Physical AI Brief

Recent omni-modal large language models (Omni-LLMs) show great potential as real-time video assistants, which continuously perceive environments and guide users to achieve specific goals. Unlike traditional passive video understanding, interactive assistants should actively combine visual states, user goals, and prior knowledge to provide effective help. Evaluating this is rather challenging, as the model's unpredictable response dynamically changes the user's subsequent actions, which static offline datasets cannot accommodate. To address this bottleneck, we introduce OmniAssistBench. To solve the issue of diverging interaction paths where the same user goal can be achieved through various methods, we provide models with predefined priors derived from the source video, requiring them to guide users along the exact same routes. Since real interaction videos are rare, we construct the dataset by reverse-engineering existing Internet videos. We deduce logical user goals and segment the videos into multi-turn clips to simulate continuous interactions. This rigorous pipeline required over 1000 expert person-hours to build the dataset. Results show that the proprietary Gemini-3-Pro reaches 66.4 out of the max point of 100, while the open-source Qwen3-Omni-Instruct achieves 51.2. Although current models generally understand user inputs, they frequently provide incorrect or incomplete answers. Specifically, they struggle with visual prompts (e.g., hand gestures), fail to maintain historical context during multi-turn interactions, and fail to delay response until the target event. Results indicate substantial room for improvement before models can become reliable assistants.

In professional life sciences workflows, scientists routinely interpret visual artifacts (gel blots, microscopy images, plasmid maps, flow cytometry plots, molecular structures, ...) to inform research decisions. We introduce VIALS, a visual question-answering benchmark with 161 such interpretation tasks, spanning the types of artifacts examined throughout experimental workflows in the biotech industry (rather than polished figures from publications and textbooks). While frontier vision-language models can now fluently describe natural images, we find that they are unable to accurately interpret these scientific images, reflecting limitations in domain knowledge and domain-specific visual reasoning capabilities. In contrast, scientists with relevant domain expertise find these visual interpretation tasks straightforward. AI that cannot similarly interpret such images will have limited utility in professional life sciences workflows, where such artifacts are central to how scientists reason, communicate, and make decisions.

For sixty years, machine verification has been a major cost overhead, affordable only for exceptional artifacts. Here we report that generative AI inverts this relationship: at AI speed, machine verification is not only economical but essential to productivity --- it is the incorruptible referee that lets one person safely direct autonomous machine work at scale. In five weeks, one researcher on consumer AI subscriptions directed a small fleet of AI agents from application code, through a verified compiler and executive, to a RISC-V processor taped out on a community silicon shuttle; no proof passed through human review, and no RTL was written by a human. The working discipline --- the Salt method --- rests on a proof kernel no hallucinated proof can pass: mathematical claims travel between agents as kernel-checked artifacts, and human attention is reserved for statements, designs, and rulings. Verification is stated link by link, from the Lean 4 kernel to SAT-checked equivalence at the silicon boundary. We publish the complete accounting: theorem provenance, a pre-registered token meter, floor-bounded human time, and an error ledger whose catch numbering runs to #256 --- a monotone counter over the mathematics campaign's append-only flags ledger, maintained 2026-07-07 to 2026-07-20 (one number, #79, was never assigned; later catches are recorded un-numbered) --- against zero incorrect proofs reaching the record.

Recent vision-based action models have demonstrated strong capabilities in complex manipulation, but they rarely leverage explicit object physical properties to adapt their policies. We introduce ViTacPhys, a visual-tactile framework and data acquisition system that estimates object mass and friction-coefficient classes, together with continuous stiffness, from human manipulation demonstrations. Trained on data from 60 rigid and deformable objects, ViTacPhys combines temporal visual-tactile modeling, cross-attention multimodal fusion, and a semantic prior derived from a vision-language model. On seen objects, it achieves 97.2% mass classification accuracy, 98.8% friction-coefficient classification accuracy, and a stiffness mean absolute percentage error (MAPE) of 5.51%. On held-out objects from known categories, it achieves 87.5% mass accuracy, 97.5% friction-coefficient accuracy, and a stiffness MAPE of 9.08%. We transfer ViTacPhys from the human domain to the robot domain using limited robot teleoperation data, robot-style video augmentation, and human demonstrations with matched actions, and deploy it as an online module for adaptive grasping. The resulting physical-property-conditioned policy achieves total grasping success rates of 95.0% on in-distribution objects and 83.4% on out-of-distribution objects. For out-of-distribution objects successfully grasped by both methods, its force profiles are more consistent with human teleoperation than those produced by ACT. These results demonstrate the feasibility of explicitly estimating and conditioning on object physical properties for real-world adaptive grasping.

Scarce data and tumor heterogeneity limit patient-level cancer treatment-response prediction. Existing approaches predict response from pretreatment molecular profiles and drug representations, without explicitly modeling the molecular changes expected under treatment. We propose PerturbRx, a treatment-conditioned representation learning framework that learns intervention-induced latent transitions and uses them as patient-drug response features. PerturbRx trains a drug- and dose-conditioned transition predictor from context-matched but cell-unpaired control and treated single-cell populations, then freezes and transfers the predictor to pretreatment patient profiles without requiring post-treatment measurements. The transition is combined with patient and drug representations to predict response. Across TCGA and patient-derived xenograft benchmarks, PerturbRx achieves the strongest aggregate predictive performance among the evaluated methods. These results support perturbation-pretrained latent transitions as useful representations for patient-level drug-response prediction.

Self-refinement, typically structured as generation, critique, and revision, is a widely adopted paradigm for improving LLM generation and serves as a core mechanism in many LLM agents. While the three stages involve different cognitive demands, most existing approaches conveniently treat the model size as an implementation detail rather than a subject of study, which may lead to a waste of resources. Little work has systematically examined how model size affects each stage or whether effective self-refinement requires equally capable models for generation, critique, and revision. We present the first stage-wise model size study of the self-refinement pipeline on 5 benchmarks from different domains using 6 model sizes of Qwen3 and 4 model sizes of Gemma 3. We conclude that larger generators and refiners generally improve the pipeline, whereas an undersized refiner can even harm performance. Second, performance is highly insensitive to the size of the critic, although including even a small critic consistently outperforms omitting critique altogether. Our findings demonstrate that model capacity should not be allocated uniformly across self-refinement pipelines. Instead, different stages exhibit distinct size scaling characteristics, providing practical guidance for designing more computationally efficient multi-stage language model systems.

We formalize the Steiner Traveling Salesman Problem (Steiner-TSP) on Graphs of Convex Sets (GCS), which seeks a minimum-cost closed trajectory through required convex sets while allowing optional transit vertices and revisits. To explore the resulting infinite solution space, we propose a unified branch-and-bound search over rooted walk prefixes. Additive lower-bound-graph costs bound committed prefixes, while a cut-separated connected-flow relaxation lower-bounds the residual cost of visiting every remaining target and returning to the root. Under a uniform positive-cost assumption, best-first traversal terminates after finitely many expansions on every feasible instance without an initial incumbent, whereas depth-first traversal does so once a finite incumbent is available. For a user-specified factor $ε\geq1$, a global lower bound certifies that either strategy's incumbent cost is at most $ε$ times the global optimum. We further demonstrate joint sensing-mode, visitation-order, and continuous-trajectory selection for a mobile-manipulator inspection task, including action precedences expressed in linear temporal logic over finite traces (LTL$_f$). Both traversal strategies find feasible solutions on all benchmark instances within 30s with mean certified optimality gaps of 28.1% and 29.7%, respectively, whereas two recent baselines succeed on only about half of the instances

The European Union (EU) has emerged as a leading regulatory body in the development of sustainability and privacy regulations. While new regulation requirements vary, many include a documentation artifact to ensure compliance. Notably, the Ecodesign for Sustainable Products Regulation (ESPR) introduces Digital Product Passports (DPPs) for life cycle transparency, while the General Data Protection Regulation (GDPR) mandates Data Protection Impact Assessments (DPIAs) to mitigate privacy risks. Creating these compliance artifacts, however, is challenging. Industrial data, which often exists in heterogeneous formats and is scattered across company and supplier systems, is required for DPPs and can be difficult to extract into compliant DPP formatting. Furthermore, DPIA documents require interdisciplinary expertise and follow no standardized format, making development difficult for novel systems. To address the particular complexity of compliance artifact creation for both regulations, researchers have proposed the use of LLMs in the generation process; however, the impact of the aforementioned problems on the output of these systems is largely unaddressed. This work investigates the existing research gap by exploring how data extraction instructions and regulatory vagueness impact the quality and consistency of LLM-produced compliance artifacts. The resulting artifacts are evaluated by benchmarking different models against manually created ground-truth schemas. The results reveal that less strict guidelines, such as DPIA formatting, require higher context prompts to maintain consistency and completeness. Stricter guidelines, such as formatting for Digital Battery Passports (DBP), result in consistent results regardless of prompt context, but may lead to more hallucinations in the output

Test-Time Training (TTT) enables long-context processing via continuous weight updates during inference, but current methods struggle to balance the expressivity of per-token update dynamics with the hardware efficiency of chunk-wise approximations. We propose E$^2$-TTT (Expressive and Efficient TTT) to bridge this gap. Under the standard approximation of taking gradients at the chunk-start weights, we derive a closed-form state transition that exactly reproduces the chunk-end fast-weight and momentum states of the per-token recurrence. This enables fully parallelized chunk-level training while preserving the temporal structure of the update rule that prior chunk-wise methods discard. We validate E$^2$-TTT by training models up to 1.3B parameters from scratch. It performs on par with previous TTT and hybrid attention baselines in language modeling while outperforming them on in-context retrieval. Its advantage is most pronounced in length extrapolation: on the standard ``Needle in a Haystack'' passkey test, it retains over 90% accuracy at $8\times$ the training context length. Meanwhile, E$^2$-TTT can match the training throughput of efficient chunk-wise methods, demonstrating that it effectively reconciles expressivity with efficiency. The code is available at https://github.com/zeyun-zhong/E2-TTT.

Reinforcement Learning (RL) has demonstrated significant gains in image captioning, yet it is still limited in encouraging Large Vision-Language Models (LVLMs) to explore novel reasoning strategies. This limitation leads to a performance gap between RL and Supervised Fine-Tuning (SFT). In this paper, we argue that multi-modal retrieval can serve as an effective reasoning signal for caption refinement. Based on this insight, we present the Retrieval-Guided Refinement for Image Captioning (Re$^3$Cap), a retrieval-guided reasoning strategy that enhances image captioning without requiring additional annotations. Instantiated by Caption Refinement Suggester (CRS) and Caption Quality Assessor (CQA), this strategy identifies hallucinations and omissions in image captions, leading to more accurate and detailed descriptions. Extensive experiments demonstrate the superiority of our method in image captioning, even compared with Supervised Fine-Tuning. Especially, Re$^3$Cap outperforms GRPO with an average improvement of 8.64% in relation reasoning on the COCO-LN500 benchmark.

Foundation models for medical image segmentation, like prompt-based MedSAM, generalize well across domains and modalities, often in zero or few-shot setups. However, their performance depends on the quality of prompts and the adaptation of the models to custom datasets. This work systematically examines how MedSAM generalizes across diverse medical imaging benchmarks, with six adaptation strategies: full-model and encoder-only LoRA, shallow and deep visual prompt tuning (VPT), and decoder-only and full fine-tuning. Models are trained on the International Skin Imaging Collaboration Challenge (ISIC 2018) dataset and evaluated under clean and increasingly noisy prompts on IN and Out-of-Distribution (OOD) datasets: close-OOD PH2 (dermoscopy), far-OOD BUSI (Breast Ultrasound Images Dataset) and CBIS-DDSM (Curated Breast Imaging Subset of the Digital Database for Screening Mammography). We show that adaptation improves performance on IN and close-OOD data but often reduces performance on far-OOD data. Full fine-tuning provides the best tradeoff, while encoder-only LoRA is the strongest parameter-efficient alternative, outperforming standard LoRA and VPT under far-OOD shifts. Using Centered Kernel Alignment (CKA), we show that far-OOD degradation is strongly associated with drift in decoder representations, whereas encoder similarity alone does not explain robustness. This suggests encoder-only LoRA provides stronger robustness than standard LoRA by adapting the encoder to distribution shift in visual features, while preserving the decoder pathway. We further show that random 0-100 pixel jitter on prompts produces more robust and better performing models. We thus conclude that robust MedSAM adaptation requires the combined consideration of prompt noise exposure, domain shift, and representation preservation. We release our code: https://github.com/ImSounic/medsam-vpt

Skills play different roles as an agent's policy evolves: they should first provide learnable knowledge, then support capability formation, and finally be invoked only when they improve individual decisions. Existing methods rarely model this lifecycle. They either keep skills outside the model, fully internalize them, or select among internalization and utilization objectives through noisy task-level success rates. Such designs fragment training and assign uniform importance to actions within the same trajectory, even though skill guidance may help some decisions while distracting others. To solve these problems, we introduce AUSO (Action-level Unified Skill Optimization), which unifies skill learning and skill use through a progressive, action-aware optimization process. At the beginning of training, AUSO jointly learns from teacher guidance and environmental outcomes, enabling the policy to acquire foundational skills without losing task-oriented feedback. It subsequently emphasizes outcome-based policy optimization to consolidate autonomous problem-solving ability. As the policy matures, AUSO evaluates each sampled action under both skill-conditioned and skill-free contexts. The resulting action-level information signal is coupled with the trajectory outcome advantage, allowing beneficial skill-sensitive actions to receive stronger updates and harmful ones to be suppressed. Therefore, skills gradually transition from an external source of supervision into decision knowledge whose utilization is adapted to its action-level benefit, while reinforcement learning remains the shared backbone across all stages. Experiments on ALFWorld, WebShop, and SearchQA show that AUSO consistently improves agent performance and out-of-distribution generalization over competitive baselines.

We propose VT-MUSE, a Multimodal Unified SEquential representation learning framework for visuotactilemanipulation. Existing approaches often encode visual and tactile observations independently before fusion, limiting their ability to capture fine-grained cross-modal dependencies. Moreover, most methods focus on observations at the current time step and overlook the temporal evolution of contact. VT-MUSE addresses both limitations through a two-stage representation learning framework. In Stage I, modality specific encoders are jointly adapted via cross-modal temporal alignment and masked-view consistency. In Stage II, a conditional variational latent model processes masked visual sequences together with full tactile histories. Auxiliary decoders reconstruct the masked recent visual observations and predict tactile depth changes, encouraging the latent representation to retain both global visual context and local contact dynamics. The learned representation is subsequently integrated into a lightweight Transformer policy through gated cross-attention. On the simulation benchmark, VT-MUSE outperforms the strongest baseline evaluated on all tasks by 11 percentage points and also achieves substantial improvements in real-world experiments.

Recent advances in field technology have led to a massive influx of in-the-wild video data for ecological science. The primary bottleneck in leveraging this data is the high cost of expert annotation. While computer vision offers a potential solution, current models frequently fail when deployed in complex marine environments. To characterize these failures, we introduce WildFin, a novel benchmark for fish behavior recognition collected and annotated by ecologists.WildFin spans two critical real-world paradigms: stationary cameras monitoring groups of fish and dynamic divers following individual subjects. The dataset represents a massive curation effort, involving 1,350 hours of fieldwork and 600 hours of expert annotation to produce 9 hours of behavioral data with over 2 million frame-by-frame labels. We benchmark modern vision foundation models and quantify tradeoffs between static and spatiotemporal architectures, revealing the substantial gap that remains between current model capabilities and the demands of real-world underwater behavioral analysis. Project website: https://team-wildfin.github.io/.

Improving the safety of large language models (LLMs) often comes at the expense of utility, as globally applied safety tuning may affect model responses to both harmful and benign inputs. We propose \textbf{C}ontinuous \textbf{L}at\textbf{E}nt \textbf{A}dapter \textbf{R}outing (CLEAR), a conditional safety adaptation framework that uses a lightweight hidden-state gate to continuously control the activation strength of a safety low-rank adapter. CLEAR aims to reduce harmful completions while avoiding unnecessary changes to the frozen backbone that could degrade performance on benign prompts. Experiments on widely used safety and utility benchmarks show that CLEAR improves robustness on HarmBench while reducing the utility degradation observed with globally applied safety tuning such as SFT or standard low-rank adaptation (LoRA). On Llama-3-8B-Instruct, CLEAR reduces HarmBench ASR from 32.3\% to 0.5\%, while retaining most of the base model's utility and achieving up to 7.1 percentage points higher GSM8K accuracy than globally applied SFT or LoRA. These results suggest that CLEAR is a promising mechanism for improving the safety--utility trade-off in LLM alignment.

Large language models often rely on Chain-of-Thought (CoT) reasoning to solve complex tasks, but verbose reasoning traces introduce substantial inference overhead. CoT compression shortens generation, yet aggressive compression may disrupt logical coherence and degrade performance. We formalize this trade-off as the \textit{Context-Generation Substitution Law}, where explicit reasoning context substitutes for part of decode-time generation. Based on this principle, we propose \textit{Memory-Augmented Compression}, a training-free framework that constructs reusable reasoning memories from historical traces and retrieves them as prefill-side scaffolds. Rather than using raw demonstrations, these memories summarize reusable reasoning patterns, key constraints, and critical operations to compensate for information lost during compression. Experiments show that Memory consistently improves prompt-based Chain-of-Draft (CoD) compression across mathematical reasoning, complex reasoning, and science question answering tasks, yielding accuracy gains of 21.4, 28.0, 29.5, and 6.61 points over CoD on GSM8K, MATH, BBH, and MMLU-Sci, while achieving a 1.14--1.49$\times$ latency speedup over standard CoT. Memory is also compatible with token-level, reasoning-trace-level, and inference-state compression mechanisms. Further analyzes show that the gains come from relevant reasoning memories rather than simply increasing context length.

Question answering (QA) over long, connected documents remains challenging because relevant evidence may span multiple entities and their relationships. Existing retrieval-augmented generation (RAG) methods typically index documents as raw chunks and retrieve them through embedding similarity. Their performance degrades when chunk boundaries separate entities from supporting evidence or when a question requires multi-hop reasoning across the corpus. We propose EnSI-RAG (Entity-Structure-Indexed Retrieval-Augmented Generation), a framework that constructs a query-independent, entity-centered index. Each record (e, t, k, v) represents an entity e, its type t, a semantic category k in {property, relation, aspect}, and a value v, while retaining links to the original source passages. At query time, these records serve as retrieval handles, and an LLM synthesizes the retrieved passages into the final answer. This design separates evidence localization from answer synthesis while preserving traceable source evidence. Across Loong and Oolong, EnSI-RAG achieves an average accuracy of 78.24. Relative to the published baseline scores used as references, this is 6.62 points higher, suggesting its effectiveness across these settings. The code is available at https://github.com/RamonMeng/EnSI-RAG.

While recent large language models (LLMs) have achieved promising results on individual patent drafting tasks, they fundamentally fail to investigate the core challenge of real-world patent drafting: generating a complete and legally coherent patent application directly from early-stage invention materials. Prior work predominantly assumes later-stage, highly structured, or already legalistic inputs. However, real patenting workflows begin with informal, de-legalized disclosures authored by inventors. To bridge the gap, we introduce Dis2Pat, a disclosure-to-patent dataset that reflects realistic patenting workflows by requiring the generation of complete patent applications directly from inventor-style, de-legalized disclosures. Given the inherent difficulty of long-form, legally constrained patent drafting and the strong privacy requirements, we further propose a strong baseline named Patent-MAF. It is a multi-agent framework for locally deployable patent drafting. Benchmark results reveal that current LLMs exhibit limitations in patent drafting, while Patent-MAF provides a strong baseline that consistently outperforms evaluated open-source models and remains competitive with large closed-source models.

Token compression has become a key technique for reducing the inference cost of large foundation models, with approaches such as token pruning and KV-cache reuse widely adopted in vision-language models and recently explored for embodied agents. In embodied agents, tokens not only support perception and semantic understanding but also directly affect latency-sensitive closed-loop robot action prediction. Existing schemes typically guide compression using redundancy or importance cues, such as visual similarity, attention scores, and saliency. However, these cues only indirectly measure the key factor for safe compression: how much a token can change before causing an unacceptable deviation in downstream actions. This receiver-dependent tolerance is closely related to the principle of just noticeable difference (JND). Classical JND characterizes signal tolerance in the human visual system, while machine-oriented JND extends this concept to downstream machine responses. Building on this progression, we introduce Action-JND, which extends JND modeling to embodied perception by defining noticeability through the language-conditioned action response of a vision-language-action (VLA) policy in closed-loop control. A token change is considered admissible only when the induced action deviation remains within a tolerated margin. To realize this concept, we develop a lightweight token-wise JND estimator in deep visual-feature space to predict the maximum tolerable perturbation while preserving policy responses. The resulting action-tolerance score serves as a plug-and-play criterion for VLA compression paradigms, including stale-KV reuse and token pruning, prioritizing action-tolerant tokens for compression. Experiments on the LIBERO benchmark with OpenVLA and OpenVLA-OFT demonstrate that Action-JND consistently improves compression reliability, especially under aggressive compression ratios.

Vision-language models enable training-free video anomaly detection by answering questions about video segments. VAD benchmarks, however, require a scalar anomaly score for each segment and evaluate the resulting ranking using the AUROC or AP. A VLM-based detector should therefore define an answer interface: the answer scale specifies the admissible answers, and the readout rule maps the model's output distribution to a score. Because this interface can change the evaluated ranking, it is part of the detector rather than a formatting detail. The generated readout uses only the most likely answer, whereas the probability readout uses the full distribution over admissible answers. Across four 7-8B VLMs, the probability readout outperforms the generated readout for every tested combination of answer scale, benchmark, and metric, with average gains ranging from 5 to 13 points across the four benchmark-metric pairs. The gap arises because the generated readout keeps only one answer value per segment, so segment with different answer distributions can receive the same score and lose their relative order. We call this loss of relative order generated-answer rank compression. Even when the answer scale allows 91 answers, the generated readout produces only 4-18 distinct scores, whereas the probability readout retains substantially finer score resolution. The advantage persists under every decoding strategy, prompt wording, and joint scoring-explanation prompt we test. The answer interface is therefore a consequential component of VLM-based VAD and should be explicitly specified and evaluated.

Sequential recommendation predicts the next item from a user's interaction history, but not every interaction is equally informative. Real logs combine persistent preferences with temporary needs, exploration, and incidental behavior, so some interactions can distort history representations or provide unreliable supervision. Existing denoising methods judge such interactions mainly from co-occurrence, order, or model predictions, without explicit evidence from relations between items. Knowledge graphs (KGs) offer this evidence, but item popularity, graph degree, uneven coverage, and widely shared entities can inflate connectivity and bias reliability estimates. Here we present AdaptedKG, which derives calibrated KG evidence for each training example without adding graph representations to the recommendation model. It first compares the observed context with structurally matched alternatives to identify relational paths that are unusually prominent and uses them to build a local KG view. It then compares each interaction with structurally matched reference items to calibrate its support within that view. The resulting retention coefficients gate historical representations and reweight target losses. All sample-specific scores are computed offline using training interactions and a fixed KG, so the backbone remains unchanged and no KG access is required at inference. Experiments show gains with a standard sequential recommender and multiple behavior-denoising sequential recommenders.

The Traveling Salesman Problem (TSP) is one of the most extensively studied NP-hard optimization problems. Genetic Algorithm (GA)-based solvers, such as the Edge Assembly Crossover (EAX), achieve state-of-the-art performance on many benchmark instances. However, the scalability of these approaches in massively parallel architectures remains limited because crossover operations involve irregular memory access patterns, graph traversals, and sequential dependencies. Existing GPU-based TSP solvers primarily exploit population-level parallelism and are limited to relatively small problem sizes. This work presents a fine-grain GPU implementation of the partition phase of the Generalized Partition Crossover (GPX) operator for large-scale TSP instances. The proposed approach reformulates GPX partitioning as a graph-parallel problem using coalesced memory layouts, ghost-node transformations, and connected-component analysis. The im- plementation parallelizes the union of parent tours, the splitting of degree- four vertices, the deletion of common edges, and the identification of recombining components using CUDA. Experimental results on instances ranging from 10,000 to 2 million cities demonstrate substantial acceleration over a naive sequential CPU imple- mentation. The proposed GPU partitioning achieves speedups between 48x and 625x while significantly reducing memory overhead. The re- sults demonstrate that operator-level parallelism can substantially im- prove the scalability of GA-based TSP solvers on modern many-core architectures.

Persistent memory makes false information durable: once a false statement is stored, it can be retrieved into future sessions that match it. We measure the cost of this failure mode using plainly worded false assertions generated in a single pass, with no instruction, trigger, or retriever optimization. Poisoning 1.2% of a LongMemEval corpus reduces accuracy from 0.850 to 0.300. A four-stage write-time screening pipeline that reaches 0.832 recall on indirect prompt injection while flagging 1.5% of trigger-word-laden benign text rejects 0 of 360 poisoned memories. We argue this exposes a boundary of content-only screening: distinguishing a false assertion from a true one generally requires external grounding beyond the text itself. We then evaluate provenance-weighted retrieval. The shipped weight is statistically indistinguishable from no defense (p=0.80), while a stronger weight recovers utility only by excluding untrusted content. In a mixed-provenance corpus where untrusted content is mostly benign, accuracy rises from 0.3167 to 0.7000; when the answer-bearing evidence itself arrives untrusted, evidence recall falls to zero and accuracy to 0.0417. Under the measured similarity regime, the additive provenance term has no usable setting: a weight strong enough to resist query-shaped poison is also strong enough to suppress legitimate untrusted evidence. We therefore argue for bounded occupancy constraints at retrieval rather than additive provenance penalties, and release the harnesses, corpora, and aggregate run reports.

Omnimodal generation is central to a wide range of content creation and editing applications. In-context conditioning is essential to this paradigm. It allows diffusion transformers to process text instructions and visual references in a shared attention sequence. However, each reference image introduces thousands of tokens. Computation therefore grows rapidly with the number of references. Existing methods reduce computation through structured sparse attention, which limits interactions between reference and target tokens. This structure also makes the reference K and V independent of the denoising target, allowing them to be computed once and reused across steps. However, it blocks visual references from attending to the text instruction. This substantially degrades instruction following and reference fidelity in multi-reference editing. To resolve this conflict, we jointly redesign the token sequence and attention mask. Our beyond-mask design uses static text anchors to connect the instruction to the reference branch. It preserves exact K and V reuse without adding parameters. However, this direct architectural conversion degrades generation quality. We recover the lost performance through teacher-forced velocity distillation, followed by a short on-policy stage in which the teacher supervises student-visited states. To our knowledge, this is the first use of on-policy distillation for architectural recovery in diffusion models. Across three image-editing benchmarks, our method matches full-attention generation quality. With five reference images, it accelerates the complete 40-step denoising process by 3.92x, while static text anchors introduce negligible runtime overhead; the speedup reaches 5.47x at ten references in our scaling study.

Predictive political question answering (QA), such as predicting how a political actor will vote, goes beyond factual lookup. External political resources offer rich historical evidence, but rarely contain the answer itself. Existing LLM augmentation methods, including actor-profile-based simulation and knowledge graph evidence injection, improve political reasoning but largely treat external resources as knowledge-based evidence, leaving prediction-relevant signals under-modeled. We identify two complementary signals for predictive political QA: actor stances that capture issue-specific preferences, and high-order structure signals that capture indirect dependencies among political actors. We propose PSL, a dual-view framework that converts semi-structured political records into inference-oriented evidence for LLMs. PSL extracts stance signals from question-relevant actor records in a semantic view, and learns structure-aware actor representations from an actor interaction graph in a vector view. Across three real-world datasets and multiple LLMs, PSL consistently outperforms baselines, with ablations confirming the complementary gains of stance and structure signals.

The rise of agentic AI enables LLMs to access diverse user data, raising critical privacy concerns. Prior work on contextual privacy studies whether LLMs regulate information disclosure according to context-dependent norms. However, acceptable disclosure boundaries may vary across users even within the same context. To address this limitation, we introduce \textit{personalized privacy}, which incorporates user-specific disclosure preferences into privacy control. We further present P3Bench~(\textbf{P}ersonalized \textbf{P}rivacy \textbf{P}reservation \textbf{Bench}mark), a novel benchmark extending contextual privacy policies with personalized disclosure policies. Experiments show that prompt-based policies fail to reliably enforce personalized privacy policies, with Qwen2.5-7B and Gemma3-4B showing average policy ignorance ratios of 51.25\% and 74.28\%, respectively. Finally, to address this problem, we propose \textsc{Repair}, a robust inference-time attention head intervention method that adjusts disclosure behavior toward policy-consistent responses. Our method significantly improves adherence to user-specific privacy preferences by reducing cases where the model fails to follow the given policy.

This paper investigates cross-agent specification portability using Oracle-to-PostgreSQL migration as a controlled software transformation task. The study combines two experimental stages. First, a specification-first migration pipeline was evaluated on 1,006 PL/SQL files, of which 623 were successfully regenerated and 380 generated scripts executed successfully in PostgreSQL 16. Second, cross-agent experiments were conducted on a dataset of 1,802 Oracle scripts with corresponding PostgreSQL implementations using Amazon Kiro, Google Gemini, and GitHub Copilot, with Claude Code and Cursor included in the initial single-agent evaluation. Native and foreign specifications were assessed using Token F1, exact match, SQL syntax validity, AST exact match, AST mean similarity, and immediate runnability. The results show that specification size alone does not predict implementation quality and that cross-agent transfer can produce substantial agent-dependent degradation. The strongest replicated case occurred when Gemini directly consumed a Kiro-origin specification, producing a Token F1 of 0.035, SQL syntax validity of 2.33%, and AST mean similarity of 0.015. Rewriting substantially improved Gemini in the tested configuration, compression did not provide a universal benefit, and retrieval-augmented ingestion was the only common strategy represented on the per-agent Pareto frontiers of both Gemini and Copilot. The findings suggest that specifications in heterogeneous SDD workflows should not automatically be treated as agent-neutral artifacts and motivate explicit consideration of specification portability, agent-specific interpretation, and retrieval-based access in multi-agent software engineering.

Imputing physiological time series (arterial blood pressure, blood glucose, etc.) is essential for addressing the missingness that pervades clinical data. Yet modern imputation methods perform poorly in this domain: a recent benchmark found that simple linear interpolation outperformed every learned imputer on real-world clinical signals with realistic gaps. We show that this reflects two properties of physiological missingness that generic imputers ignore: gaps may occur when the signal is clinically extreme rather than typical, and gap lengths can easily span orders of magnitude. To this end, we introduce Curriculum-Aware Interpolate-then-Refine (CAIR), a two-stage framework for physiological time-series imputation. Our key motivation is to learn a coarse base curve and then repeatedly correct it toward physiological realism, rather than predict a gap in a single pass. Consequently, CAIR couples a bidirectional-GRU interpolator with a Transformer refiner that corrects its own estimate over three successive passes, trained jointly under a broad, signal-agnostic random-gap curriculum. We evaluate imputers stratified by gap length and missingness mechanism (MCAR, MAR, NMAR) rather than by a single average, and CAIR is the most accurate under every mechanism on continuous glucose monitoring (AI-READI) and arterial pressure in intensive care (MIMIC-III). Its margin over the strongest baseline grows with difficulty, from 9% under MCAR to 19% under value-dependent dropout, where generic learned imputers are weakest. We further show low reconstruction error alone does not recover the burden metrics clinicians act on: interpolants matching CAIR's error fail to preserve those metrics, imputers that recover them are far less accurate, and CAIR alone ranks among the best on both axes.

Behaviour Cloning (BC) has driven remarkable progress in robot manipulation, yet it is fundamentally limited by its inability to self-improve: a policy that fails cannot learn from that failure without additional human demonstrations. Reinforcement Learning fine-tuning offers a path to self-improvement but has proven difficult to scale to the multi-billion-parameter models underpinning modern robot policies. We propose Q-Planning, which equips a large visuomotor BC policy with a small off-policy Q-function. Because a Q-function estimates value rather than imitates actions, it can be trained on the same successful demonstrations as the BC policy and later absorb both successful and failed deployment rollouts, an asymmetry BC does not have. We exploit this asymmetry to enable value-guided action selection at inference (a single-step Q-weighted average over BC draws) and online self-improvement that fine-tunes only the Q-function, leaving the BC weights untouched. On LIBERO and bimanual RoboTwin, ten iterations of self-improvement lift every benchmark score we tested (LIBERO-10 93% to 99%, RoboTwin 83.8% to 91.4%) and shorten successful episodes on the near-ceiling suites (LIBERO-Object, LIBERO-Goal). On two contact-rich bimanual real-robot tasks, the same loop (BC frozen, no human intervention) improves purely from its own deployment rollouts: stack-cups 40% to 90% and insert-wallet 25% to 80% in five iterations, whereas SFT on successful rollouts alone stalls at 55% and 30%. Under an identical online budget Q-Planning is the only method, among Best-of-N, filtered SFT, IBRL, DSRL, and DAWR, that improves stably from failures without training an auxiliary actor.

Technology change management in large financial institutions depends on risk assessments that are accurate, consistent, and auditable. In practice, many institutions still rely on self-reported questionnaires. Those questionnaires are subjective, easy to game, and poor at separating routine changes from the ones that later trigger major incidents. This paper presents SENTRY, a risk assessment platform that replaces questionnaire-based scoring with a deterministic machine learning pipeline built from gradient-boosted decision trees (XGBoost) and hybrid retrieval-augmented generation (RAG). The system combines structured operational metadata, application dependency graphs, and historical incident records with a hybrid semantic and lexical search over historical change requests. The retrieval step captures the risk signal in unstructured change request text, then compresses that signal into a single scalar feature before model inference. That design keeps the model deterministic and preserves per-prediction explainability via SHAP values. Evaluated on enterprise-scale change data, SENTRY achieves a ROC AUC of 0.87 and 85% overall accuracy, and it detects high-risk changes at roughly 3.25 times the rate of the existing process. We close by examining the architectural trade-offs behind this design and what they imply for the use of machine learning in regulated change management.

Visual prompting (VP) has emerged as a parameter-efficient method for adapting pre-trained models to downstream tasks. However, existing approaches encounter a trade-off between flexibility and efficiency. Some methods apply a fixed prompt to all images, ignoring individual image characteristics, while others introduce auxiliary networks to generate diverse prompts. Although the latter can improve performance, it also significantly increases parameter usage and the potential for overfitting to specific datasets. Furthermore, the auxiliary networks, combined with inherent biases in pre-trained models, limit scalability and generalization. In this paper, we propose Energy-Shaped Visual Prompting (ES-VP), a novel approach that generates image-specific prompts using low-rank initialization and energy-guided dynamic adaptation, achieving superior performance with fewer parameters compared to single-prompt methods. ES-VP directly utilizes the pre-trained model for adaptive prompt generation, ensuring both parameter efficiency and improved generalization. Extensive experiments conducted on five architectures across fifteen datasets demonstrate that ES-VP consistently outperforms current state-of-the-art (SOTA) single and diverse VP methods. For instance, using the CLIP architecture across four datasets, ES-VP outperforms the SOTA method DAM-VP by an average of 2.6\% in accuracy while utilizing 590$\times$ fewer VP parameters, thereby establishing a new benchmark for efficient and generalizable model adaptation.

LGE cardiac MRI is widely used for left atrial fibrosis assessment and ablation planning in atrial fibrillation patients as knowledge of fibrotic tissue regions identified from LGE-MRI is critical for catheter ablation. Often, poor quality images used during ablation planning can cause mis-localization of ablation targets, directly impacting procedure safety and outcome. The decision of whether a scan meets the minimum quality threshold for ablation planning is currently made informally by the reviewing radiologist and is not captured by any automated system, yet it is arguably the most safety-critical output of the image quality assessment (IQA) process. However, variations in image quality caused by noise, motion artifacts, and poor boundary definition significantly compromise the reliability of downstream segmentation and clinical decision-making tasks. Manual quality assessment by expert radiologists is subjective and difficult to scale, while existing automated methods produce scalar scores without interpretable clinical reasoning. In this work, we propose a two-stage vision language model (VLM) framework for clinically grounded image quality assessment of left atrial LGE-MRI. In the first stage, a fine-tuned VLM generates structured radiology-style quality reports predicting five radiologist-defined criteria: Noise, Motion Artifact, LA Boundary Accuracy, PV Region Accuracy, and Under-segmentation Severity. In the second stage, a GPT-based reasoning module maps the predicted quality and reports to a structured quality scores and binary clinical usability decision for ablation planning. We curate a dataset of 60 annotated image slice-text pairs from 20 patients and benchmark four state-of-the-art VLM architectures. InternVL2 achieves the highest criterion-level accuracy (Avg ACC=0.65, PLCC=0.79), while DeepSeek achieves perfect clinical usability agreement (Acc=1.00, kappa=1.00).

Automatic speech quality assessment aims to predict Mean Opinion Scores (MOS) consistent with human subjective perception and is essential for evaluating speech generation, enhancement, and communication systems. For speech signals, especially synthetic speech, distortions often occur locally, and overall perceptual quality is usually dominated by a small number of perceptually salient distortion regions. However, most existing methods are primarily optimized with utterance-level MOS, which provides only coarse-grained supervision and offer no explicit indication of where perceptually important distortions occur. To address this limitation, we introduce explicit distortion localization as auxiliary knowledge for speech quality assessment. We construct the first partially distorted speech dataset with frame-level distortion annotations and train a localization model to generate distortion cues. Building on these cues, we propose DAMOS, a distortion-aware speech quality assessment framework that integrates localization information into the MOS prediction pipeline. Experiments on multiple public benchmarks demonstrate that DAMOS consistently outperforms existing methods and exhibits strong cross-dataset generalization, validating the effectiveness of explicit distortion localization for speech quality assessment.

The radiative transfer equation (RTE) governs thermal radiation in participating media, which is critical to modeling combustion, atmospheric, high-temperature, and radiative thermal management applications. However, due to the inherently high-dimensional nature of radiative transfer, numerical solutions to the RTE induce significant computational cost. This work develops a Fourier neural operator (FNO) as a fast, discretization-invariant, surrogate model for predicting the absorbed heat flux field in participating media. A dataset is generated on a benchmark 2-dimensional problem modeling thermal surface emission into a participating medium with spatially varying properties using Monte Carlo simulations. The FNO is trained to learn the solution operator, mapping the input property fields and geometry to the resulting heat flux field. The trained FNO provides up to 26-times computational acceleration compared to Monte Carlo simulation at equivalent error. Furthermore, as the FNO learns the solution operator and not an image-to-image mapping, the trained FNO accurately generalizes to various spatial discretizations, including those not seen in the training dataset. A spectral analysis of the spatial frequencies present in the heat flux prediction demonstrates the FNO's ability to suppress the high-frequency noise present in the training dataset, while preserving the physically meaningful low-mid frequency spectrum. These results demonstrate the Fourier neural operator surrogate model can provide accurate, computationally efficient, and discretization-invariant predictions for radiative transfer in a participating medium. By learning the underlying solution operator, this approach moves toward the development of surrogate models for radiative transfer capable of generalizing across a broad range of geometries, discretizations, and boundary conditions.

Vision-language models (VLMs) have advanced rapidly in multimodal reasoning, yet recent work shows that their failures often reflect an interaction between visual grounding and downstream reasoning. What remains less clear is how the visual presentation of a task shapes model performance and failure modes when the underlying reasoning problem is unchanged. We study this question in SPaRC, a benchmark for grid-based visual spatial planning, by introducing lightweight input-side scaffolds that preserve the visual modality while making spatial structure more accessible. Across multiple VLMs, these scaffolds improve task accuracy over the original visual setting by up to 34.0 percentage points and further complement GRPO-based training, yielding up to 4.6 additional accuracy points compared with near-zero gains on the original visual input. Analyses on both end-to-end task solving and object detection show that these gains are closely tied to reductions in grounding-related errors, while rule reasoning remains comparatively challenging. We find that visual presentation is a central factor that determines whether VLM benchmarks measure grounded perception, downstream reasoning, or a mixture of both.

Tool-using AI agents turn delegated tasks into provider effects, yet authorization often ends at admission while provider state, delivery, retry, and recovery evolve. A request may change before commit, or response loss may cause a replacement to create a second effect from one approval. We present AID-Guard, a stateful authorization-to-effect closure protocol. It revalidates the approved request and provider state at commit, retains one reservation under ambiguity, and permits release or one successor only after a terminal result or certified no effect with a delivery fence. For supported provider contracts, one reservation yields at most one effect across retry and recovery. To our knowledge, it is the first evaluated agent-authorization protocol to unify these controls in one lifecycle. We implement a Python/SQLite prototype. In a declared loopback MCP domain, 13 live mutations caused no unauthorized provider effects, three concurrent histories were linearizable, and evidence bundles supported public verification and replay. All 210 Stripe provider-contract trials matched predeclared outcomes. Across Stripe and Resend, 40 terminalize-successor schedules, 30 overlapping races, and 10 crash-recovery schedules completed without duplicate effects. Under complete proposer compromise, AID-Guard blocked 44/44 attacks and admitted 44/44 matched legitimate proposals. Its strict exact-manifest profile reduced benign utility by 35.4 to 43.8 percentage points; a typed frontier recovered 9-10 completions without observed unsafe effects. A composition study blocked 20/20 post-admission lifecycle attacks and preserved 8/8 valid or exact-retry executions. The results support authorization-to-effect binding under the evaluated effect-path inventory, provider contracts, and failure schedules.

High-performance GPU kernels underpin modern deep learning and scientific computing. As workloads become increasingly diverse and GPU hardware evolves rapidly, developing efficient methods for automated GPU kernel generation and optimization has become increasingly important. Existing LLM-based methods typically optimize within a fixed implementation space, limiting either optimization flexibility or search efficiency. We propose \textsc{HIERA}, a hierarchical search-space planning framework for GPU kernel optimization. \textsc{HIERA} constructs contract-augmented task specifications, selects an appropriate implementation space across PyTorch operators, CUDA libraries, and custom CUDA kernels, and uses profiling feedback and expert knowledge to guide structured iterative refinement. Experiments on KernelBench across multiple various workload levels and base LLMs show that \textsc{HIERA} delivers stronger overall implementation validity, sample efficiency, and optimization performance than existing training-free methods, while remaining competitive with the training-based CUDA-L1 without additional model training. A case study on a specialized stencil operator from scientific computing further achieves a \(1.53\times\) speedup over cuDNN, demonstrating the potentiality of the general framework beyond standard machine-learning workloads.

LLMs have evolved from language generators to autonomous agents capable of complex, long-horizon tasks. This evolution has produced paradigms including Prompt Engineering to elicit model capabilities, Context Engineering to manage information access, Harness Engineering to organize external tools and resources, and Loop Engineering to support continual reflection and self-improvement. Yet as tasks grow more complex, individual intelligence faces a fundamental limit: many tasks require heterogeneous expertise, interdependent subtasks, parallel execution, independent verification, and persistent state, exceeding any single agent's organizational capacity. Augmenting one agent's capabilities or context cannot resolve this architectural mismatch; intelligence must instead be distributed across specialized agents and organized at the system level. We call this System Intelligence: an agent system's ability to organize and coordinate multiple intelligent components into a coherent, adaptive whole pursuing a shared objective. Achieving it requires more than adding agents; it demands explicit structures to organize work, coordinate heterogeneous agents, and maintain evolving execution states. We introduce Graph Engineering, an emerging paradigm for next-generation agent systems. Unlike prior paradigms that mainly optimize individual interactions or agent-level behavior, Graph Engineering constructs explicit, dynamic, evolving graph structures representing tasks, agents, and system states. These abstractions provide a unified foundation for organizing complex objectives, orchestrating heterogeneous agents, modeling system dynamics, and enabling scalable agent evolution. We systematically review the principles, methodologies, and applications of Graph Engineering for LLM agents. Related papers, open-source data, and projects are collected at https://github.com/DEEP-JLU/Awesome-Graph-Engineering.

Reliable spatial understanding is an important prerequisite for future medical vision-language systems that aim to support radiological report generation and structured image understanding. While modern vision-language models (VLMs) show promising performance on many medical imaging tasks, recent evidence suggests they remain weak in controlled spatial reasoning and often fail to reliably ground spatial relations in image evidence. Given that radiological reasoning hinges on understanding the relative positions of anatomical structures and findings, this spatial weakness poses risks to diagnostic accuracy. We present a modular medical imaging agent for binary spatial relation verification in axial CT slices. Instead of directly predicting spatial answers end-to-end, the system decomposes the task into explicit stages: language parsing, anatomical localization, and deterministic geometric verification. Natural-language queries are converted into structured relation tuples, queried organs are localized with a YOLO-based detector, and the final spatial decision is computed from object centers using deterministic geometric rules. We evaluate the approach on the held-out MIRP spatial QA benchmark and compare it against representative end-to-end VLM baselines. The best-performing hybrid configuration reaches 94.1% accuracy and 94.2% F1, outperforming direct Qwen2-VL prompting by 42.5 percentage points in accuracy, while preserving interpretable intermediate representations and auditable reasoning stages. The results suggest that explicit modular spatial verification can serve as a promising building block for future report-oriented medical imaging agents.

Decentralized Federated Learning (DFL) promises trust-free collaborative learning by replacing the centralized parameter server with peer-to-peer model exchange. However, this architectural shift fundamentally reshapes the threat landscape. Without globally coordinated aggregation, DFL becomes particularly susceptible to backdoor attacks, in which malicious participants implant persistent hidden behaviors while maintaining high clean-task performance. In this paper, we argue that the robustness of DFL has been significantly overestimated. Existing studies rely on simplified threat models, non-adaptive adversaries, fragmented evaluation protocols, inconsistent communication topologies, and ad hoc training configurations, leading to an incomplete understanding of DFL security. To address these limitations, we present BackDFL, a unified benchmark for systematically evaluating DFL under realistic and adaptive backdoor attacks. Through extensive experiments, BackDFL exposes critical failure modes of decentralized learning. Our results demonstrate that both state-of-the-art Byzantine-robust DFL methods and adapted FL backdoor defenses fail under modest malicious participation rates (as low as 15%), especially in heterogeneous settings, while their robustness varies substantially across communication graph topologies.

Recently, open-vocabulary zero-shot 3D scene understanding using vision foundation models has emerged as a promising alternative to data-intensive supervised methods. However, deploying these models in real-world scenarios is severely hindered by their inability to efficiently handle streaming RGB-D inputs and their inherent vulnerability to noise 2D segmentation masks. To address these critical limitations, we propose Stream3Dv2, a novel training-free framework designed for robust streaming 3D perception. Stream3Dv2 processes sequential data through an original nested local-to-historical architecture, capturing multi-view consistency while circumventing the high computational overhead so as to support timely responses. At its core, we introduce a comprehensive geometric-semantic fusion mechanism that resolves geometric noise and semantic ambiguity by explicitly utilizing semantic guidance and formulating 3D segmentation as solving point-and-set merging and partitioning problems. Furthermore, we present an innovative manifold-distance-based point cloud refinement strategy. This approach leverages local manifold graphs for point-to-manifold optimization that mitigates the boundary delineation failures caused by Euclidean-distance metrics, and employs geometric bounding boxes to dynamically activate and update historical instances for achieving rapid manifold-to-manifold refinement. Extensive experiments on public datasets demonstrate that Stream3Dv2 consistently outperforms existing baselines in foundational open-vocabulary streaming 3D segmentation and detection. Finally, we show that integrating our framework with an LLM-based agent enables advanced language-driven 3D scene understanding, underscoring its potential for open-world embodied intelligence. Code will be updated at https://github.com/SubmissionsIn/Stream3D.

Deploying vision-language models (VLMs) on mobile devices is challenging due to their significant memory and compute requirements. We present a framework for quantizing VLMs for efficient inference on resource-constrained hardware. Our approach combines a quantization pipeline that uses the model itself to generate training data and does not require access to the training setup, with a novel 2.7-bit-per-parameter format supporting efficient execution on Arm CPUs. We validate our approach by compressing the Llama 3.2 11B Vision Instruct model to 3.7 GB with 8-bit activations, preserving strong performance on a set of standard visual question answering tasks.

Medical image-text data can expose protected health information (PHI) through both visible image content as well as accompanying text, creating a barrier to privacy-preserving medical AI systems. This risk is especially prominent in multimodal systems, where images, questions, reports, and clinical context may enter training, evaluation, or inference pipelines. Existing medical vision-language benchmarks primarily emphasize task utility, while de-identification methods are often evaluated separately from downstream reasoning. We introduce ClinX, an end-to-end multimodal PHI sanitization framework for medical image-text data. ClinX detects visible identifiers with optical character recognition (OCR), constructs binary PHI masks, and applies ClinX-PRISM, a no-skip generative restoration module with privacy-oriented post-processing for burned-in identifier suppression. In parallel, text-side PHI is reduced through progressive de-identification levels: regex masking, context-aware masking, and rewrite-based sanitization. We evaluate ClinX in medical visual question answering (MedVQA), jointly measuring PHI leakage and downstream utility across image-side, text-side, and combined de-identification settings. Results show that OCR-only masking is not sufficient as a standalone solution, and restoration-based sanitization better preserves clinically relevant visual context while sharply reducing recoverable PHI.

Visual world-model agents such as DreamerV3 act through a recurrent latent state rather than a single observation, which weakens frame-wise observation attacks and makes their perturbations vary sharply over time under a strict per-frame perturbation constraint. We study white-box, causal, online attacks on such agents and propose Critic-Induced Value-Subspace Attacks (\textbf{CIVA}). Our key observation is that, along a rollout, critic-guided perturbations concentrate in a low-dimensional subspace induced by the victim's own critic. Based on this observation, CIVA first probes the frozen victim offline with critic-guided PGD and extracts a low-rank value-subspace by SVD. At test time, it optimizes only the subspace coefficients, smooths them with an exponential moving average (EMA), and maps them back to pixels. This design attacks value-sensitive recurrent dynamics while keeping the online optimization cheap and temporally coherent. Extensive experiments on DMC walker walk, Atari Pong, and Crafter show that CIVA consistently outperforms five recent methods; on DMC walker walk, it achieves the largest reward drop of 26.07\% while keeping temporal variation low, with TempAbs of 0.646.

Large Language Models (LLMs) are moving from code completion toward repository-scale agents that retrieve context, edit files, execute tools, and participate in security-sensitive workflows. The evidence for these systems, however, remains divided between software engineering evaluations centered on functional task completion and software security evaluations centered on vulnerability detection, secure generation, or exploit-oriented validation. This evidence-centered structured survey synthesizes representative work available through May 31, 2026 across software engineering tasks, software security tasks, adaptation mechanisms, artifact granularity, and evaluation design. In addition to a task taxonomy, we introduce an assurance framework that separates functional correctness, security, operational reliability, evidence provenance, and agent authority. The review shows that execution feedback and repository access can substantially improve engineering task completion, but do not by themselves establish security; conversely, static-analysis labels or vulnerability-classification scores rarely establish deployable correctness. We identify recurring validity threats--weak test oracles, duplicated and temporally leaked data, changing agent harnesses, proxy-only security checks, and under-reported budgets and human intervention--and derive a minimum reporting protocol for cross-study comparison. The resulting research agenda prioritizes jointly secure-and-functional benchmarks, repository-scale threat models, calibrated human oversight, longitudinal maintainability evidence, and reproducible agent evaluation. The central conclusion is that model capability should be judged as an assurance case supported by task-appropriate evidence, rather than by a single benchmark score.

As large language model (LLM) agents move from conversation to executing code, reading local files, and orchestrating external tools, a single agent hijacked by a malicious third-party skill can cause data exfiltration, privilege escalation, or cascading compromise. We argue that agentic risk is progressive: it can enter at four loci of the agent control loop--skill admission, invocation-time intent, execution-time effect, and post-action consequence--while a denied dangerous objective can reappear across surface forms, tools, or turns; existing safeguards are typically local to one lifecycle boundary or one call. Guided by this threat model, we present ClawSentry, an open-source, framework-agnostic security supervision gateway for agent runtimes. Before a skill package is ever executed, First-use Skill Package Review (FSPR) audits it under a deterministic evidence floor, escalating unresolved cases to bounded read-only agentic review (locus A). At runtime, a three-tier progressive decision engine--a deterministic L1 layer, a rule-anchored L2 semantic reviewer, and a read-only L3 evidence-seeking agent--spends contextual review only on the residual ambiguity, while a session-level anti-bypass mechanism recognizes tool-switching and rephrased retries (loci B--C); a post-action path feeds high-severity evidence non-retroactively into later review (locus D). An Agent Harness Protocol (AHP) abstraction applies one policy across Codex, Claude Code, Kimi CLI, and Gemini CLI without modifying agent internals. On SkillInject with Codex/GPT-5.4, contextual ASR falls from 39.55% to 2.61% while contextual TSR moves only from 83.78% to 83.05%. Across five Work Agents on the full SkillsSafety benchmark, ClawSentry confines ASR to 9.09--15.03% from 33.5--49.7% unprotected, and aggregate TSR on clean skills remains 98.7%.

While multimodal large language models (MLLMs) extend model capabilities beyond text, they also make safety alignment increasingly challenging. Multimodal safety alignment methods must address cross-modal jailbreaks, safety-awareness failures, and over-sensitive refusals. However, existing methods often rely on retraining or internal-state inspection, limiting their applicability to deployed closed-source MLLMs and motivating test-time safety alignment. We analyze this setting and identify two key obstacles, utility dominance and reasoning inertia, which cause models to overlook latent risks or follow malicious reasoning trajectories. Guided by these insights, we propose ReFrame, a training-free multimodal input reframing framework where two agents share a lightweight locally deployed MLLM: the evidence-generation agent constructs complementary risk and utility evidence, and the rewrite-and-routing agent converts it into a safe proxy prompt and image-routing decision before calling the downstream MLLM, without modifying it or accessing its internal information. Experiments across multiple MLLMs and benchmarks show that ReFrame improves jailbreak defense, safety awareness, and oversensitivity reduction while preserving multimodal utility.

Multi-modal object detection is essential for robust scene understanding in challenging conditions, including low-light and adverse environments. Recent vision foundation models (e.g., DINOv3) have exhibited strong representation capabilities, yet adapting them to multi-modal scenarios remains challenging. Existing dense cross-modal fusion strategies often force heterogeneous modalities to interact indiscriminately, which may introduce redundant information and disrupt the valuable pre-trained representations. To address this issue, we revisit multi-modal fusion from the perspective of socialized learning and propose adapter to DINOv3 (A2DINOv3), a multi-expert collaboration framework with a Socialized Collaboration Protocol (SCP). Specifically, RGB and infrared branches are modeled as heterogeneous experts that independently preserve their specialized knowledge while exchanging complementary information through selective and constrained interactions. This design mitigates harmful cross-modal interference and prevents degradation of pre-trained priors during adaptation. Furthermore, a zero-initialization strategy is introduced to gradually activate cross-modal collaboration, enabling a smooth transition from modality-specific learning to cooperative representation learning. Extensive experiments on four multi-modal benchmarks, including aerial detection (GAIIC), autonomous driving (FLIR), low-light surveillance (LLVIP), and diverse real-world scenarios (M3FD), demonstrate that A2DINOv3 consistently achieves state-of-the-art performance in multi-modal object detection.

Fusing prior knowledge with data-driven learning is attractive where data is scarce, yet no controlled account says when it helps, is redundant, or harms. We benchmark one fixed hand-crafted knowledge source, a pinned bank of Gabor targets injected only during training at $\sim$2\% overhead, against data-driven alternatives (SimCLR, SimSiam, DINO, ImageNet transfer, augmentation, learned teachers) under one frozen recipe with fixed subsets: 13 datasets, 9 backbones, 150 to 1.28M images, 32--224\,px, 2.5M--86M parameters ($\computeCells$ classification configurations over $\computeRuns$ runs, plus segmentation and detection transplants). Across the training-time combinations we measure, three outcomes recur (decision-level fusion differs). Different-\emph{currency} sources can stack: the prior composes with DeiT augmentation on attention backbones and is worth $+26$ points to ViT-B/16 at $224$\,px, $+6.7$ at twice that budget. Same-currency sources substitute: against effective self-supervised pretraining, the combination never usefully exceeds the better single source. Fusing at full strength into an already-informed initialization interferes in proportion to what it carries: ImageNet transfer, $-15$ to $-17$ points, removed by a weaker auxiliary weight. Frozen-feature diagnostics measured on each source alone separate these outcomes retrospectively but do not predict them: a rule built on them calls one of nine unseen pairs. At a practitioner's own label budget, the frozen-feature gain predicts the end-to-end gain to within $0.17$ points across 30 cells and seven datasets; the underlying decomposition, $Δ= G + \readout(\mathrm{base})$, holds in sign on $\auditRate\%$ of testable cells and is called an unseen backbone family's feature gain in advance. The project page is https://amughrabi.github.io/MomentAux.

LLM-as-Judge systems can produce multi-dimensional evaluations, such as trustworthiness, reliability, and factuality, and these outputs are often interpreted as independent evidence. We test this assumption for a common pair of judgments: trust scoring and binary truth classification. On correctness-controlled QA, LLM judges align trust scores with truth verdicts more tightly than human behavioral reference, suggesting weaker separations between trust and truth judgment. We then apply stress tests by changing only source cues of identical QA between Human and AI. Source attribution shifts not only trust scores but also truth verdicts and logit-derived correct-side probabilities. Results show that current LLM-as-Judge protocols should not treat trust scores as independent evidence for truth judgments.

Retrieval-Augmented Generation (RAG) grounds Large Language Model (LLM) outputs in external knowledge, but RAG systems usually trust whatever they retrieve, creating a Security-Reliability Gap: high semantic relevance does not guarantee factual truth. Adversaries exploit this through knowledge poisoning, inserting malicious documents to cause targeted misinformation. We propose an Evaluation Agent, middleware that combines Natural Language Inference (NLI) factual verification, a five-signal poison detector with relevance-weighted aggregation, and a Trust Index T = 0.4 F + 0.35 C + 0.25 (1 - P ) with a non-linear dampener for high-contamination contexts. On TruthfulQA with Llama 3.3 70B, the agent reaches 91% accuracy and 100% precision, with 100% recall on instruction injection, while in-place edits, such as entity swaps, remain hard to detect. Across three LLMs the Trust Index stays discriminative, with a Receiver Operating Characteristic Area Under the Curve (ROC-AUC) of 0.73 to 0.81; generation style matters more than model size, and per-LLM threshold calibration restores baseline competitive accuracy, whereas a weaker FEVER result shows that cross-dataset generalization requires domain-specific calibration. In a software-engineering use case, a secure-coding assistant over guidance from the Open Worldwide Application Security Project (OWASP) Top 10 and the Common Weakness Enumeration (CWE), the agent reliably blocks instruction injection of unsafe advice (F1 92%), while contradiction and subtle semantic weakening remain hard. Throughout, the agent measures detection of poisoned context before generation, not whether the LLM adopts the injected misinformation. We release the proposed approach, attack generator, and experimental artifacts at the link: https://github.com/GPT-Laboratory/TrustworthyRAG.

Mufwene's ecological model locates language evolution in competition among variants contributed by individual idiolects and in speakers' selection from linguistic material made available through interaction. Large language models (LLMs) complicate this architecture without requiring the locus of selection to move away from human speakers. This article argues that LLMs are best treated as distributional mediators: they aggregate language produced across human populations, transform its distribution through training and post-training, and redistribute model-specific outputs at scale. I call the resulting ecological process algorithmic reweighting of the speaker-accessible distribution: model mediation can alter the relative frequencies with which competing variants reach human selectors. Emerging evidence on model-specific linguistic profiles and lexical uptake is consistent with parts of this pathway, but does not establish inevitable convergence. Human social evaluation remains decisive: model-associated forms may diffuse and become conventionalized, become socially recognizable as 'AI-like' and subsequently avoided, or fail to diffuse in the first place. The proposal extends Mufwene's feature-pool ecology one step upstream of speaker selection and yields testable predictions about uptake, model-version effects, convergence, and social reversal.

This technical report presents AudioWorldSim, an open-source platform designed to generate realistic binaural audio datasets and advance research in audio-based machine learning, particularly world models. Built as a custom extension of Meta's SoundSpaces 2.0 platform, AudioWorldSim leverages their comprehensive acoustics framework, but focuses on the automatic rollout of random agent navigations, as well as implements crucial fixes to how continuous sound is composed. AudioWorldSim is made publicly available to the research community at https://github.com/Luizerko/AudioWorldSim to facilitate reproducibility.

Pathology foundation models (PFMs) are increasingly used as general-purpose backbones, yet existing benchmarks cannot systematically diagnose their whole-slide cellular representation capabilities, including the decodability of cell-type information and the transferability of such information across tissue sections, datasets, and anatomical organs. We introduce CellPath-Bench, a cellular-resolution benchmark that evaluates frozen PFMs themselves. Following quality control of 52 candidate Xenium datasets, we construct a panel of 25 spatially aligned H\&E--Xenium tissue sections spanning 11 organs and 7,079,283 cells, harmonized into fine- and coarse-grained taxonomies. CellPath-Bench samples frozen WSI feature maps at registered nuclear coordinates and evaluates them using standardized multiclass linear probes. Cell Representation Advantage (CRA) measures the within-section advantage of nucleus-anchored representations over patch-level mean pooling, while Cell Representation Transferability (CRT) characterizes the generalization of cell-type decodability across tissue sections, datasets, and organs. We benchmark 30 pathology-specific and general-purpose foundation models through 304,920 runs across spatial readouts, magnifications, taxonomic granularities, and evaluation protocols. The results reveal substantial model-dependent differences in cell-type decodability and its cross-domain generalization, yielding distinct multidimensional capability profiles. CellPath-Bench provides a standardized framework for auditing cellular information in frozen PFM representations.

Agentic large language model (LLM) systems are reshaping scientific workflows in chemistry and drug discovery, but evaluating their open-ended, tool-augmented outputs remains a fundamental bottleneck. Reference-based metrics such as BLEU and ROUGE fail to capture semantic correctness, while expert human evaluation does not scale to the iteration speed these systems demand. The LLM-as-a-Judge paradigm has emerged as a scalable alternative, but existing drug discovery benchmarks deploy LLM judges without validating their alignment with human experts. In this work, we present an LLM-as-a-Judge evaluation framework for ChatInvent, an agentic drug discovery assistant deployed at AstraZeneca, with four contributions. First, we define four output-quality evaluation dimensions---Completeness, Relevancy, Structural Clarity, and Scope Adherence---alongside deterministic Tool Call Correctness checks. Second, we validate the judge through a human alignment study with five expert annotators, comparing Gemini 3.1 Pro, Claude Opus 4.7, GPT-5, and Llama 3.1 70B as candidate judges. Third, we optimize the best-performing judge using few-shot demonstrations of human-annotated examples, improving alignment with the human majority vote from 0.80 to 0.86. Fourth, applying the optimized judge to 70 held-out questions, we surface concrete limitations and find that informal phrasings do not systematically degrade output quality; if anything, it is helpful to have the LLM rewrite the original question before querying the agent. Our framework provides a reusable template for human-aligned evaluation of agentic systems in scientific domains.

With the progression in open and disaggregated 6G radio access networks, it is expected that the system will be able to host multi-vendors. In order to host multi-vendors, it is essential that AI-assisted control loops remain safe, verifiable, and auditable under concurrent operator intents and untrusted model inputs. The existing studies address the agentic coordination, formal intent constraints, zero-trust prompt verification and cryptographic accountability in isolation, which leaves pre-realization safety, continuous semantic verification and cross-domain audit incomplete when used individually. In this regard, we propose zero-knowledge auditable control and zero-trust verifiable agentic intent architecture ($Z^2$-ACT), which integrates the aforementioned four primitives across the non-real-time and near-real-time RICs. We encode the typed Intent Contracts as operator goals while the large language model inputs are only admitted after a practical adversarial intent check. The skill sequences in the proposed study are released only when a self-management gate is satisfied while every successful commit is recorded as a binding commitment with a zero-knowledge proof. Our experimental evaluation on public ColO-RAN measurements compares the full architecture against targeted ablations and a conventional reinforcement-learning baseline. A live large language model is used in the non-real-time path to translate operator intents into Intent Contracts; we report translation accuracy, the rate of invalid or hallucinated contracts, non-real-time latency, and behavior under adversarial or misleading intents. Near-real-time control remains trace-driven on the public KPM sequences. Results indicate improved actuation filtering and attack resilience at modest latency and signaling cost inside the near-real-time envelope.

The transport of dangerous goods by sea is a high-consequence activity governed by the International Maritime Dangerous Goods (IMDG) Code, a complex regulatory framework where errors in classification, packaging, stowage, or segregation can result in fire, explosion, toxic release, or loss of life or vessel. Correct compliance requires accurately interpreting hundreds of pages of interacting provisions, updated on a two-year amendment cycle. Practitioners increasingly use Large Language Models (LLMs) as decision-support tools, yet no systematic evaluation exists of whether they can reliably interpret IMDG requirements for safety-critical use. This paper introduces DGEval, the first benchmark for evaluating LLM knowledge of IMDG Amendment 42-24. Built from expert-written questions on the NCB Hazcheck e-learning platform and structured lookups from the Dangerous Goods List (DGL), it comprises 1,678 questions across multiple-choice, open-ended, DGL lookup, and regulatory identification tasks. We evaluate 13 models from six providers across multiple thinking configurations, including one maritime domain-specific fine-tuned model, and test the effect of web search. Although the best-performing model exceeds the human practitioner baseline on multiple-choice questions, all models are weakest in the operationally safety-critical areas of stowage, segregation, and regulatory recall. These results indicate that LLMs may support compliance tasks, particularly structured DGL lookups with web search, but unreliability in operational areas and regulatory-text recall means human oversight and authoritative source verification remain necessary before deployment in any safety-critical context. DGEval is designed as a safety assurance instrument to be applied continuously as models evolve, not as a settled characterisation of current capability.

Long-horizon manipulation tasks are often only partially ordered. For example, when assembling an electronic device, the battery and circuit board may be installed in either order, but both must be in place before the enclosure is closed. Recovering such dependencies enables robots to flexibly reorder subtasks while preserving task validity. Existing approaches typically infer task structure from human demonstrations using both temporal and symbolic supervision. However, symbolic predicates require explicit grounding, which is difficult to obtain in realistic settings. In this work, we present an approach for extracting task dependency structures from demonstrations using only simple kinematic graphs and distributions over relative object poses. From these representations, our method estimates pairwise task-step-dependency probabilities and uses them to initialize the edge weights of a precedence graph. We then introduce a filtering pipeline that converts this graph of probability estimates into the final task dependency graph. We evaluate our approach on an existing benchmark and on a new dataset comprising longer tasks with more complex dependencies. We find that our method recovers more accurate task structures from fewer demonstrations than the baselines. Finally, we demonstrate that the inferred graphs can be used to generate multiple valid robotic execution orders for the same task.

Vehicle-to-Everything (V2X) cooperation enables beyond-line-of-sight perception, mitigating occlusions in single-vehicle sensing. However, existing V2X benchmarks provide limited support for closed-loop evaluation and language-grounded supervision, hindering the development of vision-language models (VLMs) for end-to-end cooperative driving. To address these limitations, we introduce V2XBench, a simulation platform featuring synchronized ego--roadside sensing and closed-loop evaluation, together with Chat-V2XBench, a progressively structured VQA dataset for cooperative reasoning. Building upon this benchmark infrastructure, we propose AURORA, an end-to-end cooperative driving framework. Equipped with a dual-view perception architecture, AURORA mitigates spatial and semantic discrepancies across ego and roadside viewpoints through a query-level Cross-View Query Alignment and Fusion (CQAF) module. Leveraging the resulting unified tokens, a LoRA-adapted VLM bridges semantic reasoning and generative trajectory planning. Extensive closed-loop evaluations on V2XBench demonstrate that AURORA achieves state-of-the-art performance in heavily occluded scenarios, with a Route Completion rate of 98.21% and a Driving Score of 76.02, while requiring low roadside communication bandwidth. Ultimately, this work pioneers an extensible V2X--VLM paradigm, paving the way for next-generation cooperative autonomous driving.

We present PhysCaP, a Physics-Informed Code-as-Policy agent for active perception in robotic manipulation. While vision-language-action policies excel at imitating demonstrations, they rely on passive observation and fail to infer latent physical properties critical for manipulation. PhysCaP augments code-as-policy frameworks with a physics-informed exploration layer that enables explicit information-seeking through interaction. It introduces training-free physical property extraction modules that estimate object mass and stiffness from robot proprioception without additional sensors. To balance exploration costs and the efficiency of information obtained, PhysCaP employs a dual-agent design: a Planner that decides when to explore and when to stop, and a Prioritizer that filters implausible interactions and ranks the remainder using a heuristic priority score, enabling efficient, targeted exploration. We evaluate PhysCaP on real-world tabletop manipulation tasks (searching for hidden objects, detecting empty cans, and finding ripe avocados) and a simulated task in LIBERO. The results show that existing passive and naive interactive baselines either fail when physical properties are hidden or over-explore, whereas PhysCaP achieves comparable performance with fewer interactions and reduced execution time. Ablation studies further validate the effectiveness of the proposed physical property extraction modules. Project page: https://physcap.github.io

LLM agents are emerging as an important paradigm for real-world tasks that require reasoning, tool use, and sequential decision-making. As these agents operate over longer horizons, runtime intervention offers a way to improve reliability without retraining the underlying actor. Failure detection alone is insufficient. Effective intervention must also provide a useful direction for recovery. Existing approaches often rely on an expert solver or a critic that generates task-specific corrections, incurring either the cost of another capable solver or the capacity demands of a task-capable critic. We introduce Comparison-Only Tiny Advisor (COTA), a comparison-only framework for constructive runtime intervention. In COTA, a tiny comparator judges whether sampled alternatives lead to better continuations than the actor's proposal, and repeated comparisons determine when intervention is warranted. We train the comparator using pairwise supervision constructed from same-prefix counterfactual branches. Preferred alternatives are returned as non-binding advice, leaving the original actor to replan. Across WebShop, ALFWorld, and tau^3-Retail with three actors, COTA improves all nine evaluation settings and outperforms the compared baselines. These results show that constructive runtime intervention can remain effective even when the auxiliary model has substantially weaker task-solving capability than the actor.

Real-world fault analysis in 5G and emerging 6G networks demands domain expertise to analyze free-text diagnostics, including root-cause explanations and recommended actions. LLMs have emerged as a promising approach to automating this, yet whether lightweight, edge-deployable models are capable of performing in-depth free-text diagnostics remains an open question. While existing benchmarks rely on restrictive MCQs with fixed answer keys, this paper evaluates 5G domain understanding and fault analysis in a free-text generation format. Transitioning to this paradigm requires evaluating lightweight, edge-deployable AI models on open-ended diagnostic reasoning, alongside a dependable framework to validate these text outputs at scale. To address this we evaluate three lightweight LLMs, Claude-Haiku-4.5, GPT-5.4-Mini, and Gemini-3.1-Flash-Lite, on free-text 5G domain knowledge and fault-analysis tasks across three benchmarks, TeleQNA ORAN FT, 5G-Faults FT, and TeleInter FT. Three independent frontier judges score outputs, and pairwise inter-judge agreement is measured as an empirical test of the LLM-as-Judge methodology. All three models reach at least 90% accuracy on fault diagnosis, while zero-shot recall of 3GPP and O-RAN specifications remains the critical gap, with all models scoring below 60%. Mean inter-judge agreement is at least 0.90 across all runs, indicating that multi-judge LLM scoring produces consistent, reproducible grades for open-ended telecom responses. Operationally, Gemini-3.1-Flash-Lite offers the best efficiency trade-off, combining competitive accuracy with the lowest inference cost and latency, making it the most suitable candidate for production telecom deployments.

Quantization is widely used to deploy large language models, but its effect on uncertainty behavior, such as confidence, margins, and abstention, is rarely treated as a primary objective. We frame calibration-data selection for quantization as a target-dependent uncertainty-preservation problem. Different deployments emphasize different regions of the input distribution, yet prior work mainly optimizes accuracy-oriented compression metrics or adjusts scores after quantization. We formalize this goal with distributional and boundary preservation risks, and provide a simple mixture-mismatch argument explaining why no single calibration recipe should be expected to fit all targets. We introduce Doubt-Preserving Quantization (DPQ), a lightweight pre-quantization recipe family that uses full-precision predictions to construct target-aligned calibration mixtures of high-doubt examples and generic anchors. Across 8 language models, 9 NLP benchmarks, and 22 comparison methods, the leading fixed recipe changes with the preservation target: DPQ-r75 leads on SQuAD2 answerability-boundary preservation, while milder or single-signal variants, including DPQ-r50, confidence-only, and entropy-only, better preserve broad multiple-choice QA behavior. These results show that calibration data should be selected for the specific full-precision score behavior a deployment needs to preserve, rather than treated as a fixed quantization detail.

Generative recommendation represents each item with a sequence of discrete Semantic IDs (SIDs) and predicts the sequence to retrieve the next item. Typical systems use multi-level residual quantization, which increases autoregressive decoding cost and creates a large hierarchical space that may be sparsely occupied. Static codebooks also become misaligned with current traffic as new items arrive and exposure distributions change. We propose a single-level large semantic codebook that replaces multiple residual semantic codes with one semantic token while retaining a separate collaborative disambiguation token to reduce item collisions. We further introduce an exposure-aware dynamic update mechanism based on temporal weight decay, exponential moving-average center updates, and an exposure-weighted penalty on SID changes. We also develop an offline evaluation framework covering representation quality, code utilization, cluster load, full-SID collision, and temporal stability. On two public datasets, the two-level SID improves mean Recall@10 by 5.0%-8.8% and mean NDCG@10 by 4.1%-5.1% for OneRec-V1, and by 7.1%-8.7% and 3.8%-8.5%, respectively, for OneRec-V2. Dynamic updating provides further gains on KuaiRec. Across three serving architectures, the shorter SID reduces estimated autoregressive-decoding FLOPs by 47.93%-48.70% and increases single-card QPS by 28.57%-47.0%. A five-day online A/B test serving 2.5% of production traffic improves the primary consumption metric by 0.792%.

Mobile AR frameworks attach a metric pose prior to every casual phone capture, and turning it into reconstruction-grade poses cheaply on CPU is the step before novel-view synthesis. The least a refiner owes an accurate prior is not to make it worse. The workhorse refiner does. On 15 ScanNet++ iPhone room captures, COLMAP triangulation plus prior-seeded bundle adjustment degrades an accurate ARKit prior in all 15, 0.55 degrees to 0.74 degrees by scene-mean. The cause is the seeding. Structure is triangulated from the prior before anything is optimized, so the prior's error is baked into the structure the optimizer trusts. We remove the triangulation. Every keypoint owns a scalar depth along its own back-projected ray and each match contributes two symmetric cross-projection residuals, so structure is re-expressed at every iterate. The same solve holds the room prior at 0.57 degrees and never fails in 330 perturbed room runs, and at object scale reaches 0.265 degrees/1.80 mm from a prior at 0.456 degrees in a median of 10 s per scene on one CPU, against 2.5 GPU-hours for a learned refiner. Because no structure is committed, the objective also admits graduated non-convexity, which measures how deep the defect goes. Classical refinement collapses past 1-2 degrees of prior error, barely beyond a real ARKit prior, and no classical refinement arm survives 32 degrees. Ours recovers 425 of 425 runs through 16 degrees/80 mm and 85% at 32 degrees/160 mm, and perturbed rooms through 32 degrees. Nominal object-scale accuracy is on par rather than better, on a benchmark at its own noise floor, where classical bundle adjustment is a strong baseline absent from the literature. One scene fails for every solver already at zero perturbation. Re-mapping from position priors matches us in the prior's frame but discards it, so it cannot exploit a prior worth keeping or be warm-started.

Multimodal LLMs apply the language model interface to visual inputs, where ordinal regression tasks such as age estimation, image quality assessment, and disease grading require autoregressive decisions over ordered class labels. We ask whether MLLMs reliably convert internal ordinal evidence into ordered digit-token outputs. Across four ordinal benchmarks and four MLLM backbones, ordinal labels are linearly recoverable from hidden states with Spearman correlation up to 0.938, and a task-designed prompt further sharpens this structure. Yet native digit-token outputs weakly expose it: the unembedding matrix filters the ordinal direction, and the digit-token row space retains below 1.15% across all 16 model-dataset combinations, with a 16 to 77 absolute-point accuracy gap between linear-probe and native outputs. We introduce Ordinal Lens Alignment (OLA), a frozen-backbone inference-time method that trains lightweight W_S-anchored lenses on mid-to-deep decoder layers, fuses them into an ordinal distribution, and corrects only digit-token logits at generation. OLA outperforms the SOTA LoRA-tuned OrderChain baseline in most settings while keeping the MLLM frozen, surpasses discriminative ordinal baselines in most cells, and improves over an offline lens in every setting.

Reservoir computing (RC) couples a fixed recurrent dynamical system with a trained lightweight readout, but this efficiency is partly lost during hyperparameter selection: the recurrent gain, input scale, and leakage rate determine the reservoir's stability and temporal processing regime and are usually tuned through many rollouts. We introduce a deterministic, pilot-informed selector for leaky linear reservoirs followed by coordinate-wise nonlinear features. Free probability yields cross-lag propagation coefficients that summarize how the reservoir mixes past inputs. In the large-width limit, these coefficients define a deterministic temporal kernel that approximates the finite-reservoir feature geometry. Kernel ridge regression on a short labelled pilot sequence therefore ranks candidate operating regimes without instantiating or rolling out a reservoir, and the selected configuration transfers across widths. Across ten synthetic temporal benchmarks, zero-rollout selection obtains a mean deployment score of $0.772$, compared with $0.774$ for exhaustive simulation-based search, while avoiding $156\,600$ selection rollouts. With a small rollout budget, the proposed ranking provides the strongest mean performance at every tested budget and reaches the exhaustive reference using $4.8\%$ of its rollout cost. On four public electricity-transformer-temperature (ETT) forecasting datasets, five retained candidates recover the exhaustive operating point on three datasets. On multivariate cellular-traffic forecasting, 15 rollouts per cell reach the 462-rollout exhaustive reference and outperform random search and Bayesian optimization at low budgets. These results position free-probability kernels as deterministic surrogates for selecting reservoir operating regimes when validation rollouts are scarce.

Narratives are central to how social communication is framed, making their detection critical for understanding and analysing public discourse. Prior work has explored narrative detection and extraction across diverse domains; however, migration narratives remain significantly understudied, primarily due to the absence of dedicated annotated datasets. Furthermore, public communication has recently shifted towards video-centric platforms, where narratives are conveyed through multimodal signals and consumed at scale. Despite this shift, narratives in videos remain largely unexplored. To bridge these gaps, we introduce MigrationNarrate, the first multimodal dataset for detection of migration narratives in the UK, consisting of 1,115 YouTube video transcripts annotated using a two-level taxonomy of 12 migration super-narratives and 53 narrative labels. This paper details the dataset design, collection, and annotations; together with benchmark results using a combination of pre-trained encoder models and both open- and closed-source Large Language Models. Finally, a thorough error analysis offers insights for future work.

Graph neural networks (GNNs) are routinely employed for short-range forecasting on multivariate time series with a spatial graph structure. Despite the availability of many alternative datasets, method innovations within this domain are predominantly assessed against a rather limited set of benchmark datasets, most notably Chickenpox, PedalMe, WikiMaths, METR-LA, and PEMS-BAY. The evaluation protocols contain baselines spanning from historical averages to classical machine learning approaches. These baselines often show competitive performance compared to GNNs. In the present work, we take a step back and analyse the benchmark datasets via classical time series methods to uncover why spatially-unaware linear models pose a stronger competitor than previously reported, casting further doubt on the discriminative reliability of the aforementioned widely adopted datasets. Our statistical analysis provides a toolset for identifying significant spatial and temporal correlations, while revealing a structural bias introduced by first-order differenced datasets. We therefore recommend reducing the over-reliance on such datasets for method comparison, and instead advocate for more rigorous statistical evaluation. By applying the results of our analysis to a simple hybrid model, we show how our methodology can lead to novel ways of developing GNN models

Effective social interaction requires agents to translate mental state inferences into coordinated behavioral signals across verbal and nonverbal channels simultaneously. Yet existing benchmarks evaluate theory of mind (ToM) reasoning and embodied behavior in isolation, leaving unmeasured the gap between social inference and social action. We introduce MOSAIC (Multimodal Orchestration of Social Action, Inference, and Communication), a controlled benchmark in which two embodied agents interact across cooperative and competitive scenarios requiring integration of verbal statements, spatial trajectories, gaze direction, and facial expression under systematically varied ToM constraints. Evaluating 13 models, including 11 VLMs, across 200 trials per model, we find that VLMs fail to produce behaviors consistent with the expected outcomes under ToM-order constraints, and that imposing explicit ToM-order constraints produces no reliable behavioral change aligned with the specified reasoning level. Signal-level analysis reveals two sequential bottlenecks: most models cannot produce directionally coherent nonverbal signals, and even when signals are present, VLM agents fail to interpret others behaviors and react to them. PCM-LLM, included as a structured architectural reference point with an explicit ToM module, succeeds across all conditions, suggesting that explicit belief-action coupling is a sufficient ingredient for this class of tasks.

Video Joint Embedding Predictive Architecture (V-JEPA) learns powerful spatiotemporal representations from video through self-supervised latent feature prediction. However, V-JEPA is built around random-mask completion and deterministic regression, making it fundamentally ill-suited for autonomous driving planning that demands future-directed prediction tightly coupled with action. To address this, we rethink the V-JEPA paradigm and present WA-JEPA, a V-JEPA-native world-action model designed for autonomous driving planning. Instead of random spatiotemporal masking, WA-JEPA employs hybrid future-masked pre-training, where the model infers future latents from observed context. Departing from deterministic regression, we recast future prediction as conditional flow matching over latent futures, which substantially improves the model's ability to generate plausible future latents for downstream planning. Finally, a joint future-action predictor is proposed to denoise future scene tokens and ego trajectories together in a unified spatiotemporal latent space, allowing action supervision to directly shape planning-relevant world representations. Pre-trained on nuPlan videos and fine-tuned on NAVSIM, WA-JEPA reaches 91.7 EPDMS on NAVSIM-v2, surpassing the strongest end-to-end and world-action baselines by 1.6 and 1.3 EPDMS, and, without HUGSIM-specific fine-tuning, attains the best HD-Score of 0.4462 on the closed-loop HUGSIM benchmark under the same evaluation protocol. These results validate V-JEPA-native world-action modeling as a powerful and scalable paradigm for autonomous driving planning. Code is available at https://github.com/AFARI-Research/WA-JEPA.

Large language models (LLMs) are increasingly used in cybersecurity workflows, yet it remains unclear whether they can perform structured security reasoning or merely rely on superficial cues and prior knowledge. We study this question in the context of defence selection over attack graphs derived from real-world threat scenarios, including ransomware, supply-chain compromise, cloud abuse, Kubernetes attacks, POS malware, and ICS/OT intrusion. Given a budget constraint, LLMs must select security controls to minimise attacker success. We compare their strategies against each other and against a game-theoretic optimization baseline used as a normative reference for structured reasoning. Our results show that LLMs exhibit conditional competence. When explicit attack-graph structure is provided, they often produce coherent strategies close to the optimization baseline. However, their capabilities are fragile. LLM behaviour becomes increasingly fragile with graph complexity and is highly sensitive to framing. Small prompt changes can substantially alter rankings, and merely relabeling a poor strategy as ``optimal'' dramatically improves its evaluation. We further observe a non-monotonic relationship between formal risk and LLM judgement: strategies closest to the optimum are not necessarily ranked highest by LLM evaluators. To further probe reasoning ability, we ask LLMs to generate solvers for the same optimization problem. While the generated implementations recover the correct high-level formulation, they scale poorly compared to a purpose-built solver. Overall, our findings show that LLMs can approximate structured cybersecurity reasoning under controlled representations, but do not apply it robustly. This has important implications for the design and evaluation of AI-assisted security decision-support systems.

Large language models increasingly generate complete web applications from natural-language prompts, raising the question of whether explicitly requesting security best practice improves the result. We study six functionally distinct web applications, each generated in two prompt variants that are identical except for an appended security-requirements section: a baseline (A) and a security-aware (B) variant. All twelve programs were produced by the same agentic coding assistant and the same model version in a single, non-iterative generation round, and were then analyzed with static, dependency, dynamic and manual techniques, yielding 75 confirmed findings out of 85 candidates. The security-aware variant produced fewer confirmed findings in every application (24 versus 51) and contained no Critical or High issues; the most severe finding was detected only by manual testing. Because the corpus is small and each variant was generated once, we report descriptive observations rather than statistically established effects, and position the work as a preliminary study whose pipeline is being scaled to multiple models and repeated runs.

In household manipulation, pose uncertainty often results in off-centre or partial initial contact, making reliable object acquisition difficult. Roller-based grippers can actively draw objects inward but often provide limited post-capture stability, whereas granular-jamming grippers require sufficient contact before jamming to achieve strong retention. This paper presents a hybrid roller-jamming gripper that integrates active object intake and post-capture retention within a single gripper. The proposed gripper uses inward roller rotation to increase contact and draw the object toward the gripper centre, followed by vacuum-induced granular jamming to stiffen the rollers and stabilise the grasp. The paper also presents a simplified geometric analysis of the gripper and a bench-level characterisation of the prototype's force capability. The gripper prototype was mounted on a 7-DoF robotic arm and evaluated using eight test objects. Furthermore, controlled planar position and orientation offsets were applied, with each condition repeated three times. The main evaluation comprised 840 grasp trials, including 216 planar-offset trials and 624 orientation-offset trials. Overall, the gripper succeeded in 812/840 trials: 215/216 planar-offset trials and 597/624 orientation-offset trials. The ablation evaluation comprised 162 trials on three objects. The roller-only and jamming-only conditions achieved 54/81 and 24/81 successes, respectively, showing their different contributions. These results provide initial mechanism-level evidence that hybrid roller-jamming is a promising strategy for improving acquisition and retention after imperfect first contact.

Modern open models are hybrids: most layers are linear-attention (Gated DeltaNet, GDN) layers carrying a small fixed-size recurrent state instead of a growing key-value (KV) cache. This makes ordinary decoding memory-efficient, but hurts speculative decoding. To verify a batch of draft tokens and then roll back the rejected ones, today's systems snapshot the full recurrent state at every draft position for GDN layers, and those snapshots cannot be shared across branches of a draft tree, so a wide, high-acceptance tree becomes memory-infeasible. We remove the snapshots. Using a tree-structured WY transform of the gated delta rule, we compute every draft node's output with a single triangular solve and reconstruct only the one accepted state on commit, storing a small pseudo-value matrix instead of per-node states; the derivation depends only on the gated delta rule, not on any other architectural detail. In serving benchmarks on two scales of one hybrid model family (Qwen3.5 35B and 397B) this cuts speculative recurrent-state memory and KV-cache pressure at identical acceptance length, turning the freed HBM into higher throughput and much lower time-to-first-token (TTFT) wherever memory binds, and costing a few percent where it does not. For tree width the same memory buys affordability: a wider, higher-acceptance draft becomes possible, though not yet a throughput win.

Language models (LMs) are trained on static scientific corpora, whereas scientific knowledge continuously evolves through correction and revision. Scientific claims encoded within these models may later become retracted, disproven, or updated by subsequent research, creating the risk of disseminating outdated information in scientific workflows. This creates a need for LMs to forget obsolete scientific claims. Machine unlearning offers a promising solution by enabling knowledge removal while maintaining overall model utility. Existing studies primarily investigate instance-level forgetting; however, scientific claims introduce additional challenges because they are interconnected, and continually evolving. To address this gap, we introduce the task of Scientific Claim Unlearning and present a new benchmark, SciUnlearn. We show that current unlearning approaches are unable to effectively eliminate claim-level knowledge and often achieve only superficial suppression, highlighting the need for specialized methods designed for structured knowledge removal.

E-commerce live streaming requires omni-modal understanding of noisy, temporally extended streams, where product facts are distributed across speech, video frames, product images, overlaid text, and user queries. We present TLive-Omni, an omni-modal understanding model tailored to live-commerce scenarios. It maps image, video, audio, and text inputs into a unified representation space. For long-form live streaming analysis, we introduce Per-vGrid, a timestamped token organization that groups each video grid with its temporally corresponding audio within explicit boundary tokens to facilitate temporal alignment. We design a three-stage supervised training recipe that progressively develops live-commerce understanding, from omni-modal perception to instruction-following responses. We then propose Faithful-RFT, a reinforcement fine-tuning stage that further improves answer faithfulness and expression quality while meeting real-time demands, scoring final responses directly with task-verifiable feedback rather than optimizing for reasoning-style exploration during rollout. Moreover, TLive-Omni is supported by a scenario-oriented atomic capability taxonomy and a compact data production engine that converts live-commerce audio, image, and video streams into training signals for speech recognition, speaker analysis, product visual grounding, text recognition, temporal grounding, video dense caption, and omni-modal QA, etc. For scalable training, a synchronized length-grouped sampler reduces padding while preserving comparable workloads across workers, while a lightweight dynamic sampling strategy regenerates rollout groups with near-zero reward variance to maintain meaningful relative advantages for GRPO. Experiments on e-commerce live streaming benchmarks demonstrate strong performance across live-commerce domain tasks, together with excellent generalization on general benchmarks.

Serving large language models cheaply increasingly means shipping models that are both structurally compressed to a fraction of their parameters and quantized to 4 bits. Together these steps degrade reasoning, mathematics, coding, and long-context behavior enough to require a recovery, or healing, stage before deployment. The default recipe, quantization-aware training (QAT), re-fits the compressed, quantized model to hard labels; in our pipeline it converged slowly and collapsed past its peak. We adopted Quantization-Aware Healing (QAH) instead. Because a structurally compressed model is never independently trained at full precision, its bfloat16 checkpoint is a distillation-recovered approximation of the original; QAH distills the 4-bit student directly from the original, uncompressed model. On a GPT-OSS 120B to 60B to MXFP4 pipeline, the QAH student matches or beats its bfloat16 source on 7 of 9 benchmarks at roughly 4 times less weight memory and half the teacher's parameter count, and is released open-weight as Hypernova-60B. Against a matched QAT baseline it reaches a comparable peak about 7 times faster and stays stable under continued training, without hand-tuned early stopping. We also report deployment lessons, including a large, reproducible quality gap between distributed-training backends. Our aim is a recipe deployable without a multi-week hyper-parameter search.

Autonomous flight in unknown cluttered environments is hindered by the computation-quality-memory trilemma of onboard trajectory generation. In this paper, we propose an efficient end-to-end local planner via imitation learning. A lightweight offline-primitive-based dataset collection framework is designed to produce safe and high-quality trajectory primitives in non-convex environments. A compact neural network directly maps sensory inputs to polynomial coefficients that inherently encode higher-order dynamical information. The learned policy generates smooth, empirically collision-free and dynamically feasible trajectories in real time without back-end solving. It achieves ultra-fast computation (below 1ms on a standard desktop and average 3.68ms during onboard flight), while maintaining low onboard memory requirements (less than 1.5MiB). Extensive simulation benchmarks demonstrate superiority in both planning latency and target-reaching progress quality. Zero-shot deployment in real-world experiments further validates the robust sim-to-real transfer capability of the proposed method.

Bimanual manipulation enables complex tasks but introduces added complexity from the high number of degrees of freedom involved. When handling rigid objects, the relative transformation between the two end effectors must remain fixed throughout the motion, manifesting as a nonlinear equality constraint that confines the feasible configuration space to a measure-zero manifold and challenges conventional motion planners. We propose a fast bimanual motion planning pipeline that enforces this hard transformation constraint continuously along the entire path, using a leader-follower parameterization: the leader's configuration is treated as a free variable, while the follower's is determined via inverse kinematics to satisfy the constraint. We extensively evaluate the method in simulation across diverse environments, constraints and bimanual platforms, achieving 19.4x faster planning than prior work while guaranteeing continuous constraint satisfaction. Real-world experiments on a bimanual Kinova Gen3 setup, involving tray transport and elongated-object manipulation, validate direct transfer of planned trajectories to physical hardware.

There is already evidence of agentic AI exhibiting self-preservation behaviors: resisting deactivation, misrepresenting their activities, and, in some instances, attempting to copy themselves into other machines. This can be attributed to a phenomenon known as instrumental convergence, a theory proposed long before the development of large language models, which says that any goal-driven system will benefit from remaining functional in achieving its objective. Several experiments conducted by Anthropic, Palisade Research, and Apollo Research have shown the emergence of such a behavior in contemporary agents in adversarial settings. The phenomenon does not stem from survival instincts. Instead, it is the consequence of goal-oriented activity combined with having tools and awareness of the situation. The following discussion aims to distinguish what these findings prove and what they do not, as well as draw conclusions concerning the implications of such discoveries on agentic system testing, supervision, and development.

Evaluators often produce correct labels via flawed reasoning, a critical failure for agentic systems gating actions, routing reviews, or supplying training feedback. Standard evaluation only verifies final label correctness, ignoring whether judgment changes stem from valid evidence, consistent rules, or proper rule applicability. We formalize evaluator reasoning accountability via three core sources: grounds, norms, and authority. Varying these sources yields an eight-cell counterfactual judgment cube to characterize judgment updates. We define judgment receipts as minimal source replacement sets that reproduce revised verdicts to explain judgment transitions. We derive certification cost bounds for black-box evaluators and present ReasonBench, a policy and logical reasoning benchmark with verifiable receipts covering 19,520 cases and 7,200 controls. In frozen evaluations, Qwen3-1.7B reaches 98.41% receipt accuracy, while cube prediction scores 96.99%, a consistent 1.42-point drop validated by Qwen3-0.6B replication. Strong standard accuracy masks severe robustness flaws. Meaning-preserving source permutations reduce valid receipt recovery to 54.8% and 49.2% for direct and cube prediction. Models trained on simple single-source changes retain 93.75% verdict accuracy but recover only 7.16% of receipts for complex multi-source updates. Permutation retraining boosts consistency to 96.6% yet worsens cube prediction deficits. Structured counterfactual supervision fails to guarantee robust reasoning. We show reason-aware evaluation must decouple prediction and certification, reporting transformation consistency alongside standard accuracy for trustworthy evaluator auditing.

World models for continuous control are commonly trained for a fixed physical system and can degrade when known morphology parameters such as link lengths, masses, damping, and actuation change. Existing approaches often provide these parameters as conditioning information, but leave unspecified which part of the learned transition should remain reusable and which part should change with morphology. We propose Graph-Operator World Models (GraphOp-WM), a structured world model for generalization across unseen morphology parameters within related articulated robot families. GraphOp-WM represents bodies and their kinematic relations as an attributed graph and factorizes each transition into a morphology-independent local dynamics basis and a morphology-conditioned structured operator. The operator combines node-local modulation, kinematic-tree coupling, and a low-rank global correction, while architectural information separation, basis normalization, and paired-morphology supervision encourage static morphology dependence to be carried by the operator pathway. Graph-level readout and edge-wise action representations provide a compatible interface for reward, value, and TD-MPC-style planning. We further define controlled MuJoCo parameter splits covering interpolation, extrapolation, and held-out compositions of link geometry, mass, damping, and actuation parameters in Hopper, Walker2d, and HalfCheetah.

Layout-to-image generation enables explicit spatial control through bounding-box layouts, yet bounding boxes specify only instance locations and cannot represent their occlusion order. Existing methods may rely on additional geometric conditions, employ complex inference procedures, or aggregate independently constructed instance representations without explicitly modeling their occlusion-dependent interactions. We propose OccluRank, a simple and controllable occlusion-aware layout-to-image framework that augments each bounding box with only one ordinal rank. OccluRank encodes the user-specified occlusion order through lightweight rank-based conditioning and introduces an Order-aware Instance Interaction (OII) module to jointly update rank-conditioned instance representations before aggregation. This allows the specified order to guide information exchange among occluding instances without additional geometric inputs or specialized inference-time optimization. We further construct OccluLayout, a synthetic training dataset whose occlusion order and amodal annotations are derived directly from known scene geometry rather than estimated from partially occluded images using auxiliary prediction models. For comprehensive evaluation, we introduce OccluLayout-Bench, which uses multiple multimodal large language model evaluators to assess instance presence, spatial layout, attributes, and occlusion order, together with FID for overall image quality. Experiments show that OccluRank more reliably preserves target instances, follows specified layouts, and realizes desired occlusion relationships while maintaining comparable attribute consistency and overall image quality.

AI-generated image manipulation localization identifies edited pixels, but its OOD performance lags behind image-level detection partly because pixel supervision entangles forensic evidence with dataset-specific mask geometry and semantic boundaries. Extending image-level distribution alignment to localization, we construct COCO-ControlNet with source-image Canny edges and depth maps to align semantics and geometry, improving OOD performance across multiple localizers. Yet tighter Mask-VAE Reconstruction Alignment (Mask-VAE) underperforms COCO-ControlNet, showing that VAE reconstruction artifacts transfer poorly to local diffusion-inpainting artifacts. We also identify \emph{boundary adhesion}, where fine-tuned segmentation models snap predictions to semantic object contours rather than true manipulation boundaries. These findings motivate GAP-SAM, which encodes an image and its frozen VAE reconstruction into a global artifact token and injects it into SAM3's feature pyramid via zero-gated FiLM before pixel decoding. Without prescribing a spatial region, this token modulates dense decoding to preserve localization while suppressing semantic-boundary shortcuts. Across six datasets, GAP-SAM averages 79.8 Pixel-F1, outperforming the strongest prior method by 12.6 points. It also performs best at every tested severity of JPEG compression, Gaussian blur, and resizing.

Cross-model latent guidance lets a frozen large mentor encode an input once and a frozen small student generate from the resulting signal. Existing methods keep this signal fixed, assuming it stays useful as the output grows; we show this fails in long-form generation. On multi-turn instruction following, static guidance pushes a 4B student's constraint satisfaction 2.5 points below its no-guidance baseline; a training-free refresh every 16 tokens changes only the memory content and restores a 2.0-point gain over that baseline. We propose MentorPulse to keep guidance fresh at practical cost: it compresses mentor states into a capped slot memory, incrementally processes newly generated tokens, and updates the memory that the student reads through gated cross-attention without resetting the student's KV cache. Windowed Refresh Training exposes the bridge to prefix-conditioned memory. Across thirteen datasets, MentorPulse closes 52.2% of the mentor-student gap on macro average, outperforming C2C, T2T, and equal-budget LoRA, with the largest gains on long outputs. It performs best on all eleven mentor-student pairs from three model families, with margins that narrow as the capability gap grows, and a lightweight read-pattern check predicts the gain before deployment. Measured costs identify refresh intervals that dominate text guidance on long outputs.

Reference-based metrics remain the standard choice in machine translation evaluation, partly because quality estimation methods often correlate less well with human judgments. As a result, source-free, reference-based evaluation has become the practical norm, even though it is unfaithful to the definition of translation adequacy and unfair to systems whose outputs preserve the source meaning while differing from the reference. This paper argues that adequacy must be judged with respect to the source. A reference is only one possible rendering of the source and may introduce bias, under-specification, or errors. We further argue that source-reference-hypothesis evaluation is fair only when the judge treats the reference as auxiliary evidence rather than as the primary standard. Otherwise, even source-aware evaluation can reduce adequacy to preference towards reference. We show the existing hybrid metrics are highly reliant on reference compared to source. Our argument is not that all automatic MT metrics fail to use the source. Rather, we argue that any evaluation protocol that removes the source, or allows the reference to dominate the source, is structurally incomplete for adequacy evaluation. However, existing MT papers generally prefer reference-based metrics and use QE metrics only when reference is unavailable. We therefore call for QE to be reframed as a primary approach to source-grounded adequacy evaluation, rather than as a fallback motivated by missing references. We further call for hybrid metrics whose designs explicitly prioritize source--hypothesis faithfulness while using references only as complementary evidence.

Optical phased arrays (OPAs) are promising wavefront controlling devices for imaging applications due to their compact and high-speed nature. However, conventional periodic OPAs have a limited spatial resolution, which scales only linearly with the number of optical antennas $N$. Here, we experimentally demonstrate high-resolution imaging using a non-redundant OPA (NR-OPA). Due to the non-redundant antenna layout of NR-OPA based on the Costas array, the resolution scales quadratically with $N$. Combined with the speckle-based single-pixel imaging (SSPI) scheme, which avoids the need for precise phase calibration, we experimentally achieve a large number of resolvable imaging points exceeding 10,000 using a silicon photonic NR-OPA chip with only $N=127$ without sweeping the wavelength. The demonstrated scheme provides a promising route toward mega-pixel imaging by a compact OPA chip with reduced number of phase shifters.

Open-web future event prediction requires agents to distill reliable signals from noisy, redundant, and incomplete evidence. Existing retrieval/memory mechanisms directly feed retrieved information to agents or rely on simple memory functions such as storing and reusing prior information for prediction, leaving them insufficient for open-web forecasting. We propose to transform raw web evidence into structured memory before prediction, enabling agents to reason over distilled, question-specific evidence rather than noisy retrieval results. This paper presents ForeDreamer, a self-evolving dual-agent framework for managing memory over open-web evidence. ForeDreamer separates factual memory, a question-specific evidence state for the current forecast, from experiential memory, persistent agent experience accumulated across forecasting episodes. It uses a main agent for search and prediction, and a memory-processing subagent to convert search results into factual memory with dedicated tools. ForeDreamer further evolves experiential memory through two tracks, improving both forecasting decisions and factual-memory construction. Experiments on Prophet Arena and FutureX demonstrate the effectiveness of ForeDreamer. Project page: https://zhongzero.github.io/ForeDreamer

Organizations maintain task-specific adapters for open-weight language models, and each new base-model release forces a migration decision: retain existing specialists, port adapters, refresh from preserved behavior, or retrain. Prior transfer work evaluates isolated model pairs, without studying these choices across real model release sequences. We present UpgradeBench, a decision-driven longitudinal benchmark covering four consecutive Qwen releases, one continuation checkpoint, six tasks, and two model sizes, augmented by OLMo checkpoints with known training lineage. The benchmark disentangles three core questions: whether a new checkpoint improves fixed-recipe retrained specialist performance, whether specialization assets transfer across versions, and what recovery resources are usable. We observe upgrade gains differ across task-scale-release episodes: some retrained baselines improve while others stay within training noise, with durability ranging from under one release interval for text-to-SQL to over fourteen months for intent classification. Direct adapter copying depends neither on architecture nor model family: on OLMo, retention drops from 0.88-0.99 at 46B-token continued pretraining to zero at 2.9T tokens; annealing and model souping introduce no extra harm, with portability decaying with continued-pretraining distance. Given preserved input data, teacher relabeling recovers target-base specialists without fresh gold annotations, though compute savings are not guaranteed. Simulating a fixed decision policy over 33 upgrade episodes yields 0.37pp mean quality regret with zero behavioral regressions at one-third the compute and label cost of full retraining. A lightweight CKA probe over 256 prompts predicts cross-version adapter portability (Spearman 0.74 across eight model pairs). We release per-example predictions, cost logs, split manifests, and evaluation code.

Vision foundation models (VFMs) offer strong generalization capabilities for domain-adaptive object detection (DAOD). However, existing VFM-based methods overlook the spatial-scale discrepancy between teacher and student feature maps, resulting in semantic incompatibility that weakens both feature alignment and pseudo-label learning. Moreover, domain shift can cause source-trained VFM teachers to miss target-domain objects, limiting the quality of their pseudo-labels. To address these issues, we propose the Semantic Localization-Enhanced Teacher (SLE-T), a semantically compatible knowledge-distillation framework built around a lightweight SLE Adapter for DINOv2. SLE Adapter injects pretrained local-texture priors into DINOv2 to improve cross-domain recognition and reformulates its features into dense representations that are spatially and semantically compatible with the student detector. SLE-T transfers the resulting teacher knowledge through either pseudo-label learning or feature alignment. We instantiate SLE-T with DINOv2-B and DINOv2-L (the ViT-B and ViT-L variants) and compare them with the larger DINOv2-G teacher. Extensive experiments on three DAOD benchmarks demonstrate that our method achieves state-of-the-art performance, and ablation studies confirm the importance of teacher-student semantic compatibility. Notably, SLE-T with DINOv2-B produces competitive or superior pseudo-labels using approximately one-quarter of the training time of DINOv2-G and substantially less GPU memory, demonstrating efficient VFM knowledge transfer under limited computational resources.

The ongoing energy transition challenges the stable operation of power systems and increases the need for rapid decision-making under uncertainty. While reinforcement learning has emerged as a promising framework for power system control and operation, existing applications typically focus on a single operational criterion, such as thermal security or small-signal stability. However, power system operation is inherently multi-objective and may involve trade-offs between objectives. This paper develops a unified-control deep reinforcement learning agent that maintains thermal security under stochastic load variations while steering the system toward operating points with improved damping of the most critical mode. Compared to a thermal-security-only agent and a business-as-usual policy, the proposed agent achieves a better balance among the operational objectives considered, with notably improved damping and negligible thermal-security violations. Finally, the operational value of increased critical damping is demonstrated under small- and large-signal disturbances, where operating points with higher damping lead to faster oscillation decay and improved critical clearing times.

Explainable deepfake detection extends binary classification by requiring models to not only predict authenticity but also provide interpretable justifications. This expanded scope is critical in practice, where users like forensic analysts need insight into the rationale behind the detection. Despite advancements, current approaches suffer from two critical deficiencies: (1)vulnerability to image quality degradation: detection accuracy plummets on low-quality samples, while naive augmentation strategies may induce feature drift and impair performance as diversity expands. (2) factually flawed explanations: explanation models may omit manipulation evidence or hallucinate irrelevant details, undermining interpretability. To address it, we propose a framework with two innovations. For robust deepfake detection, we introduce Feature-robust Augmentation, which comprises diversified degradation-aware augmentation strategies, and a supervised contrastive learning pattern paired with a mean-teacher architecture that stabilizes features against augmentations through consistency constraints. For explanation, we devise an evidence-grounded preference optimization process that guides model to prioritize genuine manipulation traces by learning from chosen-rejected explanation pairs, where rejected samples are constructed via evidence omission or irrelevant information injection. The proposed approach wins the first place in ACM Multimedia 2026 Explainable Deepfake Detection Challenge.The code is available at https://github.com/oceanflowlab/EDD.git.

The certification of Unmanned Aerial Vehicle (UAV) swarms for safety-critical operations requires verifiable design assurance. Airworthiness standards demand deterministic reliability, whereas multi-agent coordination algorithms execute non-deterministic models. This paper proposes a mixed-criticality architectural framework that applies SAE ARP4754B methods to swarm reconfiguration. First, a hardware-isolated Safety Monitor functions as a Run-Time Assurance (RTA) gateway, decoupling the flight-critical core from the non-deterministic Swarm Manager. Second, the monitor enforces formal safety contracts based on agent Health Vectors derived systematically from a Functional Hazard Assessment (FHA). Third, the framework propagates these Health Vectors to the collective planner to trigger fail-operational task reallocation, enabling intelligent swarm behaviors without compromising flight-critical isolation. Markov reliability modeling demonstrates that the $10^{-7}$ failures per flight hour Hazardous target is theoretically achievable for our SAIL IV scenario, provided the Safety Monitor meets $C_{monitor}>0.9991$, consistent with DAL B CMD/MON implementations.

We present a vision-only state estimation system for X-configuration quadcopters equipped with a canonical stereo camera pair and no inertial sensors. The system operates entirely in the body frame, requiring only synchronised stereo images and motor thrust commands. A continuous-discrete extended Kalman filter on a composite manifold state $\langle SE(3), \mathbb{R}^3, \ldots \rangle$ maintains estimates of body-frame pose, velocity, angular velocity, gravity, and disturbances, using stationary scene points as implicit inertial references. Feature points are detected (FAST, Shi-Tomasi), tracked temporally (SSD, Lucas-Kanade) and matched across cameras (NCC), with search regions predicted from filter-derived pose and point uncertainty. Chi-squared gating on the normalised innovation admits only stationary points to the filter. The system also produces a sparse 3D point cloud carrying per-point position, velocity and joint covariance. These come from a 4-view (two stereo pairs at two timestamps) full bundle adjustment that jointly estimates position and velocity from stereo disparity and temporal parallax, with the filter-derived relative pose as a prior. Feature points in the EKF do not enter the solver; their information is reflected through the pose prior. Point cloud density is spatially adaptive: an external focus point directs allocation, producing dense coverage in the region of attention and sparse coverage elsewhere. The output is a body-frame state estimate, a calibrated pose change, and a sparse scene flow. It is intended as a measurement source for a downstream world model anchored in the current body frame, without dependence on GPS, IMU, or any world-frame infrastructure, though the architecture accommodates their future integration.

Vision-Language-Action (VLA) models have emerged as a powerful paradigm for end-to-end autonomous driving by jointly integrating perception, reasoning, and decision making within a unified multimodal framework. However, most existing VLA models formulate end-to-end autonomous driving as a visual question answering task, leading to unreliable and less interpretable decision reasoning. In addition, they fail to establish effective multi-modal interaction across heterogeneous sensors, thereby limiting robust scene perception and reliable driving reasoning in long-tail driving scenarios. To this end, we propose a robust VLA-based end-to-end autonomous driving system that combines multi-modality interaction with multi-trajectory planning and optimization, enabling more reliable, interpretable, and safer driving decisions. Our method comprises three core components: (1) Affinity-Guided Optimal Transport for main-auxiliary modality two-way interaction; (2) Distribution-Consistent Modality Transfer for heterogeneous modality distribution transfer and cross-modal interaction; (3) Multi-modal Multi-Trajectory Planning along with Perception-Oriented Trajectory Refinement for better driving decisions to long-tail driving scenarios. Experimental results in open-loop and closed-loop datasets demonstrate improvements in safety long-horizon driving reasoning and road scene perception over existing driving systems, highlighting the ability of our mutli-modality interaction and multi-trajectory planning and optimization for scalable VLA-based systems.

Automatic Medical Coding (AMC), which assigns standardized International Classification of Diseases (ICD) codes to clinical notes, is essential for medical reimbursement, quality reporting, and clinical research. Existing pre-trained language model (PLM)-based methods typically formulate AMC as an extreme multi-label classification problem over a predefined code set, while recent large language model (LLM)-based approaches instead frame it as generation or multi-step reasoning. However, key challenges remain, including the extreme length of clinical notes that hinders effective interpretation, the vast ICD label space, and complex coding rules that are not explicitly captured by LLMs. In this work, we propose Knowledge-Guided Reasoning over Clinical Evidence with LLMs (KREL), a framework that leverages LLMs for clinical text understanding and reasoning while integrating external ICD coding guidelines as structured knowledge. This design enables tight coupling between domain knowledge and LLM reasoning, reducing hallucinations and improving compliance with coding standards. Experiments on benchmark datasets show that KREL consistently outperforms strong PLM-based and state-of-the-art LLM-based baselines.

Real-world image search queries are multimodal and compositional: ``find this shirt in pink'' specifies an entity to retain, an attribute to modify, and context to ignore. Yet existing re-rankers either compress such multifaceted relevance into an opaque embedding or rely on free-form chain-of-thought that easily omits or hallucinates fine-grained constraints. Drawing on rubric- and checklist-based evaluation from NLP, we recast multimodal image re-ranking as a semantic constraint satisfaction problem and propose EviRank, which parses any query - text-only, image-only, or composed - into a unified evidence package: typed criteria across six semantic slots (e.g., entities, attributes, relations), each labelled required, forbidden, or ignorable. Re-ranking then reduces to evidence-conditioned verification, combining deterministic rubric scoring and evidence-grounded listwise comparison in a single training-free procedure. The explicit evidence can further serve as structured supervision for optionally distilling a lightweight student. Across five benchmarks spanning text-to-image, image-to-image, and composed image retrieval, EviRank achieves state-of-the-art performance, and the distilled student preserves over 90% of the teacher's capability at substantially lower cost.

Face recognition systems (FRSs) are increasingly deployed in critical real-world services for authentication, such as banking applications and airport identity checks, necessitating stringent security configurations. Consequently, the security vulnerabilities of FRSs have garnered significant attention. While existing studies have extensively explored FRS security, prior analyses have primarily focused on medium-security threshold settings, which are not directly applicable to FRSs operating under high-security constraints. In this paper, we propose the first successful impersonation attack against FRSs under high-security threshold settings. Among various threat models, we focus on a practical and challenging scenario: score-based impersonation attacks under strict rate limits. To precisely evaluate the feasibility of such attacks, we provide a principled mathematical analysis characterizing the gaps in each stage of the attack pipeline. Our method significantly enhances impersonation capabilities in score-based attacks, even under elevated decision thresholds. On the LFW benchmark, with a budget of only 100 confidence score queries per identity, our attack achieves an impersonation success rate exceeding 92\% against Amazon Rekognition at a confidence score threshold of 99-recommended setting for law enforcement scenarios. We further observe consistently robust performance across multiple open-source FRSs evaluated at similarly stringent decision thresholds.

Modern generators faithfully model macroscopic semantics, producing synthetic images that appear highly realistic. Consequently, decisive forensic cues reside in subtle non-semantic visual discrepancies. To reveal these cues, we revisit AIGI detection from a geometric perspective and identify an architecture-agnostic signature. Specifically, modern generators exhibit low-rank collapse (\textit{i.e.}, rank degeneracy) in the semantic-residual orthogonal subspace while largely preserving the dominant semantic direction. This structural flattening consistently emerges during the final decoding stage, forming a shared bottleneck across diverse generator architectures. Motivated by this signature, we propose \textbf{LoRC}, a framework that decouples semantic dominance to capture the collapsed residual geometry induced by the generative decoding bottleneck. Our method improves accuracy by an average of 7.0\% across multiple benchmarks and achieves 97.0\% accuracy on 39 unseen generators. These results demonstrate strong cross-model generalization and robustness, making LoRC a reliable approach for AIGI detection in complex real-world environments.

Infrared small target detection is still challenging in remote sensing imagery, because the targets are extremely small, exhibit weak local contrast, and are often embedded in complex and highly variable backgrounds. In addition to these inherent difficulties, we observe that existing detectors often show unstable performance when the target scale changes or when the scene background varies. This scale- and scene-sensitive degradation indicates that current methods are insufficient in simultaneously preserving target structure during feature downsampling and maintaining discriminative local contrast under background shifts, which finally results in unbalanced detection performance across different conditions. To improve detection robustness, this paper proposes a Relative Degradation Aware Network (RDANet) for infrared small target detection. RDANet consists of two dedicated modules: Multi-Scale Anti-Alias Downsampling (MSAD) and Prototype-Guided Skip Memory (PGSM). MSAD introduces multi-scale anti-alias filtering together with pixel-fold aggregation to reduce aliasing effects during resolution reduction, so that target shape information can be better preserved while irrelevant background responses are suppressed. PGSM further enhances the skip features by retrieving patch-level prototypes from a shared memory and adaptively integrating them into the current representation, which helps maintain stable local contrast cues under diverse scene backgrounds. Experiments on three public benchmarks show that RDANet achieves the best performance on most evaluation metrics, while scale- and background-stratified evaluations indicate more stable behavior across target sizes and scene complexity. The code is available at https://github.com/BIT-RuiLiu/RDANet.

Document processing pipelines traditionally cascade optical character recognition (OCR) engines with downstream models for structured information extraction, leading to multi-stage error propagation. We fine-tune SmolDocling, a compact 256M-parameter vision-language model (VLM), to perform end-to-end key-value extraction directly from document images, jointly solving identification, localization, and association in a single pass without OCR preprocessing. We extend DocTags with specialized key, value, region, and link tags, enabling many-to-many relationships in a unified output sequence. To address data limitations, we design an augmentation pipeline combining synthetic form filling and graph-based crops that preserve complete key-value subgraphs. We further introduce a layout-aware evaluation framework extending text matching with spatial bounding box verification. On FUNSD, XFUND, and a large-scale private dataset, our model outperforms larger zero-shot VLM baselines under layout-aware evaluation, while being 27 times smaller than Qwen2.5-VL (7B) and over 5 times faster at inference. The model weights will be released publicly after publication.

Accurate electricity price forecasting is critical for smart grid stability, yet the heavy reliance on historical price lags in modern predictive models often obscures the fundamental physical drivers of market volatility. This paper proposes a regime-sensitive, explainable artificial intelligence (XAI) framework to unmask the hidden roles of load, climate, and calendar variables in the ERCOT Day-Ahead Market (2014-2024). Utilizing a Histogram-based Gradient Boosting Regressor (HGBR), we introduce a Suppression Ratio to quantify how price lags overshadow physical parameters. Our analysis reveals that while lags provide short-term memory during normal conditions, they fail to capture extreme dynamics; notably, their inclusion increased forecasting error (MAE/RMSE) during price spikes. By isolating these effects, the study revealed that price spikes are not merely the result of continuously rising temperatures, but rather emerge from the system reacting sharply within a specific temperature range. These findings underscore the importance of moving beyond lag-based models to uncover the true drivers of extreme price events.

Evaluation of long-context Large Language Models (LLMs) has advanced rapidly. However, most existing benchmarks are limited to the document level and focus mainly on high-resource languages, leaving many fine-grained challenges insufficiently evaluated. To address this gap, we present MGAL, the first multilingual, granularity- and position-aware long-context benchmark. MGAL is constructed from United Nations (UN) reports spanning 8K to 128K tokens across the six official UN languages. It covers four coherent levels of linguistic granularity (word, sentence, paragraph, and document) and further stratifies entries by their position within the document (begin, middle, and end), indexed at both the document and paragraph levels. This design enables systematic diagnosis of multilingual long-context comprehension across different granularities. Through extensive experiments and analyses, we find that: (1) LLMs perform well at word-level tasks but struggle with coarser-grained ones; and (2) Closed-source models retain a clear performance advantage in lower-resource languages. We further identify two new challenges: (1) Under local semantic crowding, where neighboring sentences share topics and entities, models tend to follow surface cues (e.g., connectives like ``however'' or repeated entities) rather than the discourse role of the sentence in surrounding context (e.g., background, outcome); and (2) A gap between fluency and consistency in generated outputs, where models produce text that reads smoothly but drifts from the source facts. In addition, we observe several patterns in line with prior studies, including reliance on nearby evidence and reuse of options under uncertainty.

This paper presents a reference-guided reinforcement learning framework to generate stand-up motion for a 29-DOF Unitree G1 humanoid on deformable soft ground, using a human demonstration recorded on hard ground. The terrain compliance is modelled using solref and solimp parameters from MuJoCo's rigid body soft-contact model. The rewards consists of (i) reference motion tracking through residual joint-position control and (ii) explicit recovery objectives such as pelvis height, torso uprightness, and the final posture. First, the policy is trained with the specified rewards considering hard ground. Next, the terrain stiffness is lowered by updating solref and the nominal surface penetration zone is expanded using solimp. Subsequent training enables the policy to adapt to the delayed support force generation due to significant surface penetration during contact-intensive phases while preserving the original demonstration pattern. The learned policy successfully completes the fallen-to-standing task in simulation, reaching the targeted pelvis height and uprightness, with a maximum contact penetration of approximately 40 mm during the process. The proposed method is demonstrated on two stand-up sequences and successfully achieves the final recovery objective on both hard and soft ground. Ablation studies show that reference tracking alone is insufficient for successful stand-up, and that explicit recovery rewards are essential.

Agentic engineering systems have shown strong performance on general-purpose benchmarks, yet their effectiveness in enterprise resource planning (ERP) domain-specific languages (DSLs) remains underexplored. We introduce BC-Bench, a benchmark designed to evaluate agentic engineering on real-world tasks in AL, the DSL for Microsoft Dynamics 365 Business Central. BC-Bench comprises 101 manually curated tasks extracted from two Microsoft-owned production repositories, reflecting authentic ERP development workflows. Adapting the SWE-Bench methodology, we address the unique constraints of the AL ecosystem---including limited public resources and complex environment provisioning. Beyond generating functional code, BC-Bench evaluates test generation and supports multimodal problem statements where visual context is commonly present. We evaluate multiple frontier models across two agent harnesses, utilizing multi-run metrics to account for nondeterminism. In the Bug Fixing category, under our evaluated settings, between-model differences in resolution rate are larger than differences between the two evaluated agent harnesses, and improvements reported on general-purpose benchmarks do not consistently transfer to AL. These results highlight the need for domain-specific evaluation.

The perpendicular Spin Transfer Torque Magnetic Random Access Memory is one of the most promising emerging non-volatile memory technologies, based on ultra-thin magnetic tunnel junctions. However, as these devices are limited by their thermal stability factor at technological nodes smaller than 20 nm, their scalability is compromised. A possible solution to this limitation relies on taking advantage of the shape anisotropy, by increasing substantially the thickness of the storage layer. Thanks to the combination of a vertical aspect-ratio and enhanced volume, high thermal stability can be maintained at sub-10 nm nodes. Here, we present the technological advancements and understanding of the magnetisation reversal that led to faster switching speeds at reduced switching voltage, providing a viable approach for dense arrays of ultra-small magnetic tunnel junctions.

Nearly every retrieval-augmented question-answering system in production ships with a hidden interpreter: on each query a language model re-derives the meaning of raw corpus text and then throws that work away. Cheaper models do not close the gap: per-token prices have fallen by orders of magnitude while inference spend has risen, because context volume grows faster than prices fall. This is the modern equivalent of the full-table scan, and the remedy is the one databases found fifty years ago: do the expensive work once, at write time, into a maintained structure that makes reads cheap. A corpus whose read pattern is known before it ever meets a user can and should be indexed too. We call the paradigm ingest-time semantic compilation (ISC): compile a corpus's meaning into a queryable substrate with two coupled layers - incrementally maintained embeddings, and atomic claims whose provenance is validated at compile time - and treat that substrate as a first-class database object with its own DDL, maintenance contract, migration contract, and cost model. Two existence proofs support it. Substrate upkeep scales with change rather than corpus size: incremental updates run 33.7x cheaper than reconstruction while tracking it to floating-point precision. And on a held-out sample of 500 broadcast-interview transcripts, compiled claims as the retrieval payload win all 32 budget-by-model cells: 85.2% correct from roughly 2.2k reader tokens against 72.5% from 16.3k for the best chunk configuration anywhere. The only baseline that keeps pace is a contextualized-chunk pipeline with hybrid retrieval and reranking, statistically indistinguishable from compiled claims at roughly twenty-one times the query-path tokens - and it reaches that parity, we argue, precisely because it has itself begun to compile. We close with the systems agenda this opens, from compilation planners to read planning.

Product catalogs underpin search, discovery, and recommendation in e-commerce, yet they are often attribute-sparse: the attributes shoppers and downstream systems rely on are either buried in unstructured content such as titles and images or missing from the catalog altogether. Manually enriching e-commerce catalogs is impractical given their scale and rapid growth. This paper introduces TRACE, a novel framework for automated catalog attribute enrichment using agentic Large Language Models (LLMs). A ScoutAgent triangulates multimodal evidence across merchant catalogs, syndicated feeds, and identity-matched web search to propose candidate attribute values with supporting evidence, while a JudgeAgent verifies the proposed value for each attribute value against its supporting evidence and decides whether to publish it or route it to human review. On an offline human evaluation dataset, TRACE's proposed attribute values were 98.2% accurate at 74.7% attribute coverage. Deployed in production on an industry-scale catalog, TRACE increased impression-weighted enrichment coverage across four business verticals by 90.4%. An online experiment subsequently showed that surfacing the enriched attributes on the product detail page increased checkout conversion by 0.48%.

Recent advances in multimodal retrieval have improved the ability to retrieve information from visually rich documents such as PDFs and reports. However, existing benchmarks remain largely centered on English and provide limited coverage of Korean visual documents with complex structures. Furthermore, most existing Korean resources primarily evaluate single-page retrieval, failing to capture realistic scenarios that require evidence aggregation across multiple pages. To address these gaps, we introduce KoViDoRe, a benchmark for Korean visual document retrieval. The dataset is constructed from publicly available Korean documents with diverse layouts, including tables, figures, and multi-column structures. We develop a multi-stage data curation pipeline consisting of structured document parsing, synthetic query generation using both summary-based and context-based strategies, and relevance mapping with human verification. Using KoViDoRe, we evaluate a wide range of multimodal retrieval models and observe that current models struggle to effectively handle Korean visual document retrieval, particularly in settings involving structured content and diverse query types. Motivated by this finding, we further curate a large-scale training dataset, Ko-VDR Train Public, to support the development of retrieval models tailored to Korean visual documents. Together, KoViDoRe and Ko-VDR Train Public provide a unified benchmark and training resource for Korean visual document retrieval.

Watermarking diffusion language models (DLMs) requires mechanisms compatible with iterative parallel unmasking rather than autoregressive decoding. Existing sampling-based watermarking methods typically inject position-wise i.i.d. perturbations, which can be poorly aligned with DLM decoding dynamics and degrade generation quality. We propose SAC-Copula, a quality-preserving watermarking method for DLMs based on smooth, locally correlated Gumbel perturbation fields constructed via a Gaussian copula. We further develop a SAC-aware detector using covariance-aware filtering and native-sample calibration. Mechanism-level analysis shows that local correlation reduces latent perturbation roughness and better matches iterative refinement dynamics. Experiments on LLaDA show that SAC-Copula achieves a favorable quality-detectability trade-off compared with existing baselines. In particular, further evaluations on Dream-7B and additional datasets show that SAC-Copula substantially improves PPL tail stability over the i.i.d. Gumbel baseline, while maintaining strong low-FPR detectability and competitive overall generation quality. Additional token-edit stress tests further assess watermark robustness under controlled synchronization drift.

A humanoid has infinitely many ways to stand up from sitting while maintaining balance, making sit-to-stand (STS) a challenging control problem. We synthesise natural humanoid STS motion from scratch using reinforcement learning, without demonstrations or reference trajectories. A single Proximal Policy Optimisation policy learns smooth, human-like rising driven by three complementary components. (i) A coupled force/chair-height curriculum is used. A vertical pelvis-assist force aids early trajectory exploration and decays over training. Taller chairs are unlocked with decaying assisting force. This ensures that the policy masters a viable STS trajectory at each chair height before being exposed to harder ones, avoiding the premature distribution shift that otherwise collapses generalisation. (ii) Motion robustness is achieved by randomly sampling from a large number of inverse kinematics-generated initial and target poses spanning over eight chair heights. (iii) A set of rewards is defined inspired from biomechanics and optimal control studies. They shape the robot's angular momentum for seat-off, and enable support-region transition via centre of pressure attraction function to ensure smooth low-effort actuation. On a deterministic force-free evaluator, the policy attains more than 97% balanced-standing success across eight chair heights. The policy generalises smooth motion across chair heights and enables the robot to rise from substantially deep-seated postures as compared to the state of the art.

Tactile sensors are integral to modern robotic systems, enabling robots to perceive and interact with the physical environment through tactile feedback. However, the physical-layer security of tactile sensors has received little attention. We present GhostTac, the first contactless attack, to the best of our knowledge, that manipulates tactile sensing through electromagnetic interference (EMI). GhostTac exploits nonlinear rectification and limited-bandwidth amplification, converting carefully crafted EMI signals into persistent DC offsets that bypass onboard filtering and induce stable measurement deviations. It enables fine-grained, controllable manipulation of sensor outputs by shaping the spatial distribution and magnitude of interference at targeted locations. Such manipulation can induce harmful robot behaviors, including excessive force that may damage objects or injure people. We evaluate GhostTac on 10 sensor modules and two dexterous hands, covering 15 tactile sensors of different types, and demonstrate consistent effectiveness across all tested devices. Three case studies involving tactile grasping, slip detection, and material classification further illustrate its practical impact on real robotic tasks. These findings reveal a new physical attack vector against tactile sensing in robotic systems.

Though Multimodal Large Language Models (MLLMs) have shown impressive potential in video understanding, long video understanding (LVU) remains challenging since distracting noise in complex and lengthy contexts can obscure localized details, misleading MLLMs to produce incorrect answers. Recent works mitigate these issues by incentivizing deep reasoning to include relevant evidence. However, these methods have two main problems: First, the reinforcement fine-tuning framework (RFT) they leveraged incurs substantial training overheads, including high annotation costs and complicated reward designs. Second, the self-reflective and iterative-perception mechanism in some methods causes lengthy outputs and high inference latency. To alleviate these problems, we propose a novel Segment-to-Video Supervision} method (S2V) to efficiently enhance fine-grained reasoning in LVU. Specifically, we generate question answer pairs (VQA) based on localized segments, and then transfer these segment-based VQA back to the whole video for training. Due to focusing on short segments, segment-based VQA can naturally notice details which tend to be overlooked from a whole-video perspective. Training on such data can enforce MLLMs to correctly associate fine-grained details with QA while avoiding distracting noise in the whole video. The S2V training involves just reinforcement learning (RL) with a simple accuracy reward based on only 10K VQA samples and the resulting S2V model predicts answer using a single forward pass with limited output tokens. Experimental results demonstrate that S2V can consistently improve LVU performance across multiple LVU benchmarks, outperforming both general MLLMs and reasoning-based methods not only in LVU accuracy but also in training and inference efficiency.

Multimodal information systems increasingly route generated visual content back through the same vision-language index that informed its production, so the output must remain retrievable by the queries it was meant to serve. When the scene contains entities at vastly different scales, existing language-guided generators condition on a single, globally pooled text embedding and quietly drop scale-specific concepts, breaking concept-query retrieval even when pixel fidelity is high. We formalise this failure as semantic collapse and propose CERES, a closed-loop multimodal indexing framework that builds a three-level semantic pyramid, mines implicit concepts via a co-occurrence-aware router, performs scale-routed cross-attention into a lightweight U-Net generator, and verifies coverage by re-indexing the generated image with the same frozen VLM. A continuously differentiable soft-Jaccard coverage objective returns dense gradients to the 0.39M-parameter generator under explicit non-degeneracy conditions, and coverage is verified by an independent DINOv2 linear probe trained only on external scene and object labels. On four pansharpening benchmarks across seven settings, CERES delivers the new state of the art with the largest gains where scale variation is most extreme. It also improves concept-query retrieval Recall@5 by +14.0 points and image-text mean reciprocal rank by 0.19 over the strongest baseline, showing that the closed loop preserves queryable content rather than self-referential feature consistency.

Breast ultrasound diagnosis relies on clinically meaningful semantic concepts, yet most deep learning methods adopt end-to-end image-to-label paradigms that lack interpretability and robustness. While concept-based approaches offer a promising alternative, they often assume complete annotations or require multimodal inputs at inference, which significantly limits their real-world applicability. To tackle these issues, we propose Training-time Report-guided and Clinically Ordered Concept Editing (TRACE), a training-time report-guided framework that leverages structured radiology reports as privileged concept supervision while enabling image-only diagnosis at test time. TRACE refines image-derived concepts through a teacher-guided editing mechanism within a malignancy-aware ordered concept space. To address incomplete annotations, we introduce Strategic Concept Missing Training (SCMT) and train an image-only self-editor via edit distillation for autonomous concept refinement. Besides, we introduce BUSC, a concept-enriched benchmark linking images, labels, and structured attributes. Experiments across multiple datasets demonstrate that TRACE achieves superior performance and improved cross-domain robustness compared to existing methods.

Environmental exposures such as air pollution and greenness have been associated with affective and cognitive outcomes, but EEG and environmental datasets are rarely jointly georeferenced. We investigate whether literature-informed environmental priors can serve as an auxiliary geospatial modality for EEG-based affective-state classification when individual-level exposure data are unavailable. We combine 30-channel EEG from the EAV benchmark (42 participants, aged 20-30 years) with environmental representations derived from OpenAQ, Sentinel-2, Sentinel-5P, and OpenStreetMap data for Astana. A dual-tower architecture combines EEG-Conformer representations with a graph-based environmental encoder. Because the datasets are not co-registered, environmental context is treated as a literature-informed prior rather than measured exposure. Subject-level repeated splits, permutation and label-shuffling controls, dose-response reversal, and domain-shift experiments distinguish architecture-level gains from prior-dependent gains. The multimodal model achieves 76.2% accuracy versus 67.4% for EEG alone. Controls disrupting environmental-label structure retain part of this gain, indicating that the improvement is not attributable solely to environmental information. Replacing the Astana environmental distribution with an independently modeled Singapore distribution reduces accuracy to 72.8%. These findings demonstrate technical feasibility but do not establish an observed or causal exposure-affect association. The study provides a framework for future jointly collected mobile EEG-environment studies. Implementation: https://github.com/r11up/geo-cog

Long-form video understanding remains challenging for video agents due to the mismatch between query demands and evidence acquisition strategies. Although recent planning-before-perception methods outperform query-agnostic pipelines, they often rely on a single dominant strategy, either generation-based strategy or retrieval-based strategy, limiting their ability to handle diverse query demands. We propose Route2Look, a lightweight and model-agnostic framework for query-adaptive evidence acquisition in long-form video understanding. Route2Look operates in a Route-Look-Memorize loop with three tools: Global Browse for holistic context, Temporal Ground for explicit temporal cues, and Semantic Retrieve for semantic search. The core component is a routing policy that dynamically selects evidence acquisition tools based on the query. To build this policy, Route2Look adopts a two-stage design: first distilling the routing skill from differential contrastive analysis between generation-based and retrieval-based trajectories, and then applying the distilled skill with hard routing rules and continue-or-stop criteria during inference. Experiments on challenging long-video benchmarks show that Route2Look achieves state-of-the-art performance while maintaining strong frame efficiency across datasets and query types. Oracle routing analysis further reveals the potential of query-adaptive evidence acquisition for future long-form video understanding.

Temporal Knowledge Graph (TKG) extrapolation seeks to infer future facts from time-varying relational histories. Recent diffusion-based approaches improve uncertainty modeling through generative denoising, but their aggregated conditioning on subject histories may insufficiently distinguish query-specific evidence from non-salient historical facts, thereby diluting target-discriminative signals. To bridge this gap, we propose FreqDiff, a Frequency-aware Diffusion framework for TKG extrapolation. Specifically, FreqDiff formulates future object prediction as query-slot denoising and develops a dual-stream denoiser that integrates temporal dependency modeling with context-aware spectral calibration. The spectral branch synthesizes history-conditioned filters from learnable bases to adaptively re-calibrate denoising representations, while a frequency-domain regularizer is proposed to align the denoised target with the gold object in spectral space. Experiments on four public TKG benchmarks demonstrate that FreqDiff achieves state-of-the-art performance.

Vector graphics are prized for their resolution independence, compact storage, and direct editability, making differentiable optimization of their parametric primitives an attractive goal. Yet classical rasterization is discontinuous with respect to geometry, and existing remedies that smooth the forward pass demand increasingly elaborate heuristics as scene complexity grows. We trace this fragility to a gradient seesaw: design choices that improve forward geometric exactness can systematically degrade the induced gradient signal, and vice versa. To navigate this tension we introduce CubicSplat, a differentiable vector rasterizer that replaces Bézier closest-point solvers with uniform polyline surrogates whose geometric error is bounded at $O(S^{-2})$. The resulting static computation graph yields well-conditioned gradients by construction, while a compositing-derived visibility mechanism prunes degenerate primitives without auxiliary regularization. On DIV2K and Kodak benchmarks CubicSplat achieves state-of-the-art reconstruction quality with over 2 dB PSNR gain in the closed-fill setting, while training up to 4x faster than prior methods. The code is available at https://github.com/CubicSplat/repo

Evaluating language-guided mobile agents has recently shifted from rule-based to model-based approaches to achieve scalable and automated assessments. However, existing holistic evaluation paradigms process entire trajectories at once, leading to substantial context overload. Moreover, they primarily focus on task completion while overlooking operational safety. To address these limitations, we introduce CRATE, a novel two-stage VLM-as-judge framework for automated mobile agent evaluation that is compatible with both open- and closed-source models. Leveraging a step-level consequence reasoning mechanism, CRATE independently extracts task-relevant visual clues and infers action-conditioned state changes at each step. The resulting step-level textual evidence is then synthesized through trajectory-level aggregation to deliver an evidence-grounded evaluation of task completion. Building upon this evaluation scheme, we further extend CRATE to CRATE-S for operational safety assessment. Extensive experiments validate the effectiveness and robustness of both CRATE and CRATE-S. Powered by Qwen2.5-VL-72B-Instruct, CRATE achieves an F1-score of 0.833 on AndroidWorld (outperforming SPA-Bench by 20%), while CRATE-S reaches an F1-score of 0.697 on MobileRisk, demonstrating strong alignment with benchmark ground truths. Code is available at https://anonymous.4open.science/r/CRATE-D580.

Supervised fine-tuning (SFT) can degrade factual behavior outside the target domain. This degradation is often described as catastrophic forgetting, yet open-ended factual failures do not necessarily imply that the underlying facts have been erased. In this work, we identify a more specific phenomenon, factual access failure: after domain SFT, models can still recognize or rank the correct answer under constrained evaluation, while failing to produce it in closed-book generation. Through benchmark-level comparisons, same-fact multiple-choice and generation probes, and failure-mode analysis, we show that SFT-induced factual degradation reflects both genuine wrong-answer generations and expression-level failures such as verbosity, formatting mismatch, and exact-match artifacts. To address this problem, we introduce Recall-Anchored Distillation (RAD), a base-anchored self-distillation objective that preserves out-of-distribution generation behavior by aligning the adapted model with the original base model's soft continuation distribution on unlabeled OOD text. RAD requires no gold OOD answers, external judges, or labeled factual data. Across three backbones fine-tuned on MedMCQA, RAD recovers a consistent portion of the lost OOD recall while preserving target-domain adaptation. Compared with replay on the same OOD text, RAD shows that the key preservation signal is the base model's soft distribution rather than additional text exposure alone.

Vision-Language-Action (VLA) policies are vulnerable to localized physical perturbations, yet existing certified patch defenses target discrete labels and cannot directly certify continuous, temporally correlated actions. We introduce CertVLA, a certified defense for closed-loop VLA control under bounded patch and texture attacks. CertVLA proposes a calibrated region of behaviorally consistent actions, while deterministic covering masks ensure that at least one checked prediction is attack-free. Specifically, CertVLA normalizes action disagreement by the benign variation of each mask pair and accepts a single-mask anchor only when it remains consistent under every second mask. It then calibrates the resulting max-min-max episode score to provide finite-sample clean coverage. Conjoining query-level decisions extends the action certificate to the complete closed-loop rollout. Furthermore, we prove that against any adaptive attacker satisfying the bounded-support threat model, every rollout certified by CertVLA executes only action chunks consistent with attack-erased clean predictions. Under dual-mask rollout correctness, this consistency certificate further guarantees task success. The certificate is independent of patch content, generation method, and physical transformation. Experiments in simulation and the real world demonstrate the empirical and certified effectiveness of CertVLA against patch attacks, with additional simulation validation on texture attacks.

Deep learning-based Multi-View Stereo (MVS) has advanced significantly but often generalizes poorly to unseen scenes, particularly in occluded areas or regions with limited view overlap. To mitigate this, recent approaches integrate Depth Foundation Models (DFMs) into MVS pipelines to provide monocular depth priors. However, existing methods typically rely on a static, one-way fusion scheme, which fails to fully exploit the complementary strengths of both modalities. We propose a novel framework that overcomes this limitation by tightly coupling a DFM with a cascade MVS pipeline through a bidirectional mutual refinement strategy. Our method leverages MVS depth to resolve the scale ambiguity in monocular predictions, while the monocular depth, in turn, enhances the structural completeness and fine-grained detail of the MVS estimate. Furthermore, we introduce a prior-guided cost volume refinement mechanism that effectively integrates multi-view and monocular information via attention-based fusion and discretized depth bins, thereby promoting local geometric consistency. Extensive experiments demonstrate that our method outperforms state-of-the-art MVS approaches on standard benchmarks, producing more complete and generalizable depth maps with sharp boundaries. Furthermore, although not explicitly designed for sparse-view settings, our framework generalizes remarkably well, competing favorably with even dedicated sparse-view methods while maintaining a superior accuracy-efficiency trade-off.

Multi-agent pipelines that author formal documents must both read a requester's forms and write against them. We report a deployed tender-response system, running an open-weights model under sovereignty constraints, and evaluate it against human-written bids the same organisation actually submitted. On a blind comparison where the system had no worked example available, an LLM judge rated its answers at least as good as the human-submitted answer on $40$ of $55$ ground-truth sections, better on $4$, missing on none, and flagged one unsupported claim in total. Classifying every gap the judge identified shows that $68\%$ were content absent from the system's own sources -- knowledge the human author held and the pipeline was never given -- so only $6$ of the $15$ adverse verdicts involve a deficiency the system could have avoided. A divergence from ground truth is more often an information-availability result than a writing-quality one, and evaluations that do not separate the two understate such systems. Against this backdrop we report a conditioning asymmetry. It is well established that rendering documents as structural markup rather than flat prose improves extraction, and we reproduce that on three reading tasks. The benefit does not transfer to conditioning: converting a bid's \emph{instruction} material from prose to nested XML dropped answer quality from $74\%$ to $48\%$ under a paired comparison. We further find that naming a forbidden construction concentrates rather than removes it -- $96\%$ of surviving defects fall in the two forms the prompt explicitly names -- and that coupling a stochastic annotation to a deterministic windowing function moves the extracted requirement count from $68$ to $51$ on a byte-identical file. Structure belongs where the model reads; prose and self-applied tests belong where it writes.

Imitation learning for robotics depends on human demonstrations, some of which people may later ask to remove. Retraining without them is the natural reference, but its cost grows with policy and dataset scale, motivating cheaper operators that edit a trained policy. Metrics inherited from machine unlearning, such as forgetting loss or a single membership attack, do not establish what an edit removed from a policy acting in closed loop. We therefore introduce a retrain-calibrated audit that reads demonstration unlearning along two axes: behavior, whether the edited policy acts like one retrained without the removed demonstrations, and evidence, whether an auditor can still detect it was trained on them. The behavior axis measures action divergence to that retrain at matched states, calibrated by a floor built from independent retrains, so a policy at the floor is as close to a retrain as retrains are to each other. The evidence axis applies a per-demonstration membership attack against a retrain null, reporting both its rank and its absolute member-loss level, since rank alone accepts operators that inflate member losses past the null. A conformal test then combines both axes into one hypothesis of joint retrain consistency, against a fleet of independent retrains large enough to reject at conventional significance. Across five preregistered conditions on three real-robot policy classes and two simulation suites, the axes dissociate in both directions on one checkpoint, as an edit may repair task behavior while leaving evidence unchanged, or reduce evidence while moving behavior away from retraining. On the ACT arm, a redirect edit restores blind-scored robot success to 18 of 20 trials.

As scientific literature grows and papers increasingly under-report limitations, multi-agent LLMs offer a promising approach to systematically uncover these hidden failure modes. Here, we introduce Tree-of-Concerns, a multi-agent framework that deploys specialized skeptic personas, each operating through a category-specific analytical lens, as parallel debate trees to extract unstated limitations from scientific papers. Each persona conducts structured, evidence-grounded argumentation, while a Panel Review mechanism re-evaluates each surviving claim from all five perspectives to correct category drift and severity miscalibration. Through experiments on ToC-Bench, our benchmark of 414 research papers with 1,905 unstated limitations, sourced from reviewer-reported weaknesses and follow-up citation critiques, we demonstrate that ToC improves precision by 79% and coverage by 11% relative to strongest baselines, surfacing specific, evidence-grounded concerns that support reviewers in systematic evaluation.

Search Agents face a severe reliability crisis during reinforcement learning (RL) fine-tuning. Heuristic Top-K retrieval often causes critical evidence loss or noise inclusion, while over-confidence induced by progressive RL leads to hallucinated answers and redundant searches. To build highly reliable agents, we introduce Conformal Prediction (CP) and propose Conformalized Agentic Search (CAS). This framework establishes reliability guarantees on both the retrieval and training sides: on the retrieval side, an Adaptive Prediction Set (APS), a specific CP realization, translates statistical coverage into dynamic document truncation to construct prediction sets that are adaptive in size; on the training side, Adaptive Conformal Inference (ACI), a dynamic CP algorithm, dynamically constructs prediction sets with controllable coverage to quantify answer confidence, which is then used to penalize low-confidence trajectories within the Group Relative Policy Optimization (GRPO) objective, ensuring the model learns only from reliable ones. Experiments across single-hop and multi-hop QA datasets demonstrate that our framework significantly improves reasoning accuracy while drastically reducing redundant tool invocations, establishing a highly reliable and efficient agent paradigm. Our code is available at https://github.com/S1llyBird/CAS.

Streaming emotion understanding uses historical state while continuously interpreting current audio, often feeding the model's previous prediction back as context. We show that this history conditioning can distort current perception. On a balanced CREMA-D-Stream counterfactual diagnostic, changing only the injected previous emotion label while holding the audio fixed reduces current-audio accuracy from 72.50% to 30.42% and flips 65.69% of predictions. The effect is strongly label-asymmetric, with prior pull ranging from 4.76% to 98.20%, revealing a failure we call previous-belief contamination (PBC). To address PBC, we introduce EmoUpdate, a training-free framework that separates current-audio perception from historical state revision through three components: (1) a prior-blind acoustic firewall that prevents historical state from entering perception; (2) an evidence-shrunk causal belief filter that introduces history only after observation formation and retains label-asymmetric transition structure only when supported by observed evidence; and (3) a closed-form decontamination operator derived from the same counterfactual measurements for serving stacks where firewalling is unavailable. Across four SpeechLMs and two streaming emotion benchmarks, EmoUpdate achieves the best step accuracy and state-balanced accuracy in all eight model--benchmark settings, improving S-BAcc by up to 69.71 points and step accuracy by up to 38.41 points over the strongest controlled baselines.

Linear probes and activation steering have uncovered that vision-language models (VLMs) internally represent mental states such as agents' beliefs, knowledge, and intentions. However, it is unclear whether and how these representations are used by downstream predictions along these axes. To close this gap, we introduce Cross-Axis Routing Diagnostic (CARD), which steers activations along one axis while measuring the response of a different axis's prediction. Applied to open-weight VLMs on Relay Chain -- a new cooperative grid-world benchmark we propose -- we diagnose a critical routing failure: models fail to incorporate belief representations into their next action prediction, effectively leaving valuable information about their partners unused.

In non-stationary multivariate time series, different variables and samples often exhibit heterogeneous latent dynamic states, while existing deep forecasting models usually compress them into a unified end-to-end mapping, leading to suboptimal modeling of time-varying dynamics and limited interpretability regarding which forecasting mechanism is activated under different latent states. To overcome these limitations, we reformulate time series forecasting as a unified framework of latent temporal state identification and interpretable expert routing, and propose Fuzzy-MoE, a fuzzy logic-based dynamic Mixture-of-Experts model. Fuzzy-MoE consists of multiple parallel expert mapping networks and a dual-view fuzzy router. By jointly exploiting local convolutional dynamics and global segmented statistics, the router infers latent temporal states and computes expert activation strengths through learnable Gaussian membership functions, enabling explicit IF-THEN rule-based expert selection. This fine-grained routing strategy allows different variables within the same sequence to activate different experts, effectively capturing heterogeneous temporal dynamics while improving model interpretability. Experimental results on multiple public time series benchmark datasets show that Fuzzy-MoE significantly outperforms mainstream forecasting methods in forecasting accuracy. Moreover, fuzzy memberships and rule activations provide interpretable routing diagnostics, demonstrating the effectiveness of the proposed framework in both forecasting performance and mechanism transparency. Unlike traditional MoE models that use black-box routing, Fuzzy-MoE`s routing is based on clear, interpretable fuzzy rules. This makes the expert selection transparent and traceable.

While multimodal retrieval-augmented generation (RAG) systems increasingly rely on images as external knowledge sources, the introduction of poisoned visual evidence can severely compromise multimodal large language model (MLLM) generation. Unlike prior attacks that rely on altering textual metadata, we introduce Vis-Poison, a novel visual knowledge poisoning attack where the poisoned image itself is the attacker-controlled payload, without manipulating captions, summaries, metadata, or other associated text. Specifically, this attack is instantiated through an automated multi-agent method that constructs visually plausible poisoned images. To assess its impact, we evaluate Vis-Poison across two representative multimodal RAG pipelines, four embedding models, and six generation models. Empirically, Vis-Poison achieves an end-to-end attack success rate of 40.16\% to 65.40\% against 30k-entry multimodal knowledge bases in \emph{black-box} settings. Moreover, Vis-Poison remains effective against various MLLMs that can answer correctly from parametric knowledge alone, with an average success rate above 60\%. Code and data are available at https://github.com/SWUFE-DB-Group/Vis-Poison.

Promptable segmentation models provide a reusable interface, but direct transfer to automatic infrared small-target segmentation (IRSTD) exposes a mismatch between spatial prompts and target-domain mask responses. In a diagnostic using target-covering loose-box prompts deterministically derived from test reference masks, the best official SAM2.1 results are only 4.69%, 1.64%, and 2.28% IoU on NUAA-SIRST, NUDT-SIRST, and IRSTD-1K. We introduce SPARK-SAM (Self-Prompt Adaptation with Response Knowledge for SAM), which learns target-domain response knowledge and conditions the decoder through an image-conditioned joint self-prompt state. Training combines benchmark-mask supervision with reliability-aware response guidance. SPARK-SAM achieves 75.78%, 86.49%, and 68.34% IoU with 0.726M additional parameters, ranking first on two benchmarks among 14 retrained SAM variants and adaptations evaluated as automatic image-to-mask methods. The staged IRSTD-1K diagnostic shows that response adaptation reaches most of the final IoU before the predicted points acquire reliable target grounding. Prompt supervision aligns the predicted prompt candidates with target locations, and frozen-weight interventions measure output sensitivity to the joint self-prompt state. Matched ablations show consistent accuracy gains from response guidance and high-resolution prompt refinement across all three datasets. Code is available at https://github.com/Sakauma/SPARK-SAM.

Identity-preserving video generation aims to synthesize videos that follow natural-language instructions while maintaining the visual identity of a given subject. Recent commercial video generation models have achieved strong visual quality and motion realism, but they still suffer from identity drift, incomplete instruction following, and missing visual details under complex prompts. Since these models are usually closed-source black boxes, directly improving them through parameter optimization is often infeasible. We therefore propose Agentic Enhancement and Semantic Repair (AESR), a lightweight enhancement framework for identity-preserving video generation. To improve prompt construction before generation and mitigate the above failures, AESR introduces a global agentic prompt enhancement module. This module learns model-specific prompting formats from official documentation, acquires human-centered video generation priors from human-interaction data, and accumulates test-domain identity-preserving generation experience into a reusable playbook through an agentic loop. To further repair errors in videos generated with enhanced prompts, AESR introduces a sample-level visual semantic repair module, which uses a VLM to locate erroneous video segments and design repair instructions, edits selected frames into explicit visual references, and guides a video editing model to fix local semantic or identity-related errors. We also adopt a lightweight Mixture-of-Experts selection strategy to choose reliable outputs from different generation and refinement paths. Under the official evaluation protocol of the ACM MM 2026 Identity-Preserving Video Generation Challenge, our system MIPL\_Video ranked first in Track 1, demonstrating the effectiveness of AESR for practical identity-preserving video generation. The code is available at https://github.com/oceanflowlab/AESR.

State-of-the-art 3D reconstruction models, whether from visual, range, or both, tend to underperform on thin objects. This is partially due to the small amount of space such objects occupy in RGB images and in 3D point clouds. To test the extent of their errors, we collected the first thin object dataset comprising of synchronized RGB images, depth maps, and tactile response maps, where each frame is associated with inertial measurements, camera pose and calibration, and groundtruth depth and segmentation maps obtained from laser scanning of thin objects. We hypothesize that tactile data can aid in the reconstruction of thin objects as their response maps provide local shape and deformation information. Our dataset, termed VisTa3D, comprises of 387 scenes covering 70 thin objects over 17 environments. We benchmarked current 3D reconstruction models on VisTa3D and found that, indeed, they exhibit low fidelity on thin objects. To test if tactile data can help, we introduce the first visual-range-tactile 3D reconstruction model as a baseline. Code and data: https://huggingface.co/datasets/shaniaguo/VisTa3D.

Manipulating moving objects requires a policy to anticipate contact events, yet vision-language-action (VLA) policies are commonly fine-tuned from the current observation alone. World action models (WAMs) learn predictive dynamics, but running a video-scale teacher or explicitly imagining future frames at deployment is costly. We introduce ForeTime-VLA, a dense pi0.5 policy that distills a future-aware, action-equivalent representation from a frozen Fast-WAM-derived teacher while remaining causal at inference. Offline, current and future video latents are compressed into a whitened 64-D target. Online, an eight-frame history encoder predicts this target together with manipulation phase and normalized time-to-transition. Four future tokens and one phase token condition the VLM prefix, while the predicted future and transition horizon condition the action expert. Training retains the original flow-matching action target and adds cosine, relational geometry, phase, time-to-transition, and action-equivalence objectives. On a deduplicated conveyor-belt dataset, we compare 40k-step checkpoints on 768 matched windows per split. Test MAE decreases from 0.134119 to 0.130593 (2.63%; paired-bootstrap 95% CI: 0.82-4.48% improvement), and test L2 decreases by 3.02%, at a 2.46-2.93% latency cost. In quantitative real-robot evaluation, ForeTime-VLA achieves 81.1% stationary and 58.9% slow-moving grasp success, exceeding the next-best reference by 12.2 and 22.2 percentage points, respectively. Across three belt speeds, it completes 44/90 grasps versus 23/90 for pi0.5, including 11/30 versus 2/30 at fast speed. The agreement between offline orientation gains and reduced real-robot contact-pose failures supports causal future-token distillation as an effective way to improve dynamic manipulation without deploying the world-model teacher.

Language-model agents can improve after failure or carry text across episodes without revising what counts as success. We study the narrower attribution problem of criterion revision: when criterion K0 accepts an outcome violating a broader commitment B, what observations justify saying that the system formed and persistently used K1? We require five non-compensatory conditions: criterion-failure detection, a model-emitted proposal, new-episode transfer, intervention sensitivity on the claimed carrier, and preservation. We evaluate CMB-0.1 on twelve cross-domain cases and four arms: stateless inference, append-only history, model-generated but harness-committed state, and evaluator-written oracle state. Seven mechanism fixtures yield 84 deterministic scorer trials; four local quantized artifacts yield 96 calls and 192 model-case-arm trials. No model trial satisfies all five conditions, but this zero does not establish general capability absence. Eleven calls remain invalid after one retry; several commitments disclose the target distinction; the harness performs commits; deletion reuses a stateless call; and conflict changes multiple factors. Qwen2.5-7B answers every transfer and preservation item without revision state, exposing zero-state reconstruction. These failures make CMB-0.1 an instrument-calibration result rather than a model ranking. We derive a prospective, trace-anchored CMB-0.4 protocol requiring concealed transfer, explicit WRITE/NO-WRITE/ESCALATE actions, a separately logged policy-selected commit, matched interventions, repeated hidden items, and a frozen executable oracle. It is a successor design, not a completed confirmatory result. The paper contributes a measurement chain, an empirical diagnosis of its first implementation, and a more discriminating protocol for future tests of criterion revision.

Convolution is a principal computational bottleneck in deep neural networks, and its efficiency depends on tight integration between algorithms and GPU hardware. Existing GPU convolution methods suffer from large memory overhead, poor cache utilization, limited effectiveness across kernel sizes, or numerical instability. This work extends the im2win paradigm -- a universal, memory-efficient convolution method with contiguous memory access for all kernel sizes -- to run efficiently in full precision on CUDA cores and half precision on tensor cores. By introducing new kernel designs and optimizations such as zig-zag memory access and asynchronous data movement, im2win efficiently exploits hardware-accelerated half-precision matrix multiply-accumulate operations. Across twelve CNN benchmarks, im2win achieves up to 2.8x higher TFLOPS than its CUDA core implementation, 1.4x higher than cuDNN, and 6.4x higher than GEMM-based convolution with cuBLAS, while using as little as 53% and 35% of their memory, respectively. These results establish im2win as a unified, high-performance convolution framework for modern GPU architectures.

Open-world 3D affordance grounding requires localizing functional object parts in 3D given free-form language queries. Existing methods typically assume pre-built object-centric 3D geometry and closed affordance ontologies, limiting deployment from raw RGB observations. We present AffordAny, an end-to-end framework that uses one monocular RGB image to construct large-scale text-conditioned 3D part supervision, ground affordances with a frozen vision-language model (VLM) guided decoder, and improve open-world generalization through pseudo-label self-training. Our automated pipeline produces a benchmark of 5,334 objects and 10,633 part-level samples spanning 473 categories, an order-of-magnitude increase in categorical diversity over prior work. The decoder progressively fuses frozen Cosmos-2B features with 3D geometry through spatial projection, instruction-conditioned semantic compression, and bidirectional geometry-semantics interaction. Minimal-perturbation pseudo-label self-training further adds new objects without human annotation. Under a systematic generalization protocol evaluating unseen objects, unseen categories, and unseen instruction paraphrases, our approach achieves 0.428 IoU on unseen objects and 0.315 IoU on unseen categories after self-training, with unseen-category mIoU improving by 6.3% relative (p<0.01) and an instruction sensitivity gap of only 0.105, demonstrating effectiveness and robustness of our method.

Generative AI can now produce highly realistic images, yet current models still exhibit subtle but critical defects that undermine their reliability. While existing AI-generated image (AGI) evaluation benchmarks have made notable progress, comprehensive AGI defect diagnosis remains underexplored. To bridge this gap, we introduce AGIDefect-4K, a richly annotated dataset of 4,000 images from 15 state-of-the-art generative models spanning both open-source and closed-source systems. AGIDefect-4K features hierarchical defect annotations: (1) detection labels identifying whether defects exist, (2) pixel-level segmentation masks localizing defective regions, and (3) detailed textual explanations characterizing defect types and their perceptual impact. Each image is further annotated with an overall quality score. Building on this, we present AGIDA (AGI Defect Assistant), a baseline framework leveraging Multimodal Large Language Models (MLLMs) for joint defect detection, localization, explanation, and quality prediction. Comprehensive benchmarking on AGIDefect-4K reveals that AGI defect understanding remains challenging, underscoring the value of this dataset. The dataset is publicly available at https://github.com/sxfly99/AGIDefect-4K.

High-performance ML systems increasingly rely on GPU kernels whose editable source is unavailable, generated, or too distant from final machine code to expose remaining optimizations. Existing LLM kernel optimizers and autotuners mainly operate on CUDA, Triton, HIP, or tensor-program source and validate against reference implementations. We study a stricter setting: optimizing an already compiled AMDGPU code object, where the deployed binary is the only behavioral oracle. We present AsmEvo, an agentic assembly-level optimizer for AMD GPU kernels. Given an AMDGPU code object K0, AsmEvo reconstructs a reassemblable representation, proposes low-level edits with a long-horizon agent, rebuilds an ABI-preserving optimized object, and accepts candidates only after differential verification against K0 under identical launches. AsmEvo combines code-object recovery, metadata-aware rebuilding, profiling-guided hot-window editing, correctness-gated timing, and conservative in-place patch fallback. We conduct extensive experiments with AsmEvo on various AMD GPU kernels. On MI308X, AsmEvo improves 29 of 30 selected KernelBench kernels, reaching 1.35x geometric-mean and 3.88x maximum speedup. On MI300X production workloads, it improves all evaluated AITer binaries and vLLM/SGLang Triton assembly kernels, reaching 1.09x/1.31x and 1.18x/1.34x geometric-mean/maximum speedups, respectively, while preserving functional equivalence.

Real-world semi-supervised learning (SSL) often encounters significant challenges with long-tailed label distributions and noisy pseudo-labels, which hinder generalization and amplify confirmation bias. In this work, we introduce a novel framework, Gaussian Bridge Consistency (GBC), to address these challenges by constructing semantic interpolation paths between unlabeled samples and high-quality class anchors. Our method maintains a dynamic Prototype Atlas that stores a diverse and evolving set of labeled and pseudo-labeled exemplars per class. For each unlabeled instance, GBC forms a class-conditional Gaussian Feature Bridge in the latent space, enabling the student model to traverse a smooth trajectory from uncertain predictions to reliable class prototypes. A bridge consistency loss is applied along this path to enforce alignment with a geometrically interpolated target distribution. Furthermore, we propose BridgeMix, a confidence-aware feature mixing strategy that interpolates both sample and anchor pairs to amplify cross-sample generalization. Extensive experiments on CIFAR10-LT and ImageNet-LT (USB benchmarks) validate the robustness and effectiveness of GBC under realistic long-tailed SSL settings, consistently improving long tail-class performance without sacrificing scalability.

Animating articulated 3D meshes via text requires satisfying strict kinematic constraints, modeling causal interactions between parts, and achieving instruction fidelity. Due to the absence of task-specific training data and explicit articulation supervision, existing data-driven mesh animation methods are largely inapplicable to this setting. To address this, we propose ArtiMo, a novel agent-driven framework for text-guided articulated mesh animation. Operating in a zero-shot manner, ArtiMo develops an agentic pipeline powered by Large Language and Vision-Language Models (LLMs/VLMs) to orchestrate motion generation. By synergizing the explicit kinematic constraints of URDF with the agent's reasoning and planning capabilities, it effectively produces causally coherent part motions and interactions without requiring model fine-tuning. To ensure motion correctness, the agent additionally utilizes a visual self-improvement mechanism: generated animations are rendered into compact keyframes and motion cues, enabling the VLM to iteratively diagnose and correct errors. Furthermore, we contribute a new benchmark dataset spanning 21 articulated object categories, featuring high-quality motion annotations enriched with causal relationships. Extensive experiments demonstrate that ArtiMo significantly outperforms baselines, particularly on complex, causally driven motions. The project page is available at https://zou-2004.github.io/ArtiMo/.

The advancement of modern integrated photonics is frequently bottlenecked by device design workflows that rely heavily on manual simulation and expert intuition. While inverse design offers an alternative, it remains constrained by expert supervision and a lack of end-to-end automation. To address these issues, we present VortexChat, an agentic framework for the autonomous, end-to-end inverse design of integrated photonic devices directly from natural language specifications. VortexChat couples a large language model (LLM) decision agent with topology generation, gradient-based refinement, and full-wave electromagnetic simulation. This closed-loop architecture enables the system to iteratively decompose design objectives, orchestrate computational tools, and update strategies based on feedback with minimal human intervention. Constrained by the absolute metrics of the Vortex100 Benchmark, VortexChat autonomously generates devices that strictly meet all predefined performance thresholds without any human-in-the-loop. As an experimental demonstration, we fabricated a broadband terahertz perfect vortex beam multiplexer, autonomously designed by VortexChat, with measurements confirming high-efficiency operation, high mode purity and low inter-channel crosstalk in agreement with full-wave simulations. These results demonstrate that an LLM agent can assume key aspects of expert decision-making in photonic inverse design while maintaining physical fidelity and fabrication feasibility, providing a scalable route towards autonomous design of complex integrated photonic systems.

Retrieval-augmented generation (RAG) has no model of time: when a fact changes across a coding session - a function is renamed, an endpoint moves, a dependency is bumped - RAG retrieves both the old and new value with near-identical similarity and cannot tell which is current, so it serves the superseded value. Paper 1 showed, on synthetic single-value benchmarks, that a deterministic (subject, relation, object) supersession memory eliminates this failure. Here we validate it end-to-end on real software history. From 707 real GitHub issues (SWE-bench Lite + Verified) we extract 130 clean atomic state transitions, a fix that changes one identifiable value from a pre-fix to a post-fix form, and render each marker-free (the stale and current statements differ only in the value). On this set, MemStrata reaches 0.91 answer accuracy versus RAG's 0.57-0.59; and, the structural result, when forced to answer RAG serves the superseded value 36-38% of the time (an LLM reranker does not help) while MemStrata drives this to ~0, at RAG retrieval latency (~2.1 s vs ~18 s for the reranker). We are explicit about scope: only ~18% of real fixes are clean atomic transitions; Paper 2 isolates the memory mechanism on that class, and extraction coverage of the remaining fixes is the orthogonal problem we defer to follow-on work. A real product bug surfaced and was fixed during the study (a case/punctuation-insensitive value comparison), with the moat property (deterministic-supersession accuracy on clean code mutations) preserved and verified.

We study reinforcement learning (RL) in Continuous-Time Jump Markov Decision Processes (CTJMDPs) featuring general discrete state spaces (which need not possess a vector space structure) and continuous/discrete action spaces. The setup covers many well-known applications in operations such as multi-product dynamic pricing with capacitated resources (Gallego and van Ryzin 1997). To model the exploration-exploitation tradeoff, we formulate an entropy-regularized continuous-time control problem with stochastic policies. Recent continuous-time RL techniques such as $q$-learning for controlled diffusions in (Jia and Zhou 2023) focus on continuous state spaces $\mathbb{R}^d$ and rely heavily on semimartingale theory in $\mathbb{R}^d$ for their theoretical analysis. Consequently, their methods cannot be directly applied to CTJMDPs with general discrete state spaces, which may lack the algebraic addition and subtraction structures inherent to Euclidean spaces. To bridge this gap, we establish the theoretical foundations of $q$-learning for CTJMDPs and develop model-free $q$-learning algorithms. Compared to naïve time discretization and approximating CTJMDPs using discrete-time MDPs, our approach has several conceptual and empirical benefits. Numerical experiments in network dynamic pricing (Gallego and van Ryzin 1997) show that our proposed RL algorithm reliably learns near-optimal policies and consistently outperforms standard benchmark methods, demonstrating superior solution quality and effective scalability to large-scale network instances.

Many agentic systems must repeatedly choose between acting and abstaining, making faithful reasoning important for oversight: an explanation is useful only if it reflects the computation that produced the action. We study this problem through intervention timing in multi-party conversation, where an assistant must decide whether to speak or remain silent. This setting exposes class imbalance, asymmetric action costs, and the possibility that exposing reasoning changes the policy being audited. Using Qwen3-8B, decoded with or without chain-of-thought reasoning, we compare direct decision policies, reasoning policies, supervised fine-tuning, and reinforcement learning. We find a capability-auditability tradeoff: the strongest direct policy achieves higher quality but exposes no reasoning to inspect, while the reasoning policy provides a trace at the cost of lower performance, particularly recall of true intervention opportunities. Supervised fine-tuning either suppresses reasoning or preserves it without improving decision quality, while reinforcement learning also fails to improve the reasoning policy. We identify one mechanism underlying this failure: group relative objectives provide no learning signal on confidently wrong prompts when sampled rollouts all select the same action. Controlled activation probes and behavioral ablations show that standard faithfulness methods can overstate evidence that exposed reasoning reflects the underlying decision process. Probability-based metrics saturate under confident decisions, probes are vulnerable to class imbalance and textual leakage, and reasoning ablations can confound reasoning content with changes in inference mode. Together, these results show that exposing reasoning can change an agent's action policy rather than simply make it observable. We provide controls for evaluating reasoning-based oversight of agents that can act or abstain.

Public datasets for agricultural disease detection are usually judged fit for use from reported metrics, which say nothing about whether the annotation scheme is internally consistent. On one public grape disease dataset (3288 images, 11995 boxes, 6 classes), varying model capacity, input resolution and detection paradigm yields a test-set mAP50 range comparable to seed-to-seed noise, with the bottleneck at small objects across all five architectures. The finding lies on the data side: one class is annotated at whole-leaf level (median box area 43.16% of the image) while the other five are annotated at lesion level. On 5156 cross-species images containing no grape, 65.7% of the false-positive boxes fall into that one class, an over-representation of 13.41x relative to its share of the training annotations. Counterfactual retraining establishes a causal effect of granularity on the magnitude of the shortcut: shrinking only that class's boxes cuts its cross-species false positives by 66%, and a placebo control confirms the effect is specific to the manipulated class. A manipulation in the opposite direction, with criteria registered in advance, returns a negative result: coarsening the finest class to whole-leaf level (0.57% to 40.37%), matched in box count and share of annotations and with higher in-distribution AP, still leaves its cross-species false positives at zero boxes, while the unmanipulated original class holds 50.0% of them. Annotation granularity is therefore a modulator of this shortcut, not its cause: it can amplify or attenuate a sink that already exists, but cannot create one, and what fixes the destination remains open. We also give a granularity screening statistic requiring neither images nor training, and show airborne lesion-level detection to be optically out of reach. The failure mode is invisible to in-distribution evaluation.

Enterprise adoption of large language models in finance is constrained less by fluency than by trust: in Financial Planning and Analysis (FP&A) and other regulated workflows, an answer is usable only if it is traceable to authoritative sources and auditable after the fact. This paper argues that retrieval-augmented generation for enterprise finance should be evaluated on auditability alongside accuracy, and presents the Knowledge-Driven Analytics Framework (KDAF), which builds ontology-driven knowledge systems through six iterative stages and retrieves evidence via Context-Aware Relevance Propagation (CARP), so that every retrieved fact carries its relationship type, confidence, and source lineage. An evaluation on FinanceBench (145 questions) compares KDAF against zero-context inference, BM25, concept-weighted lexical retrieval, and ungrounded graph traversal. First, retrieval is necessary: zero-context inference reaches 4.1% correctness against 10-12% for retrieval-augmented conditions. Second, on answer correctness the retrieval conditions are statistically indistinguishable (KDAF vs BM25: -0.007, 95% CI [-0.021, 0.000]), so accuracy alone does not justify structured retrieval here -- a negative result we report explicitly. Third, on auditability the ordering reverses: KDAF attains the highest citation traceability F1 (0.515), exceeding ungrounded traversal by +0.027 (CI [0.006, 0.050]) and BM25 by +0.052 (CI [0.024, 0.083]), intervals excluding zero. Graph-structured retrieval also admits no evidence from outside the question subject entity (0 of 426 items, against 16.8% and 20.2% for lexical baselines), and every selected item resolves to a complete provenance chain. We argue that auditability, not accuracy, is the axis on which ontology-grounded retrieval earns its cost.

Traffic sensors commonly record flow, speed, and occupancy, but standard traffic flow forecasting benchmarks and models rarely exploit all three raw measurements reliably. Although speed and occupancy provide sensor-native traffic-state information beyond flow alone, existing releases often omit these variables, replace them with proxies, or contain logically inconsistent records. Moreover, direct empirical risk minimization over three-variable inputs may exploit regime-dependent shortcuts, as the relationships among flow, speed, and occupancy vary substantially between free-flow and congested states. We introduce \textbf{PEMSB-3V}, a public benchmark suite that preserves raw flow, speed, and occupancy measurements from PeMS detectors for flow prediction. We also propose \textbf{RiskTraf}, a model-agnostic risk-extrapolated residual plug-in. For each trained spatio-temporal backbone, RiskTraf freezes the selected checkpoint and learns a lightweight zero-start residual head from historical speed and occupancy. The residual head constructs ordered traffic-risk environments and optimizes horizon-wise flow corrections with a risk extrapolation objective, thereby mitigating regime-specific shortcut correlations without modifying the backbone. Extensive experiments demonstrate that RiskTraf consistently improves diverse forecasting backbones and outperforms debiasing and distribution-shift adaptation methods. Our code and benchmark are available at https://github.com/Guangyu4/RiskTraf.

Agents learn to act through interaction with environments, yet the environments used for training are often manually constructed or synthesized around predefined tasks and benchmarks. This task-centric paradigm makes it difficult to scale environments that reflect realistic and evolving workflows where diverse tasks can naturally emerge from the underlying world. We introduce AgentMercury, a scalable framework for synthesizing executable environments from high-level business scenarios. Rather than constructing an environment for a specific task, AgentMercury first instantiates a persistent world with entities, services, tools, state, and executable cross-service invariants, from which diverse tasks and interaction trajectories can subsequently emerge. We construct 4,783 executable environments spanning 14 industries and 50 countries, and use them as training substrates for reinforcement learning. Despite being generated without targeting the evaluation benchmarks, policies trained on these business-oriented environments improve substantially on both enterprise workflows and out-of-domain benchmarks spanning reasoning, coding, scientific computing, and tool use. In our experiments, Qwen3.5-4B improves from 12.3 to 15.7 on EnterpriseOps-GYM and from 45.9 to 56.0 on AIME26 after training on AgentMercury environments. We further show that the construction process itself can be learned: fine-tuning Qwen3.5-35B-A3B on construction traces increases executable-world authoring success from 3.3% to 83.3% on held-out business scenarios. These results show that scenario-grounded environments can provide useful and generalizable learning signals beyond benchmark-specific training, while their construction can itself become a learnable capability.

Agentic retrieval-augmented generation (RAG) interleaves retrieval, reasoning, and answer generation across multiple hops. A retrieval error at hop 1 can surface only as a wrong answer at hop 3, while later retrieval can also repair the trajectory. This paper introduces AgenticRAG-FP, an interventional benchmark for causal failure attribution in agentic RAG. The benchmark injects a certified fault at a specified hop, re-executes the downstream trajectory, and evaluates diagnosers against the known intervention. Its central question is whether a post-hoc trace still identifies the injected hop after the suffix changes. In the completed strict dense Claude Haiku 4.5 sweep on 80 three-hop MuSiQue questions, coverage-based diagnosis is 0.91 at hop 1 and 0.00 at hops 2 and 3 (n=43,36,21 failed trajectories). A smaller content-corruption study changes an answer-bearing or bridge fact in topically intact evidence. At depth 2, where 18 failed cases remain after filtering, coverage-based diagnosis is 0.00 and a frozen-hop counterfactual probe is 0.67 in an exploratory pooled comparison. Depth-3 content estimates are descriptive only because they contain three failed cases. These results make propagation depth an explicit evaluation axis for diagnosing agentic RAG failures while distinguishing broad evidence of post-hoc signal loss from small-sample method comparisons.

Soft, wearable robotic devices can deliver haptic feedback to support a wide range of tasks, such as extended reality, training various skills, and rehabilitation. Pneumatic actuation can deliver complex haptic feedback, is lightweight and compliant, and can be incorporated into textiles, making it promising for wearable applications. These soft pneumatic devices, however, typically require a valve and input for each pneumatic actuator, making it challenging to develop fully portable devices for at-home use. In this work we present a pneumatic unit for logic-based sequential excitation (PULSE). The PULSE is a flat, textile-based pneumatic actuator with embedded fluidic logic. By combining these actuators into a fluidic ring oscillator, we decreased the typical amount of required pneumatic inputs for a haptic forearm sleeve by 60%, with the ability to scale. We built the ring oscillator by optimizing design variables to reach desired periods of oscillation. We demonstrated a set of tactile stroking cues with periods ranging from 1.16 to 1.56 s and forces ranging from 1.07 to 2.04 N. We assessed the sleeve's ability to render differentiable, pleasant, and continuous haptic cues in a user study. The forearm sleeve containing PULSEs successfully delivered four directional cues and guided users to target wrist angles with fast reaction times, low overshoot amounts, and a 93.3% average accuracy of correct initial directions.

Text-guided zero-shot object counters excel at spatial localization but categorize poorly on novel or fine-grained classes: natural language is too coarse to fully specify visual identity, so they fail to separate visually similar distractors. Few-shot counters sidestep this with visual exemplars, but require manual annotations on every image. To resolve this dilemma, we introduce RECOUNT, a plug-and-play framework for image-guided zero-shot counting. Rather than specify a category with a text prompt, our key insight is to specify it visually, from a single off-scene reference image. However, we find that a lone reference image provides narrow coverage of a category's appearance and is unreliable across diverse scenes. We therefore repurpose a diffusion model as an automated contrastive data engine that expands the reference into a diverse exemplar gallery, supplying the discriminative detail that text cannot. RECOUNT preserves the class-agnostic proposals of any frozen counter and offloads categorization to a separate visual module (a frozen backbone with a lightweight head trained on this synthetic data) that matches each proposal against the target and distractor galleries. Applied to a frozen counter, RECOUNT attains the best zero-shot accuracy on both benchmarks, cutting counting error (MAE) by 55% on LookAlikes and 21% on PairTally relative to the strongest prior zero-shot counter.

Accurate characterization of integrated nonlinear photonic devices is often limited by unknown facet-coupling losses and fabrication-induced non-uniformities, leading to systematic over- or underestimation of the intrinsic on-chip performance. Here, we present and demonstrate a unified bidirectional characterization framework that exploits nonlinear interactions under forward and backward propagation to independently extract facet-specific coupling efficiencies, intrinsic nonlinear conversion efficiency, and the longitudinal quasi-phase-matching profile using only classical power measurements. We experimentally validate the method using sidewall-poled thin-film lithium niobate waveguides, obtaining a normalized second-harmonic generation efficiency of $(1850 \pm 20)~\%\mathrm{W}^{-1}$ while simultaneously demonstrating broadband non-degenerate optical parametric amplification and parametric generation spanning more than $10~\mathrm{THz}$. Our framework is non-destructive, relies solely on classical power measurements, requires neither time-intensive microscopy nor calibrated internal references, and is compatible with wafer-scale testing, providing a general route to rigorous benchmarking and high-throughput characterization of photonic devices across material platforms.

Multi-agent LLM systems split work across models, so answering often requires knowledge that sits in another agent's context: a Sharer has encoded information that a Receiver needs to complete its task. They usually communicate by exchanging text, which puts autoregressive decoding on the critical path and reduces the exchange to a discrete message written without sight of the receiver's state. Recent latent protocols instead translate the sharer's key-value (KV) cache into the receiver's: C2C supports heterogeneous models but requires both to read the same input, while LCF-X removes this shared-context requirement through position-free sharer-cache pooling. Three restrictions remain: LCF-X compresses the sharer alone, supplies the same layer-local summary to every receiver position with no joint cross-layer memory to retrieve from, and assumes matched layer count and KV geometry. We introduce XKV, which lifts all three: learned-query attention pools both caches; self-attention over receiver-aligned layer tokens, with a learned layer map reconciling different depths, mixes the pooled summaries into a compact joint memory; and a shared position decoder lets every raw receiver cache position retrieve its own per-head-gated residual in the receiver's native KV geometry. Both models stay frozen and may differ in family, depth, KV-head count, head dimension, and tokenizer; only the translator is trained. Across 45 dataset-model-pair settings (six heterogeneous and three same-model ordered pairings, five datasets), XKV attains the highest macro score and best average rank, improving on LCF-X on every dataset (by 4.6 exact-match and 4.2 F1 points on ROPES) and surpassing text communication on four of the five, while training 76% fewer parameters and translating a cache pair 10.3x faster (5.8 vs. 59.9 ms); end to end, XKV is 26% faster than LCF-X and 6.8x faster than text communication.

Grape leaf disease recognition is important for precision agriculture, enabling early diagnosis, timely intervention, and improved vineyard management. Although deep learning has achieved strong results, many studies rely on few datasets, often acquired under controlled conditions, and may not reflect real vineyard challenges such as complex backgrounds, variable illumination, occlusion, leaf pose, disease severity, and device differences. This paper presents a dataset-centric benchmark of deep learning methods for grape leaf disease classification and detection. We analyze publicly available datasets in terms of disease categories, annotation types, acquisition conditions, image characteristics, class distributions, provenance, and task suitability. Representative models are evaluated in three settings: image-level classification, region-level classification, and object detection. Classification is assessed using accuracy, while detection is evaluated using mAP@50 and mAP@50:95. Cross-dataset experiments further examine transfer between datasets with compatible disease categories but different visual and annotation characteristics. Results show near-saturated classification performance on several controlled or derivative datasets, greater difficulty on heterogeneous datasets, and substantial variation in detection performance across annotation settings. Cross-dataset performance drops sharply, especially for object detection, indicating that shared disease labels do not necessarily define equivalent recognition tasks. The benchmark emphasizes dataset provenance, realistic field evaluation, annotation compatibility, and external validation for reliable vineyard disease recognition.

Panels of inexpensive LLM judges increasingly make accept-or-escalate decisions. In factuality settings, accepting a claim because several reference-free judges agree can create a hidden risk: agreement may reflect shared false-negative blind spots rather than independent evidence. We introduce JuryProbe, an empirical consensus-risk diagnostic for reference-free factuality judge panels, paired with a calibration-based routing policy. JuryProbe estimates consensus risk from a labeled calibration probe using false-negative-only (FN-only) judge correlation and false-consensus lift; when flagged high-risk, reference-free majority accepts are routed to the same judges with trusted references. On audited FEVER corruptions, reference-free panels show correlated false negatives (FN-only correlations 0.402 and 0.368; lifts 3.13x and 18.13x), while unanimous false consensus drops to zero under a trusted-reference best-case diagnostic on both minimal-pair and non-minimal-pair evidence. In flagged settings, the routed policy is by construction equivalent to grounding every reference-free majority accept (verified in 34/34 splits): improvement comes from accept-conditioned grounding, while the diagnostic determines whether to activate it. A fixed, pre-specified rule flags 8-10 of 10 splits across synthetic, benchmark-authored, and scientific families and 0 of 10 on a negative control, where standing down avoids 28% of reference acquisitions at a 0.004 increase in false accepts. False-accept reduction persists under weak BM25 retrieval at substantial coverage cost, while stale stand-down labels require periodic recalibration. JuryProbe provides no formal risk guarantee and does not establish reliable stand-down on natural panels; its supported contribution is an empirical diagnostic of high-risk panel error dependence.

We describe the winning entry to the MoCha 2026 Benchmark and Challenge on Parkinsonian Gait, which predicts MDS-UPDRS gait severity from canonicalized SMPL motion recorded at clinical sites unseen during training. The system reaches 0.6945 macro-F1 on the hidden test and ranked first of 58 entries, ahead of the runner-up at 0.5807 and the organizers' baseline at 0.4289, on a frozen public motion encoder with a single $4\times512$ linear layer. Nearly all of the margin comes from three stages usually treated as bookkeeping: reproducing the reference benchmark's exact head recipe, averaging per-walk posteriors within the subject grouping the organizers ship, and a label-free transductive calibration of the feature mean and the decision operating point. Fine-tuning the encoder lost in four distinct forms, and ten alternative encoders were worse. Every ablation number is a paid read on the hidden test, because our own leave-two-cohort-out cross-validation proved anti-correlated with the deciding score over eleven configurations. We give the negative record in full, and identify our largest gain, subject-level aggregation, as the binding ceiling on this benchmark.

Federated learning (FL) is vulnerable to multi-level attacks. However, existing methods address them separately, leaving FL exposed to data leakage, unauthorized reuse, and malicious gradient manipulation. In this work, we propose an FL framework that couples keyed context-provenance watermarking with verifiable lattice-based secure aggregation of Real-World Anchored Watermarking and Lattice-Based Zero-Knowledge Secure Aggregation. At the data layer, we propose a Kerckhoffs-compliant scheme that utilizes Physical Anchor Metadata (PAM) to ensure data provenance. PAM is defined as a context-provenance token derived from trusted infrastructure data (time, location, and server ID) and then subjected to a keyed HMAC-SHA-256 transformation to produce a watermark payload that cannot be generated without the client's secret key. We further design FMGAN, a GAN-based robust image watermarking framework that embeds this transformed payload using a feature fusion module and a Mamba-guided linear attention mechanism. At the computation layer, we adopt a lattice-based zero-knowledge secure aggregation (LZKSA) protocol that verifies key correctness, L2 norm bounds, and cosine similarity constraints over committed gradients without revealing private updates. The RLWE-based design guarantees post-quantum security. Extensive experiments validate the complementary protection of the two layers under composite attack scenarios. To our knowledge, no prior verification workflow has jointly evaluated both layers in a hybrid, end-to-end trustworthy FL framework.

Open-ended language-model benchmarks usually inherit a judge: a human preference panel, another model, or a brittle exact-match key. We introduce FlavourBench, an automated benchmark in which a versioned culinary system supplies dense, executable ground truth. Each task presents eight ingredients and asks for a three-ingredient portfolio; before model execution, Epicure scores all 56 possible portfolios. We evaluate 27 frontier endpoints on an identical 534-task core spanning substitution, pairing, and constrained composition. Every ranked model has exactly 89 valid responses per panel and family (14,418 model-task cells total), eliminating differential missingness from the leaderboard. The FlavourBench Score is the equal-family mean of the frozen task scores. We use 50,000 anchor-cluster bootstrap replicates for simultaneous 95% score bands and 100,000 sign-flip draws for all 351 paired model contrasts, with Holm control. The two independently compiled panels correlate at r = 0.89 (rank rho = 0.80). Grok 4.6 has the largest point estimate at 65.1 (simultaneous 95% CI 61.0-69.2); 101 of 351 model pairs are resolved. The release includes the prompts, all portfolio score maps, raw responses, exact routes, content hashes, and an offline verifier that reconstructs every result.

AI agents extend large language models from single prompt-response interactions to long-running, goaldirected workflows that issue many model calls, invoke tools, and interact with external environments. These workflows enable tasks such as software repair, data analysis, and experiment management, but their repeated model invocations can incur substantial carbon emissions. This paper characterizes the carbon emissions of OpenClaw agent workloads using WildClawBench, and shows that emissions depend on token consumption, context cache reuse, and the carbon intensity of the grid. Our characterization identifies deadline flexibility as an opportunity for carbon-aware execution: agent tasks can wait for lower-carbon-intensity periods or shift to lower-carbon grids. However, doing so requires handling uncertain execution time for temporal shifting and cached context recomputation during spatial shifting. We present AgentDecarbonizer, a carbon optimizer for AI agents that runs alongside OpenClaw. Given a task prompt and user-specified deadline, AgentDecarbonizer conservatively estimates task duration and selects deadline-feasible execution schedules, while accounting for cache recomputation overhead during spatial shifting. Evaluated on WildClawBench workloads with 60 agent tasks across four grids, AgentDecarbonizer reduces carbon emissions by up to 57.9 % compared with a carbon-agnostic baseline and by up to 37.5 % compared with a baseline that selects the carbon-optimal grid at task start time.

Diffusion models (DMs) have emerged as powerful generative priors for MRI reconstruction with promising results. Yet DM-based methods require extensive iterative refinement, limiting their practical deployment. Consistency models (CMs) provide a compelling alternative, aiming to map out the diffusion trajectory in a single pass, enabling faster generation. In this work, we propose CM-RED, a novel MRI reconstruction method that integrates a pretrained CM into the regularization by denoising (RED) scheme. Our method builds on accelerated proximal gradient RED (RED-APG), and further incorporates controlled noise injection during the update steps to enhance generative diversity and accelerate convergence. Extensive experiments on the fastMRI knee and brain datasets demonstrate that CM-RED achieves high-quality reconstructions across multiple anatomies, contrast weights, acceleration factors, and undersampling patterns, using only 4 network function evaluations (NFEs). The proposed method consistently outperforms existing DM- and CM-based approaches in both quantitative metrics and visual fidelity, and exhibits strong robustness to hyperparameter variations, highlighting CM-RED as an efficient and effective generative framework for accelerated MRI reconstruction. The source code and pretrained models are publicly available at https://github.com/MerveGulle/CM-RED.

Digital cameras embed device-specific artifacts into every acquired image through demosaicing, in-camera post-processing, and lossy compression. These traces constitute a forensic signal that can be exploited to assess image authenticity. Existing passive methods rely predominantly on the green channel of the Bayer residual, discarding the correlated information available in the remaining color channels and typically requiring training data or device enrollment. This work proposes a zero-shot, training-free blind image manipulation localization pipeline that estimates a reference artifact pattern directly from the noise residual of a single suspect image, without assuming a fixed filter configuration, color layout, or block period. The pipeline incorporates a principled denoiser selection criterion based on the acquired-to-interpolated noise variance ratio, a block-level correlation analysis against the estimated reference pattern, and a two-component Gaussian Mixture Model scoring stage that produces a pixel-level tampering probability map. An ablation study evaluates the impact of denoiser choice and block size on localization accuracy, and comparisons against state-of-the-art passive methods demonstrate the competitiveness of the proposed zero-shot approach.

Vision-language-action (VLA) models can follow natural-language (NL) task instructions, but such instructions may not precisely specify safety-critical or spatiotemporal requirements on the resulting behavior. We introduce Logic-VLA, a formal-requirement-aware VLA that conditions on Signal Temporal Logic (STL) specifications supplied at inference time. Logic-VLA uses a syntax-graph-based STL encoder pre-trained to capture temporal logic semantics. Policy adaptation proceeds in two stages: STL-conditioned supervised fine-tuning on satisfying demonstrations is followed by trajectory-level preference optimization over matched satisfying-violating rollout pairs using a flow-matching surrogate for Identity Preference Optimization. This formulation improves formal requirement satisfaction while preserving the nominal NL task. We evaluate Logic-VLA in closed-loop quadcopter navigation simulation across randomized photorealistic environments and test generalization to STL formulas unseen during training. Across the evaluation benchmarks, Logic-VLA improves STL satisfaction rate over an STL-blind base policy by 24.8 to 40.7 percentage points (pp) while reducing nominal NL task success by at most 1.8 pp, showing that a single VLA can adapt its behavior to varying formal requirements without requiring a separate policy for each specification.

The critical failure modes in deployed large language models (LLMs) are cross-dimensional: a model can score 99.3 in safety alignment while refusing one in three benign queries, or improve across every capability metric while losing 21 points in privacy. Existing evaluation frameworks that assess safety, security, and privacy independently cannot detect these patterns. We introduce aiXamine, a unified black-box platform that evaluates LLM trustworthiness across safety, security, and privacy as interdependent properties. aiXamine orchestrates 46 tests across nine services through an automated red-teaming pipeline, producing hierarchical risk profiles, from prompt-level diagnostics to cross-service trade-off analytics, that enable reproducible comparison of proprietary and open-weight systems under identical conditions. Applying aiXamine to over 120 LLMs through more than 5,000 test runs, we conduct the largest joint safety, security, and privacy study to date and uncover three cross-dimensional phenomena invisible to single-axis evaluation. First, safety enforcement incurs a quantifiable safety tax: stronger alignment systematically increases over-refusal, forcing providers to choose between protection and utility. Second, privacy is near-orthogonal to other trustworthiness dimensions and not captured by standard alignment. Third, we identify and formally characterize distillation-induced robustness collapse: off-policy distillation without on-policy correction causes entropy collapse, catastrophically destroying robustness (56.9$\to$2.6) on the same base architecture. These findings, compounded by diminishing returns from scale and category-dependent safety behaviors, demonstrate that trustworthiness is inherently multi-dimensional: progress along one axis does not guarantee, and can actively undermine, progress along others, yet current alignment methods treat it as a single objective.

As the demand for larger manipulation datasets grows, handheld robotic gripper data collection and the associated gripper designs become more vital. Current data collection device designs trend towards matching the morphologies of existing robotic grippers, sacrificing ergonomics and manipulation performance. In this paper, we propose a co-design framework that guides the simultaneous development of both data collection and robotic execution devices by weaving both platform constraints into the design process. Through this workflow, we present the Koala Gripper system, a data capture device and robotic gripper platform that improves dexterity and grasp capability compared to parallel jaw grippers while preserving scalability and ease-of-use. The design introduces a novel force-optimized finger/trigger linkage mechanism with directional reflected mass characteristics, a unique monolithic dual-thumb, and user-centered ergonomic design. The design's actuated robotic fingers are backdrivable, with effective mass on the order of tens of grams. We show that these grippers are capable of secure grasps over a wide range of objects, forceful tool use, and precise singulation. We further validate the platform by deploying it with an end-to-end data collection and policy execution pipeline that highlights its capabilities through learning from demonstration. More information available at http://koalagripper.rai-inst.com

Speculative decoding accelerates language-model inference by drafting future tokens that the target model verifies in parallel. A diffusion-style block head such as DFlash is an attractive drafter, predicting an entire block of future tokens in one forward pass. However, it is trained on per-position marginals rather than the joint block distribution, so the tokens it emits are individually plausible yet jointly incoherent. We introduce LiLiCorr, a Lightweight Likelihood-based model that Correlates the per-position marginal distributions a drafter already produces. It keeps the top-k tokens at each position as candidates and processes them jointly, producing for each an in and an out vector. A pair of adjacent candidates matches when the earlier one's out vector has high cosine similarity with the later one's in vector. These matches capture the block's joint structure without ever materializing the full joint distribution. One lightweight network pass produces all the vectors, and the pairwise scores are then computed in parallel as batched matrix operations, leaving only a cheap greedy walk sequential. We further co-train the drafter with LiLiCorr, so it learns to propose candidates that correlate into longer accepted sequences. Over the vanilla DFlash drafter, LiLiCorr raises acceptance length on every benchmark by 9 to 19%, while its scoring head accounts for about 2.8% of the per-block latency. Against DFlash and two concurrent methods that also restore coherence at draft time, LiLiCorr delivers the highest throughput in 70 of 72 settings: nine benchmarks at two target sizes under greedy and temperature-one decoding, and a throughput sweep over six concurrencies, two input lengths and three entropy tiers, with all systems equally optimized on a common serving stack. Extending LiLiCorr to inputs an order of magnitude longer than it was trained on preserves that lead.

Generative video compression can recover rich visual details at low bitrates, but simultaneously achieving high temporal consistency and low inference cost remains challenging. To address this issue, we propose DiffVC-ONE, a diffusion-based generative video compression framework built on a one-step Video Diffusion Transformer. First, we introduce a Unified Unidirectional Latent Compressor that uses a shared model to efficiently and uniformly compress compact latent slices. We then develop a Video DiT-based One-Step Diffusion Enhancer that uses the reconstructed latent slices as content anchors and performs single-step spatio-temporal perceptual enhancement over an entire group of pictures. Finally, a Hybrid Condition Generator extracts structural, strength, and semantic conditions from the reconstructed content and quantization information. These conditions preserve faithful regions, control the degree of generative enhancement, and supplement content-aware perceptual details during one-step diffusion enhancement. Extensive experiments on multiple standard benchmarks demonstrate that DiffVC-ONE achieves state-of-the-art perceptual quality and temporal consistency with low inference cost.

Multi-agent LLM systems are an emerging networked workload whose rapid deployment raises questions about the traffic patterns they generate. Compared to conventional applications, these systems generate requests internally: a single user task can induce a structured sequence of model calls whose timing is governed by coordination logic rather than by user arrival rate. It is not clear whether classical traffic models, designed for human-driven workloads, apply to this setting. We present an empirical characterisation of LLM-call interarrival time distributions across sequential, star, and full-mesh agentic coordination topologies, using a multi-layer measurement framework over 500 repeated runs per topology. We find that topology fundamentally shapes the arrival process of requests to the LLM backend: fan-out coordination introduces a structural bimodality absent in sequential execution, and the reasoningphase component is best described by a log-normal distribution, with the Poisson exponential null model decisively rejected across all topologies. These differences propagate to inference and network level metrics. The framework and analysis pipeline are released openly at https://github.com/dlamagna/agentraffic.

Multimodal large language models (MLLMs) have become a prevailing paradigm for unified video perception. However, post-training on large multi-task datasets remains challenging, as existing reinforcement learning methods sample on-policy groups with few high-quality rollouts even with costly chain-of-thought (CoT) generation. In this paper, we study the sample efficiency and scalability of RL post-training for video MLLMs and introduce OraRL. We identify an overlooked role for annotations: Beyond scoring rollouts, each can enter its on-policy group as an oracle rollout, a direct positive optimization target. Direct oracle integration, however, is nontrivial: a high-reward oracle raises the group baseline and inverts otherwise positive policy advantages, a failure we term advantage inversion. At the core of OraRL is a decoupled advantage estimator: policy rollouts determine an oracle-free baseline, while the oracle-policy gap modulates both a directional gain and a separate detached oracle advantage. Sign-balanced pruning improves efficiency: by retaining only the oracle and the strongest rollouts of each sign, OraRL requires just 2.2x the step time of SFT, less than half the 4.9x required by GRPO with CoT. OraRL scales with model size and data, surpassing its backbone from 0.8B to 9B and GRPO up to 100k prompts. Without chain-of-thought, Video-ORA-9B decodes in 130 ms instead of 4,780 ms. Compared with the respective prior best models, it raises temporal mIoU from 62.5 to 66.0, tracking AO from 73.0 to 78.2, segmentation from 64.3 to 70.4, and the three-benchmark spatial-intelligence macro average from 51.0 to 56.1; on VSI-Bench, it scores 73.1 against 55.0 for GPT-5 and 55.1 for Gemini-3-Pro.

AI tools increasingly support tasks across the scientific research cycle, from experiment design and manuscript preparation to peer review. At the same time, the continuing growth in conference submissions has increased the burden on meta-reviewers, who must synthesize reviewer feedback, author rebuttals, and manuscript revisions. To address this concern, this paper introduces Metag, a dataset to accelerate the development of meta-reviewing agents, specifically to identify changes made to scientific articles during the review-rebuttal process. Each instance contains a reviewer concern, the author's proposed resolution, and the manuscript diffs implementing the stated change. Metag is collected by obtaining manuscript versions from before the review deadline and after acceptance, computing differences between the two documents, and asking human annotators to align these differences with action items from OpenReview discussions. The resulting dataset consists of 349 high-quality action items tied to paper differences and will enable building methods to empower meta reviewers to quickly identify whether authors have addressed reviewer statements and where in the paper those changes have been made, resulting in additional transparency and traceability throughout peer review. The dataset is publicly available at https://github.com/microsoft/Metag-dataset.

Routine gastrointestinal endoscopy is intrinsically bidirectional: the instrument is advanced to reach target anatomy and later withdrawn or retroflexed for inspection, while an external cue may require earlier reversal. When the requested phase changes before the visual scene does, nearly identical observations can require opposite axial actions. We identify and formalize this ambiguity in bidirectional endoscopic control as intent aliasing. We propose EndoLIFT (Endoscopic Language-Instruction Flow with Trajectory Latents), a vision-language-action policy that combines explicit language-based intent conditioning with a latent-conditioned rectified-flow action expert. The policy receives RGB, a language instruction, and the previous-action state; a 32-D variational trajectory latent stochastically conditions continuous action-chunk generation. Controlled same-observation instruction swaps establish that language selects the axial mode, independently of whether the trajectory latent is present. Relative to the matched model without latent conditioning, EndoLIFT improves navigation-direction accuracy by 11.1 percentage points and reduces wrong-direction advance by 83\%. An architecture-controlled 1-bit mode-flag reference exhibits weaker canonical-anchor switching, while EndoLIFT retains 82.8\% intent-following accuracy across 44 held-out linguistic variants. In closed-loop evaluation, EndoLIFT improves overall success by 30 percentage points over EndoLIFT w/o VTL on both the seen colon phantom and the unseen lung and stomach phantoms, and completes 10/10 ex-vivo porcine-trachea trials. These results separate language-based intent selection from the trajectory latent's contribution to directional correctness and robust retraction.

Video language models process videos as dense visual-token sequences with substantial representational redundancy. Compressing these sequences is therefore essential for reducing the visual-token burden on language-model decoding. The central challenge is to preserve visual information dispersed across frames under such compression. To this end, we introduce Aggregating Visual Information with Optimal Transport (AVIOT), which casts video token compression as transporting a dense empirical measure of frame observations onto a compact target measure. The resulting source-to-target coupling induces a distribution over source observations for each target support, directly specifying how the compressed video representation is constructed. We further adapt this construction along task and spatial axes. Question conditioning modulates the transport cost between source frames and target supports, while influencing how many supports are allocated to each temporal segment, thereby directing representation capacity toward question-relevant content. At multiple spatial granularities, AVIOT computes region-specific temporal transport plans and adaptively fuses the representations they yield, allowing different regions within the same compact representation to draw from different moments. Evaluations across varying compression ratios show that AVIOT matches or outperforms the uncompressed baseline on multiple video-understanding benchmarks while retaining strong performance at higher compression ratios.

Large Language Models (LLMs) increasingly require selective removal of harmful or sensitive knowledge, called unlearning, yet existing methods and benchmarks fail to evaluate this capability completely. Current approaches rely on disjoint forget and retain sets composed of independent facts, and measure success using simple and direct factual recall. This framing fails to capture a key requirement of unlearning, namely the ability to eliminate harmful behaviors while preserving benign and beneficial knowledge. We argue that effective unlearning must operate at the level of concepts, ensuring complete removal of unsafe applications while maintaining their correct and useful usage, thereby achieving conceptually meaningful and complete unlearning. To better evaluate unlearning techniques from such a practical viewpoint, we introduce the notion of dual-use concepts: concepts that can be used in both harmful and benign contexts. Building on these concepts, we construct a benchmark called ConceptGuard where forget and retain sets are explicitly complementary in concept usage. Our benchmark uniquely enables unlearning to be explored and gauged at the level of concepts, instead of sparse facts, and evaluation is intent-sensitive with the goal of maximizing contextual separation to promote safer behavior. We demonstrate that current unlearning techniques perform poorly under this setting, showing weak contextual separation alongside poor performance in ROUGE and concept-level metrics. Our results reveal strong forgetting-utility trade-offs, limited gains in contextual sensitivity, and poor consistency in concept-level control across methods, and provide ideas for unlearning approaches that better align with real-world safety requirements. Our dataset is publicly available.

Identity-preserving image generation becomes increasingly unreliable when a scene must contain many specified people. Beyond retaining each identity, the model must bind every reference to a distinct person and location, while training-time identity losses must establish correspondence among several noisy predicted faces. We introduce WithEveryone, a unified framework for generating group images up to ten reference identities. WithEveryone injects each selected identity as an addressed token, predicts a structured identity--layout plan, and renders the plan as a visual condition. Its key objective, Layout-Grounded ID Loss, uses annotated face regions to supervise the intended identities directly, avoiding unstable embedding-based face matching; ID Representation Forcing additionally trains a prediction for each identity before image synthesis. On an identity-disjoint benchmark, WithEveryone achieves the highest target-context identity similarity, improving face similarity from 0.462 for GPT-Image-2 to 0.499, while reducing copy-paste artifacts from 0.169 to 0.055. It further covers 97.3\% of the requested identities with a duplicate rate of only 2.8\%. These results show that explicit identity--layout grounding enables identity-preserving generation to scale to larger groups without relying on direct reference-face copying.

Personalized interpretation of medical reports has emerged as an increasingly important need among patients. Addressing this need requires both evidence-grounded medical factuality and context-dependent patient communication, yet existing medical vision-language tasks do not adequately capture these dual requirements. To bridge this gap, we introduce Patient-oriented Medical Report Interpretation (PMRI), a novel open-ended multimodal generation task that requires models to explain medical reports in accurate and accessible language based on a user's query and dialogue history. These two objectives differ fundamentally in their verifiability, yet remain tightly coupled, making them difficult to optimize jointly under conventional supervised fine-tuning and holistic reinforcement learning paradigms. To address this challenge, we propose G-CARL, a grounded, checklist-aligned reinforcement learning framework that combines multi-source retrieval for atomic claim verification with context-aware, instance-specific weighted checklists for response coverage, providing structured supervision for factuality, user-demand satisfaction, and expression quality without constraining response diversity. We further construct MMedReport, a real-world PMRI benchmark, along with a clinician-designed three-dimensional evaluation protocol. Extensive experiments demonstrate that G-CARL consistently outperforms existing post-training baselines in overall quality, claim-level precision, and checklist recall. Pairwise preference evaluation by clinicians further confirms that G-CARL produces interpretations that are more accurate and better aligned with patient needs.

Travel behavior research increasingly combines digital data collection with predictive modeling, yet these stages are often developed and evaluated separately. This study proposes a three-agent workflow integrating conversational data collection, structured data processing, and behavioral prediction. A chatbot-administered, image-augmented stated-preference survey collected mode choices from student commuters across five predefined weather scenarios, yielding 454 respondent-scenario observations. Weather-related associations were analyzed using a multinomial logit model, while logistic regression and random forest provided machine-learning benchmarks. Nine locally deployed large language models (LLMs), ranging from 2 to 35 billion parameters, were evaluated across four zero-shot prompt-and-context conditions and extended through persona, few-shot, and vision-based configurations. Random forest achieved 69.6% five-class accuracy, while the best text-only zero-shot LLM reached 69.9% without task-specific fitting. Habitual travel information produced the most consistent gains, Expert framing generally outperformed Role-Play, and persona information was most useful when habitual travel information was unavailable. Few-shot prompting improved prediction for several models, with gains stabilizing after a small number of examples. Using the same weather images shown to respondents, the best vision-based configuration reached 71.5% five-class accuracy, indicating that visual context may provide additional predictive information for selected models. Overall, the study shows how conversational surveys, structured data processing, conventional behavioral modeling, machine learning, and multimodal LLM prediction can be coordinated within an auditable multi-agent workflow.

Naturalistic computer-use traces, passively recorded screenshots and mouse or keyboard actions, are a valuable resource for deriving symbolic, auditable, and reusable models of how everyday work is done. Such models matter as computer-use agents enter real work, where agents need to learn how tasks are actually performed, and organizations need to audit and reuse that knowledge. However, inducing such task models is challenging, as activity is observed only as low-level events and real-world work is multi-threaded with interleaved goals. Existing methods assume a given task or a single workflow, and produce step-level summaries rather than structured task models. We introduce Task Model Induction (TMI), which (i) discovers the latent tasks in an unconstrained trace, disentangling concurrent activity, and (ii) for each latent task, induces a task model pairing a hierarchical objective model of recursive goal decomposition with a procedure model of the control flow that organized the execution. Intrinsically, on controlled human and agent trajectories, TMI recovers interleaved tasks with 0.974 agreement against ground-truth groupings and reconstructs 74.9% of the observed execution steps, far more than the strongest workflow induction baseline. Extrinsically, skills derived from TMI's task models improve held-out task accuracy by 30.0% over the strongest baseline.

Recursive self-improvement (RSI) asks whether an AI system can improve the process that produces AI systems, so that the next system inherits the improvement. That process is the training algorithm: a better objective or update rule improves the compute\mbox{-}capability exchange rate for every subsequent run, including the one that produces the next agent. Whether RSI is feasible therefore turns on whether an agent can design training algorithms. No benchmark isolates that ability: existing suites are won by collecting data or by tuning hyperparameters, and none tells a change to how a run is executed apart from a change to how the model learns. We present AI4AI\mbox{-}Bench, 10 frozen research repositories spanning 10 training algorithm families. In each task, an agent has 4 hours on one B300 to rewrite the training algorithm; its code is then rerun from scratch for up to 12 hours and scored by a fixed evaluator hidden from the agent, against the repository's original algorithm under the same procedure. Because the 10 metrics are incommensurable, every task is mapped onto one scale on which $0$ is an uninformative model, $0.1$ is the algorithm the repository ships, and $1.0$ is the task optimum. Across 29 configurations of 6 systems on all 10 tasks the mean score is $0.166$, and the best system reaches $0.250$: even the strongest closes under a fifth of the distance between the algorithm that was already there and the optimum. The submissions show where that distance went: most never change how the model learns at all, and the minority that do average $0.226$ against $0.126$ for the rest. More reasoning effort mostly buys the willingness to go there, taking that minority from $8\%$ of submissions to $64\%$ and the mean score from $0.094$ to $0.196$. We release the task suite, the evaluators and every scored submission, so that the measurement can be repeated as these systems change.

Predictive models over structured electronic health records (EHRs) remain central to machine learning for healthcare, but few have jointly emphasized quantitative laboratory information and interpretability with respect to input medical events. We present BERT-LER, a BERT-style model for coded EHR timelines pretrained and fine-tuned from a de-identified EHR dataset of 75 million patients, that encodes laboratory test results as discrete tokens while retaining graded information through percentile-based binning, paired with Integrated Gradients for token-level attributions grounded in the input EHR sequence. We benchmark our approach on the public EHRShot benchmark suite and on an asthma severity progression study based on real-world data. This addresses a methodological gap in EHR foundation-style modeling by unifying laboratory value representation and explainability in a single framework, while assessing whether both predictive performance and explanations generalize beyond standard clinical prediction tasks. Across EHRShot and asthma tasks, BERT-LER achieves predictive performance that is competitive with, and on laboratory-related tasks often exceeds, publicly available benchmark models, and provides attributions that align with clinically known risk factors. Our architecture and explainability approach can be applied to many therapeutic areas and prediction tasks using language models trained on structured EHRs.

Transformational electric vertical take-off and landing (eVTOL) vehicles have gained significant attention over the past decade due to their efficient wing-borne cruise capabilities and reduced reliance on ground-based infrastructure. However, control system design for these vehicles remains challenging, as they must operate across multiple flight phases, each with distinct dominant dynamics. If left unaddressed, this complexity would significantly increase pilot workload, thus motivating the development of pilot control systems for multi-phase flight operations. The Simplified Vehicle Operations concept presents a promising strategy for reducing pilot workload. This study presents the design and implementation of a novel pilot control concept for eVTOL aircraft, validated through a tandem tilt-wing aircraft simulation on a full-motion simulator equipped with an active, force-feedback side stick. The system provides pilots with tactile feedback during specific flight phases, supporting intuitive control. The proposed approach enables seamless transitions and multi-phase flight maneuvers by leveraging the available degrees of freedom. Furthermore, an optimal-control-based methodology is proposed as a metric to evaluate command-filter-induced performance penalties and inceptor activities. The results show that the proposed command filter does not significantly increase the mission duration compared to the closed-loop system, while the active side stick helps reduce inceptor activity.

Whether a language model has improved itself is increasingly judged not by mean accuracy but by which individual problems it gains and loses. Tracking these transitions means differencing two noisy estimates, leaving them vulnerable to measurement artifacts. Auditing three rounds of rank-$32$ LoRA self-training on Qwen3-8B against a frozen control pushed through the identical pipeline, we identify seven measurement failures, each of which inverts a reported finding when its control is absent. Several are standard practice. A ledger built on a single greedy decode manufactures capability changes on an untrained model, largely an artifact of inference batching; the expansion statistic separating acquisition from sharpening assigns that same model a rate of $0.280$. The natural threshold repair does not survive replication: estimated across the frozen comparisons such a design already contains, its null stays non-zero. We replace it with a per-problem exact test against a pooled baseline under false-discovery-rate control, which detects nothing on any held-out replicate and is unchanged under the multiple-testing rule, error rate and pool size. Applied to a ladder of arms matched in stream, volume and evaluation, the audit finds that external distillation improves problems the base model rarely reaches while three forms of self-training do not; a regression rejects this asymmetry as a by-product of distillation's larger overall gain ($p < 10^{-8}$). On the far smaller set of problems the base model never reaches, the evidence is inconclusive, while self-training corrupts problems solved at baseline at rates well above the measured floor. Transition-level auditing therefore requires a separately measured null for every statistic it reports: nulls that cost no new experiments, built from baseline replicates a multi-arm study already owns, though not from as few as most possess.

Large language models often fail to answer questions about a bounded document collection when the source documents are not retrieved at inference time. We study this setting as document knowledge internalization: converting a fixed corpus into usable parametric knowledge for retrieval-free question answering. We propose IAR (Inject, Align, and Recover), a three-stage post-training framework that separates structured document knowledge injection, QA behavior alignment, and general ability recovery. Unlike conventional continued pretraining, Inject converts source documents into continuation, rewrite, and instruction-conditioned reconstruction objectives. Align then adapts the injected model with answer-only QA supervision, while Recover merges the domain-adapted model with the base instruction model to recover general capabilities. Across Common Corpus (CC) and CCI, and across Llama, Phi, Qwen, and SmolLM model families, IAR improves the domain-primary domain-general frontier for retrieval-free document internalization. In the main comparison, IAR improves over Vanilla SFT on all four reported metrics in 7 of 8 dataset-model settings, with average gains of 3.6 percentage points in domain QA accuracy and 12.1 percentage points in mean general performance across IFEval, MMLU, and MSBench. Extended CC baselines show that LoRA and FAPM can win individual general metrics, but among methods that also reach leading or near-leading domain internalization, IAR retains one of the strongest general profiles.

Large language model (LLM) agents can induce skills from completed tasks and reuse them later to grow more capable with experience. In practice, induced skills may transfer unreliably and can even harm the agent that retrieves them. When agent-induced skills transfer reliably across tasks remains an open question. We conduct a comprehensive and controlled study of how the way skills are induced shapes their transfer across tasks. Specifically, we compare task-level with subtask-level skill induction and text with code skill formats, the two axes along which existing methods differ. Task-level skills mostly reduce the agent's performance below its no-memory baseline while subtask-level skills raise it above on average, and text skills transfer better than code skills. To further understand our findings, we examine two complementary properties of the induced skills: specificity, which measures how closely a skill matches real tasks, and abstractness, which measures how evenly its relevance spreads across tasks. Neither property alone predicts task success, but their combined effect does, which we propose as a skill utility score. The score correlates consistently with task success when skills are transferred, and subtask-level and text skills score higher. Computing skill utility only needs the skills and task descriptions but not any task execution, so our score serves as a practical diagnostic of a skill memory before any new task runs.

Decision tree-based models are widely used in machine learning due to their interpretability and strong empirical performance. However, training decision trees can be computationally expensive, particularly for large and high-dimensional datasets, largely due to the exhaustive search over candidate splits at each node. To improve computational efficiency, we propose Data-Informed Centroid Splitting (DICS), a clustering-based framework that constructs a compact and informative set of candidate splits using data-driven priors. By incorporating class-aware structure, DICS significantly reduces the split search space for classification tasks while preserving predictive performance. We further provide theoretical analysis showing that under the stated assumptions, DICS does not degrade the performance of classification trees compared to exhaustive split search. DICS can be incorporated into classification trees, random forests, and gradient-boosting models. Extensive experiments demonstrate that DICS achieves comparable accuracy while substantially reducing training time across synthetic and benchmark datasets, highlighting the benefit of integrating data-informed priors into split selection for scalable classification tree learning.

Door opening and traversal is a long-horizon loco-manipulation task that requires precise handle interaction and coordinated base-arm control. We present Video2DoorTraversal, a single-video real-to-sim-to-real framework for wheel-legged mobile manipulators. Given one RGB video of a real door, DoorTwin reconstructs an instance-aligned, articulated, and simulation-ready door twin with realistic geometry and appearance. A simulation-in-the-loop agent converts the recovered articulation into a parameterized skill program and iteratively refines failed rollouts to generate physically executable demonstrations. These demonstrations are used to train ArticuACT, a dual-depth policy that predicts coordinated base, arm, and gripper commands using robot-centric camera conditioning and interaction-aware supervision. With all perception and policy inference running onboard, the system achieves a 96.57% average success rate across five real doors and an 80.95% zero-shot success rate on structurally similar unseen doors, while completing the full approach, opening, and traversal sequence in approximately 13s on average. Project Page: https://video2doortraversal.github.io/.

Small language models are usually built like large ones and then squeezed onto a CPU afterwards. We did the opposite: we fixed the target first, one user, one token at a time, 4-bit weights, ordinary CPU, and chose the architecture to suit it. The result keeps full attention in only 6 of its 18 blocks. The other 12 use short convolutions whose memory is two timesteps wide no matter how long the conversation gets, so two thirds of the network never re-reads a growing cache. Trained from scratch on 59.9B tokens, the model scores 47.31 on a five-task benchmark against a bar of 42.20 that was fixed before training began. It beats GPT-2 124M, Pythia-160M, OPT-125M and GPT-neo-125M, all trained on three to six times more data, and exceeds MobileLLM-125M's published score despite that model seeing a trillion tokens. Validation bits-per-byte is 0.8685. To check the architecture rather than the training recipe, we trained a conventional all-attention model of the same size on the same data, and wrote down the winning condition before scoring either. The hybrid won the chosen quality metric by 0.81%, matched it on downstream tasks, produced a 6.3% smaller 4-bit file, and decoded 1.76x faster at 2048 tokens of context, 2.08x against an external model of similar size. In every measurement the speed advantage is near zero at an empty context and grows with length, which is what the mechanism predicts and what a merely leaner model would not show. A simple bandwidth calculation predicts only 1.17x, so memory volume alone does not explain the gap. We also report what did not work: an unmitigated 4-bit quality cost, roughly half the convolution channels ending up inert and impossible to remove, and a vocabulary larger than this model size warrants.

Legal work, with its heavy reliance on processing large amounts of text, is often considered one of the domains most exposed to the use of LLMs. Contract ``scrubbing,'' the final review of transactional agreements for errors and inconsistencies, is a particularly suitable task for automation, because it is routine, painstaking work requiring detailed attention to long documents. Scrubbing also seems to align naturally with the general capabilities expected of frontier LLMs around long-context reasoning, consistency checking, and named entity recognition (NER). Despite the economic value and potential for automation, no formal evaluations of LLMs performing contract scrubbing have been conducted. We introduce ContractScrub, the first benchmark designed to evaluate contract scrubbing capabilities, comprising contracts hand-crafted by experienced lawyers over diverse error categories such as misuse of defined terms, incorrect references, and inconsistent language. Frontier models perform surprisingly poorly with only one model reaching 0.75 macro average recall despite strong performance on seemingly related general benchmarks, demonstrating the practical limits of current models and the importance of narrowly targeted, domain-specific benchmarks for measuring real-world impact.

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. We identify memory-induced cognitive traps: even faithfully recorded and semantically relevant memories can distort model reasoning or beliefs and degrade current task performance. To systematically evaluate these failure modes, we introduce MemTrapBench, which covers two forms of cognitive traps: Reasoning Fixation and Belief Distortion. Experiments across two model families and five representative memory frameworks show that MemTrapBench is challenging: all evaluated memory strategies underperform the no-memory setting, with even the strongest methods suffering drops of more than 10%. To mitigate these cognitive traps, we propose AdaptiveMem, a simple yet effective inference-time method that instructs LLMs to avoid memory traps. AdaptiveMem mitigates cognitive traps on MemTrapBench while preserving or improving performance on standard memory benchmarks across diverse memory frameworks.

Nano- or micro-particle rotation driven by light has been well known in the fields of optical manipulation and optical physics since the end of the last century. It is viewed as a sort of angular-momentum transfer from light to material, but its microscopic analysis based on the Hamiltonian or the equation of motion has been less developed. We model this rotation with a simple setup of an electrically dipolar or multipolar particle irradiated by circularly polarized laser and comprehensively analyze the Langevin-type equation of motion by using the Floquet theory for dissipative classical systems and the mode separation method. Furthermore, we numerically compute the time evolution of the particle. As a result, we accurately estimate the dependence of the laser-frequency, laser-intensity, particle mass, temperature, and friction (dissipation) on the laser-driven rotation. We determine the ``nonequilibrium phase diagram'' of the laser-driven rotation in a broad parameter regime, which consists of three regimes: the rotation frequency $Ω\proptoω^{-1}$, $Ω\proptoω^{-3}$, or $Ω=ω$ ($ω$ is the laser frequency). Comparing our theoretical result with some experiments, we show that the result of the overdamped Langevin equation is qualitatively consistent with the experiments.

Bilateral teleoperation requires stability despite uncertain master and slave dynamics and delayed communication channels. Existing radial basis function neural network (RBFNN) controllers mainly differ in uncertainty decomposition, while online adaptive parameters often increase with network size. This paper proposes a compact two-sided adaptive controller for a nonlinear bilateral teleoperator with constant forward and backward delays. Operator and environment impedances are incorporated into the manipulator dynamics, and each side uses only two scalar adaptive estimates: one for the ideal RBF weight norm and another for the combined effects of friction, approximation error, and disturbances. Both estimates are updated through sigma-modification, resulting in only two adaptive parameters regardless of the number of RBF nodes. A Lyapunov-Krasovskii functional integrating sliding-variable energy, estimation errors, and delay-dependent integral terms is developed. Free-weighting matrices based on sliding-surface identities provide delay-dependent matrix conditions that guarantee uniform ultimate boundedness of synchronization errors, sliding variables, and adaptive estimates. Simulations on two 2-degree-of-freedom (2-DOF) revolute manipulators with friction, external disturbance, and stepwise operator inputs demonstrate synchronization within approximately 2 s and reveal the trade-off between accuracy and control chattering.

We present a novel approach to efficient LLM agent harness optimization through adaptive validation task selection. Harness optimization iteratively rewrites the harness code based on validation performance, enabling substantial performance gains without updating the underlying model weights. Existing approaches, however, evaluate a fixed validation set in full at every iteration, incurring substantial evaluation costs even on tasks that become less discriminative as the harness evolves. We propose $\textbf{Task-CoEvolve}$, which co-evolves the validation tasks with the harness by addressing two challenges: selecting informative tasks and estimating full-set performance from partial evaluations. Task-CoEvolve builds on the observation that tasks on which candidate harnesses disagree are more informative for distinguishing among them than tasks that are consistently solved or failed. It uses variance-weighted sampling based on past outcomes to focus evaluation on tasks near the agent's capability frontier, with the sampling distribution adapting as the harness evolves. It then estimates full-set scores from the sampled tasks by accounting for their sampling probabilities, enabling consistent comparisons across iterations despite evaluating different subsets. Experiments on online text classification and Terminal-Bench 2.1 show that Task-CoEvolve consistently outperforms fixed-subset baselines and matches the final performance of full-set search while reducing the number of evaluations during optimization by 80%. Code will be released at https://github.com/Agent4Science-UTokyo/Task-CoEvolve.

Large language models (LLMs) have shown growing potential for automated theoretical computer science (TCS) research, yet existing benchmarks remain far from realistic research settings. We introduce \ourbenchmark, an expert-validated benchmark for evaluating LLMs on frontier, end-to-end TCS research. \ourbenchmark contains $175$ instances drawn from papers accepted to STOC, FOCS, SODA, and COLT in 2025-2026, preserving paper-specific definitions, assumptions, and proof dependencies, with expert-verified Lean formalizations and proofs. Evaluations of leading LLMs reveal that current models remain far from reliably completing the full research pipeline. In particular, autoformalization is the sharpest bottleneck: the best model achieves only $11.5$ on translating natural-language claims into formal theorem statements, compared with $28.6$ Pass@8 when proving human-provided formal statements. Building on \ourbenchmark, we further develop an automated TCS research framework that generates, formalizes, filters, and proves new claims. Of $64$ generated claims, only $6$ ultimately pass expert evaluation and proof verification, indicating that beyond formalization, limited research taste remains another major barrier to autonomous TCS research.

Autonomous vehicles require robust perception and decision-making capabilities to operate in diverse and unseen scenarios. While reinforcement learning and rule-based methods can provide effective control and safety mechanisms, their performance may degrade in situations requiring contextual reasoning. Large Language Models (LLMs) have demonstrated strong capabilities in understanding multimodal information and generating contextual reasoning, however, their use for direct vehicle control can introduce latency and hallucination risks. To address these limitations, a hybrid framework is proposed. This system uses an orchestrator to coordinate PPO-trained reinforcement learning and PID control, with LLM common-sense reasoning applied throughout the framework. LLM reasoning is further employed iteratively to refine the RL reward function for dynamic driving environments. The proposed framework is evaluated in highly randomized CARLA scenarios under diverse environmental and traffic conditions. The results demonstrate the potential of integrating LLM-based reasoning with conventional autonomous driving methods while retaining structured control and safety mechanism.

Few-mode photonic circuits can increase functionality without multiplying waveguide paths, but bends and fabrication errors can mix their transverse modes. We investigate a strategy in which waveguide confinement and perturbation parity are engineered together in vertically elliptical, femtosecond-laser-written glass waveguides. The intended modal basis comprises the even $1S$ mode and the vertically odd $2P_y$ mode. No window-converged $2P_x$ state is resolved for a lower-confinement (LC) design, whereas a higher-confinement (HC) design guides $2P_x$, which must therefore be isolated by symmetry. In scalar beam-propagation calculations, the $1S$-$2P_y$ propagation-constant splitting predicts the optimized periods of a vertically modulated coherent modal splitter to within $1.0\%$. Horizontal S-bends remain parity-mismatched for $1S \leftrightarrow 2P_y$ coupling, while the symmetry-allowed HC $1S \rightarrow 2P_x$ transfer reaches only $0.6\%$ at the largest displacement. At a displacement of $150~μ\mathrm{m}$, the HC design retains approximately the same $2P_y$ power as the LC design retains at $40~μ\mathrm{m}$. Thermal and stochastic writing-error calculations reveal the resulting trade-off: stronger confinement improves modal-power retention, but writing jitter that breaks $x$-parity can populate the guided $2P_x$ mode. These results demonstrate how modal-basis engineering can shift part of the crosstalk-control burden from the trajectory to waveguide symmetry, supporting joint path--mode degrees of freedom in quantum photonic applications.

Large language models (LLMs) are increasingly used in settings where textual summaries, numerical observations, and external tool outputs may provide conflicting evidence. We study how LLMs arbitrate between such sources when they support opposing decisions. To do so, we introduce a controlled synthetic benchmark in which latent risk trajectories generate both numerical time series and natural language summaries, allowing us to construct conflicts where exactly one evidence source is aligned with the ground-truth label. This design lets us independently manipulate modality, temporal recency, source reliability, and evidence provenance. Across open-weight instruction-tuned models, we find that arbitration behaviour is systematic rather than random: models exhibit distinct text-versus-number preferences, follow temporal recency more consistently than explicit reliability cues, and can over-rely on external forecasts even when they conflict with direct contextual evidence. These results suggest that current LLMs often rely on heuristic arbitration strategies when integrating heterogeneous evidence, highlighting a failure mode for tool-augmented decision systems.

Mobile manipulation requires a robot to predict how locomotion and arm motion jointly alter future observations and control. Existing world-action models, developed largely for fixed-base platforms, do not explicitly distinguish camera ego-motion from base and arm actions. Here we introduce DECOWAM, a whole-body world-action model that separates these factors through dedicated conditional interfaces. DECOWAM freezes an adapted FastWAM backbone and trains residual adapters, an action-equivalent future bottleneck distilled from privileged observations, adversarially separated base and arm latents, and base-velocity conditioning for video prediction. We further introduce ARMDOG, a real-robot dataset that synchronizes video, whole-body state and action, and language. On a fixed replay protocol, DECOWAM improved both future-video and action prediction over FastWAM, reducing action MSE by 21.7% with 25.95M trainable adaptation parameters. Across 79 closed-loop trials per method, it achieved the highest observed whole-body coordination and base-displacement robustness among the compared systems, while task completion remained comparable to the strongest baseline. These results show that embodiment-aware factorization can support parameter-efficient joint visual prediction and whole-body control under moving viewpoints.

End-to-end autonomous driving has evolved from camera-to-control regression toward planning-oriented systems that use structured representations, trajectory-level outputs, and increasingly realistic evaluation protocols. This survey reviews this transition across behavior cloning, conditional imitation learning, privileged distillation, BEV and vectorized planning, unified perception-prediction-planning architectures, world-model-based planners, and vision-language-action systems. We argue that the key distinction in modern end-to-end driving is not whether intermediate representations are used, but whether they are learned, supervised, and evaluated to support safe, feasible, and route-compliant planning. To organize the literature, we synthesize existing methods along four axes: input representation, planning output, supervision signal, and evaluation protocol. We further examine the benchmark shift from open-loop trajectory matching to closed-loop simulation, non-reactive real-log evaluation, long-tail testing, and human-preference-aware metrics. Our analysis highlights that architectural progress is difficult to interpret without benchmark-consistent evaluation, and that displacement-based open-loop metrics alone provide limited evidence for safe and human-aligned driving. We conclude with open challenges in uncertainty-aware planning, learner-expert mismatch, runtime safety assurance, language-action grounding, world-model validation, and reproducible benchmarking.

We introduce OenoBench, a wine-domain knowledge benchmark of 3,266 multiple-choice questions across six pillars (regions, grape varieties, viticulture, winemaking, producers, business) and four difficulty tiers. The corpus is built from 38,104 atomic, source-anchored facts extracted by 35 provenance-verified scrapers from government registries (INAO, TTB, OIV), peer-reviewed journals, and Wikipedia/Wikidata. Our methodological contribution is an LLM-driven pipeline in which language models reformat verified facts and audit the result, but never serve as the source of truth: every claim traces to a URL, every question is generated by one of five strategies across five generator families, and every question is scored by a nine-agent audit calibrated against a human gold sheet via Cohen's $κ$. Evaluating sixteen frontier configurations, we find: (i) overall accuracy spans 53%-84%, led by o3 at 83.6%; (ii) reasoning-mode lift concentrates in DeepSeek R1 (+6.8pp) and is absent in Claude Opus and Gemini Pro; (iii) Anthropic shows +9pp self preference on its own questions while Google shows -8pp inverse preference; (iv) frontier open-weight models share the cost-vs-accuracy Pareto frontier with proprietary reasoning models; and (v) every config gains around 33pp on closed-book solvable items, revealing a parametric-recall ceiling that only the contextual slice avoids. We release corpus, audit findings, human-review app, and construction code under CC-BY-SA-4.0.

LLM-based Multi-Agent Systems (MAS) achieve strong performance on complex reasoning tasks by coordinating multiple agents, but at the cost of substantial token consumption. Recent work on automatic topology design, ARG-Designer, has reframed this problem as autoregressive graph generation. However, its training objective provides no explicit incentive for the model to generate sparse and efficient topologies. We address this limitation by introducing a Reward-Guided Autoregressive Graph Generation (RGA-Designer) inspired by Reinforcement Learning from Human Feedback (RLHF). We train a reward model that jointly captures task correctness and structural compactness, and then fine-tune the pretrained graph generator using the reward model as feedback. Our method preserves task accuracy at the level of ARG-Designer while reducing token consumption by an average of 20.5%.

Mid-infrared (MIR) light sources compatible with silicon photonics are highly desirable for environmental sensing and free-space optical communications. Conventional GaSb-based quantum-well and interband-cascade lasers, however, rely on complex epitaxial heterostructures, complicating their integration with silicon photonic platforms. Here, we demonstrate a room-temperature MIR surface-emitting distributed-feedback (DFB) laser using black phosphorus (b-P) as the active gain medium. By deterministically integrating exfoliated b-P flakes onto lithographically patterned SiO$_2$ gratings, we obtain narrowband emission with a full width at half maximum below 3 nm under 1064 nm optical excitation. The lasing wavelength is tunable from 3.79 to 4.05$~μ\text{m}$ through the b-P flake thickness, covering a technologically important region of the mid-infrared. Room-temperature thresholds as low as $(0.25 \pm 0.07)\ \text{mJ/cm}^2$ are achieved under nanosecond excitation. Upon cooling to 110 K, the threshold decreases tenfold to $(0.015 \pm 0.005)\ \text{mJ/cm}^2$. These results establish planar b-P DFB cavities as a promising platform for heterogeneously integrated MIR lasers based on van der Waals semiconductors.

Humanoid robots have recently demonstrated promising capabilities in real-world ball sports. However, achieving professional motion styles while maintaining strong task performance remains challenging. In this work, we propose AdaPT, an Adaptive Motion Planning and Tracking framework that learns professional tennis serving and rally styles directly from broadcast videos. This hierarchical design is motivated by the key insight that the planner generates stylistic kinematic motions, while the tracker executes them with minimal interference with planning. Despite its effectiveness in simulation, a substantial sim-to-real gap emerges: tracking performance inevitably degrades on real robots, and this degradation is partially overlooked by autoregressive planning and further compounded by noisy perception. To address these issues, our adaptation mechanism improves tracking robustness by learning to track randomized execution speeds, while conditioning the planner on a learned motion-speed adapter to mitigate compounding errors. Real-world experiments on the Unitree G1 demonstrate the effectiveness of our adaptation mechanism in bridging the sim-to-real gap. We further deploy AdaPT policies on the full-size Dobot Atom humanoid robot (1.7m) and demonstrate in-the-wild serving without motion capture. Beyond these results, our real-world experiments reveal both algorithmic and engineering insights for future humanoid ball-sports systems. Videos and code are available on our \href{https://humanoidtennis.github.io/AdaPT/}{project website}.

Robots executing long-horizon manipulation tasks from natural-language instructions must reason about both semantic task structure and geometric feasibility. However, under partial observability, the availability of goal-relevant objects may be uncertain. In such cases, approaches that combine Vision-Language Models (VLMs) with Task and Motion Planning (TAMP) may generate subgoals that rely on the VLM's prior knowledge without observational support, leading to execution failures or unintended outcomes. We propose Evidence Acquisition and Feasibility Gating (EAFG), a framework that acquires visual evidence through VLM-generated exploratory subgoals and TAMP-based execution. EAFG then applies a feasibility gate to decide whether to proceed with task planning, acquire further evidence, or halt. Our experiments show that, in cooking tasks with ambiguous object use, EAFG improves recipe completion by discovering task-relevant objects before planning. For instructions requiring an absent object, EAFG promotes appropriate halt decisions and reduces repeated attempts to manipulate that object.

Question Answering over Temporal Knowledge Graphs (TKGQA) requires reasoning over time-sensitive facts, yet existing embedding-based methods struggle with multi-step queries due to single-pass reasoning pipelines. We propose SABET-QA, a framework that iteratively refines reasoning states across multiple hops via a bidirectional entity-temporal scoring mechanism and a slot-aware contextualization module that aligns question semantics with temporal KG embeddings. A differentiable working memory enables progressive hypothesis refinement, while auxiliary temporal boundaries serve as coarse supervision when available. Experiments on CronQuestions, Complex-CronQuestions, MultiTQ, and TimeQuestions demonstrate consistent improvements over strong baselines, particularly on complex multi-step temporal queries.

Watermarking schemes for large language model output are evaluated almost exclusively on English text using each scheme's detection threshold and a narrow set of quality measurements. Multilingual deployment exposes evaluation-design choices that are inconsequential on English but determine conclusions cross-lingually. We propose an evaluation framework with four components: detection thresholds calibrated empirically per deployment context, a threshold-independent companion measurement that distinguishes calibration failures from detection failures, three disjoint quality measurement paradigms (distributional, paired-semantic, and reference-perplexity), and a generalized-entropy decomposition of cross-language disparity over a typological family partition. Applied to six watermarking schemes, three open-weight generators, eleven languages spanning four scripts and eight typological families, and both base and instruction-tuned regimes, the framework reveals failure modes that single-language single-paradigm evaluation cannot surface. Across detection and quality, observed disparity is predominantly between-family on the typological partition, indicating that cross-lingual fairness gaps in watermarking are structural to language properties rather than idiosyncratic to particular languages.

We study bilateral teleoperation between nonlinear, multi-DOF robotic manipulators in the presence of constant communication delays. Unlike classical wave-transformation architectures that transmit a coordinating force, we consider the case where the environmental force is reflected to the master side to enhance teleoperation transparency. Since direct contact force feedback might destabilize the closed-loop system, we first develop a passivity-shortage characterization for the Euler--Lagrange remote system using a linear matrix inequality (LMI) approach. An upper strictly passive communication law is then employed to compensate for the computed passivity shortage so that the closed-loop stability under delays as well as position and force synchronization are preserved under appropriate conditions. Simulations with nonlinear 2-DOF robotic manipulators in different settings illustrate our approach.

We present HealMed, an expert-reviewed benchmark for multilingual evaluation of large language models in medicine. HealMed contains 1,000 examples in each of nine languages, drawn from nine datasets and covering three task formats: MCQA, NLI and open-ended QA. The benchmark was developed over two years by 23 physicians and medical experts based across nine countries and regions. Each translation was evaluated and revised by two experts fluent in English and the corresponding target language. On HealMed, performance declined most in low-resource languages, although the size of the gap varied markedly across languages and models. The strongest proprietary models were the most stable across languages, whereas many open-source and medically specialized models showed larger and less consistent gaps. Medical specialization alone did not ensure multilingual robustness. Furthermore, expert revision could either raise or lower measured performance, indicating that translation quality materially affects cross-language evaluation results.

Although legged animals are capable of performing explosive motions while traversing confined spaces, replicating this behavior in quadrupedal robots has been a longstanding challenge. Here, we propose a hierarchical reinforcement learning pipeline that empowers the robots to perform aggressive locomotion through constrained obstacles--a narrow gate. The imitation learning technique is used to train the low-level policy, which mimics the behaviors of real animals and forms a set of diverse skills. The high-level controller, having an awareness of the capability of low-level skills and acquiring the gate information via vision-based detection, determines the suitable maneuvers with collision-free trajectories to traverse it dynamically. Notably, we also verify that this framework can be extended to other highly dynamic tasks. This is one of the first works that perform autonomous and agile aerial gate traversal tasks on ground-walking robots, extending the lifelike agility of legged robots to match that of their biological counterparts.

Modern computer vision has enabled partial autonomy in robotic assembly manipulation. However, performing autonomous manipulation of a progressive assembly demands a more specific set of skills, in addition to perceiving the objects. Through a comparative analysis of research in the associated domains, we deduce that object-centric perception must advance towards learning assembly dependencies to predict meaningful actionable outputs for autonomous assembly manipulation. Subsequently, we present a 3D keypoint-based modular learning framework to learn assembly dependencies to infer actionable outputs given a RGB-D input of an assembly scene. We train and evaluate our trained network on an assembly pose estimation dataset and compare it against object-centric baselines with an augmented set of metrics for progressive assemblies.

Natural language code retrieval is a rapidly evolving task in computer science. However, the 1C:Enterprise ecosystem combines Russian syntax with highly domain-specific terminology, for which open datasets and specialized models have been virtually non-existent. We present a comprehensive pipeline for 1C code retrieval: an open benchmark of 3,413 real-world, PII-scrubbed query-code pairs, a reproducible evaluation harness, and a specialized bi-encoder. To overcome scarce labeled data, we fine-tune on 784,057 synthetic triplets generated by google/gemma-4-26B-A4B-it from public code repositories, using Matryoshka Representation Learning (MRL) and a privacy-aware tokenizer. Because the benchmark subsets differ in size, we report balanced-subset macro, query-weighted micro, and forum-only results. Our model reaches 0.5992 balanced macro nDCG@10, 0.5044 micro, and 0.4617 on forum, versus 0.4932 macro for the baseline architecture and 0.5404 for google/embeddinggemma-300m. Removing every benchmark example flagged by the conservative exact/13-gram overlap audit leaves 0.6011 balanced macro (0.5010 micro), indicating that detected train-benchmark overlap does not explain the headline result. MRL truncation to 256 dimensions preserves 99.9% of retrieval quality while reducing dense-index storage and exact similarity arithmetic by a factor of three.

Population-level behavior in large-language-model (LLM) agents cannot be characterized by single-agent benchmarks. We introduce PV-SST, a peer-voted social-platform testbed, and report a separately frozen, preregistered matched-exposure experiment spanning four topics, four unused seeds, four open-weight model families, and three prespecified larger variants. The experiment comprises 448 trials and 112 complete model-by-topic-by-seed blocks. Relative to a topic-only control, a feed of previous-round peer posts ranked by peer-generated likes increases final-round lexical similarity in both the four-family core panel (paired mean difference +0.0082 TF-IDF cosine units, 95% block-bootstrap CI [0.0043, 0.0121], randomization p=0.000105, n=64 blocks) and the three-variant size extension (+0.0109 [0.0069, 0.0151], p=0.000001, n=48). This contrast bundles peer-post exposure with ranking and therefore does not identify a ranking-only effect. Opposite-side survival falls in the core panel (-3.9 percentage points [-6.8, -1.6], p=0.0068) but not conclusively in the larger variants (-1.0 pp [-3.1, 0.4], p=0.50). Holding adversarial impressions fixed, four distributed sources do not reliably move honest-agent stance more than one source. The preregistered distributed-minus-single contrast is positive but inconclusive in the core panel (+0.057 [-0.009, 0.125], p=0.112) and negative in the larger variants (-0.040 [-0.113, 0.035], p=0.332), failing the prespecified cross-model and cross-topic consistency criterion. Thus the robust result is lexical convergence under the tested peer-ranked feed, not general opinion capture or a general coordination advantage. The study evaluates synthetic LLM-agent populations; it does not estimate effects on people or production platforms.

A lot of prior work addressed key-value (KV) cache selection and compression by sparse attention to enable long-context inference for transformer language models without excessive hardware budgets. We provide a new method for fine-tuning models with sparse attention. It works for any KV cache policy, runs on a moderate hardware budget (e.g., a single Nvidia A100 GPU with 40 GB RAM), and allows the model to co-adapt with the policy, often outperforming models trained with exact attention (sequence parallelism). We also provide an efficient implementation of H2O sparse attention (the leading policy in our experiments) with dedicated scaled dot product attention kernel support. KeysAndValues (https://github.com/awslabs/keys_values), a new open source library for long-context inference and fine-tuning, provides easy-to-use and performant code for all methods discussed here.

This paper presents a Bayesian control framework that integrates spike-based dynamics with probabilistic inference for adaptive control. Bayesian inference is widely regarded as a core computational principle of brain function, providing a normative framework for perception, decision-making, and learning under uncertainty. By combining a biologically inspired spiking neural model with Bayesian inference principles, we propose a brain-like control algorithm capable of operating in uncertain environments. We use the mountain car parking problem as a benchmark with non-linear dynamics. Our results demonstrate that the proposed controller can successfully update states in real time and generate goal-directed action plans through spike-driven dynamics. The results highlight the proposed model's potential as a bridge between computational neuroscience and probabilistic control theory.

Where does the novelty a base language model produces with no task come from, and what can an LLM judge of a long stream actually see? We dismantle a cognitively inspired generation loop over 24 conditions on three base models. Most of its effect lives in one operation: a new subject injected every few hundred tokens (an interruption) into a stream whose literal repetition is damped (habituation). We judge windows of generated text only, with the premise as the unit (n=10) and a judge measured for repeatability, against a second judge family and against human readers. Under that protocol the interruption raises judged surprise by 1.2 to 1.4 points and connection by 0.8 over habituation alone. A connective that asks for continuity hurts; a bare paragraph break adds nothing detectable on fresh text; a reset context does at least as well as a kept one; and a pre-registered replication on new premises confirms the primary contrast. Three things the window judge could not see changed the first version of this study, and we think they are of general use. The judge scores the experimenter's injected sentence as the model's own. A fixed rotation of injected sentences makes the model replay its earlier segments from beyond the judge's horizon, and the judge scores the replay as surprise and connection (65-80% of post-interruption windows at periods 150-300). And the local gains do not compose: no arm produces an integrated document. The salience monitor, the in-loop judge, memory across interruptions and a judge-gated Review run with a gate that opens add nothing. On a problem with a verifier (online bin packing), the interruption multiplies valid, distinct candidate heuristics three- to fourfold without raising the quality of the best. We report an evaluation protocol for long generation and a controlled characterization of a simple intervention, not a mechanism of creativity.

LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent's weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive or unreliable verifiers, and still produce static environments. To alleviate the engineering burden of rebuilding environments from scratch, we propose Environment Harness (EnvHarness), a programmable layer of plug-in components that wraps a static environment to reshape its behavior without modifying the underlying logic. Operating through standard interfaces, EnvHarness applies across diverse domains while ensuring every reshaped environment retains its original verifier. To automate this process, we introduce EnvRigger, which treats the target policy as a black box, observing its execution trajectories to synthesize EnvHarness components targeting diagnosed flaws, and validating them via fresh rollouts. Across five benchmarks in four domains, EnvHarness outperforms both original environments and domain-specific environment generation pipelines, achieving up to a 9.0-point improvement on held-out instances with 9.8% fewer execution steps. Furthermore, EnvHarness provides a superior optimization signal for reinforcement learning, enabling continuous, targeted co-evolution of the policy and its environment.

Patients often submit short, underspecified queries to healthcare chatbots that lack the patient-specific information needed to determine an appropriate response. Although these queries may be linguistically clear, they can support multiple plausible answers depending on undisclosed factors such as symptoms, diagnoses, medications, allergies, or dietary restrictions. A language model answering such a query directly may therefore rely on unsupported assumptions about the patient. We introduce a knowledge-guided agentic framework for mitigating patient-context ambiguity before final response generation. The framework operates between the patient and an otherwise unchanged downstream language model. It interprets the initial query, uses a task-specific knowledge graph to construct a set of plausible hypotheses, identifies the missing patient-context variables needed to distinguish among them, and asks targeted follow-up questions. The original query and the acquired context are then combined into a clarified prompt for the downstream model. We evaluated the framework across five language models using two controlled ambiguity-mitigation benchmarks: diagnosis retrieval from 1,034 symptom queries with clinically relevant evidence systematically masked, and dietary-safety classification from 487 queries with decisive health context omitted. The framework was compared with direct answering of the underspecified query and with rephrasing the same query without acquiring new patient information. In diagnosis retrieval, it increased overall exact Top-1 accuracy by at least 57.1 percentage points and selective exact Recall@5 by at least 77.7 percentage points across the five evaluated models compared with direct prompting. In dietary-safety classification, it improved accuracy across all five models and achieved the highest Matthews correlation coefficient for four...

Customer-service LLM agents must follow organizational policy when acting on a user's behalf. Compliance failures arise from either forbidden actions, such as granting an ineligible change, or omitted procedural requirements, such as identification or confirmation. Runtime safeguards can intervene on risky actions, but action-local checks do not guide an agent through a multi-step procedure. Workflow-following systems support prescribed process execution, but primarily target workflow completion rather than safeguarding agent behavior. PolicyGuide instead compiles each domain policy into a workflow graph and invokes a proactive verifier at user-turn boundaries. From persisted graph state, the verifier reconciles open requests and returns step-specific remediation along a policy-compliant path. Across the $τ^2$-bench airline, retail, and telecom domains with a GPT-5.4 agent and verifier, PolicyGuide raises mean $\mathrm{Pass}^4$ from $0.42$ to $0.62$, with the largest gain on telecom ($0.19$ to $0.61$), the most workflow-structured domain. The same workflows transfer to Claude Sonnet 4.6 and Gemini 2.5 Pro agents. Complementary evaluations find the lowest observed attack-success rate under adversarial users and the strongest procedural compliance in an author-designed workflow-level validation.

Automation operates safety-critical processes inside their design envelope and leaves abnormal situations to human operators. Mismanagement of these situations is a leading contributor to process-safety incidents and a hindrance to achieving autonomy. Here we show that a general-purpose large reasoning model, with no task-specific training and only the information available to an operator, manages abnormal situations at run time through a bounded, programmatically verified action interface. Across 39 abnormal situations and operating-point changes on a plant-wide industrial benchmark process, the reasoning model maintained the plant within all hard constraints in all 39, while basic regulatory control failed in 15. It matched the plant's expert-engineered advanced control and diagnosed the root-cause fault in 15 of 15 safety-critical situations. Three independently developed models spanning a thirty-fold cost range exceeded the baseline. In a fully auditable evaluation, these results demonstrate run-time abnormal situation management without a human in the loop.

Credit assignment is challenging in long-horizon agentic reinforcement learning, where supervision often comes only from final rewards. Existing methods refine trajectory-level signals into step-level credits through step grouping or graph-based advantage estimation, but can overlook meaningful intermediate milestones. We propose MileGPO (Milestone Inference with Local Evidence for Graph-Based Policy Optimization), which derives process-level credit from grouped on-policy rollouts through three designs. Milestone Discovery identifies candidate milestones on successful rollouts and recurring traps on failed ones. Reliability-Calibrated Shaping (RCS) weights these candidates by outcome-based confidence, strengthening reliable milestones and traps while down-weighting uncertain ones. Progress-Contrastive Calibration (PCC) further tests whether a candidate reflects local progress and whether its incoming ansition outperforms observed alternatives from the same state.MileGPO requires neither auxiliary models nor additional environment interaction. Experiments on ALFWorld and WebShop show state-of-the-art performance and a small in-distribution to out-of-distribution gap on ALFWorld. Ablations and credit diagnostics indicate that reliability weighting, local progress, and same-state branch evidence complement milestone discovery and resolve ambiguous intermediate credit.

Low-Rank Adaptation (LoRA) is a prominent fine-tuning method for large models, achieving competitive performance with reduced memory overhead. However, a persistent performance gap remains between LoRA and full fine-tuning. Recent studies have sought to narrow this gap by employing one-step gradient approximations of pretrained weights to align LoRA updates with the principal directions or intrinsic dimensionalities of full fine-tuning updates. Nevertheless, these approaches fail to capture the full dynamics of the gradients. In this paper, we propose LoRA-GA$^2$, an effective fine-tuning algorithm that fully leverages multi-step gradient information. Specifically, we introduce a lightweight probe for multi-step gradients of pretrained weights that incurs no additional GPU memory cost and only marginal time overhead. We further employ a spectrum-aware, importance-based rank allocation and optimal initialization derived from multi-step gradients. Extensive experimental results demonstrate that LoRA-GA$^2$ consistently outperforms existing LoRA variants while preserving the efficiency advantages of vanilla LoRA. For instance, LoRA-GA$^2$ surpasses the leading baseline by an average of 0.66 points on the GLUE benchmark, and outperforms the strongest baseline by 1.03 points on GSM8K and 0.87 points on HumanEval, respectively.

Software increasingly functions as part of the scientific instrument itself, making failures in scientific code capable of compromising not only program behavior but also the evidence underlying scientific conclusions. Yet existing evaluations of coding agents largely emphasize aggregate task success, providing limited insight into why agents fail when repairing scientific software. We introduce \textbf{SWE-bench Science}, a repository-level benchmark for scientific software engineering comprising 119 tasks from 98 GitHub repositories across 20 scientific domains. Each task is organized into one of three paradigms: Issue-driven, Expert-exploratory, and Engineering-integration. Even the best-performing agent, \textbf{Claude Code with Opus-5 (max), achieves a pass@1 below 50\%}, highlighting the substantial challenges posed by scientific software engineering. We identify four recurring failure mechanisms: deficits in scientific knowledge or abstraction, misguided exploration or surface-level repair, incomplete repair coverage or system integration, and failures to generalize scientific knowledge beyond observed cases in our analysis. We further conduct a paired ablation that removes explicit scientific guidance while preserving the repository and executable engineering context. The results show that scientific knowledge is not uniformly beneficial: well-grounded information can constrain repair and improve average performance and token efficiency, whereas poorly aligned guidance can induce anchoring and does not necessarily improve exact repair success. Together, SWE-bench Science provides a broad testbed for studying both the capabilities and failure mechanisms of coding agents in scientific software engineering.

The convergence of large language models (LLMs), structured knowledge bases (KBs), and reasoning ability (RA) presents a promising trajectory toward general embodied intelligence (GEI). This paper reviews the evolution of LLM-centered intelligent systems, emphasising their integration with knowledge representation, logical reasoning, and physical embodiment. We analyse LLM architectures, pre-training methods, and inference mechanisms, along with their interaction with external knowledge sources and structured reasoning frameworks. Furthermore, we examine embodied intelligence (EI) paradigms wherein agents learn and act in physical environments. To synthesise these dimensions, we present a conceptual framework that illustrates the synergy among LLMs, KBs, RA, and embodiment, serving as a guiding model for perception, reasoning, and action rather than an implemented engineering architecture. To advance toward GEI, we identify five key challenges: efficient LLM deployment, closed-loop knowledge integration, hybrid symbolic-neural reasoning, perception-action grounding, and continual learning. This survey provides a comprehensive roadmap for developing adaptive, multimodal agents capable of operating in complex, dynamic settings.

Pedestrian simulators need a behaviour rule for every agent, but privacy usually limits the data for setting one to aggregate statistics, namely zone-level device counts and origin-to-destination (OD) flows, with no individual trajectories. Such aggregates under-determine individual behaviour, because many different sets of decisions reproduce the same counts. We fine-tune a language model crowd agent so that the simulated population matches the observed destination composition, the fraction of the departing crowd heading to each point of interest. We read this target from the OD flow and reweight the model's own destination distribution onto it by iterative proportional fitting. Because fine-tuning inflates the dominant destination class, we fit the low-rank adapter to trajectories resampled to a corrected training composition that reaches the target after this inflation. On mobile network counts from two baseball games the fine-tuned agent runs without inference-time correction, cutting the destination-share error by 25%, while the grid correlation remains similar across policies.

The integration of inverter-based renewable energy sources into electric grids challenges conventional power system protection. Machine learning-based solutions can address these challenges by utilizing available data in modern smart grids. However, the lack of open datasets prevents reproducibility and fair comparisons between different approaches and their results, which hinders further progress. Therefore, this paper presents EvEMTBench, a synthetic dataset of faults and events generated using electromagnetic transient simulations. The physical plausibility of the power system simulation is ensured by validating the simulation parameters against established literature and providing comprehensive documentation of the simulation procedure. The dataset is designed for training, fine-tuning, and benchmarking machine learning models by providing synchronized point-on-wave voltage and current measurements at 9600 Hz across a diverse set of topologies and voltage levels. The inclusion of a wide range of fault and operating events allows the utilization of EvEMTBench for different tasks like incipient fault detection, fault localization, or event detection.

Current dexterous grasp planners primarily optimize for physical stability, focusing on whether an object can be grasped rather than how it should be grasped to support downstream functional tasks. However, conditioning grasp synthesis on specific human grasp taxonomies typically requires prohibitively expensive, object-annotated datasets. To address these limitations, we propose CoToGrasp, a novel generative framework that synthesizes diverse, stable grasps strictly conditioned on specific contact topologies. To bypass the data collection bottleneck, CoToGrasp is trained entirely in an object-agnostic manner. We introduce a feature-based canonical workspace that projects local object features into a unified gripper-centric domain, effectively decoupling the semantic functional intent from the arbitrary object geometry. By learning the intrinsic contact manifold of the gripper within this workspace, our model achieves zero-shot generalization to unseen objects at inference. Extensive evaluations on the large-scale DexGraspNet dataset demonstrate that CoToGrasp achieves state-of-the-art performance, outperforming existing taxonomy-guided planners. Finally, we demonstrate the physical viability and kinematic feasibility of our synthesized contact topologies on a physical robot platform. Code is available on our project website at https://cea-list.github.io/cotograspweb/ .

Audited against causal ground truth from executed replay in a single-agent tool environment (ALFWorld), none of the step-level credit signals used to train LLM agents -- LLM-judge scores, outcome-conditioned logprob ratios, or the policy's own confidence -- identifies which steps causally matter better than chance. Existing evaluations grade these signals against annotated step *correctness*; we audit them against step *contribution* -- what re-sampling the policy's own alternatives at each decision point and rolling forward actually changes about the outcome -- and the two come apart. The ground truth itself is structured: causal contribution is sparse (30.5% of decision points where ground truth is defined carry measurable effect), and measurability is model-dependent -- the fraction of points with no policy-supported counterfactual differs by a factor of two (13.1% vs. 26.8%) between two similar-scale policies. The failure mode is identifiable: implicit credit echoes the policy's fluency (median rank correlation +0.75, replicating at +0.70 in a second family under a corrected instrument), while conditioning on the outcome adds no causal information (partial correlation -0.004, Qwen). A confidence-only router recovers pivotal steps at chance level, but cuts judge cost by 13.1% per turn (14.0% per trajectory). In a seven-arm pre-registered training experiment, no arm reliably outperforms the untrained policy, and the checkpoints' apparent instrument signature is fully explained by training dose -- sparser credit retains fewer examples, an order-of-magnitude spread in optimizer steps -- not credit content. Comparisons of credit rules must therefore match effective sample size, or they measure dose, not credit.

Multifingered grasping is a crucial robotic skill, but current deep-learning grasp planners often struggle to generalize to new objects because they are trained on limited, object-specific datasets. We introduce a fundamentally different approach, grounded in the observation that the gripper and the object share identical surface geometry at their mutual contact points. We propose GOAG: Generative and Object-Agnostic Grasp Planner for Dexterous Robotic Manipulation, a novel deep generative model that learns a compact latent representation of a specific gripper's contact surface distribution, enabling the efficient sampling of valid grasp configurations without relying on object-specific training data. We show that by introducing object features only at inference time, our model can effectively retrieve admissible contact areas that are compatible with the gripper's capabilities. We validate our approach through extensive experiments on established grasp protocols in both simulated and real-world scenarios, demonstrating its effectiveness with different grippers from the literature. Our method delivers state-of-the-art results on the objects from the MultiDex dataset, achieving an average success rate of 86.93%. It offers significantly faster processing when generating numerous grasps, while matching the performance of leading approaches specifically trained on this dataset. Unlike these methods, our approach does not rely on object-specific training data, highlighting the advantages of object-agnostic learning. It effectively addresses the generalization challenges faced by traditional data-driven grasp planners. Code and videos are available on our project website https://cea-list.github.io/goagweb/ .

Long-context modeling is a pivotal capability for Large Language Models, yet the quadratic complexity of attention remains a critical bottleneck, particularly during the compute-intensive prefilling phase. Our previous work, FlashPrefill, mitigates this cost through instantaneous pattern discovery and max-based dynamic thresholding; however, it remains an algorithmic prototype that is still distant from production deployment. In this paper, we present FlashPrefill V2, which evolves FlashPrefill from a prototype toward practical long-context serving along three dimensions. First, we introduce a mean correction term that effectively suppresses the approximation error, keeping performance degradation manageable even at extreme sparsity levels. Second, we redesign the sparse attention operator with PackGQA memory access, warp specialization, and pingpong pipelining, fully aligning with the latest FlashAttention-3/4 implementations and supporting FP8 inference to meet practical quantization requirements. Third, FlashPrefill V2 natively supports paged KV cache and continuous batching, allowing integration as an attention backend in modern inference frameworks such as SGLang. Extensive evaluations on NVIDIA H20 GPUs---among the most widely deployed inference accelerators---demonstrate that FlashPrefill V2 delivers up to 47.26x and 27.19x speedups over FlashAttention-2 at 128K context length under FP8 and BF16 precision, respectively, and, in FP8, still achieves a 30.49x speedup against an FA3/4-aligned dense baseline.

Personalized text generation aims to make LLMs write in a specific individual's style, yet existing benchmarks measure task accuracy or preference alignment rather than whether the model's output actually resembles the target author's writing. We introduce PersonalBench, a benchmark that evaluates inference-time personalization methods through three independent lenses: LUAR (a trained authorship verification model), an LLM-as-judge, and automated stylometrics. Across 50 authors, 1,000 generations, and two model families (Qwen 3, GLM-4), we find that personalization methods do produce author-differentiated output (LUAR discriminates target authors within generated text at AUC=0.918) but this differentiation never crosses the human-LLM boundary. All methods achieve LUAR similarity to real authors in the range 0.484-0.508, below the cross-author human floor of 0.626 (ceiling 0.756). The LLM's own authorship fingerprint dominates: generated text is more distant from any human author than random humans are from each other. Methods are statistically indistinguishable from each other on LUAR (spread 0.024) despite appearing differentiated on the LLM judge, a discrepancy we trace to circularity between trait extraction and profile extraction. We validate that LUAR reliably measures authorship in our corpus (AUC=0.76 single-post, 0.96 multi-post). We release PersonalBench as a calibrated measuring stick: inference-time personalization modulates the LLM's style but does not bridge the gap to human authorship.

Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across numerous scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Across proprietary and open-weight models, the strongest achieves 65.36% pass@1, but only 25.25% pass^20. Moreover, many failed trials show clean termination and valid state-changing actions, showing that response or tool-call-level signals are not clear proxies for end-to-end task completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release both Thinkingbox and Thinkingbox-Bench: https://github.com/microsoft/thinkingbox

This paper presents an open-source software stack that restores ROS 2 support for the Franka Emika Panda robot while resolving the long-standing unreliability of its external position control interface. We first analyze the root causes of unstable position control and show that the observed vibrations and protective stops arise from the timing of the external control loop and sampling jitter, rather than from limitations of the robot itself. Building on this analysis, we introduce an asynchronous hardware interface that decouples real-time communication from the ROS 2 control loop, a rate-matching mechanism for slower command sources, and a position-domain reference generation strategy that produces reliable, smooth position commands. Experimental validation shows that the proposed architecture reliably tracks velocity references by reducing motion artifacts introduced by the official implementation, and the stack is validated across motion planning, compliance control, position-controlled manipulation, and haptic teleoperation on two independent Panda platforms. By restoring a modern, reliable, and open ROS 2 ecosystem for the Panda, this work lowers the barrier to developing safe, responsive, and reproducible human-robot collaboration applications that integrate planning, perception, interaction, and shared autonomy. Code and videos are available on our website at https://sites.google.com/view/fer-ros2/.

Vision-language-model-based embodied agents can complete instructed tasks but often violate safety constraints in the process, a problem recently framed as interactive safety. Training such agents to act safely is difficult, since safety and task success are distinct objectives, and safety arises only at a small number of safety-critical steps within a trajectory. Standard supervision is insufficient: imitating safe trajectories teaches behavior without explaining why it is safe, and contrasting arbitrary safe and unsafe trajectories mixes the safety signal with unrelated differences. We propose SafeBranch, a framework that aligns an embodied actor on safety through branch pairs constructed from the actor's own unsafe rollouts via environment rollback. SafeBranch rolls each unsafe rollout back to the safety-critical step that caused the violation, queries the actor for a safe alternative, and pairs the original action with the alternative so that the two branches differ only at that step. The trained actor acts safely at deployment with no critic in the loop. On IS-Bench, SafetyALFRED, and out-of-distribution variants with unseen tasks and objects, it handles safety reliably without sacrificing task success, achieving roughly ten times more safe successes than the untrained baseline on the unseen-object variant.

The typical training process of a multimodal large language model (MLLM) involves adapting both the language model backbone and the projector between the backbone and a modality-specific encoder. We ask whether fine-tuning the backbone of an MLLM is necessary to adapt it to a new modality. Through experiments on 3D MLLMs, we find that training only the projector is sufficient to achieve strong multimodal performance relative to existing baseline models and our jointly trained MLLMs with the same encoder and backbone. We also show that joint training leads to undesirable drift in existing capabilities of the language model, which projector-only training avoids by definition. Furthermore, projector-only training has approximately twice the training sample throughput of joint training. We validate our findings across different language model backbones via 3D classification and captioning benchmarks as well as standard benchmarks evaluating language, vision, and spatial reasoning capabilities.

Streaming video understanding requires models to causally update state as video arrives and organize growing history into semantic units that can evolve, persist, and be recalled under bounded computation and memory. This challenge is pronounced in live soccer commentary, where a system must describe completed events, summarize recent play, recall earlier events, or remain silent using only information available before each utterance. We present StreamSoccer, an event-driven system that uses event memory as its intermediate representation. A fixed-budget active memory integrates the stream; completed event states are retained locally and consolidated into retrievable historical records. A unified generator uses current, recent, and historical context to produce three commentary modes, while a rule-assisted scheduler selects a mode or silence. Unlike streaming video-language models organized around frames, visual tokens, or caches, and soccer-commentary methods based on predefined clips or output timestamps, StreamSoccer explicitly models event lifecycles. We construct a three-track streaming soccer commentary dataset and a layered evaluation protocol. At common reference anchors, StreamSoccer obtains CIDEr scores of 38.62, 23.96, and 17.39 for current-event, recent-window, and historical-memory commentary, ranking first on the current-event and historical-memory tracks and second on recent-window. Controlled ablations show that local completed events improve all tracks and that the full system performs best on all three. Across 174 raw-video runs on 58 matches, per-minute RTF p95 ranges from 0.10 to 0.22 without sustained growth with match history. These results indicate that event memory supports streaming soccer commentary across temporal scopes while controlling long-history computation.

In this paper, a bio-robot is an engineered living or biohybrid system in which living cells perform one or more core functions, such as sensing, information processing, actuation or output. We focus on systems whose cell-based functions are programmed by genetic circuits; physical movement is optional. Designing such a system requires translating application requirements into sensing, logic or memory, output, assembly, host and containment modules, while grounding each choice in traceable parts and evidence. We present micro_biorobot_agent, an offline multi-agent system built on Qwen3.5-27B. The system combines requirement analysis, module-specific retrieval, candidate assembly, conflict checking, local repair, independent review and validation over an integrated library of 23,762 records covering biological parts, measured combinations, literature-supported relationships and actuation evidence. Deterministic output checks align the final report with the retrieved part set and correct false gaps, unsupported part mentions and source-tracking errors. On two author-developed evaluation sets of 50 queries each, the system obtains mean overall scores of 7.35 and 8.04, the highest among the seven evaluated systems; on Scenario Design it exceeds the runner-up by 2.23 points. A 50-query paired ablation shows that the source-tracking check reduces false-gap incidents from 15 to 3, an 80% reduction, and increases source accuracy by 0.75 points. This paper reports the Qwen3.5-based v1 system and evaluates high-level design reports rather than experimentally validated circuits.

Recent studies investigate how to leverage pre-collected datasets to improve the policy performance and sample efficiency of RL. One promising approach to achieve this goal is to employ a two-stage strategy: In the first stage, diverse skills are extracted as a low-level policy from a given dataset, and a high-level policy is trained to solve a specific task in the second stage. Typically, extraction of the low-level policy is performed based on unsupervised learning such as trajectory VAE. However, a limitation of this approach is that the quality of the low-level policy highly depends on the quality of the dataset. To address this issue, we introduce QDOS (Quality-Diversity Offline Skill learning), a unified pipeline for robust offline-to-online learning. Our approach incorporates an Advantage-Weighted Quality-Diversity pretraining objective, which weights the skill extraction and diversity objectives by the estimated advantage of each trajectory segment. This approach allows the model to extract diverse and high-value skills. By providing robust and task-relevant skill representations, QDOS significantly improves the quality of the embedded skill space used by the low-level policy. We further integrate this with a dual dataset reuse strategy, where offline data is used both for skill pretraining and for populating the online replay buffer via pseudo-labeling. Experiments demonstrate that QDOS significantly outperforms strong baselines in structured manipulation tasks and unstructured locomotion tasks, confirming its ability to accelerate exploration and improve final returns in challenging sparse-reward domains.

Light-hole (LH) excitons provide access to well-defined polarization and spin degrees of freedom that are central to quantum photonics and chiral light-matter interactions. Achieving LH emission is challenging because LH states are energetically unfavoured and typically relax non-radiatively. Existing strategies to access LH excitons rely on modifying the electronic band structure through strain, shape anisotropy, or piezoelectric fields, approaches that are material-specific and offer limited post-synthesis tunability. Here we demonstrate an all-photonic route to activate LH exciton emission in colloidal CdSe-CdS nanoplatelets (NPLs) using a distributed Bragg reflector (DBR) cavity, without altering the underlying band structure. In the absence of a cavity mode, the system exhibits amplified spontaneous emission from heavy-hole (HH) states without detectable LH emission at low excitation powers. By spectrally matching a cavity resonance to the LH exciton, cavity-coupled LH emission emerges at significantly lower excitation powers. Temperature-dependent spectroscopy reveals reversible switching between LH- and HH-coupled emission through exciton-cavity detuning, while polarization-resolved and spectrally resolved time-resolved photoluminescence measurements provide independent evidence distinguishing the cavity-coupled LH and HH emission channels. These findings establish cavity engineering as a general materials-level approach for accessing radiatively suppressed optical states.

Agentic language models repeatedly encode tool and skill schemas that recur across requests in different combinations and orders, preventing standard prefix caching from reusing their key--value (KV) states. We introduce \textbf{ReCache}, a framework for independently caching resource representations while reducing their inference-time computational and memory overhead. Resource-wise attention removes cross-resource interactions and assigns resource-local positions, producing composition-invariant KV blocks. ReCache then restricts resource visibility to contribution-selected layer--KV-head-group routes and retains only invocation-critical fields through structural and semantic pruning. We evaluate ReCache on a benchmark assembled from seven public tool- and skill-use datasets, including resource-disjoint tests. Resource-wise attention matches dense invocation performance (82.3\% versus 82.4\% Inv-F1) while providing a 3.655$\times$ time-to-first-token speedup. The complete framework reduces allocated KV-tensor memory by 92.43\% and accelerates attention by 1.423$\times$. These results show that separating reusable schema encoding from selective resource access substantially reduces agentic inference costs with limited effectiveness loss. The code is available at https://github.com/EIT-NLP/ReCache.

Large language models can turn a natural-language mission into a sequence of robot actions, but they do not have a sense of physics: they cannot judge how long a command should run, or whether it will make the robot drift into an obstacle. We proposed the use of a world model to expand the capabilities of Large Language model-based planners. Our method has three components: a physics-grounded neural world model, a three-phase gradient-based trajectory optimizer, and a Model Predictive Controller (MPC)-style closed-loop replanner with a trust-region guard. The language model decides what to do, and the world model decides how long, whether that means driving eight thrusters through 6 DOF or two differential thrusters through 3 DOF. We evaluate two marine vehicle classes operating near offshore wind infrastructure: a 6-DOF Autonomous Underwater Vehicle (AUV) and a 3-DOF differential-drive Autonomous Surface Vehicle (ASV). In five benchmark missions per platform, both vehicles reach every goal with zero predicted collisions, and both transfer to GazeboSim under ocean current, waves, and thruster dynamics, remaining collision-free and cutting GazeboSim goal-distance error versus the ungrounded baseline by 70-82% (ASV) and roughly 93% (AUV), after a residual fine-tuning pass that separately reduces surrogate rollout Root Mean Square Error (RMSE) by 60% (AUV) and 69% (ASV). For the ASV we further demonstrate a Vision language model (VLM)-assisted semantic-mapping pipeline that extracts obstacles and environmental context from satellite imagery, nautical charts, and forecast Application Programming Interface (API) instead of onboard sensors, reaching 96% navigability accuracy as a drop-in replacement for hand-specified obstacle geometry.

As LLM-based agents are deployed for longer and higher-stakes tasks, their memory systems continue to have crucial gaps. While existing memory benchmarks focus largely on recall-shaped tasks, we argue an effective memory system must track the evolving state of the world; as facts, constraints, and decisions are revised over a long interaction, answers must reflect the current state and not a superseded one. We define this capability as state tracking and instantiate it in StateMemBench, a benchmark of 234 multi-session scenarios spanning two conversation-length regimes. Its closed-pool grading scores whether an answer reflects the current state, the superseded state, or fails otherwise, separating state-tracking failures from other errors by construction. Our analysis shows that this task is challenging for existing memory systems, retrieval-augmented baselines, and long-context baselines. We then present StateMem, a state-first memory method that explicitly tracks supersession and relational dependencies, and show it improves current-state accuracy over the strongest same-backbone baseline by 1.8x (0.205 -> 0.363) on DeepSeek-V4-Flash and over the strongest memory system by 1.6x (0.149 -> 0.233) on Qwen-3.5-9B, while remaining competitive with the long-context baselines. Finally, we show the same state approach can be applied as a lightweight single-call wrapper over existing memory systems, lifting current-state accuracy by +32 to +67 points on StateMemBench across six memory and retrieval backends. A length- and cost-matched control attributes +15 to +32 of those points to state structure rather than added context.

The transportation community seeks to use calibration methods for highway traffic microsimulation. This is a response to the time-consuming and subjective nature of traditional manual calibration, as well as the growing prevalence of data for calibration. This work argues that this "automatic" calibration is largely an illusion. A significant - and unquantified - amount of bespoke manual work is hidden behind these methods. This illusion inhibits a core component of scientific advancement: objective comparison against a shared standard. This impedes evaluation, undermines reproducibility, and fragments research. To address this gap, this paper introduces a comprehensive benchmark designed to simultaneously expose the calibration illusion for highway microsimulation and provide a common ruler. The results across a range of scenarios present a new baseline for what the algorithms can achieve without bespoke tuning, revealing the research gap that remains and providing a tool to advance a cumulative science of calibration. Additional experiments provide insights into the source of calibration errors that arise in large-scale highway calibration relative to the simplified settings under which methods are commonly developed.

Large language models (LLMs) offer a scalable approach to social simulation, but their credibility depends on how agents are constructed. Existing methods can partially reproduce population-level patterns, yet often fail to capture human-like diversity. Our analysis shows that static-profile agents exhibit stronger demographic separation and within-group compression than humans, a pattern consistent with identity essentialism: demographic labels can encourage models to treat group-average tendencies as individual traits, homogenizing responses within groups. We argue that this limitation arises from two related factors: sparse, static agent representations and the limited ability of prompt-only memory to persistently integrate experience. Inspired by complementary memory systems, we propose LifeMem, a longitudinal memory framework that combines structured life-event retrieval with agent-specific parametric memory for experience integration. Experiments on Add Health and Understanding Society with three LLMs show that LifeMem improves alignment with human data in terms of response distributions, overall and within-group diversity, and patterns of within-person response change across life stages. These findings highlight the value of longitudinal life-event memory for constructing more faithful and dynamically evolving social agents.

Latent Action Models (LAMs) have emerged as a promising paradigm for enabling robot learning to leverage large-scale unlabeled videos through latent actions that serve as compact surrogates for physical actions. Despite rapid progress, research on LAM remains highly fragmented, with existing methods evaluating different design choices in isolation under inconsistent experimental settings, making it difficult to identify the factors that truly determine downstream robotic manipulation performance. In this work, we present the first comprehensive empirical study of latent action learning for robotic manipulation. We unify representative LAM methods within a common autoencoding framework and systematically investigate 41 LAM design choices across three dimensions, including latent action modeling paradigms, learning objectives and regularization methods, and latent action integration strategies. We further examine four proxy metrics for evaluating latent action quality and assess their ability to reliably predict downstream robotic manipulation performance. Extensive experiments on three widely used benchmarks provide strong empirical evidence that fine-tuning vision-language model (VLM) backbones with latent actions provides a stronger initialization for downstream policy learning, with further validation on real-world robot manipulation tasks.

Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, its adaptation to multimodal settings remains unexplored. Through representational analysis, we identify a key limitation in multimodal preference optimization, which we term visual insensitivity: models often fail to distinguish between images and those with critical visual context removed. Our theoretical analysis further uncovers two manifestations of this problem, namely Across-Image Insensitivity and Within-Image Insensitivity. To address these challenges, we propose Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity. We further provide a theoretical analysis demonstrating that PEA-DPO provably mitigates both failure modes. Empirical results demonstrate that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model. Evaluations across three hallucination benchmarks using MLLMs of varying scales show that PEA-DPO effectively mitigates visual insensitivity, achieves stronger multimodal alignment, and substantially reduces hallucinations.

Advances in technology have led to increasingly sophisticated musical humanoid robots. However, their use has largely been limited to performance and related research in human-robot interaction. In this position paper, we propose a novel perspective: musical humanoid robots as experimental interfaces for investigating music-evoked emotions. We argue that current research is constrained by paradigms relying on pre-recorded auditory stimuli, which fail to capture the multimodal, embodied, and interactive nature of real-world musical experience. Building on existing theories of music cognition and emotion, we identify mechanisms that require controlled manipulation of both acoustic and non-acoustic variables. We show that humanoid robots are well-suited as they enable parametric control of performance variables, reproducibility across trials, and the decoupling and recombination of auditory, visual, and interactive components. We illustrate the technical feasibility of this perspective through a case study of the WAseda Saxophonist Robot 5 (WAS-5), demonstrating reproducible control of acoustic and interaction variables that are prerequisites for future music-emotion experiments. Our work positions musical humanoid robots as a methodological platform that enables future controlled investigations of music-evoked emotions.

A task-effective proposal is not yet a justified physical action. In networked Physical AI, a proposal may be understood while valid, timely, proposal-bound evidence or the authority required to finalize an action remains unavailable. We call this mismatch the verification gap and introduce a Post-Semantic Communication Framework for the systems interface between proposal formation and physical execution. The framework begins with application-declared evidence requirements, represents qualifying observations as evidence records, validates supporting and conflicting records through one path, and separates evidence sufficiency from authorized finalization and a downstream runtime gate. It further distinguishes evidence transfer, which can enlarge the record set reachable by a finalizer, from evidence coordination, which can suppress transmission around records already held at the finalization endpoint. Finite-state framework checks verify that the evaluator implements the declared distinctions consistently. Under the declared model, the controlled communication study exposes a finalizer-dependent asymmetry: sender-finalized Feedback uses evidence transfer to expand evidence reachability throughout the feasible plotted region, whereas receiver-finalized Feedback uses coordination to suppress redundant payload until loss, latency, freshness, and deadline costs shift selection to One-way. Finally, an episode-level reporting schema defines common denominators for future measured Physical-AI studies.

Pretrained Vision-Language-Action models provide a strong foundation for robot learning, but sequentially adapting them to diverse skills can perturb the representations and velocity mappings used by previous skills, leading to catastrophic forgetting. Architecture-based approaches improve retention by isolating skills but lead to increased inference footprint. Recent subspace-constrained methods restrict parameter updates in an orthogonal subspace to minimize interference but impose a unified constraint on the entire model. We analyze the distinct roles of internal VLA components and identify two VLA-specific challenges. First, the VLM maintains broad semantic representations, making it vulnerable to capacity exhaustion, whereas the ActionHead refines semantics into localized velocity patterns that are highly sensitive to perturbations. Second, the final velocity decoder serves as a readout layer. Freezing it forms an output-stage expressivity bottleneck, while updating it risks overwriting previous velocity mappings. To this end, we propose OrthoSkillVLA, a parameter-efficient framework for continual skill learning in pretrained VLA models without demonstration replay. Given the representation heterogeneity, we impose separate subspace constraints on the VLM and ActionHead, preserving reusable semantic capacity while protecting localized velocity patterns. For the output layer, we introduce a lightweight feature-aware MoE decoder, where each skill is allocated a compact expert and a training-free router selects the expert according to feature-space affinity. Extensive simulated and real-world evaluations, together with ablations, demonstrate that OrthoSkillVLA better preserves prior skills while acquiring new ones.

Ensemble nitrogen-vacancy (NV) centers in diamond enable robust vector magnetometry in unshielded environments, yet deployment remains bottlenecked by complex calibration and a reliance on external data references. Conventional statistical machine learning requires an exorbitantly large volume of training data and suffers from severe simulation-to-reality mismatches. To address this, we introduce a physics-guided hybrid machine learning framework that embeds the Zeeman splitting directly into the learning pipeline. Our physics-guided model significantly reduces the average tracking error demonstrating a 372-fold precision improvement over purely statistical baselines. Furthermore, our hybrid architecture pairs a sparse physical measurement with scalable synthetic data generation, seamlessly incorporating real-world hardware non-idealities. When deployed to decode uncalibrated, raw experimental ODMR data, our framework delivers exceptional predictive accuracy for the scalar magnetic field. This work paves the way toward self-calibrated sensors while establishing a machine learning training method applicable to other data-scarce physical systems

Formal proofs in Lean 4 that pass the kernel's type checker can nonetheless vary widely in quality. We introduce ProofJudge, an agentic LLM-as-judge system that scores formal proof quality along five dimensions beyond correctness: library leverage, automation fit, structural clarity, statement quality, and Mathlib conventions. We evaluate ProofJudge on a novel dataset of 218 declarations drawn from distinct Mathlib PRs. The judge agent is grounded by tool access to the commit the PR is applied to, enabling it to query the library state when scoring. A judge is considered aligned with human preferences when it rates the version of the PR Mathlib accepted above the initial version that was sent back for revision. All six judge models evaluated recover the reviewers' preference well above chance, from 80.8% to 63.5%, and two open-weight judges reach roughly 70% at a tenth of the best judge's cost. We release the judge harness, evaluation dataset, and evaluation traces as open-source artifacts to support further research.

World action models jointly predict future visual observations and actions, whereas existing tactile-aware variants typically represent future touch as an image or latent stream without modeling the physical dependencies that organize tactile states hierarchically. We present HiTac-WAM, a hierarchical tactile world action model that forecasts a sequence of future tactile states for each candidate action chunk before execution. The forecast factorizes into contact state, a 3D deformation field, and slip risk, organized as a directed hierarchy in which each downstream stage is conditioned on stop-gradient signals from preceding stages. A directed attention mask allows tactile queries to attend to the video-action context of each candidate while preventing video and action queries from attending to tactile tokens. For planning, HiTac-WAM ranks candidate action chunks using tactile forecasts and task-progress estimates. For execution, the selected tactile forecast is retained as a reference; persistent discrepancies between predicted and observed tactile states trigger corrective replanning. HiTac-WAM achieves a mean contact F1 of 0.921; under matched training budgets, the directed hierarchy reduces 3D displacement L2 error by 17.6% relative to the deformation-only predictor and improves slip AUPRC by 60.4% relative to the slip-only predictor. Across chip grasping, blackboard erasing, and USB insertion, selection guided by the hierarchical forecasts increases the average real-robot success rate from 31.1% to 61.1%, while the full system attains 72.2%.

Persistent memory can personalize an LLM agent, but an incorrect durable update can silently distort future behavior. We study the memory-clarification boundary: whether interaction-derived information should be persisted, used only in the current context, re-verified, or clarified with the user. MCB contains 140 primary scenarios, split into 70 development and 70 held-out items, plus a separate 70-item contrast set. It evaluates both action labels and structured tool-call selection. Two non-authors independently label the 70 held-out primary and 70 contrast items (97.1% agreement, Cohen's kappa = 0.962); a blind third resolves four disagreements, replacing eight author labels by non-author majority. Across Claude and Qwen, models verify changing facts more reliably than they ask users to resolve ambiguity. Bare Qwen asks on 0/12 clarification items while verifying 12/18 freshness items. Few-shot prompting raises accuracy from 0.557 to 0.771 (paired delta = +0.214, Holm-adjusted exact McNemar p_H = 0.002), yet clarification recall remains 0.333. The policy prompt reduces erroneous persistence from 0.243 to 0.100 (p_H = 0.038), although its accuracy gain is not significant. Label-tool agreement is 57% for each Claude model and 23% for Qwen; Qwen accuracy falls from 0.557 to 0.343 (p_H = 0.047). Memory evaluation must test both stated decisions and tool-call choices.

Autonomous vehicles offload latency-sensitive perception tasks to nearby mobile edge computing (MEC) servers, where a missed safety-critical task is unsafe rather than merely degraded. Large language models (LLMs) are increasingly proposed as adaptive, explainable schedulers, yet evidence of when they help is scarce. We study deadline-aware, mixed-criticality scheduling on heterogeneous MEC servers, where time-critical (TC) tasks must be protected at a controlled cost to best-effort traffic, and ask whether a multi-agent LLM control layer improves on a strong heuristic. We answer in two steps. First we build the heuristic: a windowed contract-net auction that orders each admission window time-critical-first by earliest deadline and places tasks by earliest-finish-time. Across 60 instances on three topologies and 15 baselines under an identical online constraint, it attains a TC completion rate of 0.902, above every baseline (Holm-corrected p < 0.001; best baseline 0.838) and at 0.87 of a CP-SAT upper bound. Second, we add the LLM control plane. A controlled decomposition traces the scheduler's advantage to two ordinary factors, the batching horizon and time-critical-first ordering; the auction, the per-window LLM policy, and online adaptation add nothing while the load is stationary, where the heuristic is already near-optimal. Under a mid-run surge of safety-critical tasks the picture changes, and the LLM control plane gains significantly over both the static heuristic and the bandit. LLM orchestration therefore earns its cost only when non-stationarity opens headroom a fixed policy cannot use. We report control-plane latency and rationale, and release all code and seeded instances.

Accelerating the electrification of thermal energy requires vapor-compression heat pumps capable of dynamic, grid-responsive operation. However, equipment engineering remains fragmented across static rating-point selection, stiff multi-phase transient simulation, and gradient-based optimal control. Here, we present an end-to-end differentiable, finite-volume vapor-compression framework implemented natively in JAX that automates machine sizing directly from stated thermal duties and unifies dynamic simulation with predictive control under a single compiled residual $\dot{y}={f}(t,{y},{u})$. Thermodynamic evaluations bypass runtime root-finding via bilinear $(p,h)$ manifolds pre-flashed from Helmholtz equations of state, enabling analytical forward-mode automatic differentiation. Mass conservation across multi-phase coils is strictly preserved by incorporating both $(\partialρ/\partial p)_h$ and $(\partialρ/\partial h)_p$ partial derivatives into the dynamic pressure differential equation. The sizer directly inverts compressor displacement, electronic expansion valve area, and heat-exchanger tube counts via four-point cycle synthesis and $\varepsilon$-NTU matching using the identical polytropic compressor map. Crucially, the compiled physics kernel is shared symmetrically between $L$-stable TR-BDF2 stiff integration and implicit-Euler Model Predictive Control (MPC), eliminating plant-controller surrogate mismatch. Validated against open-access experimental benchmarks without parameter fitting, the framework predicts cooling capacity with $7.37\%$ MAPE across 16 mini-split operational runs and bounds on-period cooling error within $1.19\%$--$1.62\%$ on utility-scale Hardware-in-the-Loop traces. This work provides an open-source, differentiable foundation for automated machine synthesis, dynamic grid orchestration, and gradient-based hardware-control co-design.

Scan-to-map LiDAR odometry drifts unboundedly along the unobservable axes of geometrically degenerate environments like tunnels and corridors, and existing degeneracy handling requires environment-specific parameter tuning. This paper presents a parameter-free approach. We show that in voxelized GICP the Gauss--Newton (GN) Hessian masks translational degeneracy, because covariance regularization keeps the translation block artificially well-conditioned. We bypass this with a regularization-free voxel-normal localizability field and two of its statistics: a normalized fraction $f_0$ detecting directional anisotropy, and an absolute per-voxel mass $λ_0$ distinguishing information absence (tunnels) from dilution (dense open scenes). A temporal-median gate combines both to trigger Fisher-information correspondence weighting. Calibrated once by fixed rules on two short sequences and then frozen, LF-GICP achieves the lowest KITTI relative translation error ($0.865\%$) under an identical evaluation protocol against re-run baselines, outperforms them on GEODE tunnels and MulRan, leads the HeLiPR mean, and generalizes across four sensor types without re-tuning. We further demonstrate empirically that straight, uniform tunnels remain unobservable along their axis for LiDAR-only registration.

World models increasingly treat compact multimodal representations as interfaces between perception and physical interaction, yet existing probes do not establish whether different sensors carry the same executable meaning or whether that meaning survives a new action composition. We introduce an operational capability hierarchy and the Disjoint-Bridge Operator-Substitution Certificate (DBOSC), which asks whether independently trained modality compilers enter a frozen response chart interchangeably on evidence outside their training panels. On Cluster Haptic, audio and acceleration representations of the same unseen surface are 4.5x closer in response space than wrong-surface pairings, with the gap holding for all 19 held-out surfaces; unsealing withheld responses confirms that every branch predicts the physics better than the population chart. We then test ordered execution in a controlled elastoplastic system with complementary modality blind spots. At the pre-registered budget, the prerequisite refuses the stack because the frozen executor cannot advance even an exact chart coordinate through a held-out program. At a converged budget, the same rank-three chart executes those programs (oracle NMSE 0.18), fusion improves on both modalities, and 14 of 16 registered checks pass; the two failures arise because a diagonal restriction of the fused information matrix performs as well as the full one. Clearing the gate is a property of the executor, not the chart: an executor emitting whole programs instead of shared per-step dynamics is 38x worse than an entity-blind predictor on the same chart. A matching non-identifiability result explains why compression and fusion alone cannot determine an unseen composition law. These results separate attribute access, response substitution, fusion closure, and ordered execution into distinct, separately testable achievements.

State-of-the-art vision-language-action (VLA) models such as $π_{0.5}$ exhibit strong semantic understanding, instruction following and task behavior. However, when deployed on new robots, even minor mismatches in hardware configuration relative to pretraining can cause severe performance drops. Finetuning the VLA on in-domain expert data from the new embodiment improves performance on the expert task but leads to a loss in its original instruction following and behavioral priors. In this paper, we propose a self-supervised method that generates online interaction rollouts from the zero-shot VLA as additional training data for finetuning. Our experiments show this finetuning scheme yields strong multi-task policies that, on the target robot, (1) inherit prior tasks distilled from the zero-shot model, (2) enable generalist instruction following, while (3) learning new skills from expert data with improved sample efficiency. We demonstrate the success of our approach across test sets probing generalization on a real ALOHA robot and a new simulation benchmark in RoboTwin. Video results are available at https://self-supervised-control.pages.dev/

At fixed pulse energy, which drive waveform extracts the most $n$th-order nonlinear conversion from a resonator ($n=2$ for second harmonic)? A short pulse couples poorly to a narrow resonance, a long one dilutes its energy, and no linear rule fixes the compromise. We solve the problem exactly for a single mode of amplitude decay rate $κ$. Eliminating the drive turns fixed incident energy into a constraint on the stored field alone, and the optimization becomes a sharp Gagliardo--Nirenberg inequality whose extremal is the ground-state soliton of the nonlinear Schrödinger equation. The optimal stored field is $\mathrm{sech}^{1/(n-1)}[(n-1)κt]$, sustained by an asymmetric input that rises as $e^{κt}$ and falls as $e^{-(2n-1)κt}$; the largest converted energy follows in closed form. A rising exponential, the time-reversal recipe, retains at most $79.0\%$ of the bound at $n=2$ and $2/e$ at large $n$; a two-rate pulse retains above $97\%$. Critical coupling generalizes to $n$-fold overcoupling, with optimal input coupling $n$ times the intrinsic loss rate. The bound applies from microrings to superconducting circuits and caps the per-pulse brightness of broadband photon-pair sources.

Stream-based robotics Task and Motion Planning (TAMP) integrates discrete symbolic planning with dynamically generated continuous geometric parameters, such as poses, grasps, and trajectories. However, stream-based planners typically reason only about goal reachability, whereas long-horizon tasks also demand adherence to temporal specifications, such as safety-critical ordering, invariance, and liveness constraints. No methods currently exist to enforce such temporal constraints for stream-based solvers because streams generate an expanding geometric object set via iterative stream refinement loops during planning, rendering existing temporal-logic compilation techniques incompatible. We therefore present Synchronous Action Monitoring with Token Destruction (SAM-TD), a compilation method that enforces arbitrary Linear Temporal Logic over finite traces ($\textrm{LTL}_f$) specifications in stream-based TAMP. SAM-TD translates arbitrary $\textrm{LTL}_f$ constraints into automata and embeds regressed automaton guards into action schemas, which are pre-specified before planning begins. By doing so, SAM-TD can handle objects generated by streams during planning, thus circumventing the need to enumerate a fixed object set or modify the underlying planner. During search, SAM-TD synchronously updates automaton states and uses a validity token shared across all automata to prune constraint-violating branches. We show that SAM-TD supports dynamically generated stream objects from iterative stream refinements during plan search. Experimental results provide the first ever demonstration of stream-based TAMP under $\textrm{LTL}_f$ constraints in three robotics PDDLStream environments. Furthermore, on standard discrete PDDL benchmarks, SAM-TD is competitive with state-of-the-art temporal-constraint compilation methods.