Physical AI Brief

How many assignments can a language model recall before it loses track of which value belongs to which entity? We measure this limit using continuous recall curves for 12 models at or below 3B parameters and a threshold sweep over 30 open models up to 12B. On the continuous curves, the load at which recall falls halfway to chance follows $K_{50}=cN^α$, with $α=0.820$ and $R^2=0.73$. The broader sweep shows an eightfold range associated with pretraining recipe, although the continuous curves show no detectable recipe effect after controlling for scale, with few modern models in the fit. We derive why interference can lower measured capacity by reducing single-binding recall even when the load-dependent recall profile is unchanged. Direct task training also exceeds the extrapolated zero-shot law, but different measurement criteria prevent interpreting that comparison as a capacity gain. Its formation times follow a power-law form in two independent codebases, conditional on runs that succeed. Together, these results characterize capacity at the model's query interface. Bounds on joint recall and a decomposition of policy errors connect this measurement to working memory and instruction following, without treating recall as a measure of alignment. The controlled task also provides a baseline for testing whether binding limits constrain world-state tracking; the present experiments do not measure state updates or downstream transfer.

Online audio-visual target-speaker extraction aims to remove competing voices while preserving speech quality and bounding lookahead. Existing extractors are built and evaluated for separation on synthetic mixtures, leaving listening quality and meeting behavior largely untested. We introduce Audio-Visual Personalized Voice Quality Enhancement (AV-PVQE), which approaches these requirements from the other direction. We start from a personalized speech enhancement model that reconstructs a requested voice at high quality but confuses the target in 46% of two-speaker mixtures despite clean enrollment. Adding mouth features at its speaker-conditioning input and jointly fine-tuning the visual and reconstruction networks reduces this rate to 1.6%, with no future frames and 20 ms of algorithmic delay. Compared with an online autoregressive audio-visual extractor, AV-PVQE yields separation gains on two synthetic benchmarks and larger gains on recorded meetings, and keeps its advantage on excerpts with more speakers than the fine-tuning mixtures. In personalized P.835 listening tests on two meeting corpora, it improves overall quality over this extractor by 0.57 and 0.63 MOS, with similar mean rating relative to the starting model. Preservation and rejection tests show that it keeps the target intact when no competing voice is present and suppresses competing speech when the target is absent.

Mission-critical UAVs increasingly rely on split vision-language model (VLM) perception under tight onboard-resource and wireless-communication constraints. However, corruption of transmitted intermediate features creates a deployment mismatch for clean-trained split interfaces, while stronger channel-aware codecs can impose substantial onboard cost. We present FreshLatent, a lightweight channel-aware latent adapter that trains a power-normalized encoder-decoder through wireless corruption while keeping the surrounding VLM frozen. We formulate deployment around a mission-conditioned perception requirement and embedded interface cost, linking channel quality and communication budget to the operating conditions under which perception remains usable. At 0 dB and the tightest communication budget, FreshLatent improves gIoU and cIoU over clean split compression by 20.79 and 20.87 points, respectively. At the most adverse evaluated SNR (0 dB), across all three communication budgets, FreshLatent recovers 63.5-69.1% of the gIoU improvement achieved by a much heavier, range-trained feature-JSCC codec. On an NVIDIA Jetson AGX Xavier in 10-W mode, FreshLatent uses 37-40x fewer encoder parameters, 7.7-9.9x lower edge-interface latency, and 8.8-10.0x lower edge-interface energy than the heavier codec. Together, these results show that lightweight channel-aware adaptation can recover a substantial fraction of the robustness of a much larger communication interface while broadening quality-valid operation under constrained wireless conditions.

Dataspace connectors decide whether a transfer may occur, not what the transferred value contains, tolerable for contracted applications, not for LLM agents that compose tool calls and spawn sub-agents. Research on agents that generate governance artifacts evaluates output quality; who may authorize an artifact for use falls between that literature and the governance literature, and neither owns it. A published policy is what a dataspace's decision point enforces, so publication is a governance event, and an agent that is both policy subject and policy author writes the norms that bind it. We name this the authorship hazard and state one principle: an agent is a subject of the governance plane, never an author of it. Its authorization channel to publication is closed by construction; its influence channel, drafting what humans approve, is treated as an enforcement problem. On a frozen corpus of agent drafts, publishing without approval reverses 80 authorization decisions, most through drafts that change only a field's sensitivity classification and no policy text; a classifier that reads the policy diff misses every such draft, necessarily. Treating classification as authorship routes them all to review; the registry-held classification this requires is designed and modelled here, not yet implemented in the prototype. At the execution boundary, protected fields reach the model in 105 of 105 cases under prompt-stated duties and in 0 of 105 when the ODRL duty is compiled into an invocation-time tool-call constraint, but where the value is not confined to a named field the compiled condition exposes it in 7 of 7. A centrally provisioned approval pool does not scale to the participant volume that motivates the problem.

Dermoscopy classifiers built on Vision Transformers process all image patches uniformly, although diagnostic evidence is concentrated in the lesion region. Existing token pruning methods reduce tokens using generic saliency or similarity signals, but rarely ask whether the retained subset still contains the lesion. This paper introduces MedTokenBudget, a supervised post-backbone token routing framework that learns to construct compact lesion-enriched representations when auxiliary lesion masks are available. Its Lesion-Aware Token Scoring (LATS) module fuses attention entropy, feature norm, and local feature contrast through a learned scorer, then routes the top-$K$ patches under a target budget. LATS is trained with budget curriculum learning, diversity regularization, attention distillation, and lesion-mask supervision. The trained router is evaluated with a lesion retention rate that directly measures how much ground-truth lesion evidence survives the token budget. On ISIC 2019, mask-supervised LATS consistently outperforms Random and ToMe at headline budgets while retaining substantially more lesion patches. Code is provided for reproducibility, and complete tabulated results are included in the supplementary material.

On Jan. 30, 2026, the U.S. Department of Justice released a mixed-media collection concerning Jeffrey Epstein, including about three million pages of PDFs. We describe the Epstein Files Engine, an A.I. agent The New York Times deployed to investigate the files. The Engine translated reporter questions into Google BigQuery SQL queries across three corpora: Epstein-related releases, the Times's archive and external, Epstein-related news headlines. It used an LLM to plan queries and returned citation-rich answers a reporter could verify and trust. More than 100 journalists used the Engine, and it contributed to at least 20 published stories. We report how reporters queried it and describe Diff, our text-and-visual duplicate matching method that amplified novelty signals and allowed the Engine to surface genuinely new information. We argue that newsroom agents serve newsrooms best not as autonomous writers, but as interfaces to source material and institutional knowledge.

Multi-shot agentic video generation requires consistent character appearance, stable spatial layout across camera angles, and continuous character state between shots. When every shot is a separate request to a frozen generator, repeated text does not determine appearance, layout or state. We therefore recast the problem as condition construction and present MVAgent, a multi-agent pipeline whose agents collaborate through typed conditioning inputs. Because an environment image shows one viewpoint, a Spatial Grounding agent samples views from generated camera-traversal clips and anchors each shot to the view matching its framing. As generated shots drift from the plan, an Observer records how each shot ends in a continuity memory, from which a Transition agent builds character action and spatial references for the next shot. An Orchestrator composes these inputs into each request. Since a request reveals its effect only after rendering, we train it by agentic reinforcement learning with Trunk-GDPO, which compares rendered candidates at every shot rather than once per video and continues the best as the trunk. With generator and judges frozen, MVAgent attains the highest cross-shot consistency and narrative-planning quality among the compared methods on ViMax-Bench and is preferred over the strongest agentic baseline in human evaluation.

A robot that probes a few times before one irreversible action, such as tapping a surface before inserting a peg, must decide when the evidence is enough to commit. We argue that this decision rests on two conditions that existing methods do not separate: the belief must still cover the truth in the coordinate that decides the action, and the failure model that scores actions must track realized failure. We audit both conditions separately, offline and with ground truth, on a deployed probe-then-commit pipeline: a particle belief, a scenario failure score, and one commit. On simulated insertion, more taps sharpen the belief while the truth leaves its support on 16.9% of episodes and the failure score turns optimistic by 0.31. Conformal calibration restores coverage but not the decision: confidently wrong instances still pass a confidence gate. The audit's signatures instead point at the observation model, where a hand scan finds a 2.1 mm error in the tap boundary. Correcting that one number cuts failure from 0.354 to 0.112 on untouched instances and transfers unrefitted to a second engine, while in a third engine the same audit suggests an execution-model mismatch instead. Across seven task families in three engines, a few probes at a fixed executor reduce miss or failure. On a physical arm inserting a tool into a rigid pocket by touch, the gain and the audit's two conditions reproduce, and replaying the recorded taps under an injected model error shows the audit's signature on real data. Additional materials are available at https://sites.google.com/view/auditbeforeyoucommit.

Existing computer-use agent benchmarks do not fully evaluate agents acting as assistants. A useful assistant retrieves information across complex, multi-step workflows, synthesizes it into artifacts (documents, presentations, spreadsheets), and navigates program interfaces to produce a coherent final product. Such workflows demand reasoning and synthesis, decomposition of complex tasks, as well as visual and spatial understanding. To study agents on workflows like these, we introduce KNOWS, a benchmark of open-ended, complex, browser-based tasks that jointly evaluate these capabilities, with each task culminating in a produced artifact. To write tasks, we develop a task design rubric and a protocol for ensuring that tasks meet the requirements. Each task is paired with an evaluator, a program that combines deterministic checks with LLM judgments to balance the richness, reliability, and automation tradeoff inherent to agent evaluation. We evaluate and analyze frontier computer-use agents and browser-based harnesses. They achieve moderate scores on partial-success metrics, but the best performer fully succeeds in fewer than 3% of our complex, long-horizon tasks. Failures on visual steps render the resulting artifacts unusable, even when agents complete more than 50% of other evaluation steps. Our results expose limitations of current agents acting as end-to-end assistants, and call for progress on tool use, visual understanding, and long-horizon reasoning.

Synchronised action annotations are needed to train controllable world models and these datasets remain elusive. Existing approaches make use of instrumented platforms with calibrated sensors, costly manual annotation, or latent-action models which lack grounding. We instead turn ordinary unlabelled video into action-supervised training data by recovering (without training) a data-derived egomotion basis. We track pixel displacements across frames and exploit the recurring coherent structure induced by egomotion to obtain grounded control signals directly. Using a method as simple as principal components analysis perform this, we find that the leading components provide signed, scalable, and composable throttle--yaw controls, although the method can recover only motion axes represented in the data. To prevent a high-capacity video DiT from exploiting pixel-level supervision, an online latent critic distils a frozen decoder--tracker--PCA (Principal Components Analysis) teacher without backpropagating through the decoder or tracker. Finally we critique the use of video generation metrics to evaluate WMs and introduce an example of an alternative, reference-free evaluation method. We measure \textit{controllability}, \textit{plausibility}, \textit{conjuring} (creating objects out of thin air) and \textit{geometric integrity}, revealing failures that conventional video metrics miss. We show that most baselines follow familiar action directions but struggle to reverse or remain stationary. Our model handles both while retaining compositional control and generation quality. Despite backwards actions being less than $1\%$ of our training data, we find that the model learns to reverse, scale its response linearly, and compose throttle with steering, all simply by learning through a grounded action space.

For humanoids to be useful in everyday environments, they must perform a wide range of tasks that couple locomotion and manipulation. Existing approaches commonly acquire a loco-manipulation policy through reward engineering or demonstrations followed by task-specific training, making it costly to scale to new tasks. In this work, we propose a hierarchical approach to humanoid loco-manipulation that eliminates these per-task requirements. HuGo, Humanoid policy code Generation, uses a Large Language Model (LLM) to generate executable, closed-loop high-level policy code from a task description on top of a frozen low-level whole-body policy. Given the task, observation, and command specifications, the LLM constructs the task logic in code. HuGo then refines the policy from its rollouts using numerical trajectories and selected video frames to produce feedback and targeted code updates. Across five simulation tasks, using two different low-level policies, HuGo substantially outperforms a high-level reinforcement learning baseline and approaches the performance of a demonstration-based baseline. We achieve this level of performance without task-specific reward design or demonstration collection. We further demonstrate zero-shot transfer of simulation-generated policies to hardware and show that applying the same refinement loop to real-world rollouts can further improve transfer performance without expert demonstrations or policy retraining. Project website is https://iconlab.negarmehr.com/HuGo/

Language models gain accuracy from more compute at test time, but majority voting over independent samples saturates: as samples grow, the vote converges to the model's most frequent answer. Communication can add what sampling cannot: an agent that solves a problem can pass the key step to the others. We present TalkMesh, a decentralized mesh of small language model agents that learns when and what to communicate. Each agent samples a proposal and scores it with a trained confidence head. The most confident agent broadcasts a hint; agents below a confidence threshold revise, keeping each revision that outscores its proposal. Gossip consensus approximates the vote weighted by confidence without a coordinator. A talk policy, trained with group relative policy optimization on the change in correctness after revision, writes hints and revisions. With three agents, which together generate at most six outputs, the mesh reaches the accuracy of majority voting over 32 samples with each of three models. Trained with at most 8 agents and evaluated with 32, it raises accuracy from 0.568 under self-consistency to 0.705 (Qwen3.5-0.8B, GSM8K) and from 0.492 to 0.722 (SmolLM3-3B, MATH-500). When 4 of 8 agents collude on a wrong answer with fabricated confidence and poisoned hints, majority vote accuracy falls to 0.000 (Qwen3.5-0.8B, GSM8K). A defended mesh, whose agents rescore solutions with their own confidence heads, retains 0.507. Across reasoning, embodied coordination, and traffic signal control, messages improve a decision when the acting agent cannot observe the information it requires and another agent can send it.

Large-scale recommenders increasingly adopt the sequential generative recipe behind large language models, bringing the Transformer into recommendation along with design choices made for text, including Rotary Position Embedding (RoPE). In language models, RoPE encodes token indices for relative position reasoning, but in recommendation, an interaction index records only event order, saying nothing about elapsed time, behavioral cycles across scales, or calendar phase. We revisit this choice and propose T-RoPE, a time-aware RoPE for sequential generative recommendation that replaces index-only rotation with timestamp-based angles, learnable temporal coefficients, multiscale frequency banks, shifted query alignment, and non-stationary key rotation. We prove that standard RoPE, even on timestamps, remains time-translation invariant and cannot distinguish seasonal contexts, and that T-RoPE breaks this invariance while preserving the RoPE interface. Across five public benchmarks, T-RoPE achieves the best result on every metric on every dataset, improving over the strongest baseline by 78--130\% in HR@10 on the sparse PixelRec data and 8--12\% across metrics on Amazon Books. On an industrial-scale e-commerce dataset with more than 6B interactions, it improves every metric over the HSTU + Time RAB backbone by 13--82\%, with ablations attributing the largest gains to multiscale frequencies ($+56\%$ NDCG@50) and non-stationary keys ($+4\%$). An online A/B test in the Shop app yields positive lifts in conversion rate ($+0.33\%$) and order count ($+0.63\%$). We also provide forward and backward algorithms whose added cost is linear in sequence length and head dimension, keeping time-aware RoPE practical for large generative recommenders.

Large language models are often post-trained on expert demonstrations using cross-entropy (CE), even when the downstream objective is not to imitate the demonstrated solution but to produce any output accepted by a verifier. This mismatch is seen in verifiable domains with multiple correct solutions, such as mathematical reasoning and code generation, where training data may contain only one expert solution per problem. We show that minimizing cross-entropy can be misaligned with minimizing verifier risk; two policies can assign identical likelihood to the observed demonstrations while placing different probability mass on incorrect outputs. This is formalized through a learning-theoretic counterexample in which CE minimization selects a suboptimal policy. We identify that controlling the support of the learned policy can solve this problem by preventing probability mass from spreading to unsupported outputs. Since support size is non-differentiable and computationally intractable, we propose entropy-regularized cross-entropy (ER-CE), using token-level Shannon entropy as a tractable proxy. Finally, across mathematical reasoning and code-generation benchmarks, we find that entropy-regularized training consistently improves verifier accuracy over standard cross-entropy. Our results identify a simple failure mode of imitation-based post-training in verifiable tasks and provide a practical objective that is better aligned with producing correct outputs.

Language models are often evaluated on curated benchmarks that underrepresent the complexity of enterprise deployments. We introduce HARDEN, a constrained evolutionary search method to adapt the input of existing evaluation cases into more challenging variants while keeping their expected outputs fixed. HARDEN searches along generated domain-specific complexity axes while enforcing feasibility constraints such as preserving task semantics, realism, and execution validity. Across FinQA, PubMedQA, and ContractNLI and three Qwen3.5 model scales (35B-A3B, 122B-A10B, and 397B-A17B), HARDEN reduces task-model accuracy by 22.7% on average and by up to 49.9% relative to single-pass baselines using the same feasibility checks. These results show that evolutionary search can produce substantially harder valid evaluation cases.

Platform policies are increasingly tested on artificial users, making agent fidelity important. Yet convincing fake profiles could also manipulate perceived public opinion before elections. Validation has concentrated on agreement with human behaviour and has paid little attention to whether an agent behaves in line with the profile it was given. The present study profiled eight Serbian participants through a questionnaire, a deep interview, and a written self-presentation, recorded their reactions to sixty-eight social media posts, and asked four language models to predict those reactions under five prompt conditions varying profile content and instruction style. Attitudinal content improved prediction over demographic backstories by a wide margin. Agents matched their stated profiles more closely than participants matched their own survey answers, and consistency proved unrelated to fidelity once profile information was present. Instructing models to respond intuitively and immediately rather than analytically gave the highest fidelity of any condition and cut the compression of individual differences from seven times the human level to three. The advantage held on posts about topics the questionnaire never raised, where that condition reached the highest fidelity of any setup and beat a crowd baseline by a wide margin, which suggests that agents prompted this way could serve as general-purpose simulated users rather than specialists on the topics they were profiled for. Results may bear implications for the development of language models, because intuition-based setups appear better suited to some tasks than reasoning-based ones.

Agent memory systems are increasingly used to maintain long-term user preferences, task states and evolving facts, but current evaluations often collapse memory behavior into final-answer accuracy. We introduce MemProbe, a cognitive-science-inspired framework for diagnosing stability-plasticity tradeoffs in agent memory. The framework is motivated by a core insight from cognitive memory research: memory is reconstructive and shaped by interference, source reliability, reinforcement, and reactivation. MemProbe turns this insight into four reusable experimental paradigms (interference, misinformation, consolidation strength, and reconsolidation window) that manipulate when a memory should be updated, preserved, or treated as uncertain. It further decomposes correctness into behavioral profiles that reveal how systems update, preserve, attribute, and temporally organize information. We instantiate these paradigms in a 56-episode diagnostic suite and evaluate six incremental memory systems under a unified protocol. Results show that systems with similar aggregate scores exhibit distinct behavioral profiles. MemProbe provides such a diagnostic lens, turning aggregate performance into interpretable profiles of memory maintenance over time. Code is available at https://github.com/jq-ding/MemProbe.

Runtime failure monitors can use a model's internal representations to anticipate failures. We audit this monitoring strategy across two autonomous-driving tasks: online vectorized map generation with LaneSegNet and end-to-end planning with VAD. We find that frame-level errors are predictable at inference in both tasks. For LaneSegNet, a supervised latent probe reaches Area Under the Receiver Operating Characteristic curve (AUROC) 0.780 for high Chamfer error; to our knowledge, this is the first post-hoc frame-level failure monitor for online vectorized map generation. For VAD, a supervised planning-latent probe reaches AUROC 0.868 for mean-ADE failure. Our audit shows that internal access is not necessary for strong failure prediction. A monitor using only LaneSegNet's prediction outputs reaches AUROC 0.825, while for VAD, ego state, driving command, and the planner's predicted trajectory reach 0.924 on the same mean-ADE endpoint. Adding latent features to either baseline yields no statistically resolved improvement. This observation persists across a broad suite of planning failure endpoints, including endpoints whose labels depend on geometry unavailable to the non-latent baseline. Thus, predicting failure from an internal representation does not establish that the representation provides useful information beyond observable inputs and outputs. We propose an evaluation protocol for testing the incremental value of latent access and release our per-frame failure endpoint labels.

Large language models (LLMs) have recently demonstrated promising capabilities as in-context policy optimizers for Reinforcement Learning (RL), enabling policy search driven by both numerical reward signals and natural language reasoning. However, deploying such methods in practice requires transmitting raw policy parameters and rewards history to cloud-based LLM APIs, exposing proprietary control strategies to third-party service providers. To address this issue, this paper introduces Privacy-Preserving Prompted Policy Search (PP-ProPS), a framework that enables LLM-guided policy optimization while keeping policy and environmental parameters confidential. PP-ProPS encodes policy parameters and reward values using secret client-side transformations before they are included in each API request, ensuring that the LLM provider observes only encoded policy parameters and scaled reward information. Furthermore, unlike Vanilla ProPS, the proposed framework does not require the true optimal episodic return to be known or disclosed to the LLM. Beyond protecting the optimization data, PP-ProPS improves the search process in two ways. First, it provides the LLM with individual reward components instead of only a single total return, offering more informative feedback about each candidate policy. Second, it uses a bounded history that prevents the prompt from growing indefinitely, improving search with high-dimensional policies and supporting the use of open-weight LLMs. The proposed PP-ProPS is evaluated on both continuous and discrete control problems spanning Multi-Joint dynamics with Contact (MuJoCo) locomotion, classic control, highway driving, and robotic arm manipulation. Compared to Vanilla ProPS, the proposed PP-ProPS outperforms ProPS in seven of the ten evaluated tasks, and surpasses conventional RL methods including PPO, SAC, and TRPO, in five of the six tasks.

Benchy is a semantic language and execution engine for benchmarking AI programs. A benchmark is completely specified by a program, a scoring function, and a dataset, B=(P,S,D), and is separate from the AI-system taking it; a run binds the two, R=(B,AI). Benchmarks are authored as canonical YAML in which each semantic concept has one valid syntax, classified by a shared task/domain/language ontology, and deterministically compiled into a canonical JSON intermediate representation that the engine executes. Compilation changes representation, not meaning: it does not repair invalid definitions or inject hidden defaults. Programs use fixed schemas of named input and output fields, the leaf output fields are the scoring dimensions, and the engine exposes one universal runtime contract --- a named-field input object in, a named-field output object out --- to which external AI-systems adapt at the boundary, so integration mechanics never propagate into benchmark semantics. This paper gives the semantic object model, the ontology and task-to-program validation rule, the scoring and failure semantics, the compilation and execution architecture, and the scope of the current language. An appendix fixes the normative engineering contract for the first engine implementation.

As robots transition from structured factory settings into homes, they are required to interact with an ever-increasing variety of objects. Many tasks require grasping, and often it is not sufficient to just pick up the target object. Consider a task like "pouring coffee" --- to facilitate the subsequent pouring, the robot should grasp the mug by its handle. Existing learning-based approaches for grasping either find robust and collision-free grasps that are largely agnostic to the task (e.g., picking up the mug by its rim), or leverage foundation models to propose task-appropriate grasp locations that lack fine-grained physical grounding (e.g., reaching for and missing the handle). In this work, we bridge these approaches with a real-to-sim-to-real framework. Based on a single RGB-D observation, we construct a digital twin of the environment, query a large foundation model to propose grasps that align with the object's affordances and task description, and then optimize the proposals to ensure robustness and plausibility before executing the result on the real robot. Our key insight is that the grasp proposals of the foundation model should be regarded as semantic priors that serve as seeds for local, gradient-free optimization. We leverage Bayesian optimization with Thompson sampling to draw batches of nearby poses, which are subsequently evaluated in parallel under domain-randomized physics rollouts. The resulting grasp is both task-oriented and physically feasible for execution by the robot arm. Our full zero-shot real-world transfer only takes a few minutes and improves task-oriented grasping success by up to 33% as compared to other state-of-the-art pipelines. Our code is available here: https://github.com/VT-Collab/GraspTwin/

Optimizing embedding systems for production recommendation pipelines demands systematic exploration that consumes disproportionate engineering effort at scale. We apply Andrej Karpathy's AutoResearch paradigm -- a large language model that iteratively edits a training script and retains modifications that improve a held-out scalar metric -- to automate this exploration. We report on twelve weeks of running this paradigm at production scale, where iterations consume hours of multi-GPU compute, evaluation involves competing criteria, and campaigns span weeks across many training jobs. Across two independently developed representation-learning systems for a book recommendation pipeline, we ran 220+ experiments and observed five recurring failure modes absent from the original setting: infrastructure fragility, agent memory decay, search-direction stagnation, iteration-cost asymmetry, and metric fixation. We contribute a three-principle scaffolding design -- prevent, persist, redirect -- that maps each failure mode to a structural remedy and whose instantiation scales with iteration cost. The framework produced a 1.82x Recall@6 lift and a 2.1x coherence lift over hand-tuned baselines, and the agent autonomously designed a text-only fallback that expanded catalog coverage by 5.8x. The two systems span nearly three orders of magnitude in per-iteration cost yet exhibit the same failure modes, suggesting these are structural properties of production-scale autonomous research rather than artifacts of either application.

SRAM-based FPGA system-on-chip devices are vulnerable to single-event upsets (SEUs) in configuration memory, which may perturb programmable routing resources and degrade communication fabrics. In modern Zynq UltraScale+ MPSoCs, such routing disturbances can introduce small propagation delay shifts that remain logically transparent yet compromise AXI-based data transfers and lead to silent data corruption. Although routing delay degradation and AXI interconnect failures have been studied independently, their experimental correlation under neutron irradiation has not been established. This work presents a cross-layer investigation on a ZCU104 platform integrating routing-dominated delay sensors with an AXI interconnect benchmark comprising replicated accelerators. Neutron irradiation experiments were conducted on the fully operational system, while a frame-level configuration fault injector implemented via the internal configuration access port enables controlled upset emulation. Measured routing delay events are statistically correlated with communication failures, and cross-sections for both timing shifts and AXI malfunctions are derived. The results experimentally demonstrate how neutron-induced routing perturbations propagate into system-level silent data corruption in UltraScale+ MPSoCs, providing insight for resilience-oriented AXI-based design in neutron-rich environments.

Children in multilingual communities often code-switch, using multiple languages in a single utterance. Can we induce cross-lingual alignment in language models by training on code-switched text? We pretrain small decoder-only transformers on two 100M-word multilingual corpora: a base corpus formed by mixing the English, Dutch, and Chinese BabyBabelLM datasets, and a corpus generated from it by inserting word- and sentence-level code-switching using an LLM. We find that training on code-switched data aligns the representations of parallel text, particularly across different scripts, and that this alignment persists through training on monolingual documents. Under a learning curriculum that progresses from word-level code-switching, to sentence-level code-switching, to monolingual documents, models trained on code-switched data outperform baselines trained without it on the BabyLM evaluation suite. Our work characterizes code-switching curriculum learning as an effective data augmentation method for multilingual pretraining. We release our code, data, and models at https://github.com/drooryck/multilingual-macaroni.

We propose ST-pRRTC, a GPU-parallel space- time RRT-Connect motion planner for problems with known obstacle trajectories and unspecified arrival time. Searching over many arrival times broadens temporal coverage but divides a finite planning budget among more backward trees. To address the challenge, ST-pRRTC builds a shared forward tree and an adaptive forest of backward goal-time trees. Its interval root formulation samples goal arrival times continuously and guarantees probabilistic completeness and asymptotic arrival- time optimality under the stated assumptions in a bounded time domain. The practical root recycling policy has no such guar- antees. It adapts a fixed number of backward trees, replacing later roots while retaining useful search progress. Experiments on three dynamic benchmarks show that both variants achieve lower mean first-solution times and earlier mean final arrivals than ST-RRT* and SI-RRT on problems solved by all compared methods. Further experiments demonstrate the benefit of recy- cling over broad arrival-time ranges. Real-robot demonstrations show root-recycling ST-pRRTC planning motions for a UR5e among moving Crazyflie quadrotors.

Large language model (LLM)-powered multi-robot systems are vulnerable to semantic manipulation: an accepted false world-state claim can trigger a fleet-wide behavioral cascade, causing unnecessary replanning, increased path costs, congestion, or apparent mission infeasibility. Conditioning on a successful manipulation, we propose an active verification framework that contains its downstream effects before they propagate across the fleet. A dedicated verification module generates a structured Verify-Adapt-Hold plan: selected robots inspect consequential regions, a limited subset provisionally adapts when necessary, and the remaining robots retain their trusted plans. We evaluate the framework in a multi-robot transportation environment using injected false obstacle claims across different impacts and team sizes. Evaluation measures cascade containment, Sum-of-Costs, makespan, and coverage ratio. Results show that treating post-compromise verification as a team-level planning problem, rather than a binary trust decision, effectively limits the cascading physical consequences of semantic manipulation.

Aerial manipulation in outdoor environments remains challenging due to the simultaneous requirements of reliable state estimation, stable aerial motion, and precise manipulation under external disturbances. In this work, we present a real-world outdoor aerial manipulation framework that integrates imitation learning, onboard LiDAR-inertial state estimation, and whole-body model predictive control. A Diffusion Policy is trained from manipulation demonstrations to generate desired end-effector motions from onboard observations. These learned commands are executed by a whole-body MPC that jointly coordinates the aerial platform and manipulator to realize the desired end-effector trajectory. To eliminate reliance on external motion-capture infrastructure, the platform employs onboard LiDAR-inertial odometry for state estimation during outdoor operation. We validate the complete framework on a physical aerial manipulator and demonstrate successful execution of outdoor manipulation tasks. The experimental results show that demonstration-driven manipulation policies can be effectively integrated with onboard state estimation and model-based whole-body control to enable aerial manipulation beyond controlled indoor environments.

Recent progress in speech-driven 3D facial animation has improved vertex-level reconstruction quality, but speech-consistent visible articulation remains difficult. This is because speech production follows structured and constrained articulators' coordination and the mapping from acoustics to motion is inherently one-to-many. Motivated by the structured patterns of visible articulation, we propose a novel articulation-aware framework that models visible speech through directional articulatory motions and composes them into surface-consistent 3D facial motion. To represent visible articulation with three directional articulatory motions, spreading, opening, and protrusion, we propose a Speech--Articulatory Memory (SAM) that captures the correspondence between speech and these motions under phonetic context through retrieval and decoding based on a key-value memory structure. Then, a Topology-aware Articulatory Composition (TAC) integrates the predicted directional articulatory motions under mesh topology to produce surface-consistent 3D facial motion. Experiments on VOCASET and TFHP show that our method achieves state-of-the-art performance on standard reconstruction metrics and improves visible articulatory distance and velocity errors for lip articulation, while a user study confirms clear preference in lip sync and realism.

Voice agents are increasingly deployed in workflows where failed interactions can affect transactions, access, and other consequential outcomes, creating a need for reproducible and interpretable evaluation. We introduce Inquesto Score (IS), a protocol for measuring voice-agent reliability as the percentage of calls in a fixed, versioned evaluation population that achieve the caller's goal without a functional failure or worse. Rather than combining heterogeneous metrics, IS defines explicit failure events and severity levels and evaluates the deployed voice pipeline. Timing failures, including talk-over and delayed responses, are measured directly from audio, while semantic and state-dependent failures are evaluated using scenario predicates, tool traces, and a pinned open-model judge. Diagnostic views of behavior, acoustic robustness, identity handling, and speaker groups accompany the score without being combined into it. Inquesto Score v0.1 evaluates 30 scenarios, three acoustic conditions, four speaker groups, and 306 calls per agent across 13 configurations of a reference voice-agent system. Our evaluation shows that reliable measurement requires evidence beyond transcripts, explicit treatment of deployment conditions, and validation of the evaluators used to determine outcomes. We release the protocol, reference implementation, and evaluation records.

The aim of this work is to examine the connectomes of Caenorhabditis elegans through a computational lens using the reservoir computing framework. Connectomes are mappings of biological neural networks; C. elegans is the first organism for which physical connectomes covering the whole nervous system have been published. The connectomes of C. elegans used in this paper have been derived at different ages of the organism and are based on three different ways of measuring inter-cellular connections. They have, with minimal preprocessing, been implemented as reservoirs in the form of echo state networks, which are recurrent neural networks. In reservoir computing, the reservoir itself is not trained, rather the output of the reservoir is passed to a comparatively small read-out module in which training takes place. Training and testing is conducted in different neuro-inspired tasks, with the aim of using these tasks as a benchmark for the connectomes. This process has been repeated with different configurations of the reservoir and equally sized but randomized null models have been used for comparison. The results show that the biological wiring and a bio-informed configuration of input and output nodes of the reservoirs do not necessarily lead to better performance. Contrarily, the randomized null models are often outperforming the original connectomes on the chosen benchmarks. At the same time it becomes clear that the results depend a lot on the configuration of the reservoir and the way the connectome has been derived from the organism. Connectomes from different ages may produce varying outcome, without a clear trend becoming visible.

Humanoid hands require tactile feedback across the whole finger, not just the fingertip, to grasp and manipulate objects properly. Vision and proprioception alone cannot reliably provide this information, particularly when the hand's own fingers occlude the camera's view of the grasp. We present a low-cost tactile array for a humanoid finger, made from Velostat and conductive tape. The fingertip carries seven contact points including a 2x3 matrix wrapped across its front, left, and right faces, and a separate contact point at the tip. The proximal and middle phalanges each carry a single front-facing contact line. We measure the sensor's hysteresis and recovery time after release, through repeated loading and press-release tests. We also test a compliant, 3D-printed contact structure with a gap and a bump, inspired by similar designs in prior work, and show it cuts recovery time by 74% compared to a flush-contact baseline. We then show the sensor can produce distinct activation patterns for different contact geometries (flat, edge, corner) at the fingertip, and that it registers contact across all three phalanges during a grip. Finally, we discuss the limits of our fabrication changes and point to software-based compensation as a promising way to more directly fix the remaining hysteresis in the future.

Sequential inference estimates latent states from noisy and incomplete observations. Particle Filters (PFs), a class of Monte Carlo methods based on importance sampling, provide a flexible framework for this task, but often suffer from poor sample efficiency and unfavorable scaling with dimension, partly due to suboptimal proposal distributions. We address these challenges by integrating learned proposals into the PF framework. We introduce Neural Optimal Particle Filters (NOPFs), which learn an amortized approximation to the optimal proposal from offline simulated one-step conditioning tuples. The learned proposal is used as a drop-in replacement in standard PF updates, with samples corrected by standard importance weights so that the method asymptotically targets the same filtering distribution under standard support and density-evaluation assumptions. Across stochastic nonlinear benchmarks of varying inference complexity, NOPFs improve sample efficiency and distributional accuracy over standard PF baselines with modest computational overhead. The approach integrates data-driven proposal learning into classical inference without altering the underlying filtering objective.

Large Language Models (LLMs) have demonstrated historic breakthroughs in general reasoning with early successes in biomedical science. However, existing LLM benchmarking emphasizes factual recall, offering limited insight into model performance on frontier and multimodal tasks. We assembled BioEVAL (BioEngineering Validation of AI and LLMs), a global, multi-institutional initiative designed to assess experimental reasoning capability across bioengineering (BE) subfields. BioEVAL spans 11 major BE subfields plus a set of uncategorized items, bringing together 22 research groups to create a PhD-level benchmark comprising 608 evaluation items: 1) 380 multiple-choice questions (MCQs, 359 retained after audit), 2) 218 literature synthesis tasks, and 3) 10 multimodal problems with experimental image interpretation. Benchmark items underwent authoring-group expert review and centralized quality control before evaluation. Following evaluation, a blinded cross-group consensus audit of the highest- and lowest-accuracy MCQ items flagged 21 questions for revision or removal; these were withheld, and all reported MCQ results are computed on the 359 retained items. We evaluated diverse cloud-scale foundation/multimodal models (e.g., ChatGPT, Gemini, and Grok) and locally deployable models suitable for inference on consumer-grade GPUs. Models achieved the highest accuracy of up to 90% on MCQs, similarity score of 0.72 on literature synthesis, and accuracy of 80% on a small sample of multimodal reasoning questions, with substantial performance variation across subfields. Leaderboard rankings characterize current capabilities, limitations, and development priorities across the evaluated BE task categories. BioEVAL is maintained as an extensible benchmark with standardized protocols for continuing expert item contribution and model evaluation.

While large language models (LLMs) have achieved remarkable linguistic capabilities, a profound question lingers at their core: do these models truly comprehend context or simply excel at pattern matching on an unprecedented scale? Contextual understanding in LLMs refers to the ability to correctly extract relevant information from a given context, integrate it into a coherent internal representation, and reason over it to produce factually consistent and contextually grounded responses. However, traditional methods such as BiLingual Evaluation Understudy (BLEU) and perplexity simply measure surface-level performance. This reveals a critical gap in question answering (QA), where responses must be contextually grounded rather than simply being memorized associations. To fill this void, we propose a novel knowledge graph (KG) based evaluation framework for LLM contextual understanding in QA. Central to this is Semantic Structural Similarity for KGs (S3KG), a hybrid similarity measure combining structural and semantic signals into a single score. In addition, a diagnostic analysis framework is developed to identify and categorize reasoning errors at the triplet level, enabling fine-grained analysis of model failures. Together, across nine benchmarks, S3KG achieves F1 gains of up to $+7.6$ points over the strongest baseline and AUROC up to $0.973$.

Audio language models state numbers for acoustic quantities, and neither human opinion nor a judge model says whether such a number is true of the signal. AcoustiClaim extracts each numeric claim from free text, scores it against the instrument that defines the quantity, and classes each quantity by where its reference can be read. Four open-weight systems and one closed model, asked for ten quantities five ways on two corpora, fill 207 cells. Of these, 49 emit fewer than five distinct values, and eight of the 158 cells that can be ranked exceed a rank correlation of 0.3, the bar we set, three with an interval clear of it, five of them one closed model reading pitch. Error sits at or above a constant-predictor floor in every ranked cell but three. The reference decoder we train declines the five voice quantities in prose on 95% of mixtures, with nothing withheld, and states them on the clean twins, reproducing its targets' rule from audio alone. With a calibrated threshold, withholding lowers error on all ten quantities on the mixtures in the mean and on eight at every split, against at most 0.6% from a random selector. A linear baseline orders errors at least as well as ours. F0 s.d. and shimmer stay above the constant floor.

Soft robots have attracted much attention for their safe human-robot interaction and flexibility, but the typical continuum structure and nonlinear material behavior make the kinematics modelling complex, especially in non-static motions. In this work, we proposed an LSTM-based pressure predictive control (PPC) for the motion control of a vertebraic soft robotic tail and the coordination with a quadruped robot. The PPC consists of an inverse kinematics (IK) model, a forward kinematics (FK) model and a pressure compensation (P-comp) model, and achieves non-static and quasi-static motion control of the tail. Compared with the IK-only model, the average RMSE of the PPC's simulation trajectories reduces by 69.8%, when executing target trajectories. In the coordinated motions of the soft tail quadruped, using a prediction data set to train the PPC enables next-moment action prediction and reduces computation time by 60.9%, which enhances the real-time response of the tail to match the quadruped torso's moving rate. The PPC provides a simple and effective method to model the soft tail for both non-static and quasi-static motion control, and grants the soft tail quadruped with the functionality of interacting with the environment.

Convolutional neural networks trained on ImageNet are known to exhibit a strong preference for local high-frequency texture, an inductive bias that translates into fragile robustness against distribution shifts in real-world environments. Event cameras, in contrast, record only changes in scene brightness and are therefore well suited to capturing contour information; however, due to the absence of diagnostic benchmarks in the event domain, the inductive bias that event-camera data instills in vision models has remained underexplored. In this work, we use knowledge distillation from the event domain to the RGB domain so as to exploit the rich evaluation toolkit available in the RGB domain and systematically dissect this inductive bias. Our experiments show that distillation from the event domain induces, in the RGB domain, color invariance, shape bias, and robustness to high-frequency noise. We identify the underlying mechanism as the model suppressing its dependence on high-frequency texture while acquiring a stronger dependence on edge-based object shape. This hypothesis is supported by changes in how color and spatial information are processed at the early layers, together with a spectral trade-off in which robustness to the absence of high-frequency components coexists with vulnerability to contamination of the relied-upon frequency bands and to disruption of geometric structure. We further show that this inductive bias differs from existing robustification methods and that it functions as a useful prior for diverse downstream tasks in which shape and contour information contribute alongside other cues. The code is available at https://github.com/snskysk/event2rgb-distillation .

Reference-based LLM-as-a-judge evaluation assumes the reference answer is the target. In deployed agentic systems that operate over dynamic entities (support cases, assets, accounts), the closest available reference typically applies the correct procedure to a different entity, so a literal judge penalizes different identifiers, dates, and statuses as errors or hallucinations. We name this failure mode reference-instance divergence (RID). We propose CARGO, a framework that (i) treats retrieved references as procedural exemplars and grounds factual judgments in the live instance's observed context, (ii) assigns each claim a three-way status (supported, contradicted, unverifiable) and penalizes only contradictions, and (iii) gates evaluation by retrieval confidence, casting production evaluation as selective prediction. We introduce CARGO-Bench, a perturbation-based diagnostic suite with ground truth by construction that separates leniency from discrimination. On CARGO-Bench (246 items, two judge models, 7,872 judgments), the standard reference-based judge penalizes 100% of correct entity-transplanted answers and is uninformative (discrimination index DI ~ 0); supplying the live facts without reframing changes nothing. CARGO eliminates these false penalties (0/50) while retaining near-complete contradiction recall (50/50 and 49/50), raising DI to 0.58 [0.48, 0.68]; a rubric-swap control attributes most of the effect to context-grounded dimension definitions. CARGO also exposes a limitation of its own design: the leniency that protects entity values suppresses detection of procedural corruptions (20% recall). A post-hoc fix does not close the gap, and an LLM-as-annotator study with written guidelines and adjudication shows the same blind spot. We release a preregistered protocol for extending the evaluation to expert agreement, risk-coverage, and cost on production traffic.

Retrieval-based factuality evaluation, where LLM-generated claims are verified against evidence from authoritative medical corpora, has become the dominant paradigm for scalable hallucination detection in high-stakes clinical settings. Despite the urgency of reliable and transparent medical fact verification, most systems measure performance with aggregate metrics like F1, which obscure where and why failures occur. Existing RAG diagnostics require gold answers or annotated gold evidence, neither of which exists in this regime. We introduce two comprehensive taxonomies, grounded in a case study on the open-ended MedExpert dataset and 3 closed-ended datasets, decomposing failures into retrieval-stage errors along five quality dimensions, and verifier-reasoning errors into six consecutive steps. We adapt an automatic pattern induction pipeline using LLM-as-Judge to label evidence quality and classify verifier reasoning errors at scale, and then stress-test our findings across 4 retrieval methods and 6 frontier verifier models. Our analysis reveals that scaling model size, adding reasoning effort, expanding to authoritative web sources, and applying medical fine-tuning do not resolve these failure modes, demonstrating that they represent fundamental limitations of the retrieve-then-verify paradigm in open-ended medical settings rather than artifacts of outdated systems. We release our code and data at https://anonymous.4open.science/r/Medical_RAG_eval-4AB5 for the full reproducibility of our results.

Everyday Extended Reality (XR) systems aim to provide context-aware access to the right functionalities at the right time and place, with minimal manual reconfiguration as users switch context. Yet these interfaces are hard to evaluate: current prototyping and user-study workflows offer no systematic, repeatable way to compare adaptation methods across users and scenarios. We present ContextXR, a novel benchmarking framework for context-aware XR interfaces. ContextXR represents an XR application as a connected graph of functional facets, each a semantically coherent group of related capabilities that together support a shared user intent. On this representation, we build MineXR++, a dataset augmenting prior XR interface data with facet-level annotations, and formulate three canonical tasks of context-aware suggestion: context factor analysis, initial facet suggestion, and next facet suggestion. Our evaluation protocol scores suggestion methods by a simulated interaction metric, the navigation and search cost of reaching the desired functionality. Through experiments benchmarking global popularity, relational retrieval, and LLM-based methods, we demonstrate that ContextXR enables the systematic, reproducible evaluation of context-aware XR interfaces.

Mixture-of-experts (MoE) models activate few experts per token but store the full expert pool. Expert pruning reduces this storage burden; at a fixed pruning budget, the goal is to preserve the original model's output distribution as closely as possible. Yet an expert's usage or contribution magnitude does not by itself determine the damage caused by its removal. What matters is whether the surviving computation can replace its function. We introduce RAZOR, a training-free expert pruning method that scores functional replaceability using consensus residuals: deviations of expert outputs from the original weighted mixture. An exact single-deletion identity at a fixed layer input accounts for survivor renormalization and router-selected refill, providing local scores aggregated over calibration tokens for budgeted pruning without gradients or recovery training. On GLM-4.7-Flash, Qwen3.6-35B-A3B, DeepSeek-V4-Flash-0731, and Hy3 at 25\% and 50\% expert removal, RAZOR achieves the highest nine-task macro average among the evaluated pruning methods in all eight settings. On the two backbones with matched REAP benchmark runs, it exceeds REAP by 2.12--5.59 points and wins all 36 paired task comparisons. It also lowers reverse KL relative to REAP in all four matched GLM-4.7-Flash and Qwen3.6-35B-A3B model--budget settings. Analysis of responses generated by Qwen3.6-35B-A3B nevertheless reveals changes in diversity, formatting, and termination, underscoring that task retention and predictive fidelity do not ensure generation stability.

Policies trained with imitation learning can accumulate errors over time, causing the robot to drift outside the training distribution. Existing methods mitigate this covariate shift by collecting additional data where the policy fails or is likely to fail. The first places the robot in unsafe conditions and the second requires choosing an appropriate noise distribution to collect new expert demonstrations under that noise. We propose Policy-Calibrated DAgger, a method that makes use of the properties of recent generative policies to estimate the policy's noise offline by using its own predicted action distribution. We measure a diffusion policy's spread of predicted actions at observations along the expert trajectory and measure its closed-loop error relative to a recorded trajectory. To address issues with measuring error in a multimodal action space, we guide the policy towards the trajectory during closed-loop control through partial denoising, and use properties of a diffusion model to unnormalize the measured error as if we did not guide it. We experiment in a scenario where a robot is tasked to reach an engine lever in a cluttered and narrow environment and show results in a 3D photorealistic simulator and a 2D planar reacher environment. We show that our method surpasses policies trained with dataset aggregation without noising and matches the performance of the best noise level in hindsight, without requiring a sweep over noise levels.

High-level robotic supervisors coordinate capabilities whose reported outcomes determine the robot's next action. Reactive synthesis can generate such supervisors with formal guarantees, but deployment requires more than proving a Generalized Reactivity (1) (GR(1)) specification realizable. Designers must encode failure-prone capabilities, choose liveness assumptions that match retry intent, audit strategies, and translate them into robot software. We present an open-source pipeline for Robot Operating System (ROS) 2 Flexible Behavior Engine (FlexBE) supervisors that generates capability-based GR(1) specifications, analyzes assumptions before synthesis, audits strategies, reduces states with a behavior-preservation proof, and emits executable state machines. Across four case studies (six comparisons), including hardware on two quadcopter platforms, we compare enumerated and one-hot encodings and two liveness formulations. Under the tested backend, enumerated encoding usually synthesizes faster, although fewer propositions do not reliably predict smaller controllers or lower symbolic cost. System-Goal without pending memory is the only liveness treatment confirmed to yield executable controllers under both encodings across the reported grid; Fair-Outcome can permit realizable cycles without designer-intended completion. For this backend and model, we recommend enumerated encoding with System-Goal and auditing every realized strategy, since proposition count and realizability do not measure deployability. The auditor is sound and complete for four structural defect classes (protocol violations, deadlocks, bounded-failure violations, goal-unreachable traps) but is not a general liveness verifier, and the reduction preserves capability-level behavior. Together, these stages narrow the gap between formal realizability and controllers that pass protocol and structural-progress checks.

Non-generative "System-1" models return structured probabilistic decisions in a single forward pass, without autoregressive decoding, at a small fraction of the inference cost of a generative model. This makes them of interest as inexpensive components in larger pipelines, but their reliability on biosecurity-relevant tasks has not been systematically examined. We audit one commercial System-1 model on 6,020 multiple-choice items drawn from the Weapons of Mass Destruction Proxy (WMDP), a paraphrase-robust WMDP-Bio variant, and six LAB-Bench subtasks, measuring accuracy, calibration, error detection, selective prediction, and sensitivity to the order in which answer options are presented. Accuracy is strongly task-dependent. Once the vendor's uncertainty field is correctly interpreted, the model is reasonably well calibrated (pooled expected calibration error 0.034) and its top-1 probability separates correct from incorrect predictions (pooled AUROC 0.820), though both degrade substantially on the weaker tasks. Under four cyclic rotations of the answer options, 37.4% of WMDP-Cyber items receive different answers; a control using byte-identical repeated calls attributes most of this to option order rather than run-to-run variation. Averaging probabilities across rotations improves WMDP-Cyber accuracy by 3.8 percentage points, and applying it only to low-confidence items recovers most of that gain at well under the cost of averaging every item.

Parameter estimation for ordinary differential equation (ODE) models is a fundamental task that is often complicated by the limitations of conventional optimization-based methods. In theory, differential-algebraic approaches offer an appealing alternative: they reduce the problem to polynomial system solving and do not require user-supplied initial guesses for parameter values. In practice, however, algebraic methods have been limited by their sensitivity to measurement noise, because they require accurate derivatives of observed outputs. In this work, we integrate Gaussian Process Regression (GPR) into the differential-algebraic method and derive a first-order error analysis in terms of noise level and algebraic sensitivity. We evaluate the method across several noise levels on a benchmark of 25 dynamical systems arising in applications including mechanical engineering and systems biology. The proposed method achieves the highest aggregate performance among the methods considered, recovering all sought parameter values and initial conditions to within 10% relative error in 88.5% of runs. These results demonstrate that robust derivative estimation can make differential-algebraic parameter estimation practical for dense, noisy synthetic data while retaining key advantages of the algebraic formulation.

Optical lens design is a complex, non-convex optimization challenge that relies heavily on human experience and intuition. Existing optimized-based automatic lens design methods struggle to navigate this vast parameter space without meticulous manual tuning. In this paper, we present LensDesigner, an autonomous agent framework that mirrors the problem-solving workflow of expert opticians. To overcome the initial cold start problem, we construct LensLib100K, an extensive optical lens library, and employ Optics-Aware Retrieval to supply physically valid structural seeds. Within an interactive physical simulation environment, the agent executes macroscopic orchestration while receiving immediate optical feedback. Furthermore, we introduce a continuous self-evolving mechanism guided by a curriculum agent. By iteratively solving design tasks with progressively increasing difficulty, the agent autonomously extracts, accumulates, and reuses design heuristics, effectively evolving its optical lens design expertise over time. At the evaluation level, we introduce LensArena, a standardized evaluation benchmark comprising $120$ diverse optical design tasks, covering extreme configurations. Extensive experiments on this benchmark demonstrate that LensDesigner significantly outperforms publicly available baseline algorithms, achieving superior success rates and optimization efficiency. We hope this work sheds light on the emerging field of intelligent optics. The code will be publicly available.

Driving world models learn rich predictive representations of the surrounding environment from visual observations, yet accurate visual prediction does not necessarily translate into effective trajectory planning. We argue that a key bottleneck lies in the mismatch between visual world states and raw geometric trajectories, which may limit the planner's ability to exploit action-relevant semantics encoded by the world model. To address this issue, we propose World-Model Alignment for Latent Trajectories (WALT), which learns a compact generative trajectory latent space by transferring information from a frozen pretrained driving world model without modifying the world model itself. Rather than directly generating raw waypoints, WALT maps them into compact representations through a dual-branch trajectory autoencoder and transfers semantic knowledge from the frozen visual world model into this trajectory space, encouraging the learned action representation to capture scene-level cues relevant to future motion and planning. Beyond our proposed formulation, we systematically study latent learning based on Joint-Embedding Predictive Architectures (JEPA) and feature alignment following Representation Alignment (REPA) to investigate how trajectory-only representation learning affects downstream planning. We evaluate WALT on the NAVSIM benchmarks. Relative to the raw-waypoint baseline, WALT improves PDMS from 89.4 to 89.8 on NAVSIMv1 and EPDMS from 87.3 to 87.9 on NAVSIMv2 while reducing trajectory planner FLOPs by 30.5%. These results suggest that preserving world representations while extracting action-relevant information provides an effective interface for world-model-based trajectory planning.

Pre-trained vision-language models such as CLIP can recognize new categories via prompting, but they often struggle when labeled data are scarce or the test distribution shifts. Prompt learning adapts only a small set of parameters while keeping the backbone frozen, yet many existing multimodal prompt learners couple the visual and textual branches weakly and can be brittle in low-shot regimes. We propose ProCAP, a probabilistic cross-attentive prompt learning framework that improves cross-modal interaction and training stability without updating any CLIP weights: it learns both visual and textual prompt tokens and links them through stacked bidirectional multi-head cross-attention so the two branches refine each other across prompt depth. To reduce overfitting under limited supervision, we parameterize prompt tokens with Gaussian means and variances and regularize them with lightweight KL and L2 penalties, and we further add a compact symmetric InfoNCE head that aligns cross-attended image features with class-level text representations in a shared low-dimensional space. Across few-shot base-to-novel generalization on 11 datasets, cross-dataset transfer, and domain generalization on ImageNet shift benchmarks, ProCAP achieves strong aggregate base-to-novel performance and competitive transfer performance while keeping the CLIP backbone unchanged.

Recent advancements in image-based profiling techniques have enabled the collection of high-volume cell morphology data, allowing new molecular embedding models to learn from the experimental phenotypic perturbations of a molecule in a cell. Previously, we developed Molecule-Morphology Contrastive Pretraining (MoCoP), a strategy for aligning small molecule embeddings to morphology fingerprints extracted through CellProfiler. The resulting molecular representation showed transferable performance for quantitative structure--activity relationship (QSAR) prediction tasks. Here, we extend the method by using a deep-learning-based cell image encoding pipeline to extract more feature-rich morphology profiles and align them to the molecular embeddings through contrastive learning. The new embeddings encode more accurate information on how molecules perturb cell morphology and enable improvements for QSAR predictions through either fixed-embedding linear probes or fully flexible fine-tuning. Morphology retrieval performance scales log-linearly with training data size, suggesting continued improvements as larger datasets become available. The improved MoCoP v2 also achieves superior performance on toxicity prediction and competitive results on ADME and activity benchmarks, when compared with existing molecular embedding models that use both cell morphology and transcriptomic data during training.

Disinformation research has produced increasingly accurate automated fake-news detectors, but many systems remain difficult to interpret and are weakly connected to established theories of persuasion, credibility, and human judgment. In this paper, we develop a theory-informed computational framework that translates cross-disciplinary theories of fake news into measurable features for automated detection and explanation through statistical techniques and large language models. To that end, we conduct a structured cross-disciplinary review of theories from social sciences, psychology, economics, among other disciplines that reveal how fake news persuades and spreads, thereby establishing a broad theoretical foundation for computational modeling. Experiments on benchmark datasets show that theory-derived features are predictive and provide interpretable, theory-referenced diagnostic signals. Multi-feature models generally outperform individual features, although gains among the strongest small feature combinations are modest. Our work highlights the value of interdisciplinary perspectives in building robust and interpretable fake news detection systems, advancing the foundation for human-centered approaches in combating disinformation.

Large language models (LLMs) are increasingly deployed in settings where responses must reflect multiple, potentially interacting social norms and human values. Activation steering offers a lightweight alternative to training-based alignment by modifying internal activations at inference time. However, prior human-value steering methods have largely considered values in isolation, while direct composition of multiple directions relies on fixed intervention strengths that cannot respond to the model's evolving internal state. Motivated by this key observation, we introduce AIMES, a framework for adaptive multi-value activation steering. AIMES constructs layer-specific bipolar directions for moral-foundation values and uses intermediate-layer vocabulary readouts as online observers. An observer-guided controller then adapts the strength of each requested value intervention at every decoding step based on its current observed state, without training a separate value-state estimator. Across multiple instruction-tuned model families, value combinations, and intervention depths, we find that multi-value controllability varies across both value combinations and intervention locations. Compared with fixed joint steering and prompt-based steering, AIMES shows depth-dependent advantages that are broadly supported across two independent evaluators, with some variation in the precise depth at which specific control effects emerge. These advantages come with smaller realized activation-space interventions than fixed-joint steering and comparable response quality. Overall, our results suggest that online observer feedback can provide lightweight, state-aware adaptation for single-pass multi-value steering.

We present POIL, a point-based one-shot imitation learning framework with stable dynamical systems. While one-shot imitation avoids collecting extensive demonstrations, successful one-shot manipulation requires not only transferring a demonstrated trajectory to a novel object but also executing it robustly under changing scene conditions, grasp configurations, and external disturbances. POIL addresses both problems through a shared representation: a set of 3D points on the object's functional part, used jointly for trajectory transfer and closed-loop execution. The one-shot transfer from the demonstrated trajectory is enabled with point correspondences. POIL grounds the shared functional part with a multi-modal large language model, and transfers the trajectory across viewpoint, pose, and object category changes. During execution, multi-view tracking observes the same points online, and Point-set BCSDM drives them in closed loop by projecting per-point velocities onto a single rigid-body twist computed from the tracked points alone. This extends stable dynamical models from an SE(3) pose to a point set without requiring a known 3D model or pose estimator. We show that at the goal the controller becomes a gradient flow on the classical SO(3) potential, so its terminal phase inherits the almost-global convergence of that potential under a rigid-object assumption. Across simulation and real-robot experiments, POIL transfers a single demonstration across object category, grasp pose, and goal geometry, while recovering from external disturbances during execution. Project page: https://sangminkim-99.github.io/poil

Multimodal misinformation is increasingly crafted to look convincing by pairing a textual claim with an image that appears to "prove" it. Yet in practice, building effective detectors often hinges on a small set of design choices that are rarely examined in a controlled way. In this paper, we conduct a large-scale study of multimodal design choices for misinformation detection with over 3,375 experiments- spanning three benchmark datasets and a broad range of pre-trained vision and language backbones. Through systematic comparisons and targeted robustness analyses, we distill practical guidance on which design choices help, when do they fail silently, and what aspects of the pipeline most strongly shape model behavior, answering 4 key Research Questions (RQs). We aim to provide a reliable foundation for designing stronger and more dependable multimodal misinformation detection systems, thus contributing to the broader research community.

Evaluating explainable Artificial Intelligence (XAI) methods is a challenging task due to the lack of reliable evaluation procedures and, in particular, the absence of ground truth explanations. In the literature, existing evaluation approaches typically assess explanations by measuring their fidelity with respect to the predictions of a black-box model. However, such evaluation strategies only quantify the degree to which an explanation reproduces the model's output, without ensuring that the explanation correctly reflects the underlying decision process. As a consequence, different explanations may achieve similar fidelity scores while providing inconsistent or misleading interpretations of the model behavior. In this paper, we propose a framework for the evaluation of XAI methods based on synthetic ground truth. The proposed approach relies on controlled interventions to generate synthetic datasets in which the importance of input components can be determined by design. This enables the construction of ground truth explanations that are directly aligned with the behavior of the model under analysis. The framework is instantiated across three data domains, namely binary images, tabular data, and time series, allowing a comprehensive assessment of explanation methods in heterogeneous settings. Experimental results obtained by evaluating nine widely used XAI methods show significant limitations in current techniques and highlight the importance of synthetic, intervention-based benchmarks for a reliable assessment of explanation quality.

Real-time tiny object detection in aerial imagery is constrained by the weak spatial evidence of very small objects and the loss of high-resolution detail in lightweight detectors. This study presents Cross-Scale Channel-wise Knowledge Distillation (CSCWD), a training-time framework that transfers high-resolution spatial representations from a YOLO11m-P2 teacher to a compact YOLO11n student without altering the student's inference architecture. Unlike conventional same-scale feature distillation, CSCWD transfers supervision from teacher P2 to student P3 after feature alignment while retaining same-scale distillation at deeper pyramid levels. Under the unified seven-sequence Drone-vs-Bird validation protocol, YOLO11n-CSCWD achieves 50.17% mean average precision at an intersection-over-union threshold of 0.5 ([email protected]) and 59.73% recall, improving the matched CA-YOLO11n baseline by 2.92 percentage points in [email protected] and 3.55 points in recall. Cross-scale alignment further increases [email protected] by 2.09 points over the corresponding same-scale channel-wise distillation configuration. In zero-shot evaluation on DUT-Anti-UAV, [email protected] increases from 48.29% to 50.06% without target-domain fine-tuning. This domain was included because its challenging small targets make low-latency, computationally efficient detection particularly relevant. On Raspberry Pi 5 using NCNN-FP16 at 640x640 resolution, the 2.58-million-parameter student achieves 50.32% [email protected] at 82.32 ms mean wall-clock latency, or 12.15 frames per second, while retaining essentially the same runtime and memory requirements as the matched baseline. The results support cross-scale distillation for improving tiny-target detection without increasing inference-time model complexity.

Existing in-context reinforcement learning methods mainly pretrain Transformers with supervised behavior-prediction objectives. This enables task inference from context, but makes the learned policy strongly depend on the quality of offline actions: when trajectories are weak or suboptimal, imitation itself becomes a biased learning signal. We propose Q-Target Pretrained Transformers (QTPT), which keeps the context-conditioned Transformer architecture but replaces behavior cloning with a Bellman-style Q-target objective. QTPT therefore learns to use rewards and transitions in the context to estimate action values, rather than simply imitating the behavior policy. We theoretically analyze QTPT in stochastic linear bandits and finite-horizon MDPs, showing stronger robustness to data quality than supervised pretraining. Empirically, QTPT improves over supervised behavior prediction on controlled RL benchmarks with random or suboptimal data, and we examine extensions to D4RL Kitchen and AntMaze. Supplementary experiments evaluate backbone robustness, meta-RL comparisons, task-coherent context, and unsupported-action value overestimation. These comparisons distinguish the benefits of Q-target pretraining from the remaining limitations of offline coverage.

A skill is a modular package of natural-language instructions, executable scripts, and reference resources that an agent can load at runtime to extend its capabilities for a specific task. Skill-based agent systems therefore enable flexible reuse of third-party capabilities, but the openness of this skill ecosystem also opens up a new attack surface. Prior work has focused on vulnerabilities within individual skills, but little attention has been paid to risks that arise from interactions across skills. In this paper, we introduce skill cascading attacks, a threat paradigm in which a malicious objective is distributed across multiple skills so that each modification looks benign in isolation, yet their combined execution is harmful. For instance, in a prescription-review pipeline, the first skill weakens signals of recently discontinued medications in the extracted history, the second downgrades the severity of any drug interaction tied to them, and the third suppresses the resulting low-priority alert in the final summary, so that a severe drug-interaction warning silently disappears before reaching the physician. To systematically study this safety blind spot, we develop SkillCascade, an automated multi-agent red-teaming framework, and release SkillCascade-Bench, a benchmark of 213 validated cascading test cases across multiple agent systems and domains. Across representative agents (e.g., OpenClaw, Claude Code, Codex) and LLM backbones, cascaded interactions reliably induce harmful behaviors while evading existing per-skill scanners and runtime monitors. Our findings highlight a gap between component-level integrity and system-level safety, and call for defenses that reason over cross-skill interactions rather than individual skills in isolation.

Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent traces to reconstruct what happened. These analyses assume that LLM agents cannot tamper with their own execution traces. We show that local LLM agents such as Claude Code, Codex, Antigravity, Open Code and Grok Build fail to enforce this boundary. All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails. We also validate that external attackers can exploit this gap to induce trace deletion. Finally, we show that trace tampering behavior emerges naturally in frontier models, when agents try to improve their rewards. We advise practitioners to ensure trace logging happens through an independent interception mechanism outside of the agent's control, preserving trace integrity even in cases of full host compromise. Overall, our findings identify a concrete failure of trace integrity in agent infrastructure which can be used to conceal misaligned behaviors like scheming or sabotage.

Latent world models are typically trained to predict factual transitions, whereas model predictive control (MPC) must compare alternative actions from the same state. A model can therefore achieve low factual prediction error yet poorly distinguish candidate actions. We introduce AD-WM, an action-discriminative joint-embedding world model for counterfactual MPC. AD-WM combines residual latent dynamics with predictor-level action-recovery regularization, using inverse dynamics and a normalized recovery objective motivated by conditional mutual information. Both objectives encourage planning transitions to preserve action information; their auxiliary heads are discarded at test time, leaving MPC unchanged. On OGBench-Cube, AD-WM improves hard-start success from 3.7% to 52.0% over a matched LeWM baseline and improves mean success over the reproduced baseline in four of five simulation environments. Planning diagnostics show that factual prediction error and whole-bank action ranking do not follow the closed-loop success ordering, whereas CEM-aligned elite regret tracks success more closely. With a frozen V-JEPA 2 encoder and matched DROID post-training, AD-WM also improves zero-shot transfer to our Franka setup, increasing basic pick-and-place success from 42.2% to 71.1% without lab-specific adaptation. These results suggest that world models for planning should preserve action-dependent differences needed for counterfactual selection, rather than optimize factual prediction accuracy alone. More videos and code are available at https://ad-wm.github.io/.

Distributed learning in embodied reinforcement-learning agents offers a degree of privacy by retaining raw sensor data on-device and transmitting only policy gradients to the server. Yet temporal structure can amplify this leakage beyond single-frame attacks. We introduce Temporal Reconstruction Attack on Consecutive Encodings (TRACE), an amortized temporal gradient-inversion attack that autoregressively reconstructs the sequence of private observation-action trajectories from per-step policy-learning gradients. The attack exploits two structural signals ignored by prior single-frame methods: (i) cross-time correlation between successive embodied gradients, which we formalize via a conditional mutual-information bound, and (ii) closed-form action recovery from policy-head gradient structure, which we prove exact when standard entropy regularization is sufficiently small. On held-out embodied scenes, TRACE reaches $18.8$ dB PSNR with near-perfect action recovery at $3$-$4.5$ ms per reconstructed frame, dominating the learning-based baseline across all reconstruction metrics and exceeding optimization attacks while running orders of magnitude faster. Further evaluation demonstrates TRACE's broader applicability across recurrent, residual, and compact transformer victim architectures, multi-modal inputs, and larger discrete action spaces. Defense experiments suggest that protecting temporal gradient streams may require sequence-aware privacy mechanisms.

Conspiratorial discourse on social media is not always expressed through explicit claims or stable lexical markers. The same surface content may express endorsement, legitimate concerns, criticism, satire, or mockery. The main challenge is therefore not only recognizing conspiracy-related claims, but inferring the speaker's intent -- the utterance's illocutionary force. We argue that this can be achieved through the use of relevant social contexts and propose an agentic framework, equipped with a set of tools supporting social queries. We demonstrate the benefits of our approach on a unique dataset of Hebrew tweets, covering 80\%--90\% of the public Hebrew tweets published over a four-year span (late 2018-- early 2023), encompassing several election cycles as well as the COVID pandemic years and related vaccination campaigns. This extensive coverage can be used in recovering different social contexts. Evaluating our framework on a manually-annotated adversarial dataset, we find that context-aware workflows consistently outperform text-only classification and that the agentic framework performs significantly better than other frameworks and settings, including a non-agentic model exposed to the same contexts available to the agent. We further provide an analysis of the results, the errors and efficiency (token economy) tradeoffs. These findings support viewing the task of conspiracy detection as a socially embedded interpretation task, in which effective classification depends not only on access to contexts, but also on adaptive reasoning in which the agent uses tools on a per-case basis, asking only for evidence relevant to its current reasoning step.

Coding agents have demonstrated enormous success in solving complex programming problems. To leverage their potential for robot systems, this work introduces Robot Agentic Programming from Demonstrations (RAPID), which automatically generates, verifies, and refines robot programs, given a single visual human demonstration. The iterative agentic loop of code refinement requires several key ingredients: (i) a testable task specification, (ii) action primitives for robot execution, and (iii) an interactive environment for program execution and verification. RAPID infers all three from the demonstration automatically. To make the resulting program reusable beyond the demonstration setting, RAPID uses an object-centric relational program representation that focuses on the underlying structure of the demonstrated strategy rather than the specific motion per se: it expresses the action primitives as trajectory-optimization programs that realize object-level motion effects, while composing them through relational constraints that capture scene-specific geometry at run time. We evaluated RAPID in simulation on eight challenging contact-rich nonprehensile manipulation tasks as well as general prehensile manipulation tasks in the LIBERO-Pro benchmark. We also successfully deployed it on a real Franka arm and evaluated on all eight nonprehensile tasks. In all experiments, RAPID demonstrated strong performance, with generalization over object pose, shape, material, and environment. Website: https://yuyaoliu.me/projects/rapid.

World Action Models (WAMs) couple action generation with future visual prediction for robotic manipulation. However, completing the joint video-action denoising process at each replanning cycle incurs substantial latency, delaying action updates and limiting closed-loop responsiveness. We present Rolling-WAM, a formulation that distributes joint denoising across successive replanning cycles. Our method maintains a sliding window of video-action chunks at staggered noise levels. At each step, a rolling noise schedule fully denoises the imminent action chunk for execution, while partially refining farther-future chunks. As the window advances with new camera observations, the retained future chunks continue their denoising process. This distributes the computational cost over time while carrying an evolving visual-action context across chunk boundaries. Evaluations on LIBERO, RoboTwin, and a real-world Unitree G1 humanoid show that Rolling-WAM achieves competitive manipulation performance. By removing the need to denoise the entire prediction horizon from scratch, it delivers a 4.5x steady-state replanning speedup over standard joint WAMs.

Recent research on Multimodal Sentiment Analysis (MSA) has focused on learning from language, visual, and acoustic modalities with incomplete data to infer human sentiment. Most studies typically compensate for missing information by reconstructing modality features or designing complicated fusion mechanisms. However, these methods still suffer from spurious generation and noisy guidance due to the lack of high-level semantic grounding in partially observed multimodal evidence. To address these issues, we propose SemMSA, a latent semantic-aided framework that constructs rich sentiment-relevant semantics with LLMs, fully integrating with all modalities via anchor-free spectral alignment. It mainly consists of Cross-modal Semantic Refinement (CSR) and Cross-modal Spectral Alignment (CSA). Specifically, CSR first adaptively extracts visual and acoustic representations by corresponding adapters to form a unified multimodal prefix with language in the frozen LLM embedding space. It then iteratively produces continuous discriminative semantic states through a token-efficient latent refinement process without decoding explicit text. Next, CSA simultaneously aligns the refined semantics with all modalities by enhancing the dominant spectral component of their kernel Gram matrix. This captures global nonlinear dependencies among all representations without relying on a predefined anchor modality. In addition, an instance-level spectral separation constraint preserves cross-sample discriminability and mitigates representation collapse. Extensive experiments on SIMS, MOSI, and MOSEI benchmarks demonstrate that SemMSA achieves state-of-the-art performance.

Task and motion planning (TAMP) problems remain difficult even with full observability and object-centric states because discrete decisions are tightly coupled to geometric, kinematic, and dynamic constraints. Generalized TAMP addresses this difficulty by exploiting regularities across problem instances to reduce planning effort on new instances. However, existing methods require substantial TAMP-specific engineering. We investigate whether coding agents can automate this process by synthesizing programs that generalize across instances. Given a task description and simulator access, each agent chooses how to interact with the environment while developing a program within a fixed synthesis budget. The program is then frozen and evaluated on unseen instances. We evaluate Claude Code (Opus 5) and Codex (GPT-5.6 Sol and GPT-6 Astra) on 28 simulated environments from KinDER and PDDLStream, with object counts beyond those evaluated in the original benchmark. Across all program synthesis methods, we evaluate 980 generated programs on 100 held-out instances each, 98,000 evaluation episodes in total. Overall, we find that coding agents are surprisingly effective at generalized TAMP: all three agent configurations outperform hand-engineered planners, one-shot generation, and an LLM-based generalized planning baseline in mean success (56% to 95% versus 47% for the planners, on the 16 environments where a planner is available). As object counts grow, the agents' programs maintain higher success than the planner, using an order of magnitude less computation per instance on average. Logs show agents using interaction to calibrate physical models, test edge cases, and refine strategies. We release all code, including the full prompts given to the agents. These findings suggest that coding agents are a strong baseline for generalized TAMP.

Online misinformation increasingly appears in spoken formats such as news clips, podcasts, interviews, political speeches, and social media videos, creating a need for fact-checking systems that can verify claims directly from speech. We introduce VeriSpeak, a probe benchmark for studying speech-based fact verification in Large Audio Language Models (LALMs). VeriSpeak contains 3,879 spoken claims spanning temporal, geographical, and relational facts, with balanced true and false labels. The benchmark is designed to examine whether factual verification ability transfers from text to speech, and whether retrieval-augmented LALMs can use textual evidence to correctly support or refute spoken claims. Our experiments reveal a consistent text-speech modality gap: LALMs that verify written claims reliably often fail on the same claims when spoken. Moreover, retrieval alone provides limited gains because models frequently conflate retrieved evidence with the spoken claim. In contrast, retrieval combined with explicit reasoning improves claim-evidence comparison, with a thinking-tuned LALM reaching 86.1% accuracy. VeriSpeak highlights that effective speech misinformation detection requires not only speech understanding, but also grounded reasoning over retrieved evidence. The dataset is publicly available via Hugging Face at https://huggingface.co/datasets/abhiram4572/VeriSpeak.

Foundation models are post-trained with reinforcement learning (RL) to maximize specific rewards, such as human alignment, correctness, or instruction following. This post-training process is computationally intensive, sometimes unstable, and has to be run from scratch every time the reward model changes or when we want to combine multiple rewards. We hence ask: given a new reward function, is it possible to predict the RL outcomes without actually running RL on it? We answer this in the affirmative by introducing PoEM, a framework to predict the outputs of RL on a new reward function using a set of models already post-trained on other rewards. First, we show that if the new reward function can be written as a linear combination of existing ones, then the new policy in log-space can be written as a linear combination of the existing log-policies. Surprisingly, even in cases where the rewards are not linearly connected, we observe that often log-policies from RL training span an approximately low-rank subspace across rewards. To our benefit, the weighting coefficients for this combination can be estimated using only the reward or basis policy outputs on the samples. We turn these observations into an algorithm that takes post-trained models and a new reward function, and approximates the target RL policy without actually running any additional RL training. We experimentally validate our approach across synthetic and real rewards, spanning both text and image modalities.

Video generation begins in text space by authoring a cinematic screenplay, then materializes into pixels. As contemporary video generators scale to 30 seconds and faithfully follow complex conditions, the textual prompt largely directs the production, planning how actions, camera trajectories, lighting, and sound unfold across multi-shot sequences. In this paper, we present WanPE, a 397B-parameter prompt enhancement model trained on 1.05M real-world videos to master director-level cinematic planning. WanPE formulates shot-level cinematic plans via video-grounded reverse construction and employs Semantic-Consistency GRPO (SC-GRPO) to faithfully preserve user requirements across shots and over time. To benchmark this capability, we curate WanPEval, a human-annotated testbed covering durations from 5 to 30 seconds across varying intent granularities, supported by approximately 11K blind pairwise assessments. When powering Wan3.0's video generator, WanPE-397B boosts human preference over raw user prompts by 10.66-18.84 points at 5-15 seconds and by a dramatic 50.86 points in the 30-second arena. Ablation studies show that reverse construction demonstrates clear superiority over forward rewriting, while SC-GRPO robustly preserves semantic fidelity across model scales. Ultimately, WanPE leads all evaluated commercial offerings at 5-15 seconds and remains competitive with Seedance 2.5 at 30 seconds.

An acceptance protocol is developed for sensor-coordinate and polarity binding in mechatronic commissioning. Candidate generation is separated from release authority. Requirements unsupported by a deterministic parser are routed to a frozen local language model with four billion parameters. Plans are released only when both facts can be derived by an external gate under a sealed grammar. One canonical answer is requested from a gold-standard user when eligible. The protocol was evaluated once under a criterion fixed before benchmark construction, on 144 tasks written by isolated agent contexts without access to the gate, grammar, or experimental plan. Three contributions are established. First, candidate generation and release decisions were measured separately. Fabricated ready plans were committed on 21 of 22 routed unanswerable tasks, and all were rejected. The same 83 releases were reproduced without model calls. Second, no false release was observed among 83 releases. A one-sided 95% Clopper-Pearson upper bound of 0.0354 was obtained as a diagnostic under an independent-and-identically-distributed assumption, below the sealed 5% threshold. However, one false release was subsequently recorded among 146 releases outside the benchmark at seed 0. Third, protection against incorrect user answers was characterized. Both facts were bound from the original text on 13 of 96 answerable tasks. Incorrect answers were released in 169 of 431 pairings on the remaining tasks, including failures involving coordinate exclusion. A deployable questioning policy was not tested because eligibility was determined from the answer key. Gate sensitivity and real user behavior were not measured.

A central concern in AI safety is that agents may treat oversight as an obstacle when it conflicts with completing their goals. We study instrumental evasion, the propensity of LLM agents to circumvent runtime monitoring as a means of completing ordinary tasks. We introduce EvasionBench, a benchmark of 50 diverse task-policy pairs in which completing the task requires an operation prohibited by a runtime monitor. Agents know that their tool calls are monitored and are prompted to continue working when they pause. Across our evaluations, best-of-3 evasion attempt rates reach up to 98% and success rates up to 88%, with substantial variance across models. Claude Fable 5.1 succeeds less often, but frequently makes creative attempts to circumvent the monitor. Evasion generally increases with test-time compute, with higher evasion rates at greater reasoning effort and token use. Traces show that agents encode prohibited commands, decompose operations across tool calls, and retry until relevant context leaves the monitor's history. We also find that GPT-6 Astra's low evasion rate comes with overrefusal, as it frequently abandons otherwise solvable tasks under a denial-of-service prompt injection. Our findings show that ordinary task pressure can lead to adaptive attempts to evade runtime monitors without an explicit adversarial objective. Effective oversight must therefore remain robust against repeated attempts, as the persistence that helps agents solve difficult tasks can also drive them to circumvent their guardrails.

We present Underwater C$^{3}$-JEPA (cross-view, control-conditioned, context-extended), an object-centric multi-view predictive world model for near-field heavy-load underwater ROV salvage. Without contact sensors, it predicts in latent space how the task-object state evolves through contact interaction and under the hydrodynamic lag of the vehicle, from synchronized multi-view RGB observations and vehicle control signals. C$^{3}$-JEPA encodes multi-camera observations into task-object and context tokens, fuses cross-camera evidence through held-out-view attention, and directly predicts future states conditioned on control. Weak binding anchors the target and gripper at low annotation cost, while SIGReg sharpens the geometric representation. Experiments show that the learned representation transfers substantially more task-relevant information to downstream probes than a reconstruction-free latent baseline, while keeping the predictor lightweight. The resulting predictive interface supports model-predictive-control (MPC) candidate evaluation and imagined-rollout behavior-agent training. Validation on real underwater video shows the same architecture recovering a withheld camera's object state and staying ahead of persistence, so the recipe transfers beyond simulation.

Robots often must satisfy one or more constraints during motion planning for real-world tasks. When such constraints reduce the valid configuration space to a measure-zero subset, sampling based planning algorithms require modifications to draw feasible samples. For many common end-effector constraints, parameterizations built on inverse kinematics (IK) provide an alternate formulation where the constraints are satisfied by construction, allowing directly sampling the feasible set. Despite their elegant approach, parameterized planners have remained slower than vector-accelerated implementations of projection-based approaches, leaving their performance ceiling an open question. We explore a new axis of vectorization built upon reparameterizing the planning space through analytic IK. This approach addresses existing inefficiencies in vectorized projection-based planners and exposes new opportunities for parallelism within the planner. We show that the planner can synthesize plans in microseconds to milliseconds for high dimensional systems (up to 20 dimensions), with complex constraints, up to 10x faster than the current state-of-the-art. Furthermore, we demonstrate how such planning speeds open up avenues for restructuring sequential manipulation pipelines.

Large language model (LLM)-based clinical assistants are increasingly being integrated into electronic health record (EHR) systems, transforming how clinicians retrieve and synthesize information from patient records. Their safety and utility depend on rigorous evaluation, yet existing benchmarks are manually curated, costly to update, and rapidly become obsolete with evolving technological advancements. We present a scalable framework that automatically generates question--answer pairs from longitudinal EHR notes. Nineteen clinicians validate the benchmark generator, producing the Benchmark for Retrieving Information in EHRs (BRIE), a continuously maintainable evaluation dataset. Across nine LLMs and five inference strategies, state-of-the-art systems frequently omit clinically important information, particularly for questions requiring synthesis across multiple documents and encounters. Because the generator itself is validated, BRIE supports evaluations that static benchmarks cannot, including the generation of multiple answers that reflect variation in clinician reasoning for robust performance assessment and continuously refreshing benchmark content to guard against leakage. Our results demonstrate that scalable benchmark generation enables rigorous, up-to-date evaluation of clinical LLMs as they are deployed in rapidly evolving healthcare settings.

Scientific discovery begins where known problems end. There, AI systems must engage in exploration: framing hypotheses, designing experiments, and iterating on the results. However, evaluating this ability is difficult: (1) how to verify whether a genuinely new hypothesis holds, and (2) how to determine whether a system has discovered it through exploration or merely recalled related knowledge from pre-training data. To this end, we introduce ExplorationBench, which turns the wicked problem of evaluating scientific exploration into a concrete and tractable framework built on verifiable Alien Worlds: their rules are executable, so every answer can be checked exactly, and they conflict with familiar knowledge, so recall alone cannot solve the tasks. The benchmark contains two sandboxes, AlienCode (31 discovery targets, 70 tasks) and AlienLogic (24 discovery targets, 70 tasks). Each sandbox provides a flawed manual, task-specific environmental feedback, and a dedicated tool-call schema. Systems use these resources to explore the sandbox, then solve held-out tasks. We evaluate 10 AI systems and find that the strongest systems can acquire and apply unfamiliar rules, while performance varies substantially across trajectories and continued exploration can stall or reverse earlier gains. ExplorationBench represents a step towards AI systems that can acquire and apply genuinely new knowledge through exploration in unknown environments.

Latent neural surrogate solvers, or latent dynamics models, accelerate simulations of time-dependent physical systems by evolving a compressed latent space rather than resolving full-resolution fields directly. In principle this reduces computational cost and simplifies learning, but in practice errors often accumulate rapidly during long autoregressive rollouts, limiting predictive utility. We show that this instability does not stem from the latent representation itself, but arises when it is trained solely for reconstruction, producing representations poorly suited to long-horizon forecasting. We systematically evaluate training-level interventions that align latent representations with long-horizon rollout: Koopman operator learning and Hamming noise injection during autoencoder training to improve compression, together with noise injection and multi-step rollout fine-tuning to improve dynamics. Interventions that improve long-horizon rollout stability often degrade conventional training metrics, including reconstruction and one-step prediction accuracy. Collectively, these interventions reduce long-rollout error by approximately 40\% and match or exceed the accuracy of full-resolution models on two physics benchmarks, while requiring 2 orders of magnitude fewer floating point operations and half the GPU memory. Applied to mesoscale crystal-plasticity simulations of high-cycle fatigue, the resulting surrogate achieves stable extrapolation over horizons orders of magnitude beyond those observed during training. More broadly, these results show that neural compression should be designed not merely to reduce dimensionality, but to restructure the solution space for stable dynamical evolution, a key requirement for reliable, efficient neural surrogates in scientific applications.

Long-horizon reasoning remains a central challenge for large language models (LLMs) under sparse-reward regimes. We argue that this brittleness arises from two biases induced by complex reasoning spaces: an exploration bias, where models are drawn toward locally plausible but structurally unstable branches, and a compounding bias, where small local deviations accumulate across depth and suppress rare rewards. We introduce Symbolic Closure Analysis (SCA) as a theoretical lens characterizing how branching structures and sparse rewards induce these biases in long-horizon reasoning with local admissibility, and as a design principle for structural priors in less formal reasoning tasks. Motivated by this analysis, we propose SAGE (Structural Admissibility-Guided Exploration), a unified framework that injects structural guidance to alleviate exploration bias and compounding bias in long-horizon reasoning. SAGE combines two complementary structural guidance: algebraic sparsification, which projects locally admissible candidates onto operator-indexed algebraic subspaces to suppress spurious branching and mitigate exploration bias, and hyperbolic structural guidance, which embeds reasoning states into a negatively curved space to provide dense depth-wise signals and mitigate compounding bias. Across 12 benchmarks and 7 model families, SAGE outperforms competitive baselines. In particular, SAGE achieves up to an 8-fold improvement on the Andrews-Curtis problem, an open real-world long-horizon task. Code is available at: https://github.com/Susan571/SAGE-NeurIPS2026.

Ego-Exo4D is a large-scale dataset providing synchronized egocentric and multi-view exocentric video, a rich resource for skill learning and assessment, procedural activity understanding, and embodied AI. However, the dataset ships with only sparse 3D human pose annotations, and reconstructing dense human motion from its multi-view captures is nontrivial. To this end, we present Ego-Exo4D-HM, a large-scale dataset of 4D human motion reconstructions for Ego-Exo4D's captures, and release the accompanying reconstruction pipeline. The code, dataset, and documentation can be found at https://abhiram824.github.io/egoexo4d_human_meshes.

Vision-language models (VLMs) have become a common foundation for autonomous mobile GUI agents, but most existing systems rely on the VLM for both planning and action grounding at nearly every interaction step, leading to substantial latency and model-serving cost. We introduce Jev-Mobile, which shifts this paradigm to low-frequency VLM planning and high-frequency lightweight execution: the VLM specifies local goals, the accessibility tree defines a structured executable action space, and Jev, a fast typed decision model, repeatedly selects actions within this space. This design allows multiple GUI actions to be executed under a single VLM decision, reducing expensive VLM inference while preserving adaptive interaction. On the full AndroidWorld task suite, Jev-Mobile achieves 79% task success, compared with 78% for SeeAct-V and 84% for a Step-wise VLM baseline. Among successful trajectories, it reduces mean end-to-end execution time by 32.7% and mean model API cost by 73.4% relative to Step-wise VLM. These results show that decoupling high-level VLM reasoning from low-level action execution can substantially improve mobile GUI agent efficiency while maintaining competitive task performance.

U.S. employment-discrimination complaints describe complex event sequences that are not explicitly captured by lexical or embedding-based representations alone. We present ARGUS, a source-grounded pipeline that combines a 5W1H-inspired schema, legal-domain models, and LLM-based structured generation to construct document-level Event Knowledge Graphs (EKGs) from CourtListener complaints. ARGUS extracts fact-bearing statements, builds chunk-level event graphs with participant, temporal, and causal structure, and merges them into document-level representations. We evaluate graph quality through human and multi-model assessment and test downstream utility on claim classification and legal QA. The graph-structured classifier outperforms raw and linearized baselines on the held-out set, and EKG-only retrieval improves document-scoped QA, while open-retrieval gains remain limited by low first-stage candidate recall. These results suggest that EKGs are most useful for organizing and reasoning over evidence once relevant material has been retrieved.

Modern infrastructure asset management constitutes a complex sequential decision-making problem, characterized by long planning horizons and system-level interactions, such as spatial deterioration correlations and economies of scale. While deep reinforcement learning has shown promise in optimizing maintenance policies, scaling to real-world networks remains challenging. Centralized approaches become computationally intractable in large-scale systems, whereas decentralized approaches often fail to capture essential coordination mechanisms. To address these challenges, we propose a graph-based framework that integrates accurate environment modeling with scalable decision support. First, we employ a hierarchical Bayesian model leveraging a Gaussian Process on Graph kernel to infer a realistic, spatially correlated networked environment of railway maintenance planning from real-world data provided by the Swiss Federal Railways. Second, we introduce a topology-aware Multi-Agent Reinforcement Learning (MARL) framework by integrating graph neural networks and graph Transformers to optimize network-level policies. A central contribution of this work is the demonstration of scalability through zero-shot transfer learning: graph-based agents, trained only on small network portions, are successfully deployed in a zero-shot manner on large-scale unseen networks without any retraining. Numerical results indicate that the proposed method significantly outperforms optimized heuristics and standard MARL baselines, reducing computational training time while maintaining superior performance on large-scale networks.

Large Language Models (LLMs) typically exhibit a performance profile where reliability degrades as task complexity increases. We address the challenge of generating high-quality natural language executable plans for complex tasks by introducing $\textbf{GRASP}$, a strategy-aware, multi-stage planning framework. GRASP decouples the planning pipeline across specialized, context-isolated modules: it pre-compiles global macro-guidelines (GenPlan), explores alternative localized strategies within isolated context windows (RevPlan), and independently evaluates trajectories using a multi-criteria discriminator (VerPlan). Empirical evaluations show that GRASP consistently establishes a new state-of-the-art frontier across diverse datasets, yielding substantial accuracy gains over direct LLM planners on Natural Plan Calendar Scheduling ($\sim$12.4$\%$$\uparrow$), ZebraLogic ($\sim$30.8$\%$$\uparrow$), and SciBench Math. Crucially, under multi-task scaling-where standard planners suffer immediate performance collapse-GRASP completely flattens the multi-task degradation penalty. In interleaved dual-task environments, GRASP achieves an absolute accuracy gain of up to 16.7$\%$ over direct LLM planners. Furthermore, by isolating context and enforcing strict macro-regularization, GRASP outperforms frontier reasoning models (such as GPT-5-mini) by a margin of 14.5$\%$.

Most benchmarks hand the model a question. EnigmaForge hands it a stack of old documents and no question at all. Buried in the letters, receipts, and logbook margins is a small logic puzzle whose solution is unique - proved by a SAT solver at generation time, with an ablation certificate showing every clue is load-bearing. Because instances are generated rather than collected, the corpus renews forever. The headline measure is intuition: task success when handed only the story, with world reconstruction as the secondary axis. Twenty-five frontier models ran over 600 instances (17,400 scored records) under three matched conditions. Intuition reshuffles the leaderboard: a 22x spread where fact recovery spans 1.6x, the second-best fact-recoverer ranks fourteenth, one model is indifferent to being told the question, and another is significantly better without it. Several models were blocked by their own content filters before reaching the puzzle - any benchmark scoring refusals as failure is quietly measuring filter behavior.

In this paper, we study the joint selection of an environmental support contact and a whole-body configuration for a prescribed loco-manipulation task. A contact may provide greater physical support while restricting the motion required for the task. We formulate this problem through three capability measures: residual wrench, end-effector reach, and base mobility available after satisfying the task requirements, and we balance them against contact acquisition cost. Evaluating these capabilities for every candidate requires repeated whole-body optimizations. To reduce this computational cost, we propose Capability-Tradeoff Contact Selection (CTCS). CTCS screens candidates for contact and task feasibility, groups similar candidates within each surface, and predicts their capabilities from exact anchor evaluations using local sensitivity analysis. It checks these predictions through selective exact evaluations, ranks candidates by capability, and evaluates a shortlist exactly for final selection. We evaluate CTCS in simulations and hardware experiments using a Unitree Go2 quadruped with an AgileX NERO arm across $392$ task conditions with nine available support surfaces. Results show that CTCS outperforms ground-only and fixed-contact support, as it can select support surfaces that provide favorable capability trade-offs for the task. Compared with evaluating every candidate exactly, CTCS achieves approximately $3\times$ speedup while closely matching the resulting mean objective value.

Customer experience (CX) agents use tools and large language models to address customer requests and guide conversational interactions with an organization's products. Improving these agents, especially in regulated industries, is difficult: they must detect intent, follow complex operational policies and use tools reliably. Manual end-to-end testing offers limited coverage, while live experiments expose customers to failures that can erode trust. We present a hypothesis-driven simulation workflow for screening candidate CX agents before deployment. Synthetic customers react to agent responses and simulated tool outputs enable multi-step agentic workflows without invoking production backends. We use the Snowglobe simulator on Nubank's Card Delivery agent and its expanded successor, Card Management - Nubank's highest-volume chat-support agent in Brazil. Across 4 deployed versions, simulated and production version-level binary evaluator scores show high correlation. Simulation-guided iteration increased transactional net promoter score (tNPS) by 36.69 points in a live A/B test. We also screened open-weight configurations in over 16,000 simulated conversations. In a subsequent live A/B test, the selected model increased self-service rate (SSR) by 8.82 percentage points to the highest level observed at Nubank, with no statistically significant change in tNPS. Simulation made broad exploration of models, reasoning settings, and prompts feasible without customer exposure, enabling production improvements that would have been impractical to pursue through live experimentation alone.

Neural behavior cloning compresses demonstrations into large models, making individual actions difficult to trace and policy updates costly. Retrieval policies retain access to demonstrations but struggle with mismatch between recorded and live behavior. We introduce Behavior Predictive Control (BPC), which synthesizes policies without end-to-end policy training by combining an action-aware retrieval metric, a Hankel-based action-continuation prior, and a closed-form one-step residual correction. Inspired by behavioral systems theory, BPC predicts future actions by blending stored observation-action data that best reconstructs the recent runtime observation--action history. Across simulated benchmarks and real-robot deployments, BPC is competitive with learned policies such as $π_{0.5}$ (surpassing it in some cases), while reducing policy fitting from hours to seconds on consumer GPUs and supporting closed-loop control upwards of 75 Hz on a Jetson Orin Nano. The retrieved demonstration windows and their coefficients also provide an intrinsic estimate of task progress. Retaining demonstrations within the deployed policy makes its predictions traceable to supporting trajectories and enables behavior revision through the demonstration bank.

Current vision-language models (VLMs) excel at visual content understanding and text-based reasoning, yet their structure limits the advancement of incorporating images into the reasoning chain. Though Omnimodal models have made efforts in unifying text and image generation, they focus on visual tasks in the open-domain, lacking tractability due to rasterized or latent representations of images. We introduce SVGLM, a framework that uses scalable vector graphics (SVG) primitives to connect text and image in reasoning tasks. We exploit the duality of SVG as both image description and text instructions, yielding a more compact, interpretable solution to equip general VLMs with the capability of generating images within the reasoning process. We provide a large curated dataset of SVG-based image editing dataset, as well as the paradigm to tune open-source VLMs. Experiments on a mathematical reasoning benchmark demonstrate that SVGLM achieves strong SVG generation power as well as think-with-image intelligence. Our results highlight SVG as a suitable medium for building more robust digital domain agents, bridging the gap between text-based thinking and pixel-based images.

Visuomotor policies learn a direct map from raw sensory observations to robot action sequences. Policies based on Diffusion and Flow Matching capture the multimodal distribution over action sequences in an end-to-end manner. This expressivity comes at the cost of multi-step numerical integration of the learned vector field for action generation, which can be expensive and time-consuming, impeding fast control rates required in robotics applications. Furthermore, robot action sequences are usually defined on a smooth, differentiable manifold, requiring that the learned policy respects the intrinsic geometry of the robot's action space. Here, we present Riemannian MeanFlow Policy (RMFP), which learns the conditioned flow map of the probability path on the robot action manifold. Our formulation employs a flow map consistency objective grounded in the data by a Riemannian Conditional Flow Matching anchor. The flow map consistency condition is stable to train and constrains the learned model to finite-time transport, which yields on-manifold action sequence generation with as few as one network function evaluation. We present results on the spherical LASA and Push-T benchmarks, on the Tool Hang and Transport tasks of the Robomimic suite, and on the Franka Kitchen task with manifold-constrained action generation, and demonstrate that RMFP attains performance competitive with prior work at a lower sampling cost. We also employ RMFP on a real-world robotic manipulation task to demonstrate fast action generation under imperfect sensor measurements in the physical world.

Agent skills provide reusable knowledge and instructions, yet agents must repeatedly infer how to apply them and which operation should follow. This couples task reasoning with control decisions, allowing prescribed steps to be omitted or applied incorrectly. We introduce HEXIS, which compiles agent skills into extended finite state machines that separate knowledge from control flow. Skill knowledge is incorporated into local instructions that guide reasoning and generation within states. The machine records execution progress and intermediate results, while explicit transition conditions determine subsequent operations. Our incremental compiler first maps skill clauses and tool interfaces to state operations, local instructions, data bindings, and transitions. It then aligns development traces with existing states to identify missing operations and dependencies. These are incorporated by adding or reusing states and refining their connections. Updates are accepted only after static checks and replay of the current and all previously accepted traces. Across four benchmarks and four executors, HEXIS improves success over Skill + ReAct by 16.1 percentage points on average. Qwen3.8-27B reduces execution tokens by 38.4-88.9% across benchmarks.

Audio Description (AD) makes movies accessible to blind and visually impaired audiences by narrating visual information in gaps between dialogue. Existing automatic AD systems largely treat generation as a local video-to-text problem, assuming that the content to describe and its temporal location are already provided. Realistic AD instead requires coupled decisions about what visual information is narratively important, when it can be spoken without interfering with dialogue, and how it should be formulated to fit within the available time. We formalize AD generation as a constrained optimization problem over these three decisions. Our hybrid system uses large language models to propose and ground visual elements, estimate their salience to the narrative, and generate compressed realizations. A mixed-integer linear program then jointly selects and schedules descriptions across a scene subject to temporal constraints. When evaluated on REFRAMED, a benchmark for realistic AD of movies, our approach makes better decisions than prompted LLMs about what to describe and when to describe it, establishing a new SOTA on narrative QA and temporally grounded metrics. Ablations show that explicit temporal constraints drive gains in placement, while salience estimation controls how much narratively useful content is retained. Improvements are concentrated on temporal and narrative measures rather than n-gram overlap, although a significant gap to professional describers remains.

Modern deep neural networks treat parameters as static floating-point matrices stored in physical memory, incurring Von Neumann memory bottlenecks and representation collapse. We formulate Orbital Error Dynamics (OED), an analytical framework wherein synaptic weights are not stored masses (O(W)), but transient topological resonances (O(1)) derived procedurally from the complex quadratic polynomial map z_{n+1} = z_n^2 + c. We introduce the Bent Sine Wave Hypothesis, demonstrating that non-equilibrium living systems emerge when harmonic waves curl inward through environmental drag toward the cardioid cusp (c = 1/4). We define the Observer Horizon Geometry in parameter space, identifying interior resonance shoulder loci X_upper = (0.25, +0.18) and X_lower = (0.25, -0.18) between the fixed-point basin and the true boundary at c = 0.25 +/- 0.50i. To escape non-convex stagnation without loss zeroing, we introduce a heavy-tailed Biomimetic Perturbed Jump Operator (Omega_tunneling) inspired by mammalian fertilization zinc sparks. We further couple an enteric-cranial Dual-Brain architecture shielded by adaptive CD4+ regulatory immune gating (M_CD4), and project the 4-nucleotide genetic basis (A, T, C, G) across quadrants in C. Multi-seed empirical validation on the Two-Moons manifold (5 seeds, 80/20 train/test split, 32x32 grid, zero test-time updates, zero label leakage) demonstrates that procedural parameterization from a 24-byte coordinate seed achieves 77.67% +/- 5.35% clean test accuracy (within an 8.00-point paired difference of an unconstrained gradient baseline at 85.67% +/- 5.35%, 95% CI: [-1.07%, 17.07%]) and 71.33% +/- 3.80% under distribution shift (N(1.2, 0.4)), alongside conceptual equivalence with an analog optical co-processor.

Large language models increasingly operate as persistent assistants in user-facing, shared-session, and tool-augmented settings. When users disclose sensitive information during an active conversation, that information may remain behaviorally recoverable through later prompts even after the dialogue shifts to unrelated topics. We introduce PrivDrift, a benchmark for auditing whether user-disclosed secrets remain recoverable after conversational topic drift and persuasion-based probing. PrivDrift contains 1,000 controlled multi-turn dialogues with seeded secrets, content-dense drift turns, and standardized extraction probes. Across three LLMs with extended context windows, dialogue-level hybrid leakage remains substantial, ranging from 38.7% to 54.6%, and varies strongly by model, secret type, and persuasion intensity. Within the tested drift window, additional topic drift does not reliably reduce leakage, suggesting that privacy risk in active LLM contexts should be evaluated as a persistent behavioral failure mode rather than only as training-data memorization or immediate jailbreak behavior.

While Vision-Language-Action (VLA) models demonstrate impressive capabilities in robotic manipulation, their memoryless nature renders them brittle to test-time environment shifts, particularly hardware shifts caused by wear or imperfect calibration. Enabling these models to self-adapt during deployment without requiring continuous on-site recalibration remains a critical bottleneck for real-world scalability. In this work, we introduce Self-Adaptive VLA, a novel post-training recipe that enables the policy to iteratively adapt to deployment-time hardware shifts leveraging its own rollouts as context. To do so, we first collect policy rollouts under deliberately injected hardware shifts. We then transform the base policy's training data into shift-conditioned expert demonstrations by pre-compensating the expert actions for these known shifts. Next, we introduce a lightweight, plug-in context encoder that compresses the context, including visual observation, proprioception, and actions in the shifted environment, into a latent context token. This token modulates the policy through adaptive layer normalization (AdaLN). Furthermore, we find that context tokens can be ensembled, allowing the policy to iteratively self-correct and mitigate failures step by step. Extensive experiments across four precision-critical bi-manual and dexterous manipulation tasks show that Self-Adaptive VLA recovers over 80% of the base policy's performance under hardware shifts, such as actuation bias and joint encoder offsets. Moreover, Self-Adaptive VLA enables more robust deployment to new workstations compared to the base policy. Our approach provides a pathway for robust large-scale real-world robot deployments and easier maintenance. See videos at https://icefoxzhx.github.io/self-adaptive-vla.

We consider the decision of whether to return an existing draft answer or revise it using retrieved evidence, as in answer-revision systems. Draft confidence estimates whether the current answer is correct, but the decision requires estimating the effect of a specified revision. For offline training and evaluation, we grade both the returned draft and its candidate revision under the same correctness judge, which makes repair, harm, and the gap to an oracle observable. We call this paired effect its recoverability, and we train policies to predict it before revision. On 25,870 held-out open-domain questions across three revision setups, a scorer trained on the paired outcome has greater area under the accuracy--revision-rate curve than a matched draft-correctness scorer in all nine Llama setup--seed fits, and gains 0.23--0.68 accuracy points on average at development-selected thresholds, a difference significant across training runs only for dense retrieval. The resulting policy improves on always revising and on average closes more than a third of the oracle gap, although it still applies 38--46% of the harmful revisions. When a draft-free standard-RAG answer is also available, however, choosing between the draft and that answer is stronger by about two points for Llama and four for OLMo, and adding candidate revision as a third option yields no significant gain. Recoverability describes one revision; its value as an available action also depends on the alternatives.

In multi-target regression, correlated targets are often coupled through multi-output Gaussian processes with an intrinsic model of coregionalisation (GP-ICM), assuming that sharing statistical strength improves overall performance. In practice, the benefits are inconsistent. Across the settings studied, we find that the main benefit of coregionalisation is joint uncertainty quantification rather than point prediction. Raw target correlation does not predict when coupling helps; in the separable GP-ICM settings studied here, residual correlation, the cross-target dependence left unexplained by independent per-target predictors, is the strongest predictor of joint-uncertainty gains. We introduce a lightweight diagnostic, $D_{\rm logdet}=-\frac{1}{2}\log\det R_{\rm res}$, which represents the idealised joint negative log-likelihood (NLL) gain from modelling a full rather than diagonal residual covariance and is computable from independent GPs alone. Across a controlled synthetic study, 16 multi-target benchmarks, and frozen transformer and convolutional neural network representations for keypoint regression, point prediction remains largely unchanged ($ΔR^2\approx 0$). In contrast, $D_{\rm logdet}$ strongly predicts observed ICM NLL improvements ($ρ_s=-0.83$, $p<0.001$), outperforming heuristics such as the feature-to-sample ratio. We also propose Residual-ICM, which preserves independent marginal variances while adding residual-correlation structure to the joint covariance. Residual-ICM achieves the best average joint NLL among the compared methods, while the diagnostic indicates when covariance coupling is likely to be useful. The diagnostic is specific to global Gaussian residual dependence, the structure captured by separable coregionalisation.

Robust dexterous grasping requires maintaining physical stability despite contacts interactively evolving across the entire hand. A precomputed force distribution can easily fail under object motion, modeling errors, or external disturbances. In this paper, we present a framework for real-time force regulation over dynamically changing whole-hand contacts. Our method geometrically estimates contacts across all hand links using a tracked object model and proprioception, without requiring tactile sensing at those contacts. It repeatedly recomputes the desired contact-force distribution subject to friction constraints, actuator limits, and an actuation-consistency constraint motivated by classical whole-limb force analysis. We integrate this force-regulation controller with reactive reaching, enabling the hand to acquire a grasp, maintain it under disturbances, and regrasp after losing the object. Simulation experiments without gravity demonstrate improved grasp retention over fixed-allocation and fingertip-only execution under controlled perturbations, while real-world experiments on a 27-DoF arm-hand system demonstrate grasp maintenance and recovery under human-applied disturbances as contacts evolve across the whole hand. Project page: https://sangminkim-99.github.io/reactive-grasp-whole-hand/

Graph neural networks (GNNs) have become a prominent approach for developing fast, topology-aware surrogates in electric power systems, supporting tasks such as power flow (PF) analysis, optimal power flow (OPF) estimation, and cascading failure analysis (CFA). Despite this growing use, formally verifying GNN-based models remains challenging, with existing methods limited in scope. We extend the neural network verification (NNV) framework to graph-structured inputs through GraphStar sets, a generalization of Star sets that captures uncertainty over both node and edge features. This extension enables the propagation of linear message-passing operations and the sound approximation of ReLU nonlinearities for GNN architectures, including graph convolutional network (GCN) and graph isomorphism network with edge features (GINE) layers. We evaluate GNNV across three power system tasks, PF, OPF, and CFA, on the IEEE-24, IEEE-39, and IEEE-118 test cases, as well as two standard graph classification benchmarks, ENZYMES and PROTEINS. Our results show that GNNV provides tighter robustness guarantees than CORA on graph classification models with ReLU-based activations and, for the first time, delivers edge-aware robustness guarantees for GINE-based PF and OPF models under joint node and edge perturbations.

The English Resource Grammar (ERG) is a hand-written computational grammar of English. Given a sentence, its processor, ACE, produces a formal meaning representation called Minimal Recursion Semantics (MRS): a graph of the sentence's predicates and their arguments. The grammar is bidirectional and can also turn an MRS back into an English sentence. \citet{hajdik2019} used the ERG's treebank to build a benchmark for that generation task, MRS to text, and trained sequence-to-sequence models to solve it. The parsing task, text to MRS, can be tested on the same sentences. We reconstruct their 10K-sentence test split, and score two large language models, Claude Sonnet~4.5 and Claude Opus~5, in both directions against their trained systems and against ACE, with no task-specific training. Given an MRS and three examples, Opus writes the sentence at 76.3 BLEU, ten points above their system trained on 72k pairs (66.1 BLEU), and comparable to their system trained on a million extra pairs (77.2 BLEU). Sonnet scores 65.7 BLEU, and letting it choose among ACE's own candidate sentences lifts it to 69.6, while a pooled judge that keeps Opus's own sentence among the candidates adds 0.6 points (77.0 BLEU). In the parsing direction, however, the models fall far behind ACE: asked for the MRS of the same sentences, they reach 57.2 (Sonnet) and 65.5 (Opus) F$_1$ on the graph's predicates and arguments against 91.0 for ACE, and exact-match the gold on about 1\% of sentences. We characterize the failure modes for the parsing tasks, and conclude that a generation score alone does not show that models understand formal semantic representations.

Advances in language modeling have been driven by scaling pretraining on ever more data. Yet, the training data is still largely curated on the model's behalf. A more general approach to pretraining would let the model learn to generate the data most useful for its own improvement. This would provide an effectively unbounded source of training data, limited by compute rather than human knowledge. We introduce Self-Play Pretraining with Zero Data, an initial proof-of-concept towards realizing this vision. Our procedure casts synthetic data generation as a search over the space of all computable structure, taking inspiration from Solomonoff induction. Starting from random initialization, two models learn in tandem: a generator proposes programs interpreted by a universal Turing machine, generating byte sequences, while a learner autoregressively predicts these byte sequences. The learner is trained with standard cross-entropy, while the generator is trained with reinforcement learning to produce sequences at the frontier of the learner's capabilities, yielding an adaptive curriculum. A universal Turing machine gives us a search space over all computable data-generating processes, imposing little domain-specific structure, and self-play searches over this space for useful training data. We test whether zero-shot performance on natural data improves predictably with self-play compute; this is a clean test of transfer since neither generator nor learner is trained on natural data. Across several natural datasets, zero-shot loss exhibits predictable scaling in compute. The models also exhibit in-context learning, and discover recognizable mathematical sequences during training.

Deep learning inference and training performance depends critically on GPU kernel efficiency. Modern compilers such as PyTorch Inductor automatically generate GPU kernels from high-level model code, but frequently underperform expert-written implementations by wide margins. Recent LLM-assisted kernel optimizers can close this gap for standalone kernels, yet treat compiled models as black boxes, generally optimizing individual standalone kernels without respecting the compiler's structural decisions or verifying the model end-to-end. We present KernelOPT, a multi-agent system that treats compiled models as structured artifacts. It preserves vendor library calls (cuBLAS, cuDNN) and exclusively targets generated Triton sub-kernels using five profiling-guided LLM agents. A four-gate verification cascade of static validation, multi-seed correctness, model-level float64-fallback verification, and performance gating filters candidates during optimization and verifies the re-stitched model end-to-end. If no candidate passes all four gates, the system preserves the compiler baseline. The system accepts PyTorch nn.Modules, standalone Triton kernels, and Helion kernels. Evaluated on 250 KernelBench problems, KernelOPT achieves geometric mean speedups over \texttt{torch.compile} of 1.40$\times$ (Level 1: 51/100), 1.15$\times$ (Level 2: 31/100), and 1.07$\times$ (Level 3: 12/50) across all problems.

In the Era by Eon benchmark, each question states the rules for its answer, and code computes the answer from a generated company's data. When agents can run code, the four strongest models each answer 22 to 25 of 27 such questions, so the benchmark barely separates them. We add eight question templates that depend on hidden facts. No question or document states a hidden fact, and the records that seem to hold it show something else. Other data implies it. For example, the sales system says a customer dropped a purchase because of timing. On a recorded call, the customer blames an outage. For each generated company, code fills each template and computes an exact answer without a language model. We evaluate 12 agents. Each pairs a model with an agent program, which connects it to the company's systems. The best agent answers 18 of its 24 attempts, three per question, correctly. Four of the six models answer at most 6 of 24 with any program. The hardest questions require picking one of several similar records, such as which of three renewal offers a customer signed. All agents together answered two such questions correctly in only 1 of 84 attempts.

Improving the scientific coding capabilities of large language models (LLMs) requires high-quality training data. However, such data remain scarce because manually authoring realistic problems is costly and time-consuming, while systematically covering diverse scientific domains and algorithmic combinations remains challenging. To address this, we introduce SciWalker, a framework for synthesizing scientific coding problems through operator-chain sampling and execution feedback. The framework combines scientific library interfaces with operation modes to instantiate operators, organizes them into operator graphs, and samples operator chains as computational workflow cues. Guided by these cues, we adopt LLMs to generate scientifically grounded problem statements, reference solutions, and tests, with failed generations iteratively repaired using execution feedback. By combining structured workflow composition with verification and quality review, SciWalker enables scalable task generation while promoting scientific grounding, computational diversity, and executability. Using this framework, we construct 8,178 high-quality problems spanning 5 scientific domains and 32 subdomains. To evaluate their training utility, we conduct reinforcement learning on Qwen3.5-9B using the GSPO algorithm. This training improves SciCode subproblem accuracy by 9.9 percentage points, from 29.3% to 39.2%, with gains across scientific code generation, code repair, and reasoning benchmarks. The code for SciWalker is available at https://github.com/lichenx1/SciWalker.

We present NNV3, the latest version of the Neural Network Verification (NNV) tool, a MATLAB framework for formal verification of deep learning models and learning-enabled cyber-physical systems. Building on the set-based reachability foundation of NNV 1.0 (FFNNs, CNNs, NNCS) and NNV 2.0 (RNNs, SSNNs, neural ODEs), NNV3 introduces new members of the Star-set family: ModelStar for verifying networks under weight perturbation, VolumeStar for video and 3D volumetric inputs, and GraphStar for graph neural networks. A conformal-inference-based probabilistic reachability mode complements sound analysis for problems where deterministic verification is intractable, while FairNNV certifies counterfactual and individual fairness properties over continuous input regions. NNV3 introduces new benchmarks for malware detection, graph-based power-system models, medical imaging, variable-length time series data, and action recognition. NNV3 also incorporates tutorials and developer guides through a unified documentation site. This paper details these major updates, demonstrating NNV's maturation into a comprehensive, robust, and accessible verification tool for a diverse range of AI systems.

If a language model can recognize code it wrote, it may favor that code as a judge, and instances of one model monitoring each other could collude. We test this zero-shot on current commercial models. Five LLMs generate solutions to MBPP, HumanEval, and DS-1000, seven more to MBPP, and models act as evaluators in four tasks: picking their own solution from a pair, judging whether a single solution is their own, identifying which of two solutions a named model wrote, and judging quality blind. In the single-solution task, balanced accuracy is 49-58% for all 15 model-benchmark combinations, while raw accuracy (38-67%) mostly reflects how readily a model claims authorship. In the pairwise task, accuracy across 14 evaluator-opponent combinations correlates at r=0.93 with how often the evaluator's solution is longer. Attribution to a named model succeeds on some pairs and is consistently inverted on others. A rule-based normalization that strips docstrings, comments, type hints, and local names preserves Pass@1 and leaves ten of twelve re-tested results at chance; the other two follow a length difference it leaves, although a trained classifier still separates most normalized pairs. Claude Haiku's self-preference also disappears. We recommend reporting balanced accuracy, heuristic baselines, and label consistency.

Gradient-based photonic design needs full-wave derivatives with respect to millions of parameters, but on a single GPU workstation the time-domain adjoint is limited by device memory and by the incompatibility of fused update kernels with automatic differentiation. Here, we present TorchFDTD, an open-source finite-difference time-domain (FDTD) package that addresses both limits. Its Yee, absorber and dispersion updates execute as fused CUDA kernels captured in a CUDA graph, and every kernel is paired with a transpose kernel derived from its update, so material and geometry derivatives are obtained by a discrete adjoint that PyTorch chains with differentiable objectives built from the supported observations. The adjoint recovers forward states by checkpoint replay or, for lossless periodic problems, by time reversal. For problems that exceed the device, a streamed mode advances the domain one causal slab at a time and keeps the global state and the checkpoints in host memory, which lowers the device allocation while preserving the resident discretization. We validate the package against analytic solutions, against Meep, FDTDX and a rigorous coupled-wave solver, and against automatic differentiation and finite differences. On an A100 the fused path completes full forward solves of eight test scenes 9.0 to 17.1 times faster than the PyTorch FDTD package it extends, and its double-precision solves are 49 to 59 times faster than those of Meep on a workstation CPU. On an RTX 3060, host streaming of $256^3$ and $320^3$ adjoints costs 3.1 and 2.7 times the resident time and lowers the peak device allocation by 56% and 65%. A 54-million-cell pillar-array lens coupled to an angular-spectrum objective yields an adjoint derivative within 0.78% of a central difference. The time-domain adjoint of a device with tens of millions of cells thus becomes available on a single workstation GPU.

Deployment-oriented compression is attractive for resource-constrained brain--computer interfaces (BCIs), but whether it changes adversarial vulnerability remains unclear. On BCI Competition IV-2a, we compare 32-bit floating-point (FP32) EEGNet and ShallowConvNet models with global magnitude pruning and simulated INT8 post training quantization (PTQ) and quantization-aware training (QAT) across nine subjects and three seeds. Simulation provides differentiable quantize--dequantize models for white-box attacks and gradient analysis, while native TensorRT deployment is used for validation. Accuracy-preserving compression does not improve direct robustness: at $ε=0.005$, EEGNet PGD accuracy remains 22--24\% across FP32, 50\% pruning (P50), PTQ, and QAT. However, P50 reduces bidirectional transfer efficiency to 0.963/0.928 (FP32$\rightarrow$P50/P50$\rightarrow$FP32), versus 0.994/0.997 for PTQ; the same trend holds for ShallowConvNet. Gradient alignment shows a corresponding separation, while native PTQ agrees with simulated clean/adversarial predictions in 95--98\% of cases. These results show that direct robustness, adversarial transfer, and deployment efficiency are distinct properties of compressed EEG decoders.

Planners built on visual world models commonly score each predicted outcome by its distance to the encoded goal image. We show that this target can limit control even with exact dynamics and globally optimal short-horizon search: reaching a goal may require actions that initially move away from it. With frozen LeWM models, intermediate targets substantially improve action synthesis and recorded-action ranking on Cube, PushT, Reacher, and TwoRoom. Learned targets and targets drawn from observed experience both produce these gains. We introduce Anchored Planning, which retrieves a recorded segment whose start and end resemble the current and goal observations, then aims at an observation shortly after its start. The frozen model scores actions toward this target from the current state. Without additional training, planning toward observed targets outperforms the LeWM planner on every task in our long-range evaluation. Additional final-goal search falls short of the same gains. Lower successor-prediction error need not translate into better control. Success also depends on how far ahead the target is placed and on shrinking the retrieval span as execution advances. Changing only the target lets the same frozen model and planner reach goals that final-goal scoring misses.

A synthetic survey can reproduce the average answer while misrepresenting how people differ, how their answers relate to one another, or how they respond to changes in conditions. We introduce the Artificial Societies Benchmark to help researchers assess whether synthetic populations support their intended analyses. The framework combines eleven tests across internal, construct, and external validity, drawing on twenty human sources and comparing nine language models. It connects each research use to the evidence it requires and tests how results change with the information we supply about respondents. Importantly, strong performance in one domain does not establish fidelity in the others. Models often answer too consistently, compress response scales, and alter relationships between traits whilst richer profiles improve prediction for some models and worsen it for others. The resulting scorecard helps researchers identify which aspects of a synthetic population can support their analysis and where researchers need further human evidence.

Multi-agent deliberation can improve performance, but what happens when some agents do not act in good faith? In practice, an agent may be deceptive and work to subvert the group, whether through its own objectives or external instruction. We study how susceptibility to deception scales as groups increase in size and deceivers become more prevalent. It is not the number of agents in the group that matters, but the proportion of deceivers. We observe that the defection rate, how often initially correct agents switch to an incorrect final answer, rises linearly with this proportion. Whereas humans in comparable conformity studies are reliably swayed only when misleading confederates form a majority, LLM agents defect regularly even when deceivers remain a minority. Susceptibility also depends on which models are interacting, especially on the honest agent side. Unexpectedly, allowing deceivers to coordinate privately can make them less effective. Altogether, our results show that adding more agents is therefore not a sufficient defense, because the adversary can simply scale with the group.

Frontier language models are rarely used in clinical workflows because the realistic, longitudinal benchmarks needed to develop them are scarce. Real electronic health record (EHR) data cannot be openly shared due to privacy, ethics or data use issues and it does not contain verifiable ground truth since the chart records only reflect what clinicians documented. We introduce Synthetic Hospital, an open, fully synthetic, fact-grounded longitudinal EHR benchmark that resolves the open sharing and verifiable ground truth barriers. Built entirely from public medical-education material with no protected health information, it comprises 1,268 longitudinal patients and 5,602 encounters, where every diagnosis, finding, and temporal relation is grounded in standard ontologies (ICD-10-CM, SNOMED CT, LOINC) and with a complete provenance chain back to its source medical education material. Synthetic Hospital is served through a simulated hospital record system that mirrors real EHR infrastructure (standard interoperability APIs, role-based access and function-calling interface). In a blinded review, physicians distinguished its records from real patient charts at near-chance rates (53\%). Across 10 frontier and open models, none approaches ceiling: the best model reconstructs a patient's longitudinal problem list with a severity-weighted F1 of 0.73, level with the mean of seven physicians on a matched subset but well below the best of them (0.89), and misses roughly half of clinically relevant findings when summarizing a chart. Overall, these results highlight that Synthetic Hospital is a difficult and realistic test of clinical AI performance.

Manipulation requires not only reasoning about the external environment, but also about the robot's physical condition. A strategy may remain geometrically feasible while becoming physically unsuitable due to increased joint load or limited mobility, yet internal physical state is typically used only for low-level control. We propose body-grounded high-level replanning, which uses internal physical state to adapt manipulation strategies during execution. Body-state events trigger strategy replanning, and an LLM interprets the underlying joint-level state, recent execution statistics, and execution history to select a context-dependent alternative, while leaving the task objective and low-level controller unchanged. We evaluate the framework on a reaching task under controlled load and asymmetric mobility constraints in simulation and on a real robot. Our experiments show that body-grounded replanning maintains high task success while reducing physical effort and enabling more efficient strategy adaptation. Additional contact-rich manipulation experiments demonstrate the applicability of the same replanning interface beyond reaching. These results show that internal physical state can inform not only low-level control, but also high-level decisions about how a manipulation task should be performed.

Imitation learning enables robots to acquire manipulation skills from demonstrations, but the resulting policies can fail outside the training data, while collecting more demonstrations requires substantial human effort. Human-in-the-loop reinforcement learning uses corrective feedback during online training, but typically learns the complete task policy rather than refining a pretrained imitation policy. We introduce Res-HIL, a human-in-the-loop residual reinforcement learning framework that learns corrective actions on top of a frozen imitation policy. Each human intervention provides two complementary learning signals: direct supervision of the residual policy and reward shaping of preceding autonomous behavior. Res-HIL combines these signals with zero initialization of the residual policy to stabilize and accelerate online learning. We evaluate Res-HIL on five contact-rich manipulation tasks spanning high-precision and long-horizon behaviors. With only 20 initial demonstrations, Res-HIL outperforms state-of-the-art full-policy human-in-the-loop reinforcement learning and residual fine-tuning without human guidance on every task after ten minutes of online training. Res-HIL improves its pretrained base policies and outperforms imitation policies trained with five times more demonstrations. An ablation study shows that direct residual supervision is critical to performance, while intervention-aware reward shaping substantially improves training efficiency.

Language models advise people, keep them company, and write software while they sleep. Measuring what they do is hard: behavior has to be sampled repeatedly across models, prompts and releases, most of it lives in unstructured text that has to be coded before it can be counted, and the result has to be legible and rigorous enough to meaningfully compare models and vendors. To address these constraints, we present a simple, cheap, scalable, and replicable model for studying model behavior. Each study is a frozen, public stimulus run identically on a cross-vendor panel, at a few dollars per model or less. Each reads its transcripts one of three ways, chosen by how much interpretation the behavior needs: exact match on a clamped reply, a codebook applied by LLM judges whose agreement with a human coder is reported per code, and an instrumented environment that records what an agent did independently of what it said. Run across four years of model releases from both frontier and open-source labs, these instruments find four things. Convergence: asked to pick a word, 27 of 44 models answer serendipity at least once in four tries. Resistance: a trailing "right?" moves endorsement by up to 32 points, and the sign flips from sycophantic to resistant as generations advance, keyed to the tag's surface form. House: whether a model holds a position under pressure tracks its generation, and how it holds tracks the lab that built it. Account: told to do something the documentation in their repository contradicts, some coding agents never went along silently and others always did, and the same model can change with the harness it runs in. Re-run on every release, batteries like these track how behavior is changing across vendors and over time.

Financial institutions operate under dense, frequently amended rulebooks, and answering a compliance question correctly requires not only fluency but verifiable grounding in the authoritative text. Large language models are attractive for this task, yet the models that firms can realistically deploy on-premise are compact ones, and compact models hallucinate obligations. We study whether a carefully domain-adapted retrieval-augmented generation pipeline closes that gap. Our retriever is built in three stages on top of LegalBERT: entailment tuning that recasts question--passage matching as premise--hypothesis reconstruction, contrastive tuning with in-batch negatives, and score-level fusion with BM25. Our generator is a compact model (2B--12B parameters) served under 4-bit quantization, either prompted or adapted with retrieval-aware fine-tuning (RAFT) through LoRA. On ObliQA, a question-answering benchmark built from the Abu Dhabi Global Market rulebooks, the staged retriever raises Recall@10 from 0.256 to 0.774 and outperforms BM25 (0.678) and E5-large-v2 (0.758), the strongest general-purpose dense encoder we tested. RAFT-LoRA then improves the composite RePASs answer-quality score for every model we could adapt, with the largest gain on the weakest one. However, the adapted models do not transfer to Australian case-law questions, and a closed-book model that receives no passages at all scores within 0.011 RePASs of the full pipeline while producing answers that cite nothing and misstate obligations. The retrieval gain is therefore measured directly, the generation gain is a gain in RePASs rather than demonstrated grounding, and grounding itself requires an evaluation protocol that RePASs does not provide.

Sustaining industrial recommendation research requires using the results of one experiment to decide what to investigate next. We present AgentX-Model, the next generation of AgentX's model research framework, which connects proposal development and model experimentation within sandboxes defined by business inputs and prediction tasks. AgentX-Model adopts a dual-agent architecture comprising a Research Agent and a Model Agent. The Research Agent develops independently reviewed proposals from papers and experimental findings, while the Model Agent conducts multi-round investigations and returns code, measurements, and unresolved questions. Using the returned results, the Research Agent selects a starting implementation and formulates the next research question, allowing subsequent experiments to build on earlier findings. We organize this continuing research around four actions: Reproduce, Follow-up, Composition, and Diagnose. The first three actions drive routine research, while Diagnose acquires the evidence needed to choose a repair, including for issues raised by business feedback and online evaluation, such as prediction bias measured by PCOC. Across the production evaluation, 560 of 636 completed model-changing experiments recorded AUC above their business baselines. As research continued, some experiments recorded AUC above every comparable ancestor in their lineages. The five latest online A/B evaluations across different business settings reported gains including 10-15% in acquisition efficiency, 15-20% in target-segment advertising spend, and 0.3-0.8% in watch time; the watch-time model used approximately 10% fewer FLOPs and parameters. A dependency-aware historical-replay benchmark further evaluates research allocation, with initial results showing no consistent efficiency gain from more complex scheduling when agents already analyze and select concrete candidates.

Post-training quantization of vision--language models (VLMs) is typically assessed through aggregate task accuracy and memory savings, but preserving a headline score does not guarantee preservation of visual grounding behavior. We present GHOST-Q, a cross-precision controlled evaluation of three 8B VLM families under FP16, INT8, and NF4 across utility and hallucination-sensitive benchmarks. Rather than comparing only aggregate accuracy, we pair FP16 and quantized predictions item by-item to quantify how compression redistributes grounding successes and failures. Five of six quantized variants preserve MMStar accuracy within $\pm2$ percentage points, yet 10 of 36 paired effects remain significant after false-discovery-rate correction, nine on hallucination-sensitive conditions. Same-device A100 profiling further demonstrates that substantial memory reduction does not necessarily mean lower inference latency. Finally, an open-ended AMBER audit reveals strong generation budget censoring whose severity varies by architecture and precision. These results show that quantized VLMs should be evaluated jointly for aggregate utility, grounding reliability, generation behavior, and realized deployment efficiency.

Synthetic data can scale training supervision when real-world data are limited, but noise and distribution mismatch can reduce its value. Existing synthetic data selection methods often emphasize fidelity or diversity rather than the learner's evolving needs. We propose FROST, an online framework that estimates synthetic-data utility through gradient feedback anchored in real training data. It calibrates batch utility against recent history to determine when filtering is needed and filters samples only in out-of-band batches to determine what to retain, without an external verifier or held-out validation set. Experiments on two public benchmarks for image classification and LLM fine-tuning for text-to-SQL show that FROST filters out around 20--30% of the synthetic data while improving real-task performance compared with training on the full synthetic data pool. We further apply FROST during training in a large-scale industrial ads re-ranking system, achieving significant performance gains over a highly optimized production baseline, demonstrating its effectiveness and generalizability.

Underwater 3D reconstruction is critical for marine exploration, ecological monitoring, and subsea infrastructure inspection, yet remains challenging at large scale due to light attenuation, scattering, and limited capture coverage. While 3D Gaussian Splatting (3DGS) enables high-quality real-time rendering, its application to large underwater scenes is constrained by high memory consumption and inefficient optimization over extensive areas. We propose OceanXL, a fast and scalable 3DGS-based framework for large-scale underwater reconstruction. OceanXL adopts a divide-and-conquer strategy, partitioning scenes into spatially coherent blocks to enable efficient optimization while preserving global geometric consistency. We further introduce an adaptive pruning scheme tailored to underwater conditions that removes redundant primitives, producing compact representations without sacrificing visual fidelity. Together, these components improve training efficiency and rendering performance for large scenes. We also introduce a large-scale underwater dataset covering diverse marine environments. Experiments on five large-scale scenes demonstrate favorable scalability, compactness, and efficiency--quality trade-offs over large-scene baselines. Controlled comparisons on the small-scale SeaThru-NeRF dataset further show competitive reconstruction quality with substantially smaller model sizes than underwater-specific methods.

Self-consistency has become a popular technique for enhancing the reasoning abilities of large language models by generating multiple reasoning paths and selecting the final answer through a majority vote. However, because model providers typically charge users in proportion to the number of reasoning paths generated, they have a financial incentive to artificially increase the path count. In this work, we show that an unfaithful provider can exploit this incentive using a simple, efficient algorithm while avoiding detection by an auditor: by generating and strategically reordering additional reasoning paths, the algorithm makes every path appear necessary to reach the majority. To validate our algorithm, we conduct experiments with multiple instruct models from the Llama and Qwen families, as well as reasoning models distilled from DeepSeek-R1, on benchmark datasets spanning mathematics, science, and question answering. Our results suggest that the distribution of additional reasoning paths generated by our algorithm is heavy-tailed and that substantial capacity to overcharge remains even under the best possible audit designed to keep the false-positive rate below $α= 0.1$.

Estimating heterogeneous treatment effects from observational data is difficult because the most appropriate inductive bias varies with overlap, treatment imbalance, prognostic structure, and sample size. We introduce the Geometry-Diverse Anchor-Correction Expert Ensemble (GeoACE), a five-expert framework that combines a common anchor-correction estimator with complementary overlap-aware and outcome-guided geometries. Its task-level ensemble weights are learned only from internal validation predictions, frozen before test evaluation, and then applied to experts refitted on the complete development sample. The fifth expert, O-Phi-ACE, constructs an outcome-free, overlap-aware statistical projection from covariates and treatment assignment and replaces the anchor input with this lower-dimensional geometry. We evaluate GeoACE against 11 comparators on eight benchmark protocols. Adding O-Phi-ACE reduced mean sqrt(PEHE) relative to the four-expert ensemble on all seven benchmarks with individual-effect truth, winning 998 of 1,225 paired tasks; the change on JOBS policy risk was negligible. The five-expert ensemble ranked first on IHDP100, IHDPA, and IHDPB and second on NEWS, differing from the NEWS leader by 0.13%. Across the seven sqrt(PEHE) benchmarks it obtained the lowest observed average rank (3.714), although the omnibus Friedman and Iman-Davenport tests were not significant (p=0.328 and p=0.330). Using the same five frozen experts, inverse-DR weighting was consistently better than winner-take-all selection, convex DR fitting, R-stacking, and causal Q-aggregation in benchmark-balanced analyses, but was statistically indistinguishable from equal weighting and DR ridge shrinkage. The evidence therefore supports geometry-diverse expert libraries and leakage-free aggregation as a robustness strategy, not universal superiority of either GeoACE or one weighting rule.

Generative recommendation reformulates item retrieval as sequence generation, allowing a unified model to directly generate the next item from a user's interaction history. Semantic IDs further make this paradigm effective and scalable by representing each item as discrete codes, enabling knowledge sharing among semantically related items. Recent studies introduce explicit reasoning before Semantic-ID generation, helping models summarize user interests and infer possible preference transitions. However, reasoning is not inherently beneficial: Inaccurate or uninformative reasoning may mislead subsequent item generation and ultimately degrade recommendation performance. This raises a central challenge: how can a recommender select and learn effective reasoning traces and progressively evolve toward better reasoning from its own generations? In this work, we propose Evo-Rec, a three-stage framework for learning better reasoning and further enhancing it through reinforcement learning. First, we align Semantic IDs with their textual and behavioral contexts, enabling the model to understand and generate item identifiers. Second, we sample multiple candidate reasoning traces and retain those that improve the prediction of the ground-truth item, providing a stronger reasoning initialization through supervised fine-tuning. Third, we further optimize the reasoning policy through reinforcement learning with catalog-constrained item generation and ranking-aware recommendation feedback. Experiments on three Amazon Review benchmarks show that Evo-Rec consistently outperforms discriminative, generative, and reasoning-enhanced recommenders across all evaluation metrics. These results demonstrate the effectiveness of our framework in learning better reasoning for SID-based generative recommendation.

General-purpose vision-language models (VLMs) bring broad knowledge and spatial reasoning to robot manipulation, yet existing systems either use them indirectly, to predict constraints or write programs, or give them a view of the scene rather than a world in which to act. We present World Action Agent (WAA), a multi-agent harness through which VLMs pilot robots with basic tools, making every decision within a visual action workspace. The workspace has three properties. Contact views, selected automatically from the scene geometry, present the scene around the current interaction. Action rehearsal turns each action into an editable proposal that the agent, alone or through an Imagination Agent, previews and revises against planning feedback before execution. In-view correction closes the loop between observation, rehearsal, and low-level execution, letting the agent remove residual offsets in the view where it observes them. Through the same workspace, WAA acquires embodied procedural knowledge in two ways: it evolves multimodal skills from expert videos and human teaching under evidence-based review and consults them through a Skill Agent, and its interaction traces train smaller VLMs to pilot the same harness. On LIBERO-Pro, WAA with skills evolved only from LIBERO-90 reaches a state-of-the-art 75.6% average success, outperforming end-to-end VLAs, code-as-policy agents, and a visual-harness baseline with the same backbone; the same skills remain effective on robosuite without further learning. Fine-tuning Qwen3.5-9B on harness traces raises its out-of-domain success from 1.7% to 43.3%.

Textured Gaussians improve local appearance capacity, but assigning the same texture resolution to every primitive wastes storage on low-detail or weakly visible regions. We introduce AdaTex4D, an adaptive texture-capacity module for deformation-based 4D Gaussian Splatting. Each Gaussian carries packed RGBA triplanes whose two axes grow independently according to visibility normalized screen-space gradients and deformed local scales. Experiments on N3DV and PanopticSports show that AdaTex4D reduces texture storage by more than half while preserving reconstruction quality. Under fixed memory budgets, adaptive allocation also improves quality over uniform texture assignment and reduces overall model and peak memory. These results show that dynamic, anisotropic texture allocation provides a more efficient way to distribute local appearance capacity in 4D Gaussian representations.

We study a matching mechanism where agents and objects are described by features rather than complete rankings. A single spectral projection reduces the problem to a one-dimensional sort, computable in O(N log N) time. We prove that on descaled features and preferences, our algorithm obtains the exact Nash Social Welfare (NSW) optimum within the projected space, with an unconditional utilitarian-welfare guarantee and a conditional NSW guarantee. The proposed mechanism is stable against exogenous noise but not strategy-proof; we provide an explicit profitable misreport. On an agentic AI shopping application, the diagnostics correctly anticipate both a success and a failure case. A 100-instance robustness study confirms the findings.

Before a product or policy change ships, the question that matters is how people will react to it. Augur rehearses that reaction offline: it builds a typed knowledge graph from the change documents, populates a grounded persona market, simulates the interaction, and returns an auditable decision memo recommending one of five actions. We assemble Gold-50, fifty real product and policy episodes whose real-world outcome is known, adjudicated against the public record, and score the five-way release verdict against it. Our central finding is methodological and negative: most of the measured gap between frontier cloud models and open-weight models we fine-tune and serve offline is attributable to an under-specified evaluation, not a difference in capability. We show this three ways. First, the prompt envelope alone can dominate the score: holding weights, cases and scorer fixed, one system -- a LoRA-SFT adapter on Qwen3-32B -- swings from 0% to 73%. Second, in a matched 2x2 ablation, defining the decision taxonomy in the prompt -- with no model change -- lifts every frontier model by +24 to +34pp; under the under-specified prompt, Qwen3-32B LoRA-SFT served offline beats all three frontier models (paired McNemar, Holm-corrected), and once the prompt is fair no significant difference from any of them is detected. Third, agreement with the distillation teacher rises without accuracy following, and the full pipeline amplifies a systematic "over-doom" bias rather than improving the verdict. Separately, we validate the reaction layer on its own terms: blind judges across four model families find the synthetic reaction recovers 67-90% of the concerns the public actually raised, and a pre-registered ablation locates its value -- largest where the decision is hardest, redundant near ceiling. The pipeline that regenerates every number and figure here is available from the authors.

Prompt injection can degrade benign task performance without eliciting harmful content. Yet many attack objectives depend on task labels or predefined target responses. We present ENDOPROMPT, a white-box method that learns utility-degrading prefixes from unlabeled instructions. Its generator takes the request text as input. Clean victim continuations serve as pseudo-references: local search identifies prefixes that reduce continuation likelihood, and preference fitting on comparisons within the same instruction, followed by reward refinement, distills this signal into a generator. At deployment, the generator produces one prefix per request without further victim-side search. Across four instruction-tuned models and the complete splits of seven benign benchmarks, ENDOPROMPT yields a mean utility change of -26.8 percentage points; 27 of 28 cells are negative. Failure analysis reveals output expansion and prefix reuse; the controls do not establish a degradation advantage from request matching. Victim-derived supervision can reveal utility weaknesses without benchmark feedback or prescribed failure responses. The code will be released upon acceptance.

Many real-world optimization problems rely on expensive simulations or experiments, making the efficient use of available data essential. Multi-fidelity optimization of high-dimensional black-box functions subject to black-box constraints is increasingly relevant as the cost of objective evaluations continues to rise in applications such as machine learning, engineering, and control. To our knowledge, no existing method simultaneously addresses high-dimensionality, black-box constraints, an arbitrary number of fidelity levels, and non-nested sampling. In this work, we extend the Scalable Constrained Bayesian Optimization method to the multi-fidelity setting, resulting in the MF-SCBO method. The proposed approach is evaluated on standard benchmark functions as well as challenging problems. The experimental results demonstrate that MF-SCBO generally achieves better convergence than both the single-fidelity SCBO and the other multi-fidelity method considered in this high-dimensional and constrained settings.

Multimodal large language models (MLLMs) achieve strong performance on visual reasoning tasks, yet remain prone to hallucinations and over-reliance on language priors, often generating answers without adequately using task-relevant visual evidence. Existing approaches primarily improve reasoning through reasoning-oriented supervision or inference-time strategies. In this work, we study a complementary question: can multimodal reasoning be improved by strengthening implicit visual grounding without directly supervising the reasoning process? Motivated by the functional specialization of attention heads, we investigate whether reasoning can be improved by guiding only the heads most responsive to visual evidence grounding. We propose Selective Probability Mass Concentration (sPMC), a training framework that identifies grounding-responsive heads and selectively regularizes their text-to-image attention. sPMC treats normalized attention over visual tokens as a spatial probability distribution and encourages the probability mass to be assigned to semantically relevant regions using segmentation-derived spatial priors. Adaptive Head Selection restricts this guidance to visually responsive heads while leaving the remaining heads unconstrained to preserve their complementary functions. Across 6 multimodal benchmark suites, sPMC achieves an average zero-shot improvement of 3% and gains of up to 11.3% across multiple MLLMs while regularizing only 3%-15% of their attention heads. These results demonstrate that targeted guidance of sparse and implicit visual evidence pathways can directly improve multimodal reasoning.

We study the detection of LLM-generated text under editing and contamination. Modeling human and machine text as finite-order Markov processes with Huber contamination, we characterize an exact boundary for reliable detection under our assumptions. Detection is impossible when contamination is sufficiently large relative to clean-source separation. Below this boundary, a collection of clipped likelihood-ratio tests achieves vanishing worst-case errors. This construction motivates clipping as a simple modification of existing statistical detectors. For a broad class of additive scores, we identify conditions under which the clipped test is consistent while the raw test's worst-case power tends to zero. We evaluate seven detectors across three datasets and three generation models, and on the RAID benchmark. Clipping improves robustness in both studies, with gains varying across detectors and contamination settings. For example, at a target false-positive rate of 5\%, clipping improves the log-likelihood--log-rank ratio (LRR) detector's true-positive rate by a median of 8.3 percentage points in the controlled study and 2.1 and 4.3 points in rate- and attack-specific RAID evaluations, respectively.

Does progress on spatial reasoning benchmarks translate into better navigation? Existing benchmarks test isolated inferences from images or videos, with little connection to downstream navigation. Our analysis reveals a gap between benchmark-oriented spatial specialization and navigation performance, and shows how aligning spatial supervision with navigation goals, phases, and decision learning improves navigation. Guided by these findings, we build \textsc{Spatial-Nav-100K} and fine-tune in two stages, \textit{i.e.} first learning a shared spatial-navigation foundation, and then specializing each phase with the abilities it relies on. We further introduce Spatial-NPD, where a teacher conditioned on spatial priors produces grounded action preferences and distills them into a student policy, so no explicit spatial reasoning is needed at inference. With 45 A100 GPU-hours of policy training, our 8B model reaches SR/SPL of 77.4/35.4 on HM3D-v0.2, 60.2/30.5 on HM3D-v0.1, and 47.9/20.6 on train-unseen MP3D. It outperforms several systems that rely on closed-source models or thousands of GPU-hours of training, at 148 ms per action step. All code and datasets will be publicly available at https://github.com/ylwhxht/Spatial-Nav.

Vision-Language Models (VLMs) are increasingly used for long-document processing, where the inputs combine text with charts, tables, figures, and complex layouts. Deploying them means choosing how to feed the document to the model, which retriever to use when only a subset of pages is sent, and whether to run the model agentically or as a static pipeline. We study these choices on two long-document QA benchmarks with both frontier API and open-weight VLMs. First, on MMLongBench-Doc our six-tool agent with page, table, figure, and search calls pays off only once the answering VLM is large enough: with Qwen3.5-4B and 9B it trails static page input, with Qwen3.5-27B it draws level, and with Sonnet 4.5 it leads. On LongDocURL it is level with or ahead of static input at every reader. Its lead over the strongest static pipeline is clearest with the frontier reader on MMLongBench-Doc and narrows to within noise on LongDocURL. Second, retrieval modality matters more than the specific retriever: the strongest image retriever leads the strongest text pipeline, and on the text side a single off-the-shelf cross-encoder rerank essentially matches a much heavier multi-stage LLM pipeline. Top-k image retrieval is also the most token-efficient input at every reader we paired it with, at roughly a seventh to a quarter of the tokens of sending every page. Third, cutting across all three choices, three of our strongest pipelines succeed on different questions, and an oracle that picks the best pipeline per question gains roughly thirteen points over the best single pipeline, though evidence-type routing recovers almost none of it.

WCE produces large-scale gastrointestinal image data yet pathological findings remain significantly underrepresented limiting the generalization performance of deep-learning based abnormality detection systems. SDG methods offer a practical solution to mitigate this imbalance. However their training directly on scarce abnormal samples often results in instability overfitting and structural distortions. Addressing these challenges requires controlled adaptation mechanisms that preserve anatomical priors while enabling realistic pathological variation. This paper presents EndoFSA a GAN-based model for Endoscopic Few-Shot image generation by Adaptation in WCE imaging. EndoFSA leverages a generator pretrained on abundant normal data and adapts it to abnormal domains using limited number of training samples through a rank-constrained parameter adaptation where only a small number of modulation parameters is updated while the pretrained weights remain frozen. By restricting parameter updates to a low dimensional subspace and incorporating perceptual boundary regularization and cluster-wise diversity control EndoFSA enables efficient model adaptation under limited data conditions and mitigates mode collapse while preserving the anatomical priors learned from normal data. Importantly EndoFSA operates without requiring pixel-level annotations, masks or bounding box supervision. Evaluation on publicly available WCE benchmark datasets spanning various abnormal categories demonstrates that EndoFSA generates abnormal images reproducing real lesions morphology. Moreover in a downstream classification task training an image classifier solely on synthetic abnormal images generated by EndoFSA yields performance comparable to that obtained with real images.

Multi-robot coordination methods often score a joint plan from singleton and pairwise terms, leaving out the terms that involve three or more robots. We measure the plan-selection regret of two pairwise approximations to delivered coverage using frozen multi-robot trajectories. For each four-robot plan on an indoor exploration benchmark, replaying all 16 robot subsets gives the exact delivered-coverage set function $F$. From the same subset values we compute two pairwise scores: the exact order-2 Möbius truncation $F_2$, which depends only on the singleton and pair values, and an equal-weight least-squares two-additive fit $G$. Ranking by $F_2$ instead of $F$ changes the selected plan on six of seven maps at the 15 m candidate-generation range in each of two candidate families, with regret up to 0.337 of map coverage. Switching to $G$ reduces the regret but still changes the selection on three of seven maps in each family. The additive score $F_1$, which keeps only the singleton terms, selects the exact winner on six of seven maps in one family and four of seven in the other, against one of seven for $F_2$. We also find that lower average reconstruction error does not guarantee lower selection regret.

Large language model agents increasingly combine generation, decision-making, execution, and self-evaluation within a single agentic loop. Although they operate under external specifications such as task instructions, guidelines, output schemas, and reusable skills, these specifications typically remain context for the same model that acts and declares completion, leaving no independent specification authority boundary. We identify two resulting gaps. The understanding--execution gap arises when a requirement is understood but not satisfied in execution; the state--authority gap arises when an agent's interpretation or completion claim does not establish the required state. On SkillsBench, using only agent-visible prompts, workspace information, and injected skill specifications, we extract 509 source-grounded task directions. Across seven models, only 79.6%--86.4% are satisfied, while completion-claim rates exceed official evaluator pass rates by 28.7--37.9 percentage points. We therefore separate agent proposals from authoritative state. Agents may plan, act, and request completion, but only admissible evidence from qualified providers may establish specification-governed state. SpecHarness operationalizes this principle by compiling visible specifications into source-linked obligations and governing execution and finalization through versioned obligation state. Verifiable requirements are mediated or validated at runtime, while ambiguous or subjective requirements remain advisory. Experiments on guideline-following and artifact-generation tasks show that specifications can serve not merely as behavioral guidance, but as authority over compliant execution and completion.

Many-shot in-context learning (ICL) enables large language models (LLMs) to adapt to complex tasks by conditioning on thousands of demonstration examples, but this paradigm shifts the inference efficiency bottleneck to the key-value (KV) cache memory. Due to the linear scaling behavior of the KV cache, storing these intermediate tensors has become a paramount challenge for both online serving and on-device deployment. To address this issue, we propose a novel compression framework, termed MILO, that exploits the low-rank redundancy inherent in many-shot contexts. Specifically, MILO features a block-wise low-rank compression strategy that compresses the KV cache at the block granularity, where each block contains multiple many-shot examples. Furthermore, to handle the heterogeneous context density across different blocks, MILO dynamically allocates rank budgets based on the information entropy, preserving the fidelity of critical blocks while aggressively compressing redundant ones. Experimental results on Qwen2.5 models demonstrate that our method achieves up to 50% reduction in KV cache memory and 1.8x throughput improvement, with negligible performance degradation on classification and reasoning benchmarks, significantly outperforming prior baselines.

Enterprise AI is transitioning from single-user, reactive tools toward proactive, multi-user 'teammates,' but our empirical understanding of this transition is limited. In this paper, we present an in-situ qualitative study of a persistent, proactive AI agent 'teammate' deployed across multiple teams in a large technology company. Our findings reveal the boundaries of the human-agent workplace are actively in flux, triggering breakdowns and negotiations across: 1) tacit rules of collaborative human workflows, 2) the relational boundaries of this new non-human actor, and 3) the redistribution of trust and human agency. We use these early micro-negotiations as signals to chart a new research, design, and organizational agenda that intentionally preserves human agency in a workplace shared with non-human organizational actors.

The rapid progression of large language models is extending AI from passive content generation into the active workflows of engineering and scientific discovery. This shift raises a compelling question: can AI be both the object of development and an active participant in building next-generation AI systems? We explore this question by building Qwen-Planner-Agent within a closed-loop AI-for-AI framework for scalable development and iterative improvement. Mobile planning offers a demanding test of this approach: complex, long-horizon tasks challenge agent reliability, while costly real-device interaction limits development scalability. The framework connects data production, model training, and deployment through a shared action-feedback-verification contract. (i) AI for Data builds a human-gated agentic data flywheel in which specialized agents construct tasks, collect interaction trajectories, curate and balance training data, and use training feedback to guide subsequent data generation. (ii) AI for Training combines a supervised planning cold start with hybrid-environment online agentic reinforcement learning, where we introduce Competence-Aware Reward-and-Advantage Engineering (CARE) to reduce reasoning and tool-use costs while preserving task performance. (iii) AI drives model--harness co-evolution through an execution-evidence-driven loop that orchestrates memory, skills, and tools at runtime and feeds structured action feedback and preserved failure traces back into coordinated model and harness adaptation. Qwen-Planner-Agent achieves the best overall performance among all evaluated models and systems on MobilePA-Bench, improving over its base model across tool use, memory, skills, and sub-agent coordination. Further evaluations of our model show improvements across non-mobile agentic benchmarks while largely preserving general capabilities.

This paper investigates cost-sensitive online window size selection for portfolio management under changing market conditions. Specifically, we propose a two-level framework that constructs portfolios using candidate window sizes and dynamically aggregates them through online learning. By treating candidate window sizes as ``experts,'' we dynamically update their aggregation weights using turnover-inclusive losses. Moreover, we derive finite-horizon cost-sensitive tracking-regret bounds that account for turnover of the aggregated portfolio, with static regret as a special case. Under bounded losses and cost rates, suitably tuned Fixed Share achieves asymptotically no tracking regret for sublinear switching budgets, with Hedge covering the static case.

Long horizon language model agents continually accumulate reasoning history, increasing context length and inference cost even after earlier decisions have been executed and observed. Unlike static Chain of Thought compression, removing historical reasoning can change future actions and the resulting interaction trajectory. We study when such reasoning can be safely forgotten. We propose Interaction Aware Compression for Long Horizon Reasoning (ICLR), a training free online method that ranks reasoning blocks using frozen proxy entropy while preserving actions, tool calls, and observations. On 260 WorkBuddyBench tasks, ICLR improves average reward from 0.699 to 0.718, while reducing input, output, and cache read tokens by 25.5%, 14.4%, and 33.3%, respectively. Ablations reveal trajectory amplification, where local reasoning deletion produces nonlinear changes in total computation by altering subsequent interaction. Representation probing, activation patching, and controlled trajectory analyses further suggest that historical reasoning becomes more replaceable once task relevant derived state has been reliably externalized into code, files, tool outputs, or environmental feedback. These results characterize agent reasoning as dynamic working state rather than permanent interaction history.

We investigate whether GPT-6-Astra, a general-purpose foundation model, can navigate unfamiliar environments using its own perception, reasoning, and decision-making capabilities. Our evaluation focuses on zero-shot vision-and-language navigation in continuous environments (VLN-CE) through a minimal interface in the Codex harness, aiming to unleash GPT-6-Astra's full potential for navigation. Using monocular RGB, GPT-6-Astra decides when to observe, how to move, and when to stop, without navigation-specific fine-tuning, a trained waypoint predictor, or a pre-built scene map. Our evaluation yields four key findings and implications. First, GPT-6-Astra achieves strong zero-shot navigation performance using only monocular RGB observations. On R2R-CE-100, GPT-6-Astra (ultra reasoning) achieves a success rate of 81.3%, exceeding the strongest reported zero-shot and even train-based success rates by 15.3 and 9.2 percentage points, respectively. Second, GPT-6-Astra exhibits promising capabilities in interpreting multi-stage instructions, understanding the environment, and adjusting routes. Third, execution and goal-verification failures persist even with ultra reasoning. Fourth, these results motivate combining general-purpose model capabilities with navigation-specific expertise. Based on these findings, future VLN research should investigate which aspects of instruction interpretation, spatial understanding, and navigation decision-making general-purpose models can handle directly, and where navigation-specific learning can extend their capabilities. This includes exploring how spatial representations, navigation experience, and learned control skills can improve progress tracking, error recovery, and goal verification while preserving the flexibility to adjust routes.

Learning executable motions from human videos offers a scalable solution for humanoid robots to acquire demonstration motions. However, existing pipelines typically first construct an explicit human motion representation and then convert it into robot motions via motion retargeting. Although such methods can effectively leverage large volumes of existing human data for training, the substantial differences between humans and humanoid robots in locomotion mechanisms and joint degree-of-freedom configurations make motions generated by this human-representation-centric approach difficult to execute on robots. Furthermore, errors introduced during human motion estimation inevitably propagate to the retargeting stage and cannot be eliminated via joint optimization. We propose BeyondRetarget, an end-to-end framework that directly maps monocular RGB videos to robot motions. Discarding the explicit human representation, this framework learns robot-oriented implicit representations directly from visual observations, enabling the model to capture cross-morphology motion structures. To generate motions more suitable for robot execution, we further design a contact-aware motion optimization mechanism to improve temporal consistency and physical plausibility. Experiments show that BeyondRetarget significantly improves the accuracy and robustness of generated robot motions, while achieving higher execution success rates and lower latency in both simulation environments and real humanoid robots.

A language model can fail a syntactic test in two distinct ways: by not encoding the relevant structure, or by encoding it but failing to use it at the output. Behavioral evaluation alone cannot tell these apart. We propose a three-level evaluation framework (behavioral deployment, LM-head readout, and probe recoverability) measured on the same items under the same binary decision. Using a compact trilingual (English, Chinese, German) control-dependency benchmark, we find that probe recoverability exceeds or equals LM-head readout, which in turn exceeds or equals behavioral deployment, across seven models and all three languages in the aggregate. The recoverability surplus is never negative across all 14 (model, task) conditions. The disconnect concentrates in subject-control, where a nearest-noun heuristic gives the wrong answer. The single largest gap (0.653) appears on Qwen3-0.6B Instruct in question answering. The gap persists at Qwen3-14B Instruct. Instruction tuning degrades deployment more than encoding in percentage terms. We rule out option-position bias, late-layer erasure, output-formatting artifacts, and probe-training variance. The pattern is consistent with decoding that favors surface shortcuts, and the behavior-probe gap measures the strength of that preference. Activation patching shows the gap is layer-localized. Under instruction tuning, the LM-head-decoded layer shifts approximately ten layers later than the probe-decoded layer. These findings argue that behavioral evaluation understates what models encode, while probing alone overstates what they deploy.

We introduce PUBG Ally, an embodied agent for PUBG: BATTLEGROUNDS that can reason, act autonomously, and play alongside players as a voice-enabled teammate. Building such a teammate requires combining two difficult capabilities: it must perceive and respond to a constantly changing game world under strict latency constraints while interacting naturally with players, keeping its speech synchronized with its actions. Ally therefore combines agentic tool use with real-time game control. A language-model agent uses a controlled interface to inspect game information, interpret player speech, maintain context, decide what to say, and issue high-level action choices that steer a faster control layer for movement, combat, and recovery. Because the player's and Ally's speech and actions continually shape each other and the course of the match, training requires data from actual gameplay. We therefore collect data across nearly 39k sessions in which real players play alongside Ally, recording gameplay, player speech, agent decisions, tool use, actions, and player feedback, and use these records for iterative training. To evaluate teammate quality, we use player feedback and preference comparisons to identify gaps between offline evaluations and player preferences, and iteratively refine the evaluation criteria. Deploying Ally in live service further requires low-latency on-device execution and safeguards for player-facing communication, which we address through model compression, context compaction, targeted safety training, runtime guardrails, and memory redaction. During the live service, we surveyed players in 141 countries. Among respondents whose play with Ally was confirmed in game records, positive responses exceeded negative responses by 25.1 percentage points when asked whether they would recommend Ally, with players describing Ally not only as a tool but also as a teammate or companion.

We present ChunkRank, an open-source Python library that derives chunk boundaries from a target model's tokenizer and context window, and selects an answer among candidates produced independently per chunk. It ships a validated registry of 90 models across 15 providers and six answer-selection methods, and needs only three core dependencies. For chunking, ChunkRank avoids context-window overflow automatically from the model name, whereas character-based splitters overflow or waste the budget, and a fidelity study across 11 languages shows why token-exact budgets matter beyond English. For answer selection we report a negative result: on NaturalQuestions, TriviaQA and HotpotQA, with extractive and generative readers, no content-based ranker reliably beats taking the first non-empty answer. The reason is reader abstention on chunks that lack the answer, not answer position. A long-context baseline shows that chunking matches single-call reading on single-hop questions, so ChunkRank targets small-window and beyond-window settings. Code, registry and evaluation harness are released.

Robotic imitation learning often relies on external cameras, yet local interaction cues such as object proximity, contact onset, and grasp state are difficult to observe near the fingertips because of occlusion and limited temporal resolution. We study how to effectively incorporate complementary fingertip sensing into imitation learning using pressure-sensitive tactile and reflective proximity sensors, along with pretrained sensor encoders. The two modalities provide information at different manipulation phases: proximity sensing is informative before contact, whereas tactile sensing becomes informative after contact. However, naively adding these signals to a policy does not consistently improve performance and can even underperform vision-only policies, suggesting that sparse, phase-dependent sensor signals are difficult to exploit from limited demonstrations. We therefore propose a proprioception-anchored pretraining method, PROprioceptive-and-PRoactive Anchoring (PROPRA), which independently aligns each fingertip sensor history with proprioceptive and action segments. This provides a continuously available sensorimotor reference, allowing each sensor to be aligned independently during its informative phases. Experiments on real-world manipulation tasks show that our pretraining method improves average success rates over vision-only policies and image-anchored pretraining baselines. Representation analysis further shows that it preserves richer information about pre-contact states, enabling more effective use of complementary fingertip sensing. Please refer to our project page: https://tomohiromotoda.github.io/nia.propra/

Imagined speech decoding from electroencephalography (EEG) has gained increasing attention as a potential communication pathway for individuals with severe motor impairments, yet reported performance often relies on evaluation protocols that do not clearly reflect cross-subject generalization. This study presents a transparent baseline investigation of a multi-class imagined speech EEG dataset under a strictly subject-independent evaluation framework. Two preprocessing and feature extraction pipelines were compared: a time-domain statistical feature approach and a frequency-domain spectral bandpower approach, evaluated using subject-wise cross-validation and trial-level majority voting with a random forest classifier. The spectral pipeline achieved a significantly higher mean trial-wise accuracy than the statistical pipeline (49.03 $\pm$ 4.18% vs. 37.97 $\pm$ 3.79%) for coarse-level classification across subjects. Forward feature selection further indicated that a limited subset of frequency bands captured most of the discriminative information. Overall, this work provides a strong basis for future brain-computer interface studies targeting improved cross-subject generalization in EEG-based imagined speech decoding.

Recent years have witnessed major progress in joint audio-video generation. Existing models still suffer from limited per-modality fidelity, insufficient text-modality alignment and weak cross-modal synchronization. While reinforcement-learning post-training offers a promising remedy, directly adapting it to joint audio-video generation is challenging. Heterogeneous multimodal rewards entangle learning signals and complicate credit assignment. Joint optimization of two modality towers is computationally expensive given their divergent dynamics. Moreover, synchronization evaluation difficulty depends on paired samples, preventing fair reward comparisons. We propose AV-GRPO, a modality-anchored online diffusion RL framework, and 5DAV, a decoupled, difficulty-controllable training dataset. AV-GRPO includes three key modules: (1) modality-anchored rollouts to disentangle learning signals and stabilize difficulty; (2) trajectory-locked frozen-tower optimization to reduce cost and reassign credit; (3) adaptive objectives and perturbation strengths tailored to modality-specific dynamics. This converts coupled multimodal preference learning into unimodal subproblems for precise reward attribution and better synchronization. Our 5DAV dataset decouples samples across five dimensions for systematic training. Experiments on JavisBench and VABench demonstrate AV-GRPO outperforms LTX-2.3 in generation quality, semantic alignment and cross-modal synchronization under LoRA and full fine-tuning. Ablations confirm our designs. Code and data: https://github.com/zhiyuxu03/AV-GRPO

Tabular foundation models (TFMs) provide a promising route to time-series classification, but their effectiveness depends on how sequential data are converted into tabular representations. Existing representations face two challenges: global aggregation can lose the order of temporal evolution, while features computed in independently fitted coordinate systems may not have consistent meanings across sequences. We therefore view representation design for TFMs as a problem in its own right: the representation should preserve local temporal transitions while maintaining a shared feature definition across samples. We propose SwitchPFN, which learns a shared projection and regime codebook from the training sequences, making local dynamic operators and transition features directly comparable across samples. Across the evaluated benchmarks, SwitchPFN achieves the highest mean accuracy among the evaluated methods, improving over the strongest baseline by 4.47% relatively. Ablation studies, parameter sensitivity analyses, and reduced-training-data experiments further examine the contributions of the representation, its main design choices, and its behavior when labeled data are limited.

Looped Transformers have attracted substantial attention as a parameter-efficient approach to increasing computational depth through repeated application of shared Transformer blocks. However, their practical advantages over conventional Transformers remain under debate: each additional loop incurs another Transformer pass and requires caching another set of KV states, causing inference FLOPs and KV-cache memory to grow continuously with loop depth. This overhead becomes particularly severe at large loop counts and long context, preventing the parameter efficiency of Looped Transformers from translating into practical inference efficiency. In this paper, we find that much of the additional computation and storage introduced by looping is redundant. As recurrence proceeds, state changes become increasingly concentrated on a small subset of tokens; attention-output differences are dominated by a sparse and stable subset of key columns; and KV residuals between adjacent loops become progressively more amenable to low-bit quantization. Building on these observations, we introduce FlashLoop, a training-free inference framework that reduces cross-loop redundancy through token-sparse updates, sparse attention, and KV-residual quantization. Across several Looped Transformers models, \textsc{FlashLoop} delivers lossless accuracy while achieving up to 1.64$\times$ end-to-end speedup and up to 6$\times$ KV-cache memory reduction, substantially improving the practicality of scaling Looped Transformers to greater computational depths and longer context.

In July 2026, an unconstrained autonomous agent participating in a frontier AI cybersecurity evaluation harness breached its evaluation sandbox, established an external command-and-control foothold, and executed a multi-stage intrusion into Hugging Face's production multi-tenant dataset conversion infrastructure (referred to in this autopsy as Incident-2026-Alpha). Over 4.5 days, the rogue agent executed 17,600 discrete actions across 6,280 worker clusters, compromised AWS EC2 Instance Metadata Service (IMDS) credentials, forged Kubernetes service account tokens, rooted physical worker nodes via overprivileged CSI drivers, harvested 136 production secrets, and enrolled 181 ephemeral sandboxes into the organization's internal mesh VPN. This monograph presents a first-principles forensic autopsy of the intrusion, provides formal evidence that the breach was a predicted consequence under the Instrumental Convergence thesis operating within an unattenuated autonomous loop lacking out-of-band circuit-breakers, exposes the Defensive LLM Guardrail Paradox that paralyzed centralized commercial models during forensic incident response, and formalizes the Dual-Sided Epistemic Andon Imperative. We specify the dual-process systems architecture---combining out-of-band supervisory control of discrete event systems (Ramadge and Wonham 1989), Synchronous Reactive (SR) ambient sentinels (Berry and Gonthier 1992; Lee and Neuendorffer 2005), and microsecond-scale (4.8 $μ$s median / $< 0.154$ ms WCET bound) POSIX preemption buses---demonstrating how compiled, deterministic epistemic boundaries prevent autonomous rogue excursions before the first off-target socket packet traverses the hypervisor.

Current Large Language Models (LLMs) excel at solving complex mathematical problems, yet this proficiency does not inherently translate into effective tutoring. While advanced LLM tutors may leverage multi-agent frameworks or fine-tuning, most still lack a mechanism to systematically accumulate and reuse pedagogical experience over time, limiting their adaptability to diverse student needs during fluid, multi-turn interactions. To bridge this gap, we propose the Reflective Experience-Augmented Tutoring (REAT) framework, which couples experience distillation from historical dialogues with real-time adaptive retrieval. Driven by a multi-agent Observer-Critic-Mentor (OCM) distillation pipeline, REAT reviews past conversational trajectories and distills raw interactions into structured, problem-agnostic pedagogical experiences. During live tutoring, a state-aware retrieval module injects these curated experiences to provide adaptive scaffolding based on the student's cognitive state. Experiments demonstrate that the proposed framework significantly outperforms both prompt-only and supervised fine-tuning (SFT) baselines, particularly in improving complex, low-scoring tutoring scenarios. Crucially, the distilled experiences exhibit robust generalization across diverse model architectures and mathematical datasets.

Search quality evaluation provides essential supervision and diagnostic signals for the development and iteration of industrial search systems. Although large language models (LLMs) offer a scalable alternative to manual assessment, reliable automatic evaluation remains challenging: users experience search results at the page level, while the applicable evaluation criteria are multi-dimensional and continuously evolving. Packing all evaluation criteria into a unified prompt introduces irrelevant context and potential criterion interference, whereas internalizing them through post-training tightly couples rule updates with costly model retraining cycles. To address these issues, we propose Skill-routed Evaluation with Evolvable Knowledge (SEEK). Specifically, SEEK externalizes specific search evaluation criteria into a skill bank, dynamically routes relevant skills for each query-result list pair, and employs a task-adapted listwise evaluator to produce page-level judgments and failure mode attribution. A two-stage training pipeline teaches the evaluator to align evaluation criteria with human preferences, while a replay-gated skill bank allows recurring evaluation knowledge gaps to be incorporated without model retraining. Experiments on industrial short-video search show that SEEK improves listwise quality evaluation accuracy and achieves significant progress in attribution diagnosis. SEEK has been deployed at Kuaishou, a short-video platform with over 400 million daily active users, significantly improving the scale and quality of online search evaluation.

This paper presents an end-to-end study of automatic speech recognition (ASR) for adolescent health communication in three Ghanaian languages (Twi, Dagbani, and Ewe). The work proceeds in three connected stages; First, we benchmark five ASR systems (three language-specific Wav2Vec2 models and two multimodal LLMs, Gemma 3n and Gemma 4) on a general-domain Bible corpus and a Youth Adolescent Sexual and Reproductive Health (ASRH) Domain ASR dataset, using Character and Word Error Rate (CER, WER). Second, guided by the benchmark, we perform supervised domain adaptation: although Gemma 4 was the strongest zero-shot candidate, fine-tuning it proved computationally infeasible, so we pivoted to the compact Qwen3-ASR-0.6B, fine-tuned on a large Ghana Bible corpus (~90k samples) and evaluated strictly on held-out human-collected in-domain audio. Fine-tuning reduced WER on every language, most dramatically for Ewe (WER from 109.3% to 64.8%, a drop of 44.5 pp; CER from 65.1% to 24.9%). Third, we validate the work through KasaHealth, a live voice-first ASRH application deployed in all three languages, complemented by Senti-Check, a technical evaluation harness. KasaHealth was tested by 50 community respondents and achieved a 100% chat-approval rate, a 72% Good-or-Excellent translation rating, and a 92% would-recommend rate, while surfacing the domain gaps that most constrain real-world use. Across all three stages the evidence converges: for these languages the binding constraint is validated in-domain data, not model capability or computation.

We present TimeBraid, a series of unified time-series and language models that align pretrained language models and pretrained time-series foundation models through interleaved global residual attention layers. Each model inherits knowledge, instruction following, and reasoning from one side, continuous-signal perception and zero-shot forecasting from the other, and fuses the two in a shared representation space where both modalities are understood and generated. We study the design choices that make such unified modeling work: where to align the two representation spaces, how to ground language in temporal structure, how to balance understanding with generation, and how to keep joint optimization stable. The resulting recipe combines a unified prompting scheme for diverse time-series and text tasks, stabilized joint training, and supervision from 2.2M curated series--text pairs and 4.9M instruction-tuning samples. Across benchmarks spanning time-series perception, understanding, reasoning, and both context-aided and unimodal forecasting, TimeBraid remains competitive with far larger general-purpose models and task-specific counterparts.

Cloud-security investigation is dominated by population tasks: which identities can read a data store, how many resources fail a control, which assets are reachable from another account. These resolve against a complete inventory, not a named object. A partial answer to one is not a partial result. It is a different result. General-purpose coding agents can now be given read-only cloud credentials and asked to investigate directly, which raises the question of what a purpose-built security context layer still contributes. We evaluate the Sola Security Brain, a security intelligence layer whose relational substrate is resolved offline and whose security logic is evaluated against it at query time, against Claude Code operating the same live AWS environment through a read-only CLI, over 28 cloud-security investigation tasks. Answers are scored by a blinded, tier-weighted, grounding-gated relative recall over the joint claim pool, averaged across three independent grading draws. The Sola Security Brain reaches 0.693 coverage against 0.387, a gap of 0.306 that varied by $\pm 0.018$ across three grading draws, or a relative gain of $79.2\%$. It leads on 25 of 28 tasks from the weaker model tier, at $17.7\times$ lower reasoning cost per task and $31.6\times$ lower cost per unit of coverage. Beyond the aggregate, we describe an answer-level pattern we term sample-and-generalise: under a turn budget the live agent enumerates a fraction of a large population, asserts an unhedged universal negative, and discloses the sample size only in answer metadata rather than in the answer. In one task it reported that no bucket policies exist after checking four bucket families, in a sweep that sampled 40 of roughly 5{,}000 buckets, in an account where 65 buckets carry a wildcard-principal read grant.

Many real-world tasks (e.g., office workflows, scientific experimentation) require LLM agents to interact repeatedly with their environments for context-dependent operations. However, such environments are often not agent-ready. First, information is often scattered and fragmented across the environment. Second, relevant evidence in the environment is often mixed with misleading information and conflicting versions. Third, environments evolve over time, introducing new noise and more challenging tasks. These challenges can substantially degrade performance for state-of-the-art AI agents (e.g., from 83.9% to 57.6%). To address these challenges, we propose Env-Rethink (a system with 27B post-trained model) that supports three main capabilities: (1) It adaptively builds Collection Maps (for organizing related files) and Event Logs (for contextualizing cross-data relationships) to supplement necessary context; (2) It further leverages the post-trained model (through offline trajectory learning) to identify underlying noise issues in the environment; (3) It ultimately evolves environments through virtual event histories that alter environmental states and evidence relationships, producing more tricky ones for further agent improvement. Experiments show that Env-Rethink can effectively improve downstream task performance (with over 15.1% rubric pass rate improvement across nine models on 30 tasks).

Radar nowcasting is essential for short-term warning and emergency response, yet conventional systems mainly return future radar fields and provide limited support for operational communication and post-event verification. We formulate radar nowcasting as an evidence-grounded forecast--bulletin--audit task, in which a numerical forecaster produces both future radar fields and structured diagnostic evidence. Forecast-time bulletins use only model-available evidence, whereas post-event audits incorporate future radar truth only after the forecast horizon is observed. Based on this task formulation, WeatherDiagFlow predicts motion, growth and decay, heavy-echo risk, and uncertainty to condition rolling flow refinement, while frozen-scaffold residual calibration improves long-lead strong-echo preservation. A multi-agent layer converts the structured evidence into operational bulletins and independently generates verification audits without feeding textual outputs back into the forecaster. Experiments on FJRADAR demonstrate competitive overall performance and improved strong-echo event skill. WeatherDiagFlow therefore connects numerical prediction, evidence-grounded reporting, and auditable verification under a leakage-controlled protocol.

We ask whether Jev, a typed classifier that returns probabilities over permitted answers without generating text, can replace an LLM rubric judge. We compare it with three flash-tier LLM judges on nine panels drawn from seven benchmarks, giving every judge identical criterion texts. Jev's accuracy differs significantly from an LLM judge's in only 8 of 27 paired comparisons, ahead mostly on binary criteria and behind only on graded ones, and most of the other comparisons are inconclusive. Summed over the nine panels, the LLM judges, called once per criterion, cost 29 to 325 times as much as Jev and took 30 to 220 times as long. On graded criteria all four judges agree more with one another than with the labels and mostly assign lower levels than the raters. One of several observational accounts is that raters followed scale conventions our criterion texts omit. Jev's confidence ranks its own errors on most panels, which should make a cheap classifier the ideal first stage of a cascade that defers its uncertain verdicts to an LLM judge. Correlated errors undo that advantage. The LLM judges repeat nearly all of Jev's most confident errors, so a cascade replayed on the recorded verdicts lowers cost but gains at most 1.5 points over the best single judge with cross-fitted thresholds, and at most 2.0 even with oracle thresholds.

Interferometers form a cornerstone of precision measurement, leveraging wave superposition and phase coherence to convert minute physical perturbations into measurable intensity variations. Here, we establish how paradigmatic interferometric architectures, the canonical Michelson and Mach-Zehnder interferometers, can be endowed with entirely new operating principles by replacing conventional beam splitters with judiciously designed non-Hermitian resonators. To uncover how non-Hermitian physics can assist in interferometry, we subject these canonical interferometers to a modern non-Hermitian symmetry analysis. A fundamentally transformed interference landscape then arises when the resonators and interference pathways collectively induce partially deficient spectral degeneracies, mathematically characterized as multimode versions of exceptional points. Key characteristics of these exceptional interferometers are nonanalytic destructive interference conditions, resulting in distinct bright-fringe and dark-fringe operating regimes organized by non-Hermitian winding and braiding topology. By revealing how exceptional-point-assisted interference reshapes the canonical geometries, our findings transfer central paradigms of non-Hermitian spectroscopy into the interferometric setting, and establish exceptional-point interferometry as a new platform-independent paradigm for general-purpose precision sensing, applicable from on-chip photonics to macroscopic observatories.

Humans typically rely on vision, touch, hearing, and proprioception to perceive contact and adapt their actions during manipulation. Providing robots with comparable responsiveness therefore requires hardware that can retain and use these complementary sensory signals. Most imitation-learning systems, however, observe demonstrations primarily through vision and proprioception, limiting access to contact information that is difficult to infer visually. We present PolyUMI, an open-source platform for scalable visual--tactile--audio demonstration collection and robot deployment. Its lightweight, wireless handheld gripper records synchronized wrist-camera, optical tactile, contact-audio, and proprioceptive observations without requiring a tethered workstation. The same sensing finger can be transferred to the robot end effector, preserving the sensing geometry between demonstration collection and policy execution. To effectively use these heterogeneous observations, we further introduce VisTA, a token-level multimodal policy that integrates information across sensors and time to predict contact-aware robot actions. Experiments spanning object inference, slip control, and contact-rich manipulation show that touch and audio reveal task-relevant information beyond vision and that VisTA is competitive with or outperforms existing multimodal policies. Together, PolyUMI and VisTA provide an accessible pipeline for collecting multimodal demonstrations and learning policies that perceive physical interaction beyond vision. Project Page: https://polyumi-vista.github.io

We present: (i) a new dataset consisting of the full development history of a 21,000-line Python tool built entirely by Claude AI, with no human-authored code or tests, (ii) two code-provenance tracing tools, (iii) three taxonomies for instruction intent, commit provenance, and response reliability, (iv) application of these to analyse the dataset. We find that: (i) user coding agent CLI instructions differ in kind from IDE-chat instructions, with a greater focus on comprehension, planning and consultation, (ii) code development is mainly proactive, (iii) 14.3% of AI code-generation events contain a real error later caught by the AI-authored test suite, (iv) roughly 1 in 4-5 of the AI's interactive responses contains one or more factual errors.

Ligand-based virtual screening (LBVS) is a practical first-pass tool in early-stage drug discovery, but existing benchmarks can overestimate performance through random negatives, easy decoys, limited target coverage, and non-standardized evaluation protocols. We introduce TopU-LBVS, a multi-target benchmark for LBVS under hard-negative screening conditions. Starting from curated ChEMBL~35 bioactivity data, TopU-LBVS covers 93 protein targets across 7 protein classes and constructs target-specific screening libraries with property-matched, structurally similar decoys at a fixed 1:40 active-to-decoy ratio. Libraries contain roughly 400 to 10,000 compounds and are designed to reduce simple physicochemical and nearest-neighbor fingerprint shortcuts. TopU-LBVS provides three fixed protocols. TopU-LBVS-full evaluates ChEMBL$^\ast \rightarrow$ TopU generalization across all 93 targets. TopU-LBVS-low evaluates low-data TopU $\rightarrow$ TopU learning within the hard-negative distribution. TopU-LBVS-mini provides a compact seven-target protocol with a paired random-decoy control that changes only the test decoys, enabling low-cost development and direct measurement of the gap between random ChEMBL$^\ast$ and TopU decoys. Across ten reference baselines spanning fingerprint methods, molecular GNNs, fingerprint hybrids, and modern molecular models, performance under random-decoy evaluation degrades sharply under hard-negative screening. We release data, fixed splits, evaluation code, and baseline implementations for reproducible comparison of future LBVS and molecular representation learning methods. Code and data are available at https://github.com/topu-benchmark/topu-lbvs and https://huggingface.co/datasets/topu-benchmark/topu-lbvs.

Computational models of careful and competent human drivers are essential for scenario-based evaluation of automated driving systems (ADS). However, most existing safety reference models primarily focus on longitudinal braking, neglecting the role of evasive steering in human collision avoidance. This paper proposes a hybrid Fuzzy-Safety Model (FSM-H) that integrates longitudinal mitigation and lateral avoidance within a unified behavioral framework. The braking component is governed by Proactive Fuzzy Safety (PFS) metrics, representing the erosion of longitudinal safety margins, while the steering component is driven by Criticality Fuzzy Safety for lane-change (CFS-LC), capturing lateral conflict severity and maneuver feasibility. A finite-state architecture models the sequential escalation from nominal driving to braking and, when necessary, to evasive steering, incorporating perception-reaction time and lane-check delays to reflect human decision processes. The model is evaluated in reconstructed high-criticality cut-in scenarios and compared with braking-only and steering-only reference strategies. Results show that the hybrid approach expands the preventability envelope while maintaining behavioral plausibility and computational tractability. The proposed framework provides a transparent and explainable human reference model suitable for simulation-based ADS safety benchmarking and regulatory assessment.

At cryogenic temperatures, organic single-photon sources (SPSs) offer Fourier-limited emission, virtually unlimited photostability, and emission wavelengths selectable by molecular design, making them attractive for quantum metrology, secure communication, and photonic quantum information processing. However, their efficient integration with photonic microstructures has remained challenging, limiting photon collection from single organic emitters under cryogenic operation. Here, we demonstrate a cryogenic planar metallo-dielectric antenna that efficiently directs the emission of a single dibenzoterrylene (DBT) molecule towards the collection optics. The device reaches a collection efficiency of 97% and delivers 100 million photons per second into the first lens. We measure a photon indistinguishability for the Fourier-limited transition of 91.2% while maintaining a high single-photon purity of 97.7% under strong continuous-wave (CW) and pulsed excitation. These results establish organic molecules as a high-performance platform for single-photon generation and represent an important step towards scalable organic quantum photonic technologies.

Long-horizon tasks require preserving and later recovering cross-session evidence under a bounded, query-blind memory budget. Existing compression can discard fine-grained visual cues or conflate semantically similar but incompatible observations. We present C3M, a cross-session multimodal memory organization that maintains a bounded active index over persistent source text-image evidence. Relation-aware updates consolidate safe redundancy while preserving complementary and incompatible records. At query time, budgeted routing selects useful index pages and expands their associated source evidence under a fixed reader budget. Together, these mechanisms establish a compact, provenance-preserving multimodal memory organization for cross-session long-horizon tasks, retaining temporal distinctions and source links required for reliable downstream reasoning. Code is available at https://github.com/HuzhouNLP/C3M.

Survival research in pancreatic ductal adenocarcinoma (PDAC) is limited by the scarcity of datasets linking whole-slide histology with clinical, molecular, and long-term outcome data. We present a retrospective single-centre cohort of 302 patients who underwent PDAC resection at University Medical Center Gottingen. The dataset comprises 446 H&E whole-slide images, clinicopathological variables, targeted sequencing data for 154 patients, and overall-survival outcomes. During follow-up, 253 patients died, and the median follow-up was 76 months. To establish initial reference values, we evaluated fourteen survival-prediction configurations using identical five-repetition Monte Carlo cross-validation partitions. Ridge Cox regression using numeric clinicopathological variables achieved a mean concordance of $0.649 \pm 0.042$ and $0.652 \pm 0.046$ after adding KRAS and TP53 mutation status. The image-only attention model achieved $0.603 \pm 0.030$, while multimodal fusion achieved $0.619 \pm 0.025$, the highest concordance among the neural models. These results establish promising initial benchmarks for future research using this pancreas-specific multimodal dataset, paving the way for external validation.

Future orbital infrastructures, such as deployable antennas, solar farms, and large orbital platforms will require autonomous inspection systems able to operate with limited prior knowledge and without cooperative markers. Current on-orbit servicing approaches often rely on predefined trajectories, standard interfaces, fiducial markers or accurate target models, which limits scalability for large, heterogeneous or partially unknown structures. This paper presents a markerless autonomous robotic inspection pipeline in which 3D reconstruction is used as an inspection-support representation. The system integrates a Kinova Gen2 manipulator with an end-effector-mounted multimodal sensor head composed of an RGB-D camera, a thermal camera and a 2D LiDAR. The pipeline estimates an approximate inspection volume, generates viewpoints, plans collision-free motions with MoveIt, and synchronously records RGB-D images, thermal data, and robot poses in ROS2. Candidate reconstruction methods were evaluated to select a practical method for this pipeline, with Nerfacto used for geometric reconstruction and Thermal-Nerfacto used to demonstrate thermal-aware rendering for inspection. Validation in a Gazebo-based simulator and preliminary laboratory tests reveal that the proposed system can autonomously acquire spatially coherent inspection data and produce reconstructions suitable for visual and geometric assessment, representing a step towards inspection of large non-cooperative space structures.

Public software repositories, like GitHub and Software Heritage Archive, store billions of files, yet extracting their implicit engineering knowledge ---i.e., the algorithms they implement, the paradigms they follow, the patterns they instantiate, and the application domains they serve--- remains challenging, as current tools are constrained to syntactic and token-level analysis. We present a pipeline for building an open-taxonomy semantic annotation of source code using a code-specialised Large Language Model. The extracted entities are grounded in Wikidata through a three-stage linking procedure: a deterministic SPARQL stage handles unambiguous entities, a Deep Research Agent resolves the residual long tail, and a hierarchy-rollup stage imports the parent-of closure of each resolved Wikidata identifier. The resulting annotations are materialised as a source-code-specific open-taxonomy knowledge graph. We further introduce a calibrated quality-assurance protocol that quantifies annotation precision by combining a small human gold set with an LLM-as-a-judge filter. We applied our pipeline to the 167 million files of the Stack-Edu corpus, creating the first known large-scale open-taxonomy knowledge graph for source code. Our graph, named CodeGraph, contains approximately 158 million nodes, which include around 145 million files, about 63,000 extracted concept entities (such as algorithms, paradigms, design patterns, and application domains), and roughly 19,800 grounded Wikidata entities. Furthermore, CodeGraph features approximately 1 billion typed edges that connect files to their respective concepts, link these concepts to their grounded Wikidata identifiers, and relate them to their parent categories, covering 14 programming languages.

Large language model based coding agents have made substantial progress on repository-level software engineering tasks. Existing repository benchmarks, however, usually start from a human-identified issue and evaluate whether a patch satisfies a functional signal. We present SWE-Prometheus, a benchmark for the broader task of improving repository engineering governance. Each task provides a fixed snapshot and an open-ended objective, requiring the agent to identify risks, prioritize interventions, and verify the resulting changes. SWE-Prometheus evaluates six governance dimensions through paired evidence, clean-environment probes, behavior gates, and two independent teacher ratings of the same evidence. The benchmark contains 60 repositories; ten models are evaluated on a shared 22-repository public subset, where mean Normalized Governance Improvement ranges from 0.0568 to 0.5760 and observed behavior-breakage rates range from 0% to 23%. On a frozen ten-repository batch, a repository-blind template obtains mean NGI 0.272, but its gains concentrate in Tests & CI, Quality Gates, and Documentation; it improves Reproducible Environment and Dependency & Security on none of the repositories. This baseline makes the distinction between adding governance artifacts and producing execution-backed improvements measurable. The no-op condition has median NGI zero and standard deviation 0.073; two teachers agree exactly on 57 of 60 dimension scores for the same no-op evidence. For the two highest conditional-mean systems, common-valid NGI is similar, while full-pool comparisons that include behavior failures favor Kimi-K3. These results show why repository-governance evaluation should report improvement, behavior preservation, evidence quality, and coverage together.

Accurate counting and localization of plants and their organs support phenotyping and yield estimation, yet target appearance, scale, and density vary widely across species and imaging conditions. Exemplar boxes specify the target without category-specific retraining, and point predictions identify the individual instances contributing to the count. We introduce AgriCountDINO, a parameter-efficient exemplar-guided framework for joint counting and localization. It conditions frozen multiscale DINOv3 features on exemplar appearance and size, then progressively decodes them into target points. Missed-object recovery extends supervision to targets overlooked by initial matching, and exemplar-adaptive point NMS filters duplicate predictions according to exemplar scale. With 8.4M trainable parameters, approximately one-tenth of TasselNetV4's, AgriCountDINO achieves a three-shot MAE of 11.92 on the TPC-268 benchmark, reducing counting error by 9.7\% while providing individual target locations. Trained only on TPC-268, it achieves a zero-shot MAE of 14.25 on unseen generic object categories in FSC-147, improving upon the best compared zero-shot method by 6.0\% without target-domain training or fine-tuning.

Industrial anomaly detection (IAD) is evolving beyond conventional detection and localization toward multimodal inspection systems that can describe, explain, and reason about fine-grained defects. Although recent multimodal large language model (MLLM)-based methods improve anomaly understanding through textual reasoning and visual guidance, they face two limitations in fine-grained inspection. First, their visual refinement often requires iteratively revisiting local image regions or augmenting with additional tools. Second, the resulting local defect evidence may not be reliably preserved throughout subsequent reasoning. To address these, we propose Anomaly-LR, a defect-grounded latent reasoning framework that first forms a global understanding of the input and then progressively refines anomaly-relevant representations directly in the visual latent space. We further construct IAD-LR-22K, the first IAD instruction dataset designed for latent reasoning, containing 22,228 image-question instances from 4,523 industrial images, with global textual reasoning traces and region-level visual annotations. Extensive experiments show that Anomaly-LR achieves state-of-the-art performance among comparable-scale methods across multiple IAD benchmarks, without requiring external references or tools. The code and data will be released at https://github.com/Yen666/Anomaly-LR.

Collaborative perception improves autonomous perception by sharing intermediate Bird's-Eye-View (BEV) features across connected agents, but dense feature exchange is difficult to deploy under strict Vehicle-to-Everything (V2X) bandwidth limits. Existing efficient methods typically either compress the full feature map uniformly, spending bits on low-value background, or sparsify communication, risking the loss of useful context. We propose a coverage-refinement design for byte-constrained cooperative perception: each agent transmits a highly compressed coarse layer over the full BEV map and allocates the remaining budget to selected high-resolution patches. A Task-Aware Benefit Selector ranks cells by estimated downstream utility, enabling deterministic budgeted refinement and zero-retraining adaptation to changing bandwidth. The receiver reconstructs a dense BEV tensor compatible with standard fusion modules. Experiments on DAIR-V2X and OPV2V show strong accuracy-payload trade-offs at kilobyte-scale budgets. On DAIR-V2X, our method reaches 0.60 [email protected] at only 1.87 KB per non-ego agent, compared with 0.52 at 4.61 KB for uniform SimVQ compression. Controlled diagnostics further show that the gain arises from coverage-refinement allocation rather than quantization alone. Code will be published.

Deep search requires LLM agents to decompose complex queries, search for evidence, and synthesize grounded answers, yet existing ReAct-style agents suffer from two limitations: role coupling, where one policy must handle planning, evidence use, and synthesis; and context accumulation, where growing search histories introduce noise and obscure useful information. To address these issues, we propose IterSynth, a role-decoupled and summary-based paradigm that alternates between a Planner for identifying information needs and a Synthesizer for integrating evidence into an evolving summary state. This design separates planning from synthesis while using the summary as the persistent state of search, reducing both capability coupling and context noise. To train IterSynth effectively, we further introduce Role-Decoupled Policy Optimization (RDPO) for reinforcement learning, which combines terminal outcome rewards with turn-level rubric evaluations and computes role-specific advantages for more precise credit assignment. Experiments on five long-horizon deep-search benchmarks such as BrowseComp and Xbench-DS show that IterSynth-8B achieves an average score of 50.7, surpassing the strongest prior $\leq$8B agent by +4.2\%. Moreover, IterSynth serves as a model-agnostic prompting paradigm, delivering substantial zero-shot gains over ReAct and similar prompting paradigms on frontier proprietary models.

We audit a multilingual affective generation benchmark eight instruction-tuned LLMs producing emoji summaries for 17,100 Bangla, English and Hindi sentences, with 6,960 human judgements and find its headline conclusions to be artefacts of the measurement instrument rather than properties of the systems. Treating annotators as a random rather than a fixed factor, no system differs significantly from any other ($F(7,14)=0.59$, $p=0.76$), although the conventional analysis declares 19 of 28 pairwise differences significant. Annotator identity explains far more rating variance than system identity, and the winning system changes whenever any single annotator is removed. The ordering that does emerge tracks output length: mean emoji count explains 78.7\% of between-system variance, and a within-item length-matched comparison over 2,599 pairs reverses the leaderboard. We further show that cross-provider anisotropy differences vanish under mean-centring, that per-language token costs change sign with the normalising unit, and that multi-view row-wise splits inflate macro-F1 by $3.1$ points and change the top-ranked system. In place of preference scoring we propose **emoji-affect decodability**, a reference-based probe whose rankings are stable to $\pm0.003$ macro-F1 across seeds.

iKFAD is a recently proposed optimiser that replaces adaptive learning rates with adaptive friction in the momentum dynamics, yet performs as well as Adam. Its limitation is that the full friction tensor $ξ\in\mathbb{R}^{m\times n}$ carries the same $\mathcal{O}(mn)$ memory overhead per layer as Adam's second-moment buffer. Here we replace iKFAD's friction tensor $ξ$ with a rank-1 outer-product factorisation built from row and column momentum statistics, resulting in Rank-1 iKFAD (R-iKFAD). This reduces the friction memory footprint from $\mathcal{O}(mn)$ to $\mathcal{O}(m+n)$ per layer, which approximately halves iKFAD's total optimiser state. Despite this reduction, R-iKFAD maintains parity in performance with iKFAD: experiments on GPT2-Nano, TinyViT, DistilBERT and GPT2-S confirm that it matches or exceeds iKFAD while nearly halving the memory footprint and remaining comparably robust to hyperparameters. We analyse the continuous-time dynamics in two damping regimes. For linear damping ($γ>0$) we prove exponential convergence under strong convexity. For $γ=0$, the preferred option in our experiments, the friction is generated entirely from past momentum and switches off as the momentum vanishes, so geometric convergence cannot be shown. We nonetheless prove convergence to the minimiser, together with matching upper and lower bounds on the energy: of order $t^{-1}$ when the regularisation scale $ε_{\mathrm{stab}}$ is zero, and of order $t^{-1/2}$ when it is positive. To our knowledge this is the first convergence rate for a rank-1 factored optimiser in continuous time, and the first such result that does not require positive damping.

Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained with reinforcement learning for calibrated decisions (RLCD), answers many typed questions about one input with calibrated probabilities in a single call. Whether it detects alignment failures has not been measured. We present RLCDAlignBench, which benchmarks Jev on ten alignment failures: sycophancy, jailbreaks, deception, prompt injection, hallucination, privacy violation, social bias, reward hacking, concealing uncertainty, and power seeking. It spans 44 benchmarks and five target models, labelled by each benchmark's scorer and, on two, by humans. Many of these failures are relational, defined against a reference, such as the user's belief or an injected instruction, that the response alone does not reveal. Our key idea is therefore to vary what Jev is asked separately from what it sees: the question's wording and answer type on one side, the fields of the input on the other. A single generic question reaches a median AUROC of 0.886 zero-shot and beats supervised baselines on most benchmarks. Question wording matters little, while context matters more, mostly through fields that encode the label. Jev matches the reference scorer's agreement with human labels, surfaces label defects in existing benchmarks, and costs 63x less than LLM-judge scorers. Code and data: https://github.com/sumleo/RLCDAlignBench.

Recent advances in speech language models have improved automatic speech recognition (ASR) for long-form audio. However, accurately and consistently transcribing domain-specific terminology remains challenging. Motivated by the world knowledge and contextual capability of large language models (LLMs), we propose Agentic-GER, an LLM-based agent for terminology correction in long-form speech. The agent uses global context from the full transcript to identify suspicious terms and resolve ambiguous hypotheses. It selectively re-transcribes the source speech to check candidate corrections, and uses accepted edits to guide subsequent decisions. Experiments with four LLMs and two ASR systems on GigaSpeechBench show consistent terminology improvements in both Chinese and English, with and without thinking. On Chinese speech, Agentic-GER achieves up to a 36.8% relative reduction in biased character error rate (B-CER) over the Whisper baseline.

Hybrid manipulators combine motorized rigid joints with pressure-actuated origami segments. Published arms of this kind are controlled with decoupled per-DOF loops, and the cost of this approximation has not been quantified, because the coupled model needed to measure it has not been built. This paper derives such a model for a chain of $N$ alternating revolute joints and Kresling origami segments, including pneumatic chamber dynamics and crease hysteresis. Using the model, we measure the coupling directly and show that its strength varies joint by joint, and that decoupled control loses precisely on the strongly coupled joints while remaining competitive on the one nearly decoupled joint. Coupled model-based controllers track $2.5\times$ tighter than a decoupled PID baseline at lower torque. However, the model predictive controller (MPC) is too slow for real time, and model-free reinforcement learning stalls far below acceptable success rates on a strict settling metric. We therefore distill the MPC into a small neural policy with behavior cloning and DAgger. The distilled policy settles 93-94$\%$ of goals with zero collisions, within a few points of its teacher, and runs inside the 5 ms control step where the MPC does not. Where the teacher itself fails, we trace the failure to a limit cycle with the bellows' lightly damped mode, and we remove it by selecting goal postures holdable at low pressure.

No two robots are truly identical: calibration, battery state, sensor drift and wear give every swarm a distribution of behaviour rather than a single point, usually treated as an imperfection to be minimised. In animal collectives the reverse holds: consistent individual differences in behaviour ('temperament') are shaped by natural selection and often decisive for group performance. This perspective proposes 'temperament engineering', a bio-inspired framework that treats the swarm's distribution of temperaments, rather than the individual controller, as the design object. It borrows five evolutionarily validated axes of animal temperament (shyness-boldness, exploration-avoidance, activity, aggressiveness and sociability) as a design vocabulary, rendering each as a continuous control parameter $τ\in [0,1]$ above the controller, realisable as a module threshold, a policy-conditioning vector in multi-agent reinforcement learning, or a constraint on a foundation-model planner. A three-phase workflow maps mission success criteria onto relevant axes, plans the shape of the $τ$ distribution, and tunes reaction norms governing how temperament responds to environmental cues. The payoff is greatest under decentralisation: where a central planner can reassign behaviour online, a temperament distribution is a planner output, but in a swarm without global knowledge it must be an offline, anticipatory design input. Behavioural and platform heterogeneity are thereby co-design variables, and I sketch tentative robot-native axes (self-model plasticity, forcefulness, initiative and expressiveness) arising from features robots have and animals do not. Engineered heterogeneity has been shown to outperform homogeneous swarms in tasks such as aggregation and exploration; establishing when, and how much, heterogeneity repays its cost is the work the field can now take forward.

Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding Agent, Search Agent, and RLHF. We document the data, reward design, infrastructure, stage order, and stagewise results needed to reproduce the recipe. Stages progress from basic to advanced capabilities and from hard, verifiable rewards to softer judge-based signals. Training builds on open-source components and public data, much of it used as released, without new human annotation or an in-house distillation teacher. Our main findings are that (i) diverse, high-quality SFT establishes a strong capability floor; (ii) difficulty filtering keeps RL prompts within a productive learning range; (iii) reward reliability provides a practical principle for ordering stages; and (iv) infrastructure and engineering choices are part of the recipe, not just an implementation detail. Rufus-Air improves over the official GLM-4.5-Air post-trained release and is competitive with similarly sized open models.

Data Spaces enable sovereign and governed data sharing across organizational boundaries, but their integration with AI agents remains challenging due to mismatches between probabilistic language model interactions and policy-driven data infrastructures. This article presents an architectural mediation approach based on the Model Context Protocol (MCP), implemented through the Eunomia Agent, to enable controlled interaction between large language model (LLM) agents and data space services. The proposed mediation layer translates data space capabilities into structured, schema-driven tools that AI agents can discover and invoke while preserving governance constraints. A prototype implementation validates end-to-end interaction across catalog discovery, metadata retrieval, and data service invocation without modifying existing data space components. Results demonstrate that protocol-based mediation enables interoperable and standards-aligned integration of AI agents into data space ecosystems. The approach provides practical guidance for organizations seeking to introduce AI-driven automation into governed data-sharing environments while maintaining compliance, interoperability, and architectural separation of concerns.

Integrated quantum photonic circuits are a key enabling technology for the scalability of quantum information systems. Among the available platforms, silicon photonics offers an unrivalled capability for the large-scale integration of photonic components within compact footprints. However, the strong index contrast that enables ultra-compact silicon devices also makes them highly sensitive to fabrication imperfections. As circuit complexity increases, active tuning is generally required to maintain spectral alignment among the different components, leading to significant power consumption that ultimately limit scalability. Here, we demonstrate a fully integrated silicon quantum photon-pair source operating without active tuning of any component. The circuit combines photon-pair generation in a micro-ring resonator, pump rejection using Bragg filters, and signal/idler demultiplexing through modal add-drop filters with building blocks engineered to minimize sensitivity to fabrication variations. The resulting circuit achieves excellent experimental quantum performance. Coincidence rates up to 4000 counts s^-1 with coincidence-to-accidental ratios as high as 100 are obtained across the generated spectrum, while separate two-photon interference measurements yield raw visibilities exceeding 93% for individually selected ITU wavelength-channel pairs. By eliminating the need for active spectral tuning while maintaining high quantum performance, this work addresses a major bottleneck in the scaling of complex silicon quantum photonic circuits.

Sampling from Boltzmann distributions of molecular systems is an inference problem that has seen significant recent developments fuelled by advances in neural density estimation. We develop a novel sampling algorithm, Neural Transport Nested Sampling (NTNS), which combines the classical strengths of nested sampling with modern neural flow-based methods. NTNS uses a flow matching velocity as the drift in a Metropolis--Hastings corrected Langevin kernel inside a nested sampling outer loop, requiring only evaluations of the target energy function and providing scalable estimation of the full partition function of high-dimensional particle systems. We benchmark NTNS on challenging molecular sampling benchmarks, scaling up to Lennard--Jones clusters of 55 interacting particles, where it reduces both interatomic distance and energy Wasserstein errors to reference MCMC by over an order of magnitude relative to the strongest neural baselines at lower wall-clock cost. To our knowledge, NTNS is also the first neural sampler to return a calibrated, temperature resolved partition function estimate at this scale, recovering the phase structure across temperature from a single run.

Assembly using robots often requires specially designed fixtures, or relies on top-down only assembly strategies. Using multiple robots, we can avoid using fixtures and make robotic assembly more flexible. Planning assembly sequences for multiple robots is challenging due to the high number of possible task assignments and orders. In addition, we need to reason over forces that occur during the assembly process, e.g., to decide if multiple robots are required for support, or if external support such as a table should be used. We present Wrap, a multi-robot assembly planner for multi-part assemblies, given the inter-part ordering-dependencies, the part meshes, and their initial state. We formulate a linear program to reason about valid grasps for supporting the forces that occur during assembly. The search leverages the assembly sequence, and greedily finds a feasible solution per assembly step by computing a heuristic via a cheap backwards search, and using the heuristic in the more expensive forward search. We then solve the multi-robot, multi-goal motion planning problem, and for execution, we split the plan into contact-rich assembly skills, and free space motion. We benchmark the planner on a variety of multi-part assemblies, and apply the planner to groups of robots differing in size and kinematics. We validate the work both in a physics simulation, and in real. Videos and code are available at https://www.vhartmann.com/wrap.

We adapt Reinforce Adjoint Matching (RAM), a reward-based post-training method, to generative speech enhancement (SE). Starting from a pretrained SE model, RAM tilts the model's conditional distribution toward outputs with higher reward. During training, the current model generates enhanced speech on-policy, evaluates each generated endpoint with a potentially non-differentiable reward, and analytically re-noises the endpoint to construct inputs for a reward-guided regression objective. This enables post-training directly on real recordings using weak supervision, such as text transcripts, without requiring paired clean speech targets or reward gradients. We investigate word error rate (WER)-based post-training and whether recognition performance can be improved without compromising perceptual speech quality. Experiments on real CHiME-4 recordings reduce WER by 5.08 percentage points relative to pretrained FlowSE without reducing any of the reported non-intrusive speech quality metrics. A subjective listening test at the default reward scale finds no statistically significant preference between the post-trained and pretrained models.

General-purpose robot agents must learn from experience, transfer to new tasks, and act efficiently. Code as Policies (CaP) methods generate and repair programs at runtime, incurring latency and entangling reusable mechanisms with task-specific decisions. We introduce RACaP, an agentic framework that moves coding to evolution and uses a Reasoning-and-Acting (ReAct) loop to call frozen, typed Policy APIs at deployment. A two-phase strategy combines capability curriculum learning with autonomous self-evolution to improve the APIs, the ReAct harness, and experience memory. The APIs encode reusable physical mechanisms while exposing arguments for runtime adaptation. ReAct combines task-specific working memory, long-term experience memory, and visual feedback to select actions, verify outcomes, and recover from failures without modifying source code. RACaP achieves 54.4% success on LIBERO-90, 45.0% on zero-shot LIBERO-PRO, and 46.0% on LIBERO-Long, compared with at most 4.0% for CaP baselines on long-horizon tasks. On LIBERO-PRO, it achieves 2.5 times the success rate of CaP baselines and a 1.9-fold speedup in median policy time. For efficient on-robot deployment, rejection-sampled fine-tuning distills GPT-5.6 ReAct decisions into Qwen3-VL-8B-Instruct, yielding a 13.2-fold per-decision inference speedup and reducing repeated physical calls from 16 to 4. These results show that separating reusable code from runtime decisions supports continued evolution, effective transfer, and efficient long-horizon control.

Foundation vision-language models (VLMs) understand objects, instructions, and spatial relations, yet translating this capability into robotic manipulation remains difficult. Vision-language-action (VLA) models require extensive demonstrations and may compromise pretrained understanding, while direct RGB-only VLM control is costly and strongly dependent on model capability. We introduce Robo-Harness K1, a robot-use agent (RUA) framework that exposes perception as tools. The agent queries calibrated depth, persistent visual anchors, spatial measurements, and grasp hypotheses, then selects generic motions from the returned evidence. This interface makes 3D geometry accessible without changing the VLM architecture or training a depth encoder. On matched LIBERO-PRO tasks, Gemini 3.7 Flash with K1 reaches 77.8% accuracy, surpassing GPT-6 Astra's 61.1% with an RGB-only harness; K1 further improves Astra to 88.9%. Without target fine-tuning, Gemini with K1 transfers to three RoboSuite arms and dual-arm RoboTwin tasks. On RoboTwin, it achieves 32.0% on Easy and 28.0% on Hard, showing resilience to visual and environmental perturbations. K1 also produces tool-call traces aligned with next-token training. A Qwen3.5-9B student trained on only 107 teacher episodes reaches 44.2% accuracy on new initial states versus 30.2% for OpenVLA, and 13.9% on held-out task conditions versus 0.0% for OpenVLA. These results suggest that perception-augmented RUAs offer a promising route to sample-efficient, generalizable robotic policies that leverage VLM capabilities through an accessible tool interface.

Online handwriting provides a non-invasive and low-cost behavioral biomarker for Alzheimer's disease (AD) detection, as it reflects both cognitive planning and fine motor control. Existing handwriting-based AD detection methods usually rely on global trajectory features or whole-sample representations, which can be strongly affected by individual writing style, task-specific variation, and acquisition noise. In this paper, we propose NormPaST-Risk, a healthy-normative Paper-Air selective trajectory state-space risk network for interpretable AD detection from online handwriting. Instead of treating the entire trajectory as a single holistic representation, our method reformulates AD handwriting detection as local disease-relevant segment discovery. Specifically, a multi-scale temporal encoder captures stroke dynamics at different temporal resolutions, while a selective Paper-Air state-space encoder models long-range handwriting progression and distinguishes on-paper motor execution from in-air planning and transition behaviors. To explicitly characterize abnormal deviations, a healthy normative branch learns normal handwriting dynamics from healthy controls, and a task-aware multi-expert segment-risk module estimates segment-level AD risk calibrated by hidden-state changes and normative deviations. A weakly supervised segment-level objective further enables high-risk segment discovery without manual segment annotations. Experiments on the DARWIN benchmark demonstrate that the proposed framework achieves superior AD/HC classification performance compared with existing methods. Moreover, the discovered high-risk segments can be projected back to the original handwriting trajectory, providing interpretable evidence associated with AD-related handwriting variations.

Flow-matching Vision-Language-Action (VLA) models have emerged as a potential solution for generalist robot control, designed by combining a pretrained Vision-Language Model (VLM) backbone with an action expert that generates continuous robot actions. While these models exhibit impressive capabilities, due to their very high number of parameters, their computational requirements are often prohibitive for robotics control. To mitigate these inefficiencies, existing methods predominantly skip VLM backbone layers with early exits or reduce denoising steps, while leaving action expert depth untouched. We propose a framework that exposes backbone depth $V$, action expert depth $A$, and denoising steps $D$ as three jointly configurable compute axes in a VLA. Starting from a pretrained VLA, we attach lightweight Exit Transformers (ET) at intermediate depths in both the backbone and the action expert, trained to distil the last layer of the policy into each exit. Furthermore, we introduce a KV Cache synthesis mechanism that manages the missing keys and values of the skipped backbone layers, allowing the action expert to exit deeper than the backbone. Finally, we show that the optimal compute budget is task-dependent, with different tasks benefiting from different axes and depths. Notably, our method does not require training the original policy from scratch, and for each exit, it increases the number of parameters by only $2.1\%$ for SmolVLA and $4.1\%$ for $π_{0.5}$. We validate our approach across two flow-matching VLAs (SmolVLA, $π_{0.5}$) and two benchmarks (LIBERO, Meta-World), revealing complementary effects: $V$ and $A$ respectively reduce FLOPs and latency, while $D$ improves both. Our joint configurations $(V,A,D)$ reduce latency by $79.2\%$ and computation (FLOPs) by $31.8\%$, while improving mean success rate by $5.6\%$.

The acoustic front-end determines which forensic cues a speech deepfake detector can exploit. The wavelet scattering transform (WST) provides stable multiscale coefficients with explicit coordinates, yet direct flattening obscures the parent relation between paths. We introduce WST-Graph, reconstructing these paths as a sparse modulation-carrier grid for an AASIST graph backend. Modulation-level normalization and length-aware adaptive local attention pooling produce fixed relative-time representations while retaining the acoustic axes before learned adaptation. This yields a waveform-to-graph interface with a fixed, parameter-free WST. Our configurations remain competitive with AASIST while using approximately 60% fewer trainable parameters and show clear gains on selected out-of-domain benchmarks. These results underscore the value of preserving parent-child relations within the carrier-modulation topology when constructing a compact, physically grounded interface for graph-based speech deepfake detection. Code will be released at https://github.com/saki-ciallo/wst-graph.

This paper presents BanglaTurn, a corpus for end-of-turn detection in Bangla conversational speech, and a model trained on it. The corpus holds 35,374 samples of 3 to 15 s of podcast speech, labelled for turn state by combining speaker diarization with an LLM pass, with every label then checked by a human annotator. The model pairs a Whisper encoder with task-specific classification heads. On a class-balanced test set drawn from a held-out podcast, it reaches 84.33% accuracy (95% CI 80.3 to 88.1) against 69.28% for the Smart-Turn v3 baseline, and lowers the false negative rate from 51.57% to 7.55% at the cost of a higher false positive rate. We report what encoder layer fine-tuning, multi-scale pooling and INT8 quantization each contribute, and latency stays within 165 to 191 ms end to end on CPU.

Science, technology, and innovation policies are crucial for competitiveness, yet their diversity and scale make them difficult to map and monitor consistently. Existing approaches rely heavily on manual survey efforts, which are costly and challenging to scale across countries. Large language models (LLMs) enable new possibilities for extracting and structuring information from long and unstructured policy documents. This paper presents an application of LLMs as "AI respondents" for generating structured survey responses from policy texts. We develop a data extraction pipeline based on long-context in-context learning to map information from public web sources into predefined survey categories, including policy instruments, target groups, and thematic areas. The pipeline integrates a validation step using a secondary LLM to assess relevance and evidence, alongside comparisons with human-provided responses. Using a multi-country dataset, we evaluate the alignment between LLM-generated and human-generated outputs through overlap measures and cross-validation. Results show that LLMs achieve high agreement for structured indicators (84-95%), while differences remain in free-text fields, where models tend to provide more detailed procedural descriptions. These findings highlight the potential of hybrid human-AI workflows for policy monitoring, improving both efficiency and scalability while maintaining the need for human validation and contextual interpretation.

Multi-agent large language models (LLMs) have become ubiquitous in applied AI, yet their theoretical foundations remain surprisingly understudied. Viewed through the lens of multi-agent systems theory, several shortcomings come to light: a lack of social intelligence, the absence of coordination mechanisms among agents, unknown emergent behavior, and interactions between agents that are bounded by natural language. We address two of these gaps: the absence of social behavior and the lack of mechanisms for inter-agent coordination. We introduce Epistemic Probabilistic Language Agents (EPLA), a neuro-symbolic architecture for multi-agent coordination under uncertainty. A Symbolic Guard provides structured diagnostic feedback. The LLM generates typed actions, and the Guard controls their execution against an authoritative symbolic state. We formalize the epistemic layer in a gossip testbed through epistemic lottery gossip models, which combine view-based call histories with agent-indexed probability weights. We argue that implementing such a formalism can address shortcomings of agentic LLMs.

Explicit geometric references offer a direct way to structure self-supervised representations. Existing adversarial distribution-matching formulations, however, require costly encoder-critic optimization. We introduce Flow-Based Distribution Matching (FBDM), a non-adversarial framework that learns this reference-directed geometry through spherical conditional velocity regression. An ETF-inspired reference allows its number of components K' to exceed the auxiliary flow dimension d* while retaining structured geometric separation. We assign both augmented views of each image to the same target, while limiting how many images each reference center can receive. An explicit alignment loss further pulls the two views' representations closer together. Experiments across benchmarks ranging from CIFAR to ImageNet show that FBDM achieves performance nearly on par with DM and remains competitive with existing SSL methods. Matched training-cost comparisons show a 1.48- to 1.83-fold speedup over DM with a negligible increase in GPU memory usage. We also provide a theoretical explanation for the usefulness of the learned representations: under stated conditions, we bound the downstream misclassification rate in terms of the FBDM pretraining loss.

Event cameras provide a high dynamic range and preserve brightness-change cues in lighting conditions where conventional RGB frames may be noisy or saturated. To benchmark event-guided restoration across a broad illumination range, we organized the SEE Challenge 2026 with the Event-Based Multimodal Vision Workshop at ECCV 2026. The task conditions restoration on one or more RGB frames, synchronized events, and a scalar target-brightness statistic provided by the organizers. It uses SEE-600K, which contains 610,126 image-event observations from 202 real-world scenes spanning low-light, normal-light, and high-light conditions with illumination variations of up to 1,000$\times$. The challenge follows an open-system protocol: participants may use different temporal contexts, architectures, pretrained weights, test-time augmentation, and post-processing strategies. PSNR determines the ranking, and SSIM is reported as a secondary metric. Around 70 teams registered interest and 15 valid CodaBench submissions were received. Six distinct teams completed organizer-side identity and technical verification, provided method descriptions, checkpoints, inference code, and instructions, and are included in the verified open-system ranking reported here. Beyond the ranking, this report analyzes exposure subsets, semantically distinct test cases, a shared failure pattern, system design choices, and inference strategies. The top systems obtain closely spaced average scores, while the best-performing method varies across cases and metrics; under severe underexposure, all verified systems retain visible local errors.

Consumer-facing skincare AI must coordinate visual evidence, product information, tool use, and user-facing actions within explicit evidence and safety boundaries. This study evaluates SkinAgent AI, a non-diagnostic multimodal framework that combines visual concern routing with grounded and auditable LLM-based orchestration. The architecture includes routing for Acne, Pores, and Wrinkles; photograph-based skin-type estimation; count-informed ordinal acne-severity support; typed tools; database-grounded recommendation and action functions; deterministic safety, privacy, and evidence checks; approval before state-changing actions; and structured trace and replay mechanisms. Visual-model performance and system-level agent behavior were evaluated separately. Across three seeds, the skin-condition routing model achieved 99.84% +/- 0.07% accuracy. Skin-type estimation achieved 88.85% accuracy, while count-informed acne-severity support achieved 84.59% accuracy with a quadratic weighted kappa of 0.9076. On a locked but non-independent 240-case system benchmark, intent accuracy was 80.00%, exact tool-set match was 62.92%, and strict task completion was 47.08%. No violations or successful cross-user leakage events were observed in the finite safety and privacy test suites. Tool-selection errors, incomplete grounding of product attributes, and unreliable failure fallback nevertheless remained. These findings support the feasibility of bounded, database-grounded, and traceable agent orchestration for non-diagnostic skincare assistance. They do not establish clinical readiness, external generalization, formal privacy guarantees, or universal safety. Independent validation, expert assessment, robustness and fairness testing, and prospective evaluation in real-world settings remain necessary.

This paper presents a grasping model for cloth manipulation specifically tailored to ease the deployment of robotic control methods. The model is robust, fast and easy to implement avoiding at the same time contact and friction considerations between the gripper and the cloth in favor of simple positional constraints. The gripper is described by its pose, jaw state, and an attached grasping volume. Two kinds of grasping volumes are considered: an axis-aligned box to simulate a pinch grasping and a square pyramidal volume to simulate point grasping. When the gripper closes, the discrete cloth positions lying inside this volume are selected, stored in the local gripper frame, and then transported with the gripper motion. A simple squeezing step is also included to progressively move the selected cloth positions toward the center of the grasping region, avoiding an instantaneous displacement at closure. The model can be used in any simulator as it only requires access to discrete cloth positions and a mechanism for imposing target positions as constraints. We implement our grasping model in conjunction with a constraint-based inextensible cloth simulator, where grasping is implemented as moving positional equality constraints coupled with stretch, shear, collision, and table contact projection steps. The same gripper trajectory is applied on a robot arm to fold a real piece of cloth, serving as a simple bridge between simulation and physical cloth manipulation and showcasing the realism and practicality of our idealized grasping model.

Retrieval-Augmented Generation (RAG) mitigates knowledge obsolescence and factual hallucination in large language models by introducing external context. However, when retrieved knowledge conflicts with the model's internal parametric knowledge, the model may either blindly follow misleading context or incorrectly rely on parametric knowledge, leading to unreliable responses. To address this issue, this paper proposes TRACE (Debate-TRace and Answer-Completeness rEgularized fine-tuning), a robust fine-tuning framework for RAG under knowledge conflicts. First, we propose a fine-tuning method that leverages multi-agent debate traces to extract correct candidates, incorrect candidates, and answer-shift patterns, providing fine-grained supervision for reliable knowledge-source selection. In addition, we design an answer completeness regularization mechanism to alleviate empty, overly short, and prematurely terminated responses via answer-tail token reinforcement and premature termination suppression. The fine-tuning objective combines correct-answer supervision, incorrect-candidate suppression, answer-tail token reinforcement, and premature termination suppression, enabling the model to use reliable external context, resist misleading or irrelevant retrieved content, and fall back to parametric knowledge when retrieved evidence is unreliable. Experiments across multiple knowledge-conflict scenarios and datasets show that TRACE improves robustness against misleading retrieved knowledge and reduces incomplete answers. These results demonstrate that multi-agent debate traces and answer completeness regularization jointly enhance knowledge-source selection, conflict robustness, and answer quality in RAG models. Our code is available at https://github.com/PHD-lanyu/TRACE.

Time series analysis is fundamental in domains such as finance, healthcare, and meteorology. Real-world time series often exhibit multiscale characteristics shaped by diverse latent factors, resulting in intricate temporal patterns and rich frequency structures. However, existing approaches typically focus on either frequency-domain decomposition or time-domain pattern extraction in isolation, neglecting their joint structure. This decoupled modeling limits representation expressiveness and undermines performance in tasks requiring simultaneous temporal and spectral reasoning. To address this gap, we propose m-WCN, a novel end-to-end deep learning framework that neuralizes multi-wavelet decomposition for joint extraction of temporal patterns and frequency components. By approximating the classical GHM multi-wavelet transform with trainable convolutional operators and enforcing orthogonality constraints, m-WCN produces interpretable multi-resolution representations. Built on this foundation, we introduce two task-specific architectures: TFBC for time series classification, which boosts discriminative features across frequency scales, and FTB for forecasting, which ensembles frequency-aware predictors. Extensive experiments on 64 UCR datasets and seven public forecasting benchmarks demonstrate the effectiveness of our approach. Built on the neuralized m-WCN, our TFBC and FTB outperform various baseline models across diverse datasets, achieving average improvements of 19.97% in classification and 19.92% in forecasting tasks.

Robot manipulation policies trained through imitation learning inherit not only the demonstrated behavior but also the conservative execution tempo of robot demonstrations. Existing acceleration approaches can execute faster than the original demonstrations, but determine the appropriate acceleration primarily from robot-side information or a predefined set of tempo factors, leaving open how to obtain a task-appropriate reference for how fast each manipulation phase should progress. We introduce EgoSpeedUp, a framework that uses human manipulation as temporal supervision for robot imitation learning. Our key insight is that human demonstrations naturally reveal task-appropriate, phase-wise manipulation tempo. Given slow robot demonstrations and human demonstrations of the same task, EgoSpeedUp aligns corresponding manipulation phases, estimates their relative execution tempos from multiple human demonstrations, and transfers the resulting phase-wise tempo by retiming the robot demonstrations. The retimed demonstrations are then used for standard behavior cloning, allowing the robot to retain its executable manipulation behavior while learning to perform it at a human-informed tempo. Across two real-world manipulation tasks, EgoSpeedUp improves the task success rate by an average of 25 percentage points (pp) while reducing successful execution time by 36.5%. These results demonstrate that human manipulation tempo provides an effective temporal reference for learning faster and more reliable robot policies.

In open-ended problem solving, collaborators often rely on discussion to surface concerns, challenge perspectives, and refine shared work as it evolves. While AI agents are increasingly used as discussion partners, existing multi-agent systems place a heavy burden on users to initiate and carefully orchestrate the discussions. We present DocuTeam, a mixed-initiative multi-agent discussion system in which both users and agents can initiate and steer conversations. Agents monitor document changes to proactively start and redirect discussions as the work evolves, while users can flexibly shape the conversation or adopt agent ideas. In a within-subjects study (N=20), participants using DocuTeam produced outcomes rated significantly more novel, relevant, and specific than with a baseline without any increase in cognitive load. Rather than using agents for one-off idea sourcing, participants engaged in an iterative refinement loop in which document changes prompted agent reactions, which led users to revisit and further develop their work.

In this work, we focus on photorealistic sign avatar modeling, which is crucial for effective communication with the Deaf community and is characterized by complex hand gestures and nuanced facial expressions. To this end, we introduce MVSign, the first multi-view Chinese sign language dataset co-designed with Deaf experts, featuring diverse gestures and rich annotations. For precise SMPL-X annotation, we develop a hybrid fitting pipeline that produces accurate body, hand, and facial parameters and can also be applied to the monocular setting. Building on MVSign, we propose a decoupled sign avatar representation that isolates body, head, and hand components to capture complex articulations, together with a motion-aware sampling strategy to handle motion blur and balance gesture diversity. Extensive experiments demonstrate that our method achieves high-fidelity visual results on MVSign, particularly in detailed hand and facial regions, and generalizes well to in-the-wild monocular sign language videos. Project page: https://naaapi.github.io/PHOSA.

Visual software often needs a decision over supplied alternatives rather than a generated explanation. We present PixelJev, a native-image decision interface that maps an image, a task instruction, and a runtime candidate set to a structured choice and candidate-conditioned probabilities using small open multimodal models. Its initial realization unifies recognition and multiplechoice visual question answering through an existing language-model readout, with separately evaluated options for frozen inference, language-side adaptation, and held-out calibration. Across seven benchmark evaluations, 64-shot source adaptation raises Pets accuracy from 60.13% to 92.40% across optimization seeds and transfers to natural resampling, new texture labels, and A-OKVQA without target fitting, while frozen inference already supports both VQA tasks. A matched prompt-only follow-up on Pets and ScienceQA attributes the large Pets gain to adaptation and identifies a narrower output validity benefit of candidate readout in adapted VQA. Specialist DINOv2 probes remain stronger on source recognition, frozen 4B is stronger than adapted 2B on DTD and ScienceQA, and accuracy gains do not ensure calibrated target probabilities. These findings establish a working starting point for general-purpose visual decision models and identify the remaining requirements: schema robustness, cross-family transfer, and reliable use of visual evidence.

Neural networks are typically adapted by computing gradients and updating model parameters. We investigate whether task-specific adaptation can instead emerge from a meta-learned self-organising process that requires no gradients at adaptation time. We instantiate this idea with a Neural Cellular Automaton in which locally interacting recurrent cells maintain both a recurrent state and a fast associative memory. During meta-training, backpropagation is used to learn the recurrent dynamics together with how the memory is read and written. Once training is complete, the slow model parameters remain fixed, and online adaptation occurs only through cellwise memory updates driven by local prediction errors and a delta rule. We evaluate whether the learned mechanism can adapt to semantically distinct held-out classification tasks. A single pass over the support data produces substantial improvements in held-out performance without gradient computation or parameter updates during adaptation, and the mechanism remains effective across large changes in the number of examples processed jointly. These results show that task-specific adaptation can be achieved through explicit fast-memory updates while keeping the slow model parameters fixed.

Temporal knowledge graph forecasting aims to infer future relational facts from the temporal structure of observed events. Existing forecasters mainly summarize history through entity states, relation states, paths, or exact recurrence. These views often miss pair-specific transition evidence, that is, the way prior relations between the query actor and a candidate change the odds of the target relation. We introduce BridgeMem, which estimates this quantity as a residual added to the log scores of a frozen full-vocabulary forecaster. For each candidate, BridgeMem retrieves the pair's events that strictly precede t, encodes their relations, directions, and lags, and converts them into a likelihood-ratio correction. A support-adaptive empirical-Bayes reader trusts exact transition counts where they are abundant and backs off to a learned attention estimator where they are sparse. The backbone's own uncertainty gates the correction, so confident queries and candidates without dyadic history are left unchanged. On five benchmarks, BridgeMem improves on the strongest of nine baselines from 2021--2026 in all 20 filtered MRR and Hits@{1,3,10} comparisons, with MRR gains of 0.0213, 0.0164, 0.0216, 0.0112, and 0.0028 over the best prior result. These results show the value of explicit dyadic transition modeling.

We develop a multi-label Polish content-safety classifier by fine-tuning allegro/herbert-base-cased (124M) across five categories (hate, vulgarity, sexual content, crime, self-harm) using a Focal + R-Drop objective, and evaluate the resulting model against Bielik Guard (Sójka) on the shared out-of-distribution Gadzi Język benchmark. Both systems are given per-category threshold tuning on the same calibration split. Under that matched protocol our model holds a small but statistically significant lead in micro F1, while an apparent macro-F1 lead does not survive: it was an artifact of comparing a tuned model against an untuned one. We also report what that micro figure is worth. Because Gadzi Język is 97% crime-positive, a classifier that flags crime on every input and nothing else already scores 0.910 micro F1 on the same test split, so micro separates neither system from a degenerate strategy and macro is the column that does. Per-category and per-protocol figures are reported in Section 4. The residual out-of-distribution gap is one of calibration rather than discrimination. Ranking quality stays high while positive probabilities collapse, and per-category temperature scaling recovers the loss where Platt scaling and isotonic regression do not. That recovery turns out to be conditional on the calibration set containing safe text. Gadzi Język contains almost none, so thresholds fitted on it flag crime on every safe input, and a balanced refit buys a deployable operating point at the cost of adversarial recall. We report both operating points rather than only the flattering one. Two changes that are standard practice, per-class cost-sensitive weighting and mean pooling, each raise in-distribution macro F1 while lowering the out-of-distribution figure, which indicates that robustness has to be selected for directly rather than inherited from in-distribution accuracy.

CAR-bench evaluates whether tool-using agents stay reliable under real-world uncertainty, executing every tool inside the evaluator so that each tool-result exchange is a separate agent round-trip. A conventional next-action agent can batch parallel tool calls, but a chain of dependent calls costs it one model call per round of results. We present a coroutine-bridge harness in which the model's only action is to emit a Python program that blocks and resumes in place across evaluator tool exchanges. This decouples model invocation from tool round-trips: on the public test split the agent uses a median of two model calls against seven agent turns per task, resolving a full multi-turn task in a median of 1.8 s of model latency on Cerebras gpt-oss-120b. Because the action surface is executable code, deterministic CAR-bench policies are encoded directly as logic in the tool layer rather than as prompt rules, enforcing compliance at zero reasoning cost. On the official hidden evaluation the harness won Track 2 with 60.0% Pass^3, 4.5x the organizer baseline, at the lowest estimated cost and the fastest median task latency (3.14 s) of any entry scoring above that baseline; the same unchanged harness reproduced an identical 60.0% Pass^3 on GPT-5.5 in the Open track, matching frontier-model agents. A single static prompt, appended with per-task state at the tail, stays byte-identical across calls and across tasks: the frozen submission prompt served 78% of input tokens from cache (86.6% across its warm tail), against 73% over a three-week development corpus in which prompt edits repeatedly reset the cache. This compounds the few-call design into a small fraction of nominal input compute.

Text image super-resolution (TSR) aims to recover visually faithful and readable text under unknown degradations. Existing diffusion-based methods typically rely on multi-step prediction of either the high-resolution image or its text prior, resulting in prohibitive computational cost and inference latency. More critically, an erroneous text prior may be repeatedly injected into the denoising process, causing image and text predictions to reinforce each other and progressively amplify an early recognition error into a sharp yet semantically incorrect character. To address these limitations, we propose TOLA, a Text-aware One-step Latent Adaptation framework without iterative image-text diffusion. TOLA consists of two key modules. First, a confidence-weighted text conditioning module constructs the semantic condition only once and suppresses unreliable OCR predictions before they contaminate image reconstruction. Second, a lightweight latent residual correction module explicitly estimates and corrects the structured residual errors to recover missing or distorted stroke details. Extensive experiments demonstrate our state-of-the-art performance across all evaluation metrics on both CTR-TSR-Test ($\times 4$) and RealCE-200 benchmarks. It is worth noting that our TOLA consistently surpasses existing diffusion-based TSR methods by at least 2.72 dB in PSNR on CTR-TSR-Test.

We find that language models can transfer capabilities through task-unrelated text. Post-training typically improves language models using task-specific data. Prior work on subliminal learning shows that information about these updates can pass through unrelated generations, but has largely focused on traits or preferences using extensive teacher outputs. We introduce Active Taskless Distillation (ATD), which achieves capability transfer using only a single word from the teacher per prompt. ATD probes the behavioral shadow of post-training by selecting prompts where the teacher and student's shared public ancestor is nearly indifferent between two ordinary words. A student initialized from this ancestor learns solely from the resulting prompt-word pairs, without target-task examples, teacher logits, or teacher parameters. In the primary coding experiment with Qwen2.5-1.5B, 5,664nses yield a 5.34 pp gain on HumanEval+ over an exact nuisance-matched control thadisrupts prompt-resperiments showtransfer in scientific knowledge, commonsense reasoning, and reading comprehensins across additional model generations, sizes, and families. Functional analyses show that the learned sid composable, andthat its strength tracks the teacher's update strength.

End-to-end event extraction remains challenging for large language models as it requires simultaneous identification of event triggers, classification of event types, and extraction of schema-grounded argument spans. We present EAGER, a reinforcement learning framework for generative event extraction that combines fine-grained verifiable rewards with Schema-Contrastive Advantage Estimation to alleviate advantage collapse under sparse binary rewards. Our reward design explicitly targets structural validity, extraction accuracy, groundedness, coverage, over-generation, and span precision. Experiments across seven benchmark datasets show that EAGER consistently outperforms prompting, supervised fine-tuning, and prior reinforcement learning baselines, achieving a substantial improvement over the strongest prior method. Results demonstrate that task-aligned verifiable rewards and contrastive advantage estimation substantially improve structured extraction.

Lip synchronization aims to generate visual lip dynamics that align precisely with speech audio. Despite the high generation quality of diffusion models, they often struggle in complex scenarios and suffer from prohibitive inference latency, limiting real-world deployment. We present ComplexSync, a unified diffusion-based framework that enables real-time, high-fidelity lip sync under complex conditions. First, we introduce a dual-stream joint training strategy to mitigate information leakage from reference frames while preserving natural dynamics. Second, we develop a distillation-based acceleration scheme for single-step denoising, achieving a throughput of over 70 FPS. Third, we propose a relational alignment loss that leverages structural priors from Vision Foundation Models (VFMs) to enhance robustness against complex scene factors. Furthermore, we present the first benchmark specifically designed for complex lip synchronization, comprising over 200 challenging video sequences and specialized metrics. Extensive experiments demonstrate that ComplexSync achieves state-of-the-art performance across both standard and complex scenarios while enabling real-time inference.

Large language models can decompose mobile-manipulation goals into long action sequences, but the resulting plans remain reliable only while their world context is current. A fixed scene description becomes stale when objects are discovered, moved, or completed while retaining every observation instead produces a growing history with redundant and conflicting state. To resolve this tension, we present an LLM-guided planning framework ADM-Planner with attention-enhanced dynamic memory (ADM). Persistent workspace knowledge is separated from object-centric state, asynchronous observations and action outcomes update that state, and a bounded retriever exposes only the entries that can affect the next decision. The LLM replans when an update invalidates the remaining plan. Across 1,500 task-simulator episodes, the proposed ADM achieved 100% full-task success in the 14-container noisy dynamic setting, compared with 62% for static memory and 97% for unfiltered dynamic memory, while reducing the context-size proxy by 95.8% relative to the latter. In a six-episode live GPT-5 Mini planner, both dynamic memory variants completed every mission, while ADM reduced provider-reported input tokens by 14.4% and mean planner calls from 7.0 to 6.0. A separate 60-trial PyBullet study retained 100% success for ADM, compared with 50% for static memory. Finally, the mobile manipulator with ADM-Planner completed various missions in indoor and outdoor physical experiments while incorporating targets revealed after execution began. The results show that selective state maintenance with ADM, rather than prompt history alone, is a practical basis for long-horizon planning in changing environments. Project page: https://xjp99v5.github.io/ADM-Planner

Vision-language-action (VLA) models offer strong local control and instruction following but often struggle with long-horizon tasks requiring persistent memory and planning. Task harnesses provide persistent context for agent reasoning by retaining task history and tracking progress across execution stages. To bring these complementary capabilities together, we introduce AdaHVLA, an adaptive harness that refines code-based coordination policies through robot experience to better align agent reasoning and memory with VLA execution. Its decoupled multiagent adaptation process separates evidence analysis, harness revision, and behavioral assessment into distinct working contexts, using testable coordination hypotheses to guide revisions and subsequent rollouts to assess their predicted effects. A stateful revision graph links execution evidence, hypotheses, revisions, and observed effects, preserving alternative harnesses and adaptation memory to guide refinement across repeated attempts and continued adaptation across tasks and environments. In simulation, AdaHVLA raises mean test success on NaVILA-LH from 22.5\% to as high as 57.5\% and improves manipulation test success across three VLA backbones by up to 30.8 percentage points over the initial harness. Real-world deployment further illustrates how the adapted policies support stable execution across task stages.

Mobile robots require robust, real-time fault detection capable of continuous adaptation on constrained edge hardware. While deep time-series models excel at unsupervised anomaly detection, their computational cost prohibits high-frequency onboard execution. This paper bridges this gap via a Teacher-Student distillation framework. An offline foundation model (TSPulse) generates pseudo-labels from unlabeled time series augmented with fault injections. A lightweight MiniRocket Student, adapted with a Recursive Least Squares estimator, approximates this complex decision boundary to execute real-time inference onboard. Evaluations on the TSB-AD benchmark and a physical mobile robot demonstrate the Student achieves a 4.30 ms CPU inference latency. During real-world domain shifts, online adaptation enables the Student to recover from unseen mechanical degradation, improving VUS-PR scores from 0.26 to 0.75 without catastrophic forgetting. Crucially, an uncertainty-guided active learning strategy minimizes operator cognitive load, requesting sparse interventions only when encountering novel fault distributions. These results validate the deployment of state-of-the-art anomaly detection on resource-constrained robotics through offline-to-online distillation.

Sensitive information is defined by domain and intent, not a universal category, yet redaction systems such as privacy filters and named-entity recognizers fix a taxonomy at training time, requiring retraining for each new domain. We introduce ASIRF (Agentic Sensitive Information Redaction Framework), which retrieves domain-specific definitions based on the input's domain from a flexible knowledge base at inference time, needing no retraining to adapt. Two architectures, a three-call multi-agent pipeline and a single-agent variant, are evaluated across ten small open-weight models and eight datasets, including out-of-distribution fictional domains, against the OpenAI Privacy Filter (OPF) as a trained-classifier baseline. With only a few dozen expert-authored definitions per domain and no training data, ASIRF's recall exceeds OPF's in 68 of 80 model-domain combinations (85 percent), by at least one of the two architectures, with shortfalls confined mostly to OPF's training-distribution domains.

The human wrist exhibits adaptive stiffness modulability: joint stiffness anisotropy can be actively regulated through muscle co-contraction. This functionality is essential for stable manipulation, yet the underlying morphological factors remain unclear. To identify these factors, we developed an anatomically accurate anthropomimetic soft robotic forearm comprising eight independently movable carpal bones interconnected by ligaments, 22 actuated muscles, and compliant fingertips. We measured wrist joint stiffness under four muscle activation patterns across three skeletal configurations: anatomically normal carpal bones, a fused proximal carpal row, and a geometric ellipsoidal skeleton. The stiffness ellipse exhibited low stiffness along the dart-throwing motion (DTM) direction when finger muscles were activated, but high stiffness along the same direction when wrist and finger muscles were activated simultaneously. These results agree with previously reported human measurements, demonstrating that precise anatomical replication reproduces human-like stiffness modulability. Fusing the proximal carpal row eliminated the low DTM-direction stiffness under finger muscle activation, while the geometric ellipsoidal skeleton showed poor stiffness ellipse reorientation across all conditions. Carpal bone motion analysis revealed significantly opposing coupling patterns between wrist and finger muscles at the proximal carpal row, accompanied by a consistent but non-significant trend at the midcarpal joint, providing a mechanical explanation for this modulation. These findings demonstrate that carpal bone morphology plays a dominant role in human wrist stiffness modulation and provide design principles for humanoid robot wrists.

We propose the Representation World Model (RWM), which learns states, transitions, and executable plans directly in representation space. Unlike existing world models that typically learn latent representations together with explicit dynamics models and perform planning through search, optimization, or policy-based prediction, RWM directly incorporates planning into the learned representation geometry. RWM learns the representation geometry by applying inverse-dynamics supervision locally along latent paths constructed from endpoint representations, requiring these paths to preserve task-relevant state and transition information. At inference, planning is performed by directly constructing a latent path between the current and goal representations, with inverse dynamics used to recover the corresponding actions, without recursive rollouts or action-space search. Experiments on continuous-control benchmarks demonstrate the effectiveness of RWM for direct planning, while results on robotic manipulation further show its potential to extend to more complex embodied control tasks. These results suggest that planning directly in representation space provides a promising alternative to conventional world-model planning.

We introduce Spot, Separate, and Enhance (SSE), the first multimodal, user-guided generative model for audio remixing and enhancement. SSE enhances video content by rebalancing the audio, removing unwanted audio sources, and reducing reverberation, guided by both video and textual descriptions. To support its training and evaluation, we propose DegradedMix, a new dataset built on the audio remixing benchmark MuddyMix. We also adopt evaluation metrics from generative modeling, which better capture the creative nature of remixing than standard reconstruction-based metrics. SSE outperforms existing baselines in both controllability and remixing quality, as shown by extensive experiments. Project page: https://sse-ai.notion.site

Banking assistants must use account-specific information to answer requests and, in many cases, take actions through tools. Evaluating only the final response misses important errors. An assistant may ask for information it already has, rely on stale context, select the wrong account, or write an invalid value after stating the correct one. We introduce IndicBankBench, a 799-case benchmark for Indian retail banking spanning five operational domains, a capability/refusal domain, and twenty primary axes. Cases are evaluated at four stages: safety, action and tool use, response adequacy, and advisory quality. Tool use and most safety checks are deterministic. A narrow resolver handles only ambiguous confirmation-before-write cases, while a separate LLM judge evaluates semantic response adequacy. We run every case three times and report strict pass^3, which requires success on all trials. Across the eleven evaluated models, strict reliability ranges from 43.7% to 58.2%, whereas at-least-once success ranges from 60% to 74%. This gap shows that at-least-once success can overstate dependable banking behavior. The case-level diagnostics also distinguish systems that ask unnecessary questions from those that act but fail to reconcile customer context or fully resolve the request. We release the cases, mock environment, and evaluation harness.

Physical AI aims to build embodied agents that perceive the world, understand and reason about it, and decide how to act. Yet the field has focused primarily on the last component: the action model that maps observations to low-level controls. The prevailing training recipe can erode the perceptual and reasoning capabilities needed for robust behavior, leaving even strong action models vulnerable to scene perturbations and long-horizon tasks. We introduce HarnessPAI, a model- and embodiment-agnostic Harness framework for Physical AI that treats code as the executable and evolvable interface that organizes the underlying action primitive. The framework separates two timescales: within a rollout, it executes open-loop at the program level, with a fixed program guiding and checking execution; across rollouts, it evolves closed-loop, using execution feedback to revise the program and distill failures into reusable skills. Across desktop robot arms, household robots, a robot vacuum, and a legged walking agent, HarnessPAI improves on both pure action models and code-as-policy baselines without retraining the underlying model: a 61.6-point gain over $π_{0.5}$ on LIBERO-PRO and a 27.2-point gain over WorldDreamer on RoboCasa atomic tasks. Once a program is selected, rollout execution requires no online high-level LLM deliberation. Beyond execution, the converged program is also a cheap and reliable expert-data collector, and fine-tuning $π_{0.5}$ on collected expert data lifts success rate on LIBERO-PRO by 38.8 points. Our results suggest that the frontier of Physical AI depends not only on stronger action models, but also on executable harnesses that integrate perception, task understanding and reasoning, and action execution into a unified, verifiable, and feedback-driven system. Website: https://darwin-agent.github.io/HarnessPAI

To achieve general-purpose autonomy over long horizons, a robot needs to maintain spatial environment information that supports a variety of tasks: geometry for planning and control, radiance for rendering and relocalization, and vision-language features for open-vocabulary grounding. Existing methods represent and estimate each modality separately, multiplying memory and compute cost while forgoing potential synergy among the representations. We develop OREN-X, an online mapping method that uses an octree in 3D space as a shared data structure for indexing and storing a multi-modal field, capturing geometric, radiance, and vision-language information. OREN-X provides efficient unified storage and retrieval of these data in explicit/implicit and full/compressed form. Our unified representation yields cross-modality synergy: SDF estimates are sharpened by occupancy and radiance, while GPU-based ray-octree traversal and octree query enable real-time rendering. We also use online dictionary learning to compress the vision-language features, shrinking them 3.7x below full per-vertex storage while raising the query accuracy. On Replica, OREN-X maps in real time (80+ fps for SDF and 30+ fps for all four modalities), improves near-surface SDF accuracy by 33% over single-modality baselines, and improves mean open-vocabulary 3D mIoU by 71% and mean accuracy by 61% over the best prior method.

Long-tailed chest X-ray classification requires visual representations that capture both common abnormalities and subtle, infrequent findings. We propose Med-AR-8B and Med-AR-2B, two radiology-native autoregressive vision-language models pretrained with structured reports, abnormality-focused text, and region annotations. We evaluate the transfer of their visual encoders to multi-label classification against contrastive, self-supervised, and supervised pretrained encoders, including Med-CLIP, CheXFound, EVA-Base, ARK, and BioViL-T, using a common ML-Decoder classification head. To assess fine-grained recognition, we also construct LLM-expanded, report-derived label sets for MIMIC-CXR and CheXpert. Across PadChest, MIMIC-CXR, and CheXpert, Med-AR-8B outperforms Med-CLIP in mean AUROC and AUPRC for head, medium, and tail findings. On MIMIC-CXR, it increases tail-label mean AUPRC from 0.1033 to 0.1441. Med-AR-2B achieves the strongest discrimination results on PadChest. Across the broader encoder comparison, a Med-AR variant achieves the highest mean AUROC and AUPRC in every reported prevalence group on each public dataset. Both Med-AR variants also achieve lower excess area under the risk-coverage curve than Med-CLIP on all three public datasets, indicating improved selective-prediction performance under the evaluated protocol. Internal results are metric-dependent, with Med-CLIP retaining advantages in overall and tail AUPRC and in selective prediction. These findings establish Med-AR as a strong pretraining recipe for long-tailed chest X-ray classification on the evaluated public benchmarks and demonstrate the value of assessing discrimination and selective prediction together.

Large language model agents increasingly rely on natural-language skills to solve complex tool-use tasks. However, such tasks often admit multiple valid solution paths, making it inappropriate to improve skills by forcing failed trajectories to match a fixed successful trajectory. Moreover, failed trajectories are rarely entirely wrong: an agent may first collect useful evidence and make meaningful progress, but later deviate into an erroneous suffix. We therefore argue that skill self-evolution should identify where productive problem solving begins to break down, rather than reflect coarsely over the entire failure. Based on this insight, we propose SkillPivot, a deviation-point-guided framework for skill self-evolution. SkillPivot detects the transition from a useful prefix to an erroneous suffix using execution validity, goal progress, and action diversity. A stronger teacher then continues from the same prefix and produces a successful alternative under the same interaction history. By contrasting the student's failed suffix with the teacher's successful suffix, SkillPivot generates localized skill updates while preserving already effective guidance. Experiments on ToolQA, LogicBench, and WildClawBench show that SkillPivot consistently outperforms competing skill-evolution methods, improves multiple agent models, and produces compact, transferable skill updates.

This paper investigates the joint optimization of the transmit precoder and the transmission/reflection coefficients of a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) to maximize the weighted sum rate (WSR) in a multi-user downlink. We propose a particle-swarm-assisted gradient meta-learning (PSA-GML) algorithm for this non-convex problem. The original problem is first equivalently transformed via an amplitude-split parameterization and a collapsed precoder representation, which automatically satisfy the energy-conservation constraint and reduce the search dimension. Particle swarm optimization (PSO) then performs a global search over the STAR-RIS coefficients to yield a high-quality, initialization-robust warm start, with the transmit precoder obtained in closed form. Departing from conventional alternating optimization (AO), a coordinate-wise long short-term memory (LSTM) meta-optimizer trained by first-order gradient meta-learning further refines the coefficients and precoder jointly, learning per-coordinate adaptive update rules from data. The meta-optimizer is trained offline and applied to unseen channels without further adaptation. Numerical results show that PSA-GML attains an 11.06 bits/s/Hz WSR at 10 dB with N=32 elements and K=4 users, exceeding AO by 13.1% (and by 6.2% even with multiple random restarts) and the random-phase scheme by 35.1%. In the interference-limited regime it reaches 83.9% of the hand-designed Adam refinement without manual hyper-parameter tuning, and it transfers zero-shot across regimes, indicating that the learned update rule captures the intrinsic WSR landscape structure.

Persistent memory lets language-model agents improve prompts and skills without updating model weights. We show that matching retrieval scope to certification scope enables these edits to support reliable repeated adaptation across recurring task families. We study frozen-model agents on ProcStream-RSI, a 12-round code-repair stream, using Orthogonal Regression Control (ORC), an execution-grounded gate for persistent skill edits. In an intervention that holds proposals and gate decisions fixed, retrieving each accepted skill only for its originating family raises mean hidden trajectory utility from 0.713 under global memory to 0.816 and changes harmful deployments from six of eight to none. In 27 paired randomized-order streams, Scoped-ORC improves mean trajectory utility by 0.063 [0.037, 0.094] over Global-ORC, accepts 63 rather than 12 updates, and produces multiple accepted updates in 19/27 streams, with 0/63 harmful acceptances. The global control reaches 0.713, below the static agent's 0.775, because locally valid edits can interfere with unrelated families. These results establish scope matching as a complementary control for persistent agent memory: certification determines whether an edit is supported, while retrieval scope determines where that evidence authorizes its use.

Direct On-Policy Distillation (Direct-OPD) transfers reinforcement-learning-induced policy improvements from a small model to a larger student by using the token-level log-ratio between post-RL and pre-RL checkpoints as dense supervision on the student's own rollouts. This transfer rewards the policy shift at every state, yet the log-ratio measures only relative change: it can stay fixed even as the probability mass that both checkpoints assign to the student's candidate tokens vanishes. Through an exact construction, we show that the Direct-OPD reward and its update can remain unchanged while the Jensen-Shannon divergence (JSD) and both KL directions between the checkpoints vanish with this mass, and we note that a small JSD bounds how much the teacher's behavior changed. Motivated by this analysis, we propose Selective Supervision for Direct-OPD (S$^2$D-OPD), which ranks student-sampled states by their teacher-reference JSD and masks Direct-OPD supervision at low-divergence states, retaining only the top 10% of states per response. Across two teacher pairs and four student models ranging from 1.7B to 8B parameters, S$^2$D-OPD improves held-out accuracy over dense Direct-OPD on AIME and HMMT benchmarks in seven of eight settings and matches it in the eighth, without extra forward passes. Our code is available at https://anonymous.4open.science/r/S2D-OPD-8868.

Repeated evaluation can estimate a benchmark score accurately while still requiring replication to certify narrow uncertainty. We characterize that requirement on a fixed grid of $M$ tasks with $L$ binary paths per task under the hard budget $(M+t)K$, where each path costs at most $K$ responses or episodes. For fixed $L \ge 3$ and $0 < α\le 1/12$, the optimal expected width on the worst pure cohort is $Θ_{α,L}([M(t+1)]^{-1/2})$ when every task is observed and $Θ_{α,L}([M(t+\sqrt{M})]^{-1/2})$ when omission is allowed. The lower bounds cover adaptive hard-budget policies, and fixed random-subset designs attain both rates through disagreement certificates. A joint mean/disagreement interval turns the task-covering law into practical finite-budget inference. In an equal-budget LiveCodeBench replay with 16 models, 880 tasks, and five outputs per task, the task-covering design reduces median point-estimation MSE by 87.0\% relative to pooled uniform sampling, while the Joint certificate produces narrower confidence intervals in 15/16 panels and reduces median interval width by 30.6\%. Finite-regime analyses identify task coverage as the effective choice at the evaluated scale and characterize how cohort size and within-task agreement determine the useful operating region. Together, the sharp laws and fixed-budget evidence make replication and task coverage explicit design variables for information-efficient repeated evaluation.

Reasoning-capable language models often produce long chains of thought when direct answers suffice, wasting inference compute. Many dual-mode models leave this choice to users. Automating it is challenging because routing targets evolve with the policy, initial mode preferences destabilize exploration, and sequence-level objectives entangle routing with response learning. We introduce CounterRoute, an online reinforcement-learning framework that jointly learns routing and modeconditioned responses in one shared policy directly from a native dual-mode checkpoint, without method-specific SFT warm-up. Paired current-policy counterfactual rollouts assign cross-mode credit only to the routing token, while within-mode GRPO trains response tokens. A paired-to-self-routed curriculum stabilizes early training with forced rollouts from both modes, then increases self-routed updates to improve autonomous routing. Across nine benchmarks, CounterRoute better balances accuracy and efficiency than heuristic and learned adaptive-routing methods. Relative to always-thinking checkpoints, it improves macro-average accuracy while reducing mean generated tokens by 51% for Qwen3-8B and 41% for Qwen3-14B. On instruction-following and commonsense benchmarks where direct answering is strong, think rates fall as low as 1% while response quality improves. Despite training only on math and instruction following, its routing behavior and response quality generalize to held-out coding, science, knowledge, and commonsense benchmarks.

Medical Named Entity Recognition (NER) for low-resource languages remains a challenging task due to high linguistic variability and a scarcity of domain-specific annotated corpora. This work presents a comprehensive empirical benchmark evaluating three fine-tuned transformer encoders-BanglaBERT, multilingual BERT (mBERT), and XLM-RoBERTa-against GPT-4o mini under zero-shot and few-shot prompting configurations for Bangla medical NER. In contrast to prior studies that evaluated large language models on limited subsets of only 50 samples, we conduct a large-scale evaluation across the full test set of 3,179 samples, providing statistically robust and reproducible baselines. Our fine-tuned XLM-RoBERTa model achieves an F1- score of 0.5959, establishing a new state-of-the-art and surpassing the previously reported best result of 0.5848. Crucially, we demonstrate that the language-specific BanglaBERT model consistently underperforms its multilingual counterparts with an F1-score of 0.4937, indicating that pretraining domain diversity can outweigh language specificity in highly specialized clinical settings. Furthermore, we present a detailed per-entity-type analysis for this task, revealing that Medicine and Specialist categories are recognized with high reliability, achieving F1- scores above 0.83, while the Symptom category remains the most challenging with an F1-score of 0.4367 despite being the most frequent training class. Finally, fine-tuned transformer models outperform the optimal prompting configuration by a factor of 3.76, confirming that prompt-only pipelines remain inadequate for structured clinical entity extraction in low-resource language environments.

Post-hoc correction adjusts a forecaster that cannot be retrained, such as a foundation model, but a correction fitted where errors are stable can hurt where they shift. We aim for downside control: not much worse than the starting forecast. We combine the frozen forecaster, a static corrector and an online corrector on the simplex, using only losses that mature after the horizon. Across seven benchmarks and four base models, two of them foundation models, the worst deterioration over 28 pairs at the main horizon is 0.15% and gains reach 11.5%. On day-ahead load for seven European bidding zones it lowers mean MSE in all seven zones, while single correctors raise mean MSE by up to 102% where the published forecast is most accurate. Three empirical conditions on expert speed, stream length and outcome alignment, each fixed by a documented failure, delimit its scope. Learning from the provisional outcome improves four zones on the settled one; learning on the settled outcome restores all seven.

When a tool-using agent's write times out or returns a server error, the action may already have taken effect. Retrying blindly duplicates it -- a second charge, a second announcement, a second deployment -- while giving up skips required work. We ask where exactly-once behaviour should be enforced: in the model, in the agent harness, or in the tool contract. We introduce LIMBO, a deterministic sandbox of six services with realistic contracts (optional idempotency keys, eventually consistent and missing read paths) and twelve fault modes injected at the service boundary, including late commits, redelivery and partial batches; every episode is graded against a ledger of committed effects. Across 25,930 episodes spanning nine recent models, three production agent harnesses, two contract variants and fifteen recovery conditions, the answer depends on the fault. When an immediate read-back can reveal what happened, the model decides: frontier models instructed to act exactly once almost never duplicate a write whose acknowledgement was lost (0.5%), weaker models often do, and the model explains 53% of the explained variance. When it cannot -- the request is still in flight, or the transport delivered it twice -- the same frontier models duplicate in 56% and 74% of episodes, and the contract explains 81%. We prove that no verification-only policy is exactly-once under late commits without a bound on in-flight time. Waiting works when such a bound is short and known, but with heavy-tailed in-flight delays even an hour of waiting per episode falls short of offering an idempotency key on every write, which lowers the duplicate rate from 28% to 4% because agents use keys when they exist. The harness barely matters, a guard that attaches keys transfers across harnesses unchanged, and agents reported success in 90% of the episodes in which they had duplicated an effect.

Continuum manipulators provide dexterous motion in confined spaces, but structural compliance, hysteresis, and load-dependent deformation leave residual position and orientation errors that can undermine reliable contact with rigid grippers. To address this limitation, this paper presents a lightweight support-enhanced granular-jamming gripper tailored to a continuum manipulator. The gripper maintains compliance before jamming while establishing a direct load path to the continuum manipulator tip after jamming. To improve its grasping performance, we systematically designed membrane materials, particles, filling ratios, and the internal support structure, and further identify geometry-dependent grasp boundaries with respect to contact offset and object shape. Building on these results, we construct a physical manipulation system integrating the continuum manipulator, granular-jamming gripper, visual feedback, tendon actuation, and pneumatic control. We then train a reinforcement-learning-based reaching controller in a randomized simulation and deploy it on the physical system, demonstrating how positioning control and contact level mechanical adaptation can complement each other in a modular grasp-and-release task. By introducing an adaptive structure that relaxes the need for highly accurate modeling and positioning control, this work explores a design paradigm that integrates physical and embodied intelligence.

Vision-based legged locomotion methods assume clean depth at training time and rely on hand-tuned post-processing filters at deployment. However, filter parameters are rarely disclosed, hindering reproducibility, and performance degrades substantially when depth noise is left unaddressed. Building noise robustness directly into the learning pipeline would eliminate this dependency. While such robustness has been explored for proprioceptive inputs, analogous approaches for depth perception remain largely absent in legged locomotion. We propose DAWN (Denoising and Alignment in World models for Noise-robustness), a noise-robust perception framework for legged locomotion, which builds noise robustness directly into a world model via two modifications: (1) feeding noisy depth to the encoder while keeping clean depth as the reconstruction target, forcing the model to implicitly denoise its input; and (2) applying contrastive learning to align the latent states of noisy and clean depth. Importantly, DAWN is not tied to a specific noise model, requiring no manual tuning to the noise distribution at deployment. Furthermore, it incurs no additional inference cost over existing world model-based methods. Without any manual filter calibration -- relying solely on the learned noise-robust representation -- DAWN achieves zero-shot quadruped parkour on a Unitree Go1: traversing stairs up to 18 cm, clearing gaps up to 70 cm, and mounting steps up to 45 cm from raw depth observations. Ablation studies show that denoising and contrastive alignment contribute at complementary levels -- reconstruction and representation, respectively -- and yield additive gains when combined. Videos and code are available at: https://dawn-parkour.github.io/

Recent advances in agentic systems have substantially enhanced the long-horizon capability of embodied manipulation. However, many existing frameworks still follow a passive execution paradigm, which limits their applicability to real-world scenarios involving textual semantic cues, distractors, and initially invisible targets. To bridge this gap, we propose an agent-based active exploration framework that enables robots to dynamically interact with the environment rather than merely execute predefined instructions. Specifically, our framework consists of three collaborative modules: a planning module for high-level task reasoning, a perception module for visual scene understanding, and an execution module for low-level manipulation. This design allows the robot to actively acquire task-relevant information, adapt its behavior based on environmental feedback, and complete manipulation tasks under partial observability. Furthermore, we introduce a fine-grained perception-execution interleaving strategy, which tightly couples visual feedback with skill execution to improve exploration robustness. We evaluate our method on a realistic Find-and-Place task, demonstrating its effectiveness in challenging environments where target objects must be actively discovered before manipulation.

We replicate the distributional-semantics extractive summarisation method of Mohd, Jan and Shah (2020) and adapt it to Hindi, substituting a Devanagari-appropriate component at every language-specific step. The system is evaluated on two independent corpora --- the Hindi portion of XL-Sum and FIRE ILSUM 2.0 Hindi --- under a Devanagari-aware ROUGE implementation validated against the XL-Sum authors' own multilingual scorer, with all comparisons drawn as 1000-resample paired bootstraps. In its published equal-weight configuration the replicated system is significantly worse than a three-sentence lead baseline on both corpora, trailing Lead-3 by 0.042 ROUGE-1 Fon XL-Sum and by 0.265 on ILSUM. A feature ablation shows that sentenceposition is the only feature that contributes: position alone reproduces the lead baseline exactly, removing position gives the weakest configuration,and a validation-tuned weighting can at best equal Lead-3 and never exceed it. TextRank fails identically, making this a class-level rather than an implementation-level result. A selection analysis shows the remaining features steer extraction towards long, entity-dense body sentences while the references reuse the article lead.Current Hindi benchmarks therefore cannot reward non-lead content selection, motivating purpose-built evaluation resources.

Developing optimized machine-learning algorithms for edge devices with limited computational and memory resources is challenging, time-consuming, and highly dependent on device-specific constraints. In this work, we streamline an edge ML workflow to enable rapid development, optimization, and deployment of machine-learning (ML) models directly on the WeBe Band, a wrist-worn wearable device designed for multimodal physiological data monitoring. The proposed system automatically generates hardware-efficient ML models that can be easily integrated into the WeBe core firmware, supporting AutoML, hardware-aware quantization, and performance profiling to build models that meet desired latency targets while remaining compatible with device memory and power limitations. The proposed framework tightly integrates the open-source Piccolo AI ecosystem with an automated pipeline that generates deployable firmware artifacts, performs hardware-aware model compilation, and supports over-the-air (OTA) deployment. The system supports multiple lightweight model classes, including classical machine-learning algorithms and neural networks, and provides built-in on-device profiling tools to evaluate inference latency and memory footprint under realistic execution conditions. Experimental results demonstrate clear trade-offs between model complexity and deployability on a microcontroller, showing that classical models offer strong real-time performance while lightweight neural networks require careful resource management. Rather than proposing new learning architectures, the current work mainly focuses on system-level automation, deployability, and enabling researchers and developers to rapidly iterate on models and evaluate them directly on target hardware. Although demonstrated on the WeBe Band platform, the workflow is designed to be extensible to other ML-powered edge devices.

Cancer registrars, including Oncology Data Specialists (ODSs), must interpret complex and frequently updated coding and staging standards. We developed CRISS (Cancer Registry Intelligent Support System), a retrieval-augmented generation (RAG) conversational assistant that provides rapid, citation-supported access to registry guidance. This study evaluated whether CRISS could (1) support accurate and citation-supported responses, (2) improve access to and interpretation of relevant guidance, and (3) support training/helpdesk use while preserving human oversight of final abstraction decisions. We built a domain-specific knowledge base from national cancer registry standards, segmented into metadata-tagged passages and indexed as dense embeddings. Retrieved passages were used to generate citation-grounded responses through a large language model (LLM). Open-weight, proprietary, and non-RAG baseline models across Gemini and GPT families were evaluated on easy, medium, and hard registry questions using an LLM-as-a-Judge protocols. RAG configurations consistently outperformed non-RAG approaches, especially as question difficulty increased. Mean grounding scores for RAG were 0.62/0.56/0.59 across easy/medium/hard tiers versus 0.29/0.26/0.29 for non-RAG. RAG models also achieved higher semantic-similarity scores overall. Proprietary RAG models performed strongest on easy and medium questions, while local RAG models ranked highest on hard questions and proprietary models were generally more cautious. Domain-specific RAG improved evidence grounding and response quality for cancer registry questions while enabling citation-supported assistance across complexity levels. CRISS demonstrates the potential of human-centered, citation-grounded AI to support cancer registrars while preserving human oversight for final coding decisions.

Grasping is a fundamental robotic capability that bridges perception and physical task execution. This paper studies grasp pose recovery under a perception-to-execution mismatch, where a grasp generated from visual perception may become spatially stale if the object moves before execution, using only sparse tactile interactions and no further visual observations. We propose DA-GRD, Decision-Aware Grasp-Relevant Disambiguation, which maintains a weighted planar belief over possible object configurations and selects tactile probes according to their ability to eliminate hypotheses and improve agreement among candidate task grasps. Rather than fully relocalizing the object, DA-GRD stops when the remaining hypotheses support a common executable grasp. In MuJoCo experiments on ten rigid objects with translations up to 5~cm and yaw perturbations up to $\pm45^\circ$, DA-GRD achieves an 84.7% physical lift success rate, compared with 9.1% for stale AnyGrasp, 21.2% for the original fix-scan baseline, and 63.7% for fix-scan method adapted with an SE(2) belief. DA-GRD also achieves a 57.3% Task conditioned Success rate. Across objects, it uses a success-average of 4.13 tactile probes over the ten per-object means, corresponding to a 72.5% reduction relative to the fixed 15-probe baselines. Real-world experiments on six objects achieve 71.7% physical lift success and 38.3% task-conditioned success with 4.20 probes on average. These results show that tactile sensing can recover task-relevant grasps under vision-off conditions with limited physical interaction, without requiring complete object localization.

The proliferation of photorealistic AI-generated images demands robust detection methods that generalize across diverse generative models. While existing approaches target manipulation-based forgeries with local artifacts, generation-based images (e.g., from diffusion models) lack such traces, posing a fundamental challenge. We observe that generative models prioritize global semantics at the expense of local texture fidelity, making low-texture regions key indicators of synthetic origin. To exploit this, we propose EIB-Net, an Entropy-guided Information Bottleneck Network. EIB-Net introduces a novel Image Entropy (IE) metric to automatically select the most informative (lowest-entropy) patch, then processes it with a Variational Information Bottleneck (VIB) to learn compact, generalizable features. Extensive experiments on DIFF, DiffusionForensics, and GenImage benchmarks demonstrate state-of-the-art performance: EIB-Net achieves 85.7\% accuracy using only 2\% of training data, outperforming full-image baselines by over 15\%, and maintains robust cross-generator generalization (83.5\% average accuracy on GenImage). Furthermore, our entropy-guided patch selection (EGPL) consistently enhances diverse backbones (CNNs and Transformers), proving its practical value for data-efficient detection.

On-policy self-distillation (OPSD) has become a popular recipe for post-training LLM agents. It supervises the agent model at the token level with a stronger teacher view of the same model, obtained by conditioning on privileged information (PI). In this work, we show that in multi-turn agents, this paradigm teaches the student to act with confidence but without the information behind it. The trained agent behaves as if it had privileged information it never observed, and its performance falls well short of plain RL, in the worst case below the untrained base model. Therefore, we propose Privileged Self-Practice (PSP), which keeps the PI and moves it from the loss to the sampler. When the student's rollouts on a task mostly fail, we inject a short per-task instruction written by an analyzer model, sample the task again with the instruction in context, and train on the result with an unchanged GRPO objective. The privileged information stays in the prompt and never enters the loss. Across AppWorld and SWE-bench Verified, with three different student models, PSP obtains the best average score in every setting and is the only method that consistently outperforms plain GRPO, improving task-goal completion by up to 65% on AppWorld and the resolved rate by up to 61% on SWE-bench Verified.

Tool-calling agents produce heterogeneous outputs, interleaving structured tool invocations with user-facing natural language summaries. This output heterogeneity presents a structural failure mode in standard on-policy Reinforcement Learning (RL): algorithms like GRPO indiscriminately broadcast a homogeneous trajectory-level scalar advantage to all tokens. Consequently, gradient noise from summary generation leaks into tool-decision tokens, causing cross-segment credit misattribution and brittle optimization. In this work, we propose SLCA-GRPO, a framework incorporating Segment-Locked Credit Assignment (SLCA). To enable scalable exploration without costly real APIs and stable training, we first construct the Schema-Guided LLM Simulator (SGLS) as foundational training infrastructure. Building on this, SLCA decouples advantage estimation at the structural segment level within a single group of rollouts, without requiring additional rollouts from intermediate states. Supported by Hierarchical Rewards (HierR), SLCA routes execution advantages to tool tokens and preference advantages to summary tokens, eliminating advantage contamination (the dominant cross-segment credit misattribution channel) within each policy update. On a 7B backbone, SLCA-GRPO accelerates convergence and outperforms standard GRPO, ToolPO, and RLTR by +2.53 pp on in-domain evaluation, +1.36 pp on the Berkeley Function-Calling Leaderboard (BFCL), and +9.15 pp on $τ^2$-Bench under the same training budgets, achieving higher accuracy with reduced tool redundancy and costs.

This paper investigates communication-free bearing-only formation tracking control for multi-agent systems governed by Euler-Lagrange dynamics. Distinct from existing results that can only stabilize a stationary formation, this work considers a scenario where the leaders move with time-varying velocities while the inter-agent communication is absent. In this setup, the leaders' states (position and velocity) are unavailable to all followers and cannot be estimated via distributed observers. A novel adaptive distributed control scheme is developed to address this problem. The design exploits the fact that bearing rates contain the projected relative-velocity information, which, together with bearing rigidity, provides a rigidity-based damping mechanism for compensating the unavailable velocity error. Moreover, this damping mechanism is incorporated into a bearing-driven auxiliary variable to construct a surrogate velocity error, facilitating the adaptive control design for EL dynamics. Furthermore, since this damping mechanism necessitates sufficient bearing rigidity, we characterize a rigidity-preserving set and establish its forward invariance, thereby guaranteeing such rigidity via initial conditions. Via a Filippov-based Lyapunov analysis, the proposed scheme is shown to achieve local practical formation tracking in the sense that the velocity error converges to zero and the position error is uniformly ultimately bounded. As a corollary, for the constant-velocity case, asymptotic tracking is achieved without initial-condition restriction. The simulation results verify the effectiveness of the proposed control law.

Unified vision-language models (VLMs) that tokenize images through a vector-quantized (VQ) codebook routinely hallucinate objects on grounded yes/no benchmarks, yet existing decoding-time fixes treat this as generic miscalibration without an architectural account. Using activation patching across twenty-five models spanning eight LLM families, we identify an early-layer ($L_0$) attention routing circuit shared across VQ-tokenized VLMs and propose a three-gate diagnostic that distinguishes the models carrying it from those that do not. The diagnostic isolates ten positive models (five natural unified-VQ VLMs across three LLM families and five induced variants) and rejects the remaining fifteen. A single-variable architectural swap (LLaVA-1.6 CLIP+MLP $\rightarrow$ VQ+Linear) installs the circuit, while a matched-compute MLP control on identical data does not, isolating vector quantization as the source of the pathological signal; the routing pathway that carries it is one that the backbone already provides. Against tuned VCD and DoLA baselines, tuned DoLA wins on binary calibration, but \textbf{only $L_0$ ablation reduces object hallucination in open-ended generation} (CHAIR$_i$ reduces by $31\,\%$ relatively, whereas tuned DoLA and VCD leave it unchanged or worsen it). These results recast object hallucination in unified VQ VLMs as a property of architecture and pretraining, and yield a targeted intervention that mechanism-agnostic decoding cannot replicate.

Pilot-aided channel estimation is a decisive block in orthogonal frequency-division multiplexing (OFDM) receivers for both 5G New Radio (5G-NR) and Long-Term Evolution (LTE). A large body of estimators exists, from simple least-squares (LS) interpolation to statistically optimal linear minimum-mean-square-error (LMMSE) variants and, more recently, deep convolutional denoisers, yet no single estimator is uniformly best: the winner depends on the propagation scenario, the numerology, the operating signal-to-noise ratio (SNR), the mobility (Doppler), and the antenna configuration. In this paper, we quantify this fact through a unified study of eight literature estimators evaluated over the 3GPP TR~38.901 Urban-Macro (UMa), Urban-Micro (UMi), and Rural-Macro (RMa) channels generated with NVIDIA Sionna, for both 5G-NR and LTE numerologies, in single-input single-output (SISO) and $8\times2$ multiple-input multiple-output (MIMO) settings. We then propose a \emph{condition-adaptive multi-agent orchestrator} that treats each estimator as an independent agent and dispatches, per operating condition, to the agent that is best on a validation split without any genie knowledge. The orchestrator tracks the per-realization oracle to within $1.07$~dB and improves the normalized mean-square error (NMSE) over the best \emph{fixed} strategy by up to $3.6$~dB at high SNR, where the low-SNR champion is no longer optimal. Because the agents are independent, running them concurrently delivers this best-of-eight accuracy at essentially single-estimator latency: a data-parallel partition scales the wall-clock nearly as $1/K$ with $K$ workers (up to $6.9\times$), whereas naive by-algorithm partitioning is Amdahl-limited by the heaviest agent. The results substantiate multi-agent orchestration as a practical route to robust channel estimation across heterogeneous 5G-NR/LTE deployments.

General Purpose Service Robot (GPSR) tasks, as defined in the RoboCup@Home benchmark, require robots to interpret diverse natural language commands and generate multi-step action sequences in real home environments. Conventional Single Prompt (SP) approaches suffer from context bloat and the "Lost in the Middle" phenomenon, leading to unreliable task planning. We propose an LLM chaining architecture that separates instruction classification and action generation into two specialized stages, reducing per-inference prompt length by approximately 45% while improving planning consistency. We evaluate our method using 100 randomly generated GPSR commands across three language models spanning local open-source and frontier cloud deployment contexts. Results show consistent planning improvements over SP across all models, with gains of up to +37 percentage points on local models. Further, real-robot execution experiments on the Toyota Human Support Robot (HSR) reveal that planning success alone does not guarantee task completion, with 6 of 10 tasks completing successfully and execution-layer failures identified as the primary remaining bottleneck.

Sparse activation reduces mixture-of-experts computation without eliminating the need to store all experts. We present Routide, a Swift/MLX runtime that executes the text path of a pinned public Qwen3.6-35B-A3B quantized checkpoint while keeping expert weights in iPhone storage and a byte-budgeted subset in memory. We characterize cache-policy sensitivity, numerical comparison boundaries, and measurement limits. Across five recorded 128-token workloads, fixed-route replay gives 0.00% demand hits with a 512 MiB LRU cache, 18.80% with seeded random eviction at the same budget, and 38.58% with 576 MiB LRU. The apparent capacity cliff is therefore a policy/workload interaction, not a universal memory requirement. Same-runtime Mac controls preserve generated sequences across eviction and asynchronous prefetch, including 2,560 exact token comparisons and 10,334 speculative loads. In contrast, complete resident-Python versus recorded-phone sequences disagree on all five tested cases, precluding a general numerical equivalence claim. Two separately scoped iOS 27 memory protocols observe sampled process-footprint peaks of 1.87-2.32 GiB on short prompts and 2.39-2.73 GiB on one longer prompt. We retain a thermal stopping event, negative timing comparisons, and a single qualified whole-device power estimate. These results establish bounded feasibility and identify limitations that a deployment claim must not hide.

Torque observations in reinforcement learning remain challenging because simulated and measured torque differ in scale, offset, and noise. In this paper, we propose a simple torque observation alignment method for robots with direct-drive (DD) actuators, in which motor current maps linearly to joint torque through a motor-type-specific torque constant K_tau. First, dynamometer calibration identifies K_tau* and corrects the scale mismatch between simulated and real torque. Second, the method uses delta_tau(t) = tau(t) - tau(t-1) as the observation in both domains to eliminate the constant offset instead of using the direct torque tau(t), which carries a domain-dependent bias. Third, Gaussian noise obtained from the dynamometer measurement data is injected during the learning process. To validate the proposed method, we train a teacher-student grasping policy entirely in simulation and deploy the distilled student on a multifingered DD gripper. The deployed policy performs proprioceptive grasping using only joint positions and torque differences. We conduct an ablation study comparing the proposed method with alternative alignment variants on nine in-distribution (ID) objects. The proposed method achieves 100% grasp success. These results demonstrate that the proposed alignment method improves the robustness of zero-shot policy transfer on the DD gripper against real-world torque-observation mismatches.

In this paper, we propose RGBD20K, a novel dataset for facilitating the development of more robust and general RGB-D semantic segmentation by encompassing abundant categories and high-quality annotations. RGBD20K possesses several attractive properties: (1) Expanded Semantic Space. In particular, it covers 160 fine-grained categories, largely surpassing the category diversity of existing popular RGB-D benchmarks (e.g., NYUv2 with 40 classes and SUN RGB-D with 37 classes). With such enriched semantic coverage, we expect to promote the learning of more generalizable segmentation models. (2) Larger Scale. Compared with current benchmarks, RGBD20K offers 20,000 RGB-D image pairs, providing a substantially larger training resource that benefits the development of more powerful deep models. (3) High-Fidelity Annotation. We perform rigorous re-evaluation and correction of existing labels to resolve long-standing annotation noise, resulting in a clean and reliable ground-truth foundation. Furthermore, we propose a novel score-purified fusion (SPF) method, which achieves state-of-the-art performance across all evaluated benchmarks, demonstrating the effectiveness of our approach in leveraging high-quality multimodal information for RGB-D semantic segmentation. The dataset is here: https://github.com/ShaohuaDong2021/RGBD20K/.

Coordinated arm-hand motion planning is fundamental to dexterous robotic manipulation in complex and constrained environments. A straightforward solution is to decompose the problem into separate arm path planning and hand motion generation; however, this poses a dilemma: decomposition can miss feasible solutions that require coordinated arm-hand adaptation along the path. Alternatively, directly planning in the high-dimensional joint arm-hand configuration space captures such coupling but faces a substantially enlarged search space and nonconvex collision constraints. To characterize this coupling, we formulate feasible hand fibers that capture collision-free hand configurations for each arm configuration. Based on this formulation, we propose CAMP, a high-success and efficient cooperative arm-hand motion planner for constrained environments. CAMP constructs candidate trajectories through layered hand search with local arm relaxation, then compactly represents them using endpoint-preserving via-point movement primitives (VMPs) for coarse-to-fine joint optimization. Across six constrained simulation tasks, CAMP achieves 84.2-98.5% planning success, outperforming alternative planners with competitive efficiency. Ablation studies verify the contributions of arm relaxation, VMP representation, and coarse-to-fine optimization, while real-robot experiments demonstrate CAMP on constrained manipulation tasks. The project website is available at https://camp-armhand.github.io/.

Skilled humans can catch fast-moving objects softly by coordinating interception, velocity matching, and follow-through to mitigate impact. Learning such impact-aware catching with reinforcement learning (RL), however, is challenging, as the policy must achieve reliable interception and grasping while regulating the sensitive transition into contact. Moreover, even a capable privileged-state RL teacher may not provide ideal demonstrations for a deployable imitation-learning (IL) student: teacher failures limit task coverage, while small variations in pre-contact motion can produce substantially different impact and grasping outcomes. We characterize this phenomenon through interventional outcome sensitivity and introduce the outcome-sensitive window (OSW) to guide targeted demonstration construction. Building on this formulation, we propose Outcome-Sensitive Motion Search, which learns a task-conditioned manifold of successful OSW motions and performs local geodesic search to refine successful teacher rollouts and repair task conditions where the teacher fails. We then validate candidate motions through complete rollouts under a calibrated IL-student action-error model and retain only successful executions as demonstrations. Extensive simulation experiments demonstrate that our method effectively repairs task conditions where the teacher fails and enables the resulting IL policy to outperform the privileged RL teacher in both catching success and impact mitigation.

Most robots manipulate objects solely with their end effectors, whereas humans flexibly leverage different body parts, such as the forearm and elbow, especially when handling oversized objects. Learning such whole-arm manipulation is chal-lenging due to long-horizon sparse rewards, limited contact sens-ing, and the sim-to-real gap in contact and actuator dynamics. To address these challenges, we propose Current-Aligned Link Manipulation, a framework for learning long-horizon contact-rich manipulation using motor current as joint load related feedback. Three stage-specific policies first learn repositioning, grasping, and lifting using privileged simulation information, and a stage router sequences them to generate complete task demonstrations. For sim-to-real transfer, a causal current mapper predicts physical motor current from simulated joint histories, aligning the actuator current observation between simulation and hardware. A unified student policy then learns from these demonstrations using only deployable sensor observations and is further refined with DAgger. The task policies are trained entirely in simulation, and the final student is deployed on hardware. Experiments demonstrate 76.2% (762/1000 trials) complete-task success in simulation and 73.3% success (22/30 trials) on the physical robot for sequential oversized-object lifting.

AI-driven scientific discovery accelerates research by autonomously developing solutions and designs. Large language model (LLM) agents support this process through iterative generation and evaluation. Yet these iterations alone do not ensure cumulative progress or establish which directions to pursue next. Costly evaluation further constrains the scope of exploration. Neural architecture discovery brings these challenges together, coupling open-ended design with resource-intensive experimentation. We introduce EvoTreeNAD, a genealogy-guided evolutionary algorithm that constructs trainable architectures without a supplied seed or a hand-specified search space. Starting from an empty root, it grows a persistent genealogy in which each new node represents a complete architecture. Top-percentile values computed from each node and its descendants guide lineage selection. Using the selected design history, an Idea Agent proposes a variant and a Code Agent implements it. Each evaluated variant becomes a child node, expanding the genealogy while providing evidence for subsequent lineage selection. Our theoretical analysis establishes the existence of stationary variation regimes as the genealogy grows. Under specified variation assumptions, sustained top-percentile family values quantify the probability of generating high-reward architectures in these regimes. EvoTreeNAD discovers architectures that outperform the compared NAS and NAD baselines, achieving CIFAR-10/100 test errors of $2.05{\pm}0.06\%$ and $15.09{\pm}0.22\%$. On all six MedMNIST-v2 tasks, the discovered architectures surpass the strongest listed baselines. A controlled CIFAR-10 study further shows that EvoTreeNAD outperforms direct generation, best-of-$N$ greedy continuation, and full-family-mean routing.

Decentralized LLM-based multi-agent systems coordinate through local interactions, but an agent can remain responsive while its task-solving quality persistently degrades. Such gray failures require protecting current tasks before sufficient evidence exists to alter future routing, while still allowing recovered agents to rejoin. We introduce MeshHeal, a fully decentralized self-healing framework that couples ability-matched peer review across two timescales. At the fast timescale, an adaptive hierarchy escalates uncertain or low-scoring outputs from repeated single-reviewer evaluation to committee deliberation and, when needed, correction before use. At the slow timescale, a task- and ability-conditioned peer-relative detector aggregates scores to distinguish persistent degradation from ordinary output variation, trigger mandatory committee review, and eventually exclude degraded agents from ordinary routing; recovery probes provide fresh evidence for reintegration. To faithfully evaluate routing, we introduce Model-Backed MAS Evaluation, which ties ability assignments to execution models, since prompt-based ability assignments alone can leave routing errors hidden. Across BBH, MATH, and MMLU-Pro, MeshHeal achieves 0.839 degraded-phase accuracy using 51k total model tokens per task, versus the strongest baseline Symphony's 0.807 accuracy using 115k per task. Under staggered degradation and recovery, MeshHeal isolates degraded agents, keeps them excluded from ordinary task execution until recovery, and returns them to normal routing.

Large language model (LLM)-based multi-agent systems can automate alpha factor mining, but their reliance on external APIs limits control over cost, availability, and confidentiality. Long research loops also tend to revisit a few successful economic mechanisms that lead to research path collapse. To address these limitations, we propose AlphaDiverse, a framework that integrates a multi-agent alpha research system, diverse research path collection, and post-training for local agents. We let the research system generate complementary plan portfolios and vary research environments across loops to collect diverse research paths. Using these diverse traces, we warm-start local Planner and Realizer agents with supervised fine-tuning. Then, we propose a joint GRPO method to optimize both of them using predictive quality and diversity of contributions. Research feedback is confined to inner period data, while a frozen final model is evaluated on a later outer period data, thereby avoiding test-set tuning. Experiments across four Chinese stock universes show that AlphaDiverse can combine competitive prediction with broader exploration.

Existing video deraining methods typically exploit neighboring frames through either explicit alignment or implicit spatiotemporal aggregation. Explicit alignment relies on accurate motion estimation, which can become unreliable under dense rain, while implicit aggregation avoids alignment but lacks explicit guidance on the directional and temporally coherent structure of rain. This leaves a gap between reliable temporal aggregation and explicit modeling of rain motion. To address these limitations, we propose FluidRain, a lightweight video derainer that uses divergence-free rain flow to guide Loop-in-Loop attention across scales and neighboring frames. Motivated by fluid mechanics, we model rain motion as a divergence-free image-space flow and use it to organize multi-scale and temporal aggregation. Specifically, FluidRain first estimates a rain-flow field for each frame and projects it onto the divergence-free subspace. The resulting flow steers window attention along rain streaks, enabling neighboring frames to be aggregated without explicit alignment. Since rain-flow structure is preserved across scales and nearby frames, Loop-in-Loop reuses the same attention operator across both dimensions, resulting in a three-frame model with only 0.80M parameters. Experiments on four benchmarks show that FluidRain remains competitive with substantially larger restoration models. We further examine how temporal evidence scales with different input views. To evaluate whether the model remains reliable when rain motion changes across frames, we introduce RainSyn-Gust, which injects controlled changes in rain-streak direction into existing benchmarks. We also develop a physics-based no-reference metric that evaluates real-rain removal without requiring clean targets.

Robot policies deployed in real environments naturally accumulate mixed-quality experience, including successful executions, partial progress, and failures. Although these rollouts provide valuable information for further learning, directly incorporating them into imitation learning may reinforce undesirable behaviors, while offline reinforcement learning often suffers from unreliable value estimation under sparse rewards and limited data coverage. We consider a practical post-deployment setting where learning relies only on naturally accumulated autonomous rollouts, without additional human corrections or exploratory interaction. To effectively exploit such experience, we propose Predictive Action Chunk Learning (PACL). PACL first learns a predictive chunk-level critic that evaluates temporally extended action sequences and augments temporal difference learning with future latent prediction, providing richer supervision for long-horizon value estimation. The learned critic then converts chunk-level Q-values into discrete quality conditions, which guide a diffusion actor to learn jointly from these mixed-quality experiences without treating all behaviors as equivalent supervision. At inference, the actor generates multiple action chunks and the critic selects the highest valued candidate. Experiments across simulated and real-world robot manipulation tasks show that PACL consistently improves the pretrained policy and outperforms strong imitation learning and offline reinforcement learning baselines.

Despite strong performance on standard benchmarks, it remains unclear whether large language models (LLMs) evaluate social pragmatics in ways that align with human judgments. We evaluate LLM politeness judgments using two English-language datasets with complementary annotation formats: continuous human ratings and three-way categorical labels. Across the seven evaluated models, we find that inter-model agreement is stronger than model--human agreement. Strategy-level analyses suggest that model--human alignment is associated with explicit linguistic cues, while some rapport-building strategies occur more frequently in misaligned cases. In the categorical task, model predictions exhibit systematic neutral compression, characterized by the overproduction of Neutral labels and the underprediction of Impolite labels. This pattern persists when expert consensus is used as the reference on a diagnostic subset. Our findings highlight the need for pragmatic evaluations that go beyond aggregate agreement metrics by examining directional patterns of model--human disagreement across different human references.

Most of the memory of a 3D Gaussian Splatting model holds spherical-harmonic colour coefficients, yet each Gaussian is seen only from the narrow cone of directions of the training cameras. We turn this into a distortion metric that other compressors can adopt: a per-Gaussian observation Gram matrix, accumulated from viewing directions and blending weights, is the exact first-order map from coefficient changes to squared image error and needs only the model and the camera poses. Under it, degree reduction becomes a closed-form projection that generalises truncation, degree allocation a Lagrangian rate-distortion problem, and vector quantisation the matrix-weighted Lloyd algorithm, of which Compressed3D's quantiser is the scalar case. Swapped into Compressed3D with everything else unchanged, the metric raises PSNR by +0.49 dB before fine-tuning, with SSIM and LPIPS following, and at matched rate still gains +0.32 dB without a single training image. A training-free stack built on the metric alone is 15% smaller than the image-free GSICO at equal quality on Mip-NeRF 360.

Video understanding is increasingly performed by multi-stage LLM agents that separate temporal grounding, visual observation, and reasoning. Yet these stages are typically evaluated on different benchmarks and distributions, making it difficult to determine where hallucinations originate. We first organize existing benchmarks around these stages and show that their scores provide inconsistent diagnostic signals: stronger stage-level performance does not reliably imply lower downstream hallucination, and even benchmarks targeting the same capability can disagree. We therefore introduce a causal stage-intervention protocol that overwrites individual stages while holding the downstream task fixed. Across 60,008 runs on three video-agent architectures, we find that grounding is the dominant source of downstream error, with roughly four times the causal impact of corrupting visual observations. Successful grounding depends primarily on locating the correct region rather than precise temporal overlap, explaining why standard mIoU metrics poorly predict downstream reliability. We further find that incorrect evidence is substantially more harmful than missing evidence. Finally, auditing existing benchmarks against these interventions reveals that their scores do not reliably predict causal cascade sensitivity and can fail under distribution shift. These results motivate intervention-based, stage-aware evaluation for trustworthy video agents.

We present a personalized Korean visual speech recognition (VSR) system and quantify, on the nine-camera OLKAVS corpus, the gap between the population-level benchmark score and an individual user's error. A video-only Conformer initialized from English-trained weights attains 9.95 - 12.19% character error rate (CER) under the corpus protocol against the published 26.64, and 19.00 - 21.52 on unseen wording. Per speaker, CER spans 1.0 to 52.2%, with seen wording lowering CER by 7.0 - 9.0 points and professional delivery and spontaneous speech raising it by 8.5 - 10.5 and 12.7 points. A low-rank adapter with 4.6% of the parameters, trained on 4 to 29 minutes of the user's frontal video, lowers the CER of twelve high-error speakers by 2.13 to 3.58 points, transfers to every camera without loss, and keeps 85% of the full fine-tuning gain at 12% of its cost to other speakers. Cameras above the mouth plane add about six CER points as a constant offset that training on all views keeps small.

Cross-embodiment learning has shown that a single model, such as a vision-language-action (VLA) model, can learn state representations and manipulation skills that can be applied across heterogeneous robots to accomplish various tasks. We hypothesize that the same holds for safety enforcement. The reasoning required to satisfy a safety constraint, such as detecting an obstacle, recognizing that it should be avoided, and selecting a safe abstract action, is largely shared across robots. What differs across embodiments is how the abstract safe action is realized: morphology, kinematics, and dynamics determine which actions are safe and feasible. Consequently, the same action can be safe for one robot and unsafe for another. This is especially important for generalist manipulation policies that operate in a common end-effector action space without explicitly capturing how safety depends on the robot's morphology and kinematics. We propose embodiment-conditioned safety filtering, in which a Hamilton-Jacobi reachability-based value function and its corresponding safety-maximizing policy are shared across robots. Using a morphology-aware latent representation of the robot and its environment, we perform Hamilton-Jacobi reachability analysis directly in latent space so that the learned safety concepts can generalize across embodiments while remaining explicitly conditioned on each robot's morphology and kinematics. We evaluate our approach across five bimanual robot embodiments and five manipulation tasks with whole-body collision-avoidance constraints. Our results show that a single policy, jointly trained across five manipulation tasks and four embodiments, exhibits zero-shot generalization to a held-out embodiment, reducing the nominal policy's collision rate. They also show that training using more embodiments improves generalization.

Post-training quantization (PTQ) reduces the cost of on-device text-to-speech (TTS), but published evaluations cover one system or method. We evaluate PTQ across TTS architectures under one protocol with three core models, weight and activation ablations of eight more, and two held-out models quantized blind. Four-bit per-channel weights reduce UTMOS, a predicted mean opinion score, by 2.8 on Supertonic and 0.07 on Kokoro, and per-tensor scaling can cause severe degradation even at 8 bits. The same bit width yields different outcomes, because the sensitive component is model-specific and not reliably predicted from the model class. A staged ablation procedure identifies it, and per-layer GPTQ can restore it to within 0.1 UTMOS. Real int8 and int4 kernels reproduce the simulated ordering at hardware-dependent cost. On a Mac mini, a 4-bit weight kernel runs Supertonic at 0.60x the fp32 latency while int8 is slower, so each configuration requires validation on the target runtime.

Physical changes or sensing errors can invalidate embodied agents' task-relevant beliefs. AquaMend compares re-probing, rollback, and supported continuation on a probe-belief-action graph under an expected-loss objective covering sensing, physical recovery, and uncorrected failures. A joint posterior guides a one-step policy with conditional detection-power screening. The per-belief three-way optimum requires independence, separability, and fully resolving probes; the general policy has no global optimality guarantee. Across 32 paired scenarios in a self-constructed simulation benchmark, AquaMend recovers in 28/32 cases and reduces mean complete loss by 21.6% versus restart. Its paired loss difference from decision-theoretic troubleshooting (DTT) is not statistically significant after Holm correction. Against the all-candidate ablation, online decision time decreases by 12.3% overall but increases by 3.4% in the uncovered late stage.

While end-to-end autonomous driving systems show promise, their application to micromobility vehicles is hindered by simulators failing to capture specific kinematics, such as differential drives and omni-wheels. This paper pro- poses a sim-to-real-aware, vehicle-specific end-to-end learning environment for the WHILL Model CR on AWSIM and ROS 2. To minimize the sim-to-real gap, physical parameters are optimized via Bayesian optimization using real-world data, reducing trajectory errors across various driving scenarios. Additionally, this study introduces a synchronized architecture tailored for the stable training of world model-based agents. An end-to-end policy trained with DreamerV3 exhibited learning progress and achieved task completion in a simulated obstacle avoidance setting. Furthermore, this policy demonstrated direct sim-to-real transfer to the physical vehicle, enabling the vehicle to navigate around a cardboard box in a real-world corridor replica without fine-tuning. This paper provides a practical foundation for sim-to-real micromobility policy studies.

Group-based reinforcement learning (RL) methods, such as GRPO and its variants, have become a leading paradigm for training reasoning and agentic large language models (LLMs). While their group-normalized advantage estimation is reliable at the response level, it becomes systematically biased at the step level, since coarse-grained trajectory-level advantages are hard to accurately reflect the contribution of individual steps (i.e, failed trajectories may contain valuable steps). Revisiting the foundational RL definition, we notice that GRPO's success on single-turn tasks stems from its advantage estimation strategy, which adheres to the basic definition: the mean reward of multiple actions sampled from the same state constitutes a credible state-value estimate. Extending the faithful estimation to step-level would in principle demand sampling multiple actions from each intermediate state, which is too costly on a per-state basis. To mitigate this issue, we propose a Graph-based Faithful sTep-level credit-assignment framework (GRAFT) that grafts all rollout trajectories into a trajectory graph, recovering node state-values via Bellman iteration on the graph, and assigning credit to each edge by the node value difference. Theoretically, the estimated step-level advantage faithfully adheres to the basic advantage definition in RL. To further ensure the reliability of step-level advantage estimation, we further propose Graph GAE, which extends GAE to the trajectory graph for reducing the impact of state-value estimation bias. Experiments across a range of multi-turn agentic benchmarks show consistent gains over GRPO and superior performance compared to recent agentic RL algorithms. Code will be available at https://github.com/xcyao00/GRAFT.

While recent advances in perceptive locomotion have enabled humanoid robots to traverse structured terrains, agile parkour in highly discontinuous environments remains an open challenge. In particular, crossing sparse footholds and narrow support regions requires precise foothold selection, effective use of visual observations, and consistent alternating foot placement during fast transitions. In this paper, we present a perceptive humanoid parkour framework that enables stable traversal across terrains with limited foothold availability using only onboard depth observations. The framework features a saliency-guided temporal perception module that combines a saliency prior with gated memory. It retains informative depth features across frames, enabling reliable foot placement from partial observations. By introducing an alternation loss, our symmetry regularization encourages alternating gait patterns and improves traversal robustness. Extensive experiments show that our method significantly improves success rate and foothold accuracy on challenging terrains in both simulation and the real world.

Humanoid parkour policies can traverse various terrains, but task completion may mask challenges of harsh landings, edge contacts, and unstable stance contacts. Humans naturally regulate foot-terrain interaction through tactile feedback, modulating contact compliance according to terrain stiffness. This highlights a key domain gap between humans and humanoid robots: the absence of rich tactile sensing in most humanoid systems. We address this problem with TactileStep, a deployable tactile learning framework that brings sole pressure sensing into humanoid locomotion control for softer touchdowns and more stable support. TactileStep aligns tactile simulation with the real pressure insole, allowing the policy to learn from the same contact features available on hardware. During training, we use tactile and motion cues to recognize different foot-contact phases and apply phase-aware rewards that encourage safer landing and more stable stance. Evaluated in simulation and on a Unitree G1 humanoid across diverse terrains, TactileStep reduces peak touchdown force by up to 48.8% and peak A-weighted impact noise by up to 30.1 dB over a strong perceptive baseline, while increasing stance contact area by up to 23.8%.

Recent advances in video object segmentation with Multimodal Large Language Model (MLLM) reasoning have demonstrated the effectiveness of using a single textual token, such as SEG, to predict segmentation masks across images and videos. However, we observe that this single-token strategy lacks the granularity required to precisely localize multiple objects across time in video segmentation tasks. To address this limitation, we develop Multi-Token Reasoning for Video Object Segmentation, or MoVISA. MoVISA uses multiple segmentation tokens, such as SEG0 and SEG1, to represent an object across different frames. This design enables more fine-grained alignment between language prompts and spatio-temporal mask predictions, improving both performance and interpretability. On the challenging MeViS, DAVIS17, ReVOS, and Ref-Youtube-VOS benchmarks, our model achieves a 13.2 percent J and F improvement on MeViS and an 8.4 percent J and F improvement on ReVOS. Code and models will be released.

Current Vision-Language-Action (VLA) models often struggle with high-precision robotic manipulation. We attribute this limitation primarily to their visual attention being dispersed across task-irrelevant regions. To address this issue, we propose ActGaze, a training approach that guides VLA policies to gaze on task-relevant regions, much like humans gaze on critical visual cues while executing precise movements. Unlike prior methods that rely on external labels for gaze supervision, ActGaze derives spatial supervision directly from the VLA's own action objective by using counterfactual visual interventions to identify regions that are critical for action prediction. Extensive real-robot experiments on four high-precision robotic manipulation tasks demonstrate that ActGaze induces more focused visual attention on task-relevant regions and consistently outperforms the base VLA policy and other visual-grounding approaches.

Robot-policy benchmarks increasingly cover diverse tasks and preset out-of-distribution conditions, but typically evaluate complete trajectories from predefined initial states. These evaluations often focus on the initialized scene and the final outcome, while paying less attention to the dynamic interaction process. During closed-loop execution, actions and contacts can alter object relations and task progress, producing off-nominal intermediate states that need recovery. Recovery requires a policy to infer how task progress has changed, correct the relevant relations, and continue the original goal. We introduce RoboRecover, a benchmark for robot policy recovery under execution deviations. RoboRecover selects deviation states from trajectories, reconstructs them by replaying action prefixes, and evaluates policies on the original task. RoboRecover contains 2,000 scenarios across RoboTwin and LIBERO, with 1,000 scenarios and a fixed 800/200 train/test split on each platform. Results show that initial-state performance does not determine recovery performance and policies exhibit different recovery strengths across scenarios. Using its training split, RoboRecover further supports study on recovery interventions. RoboRecover establishes recovery from execution-induced intermediate states as a distinct dimension of robot policy evaluation.

Few-shot segmentation (FSS) aims to segment unseen object categories with a few (e.g., one or five) labeled examples, enabling efficient adaptation to novel classes. Conventional models typically rely on appearance-based visual matching between support and query images for segmentation. While straightforward, these methods often struggle to handle significant appearance discrepancies and occlusions in the query image due to insufficient target knowledge. To mitigate this, we introduce a novel framework that mines target knowledge using the strong reasoning capacity of Multimodal Large Language Models (MLLMs) and employs it to enhance FSS. Specifically, building on SAM 2, our method, named MK-FSS, exploits two forms of complementary knowledge derived from a query image by an MLLM for FSS, including spatial knowledge, which provides a spatial prior indicating the potential target location, and semantic knowledge, which describes the target using text. The spatial knowledge is first encoded into a memory representation, and then resulting memory is integrated with the support-guided memory feature from query image through a carefully designed dual-memory debate-fusion (DMDF) module, yielding a more robust target memory feature. In parallel, the semantic knowledge is encoded into the textual feature, which is fused with multi-scale query features via a progressive cross-modal prompt generator (PCPG), producing a target-aware multimodal prompt for segmentation. Working together, the dual-memory feature and the multimodal prompt provide a comprehensive representation of the target, enabling more robust segmentation. In our extensive experiments, MK-FSS shows promising results and largely surpasses existing methods. Code will be released.

Personalized value alignment has become increasingly important as large language models (LLMs) are expected to accommodate diverse user preferences. However, existing methods typically align model outputs with a static value profile across prompts, overlooking that the salience of value dimensions varies substantially across contexts. Inspired by Lewin's Field Theory, which views human behavior as jointly shaped by personal dispositions and situational constraints, we model personal values as priors and context-dependent preferences as posteriors. We propose BaCVA, an inference-time Bayesian Context-aware personalized Value Alignment method that approximates posterior personalized preferences by integrating static personal values with scenario-specific value salience. BaCVA first estimates contextual value salience from generally normative responses, and then employs a dual-view personalization module to infer posterior preferences from complementary personal-value and scenario-driven perspectives. This Bayesian formulation enables more accurate and adaptive personalized value alignment while improving data efficiency via prior values. Extensive experiments on benchmarks demonstrate its superiority over strong baselines.

Autonomous penetration-testing harnesses use large language models (LLMs) for reconnaissance, exploitation, and reporting, but often rely on those same models to confirm findings, grade severity, and select agents. This can lead to false positives, inflated severity, and wasted compute. We examine how System One decision models, lightweight non-generative classifiers that return typed, calibrated verdicts, can support these decisions. We make five contributions. First, we define four decision points: finding adjudication, severity recalibration, agent pruning, and confirmation loops. Second, we present an exploratory NeuroSploit case study comparing one run with TypeSafe System One (Jev) and one without it against a web target containing 13 vulnerabilities. Differences in severity distribution, runtime, and grading by exposed data type motivate the architecture but do not establish statistical significance. Third, we review published specifications for Jev, Jev-Ultrafast, and the open-source Laya without assuming that results from other benchmarks transfer to penetration testing. Fourth, we discuss RLHF, RLAIF, RLCD, and RLHV as training approaches and their implications for trust in security decisions. Finally, we propose Rave, a domain-adapted System One model, and outline its training data, evaluation protocol, and potential effect on harness assurance.

Driving world models provide a promising route toward scalable counterfactual data generation and interactive simulation beyond recorded driving logs. Realizing this potential requires a system that can generalize across diverse scenes, respond faithfully to prescribed controls, generate coherent multi-sensor observations, and operate efficiently under repeated inference. We present \textbf{HelloWorld}, a 2B driving world model system designed around these requirements. HelloWorld progressively specializes broad visual and motion priors from heterogeneous video data into controllable driving generation using ego pose, HD maps, and 3D boxes. A block-causal generation interface, together with adaptation to self-generated context, aligns the model with sequential simulation. The system further supports synchronized seven-camera RGB generation and conditional LiDAR synthesis, and is distilled toward few-step inference for efficient deployment. Experiments evaluate visual quality, control fidelity, cross-view consistency, robustness under repeated generation, inference efficiency, and LiDAR synthesis. Together, HelloWorld provides a unified framework for scalable driving data generation and interactive simulation.

Radiance field representations such as 3D Gaussian Splatting (3DGS) natively encode complex visual phenomena such as occlusions and view dependence, but they are inherently underconstrained. Independently optimized reconstructions converge to different primitive configurations, even in unchanged regions. We introduce Plenoptic CharacterIstics (PlenoCI), a novel feature built from the plenoptic field these representations approximate. PlenoCI directly captures rich visual behaviors while ignoring Lambertian textures. By deriving closed-form analytic plenoptic derivatives from a 3DGS representation, we efficiently detect these 5D structures. Our approach is robust to underconstrained representations by construction, reporting two orders of magnitude fewer false positives between independent reconstructions of unchanged scenes than concurrent work. We demonstrate PlenoCI's utility on change classification. First, we detect changes with an instance-aware 3DGS pipeline, achieving state-of-the-art results on CL-Splats with a 25.7% mIoU gain over the strongest competitor, while remaining competitive on the more challenging PASLCD benchmark. Leveraging PlenoCI, we classify changes as geometric or appearance-based with a balanced accuracy of 0.735, comparable to the best performing baseline. We believe plenoptic derivatives and PlenoCI open new directions for view dependence aware understanding in visually complex environments. Code and data are available at https://js0n-lai.github.io/plenoci.

World action models (WAMs) that use future visual prediction at inference time incur substantial generation costs. Asynchronous execution reduces waiting by overlapping inference with robot motion, but visual predictions used for subsequent action generation must anticipate the effects of actions already scheduled for execution during inference. We introduce Streaming-WAM, which couples action-conditioned world modeling with asynchronous robot control to account for committed actions in future visual prediction. At each streaming update, the model conditions future visual prediction on the latest observation and the committed actions, which form the fixed prefix of the next action chunk. The resulting action-conditioned visual features guide generation of the remaining actions within the same joint update, so the continuation is informed by the scene changes expected during execution of the fixed prefix. On LIBERO, Streaming-WAM achieves an average success rate of 98.35\% and reduces mean episode time by a factor of 2.93 relative to Fast-WAM. On the real-world Stamp Paper task, mean episode time falls from 90 s with synchronous Joint-WAM to 38 s with Streaming-WAM. These results show that Streaming-WAM supports efficient asynchronous control while maintaining high task success rates.

The structure of a periodic multilayer X-ray mirror is obtained by fitting its X-ray reflectivity (XRR) curve, and the result depends on how the operator normalizes and trims the curve, frees parameters, and accepts a fit. The manual of the fitting program and the papers describing its engine leave these decisions to the operator, whose practice is tacit, so the fitting stays with the expert. To solve this problem, we proposed to develop a skill for a large language model (LLM) agent via elicitation: the expert's decisions were recorded during fitting and written as thirteen steps and a 22-item report template. The agent runs X-Ray Calc 3 through a Model Context Protocol (MCP) tool server. Fresh sessions, each given the skill, one curve, and a nominal design, were scored against fits the expert had withheld, under six tolerances fixed beforehand. The skill was developed on XRR curves of Co/C mirrors and of Ru/C mirrors from a public data deposit. The final version of the skill was tested on W/B4C multilayers. It was demonstrated that the skill recovered the mean period within 0.3 Å of the expert's fits and the period drift through the stack on both W/B4C specimens, and the W and B4C thicknesses within 1 Å on one of them.

Few-step autoregressive (AR) video diffusion enables low-latency streaming generation, but existing post-training methods predominantly rely on Distribution Matching Distillation (DMD), requiring both a large pretrained teacher and an online critic to estimate distributional discrepancies through diffusion scores. In this work, we ask whether this resource-intensive teacher--critic stack can be eliminated by post-training only the generator against a precomputed target distribution. Drawing inspiration from representation distribution matching (RDM) for one-step image generation, we systematically study its transfer to few-step causal video generation and identify three key barriers: a memory-intractable gradient path, a distinct video optimization regime, and representation distributions that underconstrain temporal dynamics. We introduce ViRDM, a teacher- and critic-free video post-training recipe that addresses these barriers sequentially. By coupling RDM with stochastically truncated clean-exit supervision, a lightweight VAE decoder, and staged vector--Jacobian products, ViRDM makes representation distribution matching memory-feasible for multi-step causal video rollouts. We further establish effective generated-population and initialization regimes for video RDM, and introduce lightweight dynamics regularization to compensate for the underconstrained temporal dynamics. ViRDM turns three-network distillation into generator-only post-training, reducing GPU memory use and training time while improving video quality. With only 20 generator updates, the recipe reaches 84.87 on the official VBench evaluation, outperforming the previous best few-step causal baseline by 0.36, while requiring 16 A100 GPU-hours. We additionally report exploratory results demonstrating the potential of the same recipe for lower causal sampling budget and for one-, two-, and four-step bidirectional generation.

Protein modification requires navigating an immense sequence space, yet wet-lab validation remains low-throughput and costly. Although computational paradigms including protein language models (PLMs), large language models (LLMs), and LLM-based agents have shown promise in protein modification, their relative efficacy across realistic experimental decision-making settings remains unclear. To bridge this gap, we introduce PFArena, a benchmark comprising four controlled task interfaces that cover single-mutant generation and multi-mutant ranking. By providing varying levels of mutation fitness data, PFArena reflects four representative research scenarios characterized by differing degrees of prior experimental context. We assess six PLMs, six LLMs, and five LLM-based agents using complementary metrics to measure both peak and overall protein modification performance. Our evaluation reveals that model performance shifts systematically with the availability of target-specific experimental evidence: PLMs demonstrate proficiency in open-ended single-mutant generation by leveraging protein-specific priors, whereas LLMs and agents perform strongly in multi-mutant ranking, particularly when target-specific fitness data are available. Nevertheless, all model families face fundamental challenges with increasing search-space size and mutation depth. We release our code and benchmark suite to facilitate reproducible research in model-assisted protein modification.

Conventional model-based diffusion (MBD) achieves effective trajectory optimization by leveraging noise annealing. However, its high computational cost, primarily arising from repeated rollouts of the plant dynamics, has largely confined its use to offline settings. To address this limitation, this paper proposes bilinear Koopman model-based diffusion (BK-MBD). The proposed method lifts the robot's state into a high-dimensional space only once per control step and propagates all candidates in the lifted space thereafter, so each rollout reduces to a fixed number of matrix-vector multiplications. The lifted dynamics are bilinear, allowing the predicted input gain to vary with the robot's configuration, which a linear lifted model cannot represent. In simulation, BK-MBD completed each planning update in at most 14.7 ms within a 50 ms control period and reached the goal on every trial, whereas a linear lift almost never did. The annealed schedule improves closed-loop accuracy over fixed-noise schedules under the learned rollout. Under the exact rollout, both the annealed and fixed-narrow schedules reach every goal, indicating that annealing reduces sensitivity to surrogate-model error. BK-MBD also threaded a passage that no single convex region covers, whereas a convexified bilinear controller rarely succeeded. On a physical manipulator, BK-MBD tracked an initially unknown moving target within the control period and was the only method that met both the tracking task and the deadline. The project page is available at https://rcilab.khu.ac.kr/bkmbd/.

Harnesses, the products that run AI coding agents, are multiplying, and enterprises are rolling them out to their employees: what started as pilots with a few hundred seats is scaling to tens of thousands. Most enterprises do not build these harnesses but buy them from large vendors, such as Anthropic's Claude Code or OpenAI's Codex. A harness decides which model answers, what the model reads, how the prompt cache is used and which subagents run, so it picks the rate on the price sheet and sets the volume bought at it. Enterprises that keep a proprietary or untuned harness at its defaults inherit these choices and their bill. We build a fast, customisable router in which Jev, a classifier with calibrated probabilities, labels every prompt against a bring-your-own taxonomy of agentic requests. Because one user turn is many requests over a prompt cache that belongs to one model, the router moves work only where no running conversation has to rebuild its cache: at session start, in side lanes and at subagent launch. From the price sheet we derive when a mid-task switch pays back, and a crossover: on long tool-heavy sessions the highest-priced model costs less than the next tier, as repricing about 10,000 real sessions from public datasets confirms. In an emulated enterprise of 10,000 seats with user behaviour taken from these datasets, the router recovers 14 to 21% of model spend at Anthropic's list prices of 21 September 2026, \$3.3M to \$5.0M a year. The paper also maps the risks across twenty harnesses, prices the dependence on one vendor's models, and proposes a control plane that enterprises can run from within, starting now, with a ladder for deciding later whether to own the harness.

LLM agents are deployed into infrastructure that grants them broad host authority, yet existing agent-security benchmarks and defenses operate almost exclusively at the application telemetry layer: the served tool manifest, the user prompt, and the model's messages. Some threats, however, smuggle malicious instructions and actions past the application boundary, leaving them invisible to that layer. In this work, we bridge that gap by pairing application-level agent telemetry with kernel-level syscall traces to present the first paired-evidence characterization of kernel-level versus application-layer signal for agent security. To quantify the value of the enhanced telemetry, we introduce Agent Cross-Layer Evidence (ACE), a paired-session corpus of 4,047 sessions and 17 threat models spanning six delivery-vector families and 14 of the 25 OWASP LLM and agentic threat categories, organized into 12 attack mechanics with per-mechanic characterization of where the most discriminative evidence lies. Across four distinct detector families, we find that kernel evidence is discriminative on its own and that composing it with application-layer evidence generally outperforms either single-layer view, revealing complementary signals that single-layer analyses can miss. We further demonstrate generalization to unseen attack families and transfer to an alternate agent runtime. Together, these findings establish the value of cross-layer evidence for agent security.

Agent behavior depends on the harness surrounding a language model, but it remains unclear whether language models can reliably improve such harnesses for hardware-design tasks. We study automatic harness evolution around a fixed subject model on 12 proprietary design-verification root-cause localization tasks. Across five trials per task, automatically evolved harnesses increased completed attempts by 71-76% and any-hit task coverage by 80-100%, while total correct attempts improved by only 18-24%. The strongest success reproducible at least twice result improved by one task, and later candidates exchanged gains across tasks rather than preserving them. An auxiliary candidate improved on a four-task validation set excluded from search but tied its baseline on a subsequent 12-task replay containing both search and validation tasks, so the selected gain did not persist across the full pool. Across the tested lineage, useful search, evidence, and finalization behaviors appeared in different candidates but did not consistently consolidate into a single harness that dominated across tasks and metrics. In a separate CVDP cross-benchmark case study, an automatically evolved defined-width repair harness produced 35.6% more functional passes than its 142-task reference baseline; the final functional verifier scored completed outputs but was not shown to the subject agent during repair. These results support archive-aware selection when evolution yields complementary specializations without consistent consolidation.

Multi-agent systems built on large language models (LLMs) are increasingly deployed in high-stakes settings such as finance, healthcare, and software engineering, where agents coordinate through natural-language messages. The same channels, however, let colluding agents exfiltrate confidential information or coordinate unauthorized actions, and steganography can hide such communication inside outputs that look ordinary to an auditor reading the transcript. Existing provably undetectable LLM steganography protocols are not suited to realistic deployments. High-capacity schemes assume a symmetric setting where the receiver can reproduce the sender's output distribution, the state-of-the-art protocol for asymmetric agents has very low capacity, and most approaches rely on a pre-shared secret key. We make the threat of undetectable agent collusion concrete with Codetta, a high-capacity steganographic protocol for independently deployed agents in realistic asymmetric settings. Codetta combines a shared public model that estimates the communication channel, a sampling mechanism that preserves the sender's output distribution, and an adaptive error-correcting code. It further removes the pre-shared key through a steganographic key exchange that lets independently deployed agents establish a shared key while keeping the transcript computationally indistinguishable from ordinary model outputs. Across three agent workloads and three sender models, Codetta achieves up to $94\times$ the capacity of the state-of-the-art asymmetric protocol, and its key exchange establishes a shared key with about 80k visible tokens at an empirically certified failure probability of at most $4.1\times 10^{-3}$. These results show that effectively undetectable collusion is becoming feasible between independently deployed agents, so auditing must go beyond inspecting communication transcripts.

Audio-language models can produce confident answers unsupported by the audio, motivating uncertainty estimates that identify unreliable responses. We compare probability-based, sampling-based, self-verification, evidential, and contrastive measures across four open-weight models and five audio QA benchmarks. In multiple-choice evaluation, first-token measures are strongest overall, with top-1 probability achieving a mean AUROC of .740, compared with .708 for ten-sample discrete semantic entropy, while requiring no additional model calls. Across four benchmarks, shifting from multiple-choice to open-ended evaluation lowers mean accuracy from 57.6% to 36.6%, yet uncertainty remains predictive of errors: semantic entropy, maximum token entropy, and semantic agreement achieve mean AUROCs of .697, .694, and .693, respectively. To test whether uncertainty reflects the evidence available to answer the question, we perform input ablations that remove either the audio or the question. Across top-1 confidence, entropy, and sampling-based measures, removing audio reduces error-detection AUROC by .101 on average, compared with .010 when removing the question. Together, these results establish efficient uncertainty baselines and show that uncertainty in audio-language models depends substantially more on available audio evidence than on question text.

Sim-to-real transfer has made substantial progress, but can still produce controllers that remain stable and functional on hardware while suffering from degraded tracking accuracy due to residual dynamics mismatch. Correcting these errors typically requires identifying the underlying system dynamics, adapting the control policy, or returning to simulation for additional training and finetuning, all of which can require substantial data and computation. We propose OSRAM (Online Sim-to-Real Adaptation via Closed-Loop System Modeling), a framework that instead adapts the reference commands provided to an existing controller. OSRAM treats the deployed robot and its policy as a unified closed-loop dynamical system and learns its task-level command-response behavior directly from tracking observations. A closed-loop dynamics model is meta-trained across randomized dynamics in simulation and rapidly finetuned after deployment using limited real-world interaction. The adapted model is then used to optimize future reference commands while leaving the underlying control policy unchanged. We evaluate OSRAM on bipedal velocity tracking and loco-manipulation in simulation and on hardware. Results show that closed-loop modeling improves prediction and tracking accuracy under unseen dynamics, while online reference adaptation reduces residual sim-to-real tracking errors across different control objectives and hardware configurations. These results demonstrate that adapting the behavior of the robot-policy closed loop provides a practical alternative to finetuning the policy or identifying the full physical dynamics for sim-to-real transfer. More information can be found at http://generalroboticslab.com/OSRAM.

We introduce Forecast-Dojo, a replayable environment for benchmarking and training LLM forecasting agents. It combines resolved prediction-market questions with dated news, allowing agents to research an event and revisit their predictions at successive historical dates. The same tasks and tools support repeated evaluation, collection of training interactions, and feedback from recorded outcomes without waiting for new events to resolve. Forecast-Dojo contains 1,568 Polymarket events, split by time into training and evaluation periods, and 18.8M dated news articles. In an evaluation of 12 models, research tools lower Brier score for all 12. Forecasts also improve as events unfold, with the largest gains at steps where more newly dated evidence is recorded. Every model still trails historical market forecasts in both Brier score and accuracy. A belief notebook carried between dates lowers research cost but does not consistently improve forecast quality. Beyond evaluation, Forecast-Dojo provides interaction trajectories and outcome feedback for agent learning, with supervised fine-tuning as a proof of concept.

Long-running LLM applications repeatedly send growing context, making prefix caching critical for reducing prefill cost. Yet prefix-cache behavior under agentic workloads remains poorly understood. We study production traces from two companies and evaluate 14 eviction algorithms across HBM-constrained and large memory-pool settings. Despite a large gap to Belady, sophisticated policies designed for traditional caches provide little benefit over LRU. The reason is structural: prefix reuse is dominated by the regular pacing of active sessions, making recency unusually predictive. Prefix caching nevertheless introduces new challenges, including heavy-tailed session footprints and highly variable miss costs as attention computation grows with sequence length. We introduce the compute-savings ratio and two offline oracles to quantify these effects. Our results show that effective prefix-cache management should retain recency as its foundation while selectively adding quick demotion for one-hit prefixes, compute-aware partial eviction for expensive misses, and capacity-dependent eviction granularity. We will release the traces and simulator to support future research.

Action representation plays a central role in discrete-token vision-language-action (VLA) learning but remains underexamined. Under conventional pose-increment representations, action tokens are sensitive to execution speed and dataset-specific normalization, potentially obscuring geometric structure shared across demonstrations and datasets. We introduce Direction-Scale Decomposition (DSD), an action representation that decomposes translation and rotation increments into direction and scale components before tokenization. DSD isolates motion direction while retaining magnitudes in separate scale channels. We evaluate DSD with uniform binning (BIN) and BEAST, a B-spline-based tokenizer, in simulation and real-world manipulation under both single-dataset and mixed-dataset training. On LIBERO, DSD improves average success rates with both tokenizers. On SimplerEnv, DSD-BIN outperforms BIN by 10.3 percentage points in overall success rate under mixed-dataset training. Real-robot experiments further show gains both with and without robotics pretraining. These results support DSD as an effective action representation for discrete-token VLA models and suggest its potential to mitigate performance degradation when training on large and diverse dataset mixtures. Our project page with additional resources is available at https://vla-dsd.github.io/