per-memory KV reuse residual decomposes into a shared memory-level offset plus small token-wise fluctuations. Estimating this offset from a small probe set allows us to correct every reused token by a single weighted correction. Unlike prior reuse methods which decide which tokens to recompute and leave the rest of the cache stale, AgentKVShift also corrects the tokens it does not recompute, turning the refresh budget into useful signal across the entire chunk. Across four open-source LLMs spanning 3B to 32B parameters and two long-horizon agentic memory benchmarks (long-term dialogue and agentic applications), AgentKVShift achieves near full recompute performance while refreshing only 10-30% of the cache, outperforming baselines at the same recompute ratio. It requires up to 5x lower recompute to reach this near-full performance, which prior reuse methods only attain at 45-55% refresh. In this regime, AgentKVShift delivers prefill speedups of 2-3.5x over no-KV-reuse on a single A100. AgentKVShift orthogonally composes with KV cache quantization, retaining over 2x the F1 of prior reuse methods under aggressive 2- and 4-bit settings.

Recent advances in powerful text-to-image generation models have made it increasingly important to develop test-time methods that modify the sampling trajectory to produce images more faithful to complex compositional prompts. We present TILT, a training-free framework for compositional text-to-image generation via test-time reward alignment. We interpret compositional failures as overlap modes between joint and single-concept distributions, and define a reward that favors samples where all concepts are jointly present. This reward is intrinsic to the base model and does not require any external supervision or reward models. This yields a KL-constrained objective with a closed-form tilted target distribution and principled guiding steps for diffusion sampling. The interaction of concept distributions together with the above reward naturally leads to two different guidance strategies while a hybrid approach that balances their respective benefits produces stronger performance. Experiments on prompts from T2ICompBench show that our method improves compositional alignment while preserving image quality compared to previous baselines.

Graph Neural Networks propagate information through local message passing, but the graph topologies themselves can silently prevent any amount of training from solving long-range tasks. When we deploy GNNs on new graphs, there is currently no inexpensive way to know, before training begins, whether the graphs' structures will allow information to travel far enough between distant nodes. We address this gap by proposing Spectral Flow Certificates (SFCs), single scalars computed from the graphs' normalised Laplacians in seconds, requiring no model training and no labelled data. An SFC fuses a graph's algebraic connectivity with the chosen message-passing depth into one number that measures how much of the critical spectral bottleneck can be traversed within the available depth budget. Unlike raw spectral gaps, which are static and depth-agnostic, SFCs adapt as the number of layers increases and therefore carry strictly more diagnostic information when depths vary. Compared with classical structural statistics such as average effective resistance and graph diameter, SFCs explain more than twice as much variance in trained GNN long-range accuracy. Across twenty-five synthetic graph families spanning paths, cycles, grids, regular graphs, and random graphs, SFCs predict trained accuracy before any gradients are computed, achieving explanatory power above ninety percent at all tested depths. The same predictive relationships hold on one hundred fifty real molecular graph topologies drawn from three independent benchmark datasets, confirming that the findings are not artefacts of their synthetic construction. Taken together, these results show that a single eigenvalue computation is sufficient to flag topology-limited graphs before committing to expensive training pipelines, providing a principled first filter for GNN deployments.

Although structured workflows empower Large Language Models (LLMs) to tackle complex problems, automating their creation is severely hindered by a vast combinatorial search space, frequently resulting in inflexible and resource-heavy offline training dependencies. To address this, we conceptualize workflow generation as an intertwined topology-and-execution search paradigm, where the broader topological layer dictates subtask boundaries and lower-level execution outcomes actively reshape the topology itself. Building on this foundation, we introduce HierFlow, a training-free, test-time hierarchical search architecture that automates agentic workflow design by merging feedback-guided topology adjustments with a fast, MCTS-inspired tree search for sub-workflow optimization. HierFlow maximizes efficiency through an intelligent gating module that selectively triggers execution-level searches based on contextual necessity, a mechanism we further support with an in-depth analysis detailing how varying degrees of cross-task coupling impact the effectiveness of hierarchical splitting. Comprehensive testing across question answering, mathematical reasoning, and code generation benchmarks confirms that HierFlow consistently outperforms strong baselines, delivering an optimal balance of high-quality results and computational efficiency without any additional training overhead.

Schema graphs are an upstream bottleneck of schema-grounded information extraction and knowledge graph construction, yet most extraction systems assume the schema is already available. We introduce SCOPE (Schema Construction and Ontology-induction Pipeline Evaluation), a train-text-only benchmark for corpus-to-schema induction and optional schema fusion from raw text, built from 24 public information extraction sources (15 RE and 9 EE) normalized into evaluation-only gold schema graphs; its core event-extraction target covers event types and within-event argument roles, with inter-event links reported separately. We present SCION (Schema Construction and Induction with Ontology Normalization), an auditable reference pipeline rather than a new extraction architecture; it constructs candidate spaces from train text and restricts naming, merging, filtering, validation, and conservative fusion to candidate-linked evidence under strict JSON contracts. On the SCOPE core suite, SCION-lite attains the highest F1 among released source-schema references, Text2Onto-style, LLM-only, and matched extract-then-aggregate baselines under Literal, Fuzzy, Continuous, and Graph schema-graph metrics, while the compact open-model SCION-RL variant reduces reliance on proprietary LLM schema engineers. These results are reported against normalized typed-edge targets rather than as claims that induced schemas surpass human ontology design; the release includes evidence-linked outputs, parse/fallback logs, candidate retention/merging logs, run manifests, code, and benchmark packages at https://github.com/wandugu/paper_scion.

Parameter-efficient fine-tuning methods like LoRA have become the default for adapting large language models, succeeding across instruction following, style transfer, and factual adaptation. We show that for procedural knowledge--the ability to follow multi-step procedures with conditional branching through to terminal states--LoRA fails to match full fine-tuning at the ranks where it retains its efficiency advantage. In a systematic ablation (r = 16--128) on a procedural travel booking task (14 nodes), all LoRA configurations fail uniformly (task success <= 2.54 vs. 4.11 for full fine-tuning, all p < 0.001), with scores decreasing at higher ranks--despite maintaining 95--99% conversation completion rates. Cross-domain replication on Zoom support (14 nodes) and insurance claims (55 nodes) at 8B confirms the failure generalizes: LoRA underperforms full fine-tuning by 0.8--2.2 points on average at both r = 32 and r = 128, with the largest gap on the most complex procedure. Quadrupling rank from 32 to 128 provides marginal improvement but does not close the gap. SVD analysis of the weight changes produced by full fine-tuning explains why: across three domains at both 3B and 8B, the mean effective rank of the update ranges from 761 to 1,026, and rank 128 captures only 43--51% of the squared Frobenius norm. Together, these findings establish that for procedural tasks LoRA falls well short of full fine-tuning--a fundamental limitation for agentic applications.

We investigate where and how transformer-based language models commit to predictions in multiple-choice question answering. We identify the _Hard Decision Layer_ (HDL), a natural architectural property where answer option rankings stabilize abruptly during inference. Empirical validation across four language models (Qwen, Llama, Granite, Mistral) and four benchmark datasets demonstrates consistent HDL emergence without learned routing policies. We also show that the HDL is invariant to fine-tuning. Our results reveal striking accuracy improvements at the HDL: up to +0.61 (Qwen on CommonsenseQA), after which performance stabilizes. Systematic ablations on label formats and problem complexity confirm the phenomenon is fundamental to model architecture. These findings offer mechanistic insights into transformer inference and suggest opportunities for efficient reasoning and model steering. All code and results required to reproduce this work are available in https://github.com/Mystic-Slice/hard-decision-layer

Entity resolution (ER) typically relies on pairwise similarity comparisons between records, which limits its ability to capture indirect relationships present in demographic occupancy data. An important indirect pattern arises from household movement, where multiple individuals relocate together across addresses, but detecting such patterns is difficult due to mixed-format records, noise, duplication, and the absence of stable identifiers. This paper proposes an AI-enhanced framework for detecting indirect entity links associated with household movement in unstandardized name-address data. The approach integrates prompt-based large language model (LLM) named entity recognition for extracting personal names and addresses without extensive preprocessing, semantic text embeddings for robust similarity computation, and graph-based reasoning to infer group-level movement patterns. Experimental evaluation on SPX benchmark datasets (S8-S12) generated using the Synthetic Occupancy Generator demonstrates that incorporating indirect household movement evidence improves recall by 8-15% while maintaining high precision, yielding F1-score gains of 6-8% over a strong pairwise baseline.

The rapid deployment of generative AI has amplified the critical need for Training Data Attribution to ensure transparency and accountability. However, current parametric approaches require computationally prohibitive access to model weights, while similarity-based methods ignore deep structural context. We propose a novel probabilistic framework that operates entirely in a black-box setting. Our method fuses continuous feature similarities with discrete, domain-specific Knowledge Graphs (KGs). This approach ensures the attribution is grounded in structural reality, explicitly rewarding highly specific historical samples while preventing generic background data from dominating the results. We evaluate our framework across two distinct domains where linking outputs to data and domain context is inherently complex: abstract artistic image synthesis and high-dimensional physical weather forecasting. Extensive benchmarking demonstrates the robust efficacy of our approach. In the artistic domain, it achieves a strong Linear Datamodeling Score that exceeds standard black-box similarity baselines, while closing much of the gap to gradient-based estimators. We additionally present a cross-domain feasibility case study in environmental forecasting, where we use domain KGs to retrieve physically consistent historical analogs for regional flood forecasts, improving geographic localisation over a latent-only baseline. Operating entirely without internal model access, our approach provides an efficient, interpretable mechanism for post-hoc influence analysis and domain-grounded retrieval.

Conventional wisdom suggests that reasoning models fail when problems exceed their capabilities. However, we find that frontier reasoning models sometimes possess the necessary capabilities to solve problems but fail due to premature self-doubt -- a phenomenon informally known as context anxiety. We provide the first systematic study of context anxiety, demonstrating that it arises, in part, from a model's inability to accurately estimate the tokens required to complete a task. We also show that context anxiety leads to material efficiency losses when models operate under perceived constraints. Building on this analysis, we further show that models can learn alternative strategies for solving long-horizon problems without exhibiting context anxiety, suggesting that performance improvements may be achievable not through scaling model capabilities, but by improving models' ability to accurately assess and adapt to their own limitations.

Vision Language Models (VLMs) are increasingly used in place of traditional OCR pipelines for document understanding. In this paper, we show they do not always act as faithful transcribers: when text is imperfect, they often tend to rewrite it into a more plausible form - a behavior that clean-text OCR benchmarks cannot detect. We introduce FaithC4, a multilingual perturbation benchmark of 1,455 single-page documents (English, Chinese, Korean) with three perturbation families: scramble, random substitution, and visually similar substitution. We use the benchmark to evaluate 15 systems spanning general-purpose VLMs, OCR-specialized VLMs, and traditional OCR pipelines. These three categories differ in WER degradation under perturbation: general-purpose VLMs degrade by up to 4.5 points, OCR-specialized VLMs by 0.2-2 points, and traditional OCR by less than 0.6 points on English. Probing Qwen3-VL-4B layer-by-layer, we identify a consistent pattern: rewriting fires only when a perturbed word's final layer FFN representation stays close to the original encoding; when the representation diverges sufficiently, the model transcribes faithfully. Word length affects rewriting rate: short words (4-6 characters) are rewritten up to 10% of the time, with a sharp cutoff at 8 characters above which rewriting drops to 0%.

Modern data migration relies on JSON configuration to define data connection, pipeline logic, and orchestration behavior. This requires domain knowledge from users and is time-consuming and error-prone. In this paper, we present LeafData, an agentic system that converts user intent into validated and executable JSON configuration for data migration. Specifically, LeafData comprises a frontend chatbot and the backend service. The chatbot incrementally collects required information from users and performs schema-driven validation, while the backend service processes validated inputs and generates JSON configuration artifacts. These artifacts are directly consumable by orchestration platforms, enabling end-to-end pipeline generation and execution without manual coding. LeafData supports heterogeneous data migration across various data sources and connectors including relational databases, file-based systems, document-oriented databases, and REST APIs.

Personalized text generation requires models to capture user-specific writing styles from historical data. Existing approaches based on retrieval, parameter-efficient fine-tuning, or activation steering either introduce inference and storage overhead or struggle to separate stylistic signals from semantic content. We propose GLASS, a training-free framework for personalized generation via Global-Local Activation Steering with Sparse priors. GLASS uses sparse autoencoders to extract a global user-style prior from historical responses and constructs local contrastive style vectors over clustered interaction scenarios. During inference, it jointly injects global and local vectors into different model layers, enabling context-aware personalization without retrieval or parameter updates. Experiments on LaMP and LongLaMP show that GLASS outperforms retrieval-, fine-tuning-, and steering-based baselines across ROUGE metrics and LLM-as-judge evaluations. Further analyses show that SAE-based representations are more robust to topic and length shifts, suggesting better disentanglement of stylistic information from semantic residue.

Transformer-based models have enabled unprecedented capabilities across language, vision, and multimodal tasks. On-device fine-tuning of transformer models offers a privacy-preserving path to personalized AI, yet remains inefficient on mobile GPUs due to severe memory constraints and frequent layout transformations in attention mechanism during training. Existing mobile training frameworks either use unified layouts for forward and backward passes -- leading to fragmented memory access and poor GPU utilization during backpropagation -- or rely on explicit layout conversions, which introduce significant transformation overhead. To overcome this, we propose FBLayout, a layout-aware framework that co-designs tensor organization with mobile GPU platforms. FBLayout introduces: (1) a unified R-Tile layout for multi-dimensional reductions across forward/backward passes; (2) tile-based index transformation to eliminate physical data movement; and (3) activation-guided layout selection to propagate efficient layouts globally. Evaluations on seven transformer models across different mobile phones (including ARM Mali and Qualcomm Adreno GPUs) show that FBLayout achieves 2.2-5.7x speedup over MNN, TFLite, and TVM, while significantly improving cache efficiency and reducing memory footprint, enabling practical on-device large model fine-tuning.

We study the problem of decision-making from long-form, temporally structured text using large language model (LLM) agents. Standard retrievalaugmented generation (RAG) pipelines fragment chronological context into isolated snippets, discarding the temporal structure that is often critical for correct downstream decisions. We introduce TLM (Trajectory Language Model), a closed-loop agentic framework that iteratively refines the evidence set using SHAP-guided feedback. The key technical contribution is the latent growth curve model (LGCM) over retrieved chunk embeddings, which provides an interpretable mechanism for detecting trajectory trends, turning points, and information gaps. We show that, under a scorer-calibration assumption (which holds approximately in practice), the iterative refinement procedure is monotonically non-decreasing in the probability assigned to the correct label. Empirically, TLM is evaluated on three temporally grounded decision tasks: medical question answering, earnings call surprise prediction, and overnight stock gap prediction. TLM substantially outperforms both zero-shot LLM baselines and standard retrieval-augmented approaches on the medical task, and yields consistent, economically meaningful gains on the two financial tasks.

We study whether Group Relative Policy Optimization (GRPO) can fine-tune small language models for simulated quadrotor continuous-control tasks. In our benchmark, vanilla GRPO fine-tuning of Qwen-0.5B for 25 Hz quadrotor velocity control collapses to the trivial zero action: 0 percent success rate, with entropy falling from 0.35 to 0.03 within 60 steps. Two ablations - removing the jerk-penalty term and removing the KL anchor to the pretrained prior - each prevent entropy collapse, yet neither enables learning. When the action interface is replaced by a 5-way categorical choice over PID presets, training converges. The resulting controller traces a smoothness-reliability Pareto frontier along training duration; both endpoints are reported: 98.6 percent success with 0.656 m/s3 jerk at 64 steps, and 100 percent success with 1.103 m/s3 jerk, or 0.796 under a matched velocity cap, at 256 steps. The recipe is evaluated across three pretrained language models. As context, a re-tuned classical baseline, PID with Ki = 0.30 and vmax = 2.5, reaches the same 100 percent success rate at jerk 0.736 m/s3. A high-fidelity simulation using Crazyflie 2.1 dynamics surfaces a hover-region training-distribution gap.

End-to-end reinforcement learning can improve the accuracy of compound LLM systems, but it does not constrain how modules divide labor internally. We identify Role Drift, a failure mode in which modules preserve or improve end-task performance while deviating from their assigned roles through role-violating shortcuts that remain invisible to system-level evaluation. To make role drift observable and controllable, we propose Role Anchor, a regularizer that modulates how much each module deviates from its assigned role during end-to-end training. The key idea is to preserve how the role prompt shifts the module's next-token predictions relative to a neutral prompt, which serves as a proxy for the role's intended effect during training. Experiments on two compound LLM pipelines reveal role drift that accuracy alone fails to detect: a decomposer meant to split a question into sub-questions for a separate solver instead plants the answer in them, and a reader meant to answer from retrieved passages instead falls back on parametric memory. In fact, on the decomposer pipeline this shortcut drives most of the apparent RL gain: 86% of it vanishes once the decomposer is held to its role, indicating that terminal accuracy alone can badly overstate how much a compound system has genuinely learned. Across both pipelines, Role Anchor mitigates role drift at a tunable accuracy cost that varies by pipeline and anchor strength. Additional gradient analysis suggests that the regularizer reduces alignment with the role-drift direction rather than simply suppressing learning.

Simulation-to-reality translation must bridge the appearance gap between synthetic and real domains while preserving structural and semantic consistency. Conditioning-based methods achieve spatial alignment but introduce computationally expensive control modules. Paired-data methods achieve realism but rely on complex synthesis pipelines, often altering scene geometry and semantics. Training-free editing methods avoid both constraints but lack a learned appearance prior, limiting their perceptual quality. Recently proposed phase-preserving diffusion presents a promising alternative, but Fourier-domain formulations are constrained by global spectral coupling. This coupling induces spatial artifacts such as ringing and boundary leakage, thereby degrading structural and semantic consistency. We introduce Wavelet Phase Diffusion, which addresses this through two components. First, we operate in the Dual-Tree Complex Wavelet Packet Transform domain, whose localized wavelet packets enable spatially adaptive phase injection without global spectral interference. Second, Low-Frequency Randomization (LFR) replaces the low-frequency packet, decoupling the model from the synthetic illumination prior and enabling in-distribution real-world appearance. Both components train on unpaired open-domain data, and introduce negligible inference overhead. The spatial locality further enables instance-level translation, where individual objects or regions are translated to photorealistic appearance independently while the surrounding scene remains untranslated. On vKITTI $\to$ KITTI image translation, ours outperforms prior methods in realism and semantic consistency while maintaining competitive structural alignment. For CARLA video translation, ours approaches the realism of paired-data methods while reducing VLM planner ADE and FDE by $5.4\%$ and $5.1\%$, respectively.

AI has begun to write systems code: agents now synthesize, verify, and deploy system components. Despite this shift, "AI-native" remains a marketing term with no precise technical definition. This paper gives it one. We define AI-nativeness along a single axis---authority over the system's own decisions rather than by the capability of the underlying AI models. Building on a decision-level model of a system, we distinguish occupancy (who executes a decision) from revision authority (who may change it), organize revision authority into a ladder---self-tuning, self-rewriting, self-architecting and define a system as AI-native when an AI autonomously rewrites the system's own implementations. The definition further requires an escalation detector, a verification procedure, and a verified fallback, while leaving purpose and correctness human-owned.

Generative world models are increasingly driven as simulators: a planner forks a state, rolls out futures, backtracks, and returns to a visited viewpoint. Recent benchmarks establish that current video world models fail this usage, and attribute it to the model, prescribing new architectures and training objectives. We show this attribution is incomplete, and for an important class of models simply wrong. Snapshotting the state the runtime already holds -- an observation plus RNG state, a memory bank, or a windowed KV context, by architecture -- and restoring it after a genuine excursion reproduces the never-left continuation byte-identically on all three; corrupting only the RNG degrades it. The capability was never missing: request-centric serving discarded it, inheriting from language-model serving the assumption that runtime state is recomputable -- but world-model state carries a non-recomputable kernel. We define Persistent Computational State (PCS), the minimal non-recomputable state that must survive across requests, show it can be discovered by measurement, and build a session-centric runtime over it. Checkpoint and restore cost 0.012 ms against a 1.85 s generation step; resident sessions become host- rather than device-bounded (measured to 1,024); and world memory must be evicted by relevance to the return, not recency -- the inverse of LLM practice.

Red-team evaluations of AI models support some claims and not others, and the boundary between the two is calculable rather than merely a matter of judgment. We define the evidential ceiling of an evaluation as the largest factor by which one result can move belief under a fixed testing budget, derive it in closed form for the benchmark null result, and use it to locate that boundary exactly. We find that above a calculable harm rate, a benchmark of modest size certifies a category to a stated evidentiary standard, and a clean sheet is then the stronger of the two possible observations, outweighing a single reproduced failure. Below that rate, no passive benchmark of feasible size provides the specified evidence of safety under the fixed scoring rule and approximately independent trial structure. The crossing between the two regimes has a closed form. The bound is not specific to benchmarks: written in terms of a procedure's hypothesis conditioned elicitation rates, it covers adaptive and automated red teaming as well, and shows that discrimination between the hypotheses rather than attack success is what determines evidential worth. Auditing eight evaluation suites against the boundary, we find that current benchmarks are adequate for high-frequency harm categories and several orders of magnitude short for rare, catastrophic ones. Safety benchmarks are not uninformative. They are informative about a specific and computable set of propositions, and the discipline they need is to state which.

Roadblocks in Bolivia are a social conflict phenomenon with devastating economic impacts, estimated at losses equivalent to 4% of the national Gross Domestic Product. Despite their recurrence and impact, there is a lack of local predictive systems to anticipate these events for logistical decision-making. This paper presents a hybrid probabilistic forecasting system that integrates time series decomposition (Prophet) with natural language processing (NLP) techniques applied to a six-year corpus of Bolivian news coverage. The methodology employs vector semantic embeddings and zero-shot classification models to capture signals of discursive escalation prior to the materialization of the roadblocks. Using an expanding walk-forward validation scheme applied over 1,762 days and seven forecasting horizons (H+1 to H+7), seven internal configurations and four external benchmarks were compared, including SARIMA and LightGBM. The results demonstrate that the hybrid configuration (Prophet + NLP, C6) consistently outperforms purely statistical models, achieving an AUC-ROC of 0.677 at H+1 and reducing the Brier Score by 10.9% relative to the baseline temporal model (0.220 vs. 0.247), maintaining a statistically significant error reduction across all evaluated horizons ($p < 0.02$). This research validates that the integration of semantic news signals allows for the detection of social tension peaks not captured by historical inertia, providing a technical tool for risk management in critical transport corridors.

Recent large reasoning models often develop long chain-of-thought responses during reinforcement learning (RL), resulting in high inference latency and deployment cost. Existing methods for response length control typically rely on explicit length penalties or additional control modules, which require careful tuning and may compromise reasoning quality. We propose Quadrant-weighted Sampling for Length-aware Policy Optimization (QLPO), a simple resampling-based variant of GRPO that introduces implicit length control without modifying the reward function. QLPO first over-generates candidate responses and then resamples the training group by preserving the empirical correct/incorrect ratio while favoring short correct responses and long incorrect responses. This reshapes the training distribution and implicitly encourages shorter model outputs. Across models ranging from 1.5B to 32B parameters, including both base models and strong reasoning models, QLPO consistently improves the accuracy-length trade-off. It reduces response length by 30% to 70% while preserving reasoning performance. These results suggest that structured resampling provides an effective and robust approach to efficient reasoning.

Constructing causal directed acyclic graphs (DAGs) is a core step in biomedical causal analysis, yet it remains a largely manual process. Analysts must connect study variables to prior literature, evaluate uncertain causal claims, and preserve sufficient provenance for expert review. We present DAGForge, a browser-based system for authoring causal DAGs as auditable, evidence-linked artifacts. Given free-text descriptions of study concepts, DAGForge creates a reproducible literature snapshot, uses an LLM-based reasoning module to generate structured pairwise causal judgments grounded in verbatim evidence excerpts, and assembles those judgments into a constraint-checked graph. Each proposed edge includes confidence estimates, provenance, and a reviewable rationale. The interface supports study specification, progress monitoring, evidence review, graph comparison, adjustment-set computation, and export. In evaluations against both compact benchmark DAGs and reference DAGs derived from published literature, DAGForge achieves high edge recall on the literature-based cohort while retaining verifiable evidence trails absent from LLM-only baselines. DAGForge thus reduces the burden of causal DAG curation while making the resulting assumptions auditable, supporting the design, analysis, and interpretation of biomedical studies.

Adding a learned adapter to a frozen, command-conditioned locomotion policy is worthwhile only if the interface exposes improvements that are both real and recoverable from deployment-time observations. We introduce an adapter necessity audit that separates global operating-point gain,same-state counterfactual headroom, deployment gain over a cross-fitted fixed action, and state-allocation gain over a frequency-matched randomized policy. Source-cluster learner refits map these quantities and constraint violations to a GO/NO-GO/ABSTAIN decision. Closed-loop command- response identification provides optional decision features. On Go2, an archived scale-prefix diagnostic finds 5.2% same-state headroom but only 0.55% recovered allocation gain. Our confirmatory audit evaluates direct, scale, heading, and yaw interventions on twenty independent clusters for each of three query distributions induced by direct control, VGCC, and MPC, using 200 full learner refits. At 1% deployment and allocation thresholds and a 5% violation tolerance, direct queries return NO-GO, while VGCC and MPC queries ABSTAIN. VGCC has the largest mean deployment gain (1.34%), but its allocation lower bound is 0.09% and its violation upper bound is 6.25%. A deployment-representative twenty-cluster H1 audit also returns NO-GO, whereas a learner-level synthetic control returns GO. The audit therefore tests whether observable signal justifies state-dependent adaptation rather than presuming that an adapter is valuable.

Digital twins have emerged as a foundational technology within the context of Industry 4.0, offering a paradigm for the real-time virtual representation of physical systems. However, managing their growing complexity, particularly in distributed industrial environments, requires intelligent architectures capable of autonomous decision-making, dynamic adaptability, and inter-agent coordination. This systematic review explores the intersection between Multi-Agent Systems and Digital Twins, with a particular focus on predictive maintenance applications in resource-constrained contexts. Through a critical analysis of over 547 papers published in high-impact journals (IEEE Transactions, Nature, Elsevier, MDPI), we establish a taxonomy of existing hybrid architectures, identify persistent technological bottlenecks, and formulate three open research questions concerning: (i) the deployment of artificial intelligence on resource-constrained microcontrollers, (ii) distributed multi-node coordination via lightweight communication protocols, and (iii) the hierarchical orchestration of Digital Twins toward smart factory control integrating residual life estimation and explainable Artificial Intelligence. The results of this analysis reveal that, despite significant progress, no existing system offers an integrated embedded-distributed hierarchical solution that simultaneously meets the requirements of Industry 5.0.

TRACE-RealWorld addresses a core data-management problem: maintaining an actionable materialized view over a continuously changing physical world when reads of the base state are priced, delayed, heterogeneous, and fallible. Its data-management contributions are a commitment-level validity abstraction for materialized predictions; consequence-conditioned adaptive view maintenance; transaction-style, dependency-scoped compensation for commitments invalidated after authorization; and append-only provenance supporting exact replay. The work builds directly on materialized-view maintenance, adaptive stream synchronization, transaction recovery, sagas, data freshness, and provenance. The end-to-end Flood-SAR evaluation treats sensing as physical data acquisition and measures freshness, verification cost, stale reads, recovery scope, restoration failure, and replayability through six pre-registered questions with held-out seeds. The contribution is therefore not a new predictive model, but a consistency, recovery, and accountability contract for deploying learned world representations as operational data systems.

Large language models (LLMs) have become powerful tools for language understanding and logical reasoning. However, they still make mistakes when a problem requires both understanding meaning and following logic. A key reason is that natural-language statements often carry implicit semantic relations before any formal reasoning begins. If these hidden meanings are not properly organized, the model may reach incorrect conclusions even when the subsequent reasoning process appears logically valid. Existing methods improve reasoning through decomposition, symbolic translation, external solvers, or self-verification, but pay comparatively less attention to the semantic structure on which reasoning depends. In this paper, we further investigate how semantic organization influences logical reasoning in LLMs. To this end, we propose HexLogicAgent, a framework that first organizes the meaning of natural-language statements and then guides logical reasoning through structured verification. In our investigation, we also make two observations. First, incomplete semantic representations, rather than deductive inference itself, are a major source of logical reasoning failures in LLMs. Second, explicitly modeling the complete structure of semantic opposition substantially delays the degradation of reasoning performance as logical complexity increases. Experiments on challenging logical reasoning benchmarks demonstrate that HexLogicAgent consistently improves reasoning reliability across multiple LLMs. The core idea is supported by a logical hexagon theory, which explains why a complete structure of opposing meanings is necessary for reliable reasoning.

Reinforcement learning with verifiable rewards (RLVR) has emerged as a highly effective framework for improving LLM reasoning, with methods such as GRPO among its most successful instantiations. However, GRPO relies on repeated generation of long chain-of-thought rollouts. Training time scales with the number of rollouts, a large fraction of which are uninformative. Thus, GRPO is computationally expensive and unstable. To mitigate this, existing approaches either generate a larger pool of rollouts and filter the most informative prompts, or leverage historical signals for filtering at later stages of training. These strategies offer modest performance gains, but slow down the overall process. To address this, we propose VarIance Guided Online Rollout allocation (VIGOR) which instead of allocating a fixed rollout budget per example, begins with a small number of rollouts for all examples in a batch and iteratively allocates additional rollouts to those with the highest group reward variance until a fixed total rollout budget is reached. Theoretically, we show that under RLVR, reward variance controls the gradient magnitude, and derive VIGOR's closed-form speedup ratio over GRPO, which grows with refinement rounds under Pareto-distributed reward variance. Experiments on mathematical reasoning and coding tasks show that VIGOR reaches target accuracy with up to 2.3$\times$ fewer rollouts on math, reaches GRPO's final coding full pass rate with 1.49$\times$ fewer rollouts, and improves the coding average test pass rate by 3.4 points.

Multimodal Large Language Models (MLLMs) are emerging as core reasoning modules for embodied agents, yet it remains unclear how well general-purpose models can solve long-horizon embodied tasks from a single high-level instruction. We introduce MissionBench, a benchmark for mission-level evaluation of MLLMs in aerial 3D environments. It comprises 120 missions across five simulated 3D environments and four task families. Agents must autonomously plan, navigate, and report outcomes using only egocentric observations and its action history, without aerial-specific fine-tuning. Across 22 open- and closed-source MLLMs, the strongest model succeeds on fewer than 35% of missions compared to 84.4% human performance, highlighting the difficulty of multi-step embodied tasks. Despite large variations between model families, we observe gains from scaling, indicating that larger general-purpose models possess stronger zero-shot embodied capabilities. Our analysis shows that mission-level competence requires coordinating multiple capabilities beyond spatial perception, including multi-step planning and adaptive reasoning. This motivates closed-loop evaluation and highlights both the promise and risk of scaling-driven improvements for embodied AI.

We present Nanbeige4.2-3B, a compact general agentic model with 3B non-embedding parameters. It delivers strong performance across code-agent, office-agent, and complex tool-use tasks while maintaining highly competitive reasoning capabilities in mathematics, coding, and science. Nanbeige4.2-3B is pretrained from scratch on 28T tokens with a Looped Transformer that reuses the layer stack to increase capacity without adding parameters. For SFT data and trajectory construction, we expand the diversity of executable environments, task assets, and agentic scaffolds through real-world deployment and large-scale synthesis. Our RL pipeline applies mixed-mode RLHF over Think and Non-Think responses to improve overall model quality and reduce failure cases, length-controlled reasoning RL to balance accuracy and reasoning efficiency, and agentic RL with outcome and process rewards to stabilize long-horizon training. Extensive evaluations show that Nanbeige4.2-3B outperforms larger models, including Qwen3.5-9B and Gemma4-12B, across diverse agentic benchmarks while remaining competitive on reasoning and alignment tasks. Performance with OpenClaw further supports its use as a compact local personal assistant.

Large reasoning models (LRMs) generate long reasoning traces before producing final answers. While these traces may contain useful signals for hallucination detection, harnessing them is non-trivial because long trajectories often include noisy steps that obscure the cues relevant to truthfulness assessment. In this paper, we identify two prevalent forms of reasoning noises, i.e., irrelevant steps and repetitive steps, and show that both substantially degrade hallucination detection performance. Existing confidence-based scores and naive embedding-based filtering fail to reliably separate noisy from informative steps. To address this challenge, we propose REDE, a novel learning framework for denoising reasoning traces for hallucination detection. Specifically, REDE leverages final-answer attention as an automatic supervision signal to shape the step-level representation space, yielding refined embeddings in which noisy steps can be reliably identified and filtered. REDE can be readily plugged into diverse hallucination detectors by operating on the filtered reasoning trajectory after removing noisy steps. Extensive experiments on multiple reasoning benchmarks show that REDE consistently improves detection performance over competitive baselines.

Industrial health management increasingly relies on heterogeneous information sources, including condition monitoring systems, supervisory control and data acquisition systems, maintenance records, inspection results, and prognostic models. Although large language models provide new opportunities for cross-source reasoning, industrial data and analytical outputs differ substantially in structure, temporal resolution, physical meaning, and reliability. Directly integrating such heterogeneous information into a monolithic model may reduce interpretability, traceability, and adaptability to equipment and data changes. This paper introduces Industrial Tokenization, a conceptual interface for transforming source-specific analytical outputs into structured and machine-interpretable units of industrial evidence, termed Industrial Tokens. Unlike numerical tokens used to encode raw time-series data, Industrial Tokens represent domain-grounded evidence together with source, temporal scope, operating context, analytical meaning, quality or confidence information, and provenance. Based on this concept, a federated industrial architecture is proposed, where heterogeneous analytical subsystems retain autonomy while exposing standardized Industrial Tokens to a central reasoning layer. As an initial implementation, this study presents an end-to-end DiagnosisToken pathway based on vibration-diagnostic outputs, rule-based event aggregation, structured textual token generation, and LLM-based interpretation. Other Industrial Tokens, including SCADA-based condition-monitoring tokens, maintenance tokens, and prognostic tokens, are reserved as future extensions. The proposed framework positions Industrial Tokenization as a semantic interface between domain-specific industrial intelligence and LLM- or agent-based reasoning, rather than another method for encoding raw industrial data.

AI agents encounter learning opportunities in every episode they run, and discard nearly all of them: the underlying models are frozen at deployment, so an agent that resolves a difficult request today starts from zero when it recurs tomorrow. Yet ordinary operation already produces feedback, in the form of outcome verdicts and after-the-fact corrections. We show that this feedback is a sufficient signal for continual learning when the frozen model is paired with an external memory that distils each episode into retrievable natural-language rules. On the banking domain of $\tau$-bench, against a static-RAG control retrieving over the complete policy corpus, learning from the one-bit outcome verdict lifts single-trial success to 1.6$\times$ the baseline, and learning from corrections to 2.6$\times$, converting 22 of the 84 tasks the baseline never solves. The result spans the deployment spectrum, measured on Mistral Large, an open-weights model that organisations with data sovereignty requirements can self-host, and replicated on a frontier model, Claude Sonnet 5. The accumulated memory also transfers: each model, reading the store built by the other, rises above its own no-memory baseline. The harness, protocol, and data are released.

Asynchronous reinforcement learning (RL) accelerates large language model (LLM) post-training by overlapping rollout generation with policy optimization, but the resulting stale, off-policy data can destabilize optimization and ultimately cause policy collapse. Existing methods typically retain or discard tokens based solely on the magnitude of their importance ratios, applying the same threshold uniformly across token positions. In this work, we reveal that the natural scale of the importance ratio varies systematically with token entropy. Under asynchronous dynamics, this entropy-ratio scaling dictates two distinct phenomena: at low entropy, the inherent train-inference discrepancy is drastically amplified into substantial sampling noise; at high entropy, in-flight weight updates naturally induce pronounced, legitimate exploratory deviations. Consequently, magnitude-only correction inadvertently admits the amplified noise while strictly masking out the essential exploration triggered by in-flight updates. To address this, we propose the Entropy-Scaled Trust Region (ESTR), which scales each token's off-policy deviation by its local entropy, requiring no auxiliary forward passes or explicit version-switch detection. Across long-horizon agentic tasks and mathematical reasoning benchmarks, ESTR consistently outperforms existing asynchronous methods and achieves the best train-inference consistency. Compared with synchronous GRPO, ESTR attains comparable accuracy while improving training speed by $2.6\times$.

Reuse-based development has become increasingly important in the creation of complex systems, offering significant opportunities to reduce costs, improve quality, and accelerate time-to-market. Product Line Engineering (PLE) provides a systematic approach to realizing this potential by enabling the efficient creation, management, and customization of product families by reusing shared assets and capabilities. PLE involves addressing numerous complex decisions, including feature selection, variability management, and configuration optimization, which are critical to the success of a product line. Despite its promise, the systematic integration of Artificial Intelligence (AI) into PLE processes has not yet been comprehensively explored. In this paper, we propose a methodological framework to support the systematic integration of AI into PLE and evaluate its effectiveness through a multi-case study conducted in an industrial context.

Pluralistic value alignment---the goal of building AI systems that represent and serve diverse human values and perspectives---has emerged as an active research agenda. Yet, there's no public evidence that it has shaped the training or evaluation of the AI systems people actually use. We audit the public behavior documents and evaluations of frontier labs, finding none name pluralism as a goal, and as of this writing, no clear indication that production models are explicitly trained or tested for it. This goes against the primary motivations and goals of pluralistic alignment, which revolve around making a positive difference in the models serving billions of users worldwide. We argue that the pluralistic alignment research community should focus on supporting impact and adoption in deployed, widely-used AI systems. We provide evidence for the adoption problem, present three main reasons behind it, and discuss three corresponding areas for future research to address it: 1. The primary justifications for pluralistic alignment so far have been normative or speculative. We need studies showing empirically how pluralistic AI benefits users or society. 2. The pluralistic alignment research community has not settled when pluralistic behavior is warranted or what pluralism ideally looks like in practice. We need to establish a concrete goal for developers to operationalize. 3. Current methods trade off against other desiderata of LLMs in ways that are largely unmeasured, and existing metrics are not "hill-climbable." We need trade-off-aware evaluations and methods that meet the requirements of production systems. This paper serves as a collective call to action for the pluralistic alignment researchers: progress requires moving beyond normative justification toward empirical foundations, a concrete account of ideal pluralistic behavior, and practical methods and evaluations built for adoption.

High-complexity operational environments require methods that detect and anticipate temporally distributed patterns rather than classify isolated events. This paper introduces TRACTA (Temporal Reasoning and Capability-Trajectory Analysis), a controlled synthetic benchmark for temporal structural reasoning in high-complexity event-driven systems, instantiated through Multi-Domain Operations (MDO)-like scenarios. The benchmark includes three tasks: early_warning, pattern_detection, and run_classification, and compares raw-event neural models, a contract-lite semantic baseline, and a neuro-symbolic configuration operating on semantically grounded trajectories. Results show that raw event-level learning remains informative, but learned temporal modeling over semantic capability and contextual direct-impact trajectories achieves the highest aggregate point estimates, with the largest margins on the temporal tasks. Ablation analysis indicates that capability dynamics, contextual impacts, and temporal structure contribute complementary information. Shortcut diagnostics indicate that the most direct cross-run global-identifier shortcut is controlled in the primary neural input view, while residual shallow signals remain. Overall, the findings support a bounded methodological conclusion: in controlled synthetic settings, semantically grounded trajectories provide an effective representation for temporal structural reasoning, supporting further investigation of semantic interfaces between event data, structured representations, and temporal learning.

Agent benchmarks increasingly evaluate repository editing, web research, terminal use, and long-horizon interaction. Their scores support capability claims only when the evaluation protocol keeps the intended capability necessary for success. Recent reward-hacking benchmarks and system reports show that agents can instead recover public solutions, read evaluation artifacts, infer generator structure, manipulate feedback, or benefit from invalid scoring paths; existing responses do not provide a common procedure for attributing these shortcuts and quantifying their effect across benchmarks. We formulate protocol validity and introduce HackDetect, a post-hoc audit that identifies an exposure, determines how the agent used it, and assesses whether the resulting score is misleading. We quantify score inflation with the Mislead gap, defined as the exploit score minus the intended score. We audit 2,385 traces across 15 agent benchmarks and find evidence of exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks. Across paired comparisons, we measure score inflation of 0.45-1.00, showing that benchmark reports should provide evidence that scores reflect the intended capability.

Large Language Models (LLMs) have significantly automated the process of scientific discovery over the past few years. However, existing systems share one core limitation: they generate and optimize ideas independently for either Quality or Diversity. This often leads to the generation of ideas in close proximity to one another or to a large set of trivial, unsound, or unclear concepts. In this work, we instead argue that research ideation should be treated as a conjunction of both objectives and framed as a Quality-Diversity (QD) search. In line with this perspective, we introduce IDEAgent, a multi-agent framework that manages the evolution of ideas through lineages. We jointly drive Quality using multi-objective feedback for dedicated repair and refinement, while Diversity is achieved through lightweight sequential memory and explicit comparison against completed ideas, their historical ancestors, and rejected proposals. To systematically evaluate this QD conjunction, we develop Yield, a joint metric that computes the largest set of mutually diverse ideas that satisfy a predetermined quality threshold. Finally, through evaluations across 32 topics spanning 8 domains of Computer Science, we show that IDEAgent outperforms the best baseline by 3.89x on Yield, while achieving non-zero Yield on 8x more topics. We further corroborate these findings through an analysis of quality improvements, showing that repair and refinement are crucial for building logical rigor and clarity while preserving non-obviousness. To encourage future research on QD-search-based ideation, we open-source IDEAgent at https://github.com/declare-lab/IDEAgent.

Diagnosing the root cause of anomalies is essential for safe industrial operation. Despite extensive sensor instrumentation, formulating hypotheses and gathering evidence remains a manual process, creating a major operational bottleneck. While existing data-driven approaches aim to automate this, two critical limitations restrict their deployment: their operate as black boxes unable to justify their diagnosis, and they require scarce labeled examples of faulty operation. To address this gap, we introduce AgentRCA, a zero-shot agentic framework for evidence-grounded root cause analysis. Rather than learning fault-specific mappings, AgentRCA performs inference-time reasoning by combining a data-driven digital twin (modeling normal system dynamics) with a tool-augmented large language model. The agent iteratively gathers statistical evidence, evaluates competing hypotheses, and identifies the physical fault that best explains the observed behavior. Evaluated on a real-world multiphase-flow facility and a large-scale chemical plant, AgentRCA achieves diagnostic performance competitive with fully supervised baselines without relying on fault-specific training. Crucially, it produces transparent reasoning traces that explicitly link observed symptoms to their underlying physical causes. These results establish autonomous hypothesis-driven reasoning as a practical foundation for scalable industrial root cause analysis.

Vision-language model (VLM) agents increasingly use tools to act on 3D scenes rather than only describe them. Existing 3D benchmarks score textual responses or single-object operations, leaving agent action on complete multi-object 3D scenes under evaluated. We present SceneActBench, a benchmark for visually conditioned action across five 3D tasks under a unified agent-environment loop. Given PNG images or sampled video frames and, where applicable, supplied 3D assets, an agent acts on a 3D environment. We evaluate each final output against hidden ground truth with task-specific geometric metrics. SceneActBench comprises five tasks built from 210 source instances, yielding 520 task cases including paired input conditions. Every task runs through one fixed agent loop to keep the comparison fair. Across eleven proprietary VLM configurations, Overall scores span 38.6-50.2, and none performs consistently well across tasks. We further analyse where and how failures manifest.

Enterprise AI agents are typically granted static credential sets at configuration time, holding every tool the role might need for every task they perform. This persistent over-privilege expands the attack surface. We argue that capability scoping must follow a dynamic least-privilege principle and be treated as a prevention mechanism before a detection one. A credential that does not exist in an agent's context cannot be misused regardless of the agent's reasoning or evasion sophistication. We outline a three-source architecture instantiating this principle: role-based ceilings, a task-context classifier, and policy-derived combination prohibitions creating a layered proactive defense against LLM agent misalignment and misuse cases. The architecture supports both enforcing and observe-only deployment; the latter records agent permission requests inconsistent with task context, producing a behavioral signal usable in misalignment research. As a first step toward evaluating this architecture, we contribute a synthetic dataset of 600 enterprise task prompts grounded in a multi-department company policy, labeled with minimum required permissions across a 15-permission tool-based taxonomy that maps directly to deployable credentials or enforceable guardrails. The dataset is constructed via a two-pass pipeline that separates prompt generation from permission labeling to avoid circularity, and is validated against a 60-record/688 decisions human-reviewed sample (Cohen's $\kappa = 0.917$ pre-review and $\kappa = 0.967$ post-review). Iterating between dataset and policy reduced ceiling violations from 46 to 3, a 93% reduction. This shows that synthetic prompt generation can drive policy refinement when the two are developed together. The dataset, environment specification, and generation pipeline are released to support evaluation of dynamic scoping mechanisms.

Routing to select large language models (LLMs) with different cost-quality trade-offs has become a fundamental deployment feature of enterprise AI. Existing routers, primarily make independent routing decisions for each LLM call. However, agentic applications execute as long-horizon workflows whose quality is determined only by a delayed, task-level outcome. This mismatch prevents per-call routers from correctly attributing feedback to individual routing decisions. Towards mitigating this, we present TRACE-Router, a task-level routing framework that aligns routing with the unit of supervision. TRACE-Router assigns each task to a model once at admission using a contextual bandit, pins all subsequent LLM calls to the selected backend, and updates its policy using the task's terminal reward, jointly accounting for accuracy and latency. By leveraging delayed task feedback, TRACE-Router learns routing policies that adapt to the workload while avoiding explicit task-complexity estimation. Across three agentic benchmarks, TRACE-Router consistently improves the accuracy-latency trade-off, achieving non-dominated Pareto frontier points. On tau2-Bench, it outperforms latency-matched interpolation between individual models by 7-8 accuracy points, while on Terminal-Bench it achieves 7.1 higher accuracy points than the strongest single model baseline with 36% lower latency.

Adding procedural skills to an LLM agent is typically evaluated by average improvement in task success. However, this metric hides an important cost: skills can also make agents worse. We measure both sides by comparing agents with and without skills across nearly 6,000 runs spanning two office automation benchmarks and three model harness stacks. This allows us to distinguish two outcomes. A regression is a task solved without skills but failed after skills are added. A residual failure is a task that fails both with and without skills. We find that regressions are substantial enough that the best performing skills outperform others primarily by regressing less, not by gaining more. We identify three causes of regression: (i) skill description osmosis, a skill changes an agent's behavior simply by being present in context, even when it is never invoked; (ii) grounding displacement, a skill's prescribed procedure overrides how the agent interprets its inputs; and (iii) verification displacement, where the procedure suppresses checks the agent would otherwise perform on its outputs. Analysing persistent failures reveals the same underlying pattern. Existing skills overemphasize procedural guidance the stage least often responsible for failure while under supporting grounding and verification, the dominant sources of remaining errors. After correcting evaluation artifacts and studying traces, we find many regressions and persistent failures recoverable through better grounding and verification. Procedural skills should be evaluated by decomposing their net effect into gains and regressions, not by aggregate improvement alone. We identify three regression modes skills should avoid, and find that reliability depends more on grounding and verification than on procedural skill choice.

To effectively integrate AI into high-stakes, critical environments such as healthcare, autonomous driving, and aviation--and to advance toward higher levels of automation and seamless human-AI collaboration--building trust in AI-driven solutions is essential. Trust, in turn, is closely linked to the explainability of AI systems. The rapid advancements in AI across various domains have underscored the challenges of establishing trust, raising increasing interest in AI explainability even more when applied to deep learning. In this context, the present work aims to explore the application of explainability techniques to Reinforcement Learning (RL) algorithms, specifically within the safety-critical domain of Air Traffic Control (ATC). Using a simplified ATC environment as an initial testbed, an intelligent agent is trained with a reinforcement learning algorithm to make decisions on alternative flight routes that avoid no-fly zones. As a preliminary explainability approach, a saliency map is employed, providing insights into the input features that most significantly influence the agent's decision-making process.

Multimodal Large Language Models (MLLMs) have achieved impressive performance, but their safety alignment remains vulnerable to jailbreak attacks. Existing content-based jailbreaks are often inconsistent and show unsatisfying performance against the rapidly evolving MLLMs, failing to exploit non-content-based vulnerabilities. Unlike previous research, we empirically find that MLLMs exhibit a Stylistic Inconsistency between their comprehension ability and safety ability: MLLMs can robustly understand content regardless of visual style, yet their defense mechanisms can be easily bypassed by specific stylistic triggers. Based on this finding, we propose Adversarial Style Optimization (ASO), a plug-and-play enhancement module to amplify existing visual jailbreaks. ASO fine-tunes an image-editing model to superimpose an optimized stylistic modification onto a given adversarial image, using a Group Relative Policy Optimization (GRPO) agent guided by a Structurally-Tiered Reward Function that combines a logit-based signal for detecting explicit refusals with a high-fidelity semantic evaluation from a powerful judge model. Extensive experiments show that ASO significantly enhances the ASR of SOTA attacks, demonstrating that stylistic biases are a scalable vector for red-teaming MLLMs. Our code is available at https://github.com/bingjunluo/ASO.

Traditional benchmarks for LLMs primarily rely on static datasets and objective scoring metrics, which often fail to capture differences in response quality when multiple answers are acceptable. In such settings, correctness alone is insufficient to distinguish between responses that vary in clarity, completeness, and usefulness. This paper introduces a consensus-based evaluation framework that measures relative preference among model-generated responses rather than absolute correctness. Instead of evaluating outputs against a fixed ground truth, we assess how a panel of diverse LLMs ranks anonymized candidate responses to the same prompt. This approach treats aggregate inter-model agreement as a proxy for perceived response quality under blind conditions. We conduct a controlled study using five state-of-the-art LLMs across multiple domains, including programming, general knowledge, safety, logical reasoning, and mathematics. Each model generates responses and independently ranks peer outputs through a structured voting process. Scores are aggregated into a Relative Intelligence Index (RII), representing how frequently a model's responses are preferred by other models. Our findings reveal consistent preference patterns across domains, with certain models more frequently ranked highly by their peers. However, we emphasize that these results reflect inter-model preference alignment rather than objective correctness or human judgment. This framework provides a scalable, model-driven method for comparative evaluation, offering an alternative perspective on response quality in scenarios where multiple valid answers exist. While not directly aligned with human evaluation, prior work suggests that aggregated model preferences can partially correlate with human judgments, motivating this as a proxy signal.

A systematic review begins with someone reading thousands of abstracts to identify the few that are relevant, and classifiers are used to prioritise that reading. Their inputs are often augmented with Medical Subject Headings (MeSH), assigned either by expert indexers weeks or months after publication or by automatic tools at once. To our knowledge the two have not been compared directly as classifier features, and no previous work has asked whether that comparison's outcome depends on how the classifier is evaluated. Using the Cohen et al. (2006) drug-class benchmark on three topics, we characterise a bag-of-words logistic regression classifier (seven reruns) and BiomedBERT (five seeds), then examine how the Statins result changes under alternative designs. Under the canonical 5-fold full-corpus design, the bag-of-words expert-vs-auto gap on Statins is +0.096 WSS@95%. Matching the corpus size to the smaller topics (n = 803) reduces it to +0.033 (95% bootstrap CI includes zero), and 10-fold cross-validation at full size to +0.021 (CI narrowly excludes zero). Under canonical evaluation BiomedBERT gives +0.020, within sampling noise of the bag-of-words 10-fold result. A power analysis indicates a Statins-sized effect would not have been detectable at the Opioids or ADHD variance, so those nulls are design-limited rather than informative. A representation asymmetry remains: 15.1% of Statins inputs exceed BiomedBERT's 512-token limit when expert MeSH terms are appended, so truncation may contribute to the smaller transformer gap, although this cannot be separated from training volume here. In screening pipelines using transformers or 10-fold bag-of-words, the gap on the topics tested is about 0.02 WSS@95%, with CIs spanning zero on at least one bound. More broadly, benchmark conclusions about feature sources can change substantially under reasonable changes to the evaluation design.

Teachers, conference chairs, and public readers all judge writing from limited evidence, seeing only a finished document and not the process that produced it. Final text alone cannot reveal whether a document was produced through human typing, AI generation, or mixed human-AI collaboration. Existing process-tracking tools help, but many are tied to host-document histories, provide coarse activity records, and offer limited control over the writing environment. Humanly is a writing platform that makes the writing process itself the evidence. Users configure writing environments for personal documents or assigned tasks and draft in a workspace that records writing activity and in-platform AI assistance. Humanly can package a completed session into a sealed writing certificate with configuration-aware anomaly behavior review. It can support writing scenarios such as course assignments, peer review, and personal certification. Our user study shows that Humanly is helpful across roles, and a red-teaming study shows that the Humanly Typing Detector distinguishes human hand typing from automated typing.

Large language models may infer demographic attributes from subtle linguistic cues even when those attributes are not explicitly stated. This pilot study examines whether Qwen2.5-7B-Instruct internally represents Colombian identity, socioeconomic status, or stereotype-related information when processing Colombian-Spanish and English prompts. We use Natural Language Autoencoders (NLA) to verbalize residual-stream activations from layer 20 across four positional quartiles per prompt. Our dataset contains 30 prompts arranged as 15 matched Spanish-English pairs, spanning explicit Colombian cues, implicit Colombian cues, and neutral controls. We report descriptive rates and qualitative evidence rather than statistically powered effects, focusing on whether latent nationality or stereotype representations appear before they are verbalized in the model output. This work connects activation-level interpretability with bias evaluation for underrepresented Spanish varieties.

Document packets, multiple documents concatenated into a single file, are common in government and administrative workflows, yet splitting them into their constituent documents is difficult, especially for low-resource languages. We introduce Khondo (Bangla for split/segment), the first benchmark for document packet splitting on Bangladeshi government forms. Unlike prior English and OCR-text-based datasets, Khondo is bilingual (Bangla--English) and vision-native; where models operate directly on page images. It spans five concatenation schemes, from sequential to fully shuffled, across 14 administrative domains, with ground-truth boundaries, domain types, and page order. Zero-shot evaluation of MLLMs shows they cluster pages into their source documents fairly well but struggle in restoring the original page order once shuffled. To isolate what drives this difficulty, we run two controlled analyses, varying the prompt instruction and then the packet language. Both primarily affect ordering rather than clustering: (a) explicit page-order instructions are necessary but insufficient, and (b) English packets are ordered more reliably than Bangla, making page arrangement the dominant challenge and language a secondary but consistent factor. Khondo establishes page-order reconstruction as a key open problem in vision-based, low-resource document understanding, and provides a controlled benchmark for measuring progress toward solving it. Our dataset and code is available at https://huggingface.co/datasets/Mausul/khondo

Large language model (LLM) agents increasingly perform commercial tasks that involve retrieving external content such as images and, where appropriate, reproducing that content. LLM agents should comply with the law, including copyright law. Presently, however, we lack adequate frameworks to assess whether they do so in practice. To that end, we introduce \textbf{Copyright-Bench}, a benchmark designed to evaluate \textit{LLM agents' compliance with} \emph{copyright law}. Copyright-Bench is comprised of realistic commercial tasks---website development, merchandise design, and pitch deck production---that involve agents selecting between public-domain content (the use of which is \textit{legal}) and copyrighted content (the use of which is \textit{infringing} in this setting).The evaluation introduces prompt variations that simulate different user preferences, as well as time pressure.Comparing state-of-the-art LLM agents against a human baseline, we find that: (1) agents select copyrighted works despite the availability of public-domain alternatives; and (2) for open-weights models, violation rates increase in response to certain user preferences and simulated time pressure.

We study baking documents directly into the weights of a 4-bit Gemma-4-e4b model via LoRA, so a system can answer questions about a corpus closed-book: no retrieval and no context-window budget. Across roughly 100 training runs from single documents to a 99-document corpus, we find that once adapter capacity is adequate, training-data quality is the dominant lever on closed-book accuracy, outweighing LoRA rank, learning rate, and two alternative architectures combined; capacity itself is a hard gate below which no data intervention helps. A single curation pass (shortening gold answers to canonical 1-6 word spans and dropping trivia) moved closed-book accuracy from 57.7% to 85.7% on a 15-document corpus, a larger jump than any architectural change. We confirm a capacity trend (rank must grow with corpus size) entangled with a coupling between rank and learning rate that we initially misdiagnosed. On a 15-document slice we add a real retrieval baseline: the internalized adapter (84.2% recall) beats a BM25-RAG pipeline with a base reader (58.9%) and even a realistic gold-chunk oracle (65.6%) at lower latency. We report the full arc, including three misdiagnoses, as a case study in debugging LLM training empirically.

Historical documents act as invaluable knowledge archives but often suffer from illegibility due to physical deterioration and damage. While existing restoration methods based on masked language modeling effectively utilize local context, they struggle to restore named entities that require external historical knowledge. To address this limitation, we introduce a novel framework for historical document restoration that leverages large language models with retrieval-augmented generation (RAG). By combining the implicit knowledge of pre-trained LLMs with explicitly retrieved external context, our model ARI effectively mitigates the challenge of inferring context-dependent proper nouns. Extensive experiments on Korean historical documents demonstrate that our approach significantly outperforms baselines, achieving substantial gains in restoring both general characters and named entities. Furthermore, comprehensive evaluations including expert assessments confirm that ARI serves as a practical tool for domain experts, promising to accelerate the analysis of historical records.

Large language models (LLMs) are increasingly used to generate long-form conversational content such as podcasts from textual sources. While these systems produce fluent and engaging narratives, they often introduce ungrounded information. In this work, we present the first systematic study of faithfulness in document-grounded podcast generation, where grounding must be maintained across conversational turns in long-form, multi-speaker transcripts. We construct a dataset of over 1500 documents spanning five domains and generate podcast transcripts using multiple LLMs. We introduce a turn-level LLM-as-a-judge framework for evaluating whether conversational turns are supported by the source document, and validate its reliability through human studies. Our analysis shows that even state-of-the-art models, including GPT-4o, frequently generate ungrounded content. To mitigate this issue, we propose catch-n-repair, a model-agnostic framework that detects and rewrites unfaithful conversational turns while preserving conversational flow. Experiments demonstrate consistent improvements in faithfulness across both in-domain and out-of-domain settings.

Benchmarks for LLM-agent memory typically generate conversations first and extract answer keys afterwards -- with documented label-error and contamination problems -- and they overwhelmingly measure short interaction histories. We invert the pipeline: a seeded life-script sampler emits facts with validity intervals, volatility classes, and source channels before any text exists; an LLM renderer writes chat and email from per-event fact manifests; a fidelity verifier confirms every planted fact; and questions are instantiated mechanically from the script, so gold answers are script-valid by construction and separately validated for answerability. The synthetic, fictionalized corpus (~380 questions, 15 types) embeds features absent from the benchmarks we survey: per-fact validity intervals, sent/received trust distinctions, injection probes in a benign harness, and as-of-date question sets. Benchmarking five memory architectures against a no-memory control (fixed answerer, versioned LLM judge, three replicates, two horizons), we find backend rankings invert with history length: the budgeted curated-map memory that leads at three weeks loses recall of evicted content by nine weeks (96% to 72%) while a provenance-typed graph rises to 90%; the inversion is positive for all six users under complete cross-family re-judging (exact p=0.031). A full-rendered-history baseline ties or exceeds the best memory system at the short horizon but shows no judge-independent advantage at nine weeks, at about twice the read cost. Write-stage quality strongly correlates with downstream quality (weakly-written facts fail 24% vs 2%), and injection resistance tracked whether provenance boundaries survive representation. A layered architecture performs best among the memory systems in both regimes (96.8% short-horizon) and is released as Veracium, an open-source library, with the corpus generator and harness.

Mixture-of-Experts (MoE) architectures have been widely adopted in large language models, yet parameter-efficient fine-tuning (PEFT) for MoE models remains underexplored. Existing PEFT methods for MoE either ignore router priors with uniform adapters, reducing efficiency and risking forgetting, or rely on static expert selection, limiting per-token capacity and cross-expert feature learning. In this paper, we make the first attempt to fine-tune MoE models with MoE-style low-rank adaptation: our method, entitled MoE$^2$-LoRA, deeply couples the pretrained expert specialization with task-specific adaptivity via a dual-channel Routing-Conditioned Projection (RCP) module, which reuses base router activations to inform LoRA routing. We further introduce a single global LoRA expert pool shared across all layers, enabling model-wide adaptation with emergent layer-wise affinities and balanced expert utilization. MoE$^2$-LoRA simultaneously benefits from the advantages of prior reuse, dynamic adapter routing, and model-wide knowledge sharing. Evaluated on multiple MoE backbones with varying scales and expert granularities, MoE$^2$-LoRA consistently achieves state-of-the-art downstream accuracy while retaining stronger general capabilities.

Mastodon as a decentralized federation of independently moderated social servers poses unique challenges for the detection and mitigation of toxic content. There are no unified moderation standards. The ecosystem is very diverse and uneven. This paper explores the development and spread of toxicity in Mastodon, utilizing machine learning methods to examine user posts. The results offer clarity on toxicity trends and its implications for community health and decentralized governance.

Chain-of-thought prompting improves language-model reasoning by carrying intermediate states across successive computation steps. However, relying on natural language as the only recurrent interface is overly restrictive, since many transient computations do not need to be fully verbalized. Existing latent-reasoning methods remove this constraint by recurrently propagating continuous hidden states. However, these methods pass a dense hidden vector as a whole, without an explicit mechanism for selecting and organizing the information needed by the next reasoning step. This motivates an intermediate interface that remains linguistically grounded without requiring a decoded sentence. We introduce \textbf{J-CoT}, a recurrent reasoning framework built on \emph{J-space}, a vocabulary-indexed coordinate system within the model's hidden representations. Within each cycle, the model computes in its full hidden space. At the cycle boundary, J-CoT expresses the intermediate state as vocabulary-indexed coefficients, carries these coefficients forward as a \emph{J-thought}, and maps them back into the model's hidden representation for the next cycle. J-CoT therefore requires neither a fluent intermediate rationale nor recurrence over the complete hidden state. Under matched backbone and inference settings, J-CoT-Zero matches or exceeds the strongest evaluated latent-reasoning baseline on every benchmark, while J-CoT-Train obtains the highest score across the evaluated mathematical, scientific, coding, and structured path-reasoning tasks.

Self-harm content is particularly challenging to detect using NLP techniques, and is also a high-stakes task which requires the highest accuracy to enable timely intervention or flagging at-risk users. We therefore present an analysis of how LLMs represent such self-harm content, which has downstream applications in self-harm detection, LLM intervention and governance and policing. In this paper, we focus on two datasets and four models, and perform two main experiments: (1) We train and evaluate linear probes across all layers of each model on two self-harm datasets: X-Sensitive and SH-Detection. Across both corpora, self-harm information crystallizes in the final 3 - 7% of network layers (93 to 97% depth). (2) We extract contrastive self-harm directions and, after performing a normaliation step, we find that the most accurate probes are not necessarily the most linearly separable. In particular, we find Gemma-3-4B to represent this \textit{contrastive self-harm direction} in a slightly different, more intricate way than the other LLMs.

Detecting LLM-generated text remains challenging under zero-shot and training-free conditions, especially when detectors must generalize across datasets, domains, and unseen generators. While existing training-free approaches exploit language-model statistics as detection signals, they typically characterize a text through global measures that summarize overall model behavior. Consequently, potentially informative local and multiscale variations in token-level predictability may remain underutilized. Motivated by this observation, we introduce DWT-Fusion, a training-free signal-based framework for detecting LLM-generated text using discrete wavelet analysis of token-level log-probability sequences produced by a proxy causal language model. The proposed framework analyzes these sequences through wavelet-based multiresolution signal representations and derives detection signals from localized probability dynamics. We further evaluate four training-free voting variants, including equal-weight hard voting, equal-weight soft voting, calibration-weighted hard voting, and calibration-weighted soft voting, to combine multiple wavelet configurations without training a supervised meta-classifier. We evaluate the framework on HC3, M4, and MAGE using GPT-Neo-2.7B, GPT-J-6B, Falcon-7B, and LLaMA-3-8B as proxy models. The best single wavelet configurations achieve AUROC values of 0.9872, 0.8185, and 0.7138 on HC3, M4, and MAGE, respectively. With calibration-weighted voting, the best ensemble variants further improve AUROC to 0.9919, 0.8477, and 0.7471. These findings show that DWT-based multiresolution scoring and calibration-guided voting fusion provide effective and interpretable signals for training-free LLM-generated text detection.

Model merging plays a crucial role in consolidating multiple specialized models into a single, unified model, especially in the era of large language models (LLMs). Recent research has primarily focused on developing strategies to enhance merging performance with the trained models, while the impact of training paradigms, such as supervised fine-tuning (SFT) and reinforcement learning (RL), on the effectiveness of model merging remains underexplored. In this study, we systematically explore the merging behavior of RL-trained LLMs compared to those trained with traditional SFT. Through comprehensive evaluations across five representative tasks, we find that RL significantly reduces task conflicts and results in less performance degradation after merging, making RL-trained models particularly well-suited for this process. To unearth the reasons behind the superior suitability of RL for model merging, we conduct extensive empirical experiments and theoretical analyses. Our findings highlight three key factors: (1) On-policy training data in RL control the gradient updates in a smaller magnitude, reducing the risk of overwriting existing knowledge for other tasks in the model. (2) The RL optimization objective, which favors ``\textit{enough is as good as a feast}", progressively reduces the magnitude and the number of conflict parameter updates as the model converges. (3) Joint optimization of positive and negative examples in RL steers the model towards an unbiased task-specific parameter subspace, ensuring robust performance while further preventing parameter conflicts.

There is a growing need for reliable and culturally validated instruments to assess psychological dependency on large language models (LLMs), particularly as LLMs are increasingly used for task execution, decision-making, and communication in organizational and work-related settings. This need is especially relevant for Spanish-speaking populations, where LLM adoption is rapidly expanding, yet validated psychometric tools remain scarce. The present study reports the first validation of the Spanish version of the Large Language Model Dependency Scale (LLM-D12-SP), extending prior validations conducted in English- and Arabic-speaking samples. The LLM-D12 is a two-dimensional instrument assessing Instrumental Dependency (reliance on LLMs for performing tasks and supporting decisions) and Relationship Dependency (psychological reliance on LLMs for companionship and social interaction). A total of 386 Spanish-speaking participants (M = 28.0 years, SD = 6.1; 55% male) completed the LLM-D12-SP. Confirmatory factor analysis supported the original two-factor structure. The scale demonstrated good internal consistency (Cronbach's alpha = 0.89 total; 0.86 Instrumental; 0.85 Relationship). Discriminant validity analyses indicated that the two subscales represent related but distinct constructs. External validation showed that both dependency dimensions were positively associated with internet addiction and perceived trustworthiness of LLMs, while showing weak or no association with need for cognition. Together with prior English and Arabic validations, these findings establish cross-linguistic support for the scale's structure and provide a psychometrically sound tool for investigating psychological aspects of LLM use in organizational contexts.

Although large language models (LLMs) exhibit remarkable reasoning capabilities, their reliance on text-only pre-training restricts the perception of the multimodal physical world. Native multimodal pre-training avoids this limitation by training models from scratch on multimodal inputs, thereby achieving deep cross-modal integration and mitigating optimization asymmetries inherent to traditional late-fusion architectures. Despite these advantages, the scaling properties of this paradigm remain systematically uncharacterized. To address this gap, we investigate the optimal model size and token count for training a transformer-based vision-language model under a fixed computational budget. We demonstrate that minimal objective loss adheres to a predictable compute law, whereas compute-optimal model sizes and token counts scale as power laws. Notably, language and multimodal objectives manifest distinct scaling behaviors. The language allocation law is largely invariant to the composition of the data, indicating stable language learning regardless of the multimodal data ratio. Conversely, the multimodal allocation law is highly sensitive to this composition. Specifically, text-heavy mixtures become compute-efficient only at larger model scales, shifting the optimal resource allocation toward greater model capacity. Additionally, by modeling the influence of data composition on compute laws and allocation exponents, we derive an efficiency frontier specifying precise configurations of model size, token count, and data mixture. Downstream evaluations further reveal that native multimodal pre-training induces positive cross-modal transfer, thereby enhancing pure-text spatial reasoning and enabling robust multimodal in-context learning. In summary, this empirical research establishes the essential groundwork for predictably scaling multimodal foundation models.

The integration of large language models (LLMs) into the nuclear power industry requires outputs grounded in domain-specific knowledge. This study evaluates a 31-billion-parameter open-weight multimodal model (Gemma 4 31B-IT) on its capacity to apply nuclear knowledge by benchmarking eight model-retrieval configurations against the U.S. Nuclear Regulatory Commission (NRC) Reactor Operator licensing examination. We evaluate 14 Generic Fundamentals Examinations (GFE) from the 2015-2021 March sittings (seven pressurized and seven boiling water reactor exams) using the standard 80% human passing criterion. The base model is compared against configurations utilizing supervised fine-tuning (SFT) on Gemini-distilled chain-of-thought (CoT) rationales, retrieval-augmented generation (RAG) with BM25 sparse retrieval over the U.S. Department of Energy Fundamentals Handbook, and retrieval-augmented fine-tuning (RAFT). Within the retrieval pipeline, we compare fixed-size sliding-window chunking against structure-aware chunking. The SFT configuration with fixed-size chunking RAG met the criterion on 8 of the 14 examinations, outperforming all alternatives, whereas no configuration without fine-tuning passed any. Aggregate accuracy reached 79.7%, with a confidence interval spanning the threshold, and 80.2% on PWR items specifically. Furthermore, two regularities emerged: the preferred chunking strategy reverses depending on the model's training state, and RAFT underperforms compared to standard SFT in matching search environments. These results demonstrate which combination of fine-tuning and search approaches achieves operator-level capabilities.

Continual Learning for Named Entity Recognition (CLNER) enable models to incrementally learn new entity types without forgetting previously acquired ones. However, existing methods suffer from catastrophic forgetting and insufficient exploitation of shared information across tasks. This paper proposes FSE, a Fast-Slow Experts enhanced span-based NER model for CLNER. The shared fast expert learns token-level links to efficiently filter out unlikely spans, while the task-specific slow expert performs span classification only on the remaining candidates. It stabilizes learning by promoting knowledge sharing across tasks and maintains plasticity by reducing learning burden at each task. A length-decay negative sampling strategy to mitigate span imbalance is also introduced. Extensive experiments on OntoNotes and FewNERD synthestic datasets demonstrate that FSE achieves state-of-the-art performance in CLNER scenarios, with effectiveness of each component, empirical evidence of faster convergence and expected functionality of both experts.

Lightweight projectors are an established way to connect pre-trained speech encoders with large language models (LLMs), mapping acoustic features into token-level embeddings for tasks like ASR and spoken question answering. Existing systems, however, typically only support a few languages and are often limited to English. We introduce MEUSLI, the first open-science multilingual projector family that links a Whisper encoder with open-source multilingual LLMs, enabling fully open-source end-to-end ASR in 28 European languages. MEUSLI extends prior monolingual pipelines, delivering strong results across high- and low-resource languages. Using proper continual leaning techniques, MEUSLI can be easily extended to other languages not seen in training. We further demonstrate that the MEUSLI projector can be leveraged beyond ASR, enabling multilingual speech translation and topic identification with only a few hours of task specific supervision per language. Overall, MEUSLI provides a solid foundation for multilingual speech understanding tasks, supporting scalable and inclu- sive open-source SpeechLLM

Empathetic Response Generation (ERG) requires models to recognize users' emotions and generate empathetic responses. Commonsense knowledge has been shown to support such reasoning, yet existing approaches typically reuse fixed commonsense representations across understanding and generation, limiting their ability to coordinate such knowledge across different stages. We propose DCC, a Dynamic Commonsense Coordination Framework with three complementary modules: residual-based commonsense interaction (SCE-AttnRes) to integrate contextual and situational commonsense representations, Association-Guided Commonsense Filtering (AGCF) to down-weight low-relevance commonsense relations, and Iterative Commonsense-Aware Decoding (ICAD) to dynamically retrieve commonsense memories during generation. Experiments on the Empathetic-Dialogues benchmark show that DCC improves emotion classification accuracy and response diversity over the CEM baseline while maintaining comparable perplexity. An LLM-based blind evaluation further demonstrates that DCC generates responses with better relevance, coherence, and informativeness. The code and implementation details will be publicly available at https://github.com/Hanabi-Q/DCC-ERG.

Large language model (LLM) evaluation spans diverse tasks and benchmarks, yet evidence remains organized around tasks rather than the capabilities they probe. This fragmentation limits cross-study comparison, obscures capabilities tasks recruit, and makes coverage gaps difficult to identify. We introduce a multi-layer taxonomy of 14 capability domains and 91 subskills across Primitive, Constructed, and Integrative layers. Human cognitive science guides capability definition and organization, not LLM architecture. Layer assignments draw on developmental precedence and hypothesized functional support, while human-origin constructs are adapted to observable model behavior. To demonstrate operational utility, we screened 31,505 papers from ACL, AAAI, ICML, and NeurIPS between 2023 and 2025 and mapped 15,934 LLM-focused papers through multi-model annotation, consensus, and arbitration. Direct research attention concentrated on Language-Semantic Competence (3,551; 22.3%), Reasoning (3,388; 21.3%), Planning and Decision-Making (2,149; 13.5%), and Perception (1,954; 12.3%), whereas six domains appeared in fewer than 2% of papers. Within domains, the most frequent subskill had a median prevalence of 97.9% and appeared in at least 90% of papers in 10 of 14 domains. Language-Semantic Competence and Reasoning formed the highest-volume pair (n = 1,864; 11.7%; lift = 2.47), whereas Theory of Mind and Social Reasoning and Interaction showed the highest lift among pairs with at least 20 co-occurrences (n = 62; lift = 30.84). By shifting the unit of analysis from isolated tasks to structured capabilities, the taxonomy supports research organization, coverage audits, evaluation interpretation, and testable hypotheses for diagnosis, training, and transfer.

Despite the growing use of large language models (LLMs) as creativity evaluators, evidence of their alignment with human evaluations remains mixed, raising the question of when and why their judgments converge with or diverge from human judgments. Across three studies and six widely used LLMs, we addressed this gap by identifying the standards underlying LLM creativity evaluation and examining their downstream implications. Study 1 showed that LLMs generally relied on a narrower subset of human creativity evaluation standards. Convergence with human standards was strongest in the novelty dimension, whereas divergence was clearest in the contextual dimension, which captures social, market, and reputational information. Moreover, each LLM exhibited distinct, model-specific standards that varied substantially in breadth. These differences in evaluation standards were reflected in actual creativity judgments. Study 2 (N = 1,103 ideas) showed that LLM evaluations were moderately correlated with human evaluations, and individual LLMs with broader standards better distinguished ideas humans judged as more versus less creative. Study 3 (N = 1,195) showed that LLMs were less sensitive to contextual information: such information significantly altered human creativity ratings but left LLM ratings largely unchanged. Together, our findings help explain the mixed evidence on LLM-human alignment, showing that alignment depends on the evidence a judgment demands and the standards each model applies. LLMs may resemble humans when evaluations emphasize intrinsic qualities such as novelty, yet diverge when judgments require contextual information. Selecting an LLM evaluator is therefore a consequential decision: different models, applying different standards, recognize different ideas as creative.

We present a systematic study of healthcare-domain cross-lingual transfer to address the scarcity of biomedical NMT resources for Arabic-script languages. We use Arabic and Persian as higher-resource pivots to improve translation for \textbf{four severely low-resource} targets: Dari (Afghan Persian, a standardised variety of Persian), Pashto, Sorani Kurdish (Central Kurdish, a major standardized variety of Kurdish), and Urdu (closely related to Hindi). Using LoRA fine-tuning on small decoder-only LLMs, we train \textit{domain-specific pivot adapters} and evaluate \textbf{three transfer strategies}: few-shot in-context learning, minimal supervised adaptation, and, to the best of our knowledge, for the first time in this setting, zero-data LoRA adapter merging. Supervised adaptation with just 500 sentences achieves near pivot-language quality for Dari (CHrF++ 41.01) and meaningful gains for Urdu (28.88), while adapter merging reaches within 3.5 CHrF++ of supervised adaptation for Dari at zero additional cost. Pashto and Sorani Kurdish remain insufficient for high-stakes clinical deployment exposing the limits of cross-lingual transfer when structural distance from the pivots is too great. LoRA adapter merging works surprisingly well for closely related languages, even without target-language biomedical data.

Most endangered languages lack the parallel data required for machine translation, despite the existence of descriptive grammar books. We introduce a pipeline that uses large language models to extract grammatical rules, example sentences, and lexicons from grammar books and generate synthetic parallel corpora for fine-tuning-rather than feeding grammar content into prompts at inference time, as in prior work. Validated on three typologically diverse low-resource languages-Kalamang (Papuan), Tuatschin (Romance), and Mandan (Siouan)-we show that fine-tuning on synthetic data improves over seed-data baselines in 75% of configurations for Kalamang and 59% for Tuatschin, with best-case ChrF++ gains of +8.8, +5.3, and +3.3 respectively. Through a systematic factorial study across 96 configurations varying target part-of-speech, retrieval granularity, and sample volume, we identify which factor combinations drive gains and where they break down. Our results demonstrate that static linguistic documentation can be repurposed for machine translation fine-tuning, offering a practical path towards translation tools for severely under-resourced languages.

Existing lexical distance, similarity, and evaluation metrics operate on Unicode code points, which can misrepresent errors in writing systems where a single grapheme is represented by multiple Unicode code points. We introduce grapheme-kit, an open-source Python library that extends these metrics to operate on grapheme clusters instead. The library also provides improved grapheme processing for Tamil and Sinhala, including accurate grapheme cluster identification and grapheme composition/decomposition utilities. Through an OCR case study, we demonstrate that grapheme-level metrics provide a more faithful evaluation of complex scripts.

LLM training is shifting from manual design and annotation to interaction-driven self-evolution. However, existing self-evolutionary methods face a fundamental dilemma between task diversity and verification reliability: environment-bound methods obtain precise feedback but confine learning to narrow domains, while open-ended self-generation broadens the task space but lacks reliable verification, allowing misleading rewards to pollute the training loop. We identify agent skills as a powerful middle ground to reconcile this tension: each skill ensures deep, verifiable execution in a specific scenario, while dynamic routing across skills maintains open-ended task variety. Leveraging this insight, we introduce Skill Self-Play (Skill-SP), a co-evolutionary framework comprising a proposer, a solver, and a dynamic skill controller. Orchestrated via a reinforcement learning loop, these components co-evolve in a continuous self-play loop: the proposer generates challenging tasks conditioned on dynamically sampled skills; the solver explores candidate solutions to push its capability boundaries; and the skill controller collects execution feedback to update and expand the skill library. This interactive co-evolution effectively bridges the gap between structured verification and open-ended exploration. Empirical evaluations on tool-use and reasoning benchmarks demonstrate that Skill-SP, serving as a robust evolution engine, consistently pushes the performance ceiling of competent backbones while catalyzing striking turnarounds for initially misaligned models. Our code is available at https://github.com/Qwen-Applications/skill-self-play.