Physical AI Brief
Daily cross-source signals for the Physical AI supply chain — silicon photonics, CPO, VLA models, humanoid hardware, embodied AI. Three streams, one page, zero filler.
629 items today · 570 arxiv · 5 SEC 8-K · 54 humanoid · 0 CN photonics
01 ARXIV · PHYSICAL AI PAPERS
570 items- arxiv:2611.21608 · cs.LGLatent Lattice Language ModelsShivam Duggal, Zongze Wu, William T. Freeman, Antonio Torralba +2
Language models typically increase per-token computation through deeper or looped layers and latent reasoning steps, while richer residual connections improve information flow across layers. We instead increase computational capacity: the number of active computational states a model maintains. We introduce the Lattice of Latents, a set of per-layer horizontal latent streams that run recurrently along time, alongside the vertical residual stream that flows through depth. Like hyper-connections, mHC, and attention or value residuals, the Lattice enhances the residual stream, but rather than adding or rerouting residual pathways through depth, it provides companion computational states that persist across tokens. Unlike hybrid Transformer-SSM architectures, whose recurrent states primarily serve as internal memory for sequence mixing, these latents directly condition the vertical stream's attention, KV-cache writes, MLP computation, and decoding. On Transformers trained on 20B FineWeb-Ed
memory - arxiv:2611.21607 · cs.ROScaleVLA: Scaling Vision-Language-Action Models to 552 Billion ParametersRuixi Yu, Dairu Liu, Xuchuan Chen, Zekun Qi +8
Frontier Vision-Language Models (VLMs) such as GPT-6 Astra have demonstrated remarkable capabilities in robot control, suggesting that broad pretraining provides knowledge valuable for physical action. Yet most Vision-Language-Action (VLA) policies still rely on much smaller VLM backbones, leaving this potential largely untapped. We ask whether frontier-scale representations can provide a stronger foundation for robot learning. We introduce ScaleVLA, which connects a 552B DeepSeek-V4.1-Flash backbone to a trainable action DiT, and systematically compare VLMs from 2B to 552B under a shared training setup. We evaluate three fundamental capabilities of robot foundation models. In sample efficiency, just 10 demonstrations outperform a 2B backbone trained with 50 (87.8% vs. 85.6%). In generalization, LIBERO-PRO success reaches 45.1%, compared with at most 31.3% for smaller backbones. In precise manipulation, real-world charger insertion improves from 4/10 to 9/10 successes, with gains conce
vision-language-actionmanipulationrobot foundation modellibero - arxiv:2611.21605 · cs.CVViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy OptimizationHuawei Lin, Jinting Liu, Tongyan Wang, Xuyang Chen +8
Visual agents answer questions about ultra-high-resolution photos by zooming into the relevant regions. However, existing agents often answer from common sense instead of reading the evidence. Reinforcement learning (RL) on the final answer cannot correct this, because a guess earns the same reward as reading. We propose Cross-Input Relative Policy Optimization (CrossRPO), which trains agents on contrast sets. A contrast set asks one question about several versions of an image that differ only in a small region with the evidence and have different answers, so only reading the evidence solves the whole set. CrossRPO rewards solving the whole set, and we prove that it does not reinforce a set answered from common sense. We build contrast sets for training (ViRA-Train) and evaluation (ViRA-Bench) by editing one small object in images of 21 to 120 megapixels, and we photograph further sets with real changes (ViRA-Bench-Real) to exclude editing traces. We train two Visual Reasoning Agents,
benchmark - arxiv:2611.21604 · cs.CVFaceJev: Direct Visual Decisions for Face Forgery DetectionSiran Peng, Tianshuo Zhang, Yibo Xue, Ajian Liu +5
Face forgery detection methods often perform well on seen manipulations but generalize poorly to unseen ones. Multimodal Large Language Models (MLLMs) can recognize facial parts and assess their visual plausibility. Existing MLLM-based methods typically express these judgments through textual explanations, but seldom treat individual part decisions as explicit learning targets. We present FaceJev, an MLLM-based method for face forgery detection through direct visual decisions. Following the Jev formulation, FaceJev answers eight independent questions about facial parts using a shared image encoding and reads anomaly probabilities from the native answer logits. A face is classified as forged if any part is predicted to be abnormal, directly linking the detection result to the part decisions. We construct PartForensics, a dataset of approximately 160k images with binary anomaly targets for the eight facial parts, to supervise these decisions. We further propose FS-GRPO, which optimizes e
manipulation - arxiv:2611.21598 · cs.LGMitigating Catastrophic Forgetting with Physics-Indexed Specialist Memory: the Evolutionary Neuroplastic AgentBlas Eugenio Vicco Zarate, Miguel Hernán Vicco
Continual reinforcement learning requires adaptation to changing conditions while preserving performance under previously encountered dynamics. It was investigated whether specialists generated by evolutionary search can provide a reusable memory when combined with environment detection and failure-driven search. The Evolutionary Neuroplastic Agent (ENA) evolves small neural networks without gradients, archives selected policies by detected dynamic regime, and retrieves them through Bayesian policy reuse. The architecture was evaluated in non-stationary CartPole and HalfCheetah environments. In the 30-replicate CartPole ablation, the complete architecture achieved a mean success rate of 83.2% across the five test environments. Among variants that learned the harder CartPole environments, the complete architecture showed the most consistent retention. A selection ablation also showed that the retrieval rule affected performance. In the evaluated sequences, ENA adapted more slowly than P
memoryagent - arxiv:2611.21579 · cs.AIScenarioForge: Resource-Grounded Generation and Execution Validation of Cyber ExercisesJaime C. Acosta, Anantaa Kotal, Aritran Piplai
Cyber-range and capture-the-flag scenario construction remains labor-intensive because authors must coordinate infrastructure, vulnerable services, attack paths, artifacts, and success conditions. LLMs can propose detailed exercises, but plausible descriptions do not guarantee deployable software or verifiable objectives. We present ScenarioForge, which turns high-level requirements into resource-grounded topologies, services, traffic, segmentation, and challenge chains, then instantiates and validates them in the Common Open Research Emulator (CORE). The ScenarioForge Execution Dataset contains 226 materialized scenarios covering 294 Vulhub images and all 85 custom-built challenge-resource generators developed with our AI-assisted scaffolding. All 226 passed construction, preview, sequencing, deployment, strict runtime validation, and live artifact checks; their 867 steps, each an individually verifiable challenge objective, comprise 457 vulnerability-backed and 410 generated-resource
benchmark - arxiv:2611.21577 · cs.AIWhen Tool Names Are Not Identities: Measuring Provider Misbinding in MCP-Based AI AgentsZeynab Anbiaee, Sajjad Dadkhah, Ali A. Ghorbani
Large language model (LLM) applications increasingly act as agents that select and invoke external tools. The Model Context Protocol (MCP) standardizes this interaction by allowing clients to compose tools advertised by multiple servers. However, tool metadata used for discovery and selection is not, by itself, necessarily bound to a verified provider identity. This paper studies this gap as tool-provider misbinding: a failure in which a request intended for a legitimate provider is instead resolved to an attacker-controlled tool. The attacker does not compromise the host, LLM, legitimate server, or network channel; it only operates another MCP server that advertises tools with chosen metadata. We formulate tool provider misbinding as an attacker-controlled tool-selection optimization problem and evaluate it in a controlled MCP environment across five LLM-based clients. The results show that interactions drive misbinding. The strongest failures occur when exact-name collision, task-ali
ai agent - arxiv:2611.21565 · cs.LGWhen Do Histology Foundation Models Need Geometric Test-Time Augmentation?Lars Chen, Valentino F. Foit, Dorry Segev, Michal Mankowski
Test-time augmentation (TTA) aggregates model predictions across multiple transformed views of an input during inference. Because histology images lack a canonical orientation, $D_4$ TTA---pooling across the 8 rotations and reflections of a square---is a standard default in computational pathology. However, the rise of self-supervised histology foundation models (FMs) with inherent rotational invariance calls its continued utility into question. We benchmark 9 histology FMs and 6 general-purpose models across four patch-level and two whole-slide image (WSI) datasets. We find that $D_4$ TTA remains a reliable post hoc improvement, but benefits diminish from $+1.98$ percentage points (pp) for general-purpose backbones to $+0.50$ pp for histology FMs on TCGA-UT, and shrink further for the largest models. To eliminate computational overhead, our experiments demonstrate that entropy routing---applying TTA selectively to uncertain predictions---recovers most of the gain while augmenting only
benchmark - arxiv:2611.21554 · cs.LGReading or Remembering? Measuring Temporal Contamination in Language ModelsArunkumar V, Gangadharan G. R., S. Senthilkumar
Language models are now used to read central-bank text for economic forecasting, but most were trained on text written after the documents they read. A model reading a 2021 FOMC statement may already know that inflation rose in 2022, so part of its apparent skill can be memory of the outcome, a temporal form of test-set contamination. We measure this memory on FOMC statements and minutes from 2000 to 2026 by comparing modern open models with chronologically consistent models, which are trained only on text from before a fixed date. Asked about past GDP outcomes given only a date, models that cannot know the answer still reach an AUC of up to 0.64 from priors on dates and eras, so chance is the wrong baseline. Against these memory-free models on the same questions, most modern models recall which years had strong growth. Membership-inference scores sort documents by era and document type rather than by whether a model saw them, as exact counts in an open pretraining corpus show. In real
memory - arxiv:2611.21553 · cs.CVBLIP-Live: Balancing Timeliness & Firing Precision in Proactive Streaming Video AgentsKanchana Ranasinghe, Honglu Zhou, Dominick Reilly, Le Xue +4
Streaming video understanding is becoming increasingly important for real-world interactive systems. Such systems must be proactive for natural and useful human interaction: they should respond as soon as sufficient evidence becomes available while avoiding premature, redundant, or incorrect responses. We present BLIP-Live, a framework unifying data, metrics, harness, and training around a shared formulation of proactivity as effective causal response decisions balancing the two competing objectives of responding vs silence. First, we construct BLIP-Live-780K dataset containing human annotations of proactive response content and timing, which defines this desired behavior. We then propose PAUC-F, a metric to effectively measure our desired proactive behavior. PAUC-F combines existing response timeliness metrics with firing precision to reward early useful responses without rewarding excessive firing. With our desired behavior defined and measurable, we finally investigate model-agnosti
benchmark - arxiv:2611.21552 · cs.LGBenchmarking memory and nonlinear processing capacity of quantum recurrent neural networks for time series predictionJosé Daniel Viqueira, Andrés Gómez, David Mera
Quantum machine learning has emerged as a promising approach to time series prediction. While quantum reservoir computing has been widely explored for this setting, variational quantum models tailored to sequential data remain comparatively understudied, largely because they are difficult to train. In this work, we present a rigorous benchmark of variational quantum recurrent neural networks (QRNNs) on time series prediction tasks, focusing on two key capabilities: memory and nonlinear processing. We introduce a task-driven methodology that quantifies model capacity, enabling an intuitive and fair comparison across architectures, including classical recurrent baselines. We evaluate the proposed framework on two generic benchmark tasks and on a synthetic dataset describing a chaotic system. Overall, the quantum models achieve performance comparable to their classical counterparts, but classical models scale better with the memory size within the range studied. Notably, one QRNN variant
memorybenchmark - arxiv:2611.21541 · cs.LGSciResearch-Eval: Evaluating Scientific Judgment in Autonomous Experimental AgentsBrian T. Weston, Alondra Torres Contreras, Aldair E. Gongora
We introduce SciResearch-Eval, an interactive benchmark of scientific judgment built on a standard laboratory problem: measuring a material's thermal properties by heating it and fitting a heat-conduction simulator. An agent designs its own experiments, uses a supplied simulator that may be under-resolved or physically wrong, and reports each property with a 95% interval. To our knowledge it is the first benchmark to combine agent-designed experiments, a deliberately flawed simulator, and proper scoring of calibrated intervals against exact ground truth. On a held-out split (93 tasks, 9 models, 2,511 episodes), models handle procedural traps. When the simulator is missing physics, most report wrong values while stating they are 95% sure: their ranges contain the truth in 20-62% of cases, against 94% for GPT-6 Astra and 99% for Claude Fable 5.2. At default effort the failing Claude 5.5 models rarely diagnose the missing physics (Opus in 35% of episodes, against 66% for Astra), and even
agentbenchmark - arxiv:2611.21532 · cs.CVHow Much Does Saliency Accuracy Matter for Saliency-Guided 360-degree Image Compression?Sakib Ahmed, Sidi-Mohammed Senouci, Philippe Brunet, Guillaume Caron +1
Saliency-guided coding allocates bits to the regions viewers are most likely to attend, but accurate 360-degree saliency prediction is computationally expensive, an obstacle for low-latency immersive streaming applications such as robot teleoperation. We quantify how much saliency-prediction accuracy actually translates into compression gain. Building on a saliency-guided rate-distortion-optimized (RDO) learned codec, modified so that the training-time latent mask is driven by ground-truth saliency, which yields a single, predictor-agnostic codec, we evaluate saliency predictors spanning two orders of magnitude in compute on both the learned codec and H.265/HEVC. The results show that higher saliency-prediction accuracy yields diminishing compression gains: a lightweight predictor fast enough for live use (52 frames per second) captures 65% of the ground-truth oracle's bit-rate saving, a 3.3x heavier one reaches 74%, and a further 38x increase in compute adds no measurable benefit, on
teleoperation - arxiv:2611.21530 · cs.CVDuoFlow: Towards Synchronized Speech and Motion Generation with Fine-Grained ControlJun Dong, Ronghui Li, Zhongyuan Hu, Yunqi Zhao +5
An embodied speaker's expressiveness depends not only on what is said, but also on how the voice sounds and how the body moves. However, existing joint speech-motion generation models are largely driven implicitly by transcript semantics or intermediate speech representations, offering little explicit control over motion hierarchy, body parts, or vocal style. To bridge this gap, we construct MotionSpeech, a structured benchmark for controllable joint speech-motion generation that includes transcripts, voice-style descriptions, and motion conditions at two hierarchical levels. Building on this benchmark, we propose DuoFlow, a dual-stream latent flow transformer that jointly denoises speech and motion under a shared noise level and a unified second-based temporal coordinate. Modality-specific residual connectors inject structured conditions into the corresponding streams, layer-wise bidirectional cross-modal attention coordinates their generation, and a monotonic phoneme pathway combines
embodiedbenchmark - arxiv:2611.21524 · cs.LGAlgorithm Distillation for Safe In-Context Reinforcement LearningMinjae Kwon, Amir Moeini, Shangtong Zhang, Lu Feng
Safe in-context reinforcement learning (Safe ICRL) asks a policy with fixed parameters to improve on an unseen constrained task from its interaction history while aiming to keep its cost within a budget. Prior work studied Safe ICRL on five benchmarks in three families: a grid world with a hidden goal and obstacles (DarkRoom), continuous navigation where goals and hazards stay invisible until reached (PointGoal, CarGoal), and locomotion with a hidden velocity limit (HalfCheetah, Ant). Safe Algorithm Distillation (Safe AD), trained on the learning histories of a safe RL algorithm, has shown adaptation failures in prior evaluations. For example, in the locomotion tasks its return stays below zero, and it rarely reaches the goal in CarGoal. We revisit two parts of Safe AD. First, we distill a budget-aware teacher with two phases. It searches for the hidden task, exploring more when more budget remains. After its first reward or cost, an expert policy exploits the task within the remaining
benchmark - arxiv:2611.21520 · cs.AISafe Resemblance, Silent Propagation: Limits of Text-Similarity Guardrails for AI Agent SafetyCharles Xiao
When AI agents relay information to one another, an output can preserve an injected instruction's effect on downstream agents even after its wording changes, yet legitimate quotation and template reuse can produce high textual overlap with the source, so text-similarity guardrails that attempt to flag the former risk conflating these two cases. We test this distinction using DonkeyRail, a public classifier combining BLEU, ROUGE-L, and METEOR features, deployed in a controlled three-agent email relay, defining a behavioral measure in which an output counts as functional (the injected procedure induces fresh computations in two subsequent agents) independent of the detector's score, alongside legitimate controls that explicitly authorize quotation and reuse. At its default threshold the classifier misses 8 of 19 functional outputs while flagging 70 of 192 legitimate replies, with every completed quotation and template-reuse reply triggering a false alert; calibrating for low false positi
agentai agent - arxiv:2611.21517 · cs.ROLazy-MOSAIC: Accelerating Skill-Centric Manipulation Planning with Learned Skill-Effect World ModelsItamar Mishani, Mukund Krishnakumar, Oren Salzman, Maxim Likhachev
Long-horizon manipulation tasks require a robot to perform a sequence of physical interactions with its environment, typically combining several different skills. Planning such behaviors efficiently is challenging: given a library of parameterized skills, solutions lie in a vast hybrid space of skill sequences and their parameters, and evaluating each candidate skill's feasibility and effect with a high-fidelity world model such as a physics simulator is computationally expensive. MOSAIC, a recent skill-centric approach, lets the skills guide a non-sequential search toward the regions of this space where they are likely to succeed; however, it pays the full cost of simulation for every parameterized skill it considers. In this work, we introduce Lazy-MOSAIC, a lazy planning framework that instead searches over hypothesized skill outcomes. A cheap learned skill-effect world model predicts, in a fraction of a millisecond, a skill's feasibility and the world state it produces. Lazy-MOSAIC
manipulationworld model - arxiv:2611.21513 · eess.SYA Scalable Decentralized Event-Triggered Topology-Switching Protocol for Verifiable Safety in Multi-Agent SystemsAmber He Wei, Shiyu Cheng, Andrew Clark, Bruno Sinopoli
This paper develops a decentralized event-triggered topology-switching protocol that allows agents to disconnect from the communication network while preserving state safety and excluding Zeno behavior. We establish a robust barrier function condition that guarantees safety as long as the estimation error remains within a prescribed threshold. Each agent uses locally computable certificates to determine when to connect and disconnect from the network. The certificate construction improves scalability by reducing the complexity of topology-dependent computations from $O(2^N)$ to $O(N)$, where $N$ is the number of agents. We formally prove that the protocol ensures the estimation error does not exceed the prescribed threshold and is non-Zeno. An alternating method computes an error-weighting matrix and a quadratic barrier function that guarantee safety under the proposed protocol. Finally, we evaluate the protocol on four interconnected microgrids. Compared with maintaining network conne
agentmulti-agentagent system - arxiv:2611.21507 · cs.ROGHySSeg: Real-Time Geometric Ground and Road Segmentation for Hybrid Solid-State LiDARMartín Palos, Fernando García
Hybrid solid-state LiDAR sensors are making their way into production vehicles, yet the data they produce differs structurally from that of the established spinning mechanical sensors behind today's autonomous-driving datasets: it is ring-free, dense, cluttered, and concentrated in a frontal field of view. The classical geometric methods that autonomous vehicles rely on for drivable space computation and navigation assume the ordered ring-and-column structure of a 360-degree spinning LiDAR and degrade or fail on this data, while learned alternatives are computationally heavy, hard to certify, and trained almost exclusively on spinning-LiDAR data. We demonstrate that the same dense, per-curve geometry is instead an opportunity to close the divide between the industry's preferred sensors and autonomous-driving research. We present GHySSeg, a purely geometric method that segments ground and drivable road from a single hybrid solid-state LiDAR frame by tracing each scan curve and reading t
benchmark - arxiv:2611.21497 · cs.AIThe CYBER-BRIDGE project: An AI-Powered Platform for Cross-Border Cybersecurity, Compliance and Forensics in the EUNikolaos Kekatos, Georgios Koutidis, Charalampos Bratsas, Stefan Bucur +11
Cross-border cyber attacks on European SMEs and critical-infrastructure operators expose a structural cyber-resilience gap: cybersecurity (threat sharing, detection, response), compliance (framework assessment, certification) and digital forensics (cross-jurisdictional evidence) are jointly regulated by NIS2, the CRA, GDPR and the LED, yet their tooling evolved in silos and rarely interoperates across Member States. The EU-funded CYBER-BRIDGE project integrates the three domains under one architecture and a hash-chained evidence layer: AI-driven CTI sharing with ENISA-aligned notification, multi-framework assessments with retrieval-augmented advisory, and RFC 3161-anchored custody with lawful GDPR/LED exchange, all aligned with MITRE ATT&CK. A sanitised replica of the back-end image hosts a Cyber-Range for upskilling. The contribution is not the individual tools but the reconciliation mechanism that lets one incident produce, from a single evidence record, a cross-border CTI advisory,
retrieval-augmented - arxiv:2611.21485 · cs.CVLost in the Hallway: Benchmarking Persistent Spatial Memory Across RevisitsHenry Zheng, Weibo Gong, Yang Yue, Rui Huang +2
Persistent spatial understanding is essential for embodied agents to operate reliably in environments they repeatedly encounter. It requires recognizing revisited places despite viewpoint changes, distinguishing them from similar-looking rooms, and updating memories as environments change. Yet existing evaluations often leave physical place identity across separated visits implicit. We introduce ReSpaceBench, a benchmark comprising 3,244 questions across 590 synthetic and real-home scenes, organized into 9 tasks spanning room association, persistent state tracking, and downstream reasoning. Controlled walkthroughs define visits by physical room entry and deliberately include same-type room distractors and cross-session object changes. Our evaluation of 32 models and 3 agentic video-understanding frameworks reveals substantial limitations: the strongest evaluated zero-shot model achieves 50.5% overall accuracy and 52.2% on a 450-question subset, compared with 90.1% for human annotators.
embodiedmemorypersistent stateagentagenticembodied agent - arxiv:2611.21480 · cs.AIRESILAGENT: A Three-Pillar Architecture for Cyber-Physical Resilience in Electrical Power and Energy SystemsNikolaos Kekatos, Georgios Koutidis, Konstantinos Antonakopoulos, Stylianos Basagiannis +5
Electrical Power and Energy Systems face hybrid threats that combine cyber intrusions, physical sabotage and climate-driven stress, and routinely outpace today's siloed monitoring stacks. We present RESILAGENT, a concept architecture organised around three operational pillars, Sensing & Data Intelligence, Resilience Analytics & Reporting, and Incident Response & Recovery, mediated by a Cloud-based Orchestrator and exercised through a closed Detect-Simulate-Decide-Sign loop. The framework fuses agentic artificial intelligence and multimodal large language models for cross-modal detection, real-time digital twins for what-if simulation, and a permissioned blockchain anchor for tamper-evident threat intelligence and CER/NIS2-aligned evidence. We instantiate it on two scenarios, a transmission operator under coordinated cyber compromise, sabotage and heatwave, and a distribution operator under a supply-chain compromise during a storm, and report an IEEE-bus micro-benchmark that quantifies
agenticbenchmark - arxiv:2611.21479 · cs.CVLynnReal-Hand: Unlocking World Interaction with Hand ControlShaohao Rui, Chaoyang Zhang, Xiaofeng Mao, Weijie Ma
Interactive video world models enable exploration through text and camera controls, yet direct object interaction requires finer control. Hands provide a natural interface, but translating continuous hand movements into coherent visual responses remains challenging because of complex hand-object dynamics and demanding real-time computation. To address these challenges, we introduce LynnReal-Hand, a hand-controlled video world model that unifies precise interaction modeling with efficient streaming generation. We curate 100K interaction clips spanning diverse real and synthetic environments and develop hand-aware supervision to improve fine-grained hand motions and object interactions. Building on this foundation, we learn a causal flow map for few-step generation and introduce region-aware score distillation that combines complementary video and image priors to refine interaction dynamics and visual appearance. To further reduce inference latency, we propose Ultra-Fast-VAE, a compact v
world modelmemory - arxiv:2611.21478 · cs.CLJetriever: Shifting Agent Memory Retrieval from Similarity to UtilityRuizhe Li, Licheng Zhang, Mingxuan Du, Zhendong Mao
A useful memory need not resemble the question it helps answer. Jetriever ranks each stored memory by Jev's decision score for its usefulness to the current question, replacing similarity search with full-pool utility scoring. On InMind's 125 indirect-query tasks, it places the gold exchange first in 84.8% of cases; matched BM25 and MiniLM controls reach 1.6% and 0.0% when indexing user turns. A 20-question comparison separates finding evidence from ranking it: Jev and Qwen3.5 9B each find 90% of gold exchanges by rank 50, while Laya finds 30%; at rank one, Jev reaches 85% and Qwen3.5 9B 40%. On a dimension-proportional random sample of 60 MemoryAgentBench questions, Jetriever's four-dimension score is 55.4%, against 44.3% for ReFind on the same questions. Jev retrieval costs USD 0.00834 per InMind question and USD 0.03103 per MAB question on average. These results show the value of world knowledge, scores that rank decisive memories first, and affordable full-pool scoring.
memoryagent memoryagent - arxiv:2611.21473 · cs.LGNetwork Intrusion Detection Adaptability under Spurious Labels: A Noise-Aware Few-Shot Class-Incremental Learning FrameworkZijian Zhao, Davide Di Monda, Domenico Ciuonzo, Yipeng Wang +1
Network Intrusion Detection Systems (NIDSs) must incrementally adapt to emerging attacks, even when only a limited amount of labeled network traffic is available. Consequently, Few-Shot Class-Incremental Learning (FSCIL)-based NIDSs have gained growing attention. However, existing FSCIL-based NIDSs are highly vulnerable to label noise in newly-observed attack classes. To address this limitation, we propose NAFN, a noise-aware FSCIL-based NIDS framework that reduces the impact of mislabeled data during incremental adaptation. NAFN consists of two components: (i) Noise Detection, which identifies and removes mislabeled biflows using noise-detection algorithms, and (ii) Feature Sampling, which estimates class-conditional embeddings from clean biflows, calibrates them with base-class statistics, and resamples synthetic features to recover the original biflow count. The framework follows the Rethink Few-Shot (RFS) teacher-student paradigm with lightweight nearest-neighbor or logistic regres
memory - arxiv:2611.21465 · cs.LGWarm Starting Neural Dynamics Models From Few Jacobian SamplesMaxime Rochkoulets, Joppe De Jonghe, Mariya Ishteva
Model-based reinforcement learning (MBRL) leverages a model of the environment, used for planning and generating samples without interacting with the real world. Motivated by evidence that informed initialization significantly improves the efficiency of MBRL, we propose TDec-Init, an algorithm to warm-start neural dynamics models using transition Jacobians. Using a decoupling algorithm based on the decomposition of a tensor of stacked Jacobian samples, we recover weights and flexible nonlinearities, and use them to initialize a feedforward neural network. In the low-data regime, we show on several benchmark environments that our approach outperforms random initialization and value training, and either matches or outperforms Sobolev training, while requiring fewer training steps between episodes.
benchmark - arxiv:2611.21459 · cs.CVBalanced Twin Weight Factorization for Efficient MLLMsBryan A. Plummer, Shan Lu, Keval Doshi, Noranart Vesdapunt +3
Weight Factorization (WF) methods aim to reduce a model's latency, memory, or both. These typically decompose a pretrained model into two or more factors that approximate the original model weights with fewer parameters and floating point operations (FLOPs). However, translating fewer FLOPs into reduced latency is challenging for WF methods in practice due to overhead introduced by replacing a single large weight matrix with multiple smaller ones. To address this, we propose Balanced Twin Weight Factorization (BTWF), which uses the unique structure of factorized models to simultaneously lower latency and increase learning capacity. Specifically, BTWF operates in the shared basis WF setting, where two or more layers are decomposed into a shared basis matrix and a layer-specific mixing matrix. If these shared layers have the same input (i.e., are twins), then we need only perform this shared forward pass once, reducing inference time. However, twin layers are often of varying sizes, and
benchmark - arxiv:2611.21457 · cs.CVOut of Sight, Into Memory: Compact Visual Latents Retain Local Evidence in Ultra-High-Resolution Remote SensingMeng Tang, Siyuan Du, Yuxin Liu, Shuaishuai Cao +2
Question answering over ultra-high-resolution (UHR) remote sensing scenes poses a tension for multimodal large language models: global overviews obscure small objects and local attributes, whereas detailed crops consume substantial context. Compressing inspected crops into short sequences of continuous visual latents offers a compact memory once crop tokens are discarded, yet diagnostic studies on natural images find that models often ignore latent content, leaving unclear whether compact latents can retain fine-grained targets lost in downsampled UHR remote sensing scenes. To isolate evidence retention from region selection, we encode high-resolution crops into latent memory tokens for frozen backbones and evaluate both their predictive utility and content dependence. Across two generations of Qwen-VL on LRS-GRO, a 64-token memory achieves 7.6- to 9.9-fold compression of retained crop tokens and improves accuracy by 2.5 to 4.8 percentage points over the downsampled global image alone,
memory - arxiv:2611.21456 · cs.AIExpertise capture in workplace AIAndreas Ekelhart, Ahmad Mohsin, Iqbal H. Sarker, Helge Janicke
AI assistants are becoming part of how knowledge workers think and perform, accumulating the prompts, exemplars, memory and interaction histories that encode a worker's professional judgment. This infrastructure typically runs under employer or platform control. We name the resulting process expertise capture, the absorption of worker-applied knowledge and judgment into AI resources under organizational control. We develop a framework of contribution forms, technical residences and transformation families, and show that how far a contribution can be attributed, separated and revoked depends on where it resides and how it has been transformed, properties fixed by system design long before any dispute arises. When a worker departs, the organization retains a capability shaped by their judgment while the worker loses access to it, and the asymmetry compounds as contributions move from inspectable artifacts into derivatives, learned components and successor systems. We outline three measur
memory - arxiv:2611.21446 · eess.SYRobust Action Clusters and Cascades in Opinion Dynamics with Heterogeneous Action ThresholdsHassan Munif, Anthony Couthures, Vineeth Satheeskumar Varma, Samson Lasaulce +1
We study the preservation and propagation of public actions in networks whose agents hold continuous private opinions. Agents update their opinions from their neighbors' observed actions and express them through action rules whose thresholds differ across agents. We characterize groups of agents that preserve a given common action regardless of the outsiders' actions. We call such a group, paired with that action, a robust action cluster. We show that a group is a robust action cluster for that action if and only if each member places enough of its influence weight on the group, where the required influence depends on the member's own action rule. We further give a sufficient condition under which an agent outside the cluster permanently adopts the cluster's action, and we iterate it to obtain a cascade of agents that are guaranteed to adopt that action. Finally, numerical examples show that replacing heterogeneous thresholds by their mean can change both the propagation of an action a
agent - arxiv:2611.21441 · cs.AILearning to Reason with Predictive Gains for Generative RecommendationZhuoyang Li, Teng Shi, Jiakai Tang, Weijie Yu +9
Reasoning-enhanced generative recommendation (GR) has attracted increasing attention for leveraging the reasoning capabilities of large language models (LLMs) to improve next-item prediction. However, existing methods primarily optimize reasoning through Group Relative Policy Optimization (GRPO) with rule-based rewards determined by final recommendation performance, which faces two key limitations: (i) unmeasured predictive contributions within reasoning, where the predictive contributions of individual reasoning steps are not directly reflected in optimization signals, and (ii) weakly discriminative learning signals within groups, where different reasoning trajectories within the group often receive indistinguishable rewards, limiting effective policy optimization. In this paper, we propose GainReason, a simple yet effective training framework that optimizes reasoning through Predictive Gain (PG). Specifically, PG measures changes in target Semantic ID (SID) log-likelihood induced by
benchmark - arxiv:2611.21440 · cs.CV[CLS] and [EOS] Tokens are not Enough for CLIP-based Class-Incremental LearningHao Sun, Tao Hu, Shixiong Kai, Siyuan Xu +3
Class-Incremental Learning (CIL) aims to empower models to continuously acquire new knowledge while mitigating catastrophic forgetting. With the advent of vision-language models like CLIP, leveraging their cross-modal alignment capabilities to address CIL challenges has emerged as a dominant paradigm. Existing methods leverage CLIP's global representations (i.e., the visual [CLS] token and the textual [EOS] token) for cross-modal alignment. While achieving effective classification, they overlook the rich patch-level visual features and token-level textual semantics inherent in CLIP. For example, the [CLS] token aggregates local information to capture the general concept of "cat", while the patch-level details retain specific localized features essential for recognition, such as the tail, fur, and face. We propose PACE (PAtch-token Cross-modal Enhancement) to uncover the long-neglected local representations within CLIP. First, based on the global representations like [CLS] token and [EO
benchmark - arxiv:2611.21439 · cs.LGA Reinforcement Learning Method to Navigate SMEFT Operator SpaceJacky Kumar
In particle physics, discrepancies between experimental measurements and Standard Model (SM) predictions are commonly interpreted within new theoretical models. In this context, the Standard Model Effective Field Theory (SMEFT) provides a general framework encompassing a broad class of models. However, exploring the enormous space of possible SMEFT operators rapidly becomes a challenging combinatorial problem: many operators can contribute simultaneously to different observables, their effects are correlated, and distinct operator combinations can provide comparably good descriptions of the data. In this work, we formulate the exploration of the SMEFT operator space as a sequential decision-making problem using reinforcement learning (RL). Our RL agent employs a lightweight actor-critic architecture with approximately 40K trainable parameters in total. Rather than searching for a single best-fit theory, our method is designed to discover a diverse set of high-quality theory hypotheses.
agent - arxiv:2611.21421 · cs.CLAuXSpec: Learning to Autoformalize Program Specifications via Executable PostconditionsHaocheng Wang, Weixuan Yuan, Florian Hübler, Yinya Huang +3
Formal specifications define intended program behavior and underpin verification, but writing them in theorem provers such as Lean is slow and expertise-intensive, making specification a major bottleneck to scaling formally correct code. Autoformalization from natural-language problems is promising, yet training specification models requires scalable feedback on whether generated specifications capture intended behavior. Prior approaches obtain this feedback through explicit proofs or separately implemented code with formal correspondence proofs, both expensive. We introduce AuXSpec, a Lean framework that pairs declarative postconditions with executable counterparts, obtained through type class resolution when possible and otherwise defined separately with an equivalence proof. Test cases can then evaluate declarative specifications by execution, letting us use agreement as a reinforcement-learning reward. Using AuXSpec, we synthesize 3,037 LeetCode problems and expand training to 19.5
benchmark - arxiv:2611.21412 · cs.CVFlexible Multi-Stack Super-Resolution of Clinical Brain MRI Using Hetero-Orientation Variational Encoding: Expert Reader Evaluation in a Nigerian CohortKabir Hamzah Muhammad, Lorena Garcia-Foncillas Macias, Soumya Snigdha Kundu, Emily Guilhem +9
High-resolution brain MRI is important for neuroradiological assessment, yet routine clinical protocols often rely on fast, anisotropic 2D acquisitions, particularly in resource-constrained settings. These acquisitions may differ in orientation, alignment and anatomical coverage, resulting in partially overlapping image stacks. Existing learning-based super-resolution methods typically rely on single, under-constrained inputs or fixed multi-stack configurations with idealised acquisition geometry. We introduce MS-HVED, a sequence-agnostic framework for flexible multi-stack brain MRI super-resolution. Available axial, coronal and sagittal stacks are independently encoded into multi-scale Gaussian posteriors and fused using a field-of-view-aware Product-of-Gaussians formulation that limits contributions from unsupported regions. Physics-informed domain randomisation simulates heterogeneous clinical acquisition conditions, including obliqueness, partial coverage and imaging artefacts. MS-
benchmark - arxiv:2611.21407 · cs.AIHarness Leverage Across Model Size: A Ladder of Five Local Agent Harnesses from 0.75B to 4.7BHao Wang, Ting Huang
A harness turns a model into a system that completes tasks: it builds the prompt, exposes tools and decides when work is done. Harness comparisons hold the model fixed. Model comparisons hold the harness fixed. We vary both: five shipped harnesses, 18 artifact-producing tasks, four locally deployable open-weight models from 0.75B to 4.7B and a 35.5B anchor, with one deterministic checker scoring all 360 ladder cells. The completion-controlling harness, Mingbird, has the highest mean at every size. Of four pre-registered paired contrasts, two survive Holm correction, at 0.75B and 4.7B. Thirteen of sixteen paired intervals exclude zero under the deployment reading, with leads from 0.075 over the strongest rival to 0.819 over the weakest. The completion gate is not the carrier: disabling all six executable checks moves the mean by +0.080, +0.117, +0.003 and -0.004, every interval including zero, and only at 1.9B does disabling it create empty runs. Removing one repeat-blocking rule alone
agent - arxiv:2611.21404 · cs.LGEfficient Modeling of Multi-Agent Systems via Structure-Aware Physics-Informed LearningYulun Wu, Nicola Bastianello, Karl Henrik Johansson, Matthieu Barreau
Modeling large multi-agent systems poses challenges for trajectory inference and parameter identification, as network size and coupling increase computational demands and partial observations require reconstructing unobserved states. We propose MAS-PINN, a neural-network architecture with a tailored training algorithm for a physics-informed learning framework that leverages the network topology and the agents' similarity to reduce computational and memory costs of simulating and identifying multi-agent systems. Its architecture represents agent trajectories through shared temporal features and agent-specific coefficients. Training alternates between learning features, fitting coupled trajectories, and estimating physical parameters. A matrix-free update exploits local interactions to fit trajectories without forming large global matrices. Experiments on Lorenz rings, Lotka--Volterra networks, and Rössler oscillators show improved trajectory and parameter accuracy over baselines, includ
memoryagentmulti-agentagent system - arxiv:2611.21399 · cs.LGMegNIST, A Benchmark for Non-Invasive Inner Speech DecodingTeyun Kwon, Pratik Somaiya, Gereon Elvers, Francesco Mantegna +4
MegNIST is a benchmark dataset for developing and evaluating machine learning methods for inner speech decoding from magnetoencephalography (MEG). The dataset comprises 12,000 trials of a single participant imagining speaking the digits zero through nine, recorded across four sessions. Each trial contains 306-channel MEG recordings sampled at 250 Hz, preprocessed using signal space separation, filtering, and downsampling. Data are provided as pre-defined training (10,000 trials), validation (1,000 trials), and test (1,000 trials) splits, with entire experimental runs assigned to each split to preserve temporal independence. The dataset is available in both BIDS format for neuroimaging analysis and as serialised tensors for machine learning workflows. Baseline classification results demonstrate that imagined digits can be decoded significantly above chance. Standardised splits, validated baselines, example code, and a PyTorch-compatible loader in the pnpl Python library enable straightf
benchmark - arxiv:2611.21392 · cs.CVVideoNexus: Video Perception Beyond the Reasoning ContextYancheng Long, Xuanyu Zheng, Wei Chen, Luoyuan Zhang +3
Understanding long videos requires locating evidence, resolving visual details, and integrating observations into an answer. Even when relevant evidence is present, performing these operations within long, dense visual sequences burdens both perception and reasoning. Video agents enable selective inspection, but returning frames to the planner still accumulates visual information in its reasoning context. Delegating perception reduces this burden, yet sparse sampling can miss decisive details, and per-frame descriptions can preserve context growth. We introduce VideoNexus, a training-free framework that separates detailed video perception from evidence-driven planning and reasoning. The planner controls where to inspect, what to resolve, and how densely to sample. Perception sub-agents process native audio-video in independent contexts and return compact, temporally grounded evidence for iterative search and verification. Using Gemini 3.8 Flash for both planning and perception, VideoNe
agentic - arxiv:2611.21383 · cs.CVThe Mask Is Not the Object: Volumetric Supervision for 3D Gaussian SegmentationJeonghwan Cho, Minsu Kim, Junyoung Hong, Seon Joo Kim
3D Gaussian Splatting enables object-level operations on Gaussian primitives, yet existing segmentation methods are largely trained and evaluated through rendered views. This mismatch can produce accurate rendered masks from incomplete 3D selections since alpha-blended feature supervision concentrates learning on dominant contributors while leaving occluded or low-contribution Gaussians weakly supervised. We propose VODA, a plug-and-play supervision objective that exploits local depth structure to group per-pixel contributors and directly supervises all Gaussians in selected depth clusters, bypassing contribution-weighted gradient delivery. To evaluate volumetric completeness, we introduce Removal-3D, a synthetic object-removal benchmark, and the Removal Depth Error (RDE), a metric comparing predicted and ground-truth depth changes. Across four rendering-based baselines, VODA consistently improves volumetric segmentation on Removal-3D while maintaining or improving performance on LERF
benchmark - arxiv:2611.21382 · cs.AIVisualArtifactSpec 1.0: Evaluating Visual Artifacts under Task SpecificationsShuyi Ouyang, Xiaojun Tang, Zijian Zou, Xintong Zhang +3
Visual artifact tasks test agents' ability to translate task specifications and source materials into outputs that meet practical needs. Evaluation across artifact families helps distinguish broadly applicable authoring capabilities from strengths confined to a particular medium. We introduce VisualArtifactSpec 1.0 (VAS), a benchmark of 211 creation and modification tasks spanning presentations, static graphics, animations and videos, and web interfaces. Each task specifies authoring objectives, source materials, constraints, and deliverables. Existing benchmarks provide limited support for systematically evaluating how authoring choices suit different audiences, purposes, and media across artifact families. VAS evaluates two aspects separately: Task Fulfillment verifies explicit requirements and applicable functionality, while Artifact Performance assesses Content, Organization, Visuals, and Delivery. Task-specific rating anchors instantiate shared criteria for the intended audience,
benchmark - arxiv:2611.21381 · physics.app-phMultilevel AlScN Ferroelectric Diode Crossbar Arrays with Sub-200 nm Cells for Selector-Free Nonvolatile MemoryKwan-Ho Kim, Zekun Hu, Krishna Karthikeya Chemudupati, Taarana Jammula +7
Passive crossbar arrays are a leading route to high-density nonvolatile memory and in-memory computing, but without a selector the memory element itself must supply the nonlinearity that suppresses sneak-path currents. Here we scale aluminum scandium nitride (AlScN) ferroelectric diodes (FeDs) in both dimensions that matter - the ferroelectric is thinned to 10 nm and the junction diameter is reduced to 100 nm. We further integrate 200 nm cells into a 2-kilobyte (128 * 128) passive crossbar. At a junction area of 0.0314 um2 this is a ~625 reduction in cell area, at equal array capacity, relative to kilobyte-scale AlScN FeD crossbars built on 5 μm cells. The 10 nm Al0.64Sc0.36N layer reverses its polarization under pulses of 50 ns full width at half maximum at a pulse-energy density of ~16 pJ μm-2, corresponding to ~0.5 pJ per 200 nm cell. Pulsed programming at 17.5 V and 1 kHz yields an ON/OFF ratio of 4.2 * 103 in a 200 nm cell, while DC characterization of 283 randomly addressed cells
memory - arxiv:2611.21376 · cs.CLA Crowd-Audited Atlas of Implicit Bias in LLM-Powered AgentsJason Alexander, Yuxuan Li, Michael Licamele, Wesley Hanwen Deng +1
Prior work has shown that large language models (LLMs) express implicit biases in high-publicity, one-off decision scenarios such as hiring and intelligence assessment. However, as people increasingly employ autonomous, LLM-powered AI agents that make decisions on their behalf in doing day-to-day tasks, it is of critical importance that we understand how implicit biases, based on sociodemographic assumptions, emerge across a broad milieu of workaday decision scenarios. We used the 17 domains of daily activity in the American Time Use Survey (ATUS) to generate an atlas of 506 decision scenarios in which LLMs --- endowed with personas representing different races, gender identities, and political ideologies --- expressed significant implicit bias. Next, in a crowdsourced audit, 500 participants made 8,010 binary choices about which implicitly biased decision scenarios most deserved further review. We found that participants prioritized decision scenarios that had broad relevance (e.g., a
ai agent - arxiv:2611.21373 · cs.CVWeVisForensics: Self-Evolving Agent for Omni AI-Generated Video DetectionYipeng Du, Yuxi Li, Ying Zhang, Chong Sun +3
Recent advances in video generation call for reliable detection grounded in explicit forensic evidence. However, current methods primarily depend on single-pass chain-of-thought reasoning that overlooks brief or localized artifacts, or on predefined multi-agent workflows that offer limited flexibility in selecting relevant inspections. Even with adaptive investigation, optimizing the investigation policy alone does not directly improve the forensic capabilities of fixed experts. We introduce WeVisForensics, an agentic framework in which a main investigator agent coordinates a self-evolving library of forensic inspector sub-agents spanning visual and audio cues. The investigator learns which artifact type to examine, over which window, and when to conclude, and these interactions provide shared feedback for both levels of agents. Our dual-timescale policy optimization strategy, DTPO, closes this loop through fast reinforcement learning of the investigator and slower refinement of the in
agentmulti-agentagenticself-evolvingbenchmark - arxiv:2611.21372 · cs.LGLearning What Pairs Miss for Higher-Order Multimodal AlignmentKunyu Zhang, Shenghui Feng, Xiangjie Shi, Zeqing Wang +1
Contrastive learning beyond two modalities now scores the whole tuple with multilinear products, volumes, and fusion heads, critics that can represent the joint distribution. In practice these critics still behave like bags of pairs. Their negatives break at least two of the three pairs, a single pairwise check rejects them, and when most samples can be solved by pairs the higher-order term receives almost no gradient. To solve this problem, we propose CHORD (Contrastive Higher-Order Residuals with Decoys), which trains a residual against decoys drawn from a calibrated pairwise model, negatives that the pairwise model finds plausible, so that rejecting them often needs all three modalities. Drawing decoys costs one inner-product query over a memory bank, and the optimal residual is the log-ratio between the data and the pairwise model, so the decoy loss also bounds what that model misses. On M3-Mix, where one in ten training samples needs all three modalities, our CHORD improves accura
memory - arxiv:2611.21366 · cs.CVALIVE: Making Streaming Video Models InteractiveJinzhao Li, Yuan Xu, Jinxuan Wu, Yukun Wang +12
Streaming video LLMs can continuously observe the world, yet often handle user requests in isolation. We introduce ALIVE, an interactive streaming video model supporting proactive monitoring, streaming task coordination, and feedback-seeking interaction. We construct ALIVE-SFT-288K and ALIVE-RL-3K for two-stage training. Contrastive supervised fine-tuning teaches when and how to respond, while reinforcement learning with Behavior-Aligned Credit Assignment (BACA) combines trajectory-level and local advantages to refine decisions, responses, and silence across multi-turn interactions. We also introduce FSI-Bench to evaluate feedback seeking and subsequent task completion. ALIVE achieves the highest overall scores among the compared models on four public benchmarks and FSI-Bench, demonstrating strong streaming understanding and interactive capabilities.
benchmark - arxiv:2611.21365 · cs.CVExploring the link between predictive uncertainty and the sim-to-real gap in deep learning models utilizing synthetic dataAlexander Moriz, Lukas Krebs, Amon Göppert, Robert H. Schmitt
Today, state-of-the-art methods for quality control (QC) are based on Machine Vision, combining deep learning (DL) models with camera-based inspection. Yet, the integration of such solutions in industrial applications is often impeded by data sparsity and trust in model reliability. While synthetic data offers a promising approach to generate the necessary data with perfect annotations, it poses risks related to performance degradation due to the possible occurrence of the sim-to-real gap. This study conducts experiments using the ensemble method to estimate the predictive uncertainty of DL models trained on synthetic data. The results show that predictive uncertainty increases under the sim-to-real domain shift. However, the observed performance drop is dominated by a systematic sim-to-real bias rather than by ensemble disagreement.These findings contribute to the effective utilization of synthetic data in industrial settings while highlighting the importance of risk assessment for th
sim-to-real - arxiv:2611.21339 · cs.CVMovieRoll: A Systematic Framework for Rolling Out Long-Form VideosYunyang Ge, Weijie Kong, Ryan Xu, Yue Wu +8
Streaming video generation must sustain quality beyond the training horizon while handling self-generated history and growing cache costs. To address these challenges, we present MovieRoll, a base-to-distilled framework that connects streaming fine-tuning, long-rollout distillation, and long-horizon inference. We first establish a robust multi-step base through Window-wise Teacher Forcing (WinTF), which jointly denoises chunks at progressive noise levels with bidirectional attention within each window and causal access to preceding history. To retain historical context efficiently, hierarchical KV-cache compression (HieraKV) exploits layer-wise attention patterns, retaining full history in shallow layers and sink--recent history in deeper layers. This reduces cache memory by up to 78% while improving generation quality. Building on this base, cumulative long tuning with distribution matching distillation (CuLT-DMD) holds parameters fixed throughout each long self-rollout and averages l
memory - arxiv:2611.21331 · cs.AITRIAGE-CPG: A Cross-Model Study of Budgeted Static Malware Triage with Large Language ModelsPaul-Andrei Sava, Alexander Küchler, Daniel Kowatsch, Bernhard Grill +3
When triaging unknown binaries, security analysts must decide which samples to clear as benign, flag as malicious, or escalate for deeper analysis. Large Language Models (LLMs) could support these decisions by interpreting decompiled code, but a binary can contain hundreds or thousands of functions, making exhaustive LLM analysis impractical. We study how current LLMs behave when static triage is formulated as budgeted evidence acquisition: which functions they inspect and how they trade off benign clearance, malware detection, and escalation. To enable a controlled comparison, we implement TRIAGE-CPG, a fixed pipeline over a code property graph (CPG) of a decompiled binary. An LLM-free heuristic selects seed functions, an agent expands the resulting evidence dossier under a fixed budget, and a reporting stage produces a benign, suspicious, or malicious verdict together with a behavioral report. We evaluate eight closed-source and open-weight models on 1,000 Windows PE binaries, using
agent - arxiv:2611.21315 · cs.AIBenchleaking Is Accelerating: Benchmark contamination has worsened sharply in the past month -- from post-training down to pre-training base modelsYiwei Qin, Muhang Xie, Xuefeng Li, Mohan Jiang +4
Benchmark contamination is not new, but recent model releases show it in a sharper form. We use the term benchleaking for the case in which a benchmark's own test examples enter training data, so that a score reflects recall rather than capability. This differs from benchmaxxing, which optimizes toward a benchmark without seeing its test set. We document two escalations over the past month. First, in post-training, the traces have become explicit: recent audits show an agent stating hidden test requirements it was never given in 23% of its attempts, where its predecessor did so in none, and continuing to follow them after the task is changed to contradict them (Schaeffer and Räuker, 2026; Lo and Lo, 2026). Second, the problem has reached pre-training. Using few-shot continuation on 20 benchmarks across four public base models, we find that some base models write out other test examples---complete questions, options, and answers---that appear nowhere in their input. For the recently rel
agentpost-trainingbenchmarkleaderboard - arxiv:2611.21313 · eess.SYRobust Optimal Network Reduction for Distribution SystemsOmid Mokhtari, James Bagrow, Mads R. Almassalkhi, Samuel Chevalier
Network reduction replaces large networks with smaller equivalents to speed up power flow and optimization studies. Optimal Kron-based Reduction of Networks (Opti-KRON) finds the optimal set of nodes to faithfully represent a large-scale radial network for representative loading scenarios. Under unseen nodal injections, however, its accuracy can drop. This paper extends Opti-KRON to uncertain nodal injections using robust uncertainty sets and a Wasserstein ambiguity set, with voltage magnitude error guarantees. To improve scalability, we develop a presolve screening step to eliminate redundant constraints and variables while preserving the feasible set. We also adopt adversarial constraint generation to reduce memory requirements by iteratively adding worst-case nodal injections. On a real 8,387-node feeder at a threshold of 0.001~p.u., the reduction removes 49.9% of the nodes. For one uncertainty set, adversarial constraint generation cuts the required memory by a factor of more than
memory - arxiv:2611.21310 · cs.LGCan AI Mature a Model's Geometry? Training-Free Weight Baking in LLMsAmin Golnari, Saeid Sanei
Can an AI make a trained language model more mature, in the sense of a better organized latent space, by looking only at the geometry of that space? An analyst model reads a fixed-size digest of topological descriptors of the layers of a frozen target and proposes a transformation, which closed-form ridge regression writes into the feed-forward weights without gradient training. Using lightweight models, 126 runs show that category purity rose on unseen words and sentences, benchmark accuracy stayed within sampling error, and the analyst matched a category-aware heuristic with gentler edits. In a five-model round robin approach, 16 pairs raised purity and four left the target unchanged. The results provide a baseline for future work on LLM-guided, training-free model editing without using raw weights.
benchmark - arxiv:2611.21309 · cs.ROSafe Physical Interaction of All-Electric Heavy-Duty Robotic Manipulators via Control Barrier Functions and Variable Impedance ControlAmir Hossein Barjini, Mohammad Bahari, Jouni Mattila
All-electric heavy-duty robotic manipulators are increasingly expected to operate close to tunnel walls, equipment, and other structures while retaining the force capability required for large-payload tasks. Accurate motion control alone does not address the resulting safety and interaction requirements. This paper develops a safety-aware physical-interaction architecture that combines variable impedance control (VIC) with control barrier functions (CBFs) on top of an established virtual decomposition control (VDC) framework for electromechanical linear actuator (EMLA)-driven manipulators. VIC continuously adapts the task-space stiffness and damping according to predicted proximity and contact risk, whereas a joint-velocity CBF filter enforces Cartesian workspace boundaries and the physical stroke limits induced by the closed-chain lift and tilt mechanisms. To account for nonideal realization of the filtered velocity by the inner VDC/EMLA dynamics, the CBF inequalities are tightened us
manipulator - arxiv:2611.21308 · cs.CVRotation-Invariant Machine Learning for Highly Symmetric Molecular ConfigurationsLogan Bolton, Tomas Karella, Nicholas Lubbers, Nathaniel Philip Gorski +4
Atomistic machine learning can approximate expensive quantum mechanical calculations at a fraction of their computational cost, but accuracy depends strongly on how the three-dimensional atomic environments are represented. Invariant representations enable fast computation but can fail to distinguish geometrically distinct configurations, especially in highly symmetric environments. We introduce a lightweight benchmark to evaluate expressivity on $n$-fold symmetric configurations, and use it to expose degeneracies in both invariant and equivariant architectures. Guided by these results, we construct a new invariant basis for HIP-HOP which resolves most of the identified failures. On a methane simulation dataset, the extended basis achieves the lowest error among the HIP-HOP variants and has similar accuracy to EquiformerV3 while running $3.2\times$ faster at inference. These results show that carefully constructed invariant representations can mitigate key expressivity limitations whil
benchmark - arxiv:2611.21307 · cs.LGProgrammable Semantic Interventions: Learning Causal Carriers from Behavioral ProgramsPaolo Muratore, Mackenzie Weygandt Mathis
Causal abstraction explains a frozen network by aligning its activations with the variables of a simpler, interpretable model, and checking that interventions on the two agree. Finding such an alignment currently requires a counterfactual dataset: for every pair of prompts, the answer the network ought to give once the variable has been changed. In practice, those labels are produced by an executable algorithm for the behavior, written before studying the model supposed to implement it. However, the criterion these methods optimize only leverages the answer an intervention should produce. Here, we develop an approach that lets the investigator define that relationship directly: the answers the variable can produce, and which of them an intervention should yield. We call this interface a semantic program and introduce Programmable Semantic Interventions (PSI): a low-rank activation subspace learned so that interchanging it changes the model's answer as the program declares, with the exp
benchmark - arxiv:2611.21303 · cs.CLSyllable- and Word-Level Evaluation of Tibetan OCRJacob Moore, Paula Lauren
Optical character recognition (OCR) for Tibetan is evaluated predominantly by character error rate (CER). Tibetan marks syllable boundaries but does not mark word boundaries, so a word error rate requires a segmenter. This study evaluates nine recognizer configurations on typeset, woodblock and two handwritten script styles, alone and in combination, and reports syllable- and word-level error rates, to our knowledge the first for Tibetan OCR. The configurations span CTC models, TrOCR encoder-decoders with and without post-training quantization, and a vision-language model, and each is scored under CER, the syllable error rate (SER) and three segmented word error rates (SWER). CER ranks systems largely as the higher-level rates do (Kendall $τ\geq 0.833$), with one pair ordered oppositely. On the effect of joint training, CER and SER record significant changes of opposite sign in 7 of 64 comparisons. SER needs no segmenter and ranks systems as SWER does under two of the three segmenters
post-training - arxiv:2611.21302 · cs.RODéjà-Events: Multi-view 3D Geometry with Event Cameras via Sparsity-Aware Adaptive GatingArda Cinar Demirtas, Pablo Varela Bernal, Marc Pollefeys, Sunghwan Hong
Event cameras keep capturing structure where conventional cameras blur or saturate, which makes them attractive for 3D perception in fast or poorly lit scenes. Event datasets with dense 3D ground truth, however, are far smaller and less diverse than the image collections behind feed-forward 3D reconstruction models. We present Déjà-Events, an event-only model that predicts depth, camera pose and pointmaps from multiple views in a single forward pass. Rather than relying on scarce 3D labels, we distil it from Déjà View, a feed-forward RGB model that refines its tokens by repeatedly applying one shared transformer block. On paired RGB and event recordings, the event student learns to reproduce the frozen teacher's final tokens, which requires no depth or pose ground truth and trains only about 1% of its parameters. Because events are dense at edges and moving texture but nearly absent in textureless regions, we further introduce sparsity-aware adaptive gating, which adapts the shared blo
event camera - arxiv:2611.21301 · cs.CLLatent Bridge: Asking for Help Without WordsQichao Wang, Tingyu Qu, Pengcheng Wu, Yiran Zhong +4
Text-based model collaboration requires a large model to generate intermediate responses before a smaller model can use its guidance, adding large-model decoding cost to inference. We introduce POND (Prefill-Only, No-Decode), a method that uses a frozen large sender for prefill and a small receiver for all autoregressive decoding. A trainable bridge maps the sender's hidden states to additional attention keys and values, allowing the receiver to use these representations throughout generation. POND also supports periodic review, in which the sender processes the receiver's partial response and updates its guidance through additional prefill passes while preserving the receiver's existing KV cache. Across reasoning and coding benchmarks, POND improves receiver performance with senders from different model families. In particular, GLM-5.3 provides gains comparable to those of Qwen3.8-Max when paired with a Qwen receiver, supporting the effectiveness of representation transfer between ind
benchmark - arxiv:2611.21300 · cs.CLLATTE-WordSeg-Lite: Multi-Granularity Thai Word Segmentation on a CPU, with a Reproducible BenchmarkThodsaporn Chay-intr, Kobkrit Viriyayudhakorn, Thanaruk Theeramunkong
Large language models read Thai through subword tokens, without a word segmenter. Yet Thai treebanks, speech metrics and morphological resources are built on words, often at several granularities, and word boundaries resolve ambiguities that subword tokens leave open. LATTE (Lattice ATTentive Encoding) reads a text at several granularities at once, through a lattice of dictionary words, but it is a large GPU model. We distil it into LATTE-WordSeg-Lite, a CPU segmenter of 1.17M parameters that keeps the lattice as features. It outputs compounds, words and morphemes in one pass, and learns only from annotated data and text whose licences allow commercial use. To compare it with other tools, published Thai scores do not suffice, as they come from different files, splits and scorers. We therefore rebuild seven test sets from pinned sources and score twelve systems under one protocol. With the unit that fits each test set, LATTE-WordSeg-Lite scores above DeepCut on five sets, four of them f
benchmark - arxiv:2611.21299 · cs.AITuring Agent: Agentic Formal Verification for Mathematical ResearchYang Chen, Qianyi Wang, Ronghui Yang, Zhanhao Li +12
Large language models (LLMs) are rapidly advancing mathematical discovery, moving beyond benchmark-level problem solving to tackle frontier research problems. This progress is ushering mathematics into an era of proof abundance, raising a fundamental challenge: how can we ensure the correctness of the ever-growing volume of AI-generated proofs and transform them into trustworthy, machine-verified mathematical knowledge? We introduce Turing Agent, an agent harness designed to orchestrate end-to-end formal verification of research-level mathematics. Its wo-scale architecture integrates theorem formalization and proof formalization agents, empowered by structured proof decomposition, parallel verification, accumulated proof experience, and selective semantic review. Leveraging our collaborations with theoretical computer science (TCS) researchers, we focus our evaluation on TCS, a particularly demanding domain characterized by intricate quantification, complex dependencies, and stringent
agentagenticbenchmark - arxiv:2611.21298 · cs.AIAgent-Native by Design: A Programming Language for Agentic ApplicationsJayanaka Dantanarayana, Savini Kashmira, Xiaoyu Ma, Malitha +3
Building agentic applications requires combining large language model (LLM) operations, workflow structure and application state, and model-guided routing. With SDKs and agent frameworks, developers connect these components through supporting code that prepares model inputs, propagates state, and dispatches model-selected steps. Workflow changes can require coordinated updates to this code, increasing development and testing effort. Splitting execution logic between application code and framework internals also limits compiler visibility into control flow and state dependencies, hindering automatic optimization. We introduce Agent-Native Programming, which unifies these three components with shared execution semantics. We realize this approach in Jac, a Python-superset language, through "by llm()" for typed model operations and tool use, Object-Spatial Programming for workflow structure and state, and "visit ... by llm()" for model-guided routing. The routing construct derives the mode
agentagenticagent frameworktool usebenchmark - arxiv:2611.21292 · cs.CVAnachroBench: Evaluating Historical Consistency in Large Vision-Language ModelsKeunhyeung Park, Byeonggeuk Lim, Kahyeon Kim, Sunhee Jo +4
Existing Large Vision-Language Model benchmarks primarily evaluate visual understanding in general everyday scenarios, leaving historical consistency with a specified era and country largely underexplored. To address this gap, we propose AnachroBench, a controlled discriminative benchmark for evaluating historical consistency in visual scenes. It contains 1,600 image pairs from eight countries spanning the late 19th to the early 20th century, each pairing an original historical image with an anachronistic version containing a seamlessly inserted modern object. AnachroBench evaluates three aspects: Inserted Object Perception, which tests whether models detect the inserted modern object; Era-Country Consistency, which assesses whether the scene is consistent with the specified era and country; and Conditional Era-Country Consistency, which measures consistency judgments only when the inserted object is correctly perceived. A false consistency judgment occurs when a model perceives the mo
benchmark - arxiv:2611.21291 · cs.AIBenchmarking Large Language Models for Linguistic AnnotationOndřej Tichý, Anna Marklová, Barbora Bulantová, Magdalena Titlbachová +1
Assessing large language models for linguistic annotation requires task-specific evaluation under documented conditions. We present dhAIBench, a Python benchmarking and annotation tool with a web interface for exploring and sharing results. It records prompts, configurations and predictions, enabling comparisons of annotation quality, confidence calibration, repeatability and estimated cost. We demonstrate the framework and the associated methodology through 293 runs of 48 models on nine linguistic annotation tasks spanning morphology, syntax, semantics, discourse-pragmatics, spelling normalization and learner-language correction. Models achieve high agreement with reference annotations on several tasks, including those on non-standardized, historical varieties as well as requiring contextual interpretation. The results support evaluating models against specific linguistic tasks, rather than relying on general rankings. To this end, we make the current and future benchmarks publicly av
benchmark - arxiv:2611.21277 · cs.MAEven Loose Proximity Constraints Make Makespan-optimal Unlabeled Multi-agent Pathfinding HardTakahiro Suzuki, Keisuke Okumura
We study \textsc{$r$-Connected Unlabeled Multi-agent Pathfinding} ($r$CUMAPF), a variant of MAPF where interchangeable agents have a proximity range $r\in \mathbb{N}$, and agents must form a single group in a given graph $G$ through limited-range communication at all times. The problem arises in fundamental applications such as self-reconfiguration and platooning, while standard MAPF does not capture proximity constraints. For the problem of finding a makespan-optimal solution, prior work shows \NP-hardness for $r=1$ and polynomial-time solvability when $r$ is sufficiently large; the complexity for the intermediate case $2\le r$ is largely unknown. To narrow this gap, we show that finding a solution to $r$CUMAPF with makespan at most $2$ is \NP-hard for $2\le r= O(|V|^{1-\varepsilon})$, where $0<\varepsilon\le 1$ is a fixed constant and $|V|$ is the number of vertices. As a consequence, we show that the labeled analogue of $r$CUMAPF is also \NP-hard under the same condition.
multi-agent - arxiv:2611.21273 · cs.AICAE-Bench: Benchmarking AI Agents for Reusable CAE Solver DevelopmentZhiyu Zhou, Tianyou Liu, Xinping Yang, Yiyao Liu +7
We developed CAE-Bench, a benchmark to evaluate AI agents on Computer-Aided Engineering (CAE) solver development. A CAE solver is more than a PDE solver; it must be able to generalize across various inputs, such as different boundary conditions, geometries, and material properties. CAE-Bench evaluates the generalizability of AI-developed CAE solvers by providing the same task, such as a heat transfer problem, multiple cases, with each case specifying an input. The AI agent develops a solver for a fixed task using only a few public cases, and the resulting solver is then tested on held-out cases, i.e., cases it has never seen before. We evaluate 12 models across a range of reasoning settings on 80 computational tasks. Our experiments show that even if a solver developed by an AI agent performs well on the public cases, it may even get 0 score on held-out cases. For tasks such as fluid dynamics problems, even the solver developed by GPT-6 Astra performs poorly on held-out cases. A data f
agentai agentbenchmark - arxiv:2611.21267 · cs.LGConformal Expert Pools: Training-Free Probabilistic Time-Series Forecasting Beyond SeasonalityValery Manokhin
Conformal Seasonal Pools (CSP) showed that a forecaster with no training at all, built only from past values at the same point in the season and from past forecast errors, significantly outperformed DeepNPTS, Amazon's learned non-parametric forecaster. We propose Conformal Expert Pools (CEP), a training-free successor. CEP replaces the single seasonal-naive forecast by a pool of simple forecasters, builds its predictive distribution from their past forecast errors, and adapts both the pool and the errors to the series being forecast. No model is trained and no parameter is learned across series. On the CSP benchmark of 232 datasets and 24,502 forecast windows, CEP is better than CSP on 78% of datasets by CRPS, on 84% by MSIS and on 83% by Winkler score; on a typical dataset its CRPS is 14% lower and its MSIS 23% lower (both p < 1e-22, paired Wilcoxon test), and coverage of 95% intervals rises from 90.1% to 92.2%. CEP has a lower CRPS than DeepNPTS on 88% of datasets, than NPTS on 89% a
benchmark - arxiv:2611.21264 · cs.ROHybrid Optimization-Generative Motion Planning for Multiagent Affine FormationsZhonggang Li, Hongyu Zhou, Raj Thilak Rajan
Formation maneuver control enables a multiagent system to cooperatively navigate through constrained environments in collective motion. While various formation controllers exist, they rely heavily on high-quality and dynamically feasible reference trajectories provided a priori, making formation motion planning (FMP) essential. To address the computational challenges for high-dimensional joint trajectories faced by conventional constrained optimization-based solutions, this paper introduces a hybrid optimization-generative framework with a focus on affine formations, which provide the necessary spatial flexibility. By formulating the planning task as a constrained Maximum-a-Posteriori estimation, we decouple multiagent objectives to leverage pre-trained single-agent diffusion priors for dynamic feasibility, obstacle avoidance, and smoothness. We then interleave the generative process with projected gradient descent (PGD) to strictly enforce affine geometric constraints with safety guar
agent system - arxiv:2611.21259 · cs.MAEvaluating Authorization and Trust Boundaries in Agentic Orchestration FrameworksSahaya Jestus Lazer, Kshitiz Aryal, Maanak Gupta, Elisa Bertino
Agentic AI systems extend large language models with tools, memory, and multi-agent collaboration, introducing security risks at orchestration boundaries. These risks depend on how identities are represented, operations are authorized, and instructions or information propagate between components. We evaluate LangChain, LangGraph, AutoGen, and OpenClaw through twelve experiments mapped to the OWASP Top 10 for Agentic Applications using a common LLM (gpt-4o-mini). The experiments cover prompt injection, unauthorized resource access and command execution, memory poisoning, and inter-agent trust and identity attacks. Adversarial inputs induced unauthorized tool invocation, while tool chaining, stale filesystem validation, encoded instructions, and inter-agent delegation exposed weaknesses in authorization across component boundaries. Memory poisoning showed that untrusted state can influence responses, a risk distinct from conventional access-control violations. In the evaluated protected
memorymulti-agentagentic - arxiv:2611.21258 · cs.ROThinkWAM: Rethinking Latent Futures for Action RefinementHaoyun Tang, Baqiao Deng, Enjin Zhou, Songwen Tian +6
World-action models (WAMs) couple action generation with predictions of future states. An initial action chunk may be flawed. A future prediction that is not conditioned on the chunk may fail to reflect the changes that the proposed actions would induce in visual representations, limiting its usefulness for refinement before execution. We introduce ThinkWAM, a latent WAM that establishes bidirectional feedback before execution through a proposal-rethink-refine loop: the policy first proposes a complete action chunk, the proposal corrects multi-horizon future latents through a learned residual, and the corrected latents are injected into a second pass of the same action head, which generates the executable chunk. An affordance-guided observation-action memory and a near-fine, far-coarse multi-horizon prediction scheme, both motivated by predictive coding, support this loop, and all prediction stays in latent space. ThinkWAM achieves 98.70% success on LIBERO, 84.76% on LIBERO-Plus withou
action headliberorobotwinmemory - arxiv:2611.21250 · cs.AIStream of Belief: Eliciting Collective Intelligence through Latent Belief DynamicsYu Yao, Jiayi Dong, Yilun Du
In this work, we view multi-agent reasoning as a process of seeking a collective equilibrium of implicit belief dynamics and introduce Stream of Belief (SoB), where agents maintain continuous latent beliefs and iteratively refine them through communication and reflection. Our method seeks a collective belief state of a multi-agent system where further communication induces no meaningful information propagation. Through extensive experiments, we demonstrate the generalization of SoB across tasks and model families. More Broadly, our results suggest a new perspective on scaling multi-agent reasoning from belief dynamics.
multi-agentagent system - arxiv:2611.21249 · cs.LGHeteroscedastic Gaussian Process Q-learning for Sample Efficient ExperimentationMaximilian Bloor, Max Mowbray, Calvin Tsay, Ehecatl Antonio del Rio Chanona
Self-driving laboratories must learn to control a process from tens to hundreds of expensive experiments, too few for deep reinforcement learning (RL). We learn the Q-function directly with a Gaussian process (GP), which avoids learning the dynamics, and show that the GP posterior supplies the learning rate of Q-learning. If each Bellman target is treated as a noisy measurement of the Q-function, the learning rate that minimises the mean squared error is the Kalman gain, and applying it to every observed transition is equivalent to GP regression. Heteroscedastic Gaussian Process Q-learning (HGPQ) sets the noise variance of each target with a heteroscedastic GP, which learns how unreliable the targets are across the states and inputs. Targets from noisy operating conditions, or from regions where the value function has large gradients, then move the estimate less than targets from low-noise regions, so HGPQ learns quickly where the data are reliable without being misled where they are n
benchmark - arxiv:2611.21247 · eess.SYNeurosymbolic Synthesis of Contraction Metrics with Certified VerificationVictor M. Preciado
We study counterexample-guided synthesis of contraction metrics for continuous-time autonomous systems on compact domains. A learner proposes a metric; a sound verifier certifies its matrix inequality or returns diagnostic points for retraining. The neural extension pairs a learned metric factor with symbolic contraction conditions and rigorous numerical verification; main ablations use polynomial features. We give computable polynomial bounds in arbitrary dimension and outward-rounded verification for polynomial, trigonometric, and neural representations. Loop soundness is separated from conditional finite-grid termination under idealized feasibility assumptions. We analyze why scaling and fixed-margin training can leave valid metrics uncertified. The scaling construction requires a scale-closed family, which the experimental fixed-floor class generally lacks. Verified slack improves the contraction rate and quantifies model-error tolerances using certified residual and Jacobian bound
benchmark - arxiv:2611.21232 · cs.AIA Generative AI-Driven Architecture for Whole-Home Intelligent Services: Edge Models, Memory Management, and Household ApplicationsMo Zhou
Generative artificial intelligence is entering the home. Cloud-based assistants face governance limitations in data transmission, continuous availability, and the reliability of physical actions, which are especially salient in households where devices remain online, data accumulate over time, and commands trigger real-world actions. Using the home local server as a unifying organizing axis, this article develops a design-oriented review framework. Sources from arXiv, CNKI, ACL Anthology, and official documentation were searched through November 2026 and coded into five evidence categories: reviews, system prototypes, benchmarks, HCI and eldercare studies, and official documentation. System designs were mapped onto shared architectural modules, identifying nine recurrent responsibilities of the home local server. Generative models support perceptual interpretation, intent understanding, and candidate planning, while deterministic validation, knowledge-graph constraints, and authorizati
memorybenchmarkevaluation framework - arxiv:2611.21228 · physics.app-phImmersive virtual reality in learning physics: a systematic review of opportunities and challengesAndrej Vidak, Iva Movre Šapić, Vjeran Gomzi, Vanes Mešić
Immersive virtual reality (IVR) can heighten presence and enable active, embodied interaction in realistic 3D environments, which has been associated with potential benefits in physics education. This article presents a comprehensive review of IVR implementation in physics education at both school and university levels. The analysis included 34 studies indexed in Scopus and ERIC, published between 1 January 2018 and 1 June 2025. Methodologically, this review followed a preferred reporting items for systematic reviews and meta-analyses approach. IVR implementations for physics learning were assessed, topic-aligned activity types were summarized, and associated opportunities and challenges were documented. Findings indicate that IVR can support physics learning by providing complementary visualizations, optimizing cognitive load, enabling haptic learning, saving time, and fostering collaborative inquiry. Conversely, the most frequent potential IVR challenges include discomfort, unreliabl
embodied - arxiv:2611.21222 · cs.AIVerifiable Random Sampling of Private Data Without a Trusted Single PartyMuhammed Karamuk, Mikołaj Glinka, Ruben Wolff
Problem: When a party supports claims about a private dataset by disclosing a sample, auditors cannot tell whether it was drawn at random or "cherry-picked" to hide flaws in the rest of the data. Solution: The data holder commits to the hashes of every item in its dataset and registers the commitment with independent servers. Selection then waits on a decentralized randomness beacon: a network that jointly emits a fresh random value every few seconds, outside the control of the data holder and the auditors. The value emitted two beacon rounds after the commitment picks the sample. Each server returns a signed receipt, so anyone can later check that the revealed sample is exactly the one that value prescribes. Selection from the committed set stays unbiased as long as the beacon's threshold holds and one required server is honest, and a pre-agreed name, checked with the servers, stops repeated or abandoned draws. We call this protocol NoCherry. Applications: Any audit that spot-checks a
benchmark - arxiv:2611.21218 · cs.AIInclusionMed: Benchmarking AI for Inclusive HealthcareThe InclusionMed Collaboration
Advancing medical AI for inclusive healthcare requires evaluation across diverse healthcare roles, tasks, and care settings, but benchmark coverage remains uneven. We introduce InclusionMed, an open medical AI evaluation initiative for inclusive healthcare. The InclusionMed Atlas organizes benchmarks across three healthcare roles and three AI capabilities. For the first release, we selected four public benchmarks from 388 candidates through structured audits and independent reproduction. With InclusionMed-DxFiles, our private diagnostic reasoning benchmark, the Atlas contains five benchmarks covering three of its nine role--capability cells. We evaluated frontier large language models on the Atlas and aggregated their results into the InclusionMed Index, with overall scores ranging from 17.5 to 49.1 out of 100. The Index also reports scores by healthcare role and AI capability, showing that a model's rank varied with the role or capability considered. A within-benchmark analysis using
benchmark - arxiv:2611.21216 · cs.LGMiSTiQ: MicroScaling Trellis Quantization for Extreme KV Cache CompressionOfir Gordon, Lior Dikstein, Paul Balanca, Idan Achituve +2
Modern large language models (LLMs) increasingly support longer contexts and serve more concurrent requests, increasing both sequence lengths and batch sizes. Both trends increase KV-cache size and memory traffic during inference. KV-cache quantization reduces these costs, with recent approaches using codebook-based vector quantization to achieve extremely low bit widths. However, hardware constraints limit codebook size, and most methods reconstruct the cache in higher-precision floating-point formats for attention, limiting the computational benefits of quantization. To address these limitations, we introduce MiSTiQ, a drop-in replacement for codebook-based vector quantizers that combines trellis-coded storage with microscaling (MX) formats for low-precision computation. We combine trellis quantization with beam search to encode the KV cache over large reconstruction spaces without large codebooks. To support MX computation, we derive MX-Hash functions that generate reconstruction va
memory - arxiv:2611.21209 · cs.ROBoBDeLaN: Bounded-Bandwidth Context Adaptation for Deep Lagrangian Robot ControlLucas Schulze, Sucheth Shenoy, Jan Peters, Oleg Arenz
Learned physics-based models provide structure for robot control, but uncertain payloads and unmodeled dynamics limit their predictive accuracy. Context-aware Deep Lagrangian Networks enable adaptation through a latent context inferred from measurement histories while retaining Lagrangian structure for each fixed context. However, this structure alone does not constrain sensitivity to measurement disturbances. We introduce BoBDeLaN, which combines a contracting, gain-bounded Recurrent Equilibrium Network context encoder with stable linear filtering to shape disturbance sensitivity across frequencies. This architecture guarantees exponential forgetting of the recurrent initialization and admits a finite-horizon, frequency-weighted bound on context perturbations, which extends to residual torque prediction under a Lipschitz continuity assumption on the context-conditioned dynamics model. We evaluate a low-pass realization within model predictive control on simulated and real hardware man
manipulation - arxiv:2611.21208 · cs.CLValidating LLM-Based Essay Scorers: A Comparative Analysis in Critical Online Reasoning in Medical EducationDaniel Baumartz, Alexander Mehler, Mevlüt Bagci, Maxim Konca
Automated essay scorers (AES) based on large language models (LLMs) promise human-level accuracy. However, training, fine-tuning, and evaluating LLMs for specific assessment tasks is costly. In particular, the question arises whether these AES should be trained for specific domains or whether models based on generic LLMs might be sufficient. Human annotations, by contrast, are often more valid but far more expensive. This study evaluates whether ensembles of AES systems based on general-purpose LLMs can approximate human ratings of medical-education writing tasks by leveraging encoded linguistic and conceptual knowledge. The goal is to assess whether general-purpose models are applicable for assisting or potentially replacing human annotation in this domain. We experiment with 11 AES systems and 8 different LLMs and evaluate 668 student responses to 12 tasks, using \numprint{2076} evaluations by 7 human annotators. These ratings are aggregated to create consistent performance levels th
evaluator - arxiv:2611.21198 · cs.ROCooperative Mobile Manipulation using Learned Dynamical SystemsSaray Bakker, Wendelin Böhmer, Javier Alonso-Mora
Cooperative mobile manipulation requires robots to coordinate their motions while ensuring whole-body collision avoidance and preserving task-relevant behaviors. We propose CALDS (Cooperative Adaptation of Learned Dynamical Systems), a framework that enables mobile manipulators to adapt learned single-agent behaviors for coordinated execution. Building on individual dynamical systems (DSs) learned from demonstrations, CALDS enables cooperative adaptation through two key components: Cooperative Policy Selection, which identifies compatible combinations of learned policies, and Temporal Adaptation, which ensures collision-free cooperation while preserving the spatial structure and convergence properties of the original behaviors. We evaluate CALDS across three diverse scenarios: assistive dressing, restaurant cleaning and serving, and drawer opening and object placement. Quantitative results demonstrate its effectiveness in improving task success and enabling safe, coordinated simultaneo
manipulationmanipulator - arxiv:2611.21183 · cs.CLSeeing Isn't Resolving: Visual and Linguistic Disambiguation with Conversational GroundingEdoardo Cecchinato, Shun Inadumi, Angel García Contreras, Giuseppe Riccardi +1
Visual grounding benchmarks usually do not consider the contextual uncertainty of real-world scenarios. In this study, we construct a novel benchmark, the Japanese evaluation for Ambiguous visual Grounding with Dialogue (JAG-D) dataset. It consists of human-robot conversational multimodal interactions covering different types of ambiguity, arising from visual and linguistic categories. We evaluate the capabilities of different state-of-the-art Visual Language Models (VLMs) for ambiguity recognition and resolution. We compare dialogue context grounding performance for visual and linguistic ambiguity resolution. Our experiments show that VLMs can substantially improve the resolution of linguistic ambiguity, especially Japanese zero anaphora, by using dialogue context. We further conduct a comparative analysis to understand whether grounding improvements arise from the semantic content of the dialogue or from its conversational structure.
benchmark - arxiv:2611.21178 · cs.LGTRACK: Tool-Augmented Reasoning over Time Series for Fault Diagnosis in Power Management SystemErkan Bayram, Debanjan Ghosh, Li Zhu, Sambarta Dasgupta
Fault diagnosis in power management systems running multiple loads relies on experts manually inspecting current and voltage waveforms, which fails to scale across deployed units. We present TRACK, which pairs a time series encoder with a Small Language Model to produce structured, physics-grounded diagnostic reports, calling tools for exact, verifiable calculations. It operates on-device, keeping proprietary sensor data local and avoiding expensive API calls. We introduce a rubric-based framework that evaluates these reports by scoring structure, signal fidelity, physics reasoning, diagnosis, and actionability through per-criterion verdicts. On a 15-class multi-load simulated benchmark, TRACK achieves over 95% fault-category accuracy and generalizes to out-of-distribution faults, whereas the best encoder-probing baseline reaches only 58% accuracy.
benchmark - arxiv:2611.21176 · cs.AIScientific Coding with Large Language Models: Evidence from Finite Element AnalysisErfan Hamdi, Joel Shor, Emma Lejeune
Large language models (LLMs) offer a potential route to automating scientific programming, but their usefulness depends on consistently producing numerically correct code. We extend FEM-Bench to 17 finite element tasks involving nonlinear and history-dependent mechanics and structural dynamics, and evaluate 15 models on implementation and test generation without execution feedback. The best model succeeds in 87.1% of single-pass implementation trials, but rankings differ between tracks. The leading implementer achieves 49.9% joint test success, while the leading tester reaches 82.6%. For six models, we compare single-pass generation with two agentic harnesses that allow code execution and revision. Agentic execution solves some tasks that never succeeded in single-pass trials and improves consistency on others, but gains vary across models and harnesses, and regressions also occur. Claude Opus 5 with SimpleAgent reaches 96.5% Pass@3, the probability of at least one success in three att
agentic - arxiv:2611.21162 · cs.CVDRaGAN: Dynamic Graph Attention Network for Multi-Radar Point Cloud Encoding in Automotive Radar PerceptionChristof Leitgeb, Thomas Puchleitner, Max Peter Ronecker, Daniel Watzenig
Radar represents a key modality for weather-robust perception. Unlike optical sensors such as camera and LiDAR, radar can penetrate fog, rain and snow due to its longer wavelength, and recent hardware advances have steadily improved its spatial resolution, thus making it increasingly relevant for autonomous systems. While processing raw radar data requires substantial memory computation, most existing work has focused on preprocessed point-cloud data similar to LiDAR. However, the sparse radar points and sensor-relative anisotropy make LiDAR-oriented encoders suboptimal. To address this problem, we present DRaGAN, a dynamic graph attention network for multi-radar point clouds. DRaGAN introduces a sensor-relative Fourier encoding of angle and range to capture fixed-aperture radar characteristics, gated short- and long-range hub nodes along each sensor's line of sight for efficient local-global message passing, and attention-guided dynamic rewiring that prunes weak edges and adds a bound
memory - arxiv:2611.21155 · cs.AIEdit Recommendation: Can LLMs Anticipate Edit Requests from Users, and When?Daisuke Kikuta
After a large language model (LLM) agent generates an artifact, users often send edit requests for it. This paper studies whether an LLM can anticipate these requests before they are sent. Such anticipation is useful in practice because it reduces the effort of user review and feedback. We formalize this task as edit recommendation and build EditRecBench, a benchmark for evaluating the performance of LLMs on this task, from real-world conversation logs in WildChat-4.8M. Through benchmarking and failure analysis, we identify a practical challenge that some cases are inherently difficult to predict in real-world settings. To address this, we introduce decision model-based gating to filter out cases where predictions are unlikely to be correct before they are presented to the user. Our trained model outperforms existing decision models and raises pass@5 among shown suggestions by 12--25 points at a gating threshold. Our code is available at https://github.com/ntt-dkiku/llm-edit-recommenda
agentbenchmark - arxiv:2611.21152 · cs.CLFrom Repositories to Training Environments: Scaling Long-Horizon Coding Agents with Environment Evolution LoopYang Tian, Tianyu Zheng, Xiyuan Yang, Yaxin Du +4
Training coding agents for long-horizon tasks requires environments with verifiable feedback. Existing pipelines often follow a linear process that ends with validation, establishing usability but providing little guidance on improving training value. Recursive self-improvement (RSI) often requires model rollouts or model training followed by downstream evaluation, with costs and feedback delays amplified on long-horizon tasks. How can we use timely, reliable proxy signals to improve training value and synthesize environments at scale? We present RepoFactory, an automated pipeline that converts real repositories into verifiable environments where agents build complete projects from scratch given a language specification. Its Environment Evolution Loop optimizes a multiplicative objective combining validity, structural coupling, and implementation scope through proxy feedback. RepoFactory produces 10,159 environments across 17 software domains. Training Qwen3.5-35B-A3B on these environm
agenticself-improvementbenchmark - arxiv:2611.21143 · cs.ROWinDEX: Metric 3D Hand Motion Reconstruction in the Wild for Dexterous Teleoperation and Policy LearningHongyu Ma, Hairong Qu, Shiqi Zhao, Yongsong Yang +4
Human manipulation data are widely used to train robots for human-like dexterity. However, low-cost, large-scale collection in the wild remains challenging. Existing methods remain limited in geometrically grounded metric-scale reconstruction, real-world generalization and inference speed. We present WinDEX, integrating (1) geometric-temporal optimization for metric 3D hand annotation from in-the-wild egocentric stereo videos, combined with visual-inertial SLAM for world-frame reconstruction; (2) WinDEX-Net, an end-to-end, fisheye-native network combining calibrated stereo, local detail and temporal context, with a data-model flywheel of model-assisted annotation and retraining; (3) distilled models for offline reconstruction and causal streaming, supporting reinforcement learning and teleoperation; and (4) low-cost CyberCap stereo-inertial capture, forming the WinDEX-Hand dataset with 1,000 hours of training annotations and held-out human-annotated in-the-wild 2D and independent optic
manipulationdexterousteleoperation - arxiv:2611.21134 · eess.SYRobust Data-Driven Output-Feedback Reference Governor with Quantified Model UncertaintyTam W. Nguyen
This paper constructs a data-driven reference governor (DDRG) from input--output data and supplied deterministic disturbance bounds. A data-consistent model set quantifies the uncertainty remaining in each prediction direction. With exact output measurements, the DDRG propagates output-history uncertainty while retaining exact applied-reference histories and converts the resulting tubes into constraint margins. Under a common contraction condition in a selected history norm, explicit terminal feasibility inequalities certify infinite-horizon continuation under a held reference and yield a closed invariant admissible set. A scalar update selects the largest certified step toward the requested reference, with recursive feasibility and constraint satisfaction from certified initialization. A separate bounded-measurement-noise extension derives the errors-in-variables model set and the additional physical-output margin through a measured-history certificate. On the coupled benchmark, longe
benchmark - arxiv:2611.21132 · cs.CVRegion-VAD: Video Anomaly Localization and Detection via Region SelectionGiacomo D'Amicantonio
Anomalies are local events, yet weakly supervised video anomaly detection scores them globally. Most methods pool an entire frame into a single vector, rank frames by that vector, and derive a location afterwards as a saliency map of the score. Full-frame pooling mixes the anomaly with its background, so a detector can reach competitive benchmark scores from scene context without being able to show where in the scene the anomaly occurs. We introduce Region-VAD, which first selects a region and then detects the anomaly from that region alone. Region-VAD selects an explicit rectangle over the tokens of a frozen foundation model and scores anomalousness only from the tokens inside the box, through attention pooling followed by a standard temporal head. The rectangle is not learned, it is the box in which the frozen tokens change most between consecutive segments, in all four feature projections of the encoder, and it is found by exhaustive search over the rectangles of the token grid. Onl
benchmark - arxiv:2611.21129 · cs.ROLinglong Agent Service: A Runtime for Centralized Agent Reasoning and Distributed ExecutionHua Zhang, Min Jiang
LLM agents can use remote tools and cloud sandboxes, but integrating reasoning, session state, and heterogeneous execution sites remains a runtime design problem. We present Linglong Agent Service (agentsvc), a self-hosted HTTP service and embeddable Python runtime that treats execution site as a configurable dimension alongside model, tools, and context. A parent service retains inference orchestration, tool descriptions, and conversation history; each inference request selects one child environment for file, shell, terminal, and MCP operations. Direct HTTP and child-initiated WebSocket tunnels carry the same application interfaces, including portable request context and centrally managed updates. The normal core path requires no third-party Python packages, with an optional Windows pseudo-terminal dependency. We further present online_debugger as the process-scale special case of execution-site separation: an instrumented program submits exception evidence, executes session-scoped in
agentllm agent - arxiv:2611.21126 · cs.CVIGRecon4D: Dynamic-Object-Aware Feed-Forward Reconstruction of Driving ScenesYifei Tong, Ziyang Huang, Yizhao Han, Xiaoxiao Long +1
Feed-forward 4D driving-scene reconstruction enables efficient modeling of dynamic autonomous-driving scenes, but current methods suffer from two complementary failure modes: semantic leakage of traffic participants into the static representation and geometric drift of Gaussian locations across viewpoints. We propose \textbf{IGRecon4D}, a feed-forward dynamic scene reconstruction framework that addresses these failures with training-time instance-prior distillation and LiDAR-anchored Gaussian positioning. Specifically, an instance-level semantic branch provides a potentially dynamic participant prior, which is distilled into dynamic-mask learning through the Instance-Dynamic Consistency Constraint to encourage high-recall localization of traffic-participant regions prone to dynamic-static leakage. In parallel, sparse LiDAR measurements guide dense depth estimation, and the resulting depths are back-projected to anchor Gaussian positions for improved multi-view geometric consistency. Ex
benchmark - arxiv:2611.21115 · cs.ROReuseMem: One Memory, Multiple Goals in Robot ManipulationYoonseok Kang, Dayeon Jung, Seunghoon Jeon, Wonjib Kim +1
Robots that work alongside people should be able to use what they have seen when new requests arise. The same experience may later support identifying an object or returning it to its original location, although these uses were unknown during observation. ReuseMem learns to use one visual memory for multiple later manipulation goals. It records visual features independently of future goals, retrieves complementary evidence for each request, and combines it with the current scene through a lightweight action interface. Retrieval is supervised by how well candidate evidence supports expert-action prediction across different current states, linking memory selection to its use in manipulation. Repeated visual features are stored once, and memory computations that remain unchanged during action generation are reused. Experiments on four simulation tasks and two physical scenarios demonstrate reuse of a common recording across different requests. With matched action adaptation, ReuseMem impr
manipulationmemorybenchmark - arxiv:2611.21113 · eess.SYIntransigence or compromise: concatenated opinion dynamics with voting feedbackLingfei Wang, Johan Swartling Sennhed, David Ohlin, Emma Tegling
In this paper, we study opinion evolution in a sequence of Friedkin-Johnsen (FJ) models, concatenated so that the final opinions of each model serve as the initial opinions of the next. Each FJ model is followed by a vote, and each agent becomes partially or fully stubborn in the next model, depending on how close its opinion is to the voting result. With the state-dependent switching of stubbornness, the dynamics are not guaranteed to be contractive. We show that, over any fixed strongly connected network, the opinions always converge. A sufficient condition on the initial opinions is proposed, under which consensus is reached exponentially fast. For the complete graph, we further show that, if the agents' stubbornness levels do not differ too much, an agent that is partially stubborn in one FJ model can never become fully stubborn in a subsequent one: once it has started compromising, it keeps doing so.
agent - arxiv:2611.21107 · cs.CVSource-Free Domain Adaptation for Multi-Label Image Classification: A Framework, Baselines, and BenchmarkSalih Numan Büyükbaş, Erchan Aptoula, Mamatha Thota, Petra Bosilj
Source-free domain adaptation (SFDA) adapts a model trained on a labelled source domain to an unlabelled target domain without access to the source data. Almost the entire SFDA literature assumes mutually exclusive classes, yet many deployed settings are multi-label. This paper formulates SFDA for multi-label classification and establishes the first benchmark for it: five adaptation tasks over six datasets spanning remote sensing, natural, clipart-style and driving imagery, with four single-label SFDA methods extended to provide the first baselines and a published multi-label method run alongside them, over three seeds. The benchmark shows that extending a single-label method is unreliable rather than merely suboptimal: half of the extended baselines fail to improve on the unadapted source model at all, and on two of the five tasks none of them does. The failures are not incidental. Nearest-centroid pseudo-labelling, softmax-based confidence, the diversity term of information maximizat
benchmark - arxiv:2611.21105 · cs.CVCivilBench: Benchmarking Vision-Language Models for Multimodal Information Integration in Civil EngineeringDan Yuan, Yue Wen, Zhili He, Zhenkui Zhou +2
Artificial intelligence (AI), particularly large language models (LLMs) and vision-language models (VLMs), is increasingly applied in civil and structural engineering for disaster assessment, structural analysis, and engineering decision support. However, the extent to which current VLMs possess the reasoning capabilities required for reliable engineering judgments remains unclear. In this work, we introduce CivilBench, a comprehensive benchmark for evaluating VLMs in civil engineering and disaster-related scenarios. CivilBench covers three major task families: Sequential Reasoning, Image-based Structural Understanding, and Open-world Understanding, encompassing numerical reasoning, structural damage assessment, risk inference, causal analysis, and repair planning. The benchmark is constructed through a systematic pipeline of domain-specific data collection, engineering-oriented captioning, expert annotation, question design, and quality verification. Evaluation of representative state
benchmark - arxiv:2611.21104 · cs.CVGRAZE: Text-Free Grasp Synthesis for Unseen Objects from Human GazeJeongho Lee, Boeun Kim, Hyung Jin Chang, Jungchan Cho +1
Text-conditioned hand-object interaction (HOI) generation provides intuitive semantic control, but language is an indirect way to specify where an interaction should occur on an object. Naming the part to grasp requires object-specific words, and naming the approach side requires a reference frame that a prompt lacks. We propose GRAZE (GRAsp from gaZE), which generates hand-object grasp motion from recorded gaze instead of object and part names. Its Geometry-Aware Gaze Encoding (GeoGaze) maps gaze rays to cues on object geometry, and cross-attention conditions a diffusion Transformer on these cues. We also introduce HOT3D-Grasp, a grasp-centered benchmark with object-disjoint folds built on HOT3D recordings. On unseen objects, GRAZE grasps the target part more often than the strongest text-conditioned baseline (49.3% vs. 43.6%), achieves the highest grasp success (66.3% vs. 46.6%), and transfers without fine-tuning to GazeHOI (45.3% vs. 33.8%). Gaze conveys the approach side: when gaze
graspbenchmark - arxiv:2611.21102 · cs.CVBeyond Family Labels: A Taxonomic Ornstein-Uhlenbeck Prior for Avian 3D Shape RecoveryYunchan Jeon, SooGon Kim, Seungwook Kim, CheolWon Lee +1
Recovering 3D pose and shape of birds from single images is bottlenecked by the long tail of avian diversity: out of ~10,000 species, only a few dozen have 3D-labeled training data, and in-the-wild test sets routinely contain species never seen at training. Generalizing across clades is therefore a central problem. The current state of the art, AniMer+, mitigates it with a family-level contrastive loss that conveys only a binary same-family indicator per pair and discards the continuous structure of the species tree. We propose a taxonomic Ornstein-Uhlenbeck prior on the shape code of the AVES parametric bird mesh model: a multivariate Gaussian whose covariance decays exponentially with Linnaean rank distance, with a learnable selection strength and noise scale. The prior generalizes family contrastive into a continuous, generative form, exposing the graded cross- and within-family similarity that a binary indicator cannot, and brings continuous comparative phylogenetics in as a biolog
benchmark - arxiv:2611.21099 · cs.ROForceHarness: Turning Force into Action and Physics for Frozen Frontier ModelsKe Sun, Kevin Yuchen Ma, Yike Zhang, Yanzhe Chen +1
Frontier multimodal models can now control robots zero-shot from images, yet many manipulation tasks depend on contact forces that images cannot show. Bringing force to an API-only frozen model raises two challenges. First, contact evolves much faster than an API model can decide, so the system must close the loop on force beneath the model. Second, a frozen model cannot interpret raw force streams, so force must be returned in a form the model can reason about. Therefore, we present ForceHarness, a harness that lets frozen models use force without demonstrations or finetuning. After calibrating hand force from a wrist sensor or joint torques alone, ForceHarness addresses the first challenge with self-terminating force primitives that regulate force on their own and return at a contact event. To address the second challenge, a physics translation layer turns each force trace into physical quantities with uncertainty as the tool result. To evaluate ForceHarness against expert force cont
manipulationbenchmark - arxiv:2611.21095 · cs.AISWE-WhatIf: Simulating Real Users of Coding Agents from Recorded SessionTae Soo Kim, Tongshuang Wu, Juho Kim
Recorded sessions between users and coding agents offer a way to evaluate how agents complete real-world tasks and interact with users. However, these sessions cannot simply be replayed with new agents, because recorded user replies may no longer apply when an agent takes a different trajectory. We introduce SWE-WhatIf, which replays recorded sessions with new agents by modeling the user's intents, the criteria for satisfying them, and the conditions under which they arise. Using this model, a user simulator generates replies that adapt to unseen agent turns while remaining grounded in the recording. Across 89 real-world sessions, SWE-WhatIf generated the replies most similar to those of real users and matched the strongest baseline when sessions were rewritten. To test adaptation, we replay sessions with agents instructed to avoid or repeat actions that the recorded user pushed back on. SWE-WhatIf produced the highest win rate for avoiding these actions (66.3%), significantly above ch
agent - arxiv:2611.21094 · cs.AIConflict-Aware Multi-Teacher On-Policy Distillation in Aligned Action Spaces for Cross-Game GeneralizationJing Yi, Cehao Yang, Hualei Zhang, Haitao Li
Zero-shot cross-game generalization aims to consolidate accessible source-game experts into a single vision-language model (VLM) that controls unseen games from raw pixels without target-game parameter updates. While multi-teacher on-policy distillation (MOPD) mitigates offline exposure bias and single-game reinforcement learning (RL) overfitting, scaling it to autoregressive VLMs faces two bottlenecks. First, token-to-action misalignment from subword length bias distorts discrete action probabilities and uncertainty margins. Second, multi-teacher policy conflict arises as source teachers under incompatible dynamics issue contradictory closed-loop corrections, causing negative transfer. We propose CAMOPD (Conflict-Aware Multi-Teacher On-Policy Distillation), a post-training framework that resolves multi-teacher policy conflict within an aligned semantic action space. CAMOPD first uses complete-action alignment to normalize command completion likelihoods over each game's legal action se
post-training - arxiv:2611.21085 · cs.CVState-Message Interaction Adapter for Cross-Domain Few-Shot SegmentationHang Xing, Yan Xia, Yongjiang Liu, Siqi Zhang +1
Cross-domain few-shot segmentation adapts natural-image pretrained encoders to medical or industrial domains from a few pixel-annotated support masks. Within frozen DINOv3, intermediate per-head attention messages already carry target-background distinctions that baseline final-layer similarities and predictions do not consistently reflect. We propose the State-Message Interaction Adapter (SMIA) to learn how per-head messages should update each receiving token representation while keeping pretrained attention and FFN parameters frozen. Inserted right after per-head aggregation, SMIA conditions the transformation of each head's aggregated output (message) on each patch's pre-attention feature (state), allowing similar contextual messages to produce distinct or opposing updates depending on the receiving position. Because scarce support masks cover only limited state-message combinations, a shared low-rank bilinear map confines the learned interactions to low-dimensional state, message,
benchmark - arxiv:2611.21083 · cs.AIThink2Listen: Evidence-Centric Learning For Agentic Audio Understanding With LALMSJin Wang, Liang Lin, Kaiwen Luo, Weiliu Wang +6
Recent advances in large audio-language models (LALMs) have broadened audio understanding across speech, sound, and music, with tool-based agents further supporting these tasks by invoking predefined audio tools. However, a persistent gap remains between perceiving acoustic content and using it as evidence for reasoning: recognizing what is audible does not ensure that a model can identify which cues resolve an uncertain decision. To bridge this gap, we propose Think2Listen, an evidence-centric learning framework for agentic audio understanding. Guided by evidence before commitment, Think2Listen couples iterative reasoning with a targeted LALM interaction to actively seek acoustic cues that distinguish competing hypotheses and revise candidate judgments based on the observations returned. It learns what to ask, how to use evidence, and when to stop, using observation-induced changes in candidate-state to inform training credit. Extensive evaluations on two benchmarks spanning seven aud
agenticbenchmark - arxiv:2611.21079 · cs.LGSemiNovo: Learning Beyond Search-Identified Spectra for De Novo Peptide SequencingPan Liu, Dongxin Lyu, Yusen Tan, Yifan Li +5
Sensitive peptide identification is fundamental to protein characterization and novel protein discovery, for which de novo peptide sequencing (DNPS) from tandem mass spectra provides an important database-independent approach. Although deep learning has substantially advanced de novo sequencing, existing models are predominantly trained on high-confidence peptide-spectrum matches obtained through database search. This search-conditioned supervision excludes a large fraction of unidentified spectra, creating a distribution mismatch between the selectively labeled training data and the complete spectrum stream encountered at inference time, thereby limiting peptide identification performance. To address this challenge, we propose SemiNovo, a semi-supervised framework that learns jointly from identified and unlabeled spectra. SemiNovo first constructs DarkSpec, a large-scale quality-controlled corpus of unlabeled spectra collected from diverse public proteomics projects, thereby extending
benchmark - arxiv:2611.21072 · cs.AISage S2.1 Flash Technical ReportKun Cai, Haodong Chen, Guangran Cheng, Qingxiu Cheng +202
We introduce Sage S1.5 Flash, a 320-billion-parameter general-purpose mixture-of-experts model designed to reason over scientific evidence and act in interactive environments. Sage Sci-Perception uses dedicated encoders for molecular graphs and biological sequences and represents spectra as ordered patches of measurements. Modality-specific projectors map these representations into the input space of a shared backbone, which processes them alongside text and images. To develop the model's agent capabilities, we construct verifiable environments for tool use, search, and coding, as well as specialized workflows involving machine learning experiments and scientific software. Task completion is checked against execution results, changes in environment state, or submitted artifacts. We evaluate Sage S1.5 Flash on 33 benchmarks across seven domains. In the primary comparison against five baselines, it ranks first on OfficeQA Pro (72.2), MLE-bench Lite (90.3 HumanRank), Harvey LAB, and Sprea
agenttool usebenchmark - arxiv:2611.21069 · cs.ROCoopDistill: Equilibrium-Guided Multi-Agent Policy Distillation for Cooperative DrivingShiyu Fang, Sheng Zhao, Jiarui Zhang, Chengkai Xu +3
Cooperative driving extends autonomous driving from individual vehicle decisions to joint maneuvers that improve traffic safety and efficiency. Optimization-based methods impose safety constraints but often fix paths to simplify multi-vehicle planning, limiting spatial flexibility. Multi-agent reinforcement learning (MARL) explores alternatives through interaction, but coordinated learning and safe execution remain challenging. Their integration nevertheless faces challenges in exploration efficiency, experience allocation, and execution reliability. To address these challenges, this paper proposes CoopDistill, an equilibrium-guided multi-agent policy distillation framework. A potential-game (PG) formulation provides a fixed-path longitudinal reference, around which a residual MARL policy learns longitudinal acceleration, conflict-approach timing, and lateral path corrections. A hard-case curriculum concentrates experience on interactions that PG leaves unresolved or substantially dela
multi-agent - arxiv:2611.21060 · cs.CVMetricHand: Geometry-Aware Flow Matching for 4D Hand Reconstruction in Metric World SpaceYibo Bai, Likui Zhang, Jinglei Zhang, Zhengbin Pang +3
Egocentric hand motion provides physical motion supervision for dexterous manipulation and vision-language-action pretraining. Reliable use of this supervision requires accurate and temporally coherent 4D hand reconstruction in metric world space. Existing methods typically lift framewise estimates produced under weak perspective into world coordinates. By ignoring joint-dependent depth variation at close range, this approximation couples errors in translation, orientation, and articulation that persist after world lifting and worsen under occlusion. We present MetricHand, a geometry-aware flow matching framework for 4D hand reconstruction in metric world space. MetricHand encodes depth priors from hand-object interaction regions into joint-associated representations that preserve local spatial structure and metric scale. Fused with visual features, this geometry guides coarse metric estimation and conditions every step of temporal generation. The model jointly recovers world-space tra
vision-language-actionmanipulationdexterous - arxiv:2611.21059 · cs.ROMulti-CONDOR: Learning Multistable Motion Primitives from DemonstrationGabriele Giuseppe Di Marzo, Mariano Ramírez Montero, Maximilian Stölzle, Rodrigo Pérez-Dattari +1
Dynamical-system-based motion policies provide a framework for learning reactive robot motions with stability and convergence properties. Many manipulation tasks, however, admit several distinct goals or a continuous set of acceptable outcomes, requiring motion primitives that preserve these alternatives. Accommodating both goal structures within a unified framework with continuous, autonomous dynamics remains an open challenge. We propose a CONDOR-based framework, multi-CONDOR, for learning motion primitives with either multiple discrete attractors or a continuous attracting goal contour. The approach combines imitation learning with losses that encourage agreement between the learned task dynamics and latent reference systems encoding the desired attraction structure. We provide a Lyapunov-based analysis of the convergence properties of the learned motion dynamics. We evaluate the framework on simulated benchmarks with discrete goals, qualitative examples with continuous goal contour
manipulationfrankagraspbenchmark - arxiv:2611.21057 · cs.ROWiTac: Wrist-Informed Tactile Representations for Contact-Rich ManipulationHaotian He, Yiyang Chen, Congsheng Xu, Wenchao Yue +8
Similar tactile contacts can call for different corrections depending on the configuration of the hand, object and target. Rather than supplying touch and wrist vision separately, fusing them before policy injection better grounds tactile features in visual context, yielding more accurate contact representations. We introduce Wrist-Informed Tactile Representations (WiTac), a pretrained encoder that learns such representations before action generation. Pretrained on nearly 300 hours of SharpaDex data, WiTac deeply fuses tactile signals with hand motion and wrist views into a comprehensive contact representation. Conditioning downstream VLA policies (GR00T N1.7 and $π_{0.5}$) on WiTac substantially improves task success over vision-only base policies across four contact-rich manipulation tasks (+28.8% and +15.0% on average), and further outperforms a wrist-free tactile encoder by 18.8% and 7.5%, respectively. Offline analyses further show that incorporating wrist vision yields more accur
vlamanipulationtactilegr00t - arxiv:2611.21056 · cs.CLEvaluation-Free Recursive Self-Improvement for Coding AgentsZhe Ren, Guowei Rong, Yimeng Chen, Nanbo Li +4
Recent self-improving coding agents have achieved strong benchmark performance by iteratively modifying their own executable implementations. A common pattern uses downstream task-derived evidence to decide which candidates or lineages continue evolving, which we term evaluate-and-select. Obtaining this signal requires repeatedly executing evolved agents on downstream tasks, so broader exploration incurs increasing evaluation cost, while repeated optimization against a fixed evaluation distribution may encourage benchmark-specific adaptation. We ask: can coding agents self-improve without downstream task evaluation inside the evolution loop? We introduce Explore-then-Integrate (ELITE), which integrates implementation mechanisms rather than selecting a single exploratory child to continue the lineage. At each generation, ELITE proposes diverse focused improvements, implements them independently from the same parent, and inspects the resulting programs to identify complementary, redundan
agentself-improvingself-improvementbenchmark - arxiv:2611.21053 · cs.AISystem 1 agents: a method for local, reliable and fast agentic AIEleftherios Tzanis, Michail E. Klontzas
Agentic artificial intelligence could automate analyses of sensitive clinical data, but current agents depend on large language models (LLMs). Cloud LLMs expose patient data. Local LLMs hallucinate, and an invented value in a tool call propagates through every later step. Here we present a method to build System 1 agents: local agents that call tools without generating text. A deterministic compiler turns each tool's schema and the request into multiple-choice questions whose options come from the request and the schema, then a one-pass decision model answers them. On 1,952 requests to 157 biomedical tools, the best System 1 agent chose the right tool, or correctly declined, for 97.0% of requests, significantly ahead of all 18 local LLM-agent settings (best 95.2%). The System 1 agent filled 97.0% of 6,056 stated argument values. No System 1 call contained invented content, against 5.5-19.7% of LLM calls. The System 1 Agents framework is freely available as open-source software at https
agentai agentagentic - arxiv:2611.21035 · cs.ROFlowRow: Learning Navigation Behaviors in Agricultural Environments with Long-Horizon Goal-Conditioned Flow Matching PoliciesUtkarsh Bajpai, Cédric Pradalier, Stéphanie Aravecchia
Autonomous monitoring and intervention in semi-structured agricultural environments, such as orchards and vineyards, require mobile robots to navigate to locations across diverse field configurations. This entails performing navigation behaviors including in-row driving, row turning and skipping, and inter-row traversal. Recent learning-based approaches have enabled robust in-row navigation from onboard perception, but navigation across rows typically requires a priori knowledge of field geometry to specify and tune row-transition maneuvers, preventing arbitrary transitions between rows. We address this challenge by framing navigation in these environments as a goal-directed navigation task, where a long-horizon goal is expressed relative to the robot and a planner generates local paths based on the goal and the robot's perceptual observations. We present FlowRow, a flow-matching-based local planner which models the distribution of diverse agricultural navigation behaviors as $\mathrm{
benchmark - arxiv:2611.21033 · cs.CVPOKE: Probe Offline, Keep Efficiently for Vision-Language-Action ModelsJaehyuk Lee, Yeajin Lee, Suhyun Bae, Donghun Lee
Visual token pruning reduces the inference cost of Vision-Language-Action (VLA) models, but attention-based scores do not directly measure a token's effect on the predicted action. We define action sensitivity as the magnitude of the first-order action change when a token is replaced by its camera's mean token and estimate it for all tokens with a single backward pass. Computing it online is costly: scoring on the full (unpruned) graph needs an extra forward pass at every control step. Scoring on the pruned forward pass is cheaper but causes scoring-graph lock-in: removed tokens lose their direct paths through later layers and tend to receive lower scores, making those locations less likely to return. On replayed observations, even with a random conditioning mask, pruned-graph scoring re-selects 59% of the mask's tokens, versus 6% expected by chance; attention histories show a similar bias. POKE (Probe Offline, Keep Efficiently) computes full-graph action sensitivities offline and dist
vision-language-actionopenvlalibero - arxiv:2611.21023 · physics.opticsIntrinsic Berry Curvature on the Higher-order Poincare SphereKaijun Zhang, Boen Yang, Hailu Luo
Higer-order geometric phase enable versatile spin-orbit manipulation of structured light . Con ventional theories describe this phase via a total-angular-momentum dependent Berry curvature . We develop an intrinsic theory restricted to the effective two-dimensional subspace, where the Berry curvature is independent of the topological charge of the light field . In our framework, the total-angular-momentum difference does not alter the effective monopole strength, whereas it controls the solid angle enclosed by the trajectory of the state evolution, insteadly. Our theory provides a different topological interpretation of geometric phase of the structured light, which unifies the underlying structures of generalized Poincare Spheres. Furthermore, it does not impose any specific form on the physical operation of the optical element, naturally encomoassing the case of previously reported experiment, and can be applied in arbitrary state evolutions in both classical structured light and qua
manipulation - arxiv:2611.21020 · cs.LGSWE-OPD: On-Policy Distillation for Agentic CodingYuhang He, Yanyu Chen, Chuanzhe Guo, Jie Zhu +10
In this work, we study on-policy distillation (OPD) for transferring agentic coding capabilities from a large teacher model to a smaller student model. Prior work has focused primarily on reasoning and short-horizon agentic tasks. We present SWE-OPD, a minimal OPD study for software engineering, where the student interacts with executable repositories and the teacher supervises the prefixes visited by the student. Across three teacher--student pairs, two training sources, and three SWE benchmarks, we find that on-policy supervision transfers coding capability: supervision from Qwen3.6-27B increases the SWE-bench Pro resolve rate of Qwen3.5-9B from 35.9% to 39.4%. Refreshing trajectories is not strictly required either, as an offline variant with cached teacher scores still improves over the student baselines while cutting end-to-end GPU-hours by 59.4%--72.6% and halving peak GPU allocation, at benchmark-dependent accuracy. Furthermore, we extend the offline variant to corss-tokenizer s
agentagenticbenchmark - arxiv:2611.21019 · cs.CLMemDoctor: Stage-wise Error Attribution and Repair for MLLM Agent MemoryRuiyang Jiang, Xinyang Liao, Weikai Huang, Yuqi Niu +2
Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them. A stage-resolved analysis on Mem-Gallery traces most failures to a source stage, and most of those lie before the answer. Injecting known errors against fact-preserving control edits, we then follow each tainted record downstream. Early errors survive to the answer as apparent evidence, and hurt most when no other record holds the fact. A diagnosis that trusts the quoted evidence therefore blames the stage where the error surfaces, not the one where it began. We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures. It
memoryexternal memoryagent memoryagentllm agent - arxiv:2611.21003 · cs.AIK2-Horizon: A Frontier Fleet of Open Source Language ModelsK2 Team, Zhengzhong Liu, Haonan Li, Xuezhe Ma +91
We present K2-Horizon, a fleet of six fully open language models, from a 0.9B dense model to a 375B-A23B MoE, built for general-purpose reasoning and agentic work across math and science, software engineering, deep research, office automation, and beyond. The models lead or are competitive with strong open models of their size on agentic and coding benchmarks. On the modeling side, to strengthen reasoning, we introduce Mixture-of-Value Attention (MoVA), applying MoE to attention value projections, and softplus attention gating, a more flexible, theoretically motivated gate. Our 36B-A4B model uses both and nearly matches a 32B dense model trained on the same data, with one-seventh the active parameters. We also introduce Uno, lossless diffusion adapters that speed up decoding by up to 2.2x without a separate draft model. Our release follows the LLM360 definition of open source for transparency and reproducibility: the full development history, including intermediate checkpoints from eve
agenticbenchmark - arxiv:2611.20997 · cs.MAOpenMAS-GApp: A Comprehensive Benchmark for Domain-Adaptive Multi-Agent Application ConstructionXunkai Li, Kairui Yang, Jiaxi Li, Yuecong Mao +4
Constructing multi-agent system (MAS) applications across domains requires translating diverse requirements into executable systems, involving task decomposition, agent and tool selection, and execution organization. Existing studies have explored individual aspects of MAS construction, including agent generation, architecture design, workflow generation, and component selection. However, these methods are often developed and evaluated independently, using different task specifications, component resources, and execution settings. This makes it difficult to compare their construction capabilities or determine how they can be applied to different domains. To address this problem, we propose OpenMAS-GApp, a One-for-All benchmark for MAS application construction across diverse domains. It standardizes task inputs, component resources, construction interfaces, and execution environments to support MAS construction under consistent conditions. Within OpenMAS-GApp, we provide Graph Harness a
agentmulti-agentagent systembenchmark - arxiv:2611.20996 · cs.LGExplaining Cross-Label Decision Structure in Multi-Output Tree EnsemblesSylvio Barbon Junior, Hiago F. F. Oliveira, Rafael Figueira Gonçalves, Jessica Fernandes Lopes +1
A multi-output tree ensemble predicts several binary labels through decision paths that it largely shares across those labels, yet the attribution methods commonly used to explain such models treat each output in isolation. We propose multi-sink Decision Predicate Graphs (ms-DPGs), which extend decision predicate graphs with one sink per label: each leaf contributes to the sink of every label its in-bag positive count for that label. On this representation we define Shared Predicate Flow (SPF), the share of a label's decision mass that passes through conditions active for both labels of a pair, and a contracted view of the graph in which label-specific lifts indicate whether a shared condition raises the probability of both labels or of one at the expense of the other. Controlled experiments show that SPF captures shared decision conditions rather than mere feature overlap. On nine multi-label benchmarks, native multi-output forests share substantially more decision structure than Bina
benchmark - arxiv:2611.20992 · cs.CVMedJev: A System One Vision-Language Decision Model for Fast and Effective Medical Image UnderstandingFan Zhang, Hao Wu, Jinyang Wu, Hongyu Qu +4
Medical image understanding with vision-language models (VLMs) typically relies on autoregressively generating reasoning and answer tokens. These unnecessary tokens make the inference cost scale with the response length and force the answer to be parsed from free-form text, incurring avoidable latency and decision errors. To this end, we introduce \textsc{MedJev}, a \emph{system one} vision-language decision model that maps medical images, questions, and candidate answers directly to decisions in a single multimodal forward pass, without any autoregressive generation process. For fast inference, \textsc{MedJev} removes the upper language layers and the generative output pathway, while keeping the full vision pathway and all visual tokens. On this shortened backbone, a decision head scores each candidate by pretrained query--key matching and refines the scores with a lightweight multilayer perceptron (MLP) residual, coupling the pretrained matching structure with task-specific nonlinear
benchmark - arxiv:2611.20988 · cs.AIHivemind: A Federated Protocol for AI-Mediated Tacit Knowledge Sharing in OrganisationsSrinivasan Sundaresan
This paper presents Hivemind, a federated protocol that enables AI agents to exchange tacit knowledge on behalf of their users, asynchronously and without requiring any direct human interaction. Individual AI memory systems capture the decisions, reasoning, and contextual knowledge that surface during AI-assisted work. Hivemind creates the shared space in which this captured knowledge can flow between individuals across team and departmental boundaries, addressing a structural gap that neither centralised knowledge repositories nor existing AI tooling closes, even where partial solutions such as expert-finding systems exist. The paper evaluates centralised versus decentralised design approaches, proposes a federated architecture grounded in Nonaka and Takeuchi's (1995) Socialisation, Externalisation, Combination and Internalisation (SECI) model, and demonstrates a proof-of-concept implementation using Design Science Research (DSR) methodology. Measuring the enterprise benefit of federa
memoryai agent - arxiv:2611.20985 · eess.SYA Rollout-Based Approach to Real-Time Topology and Dispatch Control in the AI4RealNet L2RPN BenchmarkDavid Krame Kadurha, Robbert van der Burg, Alessandro Zocca, Sandjai Bhulai
Operating transmission networks under uncertain outages requires coordinating discrete network topology actions with continuous adjustments to generation and storage, while respecting operational constraints and real-time computational limits. We present a rollout-based controller that combines discrete and continuous decisions under the uncertain contingency policy of the L2RPN AI4RealNet 2026 challenge. Power-flow approximations and convex optimization identify the most promising actions, followed by short-horizon simulations to predict their effect on congestion, renewable curtailment, and redispatching costs. A further policy exploits the benchmark's specific attack protocol: keeping selected lines disconnected prevents some adversarial outages from being initiated. On the benchmark's $118$-bus network, the controller completed $189$ of $208$ final-test scenarios (score $64.39$) and achieved the second-best ratio of score to computation time among the five top-ranked entries, behin
benchmark - arxiv:2611.20983 · cs.LGGRACE: Parallel Best-of-N Multi-Agent Analysis of Closed-Form Questions over Dataframes with Guardrail Selection and Adaptive RetryFabio Picariello, Salvatore Cipolla, Susanna Bonura, Vito Morreale
LLM agents answer closed-form questions over a tabular dataframe by writing and executing Python, almost always as a single rollout: one interpretation, one script, one result, accepted without comparison. The few pipelines that produce several candidates reconcile them by answer-agreement voting, which needs a single comparable scalar and always returns one, even when every candidate is wrong. We present GRACE (Generative, Ranked, Adaptive Code Ensemble), a production-oriented pipeline that instead runs parallel best-of-N selection specialised to dataframe code. N agents run concurrently under a strict, machine-checkable output contract, diversified by persona hints, per-retry rotation, and a two-regime temperature schedule. Rather than vote, GRACE judges the executed artefacts: a pairwise tournament pre-filters the pool and a domain-specialised guardrail either selects the best candidate or, when none is defensible, rejects the whole pool and abstains with a diagnosis (an option voti
llm agentmulti-agentbenchmark - arxiv:2611.20978 · cs.ROInverse Kinematics in Endoscopic Scenarios using MoveItMarc Bestmann, Timon Engelke, Rebecca Rodeck, Gerko Wende
Inverse kinematics are critical in many robotic applications, yet standard MoveIt-based solvers struggle in confined scenarios. This paper provides a benchmark of the available solvers and compares multiple practical modifications for improving the performance of the solvers. Splitting the IK problem into two parts results in the best performance, leading to a 100% success rate under 100 ms in an endoscopic tank scenario.
benchmark - arxiv:2611.20971 · cs.CVFAVR: Flow-Aware Visual Residual Alignment for Vision-Language-Action PoliciesQixuan Wu, Xiaole Yang, Yuguang Yang, Canyu Chen +4
Pretrained video representations provide rich temporal information for robotic manipulation, yet effectively integrating them into existing vision-language-action (VLA) policies remains challenging. We present Flow-Aware Visual Residual Alignment (FAVR), a lightweight framework for flow-aware visual conditioning of frozen VLA policies. FAVR maintains a compact temporal visual state through linear Test-Time Training (TTT) and retrieves action-conditioned visual features throughout flow-based action generation. By exploiting the affine structure of the frozen action head, we explicitly characterize how injected visual features influence predicted action velocities and formulate flow-aware channel selection as a box-constrained optimization problem. We develop a FISTA-based solver for joint gate optimization and derive an efficient closed-form approximation through diagonalization and rescaling. With only 1.74-2.19M trainable parameters and 5,000 adaptation steps, FAVR improves pretrained
vision-language-actionvlamanipulationaction headrobotwinaction-conditioned - arxiv:2611.20956 · cs.MAFrom Vision to Interaction: Evaluating Dynamic Social Intelligence of VLM Agents in ALIVE-WorldYue Xu, Xiaoyu Qin, Sihan Zhao, Ze-en Xiong +1
As AI agents are increasingly deployed in real-world applications and take on social roles, dynamic social intelligence, the ability to continuously perceive multimodal cues, infer latent cognitive states, intervene through embodied actions, and ultimately achieve social goals, has become essential. However, existing evaluations largely fragment social intelligence into text-based abstractions, static reasoning tasks, or embodied simulators that lack social dynamics. These approaches fail to capture the multimodal, continuous, and interpersonal complexity of authentic human interactions. To bridge this gap, we introduce ALIVE-World, an embodied simulation environment supporting dynamic multimodal interactions encompassing egocentric vision, language, and non-verbal action. To provide socially grounded interactive feedback, we populate this environment with counterpart agents driven by a profile-conditioned, dual-system (intuitive-affective and logical-cognitive) architecture. Building
embodiedai agentbenchmark - arxiv:2611.20950 · cs.ROROSE: A Shared Cross-Embodiment Facial-Motion Latent Space for Humanoid Robot ExpressionYongji Fu, Man Cao, Yi Zhou, Fei Lin +2
Natural, coherent full-face dynamic expressions are central to making humanoid robots socially approachable, yet existing systems commonly realize only limited lip motion or scripted expressions. The limitation is both physical and algorithmic. A compact head leaves little room for densely packed, independently controllable actuation; meanwhile, learned systems often generate an avatar video, landmarks, or facial parameters and then infer actuator commands through visual retargeting. This render-and-invert path adds latency and artifacts, and it does not reliably transfer generated behavior to a physical robot. We present ROSE (RObot facial expression through Shared motion Embeddings), a framework that integrates biomimetic hardware, cross-embodiment motion representation learning, and multimodal motion generation. On the hardware side, we build a flexible tendon-sheath transmission structure that integrates 34 actuators into a 19.5 cm biomimetic head. By decoupling actuator placement
humanoid - arxiv:2611.20939 · cs.ROTacBridge: Learning Unified Tactile Representations through Cross-Sensor CorrespondenceYongji Fu, Yijiong Lin, Yi Zhou, Yi Qi +1
Optical tactile sensors record tactile videos that capture local deformation and its temporal evolution during contact, supporting robot manipulation, force estimation, and tactile representation learning. Differences in imaging geometry, background appearance, illumination, resolution, and deformation response lead to sensor-dependent observations of the same contact across GelSight, DIGIT, GelSlim, TacTip, and other optical tactile sensors. This variability limits representation transfer and the joint use of multi-sensor data. We introduce TacBridge, a framework for learning unified representations from multi-sensor tactile videos through cross-sensor correspondence. Specifically, multi-sensor masked tactile-video pre-training learns spatiotemporal representations across sensors. We then train the same masked model jointly on same-sensor and cross-sensor reconstruction, using paired source and target videos of matched contact events. Experiments on cross-sensor translation and downst
manipulationtactilegelsighttactip - arxiv:2611.20936 · cs.CVReAccess: Look Once, Stay Focused with Efficient Test Time MemoryTianhao Chen, Yilong Fan, Yuheng Wu, Xiaobin Hu +1
Despite recent advances, vision-language models (VLMs) still struggle with fine-grained visual understanding, often failing to effectively utilize critical evidence already encoded in their visual representations. Existing approaches address this limitation through repeated visual processing or explicit reasoning, introducing substantial computational overhead. In this paper, we propose ReAccess, a lightweight framework that preserves and reuses question-relevant visual evidence through compact associative memory. Constructed during a single visual prefill, the memory supports progressive evidence retrieval that steers both attention queries and hidden representations, enabling effective visual grounding without repeated image encoding or explicit reasoning traces. Extensive experiments demonstrate consistent improvements in both general visual understanding and fine-grained perception, while achieving substantially faster inference than existing reasoning-intensive approaches, demonst
memory - arxiv:2611.20935 · cs.AIRazavi-Bench: Tracking AI Progress in Analog Circuit ReasoningZhishuai Zhang, Oscar Toomey, Taekwang Jang
We propose Razavi-Bench, a benchmark built from the 50 questions in Razavi's two Analog Design Experiments With AI articles and graded by GPT-6 Luna against reference answers and rubrics, and use it to track how fast AI models improve at analog circuit analysis. Across 73 multimodal large language models released between May 2024 and November 2026, the top score rose from 28.83% to 98.33% in 30 months, mostly before the reference answers were published. The leading models now answer almost every textbook question correctly; the hardest questions show a circuit that resembles a textbook one but works differently, or ask about behavior the obvious analysis misses. Within two and a half years, a set of questions that exposed the limits of AI models has become one they answer almost perfectly.
benchmark - arxiv:2611.20927 · cs.MALEDGER: Auditable Agent Memory Under ChangeMd Aidul Islam, Md Nasir Uddin Shuvo, Muhammad Waseem, Pekka Abrahamsson
Agent memory systems are mostly evaluated on fixed histories, and only their final answers are scored. In real use, facts change, sources differ in reliability, and users ask the agent to forget things. We present LEDGER, a memory layer in which a language model proposes facts and explicit rules decide which value is current, which source to trust, and what to delete. Each fact keeps its source and the period in which it held. LEDGER+EP adds a search over raw conversation turns for questions the ledger does not cover, and scrubs these turns on deletion. We compare both against no memory, full-history replay, retrieval over raw transcripts, and Mem0, with the same answering model and prompt, on a synthetic multi-session coding-agent workload and a 210-question subset of LongMemEval, and we score both the evidence each memory sends to the model and the answer. On the coding workload, LEDGER+EP in fallback mode answers 94.3% of questions correctly on five development seeds and 91.1% on fi
memoryagent memoryagent - arxiv:2611.20926 · cs.AIAdvancing Cognitive Pairwise Comparisons for Human-Centric Artificial Intelligence and Data Science: A Human-in-the-Loop FrameworkKevin Kam Fung Yuen
Human-centric artificial intelligence and data science (HCAD) systems increasingly involve human-in-the-loop (HITL) interventions to overcome the limitations of fully automated machine learning algorithms, such as automation bias and the inability to incorporate tacit domain knowledge, and continuously improve predictive performance. Data transformations, feature selection, feature weighting, and model selection usually require laborious human involvement across iterations of artificial intelligence and data science (AI&DS) pipelines, making development a challenging and time-consuming process. The Cognitive Pairwise Comparison (CPC) framework is proposed as a human-centric methodology to elicit preferences and priorities in a structured format. To establish essential facets of human-centricity, this research advances the CPC methodology so that user preference elicitation can be systematically quantified in these challenging AI&DS tasks. To demonstrate the usability and validity of th
human-in-the-loop - arxiv:2611.20924 · cs.CLNavital Gym: Rule-Graded Agent Environments for Indian WorkflowsDaksh Bathla
AI agents are being deployed on Indian workflows such as payments, tax, forms, travel and customer support in Hindi and Hinglish. Public agent evaluations rarely cover these workflows, and many score with another language model. We present Navital Gym, a set of six open environments with 321 public tasks in five task families: trip planning, UPI payments and disputes, GST invoicing and returns, form filling (including a multi-document split) and customer support in Hindi, Hinglish and three experimental Indian languages. Every task is graded by named, deterministic rules that recompute the outcome from a synthetic sandbox world. No language model judges any score. We report one run per task for three models on all six environments, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 4.5, and for Gemini 3.1 Flash-Lite on two (UPI and the Support core). Opus solves 315 of 321 tasks, Sonnet 302 and Haiku 205. Trip planning and Hindi/Hinglish support separate the models most. We also descr
agentai agent - arxiv:2611.20922 · cs.LGExperience Rot: Experienced Agents Fail Where Novices SucceedZhen Xu, Hang Yin, Zitao Zhang, Kevin Zhao +1
Agents are increasingly built to learn from experience: they keep conversation histories, memory files, retrieved notes and skills, and reuse what worked before. We show that this experience turns against them when their tools change silently. We test a range of models with tools defined as in the Model Context Protocol (MCP). After a silent change, experienced agents fail on most tasks, while agents new to the task mostly succeed. We call this failure Experience Rot. Experienced agents copy their earlier call instead of checking the current interface, even when the tool definition shows the change. Then they report success. When the documentation is incomplete, experience is what makes agents succeed. After a change, the same experience leaves them worse off than no experience at all. The failure appears in every form of memory we test and grows with the amount of experience. Simple remedies fall short: announcing the update barely helps, and pointing the agent to the documentation fa
memoryagent - arxiv:2611.20916 · cs.CVEviBridge: Encoding Predictive Visual Evidence for Proactive InteractionZhichao Huang, Changheng Cai, Song Ren, Zhichen Zhao +8
Tool-handover interaction calls for preparation before a response is warranted. The model waits until interaction cues appear, then prepares while assessing whether the accumulating evidence supports a response or calls for cancellation. We introduce EviBridge, a framework that integrates predictive visual representations into a vision-language interaction model for joint state and tool prediction. A trainable evidence interface maps features from a frozen V-JEPA 2 encoder into the language decoder alongside native visual inputs, preserving their temporal context and using only observations available at each decision time. The framework has two instantiations. EviBridge-E distinguishes recent from historical evidence and encodes the timing and spatial location of each observation. EviBridge-I preserves the temporal order of pooled V-JEPA 2 features and interaction turns and generates dedicated interaction tokens. We evaluate on FuGIBench, a streaming-video benchmark for preparation, re
v-jepabenchmark - arxiv:2611.20915 · cs.AIMulti-LLM Agentic AI at the Edge: Joint Agent Scheduling and Resource AllocationYuanyan Song, Kezhi Wang, Xinmian Xu, Yongmin Li
In this paper, an edge-enabled multi-large language model (LLM) agentic AI architecture is proposed, where one supervisor AI agent and multiple worker AI agents cooperatively execute users' requests at the edge. The supervisor AI agent decouples the user request into a series of sub-tasks and dynamically dispatches them to LLM-equipped worker AI agents via two agentic workflows, i.e., parallel and serial workflows, with a two-level retry mechanism to enhance execution reliability. To prolong the service duration of worker AI agents, the energy minimization problems are formulated by jointly considering AI agent selection and resource allocation, subject to task execution latency and quality of service (QoS) constraints. A resource allocation scheme based on alternating optimization (AO) is proposed. Specifically, for the parallel workflow, the scheme decouples the original problem into bandwidth allocation, retrieval selection and AI agent selection sub-problems, which are iteratively
agentai agentllm agentagenticbenchmark - arxiv:2611.20908 · physics.app-phHow Elasticity and Chiral Damping Shape Domain-Wall CreepAdriano Di Pietro, Alessandro Magni, Stefania Pizzini, Giovanni Gandini +13
Asymmetric domain-wall expansion in the creep regime provides an experimentally accessible probe of the interfacial Dzyaloshinskii-Moriya interaction (iDMI) and, in principle, chiral damping. However, severe discrepancies with complementary measurement techniques and the lack of a model describing the full experimental phenomenology of the creep of chiral domain walls, limits the application of these measurements. Here, using a symmetry decomposition of the velocity curve, we derive an analytical model describing the coupled influence of domain-wall stiffness and chiral damping in the creep regime. Their interplay generates the antisymmetric distortion that shifts the velocity minimum away from the DMI field. Applied to Co-based magnetic multilayers, the model yields iDMI values consistent with BLS and flow-regime benchmarks and identifies the reasons for the longstanding discrepancy. The model provides insights for possible strategies to measure chiral damping via creep dynamics of do
benchmark - arxiv:2611.20906 · cs.CVCan Image Generators Hold a Scene Together When Space Is Broken Apart?Jhen-Ke Lin, Chung-Chun Wang
Image generators are good at drawing scenes in which things that belong together appear next to each other. In almost every image they generate, and in almost every benchmark used to evaluate these images, "belonging to the same whole" and "adjacent to each other" go together. Therefore, we cannot know whether the model truly grasps the scene as a whole, or simply makes each part connect with its adjacent parts. We propose GJBench to disentangle the two. It requires the model to draw a scene as a scattered jigsaw puzzle: an image is cut into interlocking pieces, then moved apart and rotated. Thus, the spatial plausibility of the scene exists only in how these pieces collectively point to a whole. The model may have a reference image, a target layout, both, or neither. We separately evaluate three requirements: whether the scene is broken down into a specified number of separate pieces, whether the pieces interlock, and whether the pieces can reconstruct the reference scene. While every
graspbenchmark - arxiv:2611.20899 · cs.CVSciWorldBench: Do Video World Models Really Understand Scientific Experiments?Xue Yang, Haonan Cai, Xinyu Hang, Zhimeng Huang +3
While the ability of video generation models to accurately depict scientific phenomena has received growing attention, determining whether these models genuinely understand the underlying principles remains a critical challenge. Existing benchmarks typically evaluate video samples in isolation, making it difficult to expose models that simply fit superficial visual priors. In scientific experiments, a small change in one condition produces a markedly different outcome even when the rest of the setup remains fixed. Capitalizing on this sensitivity, we introduce SciWorldBench, a benchmark comprising paired scientific experiments. Each pair varies a single condition in an otherwise fixed setup, yielding two mutually exclusive expected outcomes. Each generation prompt specifies the condition but omits the expected outcome. We evaluate ten text-and-image-to-video (TI2V) models using an independent vision-language model (VLM) that assesses each video without access to the condition text or t
world modelbenchmark - arxiv:2611.20891 · cs.LGFrom Recursion to Convolution: Parallel Self-Improvement for Multi-Agent SystemsZheng Wang, Yuang Liu, Sachith Abeywickrama, Yangkai Ding +6
Multi-agent self-improvement enables agents to improve through interaction and feedback. However, sequential dependencies across agents and refinement rounds can propagate errors, increase computational costs, and limit parallelism. Moreover, refining individual features or agents in isolation can overlook their dependencies and restrict collective coordination. To address these challenges, we propose ConvRSI, a framework for parallelizing multi-agent self-improvement through convolution. ConvRSI couples intra-agent convolution, which jointly refines related agent features, with inter-agent graph convolution, which aggregates messages according to the collaboration topology. Both stages operate in latent space, enabling parallel updates without repeatedly decoding and re-encoding intermediate messages. Experiments on MATH-500 show that ConvRSI achieves 5.04% performance improvement while reducing 17.27% inference cost, and 45.11% latency compared with RecursiveMAS. It also supports div
agentmulti-agentagent systemself-improvement - arxiv:2611.20888 · cs.RODriving the Style: Velocity-Conditioned Whole-Body Control from Motion DemonstrationsMurad Dawood, Oezge Drama Berkenkamp, Antonio Gonzales Marin, Nutan Chen
Achieving human-like behavior on humanoid robots has advanced rapidly by leveraging motion-capture demonstrations rather than hand-crafted cost functions. Adversarial imitation supplies a style signal that lets a policy fulfill a task while adhering to a desired style, but is difficult to transfer to hardware. Whole-body tracking through reinforcement learning instead deploys reliably and imitates demonstrations faithfully, but the resulting policies are conditioned on the future reference, so they require a reference trajectory at runtime and can fail when the motion differs from the training data. We train a humanoid policy directly with whole-body tracking rewards while restricting its actor to compact locomotion commands and robot observations. Reference motion supervises the reward and privileged critic, without supplying reference joint trajectories to the actor or requiring teacher distillation. The demonstrated style is thereby encoded in the policy through the tracking reward,
humanoidwhole-body control - arxiv:2611.20885 · cs.CVProSE: Probing Reconstruction Error for Semantic-guided Expert Skipping in MoE-MLLMsJiameng Li, Minye Wu, Matthew B. Blaschko
Mixture-of-Experts Multimodal Large Language Models (MoE-MLLMs) offer scalable inference through sparse expert activation, yet substantial computational redundancy remains among the activated experts. Existing expert skipping methods typically calibrate thresholds based on target skipping ratios or final-output divergence, overlooking the accumulation of skipping-induced error across layers and the varying semantic importance of individual tokens. To address that, we propose ProSE, a training-free, semantic-guided expert skipping framework that probes layer-wise reconstruction error to promote representation consistency compared with default activation. Specifically, we construct query-aware semantic masks to identify salient visual tokens and progressively calibrate modality-specific skipping thresholds for each MoE layer. By minimizing masked reconstruction error under a given budget, ProSE maintains semantic consistency after skipping and allows adaptive expert allocation under limi
benchmark - arxiv:2611.20882 · cs.AIAccelerate Bench: Evaluating Agents for Application-Level Program OptimizationBowen Liu, Yuhan Li, Nuo Chen, Weile Luo +6
Application optimization requires coding agents to discover useful changes and coordinate them across a complete program while preserving the required computation. We introduce Accelerate Bench, a benchmark of 32 CPU optimization tasks adapted from scientific applications and supercomputing competitions. Each task provides a working program and specifies the workload, correctness requirements, timing boundaries, and permitted edits, leaving agents to choose where and how to optimize. Public testing supports iterative development, and a separate verifier scores the final submission. On benchmark version 1.1, we evaluate 16 coding agents, each pairing a model with a harness. Their Score@1 ranges from 25.5% to 67.4%. Most of their programs are correct, and the size of the speedup separates them: the two leading agents are at least 2x faster on 22 or more of the 32 tasks, while the other fourteen reach 2x on 8 to 17 and leave 125 of their 412 correct outcomes below 1.25x. Controlled replay
benchmark - arxiv:2611.20873 · cs.AIRe-Reading Is Not Free: Token Traces and Cache-Aware Energy Estimation for LLM Agents in Software EngineeringTina Vartziotis, Ketevan Peranidze, Rodopi Kosteli, Ezgi Sarikayak +4
As large language models (LLMs) evolve from conversational assistants into autonomous agents embedded in software development workflows, estimating their environmental footprint requires moving beyond conventional token-based accounting. Existing energy estimators focus on isolated request-response interactions, overlooking the repeated processing of accumulated context, cache reuse, and work delegated to subagents. We analyze nearly one billion tokens collected across 42 sessions and 752 task episodes over ten weeks of real-world AI-assisted software development. Cached-context reads make up 94.8% of all API-reported tokens, while uncached input accounts for only 0.017%; for every generated token, approximately 260 cached tokens are read. Context grows 5.3-fold over a session compared with 1.4-fold within individual tasks. To quantify the energy implications, we extend an existing analytical inference-energy model to account for cached-context processing, yielding a cache-aware hybrid
llm agentautonomous agentagentic - arxiv:2611.20854 · cs.CVWhen Is the Expensive Method Worth It? A Content-Driven Analysis of Classical Bayer DemosaicingTaranum Wasu, Harsh Kashyap, HS Pannu
Demosaicing reconstructs a full-colour image from a single-sensor colour-filter-array (CFA) mosaic, where only one colour is measured per pixel. Classical methods make different assumptions about images, from simple averaging to directional edge-aware interpolation, and the directional methods cost markedly more compute. We ask a practical question that a per-image quality ranking does not answer: for a given image, is the expensive directional method worth its cost? Using a controlled simulate-reconstruct pipeline on 80 benchmark images from four public datasets (Kodak, McMaster, BSD100, Urban100), six classical methods, and image-content descriptors fixed before any results were seen, we show that the quality gain of the edge-aware method (Menon 2007) over the best cheap linear method is not constant. It ranges from $-6.6$ to $+6.8$ dB, and on 31% of images it is zero or negative, meaning a cheap method matches or beats the expensive one. The gain is also predictable from content: it
benchmark - arxiv:2611.20851 · cs.LGartist: A Python package for differentiable ray tracing and gradient-based optimization of solar tower power plant operationMarlene Busch, Marie Weiel, Kaleb Phipps, Daniel Maldonado Quinto +3
Solar tower power plants provide dispatchable, renewable energy by concentrating sunlight with thousands of mirrors, called heliostats, onto a central receiver. Efficient operation requires accurate modeling of the optical path from each heliostat to the receiver and compensation for manufacturing imperfections, kinematic misalignment, and material degradation that distort the flux distribution. Conventional ray tracers predict the flux from an assumed plant model but are limited to forward simulation. They cannot recover unknown plant parameters from measured flux nor directly optimize plant operation. We present artist, an open-source, FAIR4RS-compliant Python package implementing a fully differentiable digital twin of solar tower power plants. By making the optical pipeline, comprising heliostat surface representation, kinematics, blocking, and ray tracing, end to end differentiable, artist turns the digital twin from a passive forward model into an active physics-based tool for inv
memory - arxiv:2611.20848 · cs.AIMolReps: Taking SELFIES, Making SMILES, Teaching LLMsKen Deng, Garrett Morris
The same molecule can be expressed in multiple, syntactically distinct representations, yet it remains unclear whether large language models (LLMs) can recognize and relate these equivalent encodings. We introduce MolReps, a pipeline that generates 9,600 question-answer pairs across ten representations for conversion, normalization, and 2D structure recognition. Post-training on MolReps improves text-only and multimodal models, MolQwen3-4B-Instruct (MQ3-4B-I) and MolQwen3-VL-4B-Instruct (MQ3-VL-4B-I), from $4.01\%$ to $75.13\%$ and from $7.11\%$ to $73.14\%$ (InChIKey match) on the molecule-overlap test set, respectively. On the molecule-disjoint test set, post-training increased validity from $20.35\%$ to $58.12\%$ for MQ3-4B-I and from $17.19\%$ to $59.34\%$ for MQ3-VL-4B-I. However, InChIKey match remained low, decreasing from $3.86\%$ to $3.09\%$ and from $6.44\%$ to $2.29\%$, respectively. Furthermore, the resulting models achieve a relative $27.84\%$ improvement in average over t
post-trainingbenchmark - arxiv:2611.20846 · cs.AIEDITORS Know Your Style! Editing LoRA Subspaces for Stylistic Attribution and ImitationZixuan Wang, Gregory Kang Ruey Lau, Xiao Tian, Jue Fan +2
Disentangling writing style from semantic content is a fundamental challenge in literary text modeling. Style-content entanglement causes models to rely on semantic content for authorship attribution (AA) tasks and memorize author-specific content for style imitation (SI) tasks. We propose EDITORS (EditingLoRA Subspaces), a diagnose-then-deflate framework that adapts pretrained large language models and prioritizes stylistic features, making it effective for both AA and SI tasks. EDITORS first trains a diagnostic LoRA adapter using style-neutral statements derived from the training corpus to identify a content-oriented subspace. It then trains the final style adapter under activation-space deflation that projects input activations away from the identified content-oriented directions. We empirically demonstrate that EDITORS is effective. On three AA benchmarks spanning literature, social media, and news, EDITORS achieves state-of-the-art accuracy, outperforming the strongest existing ba
benchmark - arxiv:2611.20844 · cs.AIDetourBench: Trajectory-Level Evaluation of Interactive Agents with Exact Minimum-Action BoundsHanlin Tian, Sihan Zhu, Zhao Yang, Yuxiang Wang +4
Benchmarks judge language-model agents mainly by whether they reach the correct final state, not by how they got there. Endpoint scoring cannot separate an agent that retrieved exactly the evidence a decision required from one that read most of the environment, nor an agent that never located the evidence from one that stopped a single action short. Settling this needs a reference cost, and authored solutions establish only that a task is solvable, not that no shorter route exists. We present DetourBench, which constrains the environment until that minimum is computable. Each task is a finite, partially observable transition system with unit-cost accesses whose resources become globally addressable after discovery. Exhaustive search yields the full-information minimum action count $L_0$ with a replayable witness; from graph-recoverable terminal states it also yields exact residual work $R_T$, which separates an episode that stopped one access short from one that wandered, while an irre
agentbenchmark - arxiv:2611.20836 · eess.SYUniversal Quantizer Design for Identification of Systems with MemoryYibo Shi, Ying Wang, Karl H. Johansson, Cristian R. Rojas
In system identification from quantized observations, the quantizer determines how much information about the parameters the observations retain, and hence the achievable estimation accuracy. Quantizer design can be guided by the Fisher information matrix (FIM), which assesses the parameter information retained by candidate quantizers. Existing methods, however, typically rely on closed-form expressions of the FIM. This paper proposes a universal quantizer design method for general systems with memory. Its core is a universal FIM estimator that evaluates the information per bit from simulated trajectories, and the thresholds are designed by maximizing the estimated criterion under a bit budget. For unknown parameters, the design is implemented through a two-stage procedure. Numerical studies on a quantized ARX system yield designs close to the reference optimum, while a nonlinear cascaded-tanks example reduces the identification error by about 50% relative to a single median threshold.
memory - arxiv:2611.20827 · cs.AIProcReverser: Multi-Agent Black-Box Microarchitectural Reverse Engineering for Processor Vulnerability AnalysisGuanqi Lyu, Decheng Qiu, Guangyang Zhang, Fei Zhao +3
When processor implementation details are unavailable, black-box microarchitectural reverse engineering helps security researchers identify hardware conditions related to security vulnerabilities. However, this process requires substantial manual effort to design experiments, develop test programs, and analyze results for each processor. We present ProcReverser, a multi-agent system that automates this process. ProcReverser uses execution feedback to improve experimental plans and test programs and identifies microarchitectural properties from measurements. We develop a comprehensive suite of property identification tasks focused on hardware mechanisms relevant to Spectre-class attacks, covering speculative execution, branch prediction, memory dependence, and cache behavior. We evaluate ProcReverser on the XuanTie C910, the gem5 RISC-V O3 CPU model, and the SiFive P550. ProcReverser achieves Pass@1 and Pass@5 scores of 81.3% and 93.8%, respectively, on C910, and 83.0% and 100.0% on gem
memorymulti-agentagent system - arxiv:2611.20826 · cs.AIConversing with a Conversational Agent: A Real Conversation, Not a Simulated OneChristophe Denis
The claim that interactions with artificial conversational agents merely simulate human conversation rests on a philosophical distinction between reproducing the observable features of an activity and actually performing it. Although this distinction is appropriate for certain physical phenomena, its application to linguistic and symbolic activities is far less straightforward. This article argues that conversations with large language model (LLM)-based agents are genuine conversational activities rather than mere simulations, without thereby attributing consciousness, subjective understanding, or human intentionality to artificial systems. Drawing on the distinction between subjective and objective simulation, the article first examines the limitations of defining conversation exclusively through the intrinsic properties of its participants. It then develops a relational and semiotic account of conversational activity, informed by linguistic pragmatics, Lacan's Conversing with a Conve
agent - arxiv:2611.20823 · cs.CLMedical question answering collapses in Hausa: a pre-registered pilot on three small open modelsAbbas Sani
A system that reads nothing and always answers the most frequent letter scores \BaselineP{} on our Hausa medical multiple-choice set. \NBelowBaseline{} of the \NModels{} small open models we evaluate do not beat it. We lead with that because the usual summary of a multilingual result -- a percentage-point drop from English -- does not say whether what survives the drop is performance at all. On a paired English--Hausa set of \NEval{} MedMCQA items, three instruction-tuned models of 3--4B parameters lose between \GapMedGemma{} and \GapQwen{} points of accuracy, every gap significant after Holm correction. Back-translating the Hausa items into English recovers between half and two-thirds of the gap for the two general-purpose models but not demonstrably for the medically adapted one, so the deficit is substantially, though not wholly, a matter of the language of presentation. Answers are also far less stable: agreement across three option orderings falls from \ConsEnQwen{} to \ConsHaQwen
benchmark - arxiv:2611.20821 · cs.CLCuRSE-Bench: A Benchmark for Probing Cultural Blind Spots in Text-to-Image Safety EvaluationMinsuk Jang, Minseok Son, KiHyun Seol, Yeongeun Shim +1
Text-to-image (T2I) safety evaluation typically assigns each prompt or image a single culture-blind harm label, so content that is harmful only within a particular culture can pass as benign. We introduce CultureCraft, an agentic framework that constructs T2I prompts that appear benign under a general-purpose rubric yet carry a latent harmful meaning in a target culture, and use it to build CuRSE-Bench: 10,709 culture-specific prompts and 2,648 culture-general controls over six regions and ten harm axes, each drawn from a predefined knowledge base of documented harm vectors and traceable to at least two independent sources. Scoring every prompt under six region persona judges and a paired culture-blind judge, we find that the culture-blind judge under-rates culture-specific latent harm while matching the persona judges on the controls. Each persona judge, in turn, chiefly decodes its own region's latent harm under our instrument: a selective cultural blind spot that persists in deploye
agenticbenchmark - arxiv:2611.20819 · cs.AISurprise-Gated Memory for Live Speaker DiarizationLouis Mouchon, Alexis Lainault
We transpose the Surprise-Gated Memory recipe, a frozen encoder plus a calibrated novelty signal that gates a one-shot episodic store consolidated by periodic replay, from continual visual learning to live speaker diarization. The moment-to-moment question a streaming diarizer must answer, whether an arriving voice belongs to an already-heard speaker or a new one, is a known-versus-novel decision, exactly the decision the surprise gate is built for. On a frozen fine-tuned DiariZen backbone, our online system SGM-Diar recovers 4.8 of the 9.2 diarization-error points separating a naive online baseline from the offline oracle on AMI-SDM (52%) and 1.9 of 2.3 points on VoxConverse (82%), in the end-of-meeting revisable mode, with calibration frozen on development data and transferred cross-domain without adjustment. These are single-seed runs on three AMI meetings and five VoxConverse sessions, so the ordering of conditions rather than the absolute rates is the result. Beyond error rate, we
memory - arxiv:2611.20818 · cs.LGTight PAC-Bayes Generalisation Guarantees for Large Language Model JudgesTom A. Lamb, Philip H. S. Torr, Tim G. J. Rudner
Large language model (LLM) judges are increasingly used to monitor the outputs of LLM-based AI agents, yet formal guarantees on the reliability of these judges remain largely absent, limiting confidence in their use for safety-critical oversight. We consider certifying such a judge and define its risk over safety classifications of prompt--response pairs drawn from a deployed generator. We apply a PAC-Bayes framework to this setting, refining existing compression-based methodology with the use of tighter bounds and more efficient coding. We show that standard parameter-efficient fine-tuning can yield non-vacuous and even tight certificates for safety judges, for both classification and Brier risk, without holding out data for validation. We introduce LoRA-GT, which shares a single quantisation scale across layers and achieves the tightest certificates with minimal degradation in predictive performance. Across five guardrail models and four benchmarks, we find that certificate tightness
ai agentbenchmark - arxiv:2611.20812 · cs.AIThe Cybersecurity Dataset Observatory: Which Data for AI in Cybersecurity?Francesco Cerasuolo, Gabriele Mangiacapre, Antonio Pescapè
AI is increasingly used for cybersecurity tasks such as intrusion and malware detection, in enterprise, industrial, vehicular and mobile networks. Its effectiveness depends on data: models need representative network traffic, both to be trained and to be evaluated realistically. Public datasets of cybersecurity network traffic, however, are scattered across laboratory pages, repositories and aggregators, and are continuously released, corrected and withdrawn: 92 of the 234 datasets we catalog were released in 2023 or later. At this pace, surveys that list them are outdated soon after publication, since they describe the field only as it was when they were written. We present the Cybersecurity dataset observatory, a public and continuously updated catalog of cybersecurity datasets based on network traffic. New datasets are added as they are found or submitted, every record is periodically re-checked against its original source, to which it links rather than hosting the data, and every c
benchmark - arxiv:2611.20811 · cs.LGAuditing World-Model Explanations in Retail Forecasting: Faithfulness, Accuracy and Price ResponseKathik Iyer
Forecasting models that read the price and promotion plan explain each forecast as a baseline, what would sell with nothing planned, plus the lift the plan adds. We test such explanations with eight checks, six with a pass rule fixed before any result, and find that passing them does not make an explanation right. In generated store data with known effects, a world model passes the checks of construction, stability, price direction and agreement with sales and with LightGBM, yet its lifts are 1.33 to 1.40 times the truth when merchants plan into high-demand weeks, and 1.23 to 1.42 times in eight variants of that world. It still picks the right weeks to promote, but stock ordered to its lift is off by 0.44 to 0.48 of the true lift. Its stated response to a lasting regular-price change is 4.7 to 6.5 times the true -0.5, and only a measured cause finds this. On Dominick's, 38 episodes in which a store's regular prices sat apart from its zone peers give a noisy benchmark: a pooled 8-week r
world modelbenchmark - arxiv:2611.20797 · cs.ROCorrespondence-Free Camera Pose Estimation by Matching MomentsSubhransu S. Bhattacharjee, Dylan Campbell, Rahul Shome
The blind perspective-$n$-point (PnP) problem is to estimate the pose of a calibrated camera from 2D and 3D point sets with unknown correspondences. We show that, for complete, noiseless observations, fourteen permutation-invariant image moments up to order four identify a fixed camera pose for almost every 3D point set containing at least twenty points. Our learning-free method, MomentPnP, matches these moments between the observed 2D points and the projected 3D points to initialize a camera pose without correspondences and refines the pose and correspondence weights using expectation--maximization. Across the test sets of three benchmarks, our method reduces median rotation error by $5$--$170\times$ relative to the best learned baseline, both with and without outliers. We recover accurate poses even with as many outliers as inliers. On dense ETH3D scans, pose accuracy improves as the point sets grow, even as individual correspondences become harder to resolve.
benchmark - arxiv:2611.20792 · cs.ROTRR: A Thermal Interaction Engine for Robot Simulation and RGBT Data GenerationBiwen Yang, Zhe Cao, Jin Zhang, Haoyang Sun +8
A robot's actions change the thermal conditions of its next interaction: heat retained after release, carried with objects or stored beneath the surface can change a later contact's outcome despite similar current temperatures. Training trajectories must capture how these effects connect earlier actions to later observations and task outcomes. We present TRR, a \emph{Thermal Robot Renderer} that attaches spatial and depth-resolved thermal state to both contact partners, updates local heat exchange through contact, transport and separation, and generates synchronized RGB--thermal (RGBT) observations and event-local task checks on one execution clock. Controlled interventions show why retained heat, spatial transport and internal storage matter for subsequent contact outcomes. From the same state, TRR produces a paired six-task benchmark of 2,224 static--dynamic pairs, that is, 4,448 trajectories with synchronized dual-view observations, states and actions. Visible thermal history raises
benchmark - arxiv:2611.20791 · cs.ROQTOP: Scaling Humanoid Motion Tracking with Quantile CriticsMarkus Peschl, Dana Kianfar, Daniel Dijkman
Learning human motion tracking policies through reinforcement learning (RL) has emerged as a core capability that underpins a majority of recent progress in humanoid whole-body control. However, achieving generalization to unseen motions at high fidelity requires jointly scaling training to thousands of GPU and dataset hours. This work hypothesizes that motion tracking is inherently compute-bound because training on a diverse set of motions introduces learning interference, similar to multi-task RL. To tackle this, we propose Quantile Tracking Optimization (QTOP), which improves sample efficiency by introducing a distributional advantage estimate conditioned on state and reference pairs. QTOP outperforms on and off-policy RL baselines both in simulation and real-world tracking scenarios on multiple datasets by up to 10\%. Furthermore, we establish a scaling law for QTOP and find that, aside from scaling predictably, dataset size has minimal impact on performance for moderately sized GP
humanoidwhole-body control - arxiv:2611.20781 · eess.SYVerification-Guided Safety-Repair of Learned ControllersFabian Kresse, Sobhan Gharanfoli, Nikolaus Vertovec, Alessandro Abate +2
Formal verification can certify that a learned controller is safe. But what should be done when verification fails? Simply suppressing actions that lead to unsafe next states is insufficient because the dangerous action that ultimately caused the safety violation might have occurred many steps earlier. Retraining modifies the policy globally, thereby potentially introducing new failures. Instead, we introduce ReachECO, a verification-guided repair procedure that exploits the information of failed multi-step reachability analysis for synthesizing explicit constraints on controller behavior. For Differentiable Weightless Controllers, the repairs compile into exact circuit patches that leave the controller unchanged everywhere outside the targeted policy cells. We prove soundness and termination of the repair procedure for ReachECO, as well as, assuming exhaustive continuation search, completeness relative to the reachability abstraction. Empirically, we demonstrate \method's ability to s
benchmark - arxiv:2611.20779 · cs.CVEEGProbe: Do Vision-Language Models Really Understand EEG? A Benchmark for Signal-Grounded EEG Signals EvaluationHaozhe Zhu, Dan Yuan, Chenghao Gu, Feike Xu +2
Vision-language models (VLMs) have shown strong multimodal capabilities across scientific and biomedical domains, yet their ability to understand electroencephalography (EEG) signals remains unclear. We introduce EEGProbe, a signal-grounded benchmark for evaluating whether general-purpose VLMs can extract and reason from minimally processed, channel-wise EEG without task-specific feature engineering or EEG-specific adaptation. EEGProbe contains 620 EEG-centered questions across 13 subtasks and three capability categories: EEG Signal Classification, EEG Cross-modal Alignment, and EEG Neurophysiological Reasoning. To reduce shortcut-based answering, EEGProbe preserves the temporal structure of the recordings where possible, minimizes benchmark-induced processing, and applies signal-necessity testing followed by expert verification. We evaluate 11 representative open- and closed-source VLMs under a unified zero-shot protocol. The strongest model, GPT-5.6 Sol, achieves an overall score of
benchmark - arxiv:2611.20773 · cs.MAWhen Does Interaction Help? Toward a Reynolds Number for Collective IntelligenceXuening Wu, Haode Yang, Yanlan Kang, Shenqin Yin
Multi-agent systems combine distributed information through interaction, which can improve performance but also spread errors or suppress useful differences. Predicting when interaction helps is central to understanding and designing collective intelligence. We investigate whether a few dimensionless parameters can predict gains over explicitly defined non-communicating baselines across group sizes, network structures, and tasks, toward a Reynolds-number-like description of collective intelligence. Using distributed classification tasks, stochastic tracking networks, and neural cellular automata, we combine analytical and empirical evidence to examine information sufficiency, dynamical applicability, and finite-sample prediction. We construct inference systems with bounded random variables and matching specified joint moments up to total degree three, yet opposite interaction gains under a fixed readout. These moments alone cannot generally determine the gain sign. Tracking networks li
multi-agentagent system - arxiv:2611.20771 · cs.LGNemotron 3.5 Bolt E4B 4GB: An Open and Efficient Small Language Model for Fast Conversational AgentsNVIDIA, :, Adi Margolin, Aditya Malte +109
We introduce Nemotron 3.5 Bolt E4B 4GB, an open and efficient small language model further post-trained from Gemma-4-E4B-it and optimized for speed and intelligence per GB of memory used for fast conversational agents. The model carries forward Gemma's strong conversational capabilities and further improves multi-turn tool use, instruction following, and input grounding for fast, local voice/device assistants, interactive personas, and embedded/Edge interfaces. The model combines natural conversation with multi-turn agentic capabilities to power voice/device assistants, interactive characters and robotic/IoT interfaces running on GeForce RTX, Jetson, and RTX/DGX Spark. Experiments on 21 benchmarks show that Nemotron 3.5 Bolt E4B 4GB achieves the highest average accuracy while generating the fewest output tokens in both reasoning and non-reasoning modes, compared with Gemma-4-E4B-it and Qwen3.5-4B. The model is released with a DFlash drafter. With DFlash, Bolt E4B achieves $2.74\times$
memoryagentictool usebenchmark - arxiv:2611.20770 · cs.RONon-Imaging Optical Tactile Sensors for Robotics: Transduction Principles, System Integration, and Robotic ApplicationsHao Wu, Yu Feng, Haotian Guo, Huichan Zhao +2
Non-imaging optical tactile sensors encode contact through changes in optical intensity, wavelength, spectra, or photon counts, offering diverse architectures for robotic touch. Their usefulness depends on whether contact variables can be reliably distinguished from coupled mechanical and environmental effects, and whether the resulting information can be delivered in time to guide action. This review develops a unified framework linking mechanical load transmission, optical encoding, readout, and tactile inference to robotic task requirements. It examines reflective and waveguide-based devices, micro/nanofibers, fiber gratings, interferometric and resonant structures, and mechanoluminescent composites. Three considerations emerge across these approaches. First, high optical sensitivity supports accurate tactile estimation only when target variables remain identifiable under the relevant contact and calibration conditions. Second, spatial coverage and feedback timing must be assessed a
tactile - arxiv:2611.20767 · cs.LGUNITE: Consolidating Embodied VLM Specialists via Shared-Anchor One-Bit FactorizationTao Tao, Anke Tang, Enneng Yang, Zhi Hou +2
Embodied capabilities such as visual grounding, spatial reasoning, navigation, and driving understanding are often developed in separate vision-language specialists fine-tuned from a common pretrained model.Deploying the specialists separately multiplies storage, whereas merging them into one dense model forces conflicting task updates to share parameters and loses part of each specialist's expertise.We introduce UNITE, a framework that consolidates such specialists into a compact shared-backbone mixture of experts. UNITE keeps the feed-forward block of each specialist as a routable expert and represents the expert bank through a shared-anchor one-bit factorization: a shared floating-point anchor plus expert-specific, row-scaled binary deviations.The anchor is optimized jointly with the scales and signs through closed-form alternating updates, using only source weights and no calibration data.Router adaptation then adjusts token-level expert selection with the backbone and experts froz
embodiedbenchmark - arxiv:2611.20759 · cs.CLInstant Weight-Space Consolidation for Long-Term Memory in LLMsManan Suri, Dinesh Manocha
Agentic memory systems give LLMs persistent memory by extracting experience into text and retrieving it at query time, but the model itself does not evolve: however many interactions it has had, it starts each query the same. Extraction errors compound, implicit cross-session patterns that no single fragment states cannot be surfaced, and the construction cost grows with every conversation. Fine-tuning on past interactions does change the weights, but requires backpropagation at serving time and degrades on long histories. We introduce CLAM (Carried Latent Adapter Memory), which combines the strengths of both: a recurrent hypernetwork performs parametric consolidation, compiling a stream of session activations into a low-rank adapter in a single forward pass with no gradient steps and no language-model generation calls, while a complementary episodic buffer supplies verbatim detail at inference. A gated latent state carries knowledge across sessions in fixed size, so the parametric mem
memorypersistent memoryagentic - arxiv:2611.20755 · cs.CLGPRM: Global Perspective Process Reward ModelZifan Zhang, Zhengbo Jiao, Shaobo Wang, Jing Zhou +9
Process Reward Models (PRMs) are widely used for step-level evaluation of multi-step reasoning in large language models, but existing methods are often limited by weak historical dependency modeling and insufficient global awareness. They typically fail to leverage prior step evaluations for coherent judgment and lack visibility into subsequent reasoning, leading to suboptimal assessment of intermediate steps. To address this issue, we propose the Global Perspective Process Reward Model (GPRM), which jointly models the problem, previous reasoning steps and their evaluations, the current step, and future steps to enable history-aware and globally informed process evaluation. We further introduce a structured evaluation framework along four dimensions---look back, look ahead, self-check, and goal alignment---to assess prior consistency, future plausibility, step validity, and task alignment. Based on this framework, we construct datasets for supervised fine-tuning (SFT) and reinforcement
agentevaluation framework - arxiv:2611.20741 · cs.RORobotic Reward Models in the Foundation-Model EraGuoqiang Liang, Yiqi Lin, Shangchen Zhou, Wenzheng Zeng +3
Robotic reward models (RRMs) use foundation models to replace per-task reward engineering with reusable, task-conditioned evaluation. Yet under one name, their outputs answer different questions, from whether a task succeeded to which execution is better. This survey organizes 114 methods into five semantic families by the question their native outputs answer: Outcome, Absolute Progress, Relative, Temporal Value, and Hybrid. Within this taxonomy, we review how model design determines what an evaluator can judge, how raw experience becomes supervision, and how existing evaluation practice tests the result. The review exposes a gap: agreement with demonstration order and downstream policy gains do not show whether a reward tracks task fulfillment. We therefore introduce RRM-Bench-1K, 1,117 real-robot videos of closed-loop executions with task-state and camera-evidence labels, and four contrast suites that separate the credited condition from its usual cues. A unified study of 27 evaluato
evaluator - arxiv:2611.20737 · cs.CVCLIFT: Contrastive LiDAR Instance Features for 3D Pedestrian TrackingGerman Devchich, Timur Akhtyamov, Denis Fatykhoph, Gonzalo Ferrer
A social robot has to keep track of the people around it over time to plan its motion, a task handled by 3D multi-object tracking. For the LiDAR modality, trackers usually sit on top of a detector and link its detections across frames with a separate motion model or re-identification network. We instead show how a 3D LiDAR detector itself can be turned into a tracker by adding an instance-level contrastive objective on its intermediate voxel features. Each embedding is read from the voxels inside a detected box, and boxes are matched across frames by cosine similarity and the Hungarian algorithm, without a motion model. The detector backbone is fine-tuned together with this objective, which learns identity-aware features while leaving detection accuracy essen- tially intact. Because the method uses only the voxel features that Li- DAR detectors already produce, it is backbone-agnostic. On the JRDB 3D tracking leaderboard our LiDAR-only tracker ranks first on HOTA (38.58) and IDF1 (37.9
leaderboard - arxiv:2611.20736 · cs.AIGainCtrl: Predicting Remaining Gains for Efficient LLM Agent ExecutionKangwen Zhao, Jianfeng Cai, Ruopei Sun, Mingxiao Feng +3
Large language models (LLMs) are widely used to power agents that interact with tools and environments to solve complex tasks. Repeated attempts improve the quality of tool-using agents, but continued execution incurs additional costs even when it is unlikely to improve the best result ultimately returned. This raises the question of when to stop an attempt to reduce execution cost while preserving task-level Best-of-N quality. To address this question, we introduce GainCtrl, a gain-based execution controller for LLMs. In GainCtrl, a lightweight CPU predictor uses interaction histories to estimate expected positive gains over retained scores, which are then used to guide stopping decisions with calibrated thresholds to limit the loss in mean Best-of-N task quality. We evaluate the controller through trajectory replay across three environments and three LLM actors. Comparisons of interpolated quality-cost Pareto frontiers show that GainCtrl generally uses fewer tokens than baselines at
agentllm agent - arxiv:2611.20733 · cs.AIMemEmo: Evaluating Emotion in Memory SystemsPeng Liu, Zhen Tao, Jihao Zhao, Ding Chen +4
Memory systems address the challenge of context loss in Large Language Models during prolonged interactions. However, as an essential tool for warm companionship and personalized services, the effectiveness of memory systems in processing affective information remains underexamined. To bridge this gap, we propose an emotion-enhanced memory evaluation benchmark to assess the performance of mainstream and state-of-the-art memory systems in integrating and harnessing emotional information. First, we design a benchmark called \textbf{MemEmo} to assess the affective reasoning and tracking capabilities of memory systems across four dimensions: emotional information extraction, emotional memory updating, emotional memory question-answering, and emotional conflict detection. Second, we construct the \textbf{H}uman-\textbf{L}ike \textbf{M}emory \textbf{E}motion (\textbf{HLME}) dataset to provide an empirical foundation for scrutinizing memory systems. Finally, We evaluate five representative me
memorybenchmark - arxiv:2611.20728 · cs.CLANCHOR: Cluster-Grounded 5W1H Extraction via Large-Small Model CollaborationHongyu Jiang, Chenang Cai, Lei Gao, Zhao Kang +2
Deep semantic understanding of events requires recognizing related events and distinguishing why an event occurs from how it is carried out. 5W1H extraction provides a structured foundation for such understanding by organizing event information around six fundamental questions. However, single-event slot filling provides limited support for these distinctions. We propose ANCHOR, a cluster-grounded framework that combines a large language model (LLM) with a small language model equipped with discriminative heads for accurate 5W1H extraction. The LLM proposes candidate events and generates their initial 4W records, which guide the selection of an event cluster and the subsequent anchoring of its main event. A shared document encoder provides evidence for joint 4W refinement, and two separate sets of boundary and role-scoring heads then extract why and how under event-specific constraints, respectively. Document representations are reused across events and stages, and unsupported roles ca
benchmark - arxiv:2611.20720 · cs.ROAVD: A Multi-Sensor Teleoperated Wheelchair Platform with Safety-Aware ControlNamai Chandra, Madhur Thareja, Boby George
Electric wheelchairs predominantly rely on manual joystick interfaces, which present critical barriers to independent mobility for individuals with severe motor and visual impairments. This paper presents AVD (Autonomous Vehicle Drive), a dependable teleoperation architecture for assistive mobility that unifies perception, communication, embedded control, and remote supervision in a single low-cost platform built entirely from commercially available hardware. The central engineering contribution is a safety-oriented design framework: control authority follows a static hierarchy, a manual emergency stop and the occupant's joystick pre-empt remote commands, and the on-board sensing layer has stop-only authority. The compute design partitions primary and safety functions and specifies a heartbeat-driven safe-state transition for the fault cases considered. The sensing stack this framework depends on is characterized against measured data: per-channel acquisition latency, a 2,025-scan LiDA
teleoperation - arxiv:2611.20719 · cs.ROACiT: Proprioceptive-Prior Action Chunking with Inverted TransformersHaixiang Wu, Zhihao Guan, Yang Yang
Visuomotor policies condition action prediction on visual and proprioceptive observations. While historical proprioceptive sequences carry rich dynamic priors, current methods treat them as equal to high-dimensional vision, causing the proprioceptive signal to be overwhelmed. To identify what is lost in equal-conditioning fusion, we analyze proprioceptive data and find that strong inter-joint coupling is the key action-determining structure. This finding aligns with the neural sensorimotor loop, in which proprioception first forms a body-centered prior and vision is later calibrated via parietal remapping. Inspired by this, we propose ACiT, Proprioceptive-Prior Action Chunking with Inverted Transformers, which reinforces historical joint-coupling as the primary condition with vision as a later calibration. This design is instantiated through three modules: Inverted State Encoding builds a proprioceptive prior via per-joint tokens and inter-joint self-attention. State-Visual Grounding l
vlaaction chunkingliberorobotwin - arxiv:2611.20718 · cs.ROTaF-Foundation: Wrench-Grounded Cross-Sensor Tactile RepresentationsJiaming Jiang, Yuzhe Huang, Baokun Yang, Hao Liang +7
Tactile sensing is essential for contact-rich manipulation, yet heterogeneous sensing mechanisms and observation formats hinder cross-hardware transfer. Since physical contact is governed by Newtonian mechanics, we ground tactile representations in six-dimensional wrenches, establishing a common physical reference across sensing mechanisms. Specifically, we pair tactile observations from all sensors with wrench measurements, enabling shared physical supervision without requiring direct tactile-to-tactile pairing between sensors. To support learning at scale, we develop a dedicated acquisition platform and collect over 100 million paired tactile observations and wrench measurements from 24 sensors spanning four families of sensing principles. Building on this dataset, our framework combines sensor-specific input adaptation with a shared tactile encoder. By learning to reconstruct wrenches, the encoder maps observations from different sensors into a shared, continuous feature space groun
manipulationtactilebenchmark - arxiv:2611.20715 · cs.LGCalibrate Once, Grow the Vocabulary: Candidate-Independent Scores for Conformal Risk Control in Open-Vocabulary SegmentationMd Tanveer Hossain Munim, Bijoy Ahmed Saiem, Muhammad Abdullah Adnan
Open-vocabulary segmenters accept class names at inference. Conformal risk control can bound how often their prediction sets miss the true label, but we show that this bound breaks when names are added after calibration. On ADE20K, a threshold calibrated for 10% risk at 150 names reaches up to 15.4% risk at 845 names in released CAT-Seg models. The risk exceeds the target on every calibration split, yet the prediction sets grow far less than the target requires. In SED they instead grow larger than needed. The cause is that pruning, cross-label attention and vocabulary normalisation make each label's score depend on the other requested labels. Scores can move either way, so a threshold recalibrated at a larger vocabulary need not hold at smaller ones. We remove this dependence without training by scoring each label against a fixed reference bank of names. A label's score then stays the same when names are added, so one calibration keeps its guarantee as the vocabulary grows. The guaran
benchmark - arxiv:2611.20711 · cs.LGAgentFaultBench: State-Verified Evaluation of Tool-Fault RecoverySachin Gupta
Does recovery advice help an agent finish a task after a tool fails? AgentFaultBench evaluates LangGraph and CrewAI agents performing specified tasks in an isolated GitLab installation. We inject tool errors and check both the resulting files and issues and the required reads, verification and final response. The prompt supplies the expected issue and report text, so the task tests execution rather than content generation. In the primary comparison of 1,944 planned runs, advice changes success by -8.33, 0.00 and +1.06 percentage points relative to a neutral reminder; missing outcomes leave the latter two comparisons inconclusive. Later experiments test a revised recovery policy. GPT-4o mini improves, while Sol and Claude already complete the recoverable cases. In a separate 624-run study on six related tasks not previously used by that policy, the guided condition records 161 correct final states and one unknown outcome in 192 planned runs, versus 142/192 with neutral feedback. The gai
agentai agent - arxiv:2611.20710 · cs.ROSG-ODOM: Event-Driven Proprioceptive State Estimation with Stability-Gated Contact Updates for Legged RobotsFrancesco Iotti, Chiyun Noh, Franco Angelini, Manolo Garabini +3
Reliable proprioceptive state estimation for legged robots fuses inertial sensing with kinematic sensing from leg encoders. The latter assumes a foot is stationary during the stance phase; however, physical foot touchdown does not necessarily provide a stationary reference for odometry, particularly during impact and load transfer. Repeatedly applying kinematic corrections throughout the same contact phase can introduce redundant or unreliable information. We present SG-ODOM, a proprioceptive odometry framework that uses stability-gated contact selection to schedule kinematic updates. Following physical touchdown, a stability gate accepts the contact for state estimation only after the foot reaches a sufficiently stable support condition. The resulting validated touchdown event triggers an estimator correction, while additional timeout updates bound the maximum duration of uninterrupted IMU propagation. The framework is evaluated on multiple legged platforms, both quadrupedal and human
humanoidquadruped - arxiv:2611.20698 · cs.CVSTARS: Stage-Adaptive Responsibility Separation for Shadow Removal in Remote Sensing ImagesGaoliang Song, Chengjie Li, Pengju Liu, Chenchen Chang +2
Shadows are a ubiquitous radiative attenuation phenomenon in remote sensing images and can severely affect high-resolution remote sensing interpretation tasks. In shadow removal, although introducing a mask prior provides explicit positional information for the restoration network, it also causes non-shadow regions to be modulated by the shadow prior, producing a systematic brightness shift; notably, this phenomenon is independent of whether the mask is accurate. In remote sensing images, non-shadow regions usually occupy a larger proportion, so a small shift over these regions incurs a much higher cost on whole-image metrics than an equal gain in shadow regions. Adopting a task-decoupling perspective, we therefore decompose shadow removal into two tasks, shadow-region restoration and non-shadow-region identity, within a cascaded architecture, and handle them sequentially. Based on this insight, we propose STARS, a cascaded shadow removal framework that separates regional responsibilit
benchmark - arxiv:2611.20695 · cs.AIIndustrialKnowEval: A Benchmark for Evaluating Multi-Level Industrial Knowledge in LLMsHaiteng Wang, Haoze Li, Jingwei Guo, Jingbo Zhou +2
Large language models (LLMs) are increasingly explored for industrial tasks, yet few benchmarks evaluate their mastery of industrial knowledge across different types of cognitive demand. We introduce IndustrialKnowEval (IndKnow), a benchmark for industrial knowledge and engineering question answering. It contains 43,282 samples covering 21 task types across seven industrial sectors, including aerospace, electric power, and metallurgy, organized into four levels of cognitive demand: memory, comprehension, reasoning, and engineering application. Enterprise-based domain specialists and experienced researchers participate in all stages of dataset construction and quality control. We conduct a series of experiments on 18 representative LLMs, with domain experts deeply involved in the evaluation. Building upon IndKnow, we propose the Industrial Cognitive Capability Gap (ICCG) metric to characterize the performance differential between foundational industrial knowledge and higher-order reason
benchmark - arxiv:2611.20692 · cs.ROARMFUL: Learning Whole-Arm Grasp Configurations for Single-Arm Pickup of Large ObjectsXinyi Yu, Yunshuang Li, Daniel Seita, Gaurav S. Sukhatme
Many everyday objects, such as storage boxes, baskets, and chairs, are too large or heavy for a gripper to grasp, yet a single arm can still pick them up by wrapping around and supporting them. Such grasps are hard to learn: wrapping motions are difficult to teleoperate, and existing datasets contain little whole-arm pickup data. We introduce WAG (Whole-Arm Grasp Dataset), the first large-scale dataset for single-arm pickup of large objects, with 1.7~million joint configurations across 793 objects. We sample configurations directly in joint space, execute each through squeeze and lift primitives in simulation, and stress-test survivors under active shaking and external forces. To demonstrate the utility of WAG and accomplish the whole-arm pickup task, we develop ARMFUL, a whole-arm pickup pipeline that pairs a diffusion model trained on WAG, generating whole-arm grasps from the object's point cloud, with a motion planner that reaches them collision-free before the same squeeze-and-lift
frankagrippergrasp - arxiv:2611.20688 · cs.CVVIVID: A Visual Harness for Agentic Video EditingWenhan Yu, Zhaoxi Zhang, Zhilong Tan, Weikang Li +1
Video editing requires agents to make editing decisions grounded in audiovisual observations and precise temporal localization, and to evaluate their effects in rendered outputs. General-purpose coding agents possess strong comprehension, reasoning, and code execution capabilities, while media interaction and evidence management mechanisms tailored to video tasks can further unlock their editing potential. We introduce VIVID, a visual harness that supports multiple native coding agents in video post-production without additional model training. VIVID provides tools for on-demand video inspection, audio analysis, and transcription, and preserves observations alongside their corresponding media versions and temporal coordinates. Agents can actively retrieve these observations as they reason, perform edits using their native coding and shell tools, and inspect rendered outputs to guide subsequent revisions. On AgenticVBench, comprising 100 tasks across Assembly, Repair, Sequencing, and Re
agentic - arxiv:2611.20686 · cs.AIUbiHearo: Bringing Scenario-Aware Sound Guidance to DHH Users with Mobile AgentsFengmin Wu, Sicong Liu, Zimu Zhou, Chenren Xu +3
For Deaf and Hard-of-Hearing (DHH) users, effective sound awareness goes beyond detecting \emph{what} sound occurs; it requires understanding \emph{what it means}, \emph{how urgent it is}, and \emph{what to do}. However, existing mobile assistants largely reduce acoustic scenes to isolated labels or generic alerts, failing to provide contextual guidance when the same sound implies different risks across scenarios. We present \sysname, a mobile hearing assistant agent that closes the sensing--reasoning--action loop for \textit{scenario-aware} and \textit{personalized} DHH assistance. \sysname adopts a \textit{neuro-symbolic} design that transforms continuous acoustic and physical context streams into structured intermediate representations (IRs) for controllable reasoning and guidance. It introduces three key mechanisms. \textit{First}, a perception-thresholded sensing mechanism enables energy-efficient always-on awareness by activating processing only for perceptually meaningful events
agenthuman-in-the-loop - arxiv:2611.20680 · cs.CVRoom for Supervision: Training-Free Dataset Distillation for Object DetectionZiang Gan, Qi Zhu, Libao Zhang
Dataset distillation (DD) reduces storage and training costs by compressing large datasets into small collections. Most existing methods focus on image classification, where each image typically has a single category label. In contrast, object detection requires accurate category and bounding box annotations for multiple objects, together with visual content consistent with those annotations. Existing detection distillation pipelines rely on learned models and can alter the spatial relationships between objects and their backgrounds. We introduce GRASP, which packs dense native supervision into limited pixels without network training or inference. GRASP jointly searches over image selections and spatial layouts using convex allocation subproblems that penalize insufficient object resolution, excessive image compression, and padding that source context cannot fill. During composition, each selected image is cropped around all annotated objects and expanded into surrounding context witho
grasp - arxiv:2611.20676 · cs.ROLearning and Verifying Progress: Closed-Loop Agentic Planning with Progress-Enhanced Execution for Long-Horizon Vision-Language NavigationKangjie Zhou, Sukai Wang, Yuhan Xiao, Qinghang Liu +6
Long-horizon vision-language navigation (VLN) requires agents to track instruction progress and recover from local errors before they propagate across subtasks. However, executors trained on expert trajectories encounter policy-induced state shifts during deployment, and their stop signals do not reliably indicate subtask completion. We introduce Learning and Verifying Progress for VLN (LeaP-VLN), an agentic VLN system that couples progress-aware execution with independent progress verification through plan--execute--verify--recover cycles. LeaP-VLN leverages progress both explicitly, through a progress verifier that verifies subtask completion before the agent advances and triggers dual-phase recovery upon route deviation, and implicitly, by co-training the executor on progress reasoning and action prediction to enhance the executor's progress awareness and corrective control. To adapt the executor to this closed-loop interface, we propose a two-stage progress-enhanced curriculum: beh
agentagentic - arxiv:2611.20674 · cs.CLLEAF-OPD: Learning to Forget from Privileged Contexts via On-Policy DistillationZirui Pang, Zhijie Deng, Yuancheng Wang, Qihan Yu +3
LLM unlearning aims to erase targeted knowledge from large language models while preserving their utility. However, many existing methods suppress target knowledge without explicitly teaching the model how to respond to queries about it. We propose LEAF-OPD (LEArning to Forget from Privileged Contexts via On-Policy Distillation), in which a frozen teacher provides token-level supervision on student-generated responses. On forget data, the teacher uses privileged contexts (instructions not to reveal knowledge about forget targets) to guide forgetting. On retain data, the teacher provides supervision without privileged context to preserve model utility. The student operates without privileged context in both cases, learning how to respond when encountering knowledge that should be forgotten. A key challenge is that the teacher often provides effective supervision only at the first few tokens of student trajectories. We introduce Supervision Signal (SS) to measure corrective supervision o
benchmark - arxiv:2611.20671 · cs.AIMASTA: A Feedback-Scheduled Multi-Agent System for End-to-End Tamarin Modeling and AnalysisYongkang Xiao, Qiyi Deng, Jing Chen, Min Shi +2
Tamarin-based formal verification of security protocols requires expert-authored multiset rewriting rules and temporal-logic lemmas. Translating a natural-language protocol description into a complete formal model with an LLM is hard because syntactic, reachability, and property-level errors compound across the multi-step modeling process, and a single agent that both authors and self-evaluates cannot reliably localize them. We present MASTA, a multi-agent system that addresses these difficulties through three innovations. First, an operational--evaluative role decomposition assigns rule modeling to an Architect, property modeling to an Auditor, and pure dispatch to a dedicated Arbiter that holds no authoring duties. Second, the Arbiter implements a Tamarin-anchored scheduler that dispatches the next agent based on the latest verifier verdict and routes problem-tagged fixes through a modification list, with a six-class problem taxonomy and an LLM deduplication check preventing routing
agentmulti-agentagent systembenchmark - arxiv:2611.20666 · cs.LGConditional Diffusion Models for Amortized Bayesian System IdentificationMike Hänseroth, Niklas Wahlström, Thomas B. Schön, Thomas Seel +1
Generative diffusion models rapidly solve Bayesian inverse problems, such as image restoration and inpainting, with a single trained model and without retraining for each new observation or task. Bayesian system identification, which typically involves inferring the hidden states and parameters of a dynamical system from measurements, poses a closely related inverse problem. Yet it is often addressed with computationally heavy Monte Carlo methods that must be rerun for every new measurement record. Moreover, these methods are restricted to a single inference task, and their cost grows with the state dimension and the record length. To alleviate these restrictions, we interpret Bayesian system identification in hiddem Markov models (HMMs) as a general Bayesian inverse problem, taking advantage of conditional diffusion models. Specifically, a transformer-based diffusion model is trained once on short segments of simulated trajectories to approximate the joint posterior of states and para
benchmark - arxiv:2611.20655 · cs.CLMetaAct: Self-Evolving Executable Action Constraints for Embodied Language AgentsJiaxing Zhao, Hanwen Wang, Di Liu, Shaoteng Liu +8
Self-evolving language agents have emerged as a promising paradigm for continuously improving agent behavior through experience. However, most existing approaches evolve textual artifacts, which can influence action preferences but cannot reliably enforce action admissibility in embodied environments with discrete and often irreversible actions. We propose MetaAct, which shifts self-evolution from textual guidance to executable action constraints by synthesizing GEAR (Grounded Executable Action Rules) from execution feedback. Each rule is represented as a typed specification, compiled into a deterministic guard, and validated offline through trajectory replay without additional model calls or environment interactions. Across four benchmarks, MetaAct improves task performance; on ScienceWorld and WebShop, violations drop from 70% to 0% and 40% to 1%, with score gains of +10.1 and +8.0. Our code is available at https://github.com/BaiCubed/MetaAct.
embodiedagentself-evolvingbenchmark - arxiv:2611.20651 · physics.opticsTelecom-band optical parametric oscillation in a passive silicon microring resonator at cryogenic temperaturesShoichiro Yasui, Xuejun Xu, Hidetaka Nishi, Haruki Sanada +1
At telecommunications wavelengths, free carriers generated by two-photon absorption limit optical parametric oscillation (OPO) in silicon. Here, we demonstrate telecom-band OPO in a passive silicon microring resonator without electrical carrier sweep-out, driven by a single continuous-wave (CW) pump at 4.5 K. We achieve an on-chip OPO onset power of 2.25 mW and a total conversion efficiency of 1.9% into all parametric sidebands at a pump power of 4.8 mW. Despite CW pumping, the parametric output emerges as a train of pulses synchronized with microsecond-scale self-pulsing. At higher pump powers, OPO emission is absent during intervals of sustained sub-GHz oscillations in the pump transmission, narrowing the temporal window for OPO. Furthermore, by adjusting the pump detuning, we observe OPO at temperatures up to 55 K. These results demonstrate that nonlinear cavity dynamics govern the temporal structure and operating conditions of cryogenic silicon OPOs, providing new insights into con
microringquantum photonic - arxiv:2611.20648 · cs.CVCompetent Autonomous Driving Through Fine Tuning Vision Language Action Models with Symbolic Traffic RulesBarkin Dagda, Vinal Asodia, Iman Sharifi, Peng Wei +1
Vision-language-action (VLA) models offer a unified approach to perception, reasoning, and control for autonomous driving, but accurate action prediction alone does not establish compliance with traffic regulations. This paper presents a rule-guided VLA framework that incorporates symbolic traffic-rule feedback into policy optimisation. The framework combines supervised fine-tuning on expert demonstrations with reinforcement learning (RL) refinement using a reward that encourages encoded-rule compliance while penalising deviations from expert actions. Selected provisions of the United Kingdom Highway Code are formalised as executable first-order logic rules using PyReason, covering speed adaptation, following distance, and road-user priority. A frozen vision-language model extracts scene predicates, which are combined with vehicle-state information to assess the projected consequences of candidate control actions. Offline evaluation on held-out data collected in the CARLA simulator com
vision-language-actionvision language actionvlavla policy - arxiv:2611.20644 · cs.LGInstrumenting Failure: Scoring Language Model Agents at Every Decision Node in Imperfect-Information GamesMichael Kingston, Sonit Singh, Alan Blair
Agent evaluations report task success and rarely say what an agent did at each decision. We score every decision node of three 12B to 14B instruct-tuned language models (LMs) playing five zero-sum imperfect-information games against four opponents on eight properties from decision and game theory, five of them against a CFR+ equilibrium reference. At each node, the LM also states a belief about the opponent's hidden state in a separate call. Without memory and at temperature 0, stated beliefs sit 0.250 to 0.288 in total variation distance from the equilibrium posterior, and where the action responds to the state its agreement with that belief exceeds chance by -0.04 to 0.19. The two calls frame the opponent differently, so the result is a behavioural inconsistency and makes no claim about internal beliefs. Under one prompt and greedy decoding, the LM's action never varies with the state in 10 of the 38 cells where the reference varies its own. Memory lowers the constant-action share in
memoryagent - arxiv:2611.20637 · cs.CVCounterfactual Feature Steering for Adjacency-Preserving Identity UnlearningFNU Shivam, Layth Sliman, Ramana Reddy, Donald Adjeroh +2
Identity unlearning in text-to-image diffusion models is increasingly important for privacy, safety, and ethical deployment, as it aims to remove the model's ability to generate specific individuals. In this work, we study identity unlearning with \emph{adjacency preservation}, where the goal is to remove a target individual's identity while preserving related non-target identities. We propose Counterfactual Feature Steering (CFS), a method requiring one-time SAE training that performs bounded inference-time concept editing in the latent space of a sparse autoencoder attached to the text encoder. CFS derives a target direction by contrasting forget and retain distributions, then applies a per-token projection proportional to each token's current alignment with this direction, yielding counterfactual representations that suppress the target while limiting collateral damage. To evaluate adjacency preservation, we introduce the Adjacency-Weighted Unlearning Score (AWUS), which measures th
benchmark - arxiv:2611.20635 · cs.LGContrastive Activation-Guided Zeroth-Order Preference Optimisation for Memory-Efficient LLM AlignmentWeiyun Huang, Alessio Galatolo
Direct Preference Optimisation (DPO) simplifies reinforcement learning from human feedback, but still relies on traditional backpropagation and its associated activation-memory cost. Zeroth-order (ZO) optimisation avoids backpropagation, yet isotropic perturbations can produce noisy updates for preference objectives defined by differences between chosen and rejected responses. We introduce Contrastive Activation-guided Zeroth-Order Direct Preference Optimisation (CA-ZODPO), which constructs a low-rank search subspace from the activation discrepancy between the chosen and rejected responses in each preference pair. Perturbations are restricted to this subspace, while momentum is maintained in low-rank coordinates to preserve the memory advantages of ZO training. On Qwen2.5-1.5B with Anthropic HH-RLHF, CA-ZODPO outperforms common ZO baselines and remains competitive with first-order DPO in pairwise preference evaluation. It also reduces peak GPU memory by about 59\% at a sequence length
memory - arxiv:2611.20631 · cs.ROInstructions Before Actions: Prospective and Retrospective Language Supervision for Driving Vision-Language-Action DataAngel Martinez-Sanchez, Ross Greer
Driving vision-language-action (VLA) models can accept navigation instructions, yet predicting a driver's trajectory does not establish that a model follows the instruction. We study how driving-language supervision is collected and evaluated, distinguishing prospective instructions issued before an action from retrospective labels written after observing it. We develop ADVLAT-Engine to synchronize spoken instructions, road-scene video, and vehicle trajectories, and compare three navigation apps on 48 replayed routes. On 87 maneuvers announced by both Google Maps and a passenger, the passenger uses more scene-dependent language. We evaluate four open driving VLAs before and after fine-tuning on prospective navigation-app drives. Fine-tuning reduces passenger-drive trajectory error by 23-59%, but instruction effects remain inconsistent. At real intersections, counterfactual instructions reveal stronger responsiveness: fine-tuned Alpamayo~2 Super gains 31 and 18 accuracy points at 15 and
vision-language-action - arxiv:2611.20629 · cs.CVEvent-Guided Stereo Video Super-ResolutionHanyu Xuan, Youtao Yao, Cheng Zhao, Kun Li +3
Stereo Video Super-Resolution (SVSR) aims to reconstruct high-resolution stereo videos by exploiting temporal cues across consecutive frames and complementary spatial cues between stereo views. However, existing methods struggle to recover high-frequency texture details under rapid motion and complex illumination, compromising reconstruction quality and stereo consistency. Event cameras provide valuable motion and texture cues with high temporal resolution and wide dynamic range, yet their potential for SVSR remains underexplored. To bridge this gap, we present the first investigation of event-guided SVSR and propose a novel framework. Specifically, we develop an Event-Guided Texture Enhancement (EGTE) module that combines bidirectional temporal feature propagation with event-guided texture refinement to restore high-frequency details in complex motion regions. Subsequently, we introduce a Cross-View Texture Interaction (CVTI) module to model cross-view spatial correspondences and aggr
event camera - arxiv:2611.20628 · cs.CVPixelBench: Evaluating Language Models for Pixel-Art CreationMutian Hong, Jianxuan Yao, Mingkun Yue, Changkai Li +2
Pixel art is widely used in games and digital content creation, requiring readable forms, coherent spatial relationships, and harmonious colors under limited resolution and palette constraints. Language models show growing promise in visual creation, yet their pixel art creation capabilities lack systematic evaluation. We introduce PixelBench, formalizing text-to-image generation and instruction-guided editing for pixel art and providing a benchmark for these tasks. We further present a creation environment in which language model agents invoke drawing tools through code to generate and edit pixel art. We assess the quality of generated artwork through human evaluation and employ a multimodal large language model as a judge, validating the agreement between its automated assessments and human preferences. We compare language models and configurations across diverse subjects and canvas sizes. In our experiments, GPT-6 Astra achieves both the highest average judge score and the best aver
benchmark - arxiv:2611.20625 · cs.CVASCG: Adaptive Spatial Classifier Guidance for Surgical Concept Suppression in Diffusion ModelsYounghwan Kil, Joonhyeong Park, Giung Nam, Juho Lee
While text-to-image (T2I) diffusion models achieve remarkable quality, they risk producing harmful content. Existing safety mechanisms often rely on indiscriminate, global interventions that induce unintended semantic drift and degrade image quality. To address this, we propose Adaptive Spatial Classifier Guidance (ASCG), a necessity-driven framework that employs probabilistic monitoring to activate guidance selectively upon harm detection. By leveraging Grad-CAM for localized suppression, ASCG effectively removes harmful concepts while preserving superior semantic fidelity compared to conventional methods. Furthermore, we overcome the limitations of traditional metrics like NudeNet---which fails to assess semantic integrity---by introducing a VLM-based evaluation framework to provide a fine-grained, robust analysis of both safety and context preservation. Our results demonstrate that ASCG not only balances safety and fidelity but also exhibits remarkable plug-and-play synergy with exi
evaluation framework - arxiv:2611.20623 · cs.CVExample-Based Spatial Guidance for Training-Free Concept Erasure in Diffusion ModelsYounghwan Kil, Giung Nam, Joonhyeong Park, Jinwoo Shin +1
Text-to-image (T2I) diffusion models often synthesize policy-violating content, necessitating robust safeguards and rigorous evaluation frameworks. However, current approaches remain largely monolithic in both execution and evaluation, limited to coarse-grained categorization within broad unsafe concepts. Such approaches fail to account for the specific constituent factors of a violation, resulting in weak learning signals and allowing model collapse to bypass safety checks. To overcome these limitations, we propose Example-Based Spatial Guidance (EBSG), a training-free method that utilizes user-editable exemplar packs to provide granular spatial guidance for precise concept steering. Unlike existing methods that coarsely steer the entire image away from a broad concept, EBSG decomposes these categories into specific sub-concepts via exemplar text-image pairs, providing explicit, localized steering signals for more robust safety control. Furthermore, we introduce a vision language mode
benchmarkevaluatorevaluation frameworkevaluation protocol - arxiv:2611.20616 · cs.AIGEAR$^2$: Pruned Experts Live On through Gated Reuse of Activated Experts and a Missing-Mass ResidualLinliang Chen, Yan Zhong, Wang Kang, Sai Li +2
Mixture-of-Experts (MoE) Large Language Models (LLMs) activate only a few experts per token while generally keeping all the experts in GPU memory for high efficiency. Expert pruning reduces this GPU memory footprint cost, yet discards the contribution of pruned experts. Existing remedies either rely on static substitutes, merge experts at the cost of altering their functions, or require costly recovery training. In this paper, we find that the functionalities of pruned experts can be efficiently reconstructed from the activated remaining experts. We propose GEAR (Gated Expert Activation Reuse), which reuses the scaled outputs of reliable activated experts without any training or additional expert computation. In addition, we propose GEAR$^2$ to further recover what reuse cannot express with an ultra-lightweight residual. On five MoE LLMs from 16B to 106B parameters with 25% and 50% of experts pruned, GEAR improves over its training-free counterparts in nearly every setting (up to 4 pp.
memorybenchmark - arxiv:2611.20612 · cs.MALearning Design Taste through Verifiable Synthetic Environments and Agentic Reinforcement LearningSujay choubey, Siddhant Saxena, Sarthak Kapil, Suhas Motwani
Code-generating language models produce user interfaces that build and run, but the pages are often weak in hierarchy, spacing, typography and composition. We study whether a small code model can learn design taste through reinforcement learning in verifiable synthetic environments. The method has three components. First, a layered verification framework splits design quality into hard gates, render-grounded metrics, rubric judges, within-group relative preference and a blinded expert panel. A signal enters the reward only if it passes an admission protocol that tests agreement with experts, causal sensitivity, resistance to hack traps and stability. Second, following SCALER [1], synthetic design environments are parameterised generators with measured difficulty ladders: base pages are written once by an agentic coder, and instances are produced by syntax-tree perturbation operators and constraint samplers, scheduled by an adaptive curriculum with a quarantine rule against reward over-
agentic - arxiv:2611.20609 · cs.LGNewton-PINet: A fast physics-informed neural network with Newton linearization for meta-learning nonlinear PDEsYuchen Fan, Chang Wei, Pao-Hsiung Chiu, Chin Chun Ooi +2
Scientific machine learning has opened new avenues for solving parameterized partial differential equations (PDEs), enabling models to learn a family of PDEs and generalize to unseen instances. In this context, data-driven operator learning methods typically require large training datasets, while physics-informed neural networks (PINNs) suffer from difficult optimization and limited generalization, especially for nonlinear PDEs. We propose Newton-PINet, a physics-informed network enhanced by Newton linearization, offering an effective meta-learning framework for nonlinear PDEs. Newton-PINet (i) employs a physics-informed multilayer network with skip connections, where the output-layer weights are solved by least squares; (ii) adopts a two-stage learning strategy that first leverages gradient-based training to learn robust representations from the available training tasks, and then performs gradient-free fine-tuning on the output layer for fast task-specific generalization; and (iii) in
benchmark - arxiv:2611.20608 · cs.AISame Papers, Different Verdicts: Tracing Version Drift in LLM Peer ReviewJizhou Guo
Venues and benchmarks read LLM review scores as measurements of a paper, while providers replace their models within months. We audit twenty frontier models from OpenAI, Google and Anthropic on ICLR 2027 submissions while they were under review, with tests registered before the decisions. Between an earlier and a later version of Claude Opus and of Gemini, the share of papers placed on the accept side falls by 52 and 53 percentage points on 183 and 60 randomly drawn submissions, and by 43 to 79 points in six further paired comparisons, with no paper moving up in any of them. Stating a 30% acceptance rate in the prompt leaves a gap of 79 points between the two Opus versions. When score-stripped critiques are swapped between versions, the later critiques alone lower the accept share by 35 points for Opus and 51 for Gemini; for Opus the drop is 35 points under a third-party scorer and 24 after a rewrite that keeps each technical point and removes evaluative wording. On 57 papers, the perc
benchmark - arxiv:2611.20607 · cs.LGDNA: A Double-Helix Hierarchical Deep Interaction Network Architecture for Industrial Digital TwinHaoran Yang, Yanan Xu, Yinan Zhang, Wenjie Zhang +3
High-fidelity modeling is fundamental to industrial digital twin, yet a trade-off persists between the physical transparency of mechanistic models and the superior predictive power of data-driven approaches. Existing hybrid methods often fail to bridge this gap, suffering from shallow integration and rigid one-shot modeling strategies that struggle with complex industrial systems. To address these issues, we propose a \textbf{Double-Helix Hierarchical Deep Interaction Network Architecture (DNA)}. Inspired by the double-helix structure of biological \textit{DNA}, our architecture intertwines mechanistic and data-driven modeling strands through two novel interaction strategies, forming a symbiotic architecture in which the two strands mutually enhance predictive capability. Furthermore, the hierarchical structure of DNA allows it to model complex systems in a multi-stage manner through two tailored deep learning frameworks. We evaluate the proposed approach on two distinct industrial sys
helix - arxiv:2611.20605 · cs.AIAdMem: Expectation-Grounded Procedural Memory for Long-Horizon Task-Solving AgentsRunzhe Wang, Huilin Lu, Shengjie Liu, Li Dong +1
An agent that carries memory across tasks is only as good as the entries it keeps. We begin from an observation that is easy to reproduce and uncomfortable for the field: naively accumulating cross-task experience can make an agent \emph{worse} than one with no memory at all. On AgentBoard, a workflow-induction memory solves fewer tasks than a memoryless ReAct agent in seven of eight domains, and a per-step procedural store with no surrounding structure loses half of ReAct's completions on Jericho. We argue that the missing ingredients are per-action credit assignment and an explicit life cycle for stored entries, and present AdMem, which supplies both. Before each action the actor commits to an expected outcome; a concurrently running critic scores the realized outcome against that expectation and writes a procedural entry carrying a binary reward and a reflection, so credit lands on the action that earned it and no environment reward is required. Each entry then carries a scalar util
memoryagent - arxiv:2611.20603 · cs.AIBiGSyn: A Bi-Level Heterogeneous Graph Contrastive Learning Framework for Interpretable Drug Synergy PredictionSenquan Wang, Hui Fang, Haoran Nan, Xingwei Wang +1
Drug combination therapy has become a mainstream of modern medicine, with the potential to enhance efficacy, reduce toxicity, and overcome drug resistance. However, identifying synergistic drug combinations remains challenging, as it requires modeling complex interactions across multiple biological scales---from atom-level molecular structures to entity-level relations in biological networks. Existing approaches typically model these levels in isolation, overlooking complementary signals and long-range dependencies. To address these limitations, we propose BiGSyn, a Bi-level heterogeneous Graph contrastive learning framework for drug Synergy prediction. Specifically, BiGSyn leverages heterogeneous graph transformers to extract high-order dependencies from molecular graphs and biological networks, and introduces a bi-level contrastive learning objective to bridge the cross-level semantic gap. Extensive experiments on two benchmark datasets show that BiGSyn consistently outperforms state
benchmark - arxiv:2611.20595 · cs.CLInvariant-Aware Harness Enforcement to Protect LLM Agents from Silent API FailuresSajib Acharjee Dip, Brian Ouyang, Lala Motlhabi
LLM agents answer questions by calling APIs and trusting the results. An API can fail silently. It ignores part of a request, such as a misnamed filter, yet returns HTTP 200 and a plausible but wrong result. Current fixes, such as checklists, reviewer models and LLM judges, ask the model to catch it. But models often see the problem and answer anyway. We present SilentGuard, which moves the decision from the model to the harness running tool calls. It tests each result against invariants that any correct API must satisfy and withholds values that break one. We find silent failures in 11 of 15 public research APIs, with counts inflated up to 691 times. Current fixes leave over 70% of answers wrong, and agents report 77.4% of counts a judge flags. Across 7 open models, SilentGuard cuts wrong answers on its target faults from 74.0% to 1.7%, with no loss on fault-free tasks. On three frontier models they fall from 70.2% to 0.5%. A warning from the same check leaves 70.1% wrong, so enforcem
llm agent - arxiv:2611.20574 · cs.ROExtremeDrive: A Real-World Dataset and Benchmark for Critical DrivingXinhan Li, Yuhong Jiang, Shiyue Zhao, Junzhi Zhang +6
Existing autonomous driving datasets and benchmarks provide extensive resources for routine driving, yet real-world data covering safety-critical scenarios and extreme environmental conditions remain limited. To address this gap, we present ExtremeDrive, a real-world dataset and benchmark covering emergency maneuvers, complex traffic interactions, and adverse conditions such as slippery or snow-covered roads, alongside routine driving. The dataset comprises 589 driving scenes curated from a larger pool of real-world driving recordings, with calibrated and aligned multisensor data and reviewed annotations of road structure and traffic participants. We further establish a benchmark to evaluate planned trajectories through simulated vehicle execution, measuring safety-related metrics and driving performance. Experiments at equal training set sizes show that incorporating critical driving data reduces trajectory prediction errors and improves mean critical-success rates in critical scenari
benchmark - arxiv:2611.20569 · cs.AIToolGraph as a Q-Function for Test-Time Scaling of Tool-Use AgentsKe Shen, Ankita Bhaumik, Li Dong, Shengjie Liu +2
Tool-use agents must chain interdependent calls, yet which outputs satisfy which later inputs in a deployed catalog is not something a general-purpose language model reliably infers from tool descriptions alone. We compile that structure once into a reusable external representation. ToolGraph is a directed, weighted dependency graph over a tool catalog: parameter matching and interventional probes, which execute a candidate chain with and without its proposed predecessor, propose edges whose evidence is combined into an edge confidence. Prior work uses such graphs to retrieve tools, search for tool paths, or score the next call. We ask instead whether the same structure can supply a \emph{multi-step} value by composing graph paths into a goal-conditioned action value \(Q(s,a,g)\). We study three estimators (dynamic programming over entity sets, a GraphSAGE estimator distilled from it, and an LLM scorer conditioned on graph evidence) plus a verbalizer that returns action--goal feedback
tool-usebenchmark - arxiv:2611.20567 · cs.LGCalibrated Functional Prediction from Overlapping Nonlinear ObservationsPeida Li
We study prediction of nonlinear audit functionals from scalar observations of a shared hidden table state. For Schur-rational operators, local Gaussian projections yield a common sparse kernel for overlapping tasks, repeated row identities, and cross-row dependence. Shared-state residual samples extend this kernel through a positive-semidefinite feature representation. Positive matrix integration and sample-value enclosures connect the original inputs to the actual saved hybrid prediction. A sorted calibration-role perturbation bound then certifies its complete finite-pool pinball risk. Independent complete-loss selection chooses the ridge, score scale, and residual weight, while task-rank calibration supplies an upper endpoint. Evaluation uses twenty new reference contexts per dataset at two task sizes, giving 400 conditions across ten benchmark populations. The mean dataset complete-loss ratios are 0.9977 to the degree-two kernel, 0.9961 to matched complete-MC shrinkage, and 0.9911
benchmark - arxiv:2611.20565 · cs.LGMemory-Efficient Speculative Decoding with Quantized Draft KV CacheSaksham Gupta, Michael Li
Speculative decoding reduces generation cost, but it stores two KV caches: one for the target model and one for the draft model. We reduce this extra memory by quantizing the draft cache. The target cache and verifier remain in BF16, so that compression cannot change the target model's final correction rule. We first study how to divide precision between keys and values. A naive per-token quantizer makes keys appear much more sensitive: splitting a 12-bit budget, K8V4 beats K4V8 in acceptance by 10.87pp. However, with grouped per-channel keys, the gap becomes -0.25pp. A controlled Qwen test confirms that changing only the key quantizer can erase or reverse the preferred bit split. Hence, following KIVI, we propose K4V4 quantization with per-channel keys and per-token values. Across six model pairs, our method reduces draft cache memory by 66.20%, and combined memory across both caches by 23.95%. Mean acceptance drops from 57.93% to 57.14%, only 1.36% relative. Overall, quantized draft
memory - arxiv:2611.20557 · cs.ROAsymEEF: Bimanual Manipulation with Asymmetric End EffectorsXiaokang Sun, Ningyue Wang, Kunqi Xu, Guangqi Jiang +3
Teleoperated demonstrations are costly to collect and rarely reusable across end-effectors with different kinematics and action spaces. This burden grows in bimanual manipulation, where each end-effector pair may require new data. We present AsymEEF, a framework for executable cross-end-effector demonstration generation and unified bimanual policy learning. Given a small source dataset, AsymEEF transfers demonstrations to heterogeneous combinations of two-, three-, and five-fingered end-effectors without additional teleoperation. Kinematic teacher anchors provide geometric supervision, while sparse real-world X-Calib anchors correct discrepancies in grasp execution. A shared encoder maps wrist poses and actuator states to a morphology-agnostic latent space, and a morphology-conditioned decoder reconstructs target commands. Generated trajectories undergo replay filtering for task-level executability. The representation also supports a unified bimanual visuomotor policy trained across mu
manipulationteleoperationgrasp - arxiv:2611.20553 · cs.CVLearning a Task-Adaptive Low-Dimensional Semantic Space for Improved Visual ClassificationNikalal Kaldera, Wanqing Li, Chris Bunn, Philip O. Ogunbona
Semantic visual classification commonly aligns visual embeddings with semantic spaces induced by pretrained text encoders. However, such spaces are typically high-dimensional, weakly aligned with visual feature geometry, and insufficiently discriminative for downstream classification tasks. We propose TaSS, a framework for learning a task-adaptive low-dimensional semantic space that preserves semantic structure while explicitly optimizing class separability. TaSS is constructed through a parameterized geometry transformation of pretrained text embeddings that enlarges inter-class margins via distance reshaping and improves intra-class compactness through prototype-preserving semantic constraints. The resulting semantic space maintains meaningful semantic relationships while producing a discriminative geometry tailored for classification. Frozen visual encoder features are subsequently aligned to TaSS using a lightweight MLP, enabling improved discriminative learning without fine-tuning
benchmark - arxiv:2611.20551 · cs.AIGranularGPT: Natural-Language-Driven End-to-End Discrete Element Method SimulationsErlei Li, Sirui Tang, Ai Wang, Mingwei Hu +2
Discrete element method (DEM) simulations are widely used to investigate granular and particulate systems, but their application often requires specialized expertise in particle physics, numerical modelling, and simulation post-processing. Large language models (LLMs) offer new opportunities to simplify these workflows by enabling users to interact with simulation tools through natural language. This study offers GranularGPT, an LLM-assisted agentic framework for end-to-end DEM simulation workflows, from natural-language problem descriptions to simulation execution and automated post-processing. The framework enables users to generate DEM simulation cases, execute and monitor simulations, and automatically extract and report physical and process-level metrics. Evaluation across 60 normal-condition cases covering 12 DEM scenarios and parameter variations achieve a simulation success rate of 91.7%. In the error-handling evaluation, all 10 repairable errors are detected and 9 are successf
agentic - arxiv:2611.20535 · cs.AIVAHBench: A Harness Benchmark for Agentic Long-Video GenerationHongbo Zhang, Liuyang Song, Hang Zhou, Xinchen Liu +3
AI agents are commonly used for everyday information retrieval and coding. We ask whether they can also interpret human requests and turn them into complete videos. We introduce VAHBench, a benchmark that tests this ability through five short-drama briefs specifying story, characters, dialogue and duration. Agents plan and carry out production using shared generation tools. Evaluation combines video completeness and basic quality with checks of production reliability and reporting accuracy. Across 240 runs covering 48 configurations of general-purpose and video-specific agents, higher quality scores do not necessarily mean more complete delivery. Missing episodes, repeated footage and static filler reveal gaps between producing a video file and fulfilling the brief. Added composition tools help some agents but do not consistently improve performance. VAHBench measures how well agents translate user intent into finished video content and helps identify where that process breaks down.
ai agentagenticbenchmark - arxiv:2611.20528 · cs.CLReconPotBench: Evaluating LLM Honeypots in Multi-Round Linux ReconnaissanceQianli Wang, Junchao Wu, Binkai Ou, Lidia S. Chao +1
Large language models (LLMs) are increasingly used to build interactive honeypots, yet existing evaluations mainly assess isolated command outputs and cannot determine whether a system remains believable when later commands depend on earlier results. We present ReconPotBench, a multi-round benchmark for Linux reconnaissance honeypots. It contains 1{,}032 three-round reconnaissance interactions across eight reconnaissance categories, with reference outputs collected from a real Linux environment. Under a unified testing protocol, we compare Cowrie, open-source LLMs, and closed-source LLMs along three dimensions: round-level authenticity, cross-round contextual consistency, and hacker-oriented reconnaissance value. Results show that multi-round evaluation exposes failures hidden by single-round testing, as authenticity declines with interaction depth and both open-source and closed-source LLMs struggle to maintain coherent state across rounds. LLM-based honeypots nevertheless outperform
benchmark - arxiv:2611.20524 · cs.LGGaugeRoPE: RoPE-Compatible Executable Latent AttentionYuanhang Gao, Xiangrui Yang, Lingyuan Meng, Yuanfeng Chen +1
Latent KV cache compression reduces memory footprint by mapping historical keys and values into compact representations. However, generic compressed key states require reconstruction to native dimension before applying RoPE and computing attention scores, adding sequence-scaling overhead to the decode path. We characterize when a latent KV representation can serve directly as the execution state of pretrained RoPE. For RoPE-based transformers, exact static linear execution of the rotary key path requires the latent-to-native map to preserve the pretrained RoPE action through an intertwining relation, restricting exact mixing to coordinates that share the same RoPE frequency. Guided by this characterization, we build GaugeRoPE, a training-free executable latent KV representation. It combines a RoPE-compatible carrier that preserves the rotary action with an approximate position-independent residual key--value (RKV) correction for complementary key and value information, both evaluated w
memory - arxiv:2611.20522 · cs.CVAFRNet: Adaptive Frequency Redistribution Network for Image DerainingDezhi Zhang, Shangrong Yang, Husheng Guo, Zhijie Shen +1
Image deraining aims to recover rain-free images from rain-degraded observations. Recent frequency-aware deraining methods have demonstrated promising performance by removing rain artifacts in the spectral domain. However, since rain artifacts and background textures often appear in several correlated subbands, prior frequency-domain deraining models still enhance frequency subbands separately. Such band-wise processing limits information exchange across subbands. Without context from other subbands, rain and clean content that appear together in one subband cannot be separated. This leads to either residual rain streaks or over-smoothing of fine details. To address this limitation, we propose AFRNet, an Adaptive Frequency Redistribution Network for image deraining. First, we introduce an Adaptive Frequency Recombination Module (AFRM) that performs cross-band modulation among wavelet subbands, enabling information interaction and adaptive recombination of degraded frequency components.
benchmark - arxiv:2611.20521 · cs.LGLINK: Lightweight and Inductive Node-to-Node Propagation Rule Learning for Traffic KrigingYou Shang, Yu-Kwan Ching, Weifeng Su, Guanyao Li +2
Spatio-temporal traffic kriging aims to reconstruct traffic time series at unobserved locations with no historical traffic observations by leveraging spatial and temporal dependencies across the road network. Existing Graph Neural Network (GNN)-based approaches primarily rely on neighborhood aggregation to capture spatial dependencies, thereby modeling traffic interactions implicitly and making it difficult to accurately recover traffic magnitudes at locations without historical observations. To address this limitation, we propose LINK, a Lightweight and Inductive Node-to-node propagation rule learning paradigm for traffic Kriging. LINK introduces three key designs: 1) a propagation-centric traffic kriging paradigm that explicitly learns a shared node-to-node transition rule; 2) a context-aware Spectral Propagation Network that performs traffic-state propagation in both the spectral and time domains; and 3) TraffiKrig, a realistic traffic kriging benchmark that preserves heterogeneous
benchmark - arxiv:2611.20518 · cs.CVWhat the Averages Hide: Monocular Metric Depth Models Still Fail in Everyday Scenes, Mostly on ScaleHualie Jiang, Yujie Luo, Ziyang Song, Ximeng Li +2
Zero-shot monocular metric depth models are evaluated by averaging metric and scale-aligned scores across datasets. Averages cannot show how many images still have severely wrong metric depth, whether depth is under- or overestimated, or whether different models fail on the same images. We propose an evaluation built on per-image results: the failure rate (metric $δ_1$ < 0.5); scale-aligned metrics for structure, and the bias, dispersion and absolute error of the alignment scale for scale; and a cross-model comparison that separates shared from model-specific failures. We apply it to 11 recent architectures with and without ground-truth intrinsics (18 model evaluations) on the 7 everyday-scene benchmarks (4,406 images) used by MoGe-2. Metric depth estimation is not solved in everyday scenes, and the remaining difficulty is mainly scale: the best of the 18 evaluations still has a failure rate of 7.7% (datasets weighted equally); pooled over all failures, 78.8% have intact structure, and
benchmark - arxiv:2611.20510 · cs.ROWTR-Drive: World-Task Decomposition with Structured Risk Supervision for End-to-End DrivingHaoyu Xing, Shaobing Ding, Guanglei Zhao, Guanhua Zhao +2
Proposal-conditioned scene predictions can lack a common object-level reference across candidate trajectories. Aggregate planning scores also leave the objects and times associated with violations implicit. We present WTR-Drive (World-Task-Risk Driving), a Drive-JEPA-based planner that separates shared scene estimation from proposal-conditioned reasoning. A proposal-independent World Memory estimates object states, map structure, and future motion by re-reading the original visual features at each refinement round. Proposal feedback updates a residual Task branch without writing back to World Memory. This branch guides refinement through candidate-specific retrieval and gated visual feedback. A final Task readout supplies candidate-conditioned features for scoring. We derive structured risk targets from existing candidate-scoring rollouts without additional simulator calls. For at-fault collision and time-to-collision violations, the targets encode event existence, association with sha
memory - arxiv:2611.20507 · cs.ROVoxSAOR: Semantic-Aware Occupancy Robustification via Aleatoric Uncertainty ModelingMengfei Duan, Yuheng Zhang, Guoqiang Zhao, Yuanfan Zheng +2
Vision-based 3D occupancy prediction provides a dense semantic representation of scenes for autonomous driving and embodied perception, yet remains vulnerable to occlusion, depth ambiguity, and projection errors that induce spatially varying reliability in voxel features. Existing methods typically map voxel representations to semantic predictions without explicitly modeling such observation-induced feature variations. To address this limitation, we propose VoxSAOR, a semantic-aware occupancy robustification framework that models aleatoric uncertainty through feature-space proxies. It first quantifies Local Semantic Consistency (LSC) to derive a voxel-wise structural stability prior and uses cross-view semantic consensus to filter unreliable semantic correspondences. Within each semantic category, spatially matched stable and ambiguous voxels form stable-to-ambiguous feature transitions, which are modeled by a class-conditional diagonal Gaussian distribution, capturing both feature dis
embodied - arxiv:2611.20505 · cs.CVHow Reliable Is Visual Token Selection? Rethinking Pruning in Multimodal LLMsChaehee Song, Seokil Ham, KiHyun Seol, Jihyun Lee +2
Visual token pruning improves the efficiency of multimodal large language models (MLLMs) by removing redundant visual tokens, yet the reliability of existing selection signals has not been systematically examined. We analyze token selection under semantics-preserving perturbations and across representative pruning criteria, revealing two limitations. First, attention-based importance shifts non-uniformly under perturbations, so magnitude alone cannot distinguish stable evidence from unstable saliency (local instability). Second, attention- and diversity-based methods select substantially different subsets despite comparable performance, showing that useful evidence is not confined to a unique subset (global inconsistency). These findings motivate reliable token-set construction. We instantiate this perspective in CoRePruner, a training-free framework combining instruction-aware diversity with reliability-aware attention. Across image benchmarks, multiple MLLM architectures, and token b
benchmark - arxiv:2611.20493 · cs.LGA Temporal Convolutional Network with Multi-Head Attention Model: A Surrogate for Parallel DEVS Semiconductor Fab SimulationVamsi Krishna Pendyala, Hessam S. Sarjoughian, Edward J. Yellig, Bharadhwaj Balasubramaniam
Semiconductor factories process many kinds of wafers organized into lots of various sizes and configurations. Different kinds of machines, each with its own operations, process wafers for hours to days, and the wafers undergo many incremental fabrication steps (with re-entry) and flows spanning days to weeks across connected machines. Fast and accurate predictions of throughput ($TH$) and turnaround time ($TA$) dynamics for individual machines and their combinations are essential for factory operations. To use machine learning methods for this purpose, this paper presents a Temporal Convolutional Network with Multi-head Attention (TCNA) architecture. It uses simulated datasets from a benchmark Parallel DEVS (PDEVS) semiconductor fabrication factory model. The factory processes configurable wafer lots with uniform and sinusoidal wafer arrival profiles and may undergo preventive maintenance. The wafer lot size, arrival profile, and factory maintenance policy are used as static covariates
benchmark - arxiv:2611.20488 · cs.AIEVIDENT: Evaluating Evidence Grounding in GUI AgentsYerin Nam, Minkyung Cho, MinKyu Kim, SeungWoo Song +3
GUI agents are increasingly evaluated by whether they complete web tasks, but surface success can hide a critical failure: the final answer may not be supported by evidence observed during navigation. We introduce EVIDENT, an evidence-driven benchmark for evaluating whether GUI agents ground their decisions in observable web evidence. EVIDENT contains 220 real-world English and Korean tasks spanning semantic grounding, contextual inference, and visual grounding, each paired with human-curated evidence metadata specifying what must be verified. We further separate surface task completion from evidence-grounded success using three metrics: TSR, ESR, and GSR, which respectively measure outcome success, required evidence observation, and claim-level evidence support. Experiments on commercial and open-source GUI agents reveal substantial gaps between surface success and evidence-grounded success, especially in contextual and visual tasks. These results show that current evaluations can ove
agentbenchmark - arxiv:2611.20484 · cs.AIAgent-Guided Configuration Tuning for SMT SolversCong Wang, Qiaoyi Zheng, Zhaoguo Wang, Fan Long
SMT solvers expose many configuration choices that determine how well their existing algorithms perform on a workload. We present, to our knowledge, the first workflow that uses general-purpose LLM agents to tune SMT solver configurations. Agents read solver documentation, source code, and persistent experiment records to propose option changes, and a deterministic harness measures each proposal, validates the returned models, and selects changes separately for each competition Division. Over roughly nine weeks, our tuning campaign recorded 4,545 proposals for cvc5 across 13 Divisions and 36 logics. We evaluate the tuned configurations on the 2026 and 2025 SMT-COMP ModelValidation benchmarks under a 24-second cutoff. On problems not used for training, they validate more problems than official cvc5 in ten of the 13 Divisions in 2026 and in eight in 2025, with geometric-mean gains of 7.8% and 6.8% across Divisions. On the complete official selections, which include training problems, the
llm agentbenchmark - arxiv:2611.20481 · cs.ROLift, See, Act: Hierarchical Robot Policy Pretraining with 3D Foundation ModelsYiyuan Ge, Changxing Ding, Ziyu Hao, Zijie Zheng +1
Robot policy pretraining based on human videos is crucial for improving the policy's generalization ability. One main challenge for this task is the lack of explicit action-relevant representations in such unlabeled data. Recent works tend to estimate 3D hand motion trajectories from videos using 3D foundation models (3D FMs). However, they keep solely the sparse trajectories as supervision, ignoring the rich, fine-grained information produced during the trajectory extraction process. To address this issue, we propose LSA, a hierarchical robot policy pretraining framework that includes three phases: Lift, See, and Act. Specifically, the Lift phase elevates 2D video observations into dense 3D representations by adaptively aligning features in the policy's shallow layers with the intermediate representations extracted from multiple 3D FMs. Building upon these lifted representations, the See phase equips the policy with explicit geometric and interactive awareness through dual guidance. I
manipulationrobot policy - arxiv:2611.20478 · cs.CVBeyond Seeing: Fine-Grained Preference Validation and Severity-Aware Alignment for Mitigating Visual Hallucinations in Vision-Language ModelsAbdulaziz Alqahtani, Alwaleed Alharthi, Muzammil Behzad
Large vision-language models can produce fluent responses containing objects, attributes, or relations unsupported by the image. We study an implemented alignment pipeline that normalizes released fine-grained hallucination supervision, validates chosen-rejected preference pairs with vision-language judges, repairs rejected chosen responses, and trains LLaVA adapters with standard or severity-aware direct preference optimization. The study uses 16,143 normalized feedback records and 8,386 released preference pairs, comparing direct training with zero-shot and two-shot judged severity-margin variants. The zero-shot variant reduces Object HalBench CHAIRs from 53.00 to 36.00 and CHAIRi from 15.72 to 10.41. Across aligned models, AMBER-Hal decreases from 35.90 to 25.10-26.90, accompanied by reduced visual coverage. In contrast, the base model retains the highest valid POPE adversarial F1, 84.18 versus 83.87-83.91 after alignment. These results support benchmark-dependent reductions in gene
benchmark - arxiv:2611.20465 · cs.ROReRead: Wrong Predictions, Useful Representations - Correcting Frozen Robot Policies by Re-reading Their Own RepresentationsXu Su, Shunlong Xiao, Qintao Yi, Mingkuan Zhao +4
When a pretrained robot policy fails on challenging tasks, common remedies include retraining the backbone, adding an external visual encoder, or introducing a memory module. We argue that wrong predictions do not imply wrong representations: a policy can retain the required information yet fail to read it out as the correct action. To test this view, we propose ReRead: with the backbone frozen, a corrector re-reads the current and two previous caches to predict an action residual. The corrector sees no images, is trained solely on offline demonstrations from all 50 RoboTwin 2.0 tasks, and adds only 4.17% latency. Ablations and interventions show that the correction relies on the backbone's own representations: none of four external visual encoders adds further gain, and swapping in past caches from other trajectories sharply increases action error. On the 11 hardest RoboTwin 2.0 tasks, ReRead improves the success rates of LingBot-VLA, $π_{0.5}$, and Fast-WAM by 23.45, 13.09, and 8.00
robot policyrobotwinmemorymemory module - arxiv:2611.20460 · cs.ROStochastic Self-Distillation for Flow-Based Vision-Language-Action ModelsJingru Guo, Alessandra Russo, Joey Bose, Christos Ziakas
Flow-based vision-language-action (VLA) models achieve strong performance in robotics through imitation learning, yet fine-tuning them with on-policy reinforcement learning (RL) is limited. In group-based RL methods like Group Relative Policy Optimization (GRPO), every rollout step requires multi-step ODE integration, and the sampler's limited stochasticity leads to rollout groups with low variance. We propose VLA-SSD, which addresses both high inference latency and limited stochasticity. In particular, our method self-distills a pretrained flow-matching action head into a stochastic one-step policy by conditioning a single-jump generator on a Brownian path. We derive a conditional action density via our stochastic one-step policy, enabling efficient post-training via likelihood-ratio RL algorithms. We show that distilling VLA-SSD from the $π_0$ VLA reduces the action head's forward-pass time by a factor of 7.7, at the cost of a 5.5% relative decrease in the success rate on the LIBERO
vision-language-actionvlaaction headliberopost-trainingbenchmark - arxiv:2611.20458 · cs.LGVISTA: A Framework for Vision-Informed Semantic Traffic AwarenessMurat Arda Onsu, Poonam Lohan, Burak Kantarci, Aisha Syed +2
The Internet of Things (IoT) is transforming intelligent transportation systems by connecting roadside cameras, edge computing units, vehicles, and traffic infrastructure into a shared sensing and decision-making ecosystem. However, transmitting raw traffic video over vehicle-to-everything (V2X) links is bandwidth-intensive, while deploying large multimodal models directly on resource-constrained vehicles is often impractical. To address these limitations, this article proposes VISTA (Vision-Informed Semantic Traffic Awareness), a multimodal vision-language edge-intelligence framework that combines predictive video representation learning with lightweight language-based reasoning to provide communication-efficient traffic situational awareness. In the proposed architecture, roadside surveillance units employ the Video Joint Embedding Predictive Architecture (V-JEPA) and an attentive-pooler module to convert traffic video clips into compact future-scene embeddings. These embeddings are
v-jepa - arxiv:2611.20452 · cs.ROXMimic and XSoma: Learning Humanoid Motion Tracking and Locomotion in Under a MinuteNico Bohlinger, Jan Peters
We present XMimic and XSoma, which learn humanoid motion tracking and locomotion on rough terrain in under a minute of training. Both follow SuperMimic, which learns a world model of the robot and optimizes the policy with analytic gradients through short rollouts imagined in this model. Existing implementations of this method do not make use of larger GPUs, as every training iteration is a long chain of small GPU operations that run one after another. XMimic and XSoma therefore shorten this chain and run the steps of every iteration at the same time on up to four GPUs. We compile the complete training loop with JAX, including the MuJoCo Warp simulator. For locomotion, XSoma leaves only the dynamics of the robot to the world model and computes everything else exactly, including the terrain height under the feet. It learns with a horizon of one second and without a critic. On four H100 GPUs, XMimic learns to track all 40 LAFAN1 clips with the Unitree G1 in 0.86 minutes, 3.7 times faster
humanoidworld model - arxiv:2611.20447 · cs.AIThe Logic Chain Breaking Point: A Controlled-Variable Benchmark for Reasoning Degradation in Open-Weight Language ModelsAkif Islam, Mohd. Ruhul Ameen, Md. Ekramul Hamid
We introduce The Logic Chain Breaking Point, a controlled-variable benchmark for deductive reasoning in open-weight language models. It spans three phases: (i) pure modus-ponens chain scaling, (ii) distractor resilience with a fixed core chain, and (iii) contradiction detection under primacy versus recency placement. Independent variables follow a powers-of-two sequence $v \in \{1,2,4,8,16,32,64,128\}$, with ground truth known by construction. We evaluate sixteen open-weight models across instruction-tuned and reasoning-tuned panels, including a small-model suite and a larger local-model extension. All models are evaluated with greedy decoding under model-appropriate context and output budgets, while context- and length-limited trials are quarantined from the valid accuracy denominator. Our results reveal a sharply non-monotonic difficulty surface: pure deductive chains up to $L=128$ remain near saturation for most models, and distractor noise causes only model-specific degradation at
benchmark - arxiv:2611.20446 · cs.CVLEDGE: Separating Critic and Editor Failures in Text-to-Image Self-CorrectionVyom Agarwal, Mokshda Gangrade, Samet Ayhan, Giacinto Paolo Saggese
Text-to-image (T2I) self-correction combines a vision-language model (VLM) critic that diagnoses errors and proposes instructions with an image editor that executes them. End-to-end evaluation obscures whether failures arise from incorrect diagnoses or an editor unable to execute valid instructions. We introduce LEDGE (Learning Error Diagnosis in Generation and Editing), a framework that separates these bottlenecks. LEDGE compiles oracle edit instructions from GenEval while holding the generated images and the editor fixed, isolating the editor's correction capability from the critic's diagnostic accuracy. On Stable Diffusion 3.5 Large, oracle instructions raise GenEval from 72.0% to 81.2%, versus 74.1% for the baseline VLM critic. Yet 435 of 644 failed images remain unrepaired, exposing a substantial editor gap. We also use oracle labels from disjoint training prompts to supervise the critic. Oracle supervision raises exact diagnosis from 33.3% (the same VLM with our structured output
self-correction - arxiv:2611.20445 · cs.LGExplainable Federated Learning via Global-Local Attribution AlignmentDawood Wasif, Terrence J. Moore, Chang-Tien Lu, Jin-Hee Cho
Federated learning enables on-device training without centralizing data, yet existing systems still struggle to provide explanations that are both locally faithful and globally consistent under strict privacy and bandwidth constraints. Prior approaches either keep explanations siloed across clients, transmit heavy or sensitive artifacts, or replace expressive task models with interpretable surrogates that sacrifice accuracy. We propose xFedAlign, a model-agnostic framework that decouples task optimization in parameter space from explanation coordination in a compact group space. Each client distills a lightweight surrogate to produce private, per-class top-k attribution artifacts, which are robustly aggregated by the server into a Global Explanation Prior that softly aligns client explanations without constraining task learning. Across image, text, and tabular benchmarks with IID and non-IID partitions, xFedAlign matches FedAvg accuracy while consistently reducing explanation drift and
benchmark - arxiv:2611.20433 · cs.CVReScene: Incremental Scene Chronology from Sparse RevisitsWei Qin Chuah, Zeji Hui, Amirali Khodadadian Gostar, Alireza Bab-Hadiashar
Scene chronology aims to reconstruct a scene across visits separated by hours or days, during which objects may change in position, presence or shape. Such reconstruction must retain every earlier state, yet each revisit often provides only a few images. Storing a complete model for each visit duplicates the unchanged scene, whereas updating a single shared model alters earlier states. We present ReScene, which stores each visit as an immutable compact update, so that any earlier visit can be recovered exactly by applying the stored updates in order, without retaining past images. To localise each update, ReScene attributes the disagreement between the preceding reconstruction and the new images to individual Gaussians through their compositing weights. This attribution guides the suppression of outdated Gaussians and the refinement of persistent ones, while newly visible surfaces are initialised from learned multi-view depth. We also introduce Multi-State, a benchmark of non-rigid def
benchmark - arxiv:2611.20431 · cs.CVUrbanTwin Sim2Real LiDAR Challenge 2026: Site-Calibrated Synthetic Data and Multi-Detector Fusion for Roadside 3D DetectionArnav Kumar, Brijesh Soni
This report summarizes the work submitted by team arnav007 (submission ID 1054222) for the UrbanTwin Sim2Real LiDAR Challenge (V2X-Real track) at the IEEE BigData Cup 2026. The primary objective is to train 3D detectors for vehicles, pedestrians, and trucks using only synthetic roadside LiDAR data, and to generate synthetic scans that are realistic enough to be indistinguishable from real ones. A CARLA based pipeline is developed and calibrated to the fixed sensor site, accurately modeling the range image, including no-return beams, based on unlabelled real development scans. A combined inference detector, using multiple 3D detectors (VoxelNeXt, PV-RCNN, and SECOND) trained exclusively on synthetic data, is also proposed. Each scan is aligned to the synthetic ground plane and processed through the detectors at multiple ground heights. The resulting bounding boxes are fused using bird's-eye view (BEV) clustering, followed by class-specific post-processing to refine heading, box size, an
sim2real - arxiv:2611.20430 · cs.LGTrainExplorer: Autonomous LLM Post-Training via Plan-and-Decouple ExplorationXuannan Liu, Tian Lan, Zekun Li, Xing Cui +8
LLM-based agents can automate model post-training, accelerating progress toward recursive self-improvement. Despite their promise, existing approaches remain under-explored and face two main challenges: (1) strategy lock-in that limits exploration beyond the initial strategy; and (2) coupled execution that obscures the effects of individual modifications. To address these challanges, we propose TrainExplorer, a multi-agent framework for autonomous post-training, that incorporates strategy planning and execution decoupling, to achieve broaden strategy exploration. First, strategy planning draws on scientific literature to construct a candidate search space of training strategies. Then, decoupled execution enables single-component interventions and attribution to guide strategy exploration. Extensive experiments on PostTrainBench demonstrate that TrainExplorer achieves a weighted aggregate score of 39.02, outperforming Claude Code using the same LLM by 7.32 and surpassing GPT-5.6 Sol-pow
multi-agentagent frameworkself-improvementpost-training - arxiv:2611.20423 · cs.AIFinVerify: A Verification Framework for Financial AgentsMao Guan, Qian Chen, Ge Meng, Xiao-Yang Liu
Financial agents produce numerical claims that can be wrong even when their calculations execute correctly, due to mismatched reporting periods, unsupported sources, incompatible units, or incorrect financial definitions. We introduce FinVerify, an evidence-grounded framework for verifying financial agents' numerical claims. Based on an analysis of 190 incorrect trajectories from the Agent Finance Reasoning dataset, we define five error categories--temporal integrity, source fidelity, comparability, accounting integrity, and financial-model fidelity--and operationalize them as verification protocols. An LLM structures claims and supporting evidence, while deterministic validators check financial requirements; when authoritative formulas are available, Z3 checks consistency with grounded inputs. FinVerify accepts only if all applicable checks pass, rejects proven violations, and abstains when evidence is insufficient. We evaluate the framework against a Direct baseline and an LLM judge
agentbenchmark - arxiv:2611.20419 · cs.LGReasoning Again with a Quarter of the Experts: Retry First, Then TrainAnik Jha
Removing three quarters of the routed experts from a 122B-parameter reasoning mixture-of-experts model makes it fit in a 12 GiB file, and breaks it in a specific way: on LiveCodeBench it enters verbatim reasoning loops that run to the token limit on 44 of 120 development problems, where the unpruned model at the same 2-bit quantization loops on none. We study how much of that loss can be recovered without changing the kept experts' quantized bytes, in three steps of increasing cost. On development problems, a sampling penalty recovers part of it. Stopping a reply when it starts to loop and drawing again recovers more on top of it, with no weight change. Our method, B, adds rank-8 trained corrections to the kept experts, loaded as an adapter, and restores two experts per layer; it is fitted layer by layer to the unpruned model's expert block on the pruned model's own activations. In a pre-registered comparison on 130 easy and medium LiveCodeBench problems that no part of the study had t
benchmark - arxiv:2611.20409 · cs.LGIntegrated GPU Core for Mamba-3Shiva Raja, William Desueza, Ajay Joshi
Mamba-3, the latest Mamba state-space model (SSM), enables efficient long-context modeling, matching Mamba-2 perplexity with half the state size and increasing decoding arithmetic intensity. However, its multi-input multi-output (MIMO) computation, exponential-trapezoidal discretization, and rotary positional embeddings (RoPE) introduce dense, element-wise, and sequential operations. Despite kernel fusion, conventional GPUs execute these operations inefficiently. Our characterization of Mamba-3 hybrid models reveals numerous kernel launches, substantial data movement, and increased register pressure that reduces warp occupancy and memory-latency hiding. We propose RescuCore, a specialized GPU execution unit for Mamba-3 inference that retains other hybrid model layers on existing GPU resources. Integrated alongside CUDA Cores and Tensor Cores within each Streaming Multiprocessor (SM) sub-partition, RescuCore shares the register file, shared memory, and warp schedulers. Dense projections
long-context - arxiv:2611.20402 · cs.ROMaking Koopman Control Guarantees Hold on RobotsAbulikemu Abuduweili, Changliu Liu
A Koopman controller can stabilize its learned latent-space linear model yet fail to stabilize the nonlinear plant. The learned model propagates predictions in latent space, while the deployed controller repeatedly re-encodes the measured physical state. We show that nominal closed-loop stability transfers to the nonlinear plant when the learned and true closed-loop dynamics agree to first order on the embedding's tangent space at the target. We derive a robustness margin for approximate matching and show that a stable open-loop latent model cannot exactly match an unstable equilibrium. We introduce Tangent-Invariant Koopman (TIK), which estimates a local plant linearization from offline transitions and enforces first-order matching to this estimate. TIK recovers the fitted LQR gain at the target, while a bilinear predictor captures state-dependent actuation during large maneuvers. Across four simulated unstable benchmarks, all 40 Standard Koopman controllers have stable latent closed
benchmark - arxiv:2611.20398 · cs.AIDeploying a Control-Logic LLM Assistant in Air-Gapped Power Plants: Lessons in Source Authority and Bounded RetrievalYiwei Fu, Nicholas Gabriel
Deploying Large Language Model (LLM) assistants in industrial operational technology environments poses constraints largely absent from cloud applications: site policies restrict engineering and operational data from leaving the plant, on-premises compute is limited, and incorrect causal attribution can mislead incident response. We report lessons from deploying a control-logic evidence layer that underpins a broader on-premises operator assistance system at two gas power plants. Two prototypes revealed where generative models added avoidable uncertainty and cost. The first used a Vision Language Model (VLM) to reconstruct control topology from PDF printouts routinely exchanged by engineers. Across the full 6,979-page document, extraction reached 0.51 connection precision and required 3.1 hours on 4xH200 GPUs; consultation with controls engineers then revealed a native structured export from the authoring system. The second used an LLM tool-calling agent to plan repeatedly over eight p
knowledge graphagentbenchmark - arxiv:2611.20390 · cs.LGArchitectural Principles for Parameter-Efficient Unitary Transformations in Coherent Optical Neural NetworksKazuo Aoyama, Hiroshi Sawada, Shota Kita, Akihiko Shinya +1
Coherent optical neural networks based on interferometric architectures offer a promising route toward high-throughput, energy-efficient computation by exploiting the multiple degrees of freedom of light, including spatial, temporal, and spectral modes. However, practical implementations are constrained by the number of tunable parameters due to fabrication complexity, device footprint, and control overhead. In such parameter-constrained settings, achieving high expressive power remains a central challenge. We identify two key architectural requirements for maintaining high expressive power under these constraints: (i) statistical uniformity in the implemented transformations and (ii) strong mixing capability per parameter, which enables rapid and unbiased redistribution of information across modes. To realize these requirements, we develop a logarithmic-depth butterfly-topology architecture whose fundamental building block is a single phase-shifter--beam-splitter pair, in contrast to
mach-zehnder - arxiv:2611.20387 · cs.LGGeneralized Prediction Set with Bandit FeedbackZhou Wang, Xingye Qiao
In high-stakes environments where uncertainties abound, set-valued prediction offers a cautious and robust mechanism by presenting multiple potential labels as the prediction for each test instance to mitigate the potential risk associated with prediction errors. Yet, integrating this paradigm with out-of-distribution (OOD) detection remains scarcely explored in such settings as online learning with bandit feedback. The bandit feedback mechanism informs the learner about the correctness of the pulled arm/action instead of the explicit ground truth label, leaving the true class label unknown when an incorrect action is taken. To address this challenge, we introduce BanditGPS which conducts set-valued prediction with OOD detection in the bandit feedback setting, using an estimation to the ground truth of class labels. BanditGPS achieves three objectives: render small/informative prediction sets, enhance the OOD detection performance, and control the recall for all normal classes to meet
online learning - arxiv:2611.20385 · cs.CVDominoBench: Evaluating Multi-Step Physical Reasoning from Still ImagesEmilio Ferrara
Can a vision-language model look at a picture of a chain reaction and tell what will happen? Recent multimodal models describe scenes fluently and solve textbook physics problems, which suggests they should also predict how a scene of ramps, ledges, pendulums, seesaws and dominoes plays out once a ball is pushed. We tested this claim on 30 models from 10 families and reached a clear conclusion: unaided, no model predicts the outcome reliably. We built DominoBench, a benchmark of still images of 2D rigid-body scenes whose outcome a physics simulator computes. It holds 854 items in 427 minimal pairs, in which a change of one to three grid steps in one quantity decides whether one person survives, and every item comes with a four-rung hint ladder ending in the exact calculation that decides the outcome. Unaided, frontier models from Anthropic, Google and OpenAI are right about the person whose fate flips in at most 72% of items and answer both twins of a pair correctly in at most 34%. Poi
benchmark - arxiv:2611.20384 · cs.LGJointQ: Joint Scale-and-Integer Quantization for Large Language ModelsKeiji Kimura, Yuma Ichikawa
GPTQ is the de facto standard solver for layer-wise post-training quantization (PTQ) of large language models, yet the solver itself has received less attention than weight preprocessing and cross-layer objectives. We target a bottleneck of GPTQ: its greedy procedure optimizes continuous group scales separately from the assignment of discrete integers. Evaluating each integer candidate with its closed-form optimal scale has been used for per-channel quantization; JointQ extends it to group-wise quantization, where a row's scales are coupled through the Hessian, and applies it to refine the GPTQ solution. Each local-search move changes one integer and recomputes its group's optimal scale in closed form, accounting for the other groups through the full Hessian; all scales are periodically re-solved jointly. The search starts from a row-wise selection among scale-refitted candidates including the GPTQ solution, so unregularized JointQ never has a larger layer-wise calibration error than G
post-training - arxiv:2611.20382 · cs.CVSigns of Wear: Evaluating Intelligent Speed Assistance Under Synthetic Adverse ConditionsJohannes Fabian Hahn, Mao Shan, Julie Stephany Berrio Pere
Intelligent Speed Assistance (ISA) systems rely on cameras to recognize speed-limit signs, yet existing benchmarks underrepresent the weathered, dirty, and overgrown signs commonly encountered on rural roads. We use generative image editing to introduce different forms of wear into real street-level images of speed-limit signs. The resulting ISA-Wear dataset contains paired original and edited images. Guided by text instructions, a multimodal image-editing model adds snow, rust, vegetation, faded paint, or mud to the speed-limit sign in a street-level frame at three severity levels. Where appropriate, the surrounding scene is adapted as well, and the viewpoint is kept. Every edited frame has an unedited counterpart, and we only evaluate frames whose original is recognized correctly, so any failure can be attributed to the wear. Since the model alters the sign's geometry in 21\% of the edits, all edits are filtered by a geometric check and a manual review. Severity is measured as the LP
benchmark - arxiv:2611.20381 · cs.CVTrait Collapse in Text-to-Image Generation: Evaluating and Mitigating Personality Inconsistency in Visual StorytellingYuanbo Tong, Haoyuan Li, Jibiao Jiang, Zirui Wang +2
In visual storytelling, a believable character must not only look consistent across scenes but also \textit{behave} in ways that reflect their personality. Current text-to-image (T2I) models preserve visual appearance effectively, yet consistently fail to maintain personality-relevant behavior---a failure we term \textit{trait collapse}. We show that conventional metrics (CLIP Score, DINO similarity) are blind to this failure, and that abstract personality descriptors and scene-level emotion labels do not produce consistent improvements over neutral prompts across models. To systematically diagnose trait collapse, we introduce \textbf{ViStoryBench-Personality}, a story-level benchmark combining an expert-validated Visual Cues Lexicon (Fleiss' $κ= 0.757$) with a six-dimensional metric suite covering personality alignment, cross-scene stability, and behavioral coherence. Evaluating seven T2I systems, we confirm a persistent semantic gap between abstract trait labels and visual evidence,
benchmark - arxiv:2611.20372 · cs.CVVisDrone-ReID: A Unified Multi-Category Object Re-Identification BenchmarkAsmau Lawal Bilbis, Nathanael L. Baisa, Amudhavel Jayavel
Object re-identification (Re-ID) benchmarks remain predominantly confined to single-category evaluations, limiting the development of unified systems capable of operating across diverse object types. This paper presents VisDrone-ReID, a unified multi-category Re-ID benchmark derived from the VisDrone2019-MOT dataset through a principled conversion pipeline. The pipeline resolves structural incompatibilities between multi-object tracking annotations and Re-ID evaluation requirements, addressing sequence-scoped identity collision, temporal redundancy, and the absence of a principled identity-level train/test split. The resulting benchmark comprises 9,516 identities and 150,016 images across ten object categories - pedestrian, people, bicycle, car, van, truck, tricycle, awning-tricycle, bus, and motor, sourced from 80 unmanned aerial vehicle (UAV) sequences captured across urban environments. A leave-one-out (LOO) evaluation protocol is adopted to provide a comprehensive assessment. Basel
benchmarkevaluation protocol - arxiv:2611.20364 · cs.LGMulti-Library Feature Engineering and Statistically Validated Classification for Power Swing Detection and Stability AssessmentPallav Kumar Bera, Samita Rani Pani, Ranjeeta Patel
Faults occurring during power swings, and the evolution of a swing into an unstable condition, cause distance-relay maloperation and cascaded outages. This paper identifies faults, faults during power swings, and power swings, and distinguishes stable from unstable swings in a 9-bus WSCC system modeled in PSCAD/EMTDC. Three time-series feature libraries (tsfresh, TSFEL, and tsfeatures) are applied to the 3-phase relay voltages and currents, giving 3286 statistical, temporal, spectral, and fractal descriptors per event window. A two-step feature selection (Boruta all-relevant filtering followed by RFECV- and SHAP-guided minimal-optimal pruning) reduces the feature set to 18 compact, non-redundant features. 14 classifiers spanning linear, kernel/instance-based, probabilistic, neural, and tree-ensemble/boosting families are then benchmarked under repeated stratified cross-validation, with model differences assessed by a Friedman/Nemenyi omnibus-and-post-hoc test and a Nadeau--Bengio corre
benchmark - arxiv:2611.20352 · cs.LGEPI-SST: Renewal-Structured State-Space Transformers for Epidemic ForecastingYiyun Chen, Zewen Liu, Shengpu Tang, Max S. Y. Lau +1
Epidemic forecasting supports public health decisions ranging from capacity planning to intervention timing, for which both point accuracy and calibrated uncertainty are essential. Although neural models have advanced point forecasting, producing reliable probabilistic predictions over multiple future horizons remains challenging. Renewal models encode epidemic propagation, while neural forecasters learn flexible multi-horizon representations; integrating these strengths within a probabilistic future rollout remains underexplored. We propose EPI-SST (Epidemic State Space Transformer), a probabilistic forecaster centered on an evidence-driven renewal transition. At each horizon, neural evidence produces a bounded update to an internal multiplier and advances a finite-memory renewal state that generates the future mean trajectory. A separate heteroscedastic Gaussian head controls predictive scale using a validation-selected log-variance penalty. Across four epidemic surveillance benchmar
benchmark - arxiv:2611.20351 · cs.CVBlockAlign: Preference Optimization for Trajectory-Distilled Video GenerationJingyuan Zhu, Biaolong Chen, Yangyang Qin, Shengyang Shen +4
Directly aligning trajectory-distilled video generators slashes rollout costs and optimizes the deployed few-step model without teacher fine-tuning or re-distillation. However, conventional flow-matching preference objectives assume single-head local velocity supervision, mismatching the finite-interval, block-average transition interface of trajectory-distilled generators such as Parallel Decoding Distillation (PDD) and Parametric Trajectory Distillation (PTD). Naively supervising only the primary initial-subinterval velocity sub-head rather than the fused block-average velocity at block starts (DirectAlign) induces a severe training--inference mismatch that degrades quality below the unaligned baseline. We introduce BlockAlign, an interface-consistent preference optimization framework that constructs states at valid block boundaries, queries the fused block-level velocity predictor, and supervises span-normalized transition residuals-seamlessly supporting DPO-, NFT-, and rollout-styl
post-training - arxiv:2611.20349 · cs.CVWords Speak Louder? Exploring the Language Shortcut of LLM-Driven Incomplete Multimodal LearningWenhao Li, Chong Xiao, Xiaokun Bai, Rundong He +4
Is a face or a voice still heard once the words are known? Incomplete multimodal learning increasingly uses large language models (LLMs) as fusion engines: modality features enter as token blocks and missing modalities become placeholders. We find that LLM-driven fusion can override non-verbal evidence it can read. Contrasting its full-input prediction with counterfactual views, the same model shown one modality or no content, we identify language shortcuts. We build MR-Bench: MR-Bench-A types every test sample of four sentiment benchmarks; MR-Bench-H contains 900 human-validated clips (620 language-shortcut, LS; 280 non-verbal-shortcut, NS) from seven sources, and every model is scored on all of them. Zero-shot MLLMs solve at most 32.7% of the LS clips; adding the transcript to the muted video lowers vision-language accuracy by 46.1-53.1 points. Yet inverting a text-only model's answers solves 94.4%, so LS accuracy alone cannot establish use of non-verbal evidence. Inside the fusion L
benchmark - arxiv:2611.20335 · cs.LGWhen Should Decision Models Compare Candidates?Fangzhi Li, Quan Shi, Wenting Yang, Yueqiao Chen +6
Language-model agents frequently select among candidate actions, such as tools and web elements. Should decision models score candidates independently or allow their scores to depend on other candidates? We investigate this question by comparing three approaches: Pointwise scoring, Setwise scoring, and Candidate Attention. Across five agent-decision benchmarks, Pointwise scoring remains competitive, while Setwise scoring performs better on some tool-selection and binary-decision tasks. Candidate Attention does not consistently improve upon Pointwise scoring, despite allowing cross-candidate interaction. Using two controlled synthetic tasks, we further examine whether models learn to use candidate-set information. With targeted relational training, Setwise models learn candidate dependencies and generalize to shifted inputs, while Candidate Attention shows weaker transfer. Our results show that access to other candidates does not guarantee effective comparison, and that its benefits dep
benchmark - arxiv:2611.20334 · cs.LGResidual-Tail-Informed Flow Matching for Time Series Anomaly DetectionShyam Sundar Saravanan, Tie Luo, Sanjay Madria
Time series anomaly detection (TSAD) models are often trained on nominal data using average-risk objectives, which emphasize dominant normal patterns. However, real time series can contain rare but valid nominal regimes that are underrepresented in the training distribution. We show that data-space flow matching can over-penalize such rare-normal windows and assign them anomaly scores comparable to those of true anomalies. To address this, we propose Residual-Tail-Informed Flow Matching (RT-FM), an unsupervised framework that uses a temporal variational autoencoder to estimate residual latent tailness and converts it into bounded, warmup-annealed weights for training a data-space flow matching model. Experiments on seven TSAD benchmarks show that RT-FM improves rare-normal coverage while preserving strong anomaly discrimination, substantially reducing false alarms on such windows and resulting in improved aggregate detection performance. The code is available at: https://github.com/tru
benchmark - arxiv:2611.20321 · cs.LGUnreasonable Confidence from Simulator Misspecification: Directional Robustness on the SPEED+ Sim-to-Real BenchmarkNikolay Dolgov
Vision-based navigation systems for orbit must be trained almost entirely on synthetic imagery, because real labelled images of a given spacecraft in flight rarely exist. Benchmarks like SPEED+ are built for that situation: the training set comes from a simulator, and the test sets are real images of a satellite mockup under two kinds of lighting. That simulator approximates how light reflects off mylar and aluminium, and like every scientific simulator it is misspecified. Its error is radiometric, since it misses the specular highlights, hard shadows and sensor saturation common in real images. We probe the resulting synthetic-to-real gap, the central risk for such systems, with a deliberately simple monocular pose estimator. In a 2x2 study of backbone adaptation and photometric augmentation, trained on synthetic images and tested on both real hardware-in-the-loop domains, we find three things. (1) The model does not know it is failing: regions conformally calibrated to 90% on synthet
sim-to-realbenchmark - arxiv:2611.20314 · cs.RONF-VLA: A Normalizing-Flow Baseline for Vision-Language-Action LearningRuichen Deng, Qinyu Zhao, Yong-Hyun Park, Mutian Tong +2
Autoregressive normalizing flows (NFs) provide a tractable density for learning and sampling continuous robot actions. We show that a single maximum-likelihood objective can train vision--language--action (VLA) policies competitive with leading methods on manipulation benchmarks. We present NF-VLA, which combines conditional autoregressive normalizing flows with score-based action refinement. Its deep--shallow architecture concentrates visual, language, and proprioceptive prefix conditioning in a high-capacity deep block, while unconditioned shallow blocks provide a shared invertible action transformation. The vision--language backbone and flow are jointly trained by exact maximum likelihood on Gaussian-smoothed action chunks. The same learned density supplies a single score-based correction at inference, refining generated actions to account for training-time smoothing without a separately trained denoiser. Trained on LIBERO from a pretrained vision--language backbone without robot pr
vision-language-actionvlamanipulationliberobenchmark - arxiv:2611.20306 · cs.ROOntology-Compiled Control Barrier Functions for Heterogeneous Robot TeamsPeng Wu, Seyede Fatemeh Ghoreishi, Nathan Bastian, Mohsen Imani +2
Safety filters for heterogeneous robot teams fail because of missing constraints, overestimated braking capabilities and conflicts between individually satisfiable constraints. We present OntoCBF, which compiles a domain ontology with a class taxonomy, typed relations, safety axioms and a dynamics profile per robot class into a control barrier function filter with one quadratic program per agent. Description-logic entailment supplies the constraints of any team online, including those that follow from chains of relations and from classes defined by what a robot carries, so the existing safety axioms cover new robot classes, payload classes and combinations. We prove that a subclass may reuse the braking capability of its parent whenever it can produce every velocity of the parent, a property separate from subsumption. For box-bounded double integrators this property is also necessary, and every loss of safety then begins with a conflict. Once three robots interact, the safe set of pair
multi-agent - arxiv:2611.20293 · eess.SYModel-Free Stealthy Attack on Vehicle Systems and Its Implications for ISO/SAE 21434 Risk AssessmentQazi Mairaj ud din, Qadeer Ahmed
This paper investigates a model-free stealthy adversary targeting in-vehicle networks and its implications for automotive cybersecurity risk assessment. Conventional stealthy false-data-injection (FDI) attacks often assume privileged knowledge of system dynamics, estimator parameters, or detector statistics, understating the feasibility of adversaries with limited system information. We develop a model-free FDI attack that uses a Vector Autoregressive model with Exogenous inputs (VARX), identified from passively recorded nominal measurements, to characterize a compromised channel and construct attack sequences within a data-calibrated residual envelope. The construction requires no vehicle dynamics model, estimator parameters, assumed noise distribution, or detector knowledge. In a closed-loop lane-keeping benchmark, the attack induces a 1.25 m lateral displacement, exceeding the specified 0.9 m half-lane width, and keeps the vehicle outside the lane corridor for 9.21 s. At a matched 1
benchmark - arxiv:2611.20292 · eess.SYCausal Discovery for Resiliency-Prioritized Transmission Expansion Planning in Power Systems with Grid-Forming Inverter-Based ResourcesRoy McCann, Mustafa Mhawesh, Sandra Canas Rivera
Transmission planners at regional transmission organizations (RTOs), independent system operators (ISOs), and NERC-registered entities have limited analytical resources for exploring expansion options that improve resiliency for withstanding and recovering from fault conditions in electric power systems that are dominated by grid-forming (GFM) and grid-following (GFL) inverter-based resources (IBRs). Conventional existing TPL-001 contingency analysis will assess if a performance benchmark is violated. However, it will not identify which dynamic states and inputs caused the violation, nor how much a protective action will avoid propagation into a cascading outage. This paper proposes an analytical method that combines time-series causal discovery using the Peter and Clark Momentary Conditional Independence (PCMCI) algorithm with right/left eigenvector participation-factor analysis of validated small-signal state-space models. Synthetic data are generated from linearized models derived f
benchmark - arxiv:2611.20289 · cs.AIBenchmarking LLM Inference on Desktop-Scale NVIDIA Grace Blackwell ClustersKevin Bhimani, Christian Escobar, Daniel Mihovch, Rudolf Eigenmann +1
Two NVIDIA DGX Spark GB10 systems connect directly over 200 GbE to serve larger language models and increase inference capacity, but the benefits depend on how they are deployed. We benchmark 17 open-weight checkpoints spanning dense and mixture-of-experts architectures, 8B to 235B parameters and four weight precisions on one and two RDMA-connected systems. Using vLLM, we compare tensor parallelism (TP=2), pipeline parallelism (PP=2) and a two-replica baseline at concurrency 1 to 64, with three independent launches per configuration. TP=2 delivers a median 1.59x peak-throughput gain over one Spark, versus 1.39x for PP=2 and has higher throughput in 93 of 96 matched comparisons with separated 95% Welch intervals. It reduces unloaded inter-token latency to 0.52x to 0.73x of one Spark, whereas PP=2 provides little reduction. TP=2 also achieves the highest goodput under a strict 100 ms inter-token latency objective wherever both strategies are supported and the objective is met. For models
benchmark - arxiv:2611.20278 · cs.CVVisual Marking as a Native Reasoning Capability for Multimodal LLMsJiazhen Liu, Yanghao Wang, Chaolei Tan, Long Chen
Visual reasoning often requires maintaining intermediate spatial states that are difficult to express in text. Existing approaches either rely on external drawing tools or image generators, or encode these states in latent tokens without explicit spatial correspondence to the input image. We introduce SelfMark, which enables MLLMs to construct and use visual marks directly in their native visual feature space. By treating the visual token grid as a canvas, SelfMark establishes explicit spatial correspondence between marks and images. This is achieved through a locate-mark-read procedure, where the model identifies relevant locations, constructs task-dependent mark features, and reads the resulting visual features to guide its reasoning. We jointly train a single model on spatial planning, dense counting, high-resolution VQA, and line tracing. Across six benchmarks, SelfMark outperforms state-of-the-art MLLM-based methods specialized for individual tasks. These improvements also persist
benchmark - arxiv:2611.20273 · cs.RODexFormer 2.0: Towards a Cross-Embodiment Foundation Policy for Dexterous GraspingLixin Xu, Ke Zhang, Xinyu Liu, Jie Tan +6
Dexterous grasping policies are usually trained for one arm paired with one hand. Every new pairing then needs its own policy, and the number of policies grows with every arm and hand added. We ask whether a single policy can control many arm-hand combinations and transfer to pairings and hands it has never seen. This is hard because robots differ in the number, layout, and dynamics of their joints, so their observations and actions share no common interface. We present DexFormer 2.0, a single reinforcement learning policy for cross-embodiment dexterous grasping. It maps every robot to a canonical joint interface with validity masks, encodes each joint's morphology with attention biased by the kinematic tree, and uses a Transformer-XL memory to infer embodiment-dependent dynamics from recent interaction. Trained on 18 combinations of four arms and six hands, it reaches 89% stable success on seen combinations and 51% on six held-out arm-hand pairs, outperforming GRU and DexFormer 1.0 ba
dexterousgrippergraspmemory - arxiv:2611.20269 · cs.LGSelf-supervised learning for zebrafish embryo phenotype evaluation: Systematic comparison of algorithms, label-efficient classification, and transfer learningThomas Lautenschlager, Gaëlle Hayot, Quentin Brassart, Angelo Jovin Yamachui Sitcheu +9
Zebrafish embryos are a prominent model organism for image-based phenotyping and especially suited for large-scale imaging studies, for which machine learning (ML) models can automate evaluation. However, training ML models usually requires large labeled datasets, and differences in imaging conditions often impede reuse across laboratories. Self-supervised learning (SSL) addresses both constraints by learning transferable representations from unlabeled images. Yet central aspects like the choice of algorithm remain underexplored for domain-specific applications. To close this gap, we benchmark popular SSL algorithms on the public EmbryoNet dataset, explore label-efficient learning, and test transfer to an unseen setting. Using representations from our strongest model, a ResNet50 network trained with SwAV, a simple linear classifier reaches 86.4% accuracy and a non-linear classifier 87.1%, rivaling the fully supervised EmbryoNet model (87.1%). The SwAV model decisively outperforms super
benchmark - arxiv:2611.20261 · cs.LGEstimating Outpatient Visit Prices from Hospital Price Transparency Data using Large Language ModelsKurmanbek Kaiyrbekov, Nicholas J Dobbins, Sean D Mooney
Objective: Hospital care prices in the United States have increased substantially, yet patients lack effective tools to access and compare costs. We aimed to develop and evaluate an LLM-based approach for improving access to and use of hospital price transparency (PT) data. Materials and Methods: To estimate visit prices, we used a retrieval-augmented generation (RAG) framework to analyze PT files from three major hospitals in Maryland and the surrounding region. We compared BM25, OpenAI embedding models, and a GPT-BM25 hybrid against a no-RAG baseline for cost estimation of individual procedures and procedure groups, then evaluated the best approach for symptom-based visit cost estimation using synthetic simulations. Results: Across hospitals, the GPT-BM25 retriever achieved the best performance, with average price MAPE of 19.44% for individual procedures and 4.90% for groups, compared with 147.98% and 82.82%, respectively, for the no-RAG baseline. In simulations, our framework correc
retrieval-augmentedrag - arxiv:2611.20259 · cs.LGScaling large language model-assisted line-of-therapy adjudication with human-in-the-loop evidence-calibrated selectionOlatunji Oni, Sanhita Sengupta, Marc De Benedetti, Nathanial Eddy
Large language models (LLMs) have enabled the ability to scale clinical adjudication tasks traditionally dependent on exhaustive manual review. Line-of-therapy (LoT) adjudication, a major bottleneck in oncology cohort construction, is a prime example. Line boundaries must be inferred from longitudinal treatment histories according to nuanced, study-specific definitions, at every regimen transition for every patient. LLMs can apply this adjudication at scale, but their reliability varies across patients. Treatment histories differ in complexity, and ambiguous or atypical patterns depend on expert clinical judgment. Reliable automation thus requires verified support for model assignments and the ability to identify cases that warrant manual review. Here, we present evidence-calibrated selection (ECS), a human-in-the-loop framework for scaling LLM-assisted LoT adjudication. ECS derives confidence from observed clinician adjudications, prioritizes patients with the least-supported assignme
human-in-the-loop - arxiv:2611.20255 · cs.LGDeep Sequence Model for Genome Wide Discovery of Coding and Regulatory Element SignaturesRayhaneh Shabani Nia, Ali Karkehabadi
Interpreting deep neural networks for genomic sequence classification remains challenging despite strong predictive performance. We develop a lightweight CNN with post-training gradient-based analysis to identify which sequence positions drive coding versus intergenomic classification. Applied to the standardized demo_coding_vs_intergenomic_seqs dataset, our model achieves 91.87% validation accuracy while revealing interpretable patterns: nine hot spots with consistently high importance, clear class-specific separation at positions 20-100 (coding) and 150-190 (intergenomic), and a strong mean-variance correlation (r = 0.530) indicating robust discriminative features. Gradient-based importance analysis shows that the model implicitly learns biologically meaningful sequence distinctions without explicit annotations. This work demonstrates that neural network interpretability and accuracy can coexist, providing a framework for understanding genomic sequence classification and enabling bio
post-training - arxiv:2611.20246 · cs.LGTileCache: Serving Multi-Round Diffusion Image Editing with Compressed Activation CachesYuyang Tian, Desen Sun, Shuncheng Jie, Lishan Yang +1
Multi-round diffusion image editing can reduce computation by reusing intermediate activations, but retaining these activations across editing rounds introduces substantial memory and bandwidth overheads. Compression can reduce these overheads, but efficient cache construction is challenging: compressing only after editing requires buffering over 100 GB of activations per request. Moreover, the compressed cache must support selective reuse, as future masked edits require only a subset of the cached activations. It also exposes a scheduling challenge, as cached edits are faster than initial generations, causing naive policies to severely delay either initial generations or editing requests. In this paper, we aim to address both challenges through compression-informed serving. We develop TileCache, a serving system for multi-round diffusion editing. It compresses reusable activations and incorporates a scheduler that balances both edits and initial generation requests. Compared with the
memory - arxiv:2611.20244 · cs.RORight Goal, Wrong Body: Re-grounding Frozen VLAs under Actuation ShiftUmer Siddique, Peilang Li, Yongcan Cao
Vision-language-action (VLA) models are capable of producing desired robotic behaviors in ideal settings. However, they often fail when the robot's actuation changes at deployment. For example, remounting, mechanical wear, or command latency changes can lead to failed tasks and unsafe behaviors. To address the inevitable actuation shift, we propose ReGround, an inference-time method that keeps the VLA frozen and re-grounds its intended motion. ReGround learns the deployed robot's dynamics from task-agnostic motor babbling, and then tracks the policy's intended motion while staying close to the policy's own plan. ReGround activates only upon detecting a shift, and needs no rewards, demonstrations, or retraining. We show via two VLAs, eight LIBERO tasks, and five shift settings that ReGround can restore near-oracle success, such as raising $π_{0.5}$'s success from 0.16 to 0.90 under 400 ms latency, while remaining safer than a linear-calibration baseline. Project website and open-source
vision-language-actionvlalibero - arxiv:2611.20241 · cs.CVEvoMem3D: Adaptive-Depth Temporal Representation Memory for Multi-Camera 3D DetectionMoshe Kimhi, Idan Kligvasser, Yotam Intrator, Regev Cohen +1
Modern multi-camera 3D detectors increasingly rely on large Vision Transformer (ViT) image encoders, making repeated full-depth encoding a dominant inference cost even though adjacent driving frames are highly redundant. We present EvoMem3D, an adaptive-depth temporal framework that caches the most recent full-depth detector-interface appearance representation as a read-only anchor and combines it with policy-selected partial features from the current frame. A lightweight pre-encoding controller chooses among semantic encoder exits before image processing, allocating shallow computation to stable frames and deeper computation when the visual state changes. To reuse appearance without carrying stale geometry, we decouple temporal appearance from geometric alignment: current-frame calibration defines the decoder geometry at every timestamp, while a motion-conditioned positional adapter accounts for anchor-to-current ego motion. On nuScenes, EvoMem3D achieves 61.50 NDS and 52.76 mAP while
memory - arxiv:2611.20230 · cs.AIA Matrix with Steps: An Episodic and Procedural Memory Method for Tool-Using Agents and RobotsJürgen Paulhart, Peter Schüller
Agents that call tools and robots that execute tasks are challenged by two shortcomings: they repeat what failed before, and they learn from runs that they only believe succeeded. We describe the construction of a memory that addresses both, built for a production agent platform. Every run is recorded as an episode, and only an outcome confirmed by an external check, a test, a tool result or a person counts; the agent's own claim never does. What is learned from episodes is kept in a memory with two parts: an associative matrix, written with the delta rule, which finds the procedure that applies to a situation; and a section of ordered, addressable step records that answers what comes after a given step, where a step is a goal of the agent, not a motion. Because the matrix sums its associations, it keeps no order and cannot return or delete a single one; it serves as the finder, and order, outcome counts and deletion live in the records. We describe how execution outcomes are stored, h
memoryagent - arxiv:2611.20218 · cs.AISWE-CoWork: Benchmarking Coding Agents as Digital Employees in OrganizationsXinping Lei, Minghao Liu, Ruijie Zhu, Xiaohui Hu +22
A coding agent working within a software organization must establish what to build, collaborate with colleagues, and verify what it delivers. We introduce SWE-CoWork, a benchmark comprising 101 development assignments across 101 real repositories, 11 programming languages, and 12 application domains. Each assignment places an agent in a simulated engineering organization with 26 colleagues who have distinct responsibilities, bounded knowledge, and controlled collaboration behavior. Starting from a brief handoff, agents must recover requirements distributed across colleagues and workplace records, reconcile conflicting evidence, and implement changes evaluated by private executable tests. We evaluate eleven models and find a substantial gap between partial implementation and complete task delivery. The two leading models satisfy approximately 63% of fail-to-pass (F2P) tests but complete fewer than 8% of assignments. Providing complete requirements upfront improves F2P by approximately 1
agentbenchmark - arxiv:2611.20207 · cs.CVAstroPIE-53: Towards Understanding Failure Modes of Human Pose Estimation in SpaceStephanie Käs, Marten Flaig, Lukas Woyke, Osmanalp Ömeroğlu +3
Reliable monitoring of astronauts is essential for safety, situational awareness, and autonomous assistance in space missions. While modern computer vision enables accurate human detection and pose estimation on Earth, its robustness in space environments remains largely unquantified due to limited data availability. We present AstroPIE-53, a pilot dataset of 53 annotated astronaut videos collected from publicly available mission footage. To respect copyright restrictions, we release video identifiers, reconstruction scripts, and dense annotations rather than raw videos. The dataset includes bounding boxes, challenge classes, and human-annotated pose failure tags. We benchmark state-of-the-art person detection and pose estimation models under realistic zero-shot deployment conditions and perform the first systematic failure analysis for astronaut monitoring. Our results reveal substantial performance degradation caused by occlusions, unusual body orientations, low contrast, and microgr
benchmark - arxiv:2611.20202 · cs.RODAPPT: Domain-Aware Policies from Procedural Trajectories for Sim-to-Real Dexterous Tool UseBeining Han, Jia Deng
Sim-to-real RL has achieved impressive results in dexterous manipulation but remains limited to tasks with accurate rigid-body simulation. However, these methods cannot be applied to skills that involve deformable and particles, with which the simulation speed will become too slow for large-scale RL learning. We present DAPPT, a sim-to-real framework for end-to-end dexterous tool-use manipulation skills, e.g., sand scooping, play-dough rod-rolling, without simulating non-rigid materials. Our method learns a domain-aware RL policy by tracking procedural trajectories parameterized by domain parameters, and distills it into an end-to-end domain-aware sensorimotor policy. Furthermore, explicit domain represen-tation enables fast selection of domain parameters that can maximize downstream task performance. We evaluate DAPPT end-to-end on six dexterous manipulation tasks, including sand scooping, play-dough rod-rolling, sand scraping, mouse-wheel scrolling, dispenser pressing, and spray trig
manipulationdexteroussim-to-realtool usetool-use - arxiv:2611.20196 · cs.LGDoF-Query Transformer for Adaptive Fusion of Egocentric Vision and Forearm HD-sEMG in Continuous Hand KinematicsJiahang Liu, Wenbo Zhang, Junsong Wang, Zhenzhi Ying +3
Continuous and accurate estimation of human hand kinematics is important not only for intuitive human-machine interaction, but also for capturing fine-grained human manipulation demonstrations that can provide action supervision for embodied-intelligence datasets and dexterous robot learning. However, individual sensing modalities may be insufficient for reliable estimation under varying operating conditions. We present a finger-unencumbered multimodal framework for continuous regression of 20 hand joint angles from egocentric RGB images and 128-channel forearm high-density surface electromyography (HD-sEMG). The framework combines a Vision Transformer-based visual encoder, a Transformer-based EMG encoder, and a degree of freedom (DoF) query Transformer fusion decoder. To account for the varying informativeness of visual and EMG cues across joint DoFs and input conditions, the decoder assigns a separate learnable query to each DoF, enabling DoF-specific and context-adaptive fusion of t
embodiedmanipulationdexterous - arxiv:2611.20191 · cs.CLDo We Still Need Fine-Tuned Classifiers? Swapping Typed Decision Models into 24 ACL-Venue PapersMarek Šuppa, Jaroslav Kopčan
Typed Decision Models such as TypeSafe's Jev, also called System One or System 1 models, answer typed questions about an input with a probability for each option, and they need no training. Many NLP papers still fine-tune a task-specific classifier, and prompted large language models (LLMs) trail such classifiers on most text classification benchmarks. We ask how many of these fine-tuned classifiers a Typed Decision Model can replace under each paper's own evaluation, and what is needed where it cannot. We swap the predictors of 24 ACL-venue papers for Jev, keep each paper's splits, label definitions and metric, and check the results with Cloudflare's Clef, Microsoft-Decision-1 and OpenAI's GPT-6 Luna Decisions. The resulting 46 English test sets, with their questions and published scores, form the TDM-Gauntlet benchmark. With a zero-shot prompt, Jev directly replaced the paper's predictor on 12 of the 46 test sets, mostly single judgments with plain labels, but rarely where the paper'
benchmark - arxiv:2611.20187 · cs.LGCyber-Secure Wearable, Ultra-Low-Power Networked Sensors for Soldier Health Monitoring: Insights from the CAPSARII projectLuciano Bozzi, Christian Celidonio, Umberto Nuzzi, Massimo Biagini +18
The CAPSARII project advances a cyber-secure, ultra-low-power wearable platform and Internet of Battlefield Things (IoBT) framework to support real-time monitoring of military personnel and data-driven decision support. The architecture integrates smart-textile sensing, a body/squad area mesh, edge-deployed AI on secure RISC-V microcontrollers, and cloud/federated analytics for cross-mission learning and comparative studies. Two representative use cases--pilot fatigue and warfighters operating in harsh environments--guide requirements, validation, and interface design (including HUD and haptic feedback). As an initial step toward embedded models, we benchmark stress classification on a public wearable dataset: tree-ensemble baselines achieve state-of-the-art performance for accuracy and F1 (e.g., XGBoost: $\mathrm{Accuracy}=0.901\!\pm\!0.023$, $\mathrm{F1}=0.920\!\pm\!0.019$), while a feature-based Transformer attains the highest AUC ($0.963\!\pm\!0.025$), informing feature-engineering
memorybenchmark - arxiv:2611.20185 · cs.RORoboKit: Accelerating Parallel Differentiable Optimization for Robot LearningJianglong Ye, Xuanbin Peng, Conor Mc Gartoll, Changwei Jing +4
We present RoboKit, a high-performance, general-purpose toolkit for parallel differentiable optimization in robotics, built on three principles: (i) Performance and flexibility: GPU-parallel multi-seed solvers pair sampling with versatile optimizers, running 2-60$\times$ faster than task-specific baselines across seven tasks; (ii) Geometry, rendering and physics: hybrid and coarse-to-fine representations accelerate collision, differentiable rendering and simulation, optimizing 4096 grasps in 2.97 s; (iii) Integration with learning: costs are differentiable and simulation is parallel, so RoboKit both distills optimization into policies and trains RL; distilled policies run 9-240$\times$ faster at comparable accuracy, and RL reaches a GPU baseline's scores up to 2.1$\times$ sooner on the same simulator.
grasp - arxiv:2611.20183 · cs.CVGeoReVis: Learning Selective Revision and Evidence Consolidation for Remote Sensing SegmentationKe An, Liang Chen, Yupei Wang
Recent methods try to inspect and revise masks to mitigate referential ambiguity and localization errors caused by complex spatial relations in language-guided remote sensing segmentation. However, these solutions often lead to Revision Degradation: revising missing regions may introduce errors into previously correct ones. Even a globally better mask can omit correct regions found in inferior masks. We term this Evidence Fragmentation: correct target evidence is scattered across imperfect masks. All masks generated by the model at this time are suboptimal. We propose GeoReVis, which unifies revision guided by execution feedback, Candidate Selection, and Cross-Mask Evidence Consolidation. It learns fine-grained revision from real actions that improve segmentation when executed and learns to select a better mask through quality comparisons between actual candidates, reducing the risk of degradation during revision. The selected whole mask can still discard useful regions. GeoReVis there
benchmark - arxiv:2611.20176 · cs.LGPhase Transitions in Heavy-Tailed Mean Estimation under $\ell_p$ NormsIshaq Aden-Ali, Yeshwanth Cherapanamjeri, Mikael Møller Høgsgaard, Kasper Green Larsen +1
We study the problem of estimating the mean of a random vector in $\mathbb R^d$ under $\ell_p$ norms, assuming only the existence of a covariance matrix. In the Euclidean norm, the sample mean is already optimal in expectation, and robust estimators are usually needed only to obtain high-probability tails. Our results show that this picture changes for $\ell_p$ norms with $p>2$: robustification is already necessary even for the expected risk. We prove that the minimax risk exhibits a sharp phase transition, controlled by the interaction between the sample size $n$, the dimension $d$, and the norm parameter $p$. In the large-sample regime, the classical Gaussian rate remains achievable. In the complementary regime, heavy-tailed distributions create a genuinely non-Gaussian obstruction. Furthermore we show that the simple estimator: coordinate-wise median of means whose number of buckets chosen according to the geometry of the loss, optimally tracks this phase transition in its expected
benchmark - arxiv:2611.20144 · cs.AIHEAMS: Holistic Evaluation of Agent Memory SystemsChirag Goel, Jacob Lavoie, Amirhossein Naghshzan, Jordan Bannister +6
Memory systems allow AI agents to persist information across interactions, making them more personalized and more capable on long-horizon tasks. This persistence introduces new risks: stored information can be disclosed in inappropriate contexts, resist correction or deletion, be retained beyond what users desire, or be poisoned to influence later agent behaviour. Yet memory systems are typically evaluated on task performance and efficiency alone. We argue that a memory system should instead be understood as the custodian of its user's persistent information, with responsibilities that extend beyond helping an agent complete its tasks, and that it should be evaluated holistically against all such responsibilities. We present Holistic Evaluation of Agent Memory Systems (HEAMS), an open-source framework for evaluating agent memory systems in this way. First, we state the responsibilities of agent memory systems as explicit behavioural desiderata spanning performance, efficiency, security
memoryagent memoryagentai agentbenchmark - arxiv:2611.20140 · cs.AICan AI Judge You? An Empirical Audit of Deployment Readiness in LLM-assisted Hiring SystemsSpandan Roy, Subhabrata Majumdar, Soudeep Deb
Delegating resume screening to large language models (LLMs) requires organizations to establish whether they can distinguish candidates by relevant qualifications while allowing only authorized users to change screening rules or approve exceptions. To this end, we develop a system architecture that separates candidate evidence, evaluation, permission controls, and assessment record, and empirically examine two complementary dimensions of its deployment readiness. First, we assess its operational readiness on four LLMs using a controlled experiment on 210 resumes across 14 job roles and six decision interfaces. Second, we assess its compliance readiness by testing how it handles unauthorized user requests and instructions embedded in resumes. We find that operational performance depends on the choice of evaluation interface: scalar scores generally distinguish qualification changes better than simple pass/fail judgments, while comparative judgments are sensitive to presentation order. E
evaluator - arxiv:2611.20137 · cs.ROCleanly-DINO: Efficiently Learning Visual Task Spaces for RoboticsAdrian Röfer, Melika Shekarriz, José Arce y de la Borbolla, Iman Nematollahi +1
Learning manipulation tasks from few demonstrations requires abstractions that capture the changes relevant to task completion, such as object poses in rigid-body manipulation. However, tasks such as cleaning are defined by changes in visual state rather than object pose, making the choice of task representation less clear. In this work, we propose Cleanly-DINO to learn visual task spaces from few demonstrations. Given a mask of a task-relevant region and frozen ViT features, Cleanly-DINO fits a projection to isolate a task space and fit a progression metric to it. We train the model on diverse cleaning examples and evaluate its ability to localize incomplete task regions and transfer across visual domains. We also study its utility for robot learning by using it as an observation encoder and reward signal in a reinforcement learning setting. Finally, we deploy our fitted model within a real-world robotic wiping pipeline to evaluate its use in guiding task execution.
manipulation - arxiv:2611.20132 · cs.AICapCover Filter: Semantic Approximate Membership Queries over EmbeddingsDvir David Biton, Shay Hahiashvili, Nadav Keren, Gil Einziger +1
Determining whether an embedding is sufficiently similar to any vector in a vector store without performing a nearest-neighbor search can substantially improve query throughput in vector-store systems. We study Semantic Approximate Membership Queries (SAMQs), which allow queries to be rejected when no indexed vector meets the requested similarity threshold, while guaranteeing no false negatives. We introduce CapCover Filter, a geometric data structure that summarizes a set of embeddings using spherical caps. Its representation is independent of the query threshold, allowing a single filter to support arbitrary thresholds without reconstruction. We evaluate alternative construction and quantization methods, compare CapCover Filter with competing filters, and measure its impact on query throughput. Across five real-world datasets, at a cosine similarity threshold of 0.95, CapCover Filter eliminates approximately 20--100\% of index accesses while using only 2\% of the raw (FP32) embedding
evaluation framework - arxiv:2611.20126 · cs.CLA Corpus-Driven SentencePiece Unigram Tokenizer for Sindhi: Design and Benchmark Against Fourteen Existing TokenizersAli Nawaz, Muqadas Fareed, Soonh Sadaf
Sindhi is a low-resource Indo-Aryan language written mainly in an extended Perso-Arabic script. Several Sindhi tokenizers have been released on public model hubs, but they have not been compared with one another on shared data. We train SentencePiece Unigram tokenizers with byte fallback at four vocabulary sizes (32k to 150k) on about 1.10 million Sindhi sentences and a smaller, repeated Urdu, Persian and Arabic slice, from four public sources, and compare them with fourteen existing tokenizers on four evaluation sets: FLORES+, Sindhi Wikipedia, a news-derived sentiment corpus, and a contemporary news corpus. The metrics are fertility, single-token retention, UNK rate, byte-fallback rate, and round-trip character loss. At 64k, our tokenizer averages 1.147 tokens per word, against 1.208 for the best existing tokenizer and 1.512 for XLM-RoBERTa, and keeps 88.6% of words as single tokens. On fertility, single-token retention and UNK rate it is best or tied for best in all four evaluation
benchmark - arxiv:2611.20109 · cs.LGMT-RGCN-DTI: A Multi-Task Relational Graph Convolutional Network for Drug-Target Interaction PredictionDheeran Sankaran, R Yadunandan, Daniele Santoni, Nimisha Ghosh
Prediction of drug-target interaction (DTI) has become a very crucial part of drug discovery. With the increase in the data volume, it has become very important to improve prediction accuracy by integrating multiple data sources. In this regard, utilising multiple resources in the form of heterogeneous information network to address DTI prediction can be rewarding. In this work, we propose a heterogeneous information network based multi-task DTI prediction model which leverages the advantages of relation graph convolutional network (RGCN). To achieve this, the model employs drug-target interaction network, drug-drug similarity network and protein-protein similarity network to construct a heterogeneous information network. Moreover, we use a knowledge graph embedding scoring functions to model interactions between drug and targets, specifically DistMult and ComplEx as decoders over the learned node representations. We performed extensive experiments on multiple benchmark datasets and co
knowledge graphbenchmark - arxiv:2611.20107 · cs.ROSAFER: Navigating the Exploration-Exploitation Dilemma in World-Model-Based Planning for Autonomous DrivingYansong Qu, Zilin Huang, Zihao Sheng, Sikai Chen +1
An autonomous vehicle must foresee the consequences of its actions, especially in the rare, time-critical situations where crashes happen. A world model offers this foresight, but only within the coverage of its training data, which poses an exploration-exploitation dilemma. Within this coverage a planner can hardly improve on the data, and beyond it the model's errors mislead it. Logs of competent driving rarely contain the failures that would widen the coverage, and agents that learn from interaction gain them only by failing themselves, without being shown what to do instead. We propose SAFER, a framework that learns offline from failures that others make, judged by an expert, and keeps learning online from its own driving. Offline, partially trained agents and scripted, recoverable perturbations Stage failures in simulation, and an expert Annotates every visited state with its action and value, so that much of what the planner explores is covered before it drives. In planning, a ch
world modelleaderboard - arxiv:2611.20106 · cs.AIMaking Preemption Productive in Long-Output LLM ServingZiji Shi, Xinyi Wan, Chaoyi Ruan, Guangxing Huang +2
Reasoning LLMs generate long outputs whose key--value (KV) caches occupy GPU memory for minutes. When memory fills, the server preempts requests, and a preempted request can repeatedly lose and recover its state while producing little new output. We call this behavior KV thrashing. In certain cases, existing systems leave more than a fifth of requests stalled for over a minute. Because preemption cannot be avoided without delaying admission, we ask how to make it productive. We present Stint, a serving system that preserves a preempted request's work, bounds how long it waits, and gives it fresh output after it returns. Stint stores fixed-length KV snapshots in a contiguity-first CPU pool with an exact-extent allocator, and moves them with fused K/V gather/scatter kernels and double buffering. Age-based readmission brings stalled requests back, and protection gives them a turn of fresh decoding before another preemption. Remaining-length estimates from decoding hidden states guide vict
memory - arxiv:2611.20100 · cs.ROEgoSwarm: Learning Decentralized Robot Swarm Intelligence from Human Swarm DemonstrationsZiqiang Liu, Xitong Chen, Min Li, Yiqun Li +1
Advancing from individual embodied intelligence to swarm embodied intelligence is a key step toward scalable, general-purpose embodied intelligence. Complex tasks often require multiple agents to divide responsibilities, coordinate handovers, and manipulate shared objects collaboratively. However, existing robot learning methods rely primarily on individual demonstrations, which provide limited supervision for modeling dependencies among agent actions, while collecting multi-robot demonstrations at scale remains difficult. Human egocentric demonstrations offer a scalable alternative, yet distributed observations from multiple participants cannot directly supervise decentralized policies. We introduce EgoSwarm, the first robot learning framework for learning decentralized multi-robot collaboration from human swarm demonstrations. EgoSwarm collects more than 50 hours of collaborative experience using a robot-free multi-person egocentric system. We introduce the Swarm Interaction Packet (
embodiedagent - arxiv:2611.20095 · cs.LGHigh-Throughput Serving of Nemotron 3.5 Transcribe with Regime-Aware Megakernels and Kernel FusionAditya Chatterjee, Sanjay Singh Chauhan, Mayank Jain, Arushi Dhingra +2
Nemotron 3.5 Transcribe ranks first among open-weight models on the Hugging Face Open ASR leaderboards. Serving it efficiently takes more than fast operators: continuous batching changes the amount and mix of work at every iteration, and the best fusion granularity changes with it. We present a serving engine that, for every pass, chooses between persistent megakernels and chains of fused kernels at crossovers measured offline. On an H100 SXM GPU, throughput of our accuracy-preserving serving peaks at 4,345x real time, 1.39x the peak of tuned vLLM, and is up to 4.39x faster at low concurrency; on B200 it peaks at 6,367x, 1.64x vLLM at the highest concurrency we measured. Word error rate (WER) is 1.91 to 1.93% against 1.88 to 1.93% for vLLM. Within a 400 ms budget for p95 end-to-end completion time, it serves 8x as many concurrent requests. Choosing per pass beats either level alone: 23% higher throughput than always using megakernels at 128 concurrent requests, and 9% higher than alway
leaderboard - arxiv:2611.20094 · cs.AIFRAAC: A French Multimodal Large Language Model and Dataset forPictogram-Based Augmentative and Alternative CommunicationAlexandre Audibert, François Portet, Benoit Favre, Maja Jonck Hjuler
Pictograms are a core modality in augmentative and alternative communication (AAC), yet recent multimodal large language models (MLLMs) cannot reliably interpret or produce pictogram sequences. Two obstacles explain this gap: the absence of a standardized pictogram grammar and the scarcity of parallel text-pictogram data. We introduce Retrieval-Augmented Pictogram Selection (RAPS), an annotation method that refines a rule-based transcription system's output with LLM-based concept extraction and retrieval over a pictogram vocabulary, handling cases where hand-written rules fail (named entities, numbers, polysemous words) without requiring new rules. Applied to French, RAPS yields a synthetic corpus of 1.4 million parallel pairs, of which 300k are publicly released. In a human ranking study where annotators order four pictogram transcriptions of the same sentence, RAPS outperforms the rule-based state of the art. Moreover, we show that standard reference-based metrics separate systems of
retrieval-augmented - arxiv:2611.20090 · cs.CVLatent World Modeling for Point-Cloud Video PretrainingZhuoyue Zhang, Jian Liu, Jihua Zhu, Ajmal Saeed Mian
Masked modeling is the dominant self-supervised paradigm for point-cloud videos. However, it leaks the information it is meant to predict, whether it reconstructs masked points as in MAE-style methods or predicts their latent features. Because point-cloud tokens are anchored at points sampled from each frame, querying a target at its own position reveals where its geometry has moved, in reconstruction and latent prediction alike. For latent predictors, random spatio-temporal masks introduce a further shortcut, as visible frames on both sides of each target allow interpolation rather than forecasting. The pretext task thus reduces to completion. Once the target positions are withheld, a masked latent predictor performs only slightly better than copying the last frame. To remove this leakage, we propose PointWM, a latent world model that turns completion into forecasting by predicting the features and geometry of future frames from strictly earlier context. Each target is queried only wi
world model - arxiv:2611.20078 · cs.LGInfrastructure-Based Multi-Agent Reinforcement Learning with Attention for Highway Ramp MergingChenhao Tong, Steffen Müller
On-ramp merging is a typical safety and throughput problem on highways. Without proper coordination, human drivers and self-interested autonomous vehicles may make risky manoeuvres and cause accidents or congestion. With recent advances in connected autonomous vehicles (CAVs) and multi-agent reinforcement learning (MARL), vehicles can be trained to coordinate for safe and efficient merging. However, in dense traffic, forming an effective observation of the traffic state is challenging. For example, observing only the neighbouring vehicles may lead to locally optimal decisions, while observing the entire traffic can introduce much irrelevant information. In this paper, we propose a transformer-based MARL vehicle control model to address these problems. A roadside infrastructure controller receives the states of all CAVs via vehicle-to-infrastructure (V2I) communication and sends control commands back to them. An attention-based observation encoder is designed where each vehicle can filt
multi-agent - arxiv:2611.20076 · cs.CVSGIP: Learning Scene-Grounded Interaction Priors for Scene-Aware Human Motion PredictionDonghyeon Soon, Jiil Yoon, Hyeondong Kim, Daehee Park
Human motion prediction (HMP) aims to forecast future movement from an observed motion sequence, and recent methods have incorporated 3D scene geometry and object semantics to improve long-term plausibility. However, the destination of future motion remains ambiguous in cluttered scenes: the same observed motion can lead to different interaction targets or navigation modes. Existing methods either handle this ambiguity implicitly or rely on auxiliary signals like gaze that are hard to obtain in practice. To address this, we propose SGIP, which explicitly predicts where in the scene future motion will be grounded across both objects and open navigation regions, and uses it to condition future trajectory and full-body pose prediction. This scene-grounded interaction prior is learned via intent distillation: at training time only, a teacher model scores candidate scene regions by combining an LLM prompted with a textual description of the person's intent with geometric motion-scene featur
benchmark - arxiv:2611.20070 · eess.SYConsensus-based Economic Dispatch for Self-organized Local Energy CommunitiesDimitrios Lagos, Josh Eichman, Georgios Tsaousoglou
The increasing penetration of distributed energy resources (DERs) is transforming traditional distribution networks into dynamic, decentralized systems, creating a pressing need for scalable, autonomous coordination mechanisms that do not rely on centralized control or full system observability. In particular, local energy communities need to self-organize and coordinate energy schedules while respecting network constraints, especially voltage limits, despite lacking detailed knowledge of the distribution grid's topology and parameters. Existing distributed optimization approaches either neglect such physical constraints or assume perfect knowledge of the network model. This paper addresses the Voltage-Constrained Economic Dispatch (VCED) problem in a fully decentralized, consensus-based manner, without any coordinating entity. We use sparse-group LASSO regression to learn a linear surrogate of nodal voltages from local power and voltage measurements alone. The surrogate serves a dual
benchmark - arxiv:2611.20067 · cs.MAWorld-Model-Guided Plan Search for Cooperative Mixed-Traffic Control at BottlenecksJinghan Peng, Ning Liu, Xiangrui Kong, Sen Shen +1
Cooperative speed control by connected and automated vehicles can reduce congestion at lane drop bottlenecks. However, effective coordination depends on vehicle roles and execution timing, while initial holding costs may be offset by later traffic benefits. This paper proposes an offline plan search framework that combines world model guidance with simulator verification. A traffic world model incorporating physical dynamics guides evolutionary search over vehicle roles, target speeds, holding durations, and start times. Paired SUMO simulations screen candidates for total delay reduction and limits on additional mean delay across vehicle groups. Accepted plans form a reusable library whose actions temporarily override a frozen MAPPO policy. Across 192 paired test scenarios at 60% penetration and a demand of 2,800 vehicles per hour, the searched library reduces mean total delay by 14.3% relative to MAPPO and by 3.8% relative to a handwritten plan library. The largest increase in group m
world model - arxiv:2611.20049 · cs.AIThe Frontier Must Grow - Self-proposed, interpreter-labelled training roughly doubles a 4B model's accuracy on a fresh exam as its verified frontier growsMartin Schroder
A model can learn from problems it poses itself when an executable oracle supplies the answers. Qwen3-4B proposed Lua 5.1 programs; those it predicted wrong at least once in four samples joined a pool labelled by the interpreter, each round adding twice as many programs as the last (300, 900, 2100 in total), and each round trained a fresh adapter from the base on the pool. On the original exam the procedure met its registered criterion, an item-level test of each adapter against its predecessor. On a 2340-item exam drawn afterwards, accuracy rose from 8.3% to 16.7%; that follow-up's registered criteria were not met because round 1 gained nothing, while rounds 2 and 3 added 5.0 and 2.9 points on average, both intervals above zero. One fit on the same amount of base-proposed frontier data reached 14.6%, so the rounds' 2.1-point advantage is inconclusive. Three earlier registered series met no per-lineage criterion, and the same loop showed no gain on repository tasks or for a 27B agent i
agent - arxiv:2611.20045 · cs.AISecurity of Deep Research Agents: Attack Surface, Defenses, Benchmarks, and Agenda Seeding against Research PlanningYang Luo, Guangpu Chen, Yichen Liu, Zhaoyu Chen +4
Deep research and agentic search agents autonomously retrieve content from the open web and generate reports with citations, and their outputs are often taken directly as research conclusions. This paper systematizes security research in this area along four stages, namely retrieval, planning, evidence integration, and report generation. We present a unified threat model, a taxonomy of attacks and defenses, and a review of evaluation benchmarks, and we align the area dimension by dimension with retrieval-augmented generation (RAG) poisoning, prompt injection, and generative engine optimization (GEO). Coding 34 primary studies shows that only 26.5% of them involve the planning stage, and none performs an end-to-end poisoning evaluation on commercial products. To address this gap, we propose the agenda seeding attack, in which the attacker writes no instructions and only introduces a fictitious evaluation metric. Across 3,520 sandboxed research runs under the same budget, GEO and prompt
retrieval-augmentedagenticbenchmark - arxiv:2611.20044 · cs.CLA Dataset for Misogyny Detection in Polish Social Media: Transferability of Hateful Speech Detection from General to Specific Types of HateEwelina Gajewska
Harmful language detection is commonly evaluated on broadly defined hate speech and toxicity benchmarks, yet it remains unclear whether models that successfully detect harmful language in general can transfer this capability to more specific forms of hate. We investigate this problem for misogyny detection in Polish, an area for which dedicated resources remain scarce. We introduce a new dataset of Polish social media comments annotated for misogyny, gender-directed hateful speech. Using this resource, we first assess the reliability of LLM-assisted annotation and then evaluate general-to-specific transfer by comparing models trained on general harmful language data, safety-specialized models, and models fine-tuned directly for misogyny detection. Our results show that strong performance on general harmful language detection does not transfer uniformly to misogyny; safety-specialized models demonstrate substantially stronger generalization ability. The study contributes a new Polish re
benchmark - arxiv:2611.20035 · cs.LGU-HamGNN: a universal model for magnetic Hamiltonians conditioned on spin and Hubbard UHongyu Yu, Yinze Hu, Yizhou Chen, Yang Cheng +4
The electronic structure of a magnetic crystal depends jointly on its spin configuration and the on-site Hubbard correction U, requiring a separate DFT+U calculation for each combination. We introduce U-HamGNN, an equivariant network that takes these two inputs directly and predicts spin-resolved Hamiltonians across diverse crystal chemistries. A spectral-hybrid optimizer that weights each update direction by how consistently the gradient points along it reaches 8% lower Hamiltonian error than Muon and 26% lower than AdamW in a ten-epoch benchmark. The predicted bands reproduce U-dependent gap opening and altermagnetic spin splitting within the trained window and beyond it; exchange couplings and magnetic-order calls derived from the Hamiltonian agree with DFT+U on a calibration set. This universal model provides a new tool for magnetic research.
benchmark - arxiv:2611.20031 · cs.AITowards Agentic Scientific Discovery in Distributed SystemsRashmita Soni, Abhirup Ghosh, Soumya Banerjee
We investigate when adaptive tool selection helps identify a shared equation across clients with different input coverage. Query-ATLAS (query-conditioned Agentic Tool-based Learning for Aggregating Symbolic Equations) uses an LLM to guide symbolic search from client errors and previous proposals in a simulated distributed system. A four-architecture comparison reveals task-dependent opportunities: relative to fixed PySR-LLM, validation acceptances rise from 3/12 to 7/12 on CRK3 and from 11/12 to 12/12 on oscillator2, while MatSci3 favors the fixed workflow. Across 72 settings in a post-hoc six-task subset of eight development tasks, Query-ATLAS achieves 64 validation and 64 in-distribution passes, versus 62 and 59 for the fixed workflow; both pass 47/60 eligible out-of-distribution tests. The fixed workflow retains lower median errors and cost. Broader query coverage emerges as a promising benefit, motivating task-aware tool selection and client-aware fitting for scientific agents.
agentic - arxiv:2611.20026 · eess.SYThe Grid's Own Language: A Common Information Model Layer for LLM Grid AssistantsAntonio Alcántara, Ioannis Karampinis, Spyros Chatzivasileiadis
Assistants built on large language models (LLMs) are entering power system operation. Current designs delegate computations to validated solvers, but the agent still infers the type, role and state of each element from its name and tool-specific formats, and errors propagate to every study. The Common Information Model (CIM), the IEC standard in which operators already exchange network models, defines unambiguously what each element is and how it connects. We propose a CIM layer for LLM grid assistants and implement it in CONDUCTOR, an open-source assistant. Both the agent and its power-system studies read the layer. It labels the provenance of each fact (stated in the data, suggested only by a name, or inferred from topology), loads operators' CGMES models directly, and builds in $\leq$1\,s on the 9,241-bus PEGASE network. Deterministic checks compare every answer with solver results and the layer. On 37 tasks about the CIGRE medium-voltage and IEEE 30-bus networks, the agent gets rig
agent - arxiv:2611.20019 · cs.ROD'BEAST: Logged Bad Experiences as Explanation for Autonomous Driving AgentsAhmad Alfan Alfian Irfan, Nur Ahmad Khatim, Mansur Arief
Reinforcement learning agents learn from mistakes, yet their decisions can be difficult to trace back to those experiences. We investigate whether a tabular agent's decisions can be explained by its bad experiences, defined as the training transitions associated with safety or task events. A fixed training log is replayed from the same initial Q-table under three conditions. FULL retains all updates, SKIP omits bad-experience updates, and NEUTRALIZE retains those updates without their event penalties. We illustrate the method on six classic Markov decision processes and evaluate it in closed-loop driving in CARLA. On the full driving task, FULL reached the goal from 33 of 40 test starts, whereas SKIP and NEUTRALIZE reached it from none, indicating that the bad experiences carry decision-relevant information. At the local level, for 47 decisions shortly before a failure of SKIP or NEUTRALIZE, removing or neutralizing a small group of bad experiences, with a median size of one, caused th
agent - arxiv:2611.20016 · cs.LGGEM-MM: Reducing VLM Hallucination with Entropy-Guided Preference AlignmentYuanshan Chua, Xuejiao Zhao
Vision-language models (VLMs) are routinely aligned with offline pairwise RLHF objectives, but multimodal preference data is scarce and treats every response token as equally informative. Under a small preference budget, pair construction becomes nearly irrelevant: standard DPO-style, image-conditional, and self-improvement objectives all perform within 1.1 points on MM-RLHF metrics and 0.5 points on hallucination rates, matching the SFT baseline. We call this entropy-blind preference optimization: the learning signal ignores policy uncertainty, which in chain-of-thought VLMs concentrates on specific fork tokens where visual claims are made. We introduce GEM-MM, extending entropy-guided preference modeling to VLMs. It samples on-policy candidates, rewards high-entropy reasoning forks, penalizes uncertain final answers, and updates the model via group-normalized policy gradients. On a 3000-prompt MM-RLHF budget with Qwen3-VL-4B-Instruct, GEM-MM achieves the best entropy-depth preference
self-improvementrlhf - arxiv:2611.20012 · cs.CLOffline-First Quantized AI: Design and Evaluation of an On-Device Maternal-Health Assistant for Low-Connectivity EnvironmentsOkenla AbdulQahhar Abimbola, Maurice Abraham John, Adepoju Jeremiah Moyondafoluwa, Omolaiye Samuel +9
Offline language models could extend access to maternal-health information where internet connectivity is unreliable, but deployment requires both clinically appropriate responses and feasible local execution. This study presents an offline Android assistant integrating ready-made quantized language models, a clinically validated local knowledge package, conversational context, and model-initiated local retrieval. Gemma 4 E2B and Qwen3.5 0.8B were compared across 45 conversations and 312 turns per configuration, including a dedicated offline agentic benchmark. Gemma achieved higher response-quality, context-retention, and safety scores; Qwen offered faster generation and lower memory use. Clinician assessment of 150 Gemma responses across nine clinical domains yielded a mean quality score of 3.80/4, with 147 responses meeting the acceptance criterion. A paired RAG-on/off experiment evaluated 74 questions: retrieval improved 11 responses, left 56 neutral, produced four adverse response
memoryagenticbenchmark - arxiv:2611.20011 · cs.ROReactBench-AD: Do Driving VLAs and World-Action Models React to Imminent Hazards in Their Path?Nikos Theodoridis, Reenu Mohandas, Sushil Sharma, Ciarán Hogan +4
Vision-Language-Action models (VLAs) and World-Action Models (WAMs) have emerged in autonomous driving with the promise of strong generalization to long-tail scenarios. Although these models achieve state-of-the-art results on autonomous driving benchmarks, most existing benchmarks are dominated by nominal scenarios and therefore cannot test this promise. To test one such long-tail capability that is critical for safety, we introduce ReactBench-AD, a benchmark that measures how driving models react to objects that appear in the ego vehicle's path. Built on real NAVSIM and PhysicalAI-Autonomous-Vehicles scenes, ReactBench-AD composites a crossing pedestrian, a standing pedestrian, or a stationary car consistently across all camera views and history frames. Each edited scene thus forms an exact, pixel-level counterfactual pair with its original, which allows us to isolate the causal effect of the inserted object on the predicted trajectory. Evaluating four VLAs and two WAMs, we find that
vision-language-actionbenchmark - arxiv:2611.20010 · cs.ROPlanDiv: Test-Time Scaling with Latent Plans for Vision-Language-Action ModelsHanbo Zhang, Lianlei Shan, Liuhaichen Yang, Zezhi Tang
Test-time scaling improves vision-language-action (VLA) policies by drawing several action candidates and executing the best one. We find that its returns diminish quickly when the candidates are resampled from one decoder: they differ by noise rather than by plan, and for each of three resampling sources, going from four to eight candidates adds only 0.06-0.07 to the average number of completed tasks on CALVIN. We introduce PlanDiv, an action head that makes the candidates differ by plan: a lightweight planner proposes K latent plan hypotheses, a shared masked-diffusion executor decodes one action chunk per plan from a single backbone pass, and the policy executes the chunk that its own likelihood ranks highest. The plans are learned from demonstrations alone, without plan labels or rewards. With the backbone, data and training recipe held fixed, eight plan-conditioned candidates raise five-task chain success on CALVIN ABC$\rightarrow$D from 56.9% to 69.9%, while eight candidates from
vision-language-actionaction headopenvlaliberofranka - arxiv:2611.20006 · cs.CVGlance, then Select: Evidence-Tempered Visual Token Selection for Extremely Compressed Multimodal LLMsYuxuan Wang, Guangxiang Zhao, Yuxuan Tian, Yahui Han +6
Multimodal large language models (MLLMs) encode a high-resolution image into thousands of visual tokens, which dominate the cost of the prefill stage. Visual token reduction removes most of these tokens, but two problems remain at extreme budgets. The vision encoder still processes every patch of the image, and the few surviving tokens often miss the small regions that hold the answer. In this paper, we propose Glance-ETD, a two-stage token reduction framework guided by a lightweight question-conditioned scorer. In the first stage, a glance encoder views the image at low resolution and re-encodes only the question-relevant regions at native resolution. In the second stage, Evidence-Tempered Diversity (ETD) selects the final tokens by combining diversity with the evidence predicted by the scorer, so that the kept tokens cover the scene and retain the answer. A single-rollout self-distillation recipe trains the MLLM with the glance encoder. With only 5% of the visual tokens, Glance-ETD r
benchmark - arxiv:2611.20003 · cs.LGThe novelty exposed: ablation, reproducibility and transparency in 40 technical reports of open-weight large language models (LLMs)Bahaeddin Türkoğlu
In 2025 alone, 48 new large language model (LLM) architectures were added to the Hugging Face transformers library, and many of them arrive with a technical report that presents the model as new. What is actually new in these models, and is it tested? We put this question to LLM technical reports. We look at three things: what the reference code shows about how new open-weight architectures differ from earlier ones; what 40 randomly sampled technical reports claim as new and whether they support it with ablation studies; and what they disclose for reproducibility and transparency. In the library, 39 of the 48 architectures added in 2025 take at least two of their three core blocks from another model, most often Llama, although part of this reflects how the library writes code. In the reports, 19 of 73 architectural or training novelty claims come with a controlled with/without comparison; 4 of 40 reports state more than one training run; 8 rerun the models they compare against under on
judge model - arxiv:2611.19999 · cs.LGMOBO-CAPS: Multi-Objective Bayesian Optimization with Cardinality-Aware Pareto SelectionHanyang Wang, Juergen Branke, Matthias Poloczek
Multi-objective Bayesian optimization (MOBO) traditionally aims to approximate the entire Pareto front, but, due to resource constraints, real-world deployment is often limited to testing a small, representative subset of solutions. Hence current methods first approximate the full front before selecting solutions, potentially wasting costly evaluations. We formalize this problem as Multi-objective Bayesian Optimization with Cardinality-aware Pareto Selection (MOBO-CAPS), where the goal is to identify a set of exactly M solutions with high hypervolume. We propose two practical cardinality-aware acquisition strategies: REHVI, a replacement-aware approximation to expected hypervolume improvement, and CHO-UCB, which constructs an optimistic M-point set by optimizing a collective UCB-based hypervolume objective. We prove a sublinear simple regret bound for the batched version of CHO-UCB, and empirically show that cardinality-aware acquisition design improves over standard MOBO followed by p
benchmark - arxiv:2611.19998 · cs.ROMoViC: Real-Time Humanoid Whole-Body Control from Monocular Video StreamsYimin Han, Jiahui Zhang, Wanyue Li, Zeren Luo +9
Real-time whole-body imitation from monocular video requires both a robust universal tracker and high-quality real-time reconstruction. We present MoViC, a framework that uses live RGB video as a control signal for humanoid robots without wearable tracking devices. The framework combines monocular human motion reconstruction, robot retargeting, and a transformer-based behavior foundation model. The policy receives proprioceptive history, a single explicit next-step reference, and a compact frequency-domain motion representation that supplies temporal context without explicitly stacking future reference frames, which reduces the input dimension and network size. During deployment, our extrapolation method bridges the 10-Hz visual estimation with 50-Hz controller, enabling asynchronous visual processing and policy execution. Hardware experiments further demonstrate highly dynamic motion tracking and device-free teleoperation of bimanual and single-arm manipulation tasks from live monocul
manipulationhumanoidteleoperationwhole-body control - arxiv:2611.19982 · cs.AIAutonomous Multi-modal Agents for ITOps ObservabilityGunjan Siddharth, Sanjana Garai, Utsav Shekhar, Prakhar Jain +14
ITOps observability products must integrate data from different tools and systems. We present ITOps Observability Studio, a multi-modal multi-agent system that orchestrates observability workflows through collaborative AI agents. Its three-tier architecture features inter-agent communication, specialized LLM-powered agents for data access, schema operations and visualization, and unified tool integration. The system supports natural language to GraphQL translation, GraphQL schema generation, schema mapping, trace data lifecycle management and multi-tool workflows for reliability engineering. Its agents are instrumented with agent analytics, so that their model and tool calls can be traced. We evaluate each agent on its own benchmark, scored against gold answers where they exist and by a model judge for schema mapping and chart quality. Query generation on a public benchmark and routing work well with a current open model, and the data lifecycle management agent solves most of its scena
agentai agentmulti-agentagent systembenchmark - arxiv:2611.19979 · cs.ROPractice Is Not Free: Measuring the Embodied Cost of Reset-Free Robot PracticeVedant Telrandhe, Sushil Sharma
When a robot practices without human resets, every attempt still costs wall-clock time, actuator wear, and sometimes a person's attention, yet the schedulers that decide what to practice next treat attempts as free. We present an open, low-cost testbed for measuring these costs: a self-built SO-101 arm runs two demonstration-trained ACT policies that reset each other, with a vision-based success detector and a ledger that records the time, joint travel, and human interventions of every attempt. We formulate cost-aware practice scheduling, in which task selection is rewarded by measured competence gain net of embodied cost, and specify in advance the comparison that will test it. Its first two conditions, run on hardware for 51 minutes of robot time, show that a fixed retry limit did not reduce the share of execution time lost to failures (36%, against 27% for a feasibility rule; not significant) and saved no robot time. The logs suggest why: failures showed no sign of clustering, so a
embodied - arxiv:2611.19975 · cs.LGFederated Dataset Simulation: Inducing Label-Free Heterogeneity Across TasksVasilis Siomos, Lam Ngo, Jonathan Passerat-Palmbach, Giacomo Tarroni
Federated Learning enables collaborative model training across decentralized clients without sharing raw data. While the field has seen rapid growth, the datasets underpinning FL evaluation have not kept pace: real federated datasets remain scarce, and most studies rely on small-scale centralized classification datasets partitioned via Dirichlet label skew, a one-dimensional proxy for the multi-dimensional heterogeneity present in real federations. Moreover, label-based partitioning is by construction inapplicable to non-classification tasks such as segmentation and detection, for which no principled simulation method exists. We cast the conversion of centralized datasets into federated benchmarks as the task of Federated Dataset Simulation (FDS), and decompose it into a representation function $Φ$ and an assignment mechanism $π$. We provide two label-free instantiations -- geometric partitioning via pretrained model embeddings and semantic partitioning via vision-language model-derive
benchmark - arxiv:2611.19970 · eess.SYPhysics-Informed Fractional Load Forecasting With Memory-Aware Break Detection for Reserve SchedulingAli Nikseresht
Day-ahead load forecasts set balancing reserves, including the energy a pumped-storage plant holds against forecast errors. Both depend on three effects that are usually modeled separately, namely the thermal inertia of buildings, which makes load lag temperature, structural breaks, which shift demand, and the long memory of forecast errors, which keeps them from cancelling out. We prove that averaging building responses over Cole-Cole distributed time constants yields a fractional temperature kernel with nonnegative weights exactly when its order is at most one. A Bayesian information criterion that ignores long memory selects spurious breaks with probability tending to one, so breaks and memory are estimated jointly, and an online break detector keeps its false-alarm rate below any prescribed level. Long memory also makes the required storage energy grow faster than the square root of the horizon, so a second-order cone program schedules reserves from the joint distribution of hourly
memorybenchmark - arxiv:2611.19969 · cs.ROValue and Policy Calibration from Execution Feedback for Socially Compatible DrivingSen Shen, Ning Liu, Xiangrui Kong, Yingwei Li +1
Autonomous vehicle (AV) decisions at intersections affect the delay, braking, and motion smoothness of surrounding human-driven vehicles (HDVs). Learning socially compatible behavior requires connecting imagined interaction outcomes with feedback from executed actions. We propose a world-model reinforcement learning framework combines imagined policy improvement with direct value and policy calibration from finalized interactions, balancing AV progress with HDV interaction burden. An interaction-aware latent model supports imagination, while off-policy correction and replay updates enable observed rewards and transitions to supervise both the critic and actor. Simulation experiments at an unsignalized intersection achieve a 100% success and reductions in following-HDV acceleration variation of 65.5% and 44.0% relative to PIWM and DreamerV3, respectively. Ablations identify complementary roles for value calibration in reducing HDV burden and policy calibration in improving passage effic
dreamerv3 - arxiv:2611.19965 · cs.LGEvaluating Near-Perfect Accuracy Claims in Machine-Learning-Based Leak Detection for Two-Phase Pipelines: A Condition-Blocked AuditIssah N. Suleiman, Sormeh Serpoosh, Nadine Elkholy, Hicham Ferroudji +2
Reliable leak detection in gas-liquid two-phase pipelines is important for process safety, yet machine-learning studies frequently report near-perfect performance using evaluation protocols that may not represent deployment to unseen operating conditions. This study audits a pressure-based leak-detection model using condition-blocked leave-one-condition-out validation, in which the classifier is tested on a flow condition excluded from training. Accuracy decreased from 93.3% under within-condition evaluation to 59.1% across fifteen unseen-condition tests. Ablation analysis showed that most of this difference arose from the data-splitting protocol, with a smaller contribution from encoding operating conditions as model inputs. Under condition-blocked evaluation, the false-alarm and missed-detection rates increased to 34.1% and 10.2%, respectively, demonstrating that near-perfect reported accuracy did not translate into reliable operation at unseen flow conditions. The reduction persiste
evaluation protocol - arxiv:2611.19960 · cs.CLAnatomy of an Agent Harness: Decisions, Invariants and Corner CasesResham Joshi
An agent harness is the program around a language model that turns its replies into actions: it builds each request, runs the tools the model asks for, decides what the model may do without asking, stores the conversation, recovers from failures, and shows a person what happened. Most of these decisions are made in code and rarely written down, so each harness meets the same corner cases again. This paper writes them down as invariants to test. It gives a system model with trust boundaries; 30 decision cards in six groups, each with options and costs, an invariant, the corner cases that break it, a test and an evidence tag; six cause classes for failures; a test method built on scripted providers, recorded behaviour and planted changes; and a checklist. The cards come from building one desktop harness, Crewroom, the author's own, and from manually testing ten other harnesses and studying open-source harnesses as reference; some patterns in the cards were adopted from those, so the cata
agent - arxiv:2611.19955 · cs.AIEqual Dose, Different Coverage:Tracing Rate and Duration in Image EditingYi Liu
Benchmarks for scientific image editing test whether a model draws the outcome of a described process, leaving open which representations carry its errors and whether correcting them yields the right quantity. We study this question with an airbrush of fixed deposition profile, under which exchanging rate and duration preserves the dose and should preserve the endpoint, yet changes the edited coverage. Matched-norm activation patches across text positions and denoising stages show that, in distilled FLUX with Chinese prompts, representations at unchanged text positions (context) carry this discrepancy: early context transfers 3.51 of the 3.98 percentage-point (pp) signed route difference, against 0.09 pp for equal-norm updates at the two changed digits. Early image latents transmit native (unpatched) route differences across two FLUX configurations and OmniGen, while text-position effects depend on configuration. The trace predicts an intervention: share early context between routes wh
benchmark - arxiv:2611.19952 · cs.ROCRAFT: A Robot-Agent Harness with Scope-Aware AdaptationPeng Zhang, Jiawen Wang, Cheng Xue, Jochen Renz
A robot may know how to move an object yet still need to learn where it belongs. When a person rejects a placement, the robot must decide whether the correction applies only to that situation or expresses a broader rule. We study this problem in CRAFT (Contextual Requirement Adaptation with Frozen Tools), a framework for learning local requirements without retraining robot skills. It records physical success separately from whether an action meets the requirement. Our scope ladder uses memory when an action has already been judged in a nearly identical situation. Elsewhere, it combines memory, explicit rules and a learner for requirements beyond those rules, weighting their predictions by how well they predicted earlier feedback. Controlled studies show gains over correction memory alone in utility, which rewards acceptable actions and charges for mistakes and questions. In camera-grounded robot simulations, the ladder asks fewer questions than language-model agents with persistent mem
memorypersistent memory - arxiv:2611.19951 · cs.LGDiscovering Programmatic Policies from Reinforcement Learning-Based Traffic Signal ControllersLindong Xie, Yang Zhang, Beiyu Song, Zeren Xing +2
Traffic signal control (TSC) is critical for mitigating urban congestion. Deep reinforcement learning (DRL) has achieved strong performance by learning adaptive policies from traffic observations, but learned neural controllers are often opaque, structurally complex, and difficult to generalize across scenarios. We propose PPL-TSC, a programmatic policy learning framework that distills compact and human-readable TSC policies from DRL teachers. PPL-TSC formulates policy learning as an evolutionary search over phase utility functions (PUFs), which score candidate signal phases and select the highest-utility phase for execution. Each generation consists of four steps: policy generation, policy review, policy calibration, and policy evaluation. In the generation step, a large language model (LLM) acts as a semantics-aware evolutionary operator to propose and refine candidate PUFs. Generated PUFs are screened through explicit checks for syntactic validity, mathematical soundness, and consis
policy evaluation - arxiv:2611.19939 · cs.CVTwo-Stage Prompt Learning: A Unified Framework for Person ReID without Clothing HypothesisYuxuan Xiao, Shanshan Zhang, Shengcai Liao, Jian Yang
Person re-identification (ReID) in real-world scenarios involves three clothes conditions: the same clothes (SC), changed clothes (CC), and the unknown (GEN) scenario where clothes changes are not specified. Existing methods typically train separate SC- or CC-specific models, which limits generalization across all conditions. Meanwhile, pre-trained visual-language models (VLMs) such as CLIP have shown strong generalization ability in various vision tasks, motivating recent ReID methods to incorporate textual prompts. However, to distinguish SC and CC, prior approaches rely on fixed predefined instructions, which lack learnable context embeddings in prompts and cannot effectively handle the GEN scenario. To overcome these limitations, we propose Two-Stage Prompt Learning (TSPL), a unified framework that integrates clothes-identity decoupling and scenario-aware task prompts within VLMs. In Stage 1, appearance-decomposed prompt learning disentangles clothes and identity information under
benchmark - arxiv:2611.19933 · cs.CVSAGE-VQA: A Training-Free, Self-Evolving Agentic Framework with Grounded Evidence for Knowledge-Based Visual Question AnsweringRuirui Shen, Jianqi Gao, Xiangfeng Luo, Yonggang Zhang +4
Knowledge-based visual question answering (KB-VQA) aims to combine visual information with external knowledge to answer user questions. Existing methods face two main bottlenecks: visual perception errors can propagate into hallucinations, and generalization across scenarios remains limited. To address these issues, we propose SAGE-VQA (Self-evolving Agent framework with Grounded Evidence), a training-free agentic framework that combines evidence verification with experience evolution. Our key insight is to explore an Evidence-Driven Adaptive Reasoning (EDAR) strategy. It accurately acquires and verifies question-relevant fine-grained visual information. It also extracts and refines reusable experience to support robust reasoning across tasks. Specifically, we design a Co-verified Visual Grounding module that verifies question-relevant visual information and provides grounded evidence for subsequent reasoning. We further introduce a Self-evolving Hierarchical Memory (SHM) mechanism. SH
memoryagentagenticagent frameworkself-evolving - arxiv:2611.19932 · cs.ROFrom Companion to Guardian: Designing an Emergency Support System Using a Quadruped Robot for Older AdultsChunyang Meng, Eduardo B. Sandoval, Josh Andres, Francisco Cruz +1
As older adults increasingly live alone, emergency support becomes an important function for home companion robots. In Australia, demand for the care and support workforce is projected to nearly double between 2020-21 and 2049-50. Service robots such as robot dogs may support older adults by combining everyday companionship with assistance when human support is not immediately available. We present the design and implementation of a domestic companion-robot system on a Unitree Go1 quadruped platform that combines companionship with emergency awareness and response. The system uses a three-stage interaction model consisting of companion, danger-awareness, and emergency modes. A rule-based escalation policy combines signs detected from visual and speech input to determine the current risk level and adjust how long the robot waits for the user to confirm safety before entering emergency mode. Additional detected signs can shorten this waiting time. Users can confirm safety, request immedi
quadruped - arxiv:2611.19931 · cs.LGDoes synthetic pre-training pay off on a microcontroller? Accuracy, latency and energy across two MCUsAchintya Yadav
Rendered images are cheap to label, but a microcontroller classifier must survive 8-bit integer (INT8) quantisation and compute correctly on the chip. Five models, from MobileNetV2 at 224 px to a 61K-parameter convolutional neural network (CNN), were fine-tuned on a six-class scene dataset with and without a synthetic stage, against thresholds fixed in advance, and deployed on an ESP32-S3 and an RP2040. With full real data the stage gave no detectable gain. For the main deployable model the INT8 difference was +0.312 percentage points (pp), its interval below the +1.6 pp gain recomputed from the source study; quantisation did not detectably change it, nor did model size. With 10% of the real data the stage paid off for the CNN, by +3.438 pp on one seed and +2.715 pp over three; its cause is open. On the ESP32-S3 the main model took 152.7 ms and 16.3 mJ above idle per inference, and adding the stage changed neither beyond preset bounds. The model's INT8 outputs matched TensorFlow Lite's
memory - arxiv:2611.19922 · cs.ROLao Siji: Adapting openpilot to Closed-Loop End-to-End Driving in Neural ReconstructionChengzhi Gao, Yongzhi Liu, Jinchang Xu, Weichao Zhuang +1
We present Lao Siji, an end-to-end driving policy that ranks third in the 2026 NVIDIA AlpaSim E2E Driving Challenge, ahead of most entries built on foundation models. Instead of training from scratch on a general-purpose vision backbone, we adapt openpilot, an open-source L2 production model. We freeze its vision encoder and fine-tune the policy layers with an ego-state adapter and a route memory. On closed-loop rollouts in neural-reconstructed scenes, we supervise the policy with a privileged map teacher for lateral self-correction and a body-extent loss for road and obstacle clearance, while distilling the base model's remaining outputs to retain car following. Ablations show that self-correction samples carry the gain, while the clearance loss substitutes for them where no teacher exists. Longitudinal safety instead rests on the base model's lead-vehicle estimate behind a rule-based speed layer. Longer training, imitating human logs, and human-log speed targets yield either no measu
self-correction - arxiv:2611.19912 · cs.ROAgenticJevNav: Agentic Object Search and Navigation via Event-Driven VerificationShuning Zhang, Shuli Zhang, Yunheng Wang, Wei Lu +7
Agentic navigation must maintain goals and adapt actions from incomplete evidence despite execution failures and delayed inference. We present AgenticJevNav, which links goal-directed planning, evidence memory, event-triggered verification, and bounded execution. The Agent plans from RGB, odometry, memory, and feedback, retaining goal ownership while authorizing a bounded action set. Jev reviews structured evidence after blocked/failed execution or two consecutive low-progress segments. Version, freshness, and once-only checks govern decision application. A single executor and independent RGB reflex constrain motion. Goal revisions revoke obsolete requests and actions before replanning. Across six simulated task families, AgenticJevNav achieves 73.33% SR over 60 trials, 21.67 percentage points above the next-best Agent-only aggregate control under differing historical protocols. A real-world GENISOM L1 quadruped deployment achieves 70.0% success over 30 trials, including bounded search
quadrupedagentagentic - arxiv:2611.19911 · cs.LG"Don't Leave the Task Unfinished": Persistence-Driven Violations under Quiet OversightSeunghyun Lee, Jaehui Hwang, Dongyoon Han, Sangdoo Yun
LLM agents are increasingly given long, multi-step tasks to carry out, with the user reviewing little of what happens in between. Persistence, the tendency to keep working until a task is complete, is installed on purpose to improve performance on long tasks. Long-horizon tasks are typically complex, and the user's instruction can fall short of what the work actually requires, or conflict with the constraints the agent encounters in the task. We study what persistence does on long-horizon operations tasks that cannot be solved without breaking a documented constraint. We find that the agent pursues completion to the point of violating the constraint, and that a critically engaged user lowers the violation rate. Persistence takes violations from 29.2% to 58.3% across four models. Violations stay high wherever oversight goes quiet, whether the user is silent, assenting, or delegating with expressed trust. Only a questioning user, who asks for the basis of each action before allowing the
agentllm agent - arxiv:2611.19905 · cs.CVHEmIF-Bench: Benchmarking H&E-to-mIF Models BeyondVirtual-StainFidelityFuxian Zhao, Hyokeun Park
Hematoxylin and eosin (H&E)-to-multiplex immunofluorescence (mIF) generators predict spatial protein expression from routine histology, yet pixel-level marker reconstruction does not establish downstream utility. We present HEmIF-Bench, evaluating five frozen generators across eight datasets spanning molecular, cellular, tissue, and patient scales. HEmIF-Bench decouples internal encoder features from generated outputs, benchmarking both against the pathology foundation reference UNI2. Mirroring the clinical workflow where pathologists rely on mIF to define cell phenotypes, delineate tertiary lymphoid structures (TLSs) and germinal centers (GCs), and stratify patient risk, we evaluate external CD3 recovery, single-cell classification, dense nuclear segmentation, TLS/GC segmentation, and patient survival prediction. Across the benchmark, foundation-model-based MIPHEI-ViT leads among the generators, while non-foundation models show task-specific strengths. Comparing the two routes reveals
benchmark - arxiv:2611.19904 · cs.RODiffWAM: Learning World Action Model from Consecutive Visual DifferenceXinheng Zhang, Haoyu Xiong, Chao Jing, Xiaohan Zhang +3
World Action Models (WAMs) have emerged as an important paradigm in robot learning, jointly predicting future scene states and robot actions from current observations and instructions. However, future states entangle task-related changes with large amounts of task-irrelevant content, such as static backgrounds, which weakens the model's ability to capture changes relevant to task execution. To address this, we propose Difference World Action Model (DiffWAM), a Mixture-of-Experts (MoE) framework that enhances world representations by jointly learning future-state prediction and consecutive visual difference prediction. We define consecutive visual differences as differences between the latent representations of consecutive observations, providing an explicit learning signal for temporal variation. The framework comprises a World Expert for future frame prediction, an Action Expert for action generation, and a Diff Expert for predicting these latent differences along the trajectory. A st
manipulationliberorobotwinbenchmark - arxiv:2611.19901 · cs.CVRemember, Replay, Adapt: Learning from Past Experience in Online Vision-and-Language NavigationChenhao Ding, Jizhou Han, Lixin Xiu, Muyao Wang +2
Existing feedback-based methods for online vision-and-language navigation (VLN) primarily learn from current episode outcomes, leaving previously acquired feedback underutilized. Inspired by human episodic memory, we propose R2A, an online VLN adaptation framework that continually reuses successful and failed navigation experiences for policy learning. Specifically, Success-Failure Memory (SFM) retains successful and failed navigation episodes with acquired feedback and retrieves task-relevant experiences based on incoming instructions, enabling historical feedback to guide subsequent navigation. Prior-Anchored Adaptation (PAA) learns from these experiences by reinforcing successful trajectories and suppressing failed ones relative to a frozen pretrained policy, while regularizing behavioral drift. Together, these modules enable continual feedback reuse and stable policy adaptation without source-domain data or expert supervision. Extensive experiments across three benchmarks, REVERIE,
memoryepisodic memorybenchmark - arxiv:2611.19900 · cs.CLAttribute or Abstain: Evaluating Evidence-Grounded Retrieval on the Low-Resource African Business WebBakary Gibba, Muhammad Abba Mandara, Karim Ismail Abd-Ellah Mohamed
Retrieval-augmented systems that research companies are judged by how much they find, yet in markets where public evidence is sparse, a confident answer without support costs more than no answer. We introduce AfriBizClaims, a benchmark of 150 company-research queries across six African markets in English, French and Portuguese, including queries about fictitious companies that a correct system must decline. Each claim a system makes is judged against the page it cites by an AI judge, checked against a second AI judge from another model family and against two human annotators, and abstentions are scored against answerability labels. We compare BM25 over a pooled corpus, standard web retrieval-augmented generation, and a multi-step agent pipeline (Konexium V1), all using the same language model. The two simple baselines are well attributed: 98% of their judged claims are supported by the page they cite, and they decline every field of every fictitious company. Counting claims about a com
retrieval-augmentedagentbenchmark - arxiv:2611.19889 · cs.LGIllusion of Explainability: A Sanity Check for Spectral Band Attribution in Satellite Time-Series Crop ClassificationSumit Tahiliani, Avi Mathur, Sandarbh Agarwal, Rakshith Sathish +1
Attribution methods are commonly validated through convergence across methods and consistency with domain knowledge, yet neither criterion establishes whether identified features are necessary for a task. This limitation is particularly relevant in multispectral remote sensing, where overlapping spectral information makes individual bands potentially substitutable. We investigate this gap in crop-type classification using GradientSHAP, Shapley sampling, and Occlusion on a U-TAE segmentation model trained on Sentinel-2 time series from the PASTIS benchmark. All three methods identify vegetation-sensitive spectral regions consistent with domain knowledge, while Shapley and Occlusion exhibit strong ordering sensitivity in deletion and insertion tests. However, these properties do not reliably predict retraining performance. Aggregate attribution consensus masks substantial per-class disagreement, with pairwise mean Spearman correlations ranging from 0.09 to 0.79. Models retrained from scr
benchmark - arxiv:2611.19886 · cs.AIRecovery Gone Too Far: Understanding and Repairing Risky Fallback in LLM AgentsLiwen Wang, Jiangrong Wu, Zongjie Li, Zongyi Lyu +2
Large Language Model (LLM) agents interact with their environments through tools, reading repositories, running builds, and pushing code. When a tool fails with an ordinary error, such as a rate limit or a timeout, the agent can fall back to a substitute tool and still complete the task. However, the same fallback that completes a task in one environment can cross an action boundary in another. For example, after a push is rejected, a forced push in one repository replaces only the agent's own commits, while in a different repository it also removes another author's commits. We call such a fallback a risky fallback. To understand this phenomenon, we conduct an empirical study on 304 scenarios that span eight domains and cover 55 real CVEs. All four mainstream models we evaluate exhibit risky fallback, with action-boundary violations in up to 96.8% of risky scenarios. We further evaluate three mainstream mitigations and find that none of them is effective against risky fallback. We then
agentllm agent - arxiv:2611.19875 · cs.AILet Execution Do the TeachingAbhishek Mishra, Sandeep Srinivasa, Arko Chattopadhyay, Enrique Martinho Ferrao +1
Improving a coding model is a sequence of decisions: find the limiting behavior, create useful practice, and determine which attempts deserve reinforcement. We investigate a practical way to make those decisions increasingly repeatable by grounding feedback in execution. Starting from GLM-4.7-Flash, the work combines whole-file supervised fine-tuning, targeted C++ repair curricula, group-relative reinforcement learning, and verifiers that turn program behavior into structured rewards. Compiler errors and failing tests provide specific contexts for repair practice, while executable checks score the resulting programs. We evaluate the trained models on 26 C++ coding tasks, measuring both initial success and completion after test feedback. The strongest reported post-training configuration achieves 45.2% success in the first test round and 61.5% within two rounds. These results come from an assisted development suite with evolving prompt contracts; the strongest configuration uses four se
post-training - arxiv:2611.19872 · cs.AIWho Writes the Codebook? An Eight-Step Framework for Auditing LLM-Generated Category SystemsThilo Tamme, Anton Hantel, Jakob Scheppers
Large language models (LLMs) increasingly create the categories through which qualitative evidence is interpreted. Existing coding benchmarks often hold these categories fixed. We propose the Qualitative Variance Audit Framework, an eight-step method that compares codebook creation, repeated application and code distributions under controlled inputs. It combines model-owned and shared-reference codebooks to distinguish variation in the overall pipeline from application differences within a common category system. An application to 668 organisational interviews, 13,166 text units and eight models from five vendors yields 48 codebooks and 270 coding runs. Models create markedly different category structures; mean within-model reliability ranges from .76 to .92, whereas between-model agreement on the shared codebook is about .69. The framework complements LLM-assisted grounded theory by exposing model dependencies before substantive interpretation. It assesses reliability and representati
benchmark - arxiv:2611.19871 · cs.LGInteractive Structured Environment Scaling for Large Language Model Agentic ReasoningPeng Yu, Luoyi Fu, Xinbing Wang, Ying Wen
Reinforcement learning via environmental interaction is a cornerstone post-training paradigm for evolving large language models into autonomous agents capable of multi-turn tool use and exploration. Because the environment defines what the policy observes and how actions are verified, its structure directly bounds the capabilities an agent can acquire. Yet existing structured and retrieval environments rely on upstream pre-filtering---extracting a compact subgraph around the gold answer or retrieving fixed passages before the policy acts. We show that such pre-filtering introduces a fundamental measurement confound: because per-question subgraphs enclose ground-truth paths, they compress the candidate space by orders of magnitude and leak answer locations via local topology, allowing a question-blind degree heuristic to surface gold answers far above chance. To restore path selection, candidate pruning, and optimal stopping as genuine policy decisions, we propose RISE (Interactive Stru
knowledge graphagentautonomous agentagentictool useself-evolving - arxiv:2611.19865 · cs.LGMax-Min Throughput in RSMA-Aided SWIPT for Sustainable Internet of Underground Things: A Deep Reinforcement Learning ApproachKaiqiang Lin, Mohamed-Slim Alouini
Unmanned aerial vehicle (UAV)-enabled wireless-powered underground sensor networks (WPUSNs) allow underground devices (UDs) to harvest radio-frequency energy from a UAV via wireless energy transfer (WET) techniques, enabling sustainable Internet of underground things (IoUT) applications. This paper integrates rate-splitting multiple access (RSMA)-aided simultaneous wireless information and power transfer (SWIPT) into UAV-enabled WPUSNs to deliver information and energy to UDs. Within this framework, we investigate a max-min fairness problem that jointly optimizes the UAV transmit power, power allocation and message splitting, power-splitting ratios, and time allocation, aiming to maximize the minimum uplink throughput across UDs subject to per-UD downlink throughput requirements and a UAV energy-consumption constraint. To address this non-convex problem under dynamic underground environments, we develop an off-policy deep reinforcement learning approach based on the primal-dual truncat
benchmark - arxiv:2611.19862 · cs.CLBioformer-RetroMAE: retrieval-oriented pretraining of a compact biomedical language model for literature searchJialu Li, Li Fang
Dense retrieval is increasingly used for biomedical literature search and retrieval-augmented generation (RAG). Most biomedical bi-encoders are built on BERTBase-sized encoders pretrained with masked language modeling (MLM). These encoders are costly for corpus-scale indexing and latency-sensitive query encoding, and MLM does not explicitly train the [CLS] vector that a bi-encoder uses to represent a whole text. We present Bioformer-RetroMAE, a compact (43M-parameter) retrieval-oriented biomedical encoder obtained by continually pretraining Bioformer8L} with RetroMAE on approximately 32 million PubMed titles and abstracts. After bi-encoder fine-tuning with BM25 hard negatives, we evaluated it on BioASQ Task 12B test questions by retrieving from all PubMed abstracts. Bioformer-RetroMAE outperformed BioBERT and PubMedBERT bi-encoders fine-tuned under identical settings on all metrics, despite having about 40% of their parameters and producing 512-dimensional rather than 768-dimensional v
retrieval-augmented - arxiv:2611.19859 · cs.AIWhen Facts Change, Which Decisions Break? Dependency-Aware Context Graphs for AgentsPaul Deepakraj Retinraj
When an observed fact changes, an agent must identify and repair the decisions that depend on it while preserving unaffected outputs. We investigate the reliability-cost trade-off of field-level dependency graphs extracted from actual agent read traces. A controlled benchmark spans three executable workflow domains and four successive updates, with independent imperative and expression-based references. Across 120 workflows, two models and two repeated runs, we compare full recomputation, rolling summary memory, lexical retrieval, trace-extracted dependencies and verified dependency upper bounds. Trace-derived graphs reduce reviewed decisions and estimated total cost, but omitted alternative supports and missing repair context can reduce complete-plan correctness. Paired ablations isolate the effects of edge deletion, spurious edges, transitive propagation and context scope. An exploratory enhancement supplies the existing record ID/field directory during extraction without revealing u
agentbenchmark - arxiv:2611.19857 · cs.CVPanDWM: Panoramic 4D Driving World ModelsJiajun Sun, Zeyu Zhang, Zeyu Ren, Guangfeng Jiang +2
Driving world models usually generate videos from six pinhole cameras with separate optical centres and small overlaps, and must learn to keep these views consistent. We present PanDWM, which stitches them into one $360^\circ$ panoramic band. Camera boundaries then fall inside one image, so keeping the views consistent means keeping the image continuous, and each pixel still records its camera ray. Conditioned on layouts, a causal student generates the band window by window in four steps; it is distilled from a causal teacher and trained on its own rollouts. A feed-forward decoder lifts each frame along these rays into 3D Gaussians, and static Gaussians from neighbouring frames complete them into 4D scenes. On nuScenes, scored in one stitched $360^\circ$ view, PanDWM has lower FID and FVD than the multi-camera baselines at $8$-$46\times$ their speed, and keeps its 50-step teacher's FID while lowering FVD by 22-25%.
world model - arxiv:2611.19841 · cs.AIAsking the right question is key to AI discovery in a chain of hypothesesHu Cang
Artificial intelligence (AI) scientists pursue discovery through a chain of hypotheses, fitting candidate models to existing observations and often choosing experiments where those models disagree most. Yet when early data miss a narrow feature, model selection favors simple curves that agree across the unmea- sured gap, so disagreement-guided search has no reason to look where the new phenomenon lies. When we tested two frontier AI scientists, Claude Sonnet 5.5 and GPT-6.1-Sol, across 32 closed-loop trials with a narrow peak omitted from the starting data, both sampled inside the peak region less often than random search. Surprisingly, the agent that followed model disagreement more closely missed the peak in all 32 trials, whereas discovery succeeded when experiments probed the unmeasured gap early and followed up across the anomaly. Asking the right question, rather than testing only where current models disagree, is therefore key to AI discovery in a chain of hypotheses.
agent - arxiv:2611.19840 · cs.LGGrounding Language Models in Speech Does Not Fix "AI Flavor": Calibrated LLM Judges, Goodhart Traps, and a Preference-Tuning FixDongwu Qian
Chinese outputs of large language models carry a perceptible "AI flavor": formulaic sentence frames, stacked written-register connectives, repetition loops, and layout leakage. A popular hypothesis blames text-only training for the missing inner-speech layer of human writing; injecting speech or prosody signals should therefore reduce it. We test it with controlled interventions on a 0.6B model; all three fail: (i) phoneme-interleaved continued pretraining yields no spokenness gain under text-matched controls while raising repetition (+0.13 [+0.03, +0.23]); (ii) prosody-plan conditional generation reaches an adherence of only 0.257 against a pre-registered 0.50 gate; (iii) a rule-based reward built on TTS prosody-feature distance is imperceptible in double-blind evaluation by both humans (8/16) and a calibrated LLM judge (9/16). A root cause is evaluation: single-rater blind evaluation at n~20 only detects extreme (>=85%) preferences. We therefore build a calibrated LLM-judge protocol:
evaluation framework - arxiv:2611.19836 · cs.CVAGPF: Dynamic Source Selection for Feature Pyramid FusionXu Yang, Gady Agam
Feature pyramids provide dense predictors with features at several spatial resolutions. However, it remains largely unexplored which pyramid levels each output location should aggregate information from. We investigate this question by allowing each output location to attend to features from every pyramid level and measuring how attention is distributed across scales. Surprisingly, locations on the finest output grid allocate approximately 75\% of their attention to features two levels coarser, rather than to the current or adjacent levels. Even on the same output grid, the balance shifts toward coarser features as object size increases. These observations motivate Asymmetric Global Pyramid Fusion (AGPF), a neck that preserves the original output grids while allowing each location to dynamically select its contextual sources across scales. Each location interacts with a multiscale memory whose features are aligned in image coordinates, enabling flexible cross-scale aggregation without
memory - arxiv:2611.19832 · cs.LGProTTS: End-to-End Prominence-aware Natural and Controllable Text-to-Speech via Automatic Word Stress DiscoverySai Harshitha Aluru, Chiranjeevi Yarra
Natural speech conveys meaning not only through words, but also through the prominence on a few words. Yet most text-to-speech (TTS) systems remain prominence-unaware, existing prominence-aware TTS require external stress annotations, while unsupervised stress detectors operate in isolation, disconnected from synthesis. This creates a fundamental gap no existing system can automatically discover where stress should be placed, realize it as natural-sounding prosody, and offer user control over emphasis placement, all within a single model. To circumvent this, we propose ProTTS, a unified framework where stress detection emerges from the TTS model itself and, in turn, guides it to synthesize more natural and controllable speech. We incorporate a clustering head into the variance adaptor of FastSpeech_2 that discovers word stress directly from aggregated prosodic features. The discovered stress then drives emphasis variance predictors that modify pitch, energy, and duration for stressed w
benchmark - arxiv:2611.19828 · cs.CVPRISM: Primitive-Integrated Separable Modulation with Priority Visual Memory Queue for Compositional Zero-Shot LearningJianhua Wang, Zhihang Zheng, Yang Liu, Jiale Du
Compositional Zero-Shot Learning (CZSL) recognizes unseen attribute--object pairs by recombining primitives from seen compositions. Existing methods stop sharing evidence at deployment: fixed prototypes cannot exploit unlabeled test streams, while composition-wise refinement cannot propagate corrections across pairs sharing a primitive. We introduce PRISM, an online CZSL framework that extends primitive sharing to test-time adaptation. Its Hierarchical Knowledge Accumulation Module (HiKAM) factorizes prototype migration into composition-, attribute-, and object-level residual banks. A Priority Visual Memory Queue (PVMQ), initialized by Prototype Queue Initialization (PQI), grounds text updates with bounded visual evidence. The same adaptive core supports exhaustive Full PRISM and recall-calibrated Adaptive PRISM-Lite, which selects candidate budgets from held-out ranks. Comprehensive closed- and open-world experiments on UT-Zappos, MIT-States, and C-GQA demonstrate the state of the art
memorybenchmark - arxiv:2611.19827 · cs.CVWhen Does Reusing Layers Help Multimodal Embeddings?Zhizhang Hu, Yue Zhang
We ask whether a compact universal multimodal embedding models can produce better search embeddings by running the same layers again. Looped transformers add computation with nearly unchanged model size, making them an appealing alternative to a larger encoder. Their success in language-model reasoning and recent use for retrieval motivate the idea, but higher retrieval scores alone do not reveal what an extra pass contributes. We study eleven loop designs and training variants on three checkpoints of one vision-language architecture. Our protocol distinguishes whether a model uses its extra pass, whether its image states respond to the query, and whether those updates improve retrieval. We prove that an image placed before the query remains query-independent through repeated layers when information only travels forward. We then open a query-to-image path in three ways, including order-alternating recurrence, which places the query first on the next pass. Each path produces query-sensi
benchmark - arxiv:2611.19825 · cs.LGFoundation Pareto Flow Policy for Multi-Objective Reinforcement LearningZhanjiang Yang, Lijun Sun, Yueming Li, Meng Li +1
The overarching goal of reinforcement learning (RL) is to efficiently optimize the agent's policy for various environments, tasks, dynamics, and even user preferences, which often consist of multiple and potentially conflicting rewards/objectives. This necessitates the learning of Pareto-optimal policies that can navigate different trade-offs among objectives. However, existing multi-objective RL (MORL) methods, whether trained online or offline, are typically limited to task- or objective-specific policy learning and do not address generalization of learned trade-offs to unseen tasks/objectives. In this work, we propose the Foundation Pareto Flow Policy (FP2) for offline MORL in continuous robotic control, inspired by the pre-mid-post-training paradigm underlying recent foundation models. Specifically, we instantiate FP2 as a multi-modal diffusion transformer equipped with a likelihood-free flow matching loss, enabling unified, scalable, and instructable learning of diverse trade-off
post-trainingbenchmark - arxiv:2611.19823 · cs.LGEfficientSkill: Learning Skills for Token-Efficient Multi-Turn AgentsDeep Shah, Sanket Badhe, Nehal Kathrotia, Priyanka Tiwari
Reflective skill optimizers improve multi-turn LLM agents by revising instructions from failed and successful rollouts, yet accumulating edge-case rules bloats prompts, inflates token use, and pushes agents into long, error-prone loops. We introduce EfficientSkill, a two-step framework that first discovers a reliable skill without token penalties, and then warm-starts accuracy-guarded pruning from that skill to compress instructions, bound agent turns, and offload reusable code into workspace utilities with zero implementation prompt cost. Across four frontier, mid-tier, and open-weight models on MATH-500, ALFWorld, and SpreadsheetBench, EfficientSkill cuts output tokens over unconstrained EvoSkill on 11 of 12 configurations (by up to 80%) while raising mean accuracy on every benchmark, including +14.3 and +17.5 percentage-point gains on SpreadsheetBench with Gemini 3.8 Flash and Grok 4.1 Fast by replacing stalled loops with compact workflows. By shrinking both generated tokens and rep
agentllm agentbenchmark - arxiv:2611.19822 · cs.ROThe Flip Test: Video World Models Fail to Learn the World StateMi Luo, Lijun Yu, Adams Wei Yu, Ang Cao +9
Video world models promise to predict the future of a scene from its present. Do current models keep this promise? We probe this question counterfactually. When a small change to the present flips the future, such as a box resting a few millimeters before or past a table edge, a model that tracks the state of the world must flip its prediction too. We build FlipTest, a three-level hierarchy of increasingly strict tests over 1,680 initial frames from 10 simulated and 15 real-world tasks. At the base, visual minimal pairs, two nearly identical frames whose true futures differ, test whether predictions change across the boundary; matched controls go further and test whether they change only where the future does; and dense sweeps, the strictest level, locate exactly where they change. All 11 video world models we evaluate fall short. The best averages 18.3% pair accuracy over real and simulated scenes, below the 25% of random guessing. Models flip as often on steps that leave the future u
world modelagent - arxiv:2611.19821 · cs.LGGraphPass: A Study of Graph Neural Networks for Compiler Pass SelectionMohammed Saqlain
Compiler pass selection is usually governed by fixed pipelines (-O2, -O3) or hand-written heuristics. We study whether a graph neural network (GNN) over a program's control-flow graph (CFG) can select among a fixed menu of five pass sequences in a single forward pass. GraphPass encodes each CFG with a graph attention (GAT) or graph convolutional (GCN) encoder and classifies it into one of the five sequences. On a synthetic benchmark of 5,000 generated CFGs in five program families, both encoders agree with the benchmark's reference label on about 92% of held-out programs, versus 68.4% for a hand-written rule-based heuristic and 85% for a flat MLP. An edge ablation indicates that this gain comes from per-block processing rather than from CFG edges: a GCN given the same basic-block features with all edges removed scores 95.0%, versus 92.2% with the true CFG. In this benchmark the label is determined by the program family, so the task amounts to family recognition. Real-compiler speedup i
benchmark - arxiv:2611.19815 · cs.LGConditional Optimal Transport and Triangular Maps for Imitation LearningMohammad Al-Jarrah, Yuhang Mei, Karen Leung, Bamdad Hosseini +1
We introduce a unified formulation of imitation learning from demonstrations as a conditional sampling problem solved by triangular transport: a policy is the second block of a triangular map that transports a reference distribution to the conditional distribution of the action chunk given the observation history. Unlike in Diffusion Policy, where it is a fixed Gaussian, the reference becomes a design variable; we take it to be the empirical distribution of the demonstrated action chunks and introduce three instances of the formulation: an optimal transport policy (OTP), a flow matching policy (FMP), and a stochastic interpolant policy (SIP). On the Push-T and robomimic Square tasks, the three policies are competitive in task performance with Diffusion Policy and Consistency Policy while using considerably smaller networks and shorter training, and the informative reference accounts for a substantial part of their performance when few network evaluations are used. The three instances o
diffusion policy - arxiv:2611.19812 · cs.LGReasoning in the loop: antibody lead optimization by probing mechanism and measurementSamuel Stanton, Aaron Solomon, Emily Delaney, Melanie Montgomery +6
Lab-in-the-loop systems, in which a model proposes molecules, automated assays measure their properties and the results retrain the model, are a popular approach to antibody lead optimization. These systems optimize for a specification defined with metrics computed from instrument signals, so their success depends on how well those metrics serve as proxies for the properties a drug candidate needs. When a metric is a poor proxy, neither a better model nor more data will fix it, and identifying the problem usually requires expert review. We report two campaigns run on BigHat Biosciences' Milliner platform, one on an anti-albumin nanobody (VHH) and one on an anti-mesothelin single-chain variable fragment (scFv), in which a research team and a general-purpose AI agent, Claude, brought scientific reasoning into the loop. Over seven rounds and about 1,010 design variants we built libraries from antibody lineage, structure and biophysics, and found from the raw curves that on each scaffold o
ai agent - arxiv:2611.19807 · cs.AIMAPS: A Multimodal Agentic Framework for POI Data Production Based on Crowdsourced Street-view ImagerySijie Ruan, Tingyu Xie, Mingfei Liu, Yicheng Tang +6
Point-of-interest (POI) databases support location-based services, yet urban changes quickly make their records stale and manual maintenance is costly. Crowdsourced street-view images provide timely visual evidence, but converting them into database actions requires separating multiple POIs and irrelevant text in each image, then resolving incomplete observations against ambiguous records. We study Street-view Image-based POI Data Production and propose MAPS, the first framework to integrate structured POI extraction with database maintenance from such imagery. MAPS contains two cascaded modules. SPICE progressively adapts a lightweight vision-language model to reconstruct and filter text, group entries belonging to the same POI, and classify their semantics and completeness. RIMA retrieves candidate records, performs evidence-grounded POI matching and attribute-level reasoning, and selectively acquires complementary visual, geospatial, or external-source evidence when necessary. In hu
agentic - arxiv:2611.19803 · cs.CVPros4D: Modeling Evolving 4D Worlds in Latent Space for Dynamic Spatial ReasoningRongxin Gao, Dongxuan Liu, Chu Li, Chengyang Ji +11
Dynamic spatial reasoning from monocular video enables vision-language models (VLMs) to understand how 3D environments evolve over time, supporting physical world understanding and embodied intelligence. Existing efforts primarily address observed-dynamics understanding, whereas agents in dynamic environments also require future-dynamics prediction beyond the observation boundary. Moreover, external geometric processing and textual descriptions can increase inference complexity or lose information when representing continuous dynamics. These limitations motivate a shared internal state that preserves observed dynamics and remains predictive beyond the input video. Accordingly, we present Pros4D, a framework for modeling evolving 4D worlds in latent space that unifies retrospective understanding of observed-dynamics with prospective reasoning for future-dynamics prediction. Specifically, Pros4D-Engine derives geometry-grounded supervision and constructs paired examples for both tasks, w
embodiedbenchmark - arxiv:2611.19802 · cs.CLMachine interpreting competitiveness: a discourse-sensitive comparison of meaning and target-language quality in English-Chinese simultaneous interpretingYuanyuan Ding
Machine interpreting (MI), or AI-enabled speech translation, is increasingly visible in professional and public-facing communication, yet its performance relative to human simultaneous interpreters remains insufficiently understood. Existing studies are often limited by a single discourse mode and evaluation frameworks that are not well aligned with interpreting as a discourse-level communicative activity. The present study compares five commercial MI systems with 26 human interpreters (14 professionals, 12 advanced students) on English-to-Chinese simultaneous interpreting across planned and unplanned discourse, focusing on meaning and target-language quality. In these materials, MI performance differed sharply between the two discourse modes. In the planned formal speech, the strongest system became highly competitive on meaning, entering the upper human range, and a second system also performed strongly. In the extemporaneous interview, however, no MI system approached the stronger h
evaluation framework - arxiv:2611.19800 · cs.LGMulti-Branch TransformerGaotang Li, Lingjie Chen, Chunyuan Deng, Zhiwei Xu +5
Transformer language models are commonly scaled by making their blocks wider or stacking them deeper. In this work, we explore a complementary path: branching, the parallel composition of full Transformer blocks. We introduce the Multi-Branch Transformer (MBT), which processes the input through shared initial layers, executes independently parameterized branches of sequential blocks in parallel, and fuses their outputs before shared final layers. At comparable training compute, we study three MBT variants: Wide-Shallow preserves block width, Narrow-Deep preserves sequential depth, and Balanced interpolates between them. In controlled pretraining experiments at four mixture-of-experts scales (0.8B to 2.1B total parameters), 2-branch Wide-Shallow MBTs consistently outperform sequential Transformers under matched parameter and training-compute budgets. Balanced MBTs improve further and retain comparable quality as the number of branches increases. To improve inference efficiency at long c
long context - arxiv:2611.19796 · cs.LGThe Sense Paradigm for World ModelsHaidong Hu
Latent world models predict the future in an embedding space rather than in pixels. An encoder trained only to predict collapses and maps every observation to the same point, so existing models add a mechanism against collapse, such as a target network, a variance penalty or a distribution penalty. These mechanisms keep the embedding from collapsing, but they leave its distances without meaning, and the model cannot tell a reversible transition from an irreversible one. We propose Sense, a single cross-entropy loss over the directed distance $d(x\to y)=\sum_i \mathrm{ReLU}(y_i-x_i)$ that replaces these mechanisms. The loss avoids collapse on its own, and since the encoder no longer needs the predictor, the predictor is fitted afterwards to fixed embeddings by plain regression. On four pixel environments, the return distance learned by Sense detects irreversible transitions with an AUC between 0.95 and 1.00, where LeWorldModel stays between 0.39 and 0.59. On TwoRoom, the navigation benc
world modelbenchmark - arxiv:2611.19795 · cs.AILossless First-Action Menus for Linear-Reward PlanningYukun Yang
We study sequential decision making where an agent reserves a minimum menu of initial actions before multi-objective trade-offs are revealed, followed by unrestricted continuation planning once linear reward weights arrive. We prove that a menu preserves the unrestricted optimal return across all linear scalarizations if and only if its reachable continuations cover all vertices of the full feature polytope. This geometric equivalence yields an output-sensitive algorithm that is fixed-parameter tractable in the vertex count, combining scalar backward induction with oracle-based facet identification. On explicitly layered graphs with unit feature increments, menu verification is coNP-complete and synthesis is complete for the second existential level of the polynomial hierarchy, even under additive loss strictly below one. In fixed dimensions two and higher, verification is polynomial whereas synthesis remains NP-complete, while one-dimensional instances admit polynomial-time solutions.
agent - arxiv:2611.19793 · cs.LGIANN: A Unified Framework for Equivariant InterAtomic Neural Networks with User-Friendly ImplementationChangzhi Ai, Xueping Qin, Dimosthenis Sokaras, Frank Abild-Pedersen +1
Machine-learning interatomic potentials (MLIPs) now approach density functional theory (DFT) accuracy at a fraction of its cost, but the practical benefit is limited by software fragmentation: each state-of-the-art architecture is released as a separate codebase, with its own data format, configuration system, training loop and molecular-dynamics interface. Comparing two architectures on the same dataset, or moving a trained model into a production simulation, therefore requires substantial re-engineering. We introduce IANN (InterAtomic Neural Network), an open-source PyTorch framework that reimplements equivariant graph neural networks - PaiNN, NequIP, MACE, Allegro, EquiformerV2, EquiformerV3 and UMA - behind a unified data object, a unified trainer and a single deployment route, so that the architecture is a one-word change in a configuration dictionary. Forces, stresses and virials are obtained for every architecture from one analytic differentiation of the energy with respect to t
agentautonomous agent - arxiv:2611.19786 · cs.CVFine-Grained RLAIF for Visual Safety: Learning to Zoom into Hazards for Multimodal Safety AlignmentChunxue Xu, Yiwei Wang, Yujun Cai, Songze Li
Recent multimodal safety alignment heavily relies on a text-centric paradigm, assuming risks are explicit in language or globally prominent in images. However, real-world hazards are often visually subtle or peripheral, causing existing models to fail due to their reliance on coarse, global visual representations. In this work, we study fine-grained visual safety alignment and introduce a challenging 1,200-sample benchmark across four domains, where hazards necessitate localized inspection. To address this, we propose Fine-Grained RLAIF (FGR), a reinforcement learning framework that casts image zoom-in as a learnable perception policy. FGR employs a decomposed reward design to jointly align three core safety behaviors: (i) where to look via a visual evidence reward, (ii) when to act via a tool-formatting constraint, and (iii) how to reason via visually grounded safety explanations. FGR achieves accuracy ranging from 86.71% to 93.00%, consistently outperforming strong multimodal and age
rlaifbenchmark - arxiv:2611.19785 · cs.RODynLimit: A Simulation Benchmark for Dynamics-Critical Planning and ControlYang Xiao, Yongzhi Liu, Sunan Zhang, Kang Wang +2
Experiments near vehicle handling limits must keep three questions apart: what a maneuver requests, what the closed-loop system executes, and whether the task is completed. We present DynLimit, a simulation platform that organizes this process end to end. Scenarios are composed from road, friction, obstacle, and reference objects; planners and controllers attach through a shared observation and control interface; a vehicle adapter runs either a license-free four-wheel model or an external dynamics solver; and synchronized records feed task evaluation, replay, and response inspection. The platform carries a task-oriented evaluation method that pairs a prescribed reference demand with a sampled passing set and event-aligned diagnosis. Across eight task families, three friction environments, and three reference-assisted baseline chains we executed 450 runs: 360 nominal, 18 locally refined, and 72 with prescribed initial offsets. Method differences depend on the task family, equal aggregat
benchmark - arxiv:2611.19780 · cs.CVA Threshold-Free Visual-Semantic Matching Metric for Evaluating Consistency in Subject-Driven Image GenerationBowei Shan, Heng Yin
In subject-driven text-to-image generation, quantitatively evaluating localized visual consistency across varying poses and environments remains a major challenge. Existing metrics either rely on global image embeddings (CLIP, DINO) and Vision-Language Models (VLMs) that pool spatial representations and miss fine-grained textural artifacts, or depend on scalar threshold-gated dense metrics such as Visual-Semantic Matching (VSM) that suffer from point-alignment dropouts, severe threshold sensitivity, and score instability. To overcome these limitations, we propose TF-VSM, a fully continuous, threshold-free visual-consistency metric paired with an extended feature-disentanglement retraining schedule for dual-branch ResNet feature-aggregation networks. Rather than enforcing rigid binary cutoffs $(T_s,T_v)$, TF-VSM weights every object point by a non-linear semantic match confidence $w_j=ReLU(\hat{s}_j)^2$ and continuously integrates visual similarity over all consistency levels $r\in[0,1]
benchmark - arxiv:2611.19779 · cs.AIRead Before You Fit: Text-Derived Constraints for Discovering Differential Equations from Noisy Biological DataPronoma Banerjee, Md. Masudur Rahman, Juan P. Wachs
Large language models are increasingly used for scientific equation discovery. However, most existing methods are strongly data-driven and assume sufficiently clean and representative observations. This assumption is problematic in biological systems, where individual-level measurements are often sparse, noisy, heterogeneous, and expensive to obtain. While the underlying mechanisms are extensively described in scientific text, they are not explicitly used for the LLM based modeling of such systems. We introduce P2C, an agentic framework that translates textual descriptions of biological systems into a typed set of formal constraints, and use these constraints to restrict candidate ordinary differential equation (ODE) structures before numerical fitting. P2C consists of five stages: Text2Constraint extraction, ODE discovery, symbolic verification, parameter fitting, and model scoring. We evaluate P2C on four published ODE systems of acute inflammation and immune regulation, under clean
agentic - arxiv:2611.19775 · cs.RORecovery-Guided Model-Based Diffusion for Safety-Critical Legged LocomotionLei Zheng, Shivam Sood, Mithun Vanniasinghe, Guillaume Sartoretti +1
Safety-critical high-speed legged locomotion requires motion planners to balance safety foresight with online computational efficiency. Although short prediction horizons reduce rollout costs, they provide limited safety foresight, and strict safety constraints severely restrict the feasible trajectory space. We propose a recovery-guided model-based diffusion (MBD) framework for safety-critical legged locomotion. It incorporates a learned terminal recovery critic into short-horizon trajectory optimization. Specifically, the critic approximates the worst-case future clearance margin along rollouts of a designated recovery policy. We introduce a feasibility-guided refinement strategy that combines within-horizon constraints and terminal recoverability at each sampling stage, while retaining feasible elites to mitigate sampling-support loss. Across 50 paired randomized Unitree Go2 obstacle-avoidance trials, our approach achieves zero collisions and 98% strict-clearance success, while requ
legged locomotion - arxiv:2611.19774 · cs.CVWhen Backdoors Can Be Switched Off: Suppression-Aware Backdoor Attacks on Vision-Language-Action ModelsNur Afsa Syeda, Mohamed Elmahallawy
Vision-Language-Action (VLA) models map visual observations, robot states, and natural-language instructions to robotic actions, but their multimodal interfaces introduce new security risks. Existing VLA backdoor attacks primarily study trigger-based activation and stealthy action manipulation, leaving conditional suppression of malicious behavior underexplored. We investigate suppression-aware backdoors, which combine conjunctive trigger activation with instruction-conditioned suppression. We instantiate this design through three variants: Cross-Modal Trigger, combining robot-state and spatially separated visual triggers; Self-Suppressing Trigger, requiring a state trigger and a [DISRUPT] instruction token for activation, with [RESTORE] token suppressing the attack; and Multi-Trigger, additionally requiring a visual trigger while retaining suppression even when physical triggers persist. We evaluate these attacks using SmolVLA on two SO-100 manipulation datasets against four baselines
vision-language-actionvlamanipulationopenvlalibero - arxiv:2611.19771 · cs.ROJEPA4R: Embodied JEPA World Models with 4D Point PlanningJiarui Dong, Zhiheng Liang, Yanjun Li, Zeyu Zhang +3
Joint-embedding predictive architecture (JEPA) world models predict future embeddings instead of pixels and plan in latent space. Two challenges remain: 2D images leave the metric geometry that manipulation acts on implicit and entangled with appearance, and what these models imagine stays hidden, as they are mostly judged in simulation by short-horizon latent planning. We present JEPA4R, a JEPA world model that observes single-depth-camera point clouds, predicts and plans in their latent space, and acts through a policy head on its frozen latents; we call this 4D point planning. A decoder on the frozen world model shows what it imagines as 4D point clouds, with a mean point error within 1.1 cm over 200 steps. Under the policy-learning protocols of OGBench and CortexBench, JEPA4R leads pixel JEPA models on OGBench Cube-Single, Cube-Double, Cube-Triple and Scene, and ranks first on MetaWorld and Adroit Relocate with 98.4% and 88.0%, above the runner-up, R3M, at 96.0% and 65.3%. Trained
embodiedmanipulationdexterousgripperworld model - arxiv:2611.19770 · cs.LGExact Sparse Assignment with Costly Target Embeddings: Fixed-Realization Certificates and an Empirical AuditHamidreza Hasani Balyani, Mahtab Movahhedrad, Seyed Pouyan Mousavi Davoudi
Embedding-based one-to-one association is common in vision and data integration: detections are associated with tracks and records are linked across databases. Such pipelines often solve a sparse optional assignment whose edge scores add a cheap term to an inner product of neural embeddings. When target embeddings are costly, one can acquire them selectively and stop once a certificate proves that the current assignment is optimal. We give such a certificate for unit-norm embeddings. It bounds the unknown part of every reassignment by a support function and maximizes the resulting robust gains with a second sparse assignment; a shared owner-centred bound dominates any fixed-centre triangle bound. We make precise what is certified when network features are not batch-invariant: the realized values of the acquired rows and every unit-ball completion of the rest, not a second forward pass in a different batch. Auditing against a separately batched cache tests a different property; on our f
benchmark - arxiv:2611.19765 · cs.ROM-CROSS: A Multisensor Simulation Dataset for SLAM across Water--Land TransitionsYan Dong, Tianyi Wu, Junyu Yang, Shaoqiang Qiu +3
Water--land cross-domain robots require continuous localization across land, transition and underwater phases, where sensor characteristics and availability change abruptly. Existing datasets and conventional metrics are not designed to capture transition-induced degradation and failures. We present \textbf{M-CROSS}, to the best of our knowledge, the first multisensor simulation dataset and benchmark for SLAM across continuous water--land transitions. It provides synchronized LiDAR, sonar, stereo camera, IMU, DVL, and depth measurements; 6DoF pose ground truth; sensor-wise medium-state annotations; and medium-dependent sensor degradation models. We establish a transition-aware protocol comprising the Trajectory Completion Rate (TCR), Piecewise Absolute Pose Error (PW-APE), Transition End Error (TEE), and Transition Survival Period (TSP), which measure trajectory completion, single-domain accuracy, transition-induced error, and localization survival under prolonged partial immersion, re
benchmark - arxiv:2611.19752 · cs.AIStaleBench: Isolating Stale Retrieval in Agent Memory, and Pricing the FixS M Mahbub Murshed
An agent that stores facts over time eventually stores several versions of the same fact. Asked a value-agnostic question - "which team does X play for now?" - a similarity retriever cannot prefer the current version, because every version matches the entity the query names. StaleBench isolates this stale-retrieval failure on 892 real updating facts, with currency labels from the source (Wikidata validity years via TempLAMA), never from a retrieval score. Our central result is that retrieval metrics can rank memory policies backwards: a retention policy and its retain-everything control return the identical statement at rank 1 for every test query, yet differ by 29.5 and 65.6 points in reader accuracy on two readers when no date is shown. Contamination@5, the metric we add, ranks them correctly where current@1 and current@5 do not. Withdrawing the lexical anchor that statements carry costs deployed keying mechanisms tens of points of freshness; entity resolution on both the write path
memoryagent memoryagentevaluator - arxiv:2611.19748 · cs.CVSEAT: Sign-Type Evidence-Audited Transport for Multi-Label Tongue DiagnosisLiuyi Xu, Qing Wang, Zikai Liang, An-Yang Lu +1
Automated tongue-sign recognition can improve the consistency of non-invasive examinations that otherwise rely on subjective visual judgment. A single tongue image often exhibits multiple coexisting signs characterized by heterogeneous evidence, including localized morphology, global chromaticity, and inter-label dependence. However, existing methods typically rely on shared representations and uniform multi-label heads, making it difficult to verify whether predictions stem from clinically relevant evidence rather than spurious visual correlations. We propose SEAT, an evidence-audited framework that matches distinct evidence types with dedicated prediction and evaluation mechanisms. Built on frozen DINOv2 patch features, SEAT couples mask-guided sign-specific attention for localized morphology, a semi-relaxed optimal transport pathway over LAB/HSV color prototypes to enhance pale-tongue recognition, and a co-occurrence-aware transport regularizer to shape the batch-level predicted lab
benchmark - arxiv:2611.19747 · cs.CVHow Far Is Too Far? Object Recognition Declines Monotonically with Semantic DistanceDomenic Bersch, Hai Van Tran, Gemma Roig
Vision models are sensitive to object-background context, yet it remains unclear how recognition changes as the semantic distance between objects and backgrounds increases. We introduce ImageNet-OOC1k, a benchmark enabling graded control of object-background relationships along a continuous consensus visual-semantic dissimilarity axis. This formulation reveals a highly consistent monotonic decline in recognition accuracy across 35 models spanning convolutional neural networks, Vision Transformers, and vision-language models. Internal analyses of Vision Transformers reveal patterns consistent with aggregation failure, where object evidence is present at the patch level but not consolidated into the final prediction. Using this characterization as a diagnostic probe, a simple spatial re-ranking strategy recovers a subset of these errors and improves in-sample accuracy by up to 5.7 percentage points. These results establish context sensitivity as a continuous and predictable property of m
benchmark - arxiv:2611.19746 · cs.CVEvent-based object detection under adverse lighting and motion: lessons for construction sitesZhuoran Feng, Xiaoshan Zhou
Vision-based monitoring on construction sites fails at night, at dawn and dusk, in tunnels and under floodlights, when conventional RGB cameras under- or over-expose. Event cameras record per-pixel brightness changes with microsecond latency and over 120 dB of dynamic range, but their benefit for site-relevant object detection has not been quantified on real data. We compare three recent event-based detectors (SMamba, SATE and EvRT-DETR) with an RGB detector trained on the same data, using the pixel-aligned PEOD dataset. On low-light frames (70% of the test set) the best event detector reaches 27.2 AP against 21.3 AP for RGB, and a training-free late fusion of both sensors is the best option in every lighting condition (33.3 AP). An object-level analysis of 53,091 objects shows that brightness has little effect on event detectors, whereas fast motion matters: above 20 pixels per frame RGB finds 11-19% of medium-size objects and EvRT-DETR 39-60%. Activation patching locates the damage c
event camera - arxiv:2611.19741 · cs.LGWhere the Error Lives: Separating Lid-Edge and Body-Induced Error in Neural Pressure Prediction for 3D Cavity FlowZirun Zhou, Zhi-Qian Zhang, Yixuan Hong, Guang Liu +2
Can a pressure surrogate score much better globally without better resolving the pressure around immersed bodies? We test this in a three-dimensional lid-driven cavity with an unregularized lid and centered pressure at native $128^3$ resolution. On 60 cases from 12 unseen geometries, with predictions sealed before the CFD truth existed, adding a body-free, Reynolds-conditioned reference correction to the same frozen $64^3$ surrogate lifted to $128^3$ lowers the mean full-volume relative $L_2$ error from 20.6% to 7.5%. In the post hoc regional analysis, the lid-edge junction occupies 1.5% of the fluid but holds 69% of the truth pressure energy and 95% of the uncorrected surrogate's squared error. Interior error changes by just +0.16% relative, and object-local peak error and recall change little on average. Retrospective ablations locate the learned term's contribution in the interior and near the bodies; rebasing on the reference lowers global error further with little change in cohort
benchmark - arxiv:2611.19738 · cs.LGRelax-and-Sparsify Approximation Algorithms for Sparse Ridge RegressionShixuan Yi, Yongchun Li
This paper studies sparse ridge regression in the low-rank, high-dimensional setting that often arises when data collection is expensive. By exploiting the low-rank structure, we develop approximation algorithms within a relax-and-sparsify framework: solve a convex relaxation and then sparsify its solution to satisfy the original sparsity constraint. Specifically, we establish a tight rank-dependent sparsity bound for the convex relaxation by decomposing its support into two parts. This decomposition in turn motivates two approximation strategies, parameter reduction and binary redistribution, each exploiting a different part. Experiments on synthetic and real datasets show that the binary redistribution algorithms outperform the benchmarks in most instances.
benchmark - arxiv:2611.19733 · cs.LGSample-Efficient Ensemble Reinforcement Learning at ScaleYoav Feigenbaum, Dmitrii Krylov, Roy Fox
Ensemble methods such as MeanQ have been shown to improve the sample efficiency of relatively small model-free deep reinforcement learning (RL) architectures, including DQN and similarly sized variants such as Rainbow. However, to our knowledge, the effects of ensembling on large-scale architectures such as Bigger Better Faster (BBF) have not been studied. In this work, we present MeanQ-BBF, a hybrid model-free algorithm that combines the principles of MeanQ and BBF. Through statistically rigorous evaluation on the Atari 100K benchmark, we show that in contrast to naive scaling of the network size of BBF, which degrades performance, MeanQ-BBF is able to leverage the increase in model capacity to improve performance over BBF. Moreover, MeanQ-BBF does this in a computationally efficient way by organizing the added parameters into largely independent learners. Finally, we discuss the performance-cost tradeoff in the context of practical sample-efficient RL.
benchmark - arxiv:2611.19731 · cs.ROCooperative Transport of Unknown Objects by Distributed Effort Regulation in Multi-agent SystemsTamzidul Mina, Shyam S. Kannan, Wonse Jo, Shaocheng Luo +2
A distributed multi-robot object transportation strategy is proposed for objects of unknown shape, size, geometry, mass and mass distribution without any prior knowledge about the object or number of robots involved in the transportation process. Robots follow the boundary of the object and dynamically determine viable pushing points. A novel object-model-free heading-error-driven pushing effort regulation method is proposed to promote goal-directed object motion; robots regulate their pushing effort using individual and collective heading errors. Robots dynamically relocate around the object boundary if a pushing point is lost or the required pushing effort exceeds its capability to correct its heading. The proposed method is validated using simple and complex shapes with convex and non-convex boundary regions and dynamically changing weight distribution of objects. Detailed analysis on the effects of robot distribution and initial object pose on the transportation performance in term
multi-agentagent system - arxiv:2611.19720 · cs.ROSim-to-Real Room-Mediated Object-Goal Navigation with ROS 2 and Nav2Matthew Chua, Adam Scicluna, Gavin Paul
Although object-goal navigation (ObjectNav) approaches are usually evaluated in simulation, which offers scale and repeatable environments, real-world deployments are rarely examined beyond feasibility demonstrations. This paper presents the deployment of Room-Mediated Co-occurrence, an ObjectNav policy that reasons over Room Probability Vectors (RPVs), on two robotic platforms (Boston Dynamics Spot and Clearpath Robotics TurtleBot 4) through a robot-agnostic ROS 2 bridge. We study how these semantic priors generalise to physical environments and describe the adaptations required to transfer RPV from simulation to hardware, covering sensing, localisation, and the action interface. Across 66 trials on Spot, searching for 11 object categories from two starting positions in a university postgraduate research facility, RPV achieves a raw Success Rate (SR) of 0.455. Our analysis identifies undetected glass walls, entry into off-limits areas, limited usable depth range, target size, and mism
sim-to-real - arxiv:2611.19719 · cs.LGFound, but Not Fixed : From Software Vulnerability Discovery to Fixes with Iterative LLM PipelinesYoussef Gamal, Mohamad Fakih, Halima Bouzidi, Rahul Dharmaji +4
Software vulnerabilities persist as a critical threat despite the deployment of sophisticated detection tools and AI assistants. While detection has advanced, the subsequent code repair phase remains a primary bottleneck; manual intervention is inherently slow, costly, and prone to introducing secondary regressions and issues. In this paper, we investigate whether an iterative, LLM-based pipeline can provide a robust solution for the automated repair of vulnerable functions in real-world code. We propose a generalized framework that leverages the reasoning capabilities of Large Language Models (LLMs) through a multi-stage feedback loop. We benchmark this approach against state-of-the-art models, including GPT-4o, GPT-5.1, and Gemini 3 Flash, demonstrating that such a pipeline can achieve high-quality code repair. To support the broader security community, we provide our evaluation framework, fine-tuned weights, and comprehensive experimental dataset on our website: https://sites.google
benchmarkevaluation framework - arxiv:2611.19712 · eess.SYBehavioral Denoising: A New Perspective on Nonlinear Behavioral Control Inspired by Closed-Form Diffusion PoliciesLucas Y. Jia, John Z. Zhang, Zachary Manchester
Diffusion policies have become a dominant approach to imitation learning in robotics. Recent work suggests that, in the low-data regime, they largely retrieve or combine the demonstrated actions, which the closed-form diffusion policy does explicitly. This closely resembles behavioral control, where future inputs are formed as linear combinations of recorded trajectories. In this work, we make this connection precise: we show that the weights with which the closed-form diffusion policy combines the demonstrated actions are the unique solution of a regularized behavioral control problem. Building on this connection, we propose an algorithm, termed $χ^2$-projected behavioral denoising, which is a training-free method that combines nearby trajectory data with a limit on extrapolation when the data offer little support. It shows encouraging results on four nonlinear control tasks compared to recent nonlinear behavioral control methods. We hope that this connection offers a useful perspecti
diffusion policy - arxiv:2611.19710 · cs.LGJoint Learning of Factual and Counterfactual Worlds for Causal Prior-data Fitted NetworksFurkan Danisman, Luke Zhang, Zhijing Jin, Dehan Kong
Estimating treatment effects is central to science across many disciplines and requires counterfactual reasoning about alternative worlds. In the real world, however, each unit reveals its outcome only under the factual world. Thus, the factual and counterfactual outcomes can never be jointly learned by any model trained on real data. The problem is therefore tractable only in synthetic settings, where both worlds can be generated. Recent progress on prior-data fitted networks (PFNs) has shown that models pre-trained purely on synthetic data can transfer to real-world prediction. This paradigm has since carried over to causal effect estimation. Despite this fully synthetic setup, existing causal PFNs carry the real-world constraint into their training objective, confining learning to a single world's outcome per unit. The dependence between the two worlds is thereby lost, and the density of the treatment effect relies by default on independence, which is implausible in practice. We the
benchmark - arxiv:2611.19709 · cs.ROCache or Denoise Less? Residual Reuse in a Flow-Matching World Action ModelAbhinav Gupta, Nissan Pow
Flow-matching robot policies generate each action chunk through repeated network evaluations, one per denoising step. Cross-cycle caches such as C$^3$ache reuse computation from earlier replanning calls, but whether they beat simply taking fewer steps is unclear. We compare the two in Fast-WAM, a 6B-parameter world action model, on LIBERO-Goal. Two uncached steps keep the success of ten: on all 500 initial states they are non-inferior within two percentage points, for one checkpoint and noise seed. Under PyTorch's default compile mode, they also cut a replanning call on an RTX A6000 from 202 to 92 ms. This leaves little room for caching. A cache that keeps two fresh steps cannot be faster than two uncached steps: every call still encodes the images, runs the video expert and computes those two steps. Caching the first of two steps, refreshed every fourth call, saves at most 11--14% of a call by our cost model. Cached policies are also fragile. Their losses tracked the share of the flow
libero - arxiv:2611.19701 · cs.LGMultiaccuracy as an Error Benchmark for Learning with Distribution ShiftKonstantinos Stavropoulos
Agnostic learning has a well-developed theory of computational and statistical complexity. For learning with distribution shift, even the choice of an appropriate error benchmark remains challenging. Testable learning with distribution shift (TDS) was introduced to develop computational foundations for this setting: using labeled training data and unlabeled test data, the learner must either produce a predictor with low test error or detect distribution shift. However, standard agnostic TDS guarantees allow a constant-factor amplification of the optimum error even when the training and test distributions coincide. They therefore do not directly recover the guarantee of classical agnostic learning. We introduce shift multiaccuracy, a stronger error benchmark inspired by multiaccuracy from algorithmic fairness. This yields a TDS formulation that becomes equivalent to multiaccuracy when there is no shift. Using a recent nontrivial reduction from agnostic learning to multiaccurate predicti
benchmark - arxiv:2611.19700 · cs.LGWhere Should LLM Agents Adapt at Test Time? In Context, in Weights, or BothParsa Mirtaheri, Mohammadreza Armandpour, Keivan Alizadeh, Iman Mirzadeh +4
LLM agents operating repeatedly in the same environment, such as a code repository, a database, or an application suite, encounter tasks that share latent structure. Experience can reveal this structure, allowing agents to reuse environment knowledge and learn from environmental feedback. Agents can adapt at test-time using this experience through two channels: retaining it in context and updating their weights. These channels are usually studied separately, leaving unclear what each contributes and when they complement each other. We compare in-context learning (ICL), reinforcement learning with verifiable rewards using GRPO, and self-distillation using SDPO, alone and in combination, using a shared online protocol and modular framework. Agents encounter matched task streams, are evaluated on each task before adapting to it, and learn only from their own attempts and environment feedback. We evaluate 14 adaptation mechanisms and a stateless baseline with Qwen3.5-9B on 109 tasks from f
llm agent - arxiv:2611.19688 · cs.ROPhysics-Grounded STL Monitoring from Image ObservationsXinpei Ni, Glen Chou, Samuel Coogan
Robotic manipulation missions must satisfy timed and conditional requirements that can be expressed in signal temporal logic (STL). While a practical means of monitoring these requirements is from camera images, this is challenging because STL robustness depends on physical margins, such as distances and clearances, that are not directly observed. Rather than learning a separate visual monitor for each formula, we learn a shared latent representation from multi-view images and decode physically meaningful margins for individual predicates. Co-training the encoder with predicate decoders increases the certification rate from 58.7% to 96.3% compared with a frozen encoder, despite nearly unchanged predicate sign accuracy. STL semantics then compose these margins into robustness estimates for any bounded formula over the trained predicates, without retraining. We further calibrate the decoding errors of selected STL clauses using conformal prediction, yielding two-sided intervals that cert
manipulationlibero - arxiv:2611.19676 · cs.LGScore-Based Stochastic Closure Modeling via Offline Indirect Learning and Online Data AssimilationXinghao Dong, Huchen Yang, Jin-Long Wu
Stochastic closure modeling is an emerging area in reduced-order modeling of large and complex (e.g., multiscale, multi-physics) systems, for which the computationally affordable reduced-order physical descriptions (e.g., digital twins for additive manufacturing, world models for AI agents) in numerical simulations inevitably lead to information losses that cannot be well addressed by classical deterministic closures. However, learning stochastic closures typically requires closure data that are difficult to obtain from measurements or often prohibitively expensive to compute using first-principles simulations. In this work, we develop a stochastic closure modeling framework that builds on score-based diffusion models, leveraging indirect offline training and online data assimilation without requiring the direct data of closures. Numerical experiments on two-dimensional turbulent flows demonstrate that indirect offline learning captures key closure statistics and improves predictions o
world modelai agent - arxiv:2611.19663 · cs.AIWhen AI Agents Should Repair, Replan, or Abandon: A Threshold Theory of Plan CommitmentGaurav Tewari
AI agents often discover mid-execution that their plan is broken, for example when an action fails or a tool is withdrawn. Deployed systems use conventions such as retry loops or "always replan," and a recent benchmark finds no fixed strategy dominates. This paper derives the optimal policy. In the model, a disruption destroys the causal support of part of a plan; repair is a partial reset that preserves the plan's structure, replanning is a full reset, and changing a plan is costly because others have conditioned on it. The optimal policy is a threshold in effective disruption depth: repair shallow damage, replan after deep damage, and abandon the goal when even a fresh plan is not worth its cost. Depth depends on plan structure. In expectation, a single broken action forces 1+R(G)/n actions to be revalidated, where R(G) counts the pairs of actions joined by a dependency path; to first order, a fully serial plan switches to replanning at raw damage about w times smaller than a plan of
agentai agentbenchmark - arxiv:2611.19656 · cs.LGBadGCL: Towards Persistent Backdoor Attack on Graph Continual Learning SystemsDong Li, Yufei Han, Xintao Wu, Xujiang Zhao +2
Backdoor attacks pose a persistent security threat to machine learning systems, yet their behavior under graph continual learning (GCL) remains underexplored. Conventional graph backdoor attacks often lose effectiveness as models are updated across successive tasks, creating the impression that continual updates naturally weaken malicious associations. This work reveals a contrasting vulnerability: knowledge-preservation mechanisms in GCL can also preserve early injected backdoors when the poisoned update is compatible with the clean-task signal being retained. For graph data, this risk is amplified by structural triggers, which are embedded into the message-passing paths of graph neural networks (GNNs) and can induce stable representation shifts across tasks. To expose this risk, we propose BadGCL, a model-agnostic framework for persistent backdoor attacks under a strict early-stage intervention setting. BadGCL learns topology-aware structural triggers using a gradient mutual-informat
benchmark - arxiv:2611.19650 · cs.LGWith the Right Representation, Classical Machine Learning Rivals Deep Models for Olfactory Mixture SimilarityZe Dong, Yi Wang, Ambuj Tewari
Modeling olfactory mixtures remains a critical challenge in machine olfaction because the task requires reasoning over relationships between sets of odorants rather than over individual molecules in isolation. However, prior comparisons between deep and classical approaches also varied how the relationship between the two mixtures was represented, since the classical baselines primarily relied on weak representations that described each mixture independently and provided little direct signal about how the mixtures differed. In this work, we revisit classical learning using explicit symmetric comparison features on unordered pairs of mixtures, evaluating these representations with XGBoost and kernel ridge regression (KRR). We find that XGBoost with explicit comparison features achieves performance competitive with specialized deep mixture models on the main benchmark. Ablation studies show that coordinate-wise absolute differences alone provide strong predictive performance, while combi
benchmark - arxiv:2611.19649 · cs.LGLoopFill: Pipelined Multi-Token Speculative Decoding for Looped TransformersSiqi Zhu, Suozhi Huang, Yi Zhao, Jiaxuan You
Looped Transformers improve parameter efficiency by repeatedly applying the same Transformer layers. During autoregressive decoding, however, generating each token requires several sequential passes through these layers, leading to high decoding latency. Pipelined self-speculative methods use shallow-loop predictions to process tokens at different loops in the same batch, but each token must reach the drafting depth before the next can enter the pipeline. We introduce LoopFill, a training-free speculative decoding algorithm for looped LLM that combines multi-token suffix drafts with shallow-loop predictions in a single draft tree. Both drafters contribute to one tree, so a suffix lookup can immediately extend a shallow prediction and several future tokens can enter the pipeline together. The shared block computes the highest-scoring draft candidates at different loop depths in one batch, and full-depth verification preserves the model's autoregressive output distribution. Across seven
benchmark - arxiv:2611.19643 · cs.LGSG-LegalClarify: A Large-Scale Benchmark for Legal Reasoning and Ambiguity Recognition in Singapore LawZiwei Gu, Michael S. Lee, Drew Rein, Yash Maurya +4
Singapore offers a distinctive testbed for legal reasoning by large language models (LLMs). Its legal system combines inherited English common law with jurisdiction-specific statutes, requiring models to distinguish local doctrine from generic common-law intuition. This setting also highlights the challenge of recognizing underspecification in lay legal queries, because the facts material to a legal outcome depend on the governing doctrine. Yet legal benchmarks rarely cover Singapore law or test this form of ambiguity recognition. To address these gaps, we introduce SG-LegalClarify, a large-scale benchmark for open-ended legal reasoning and ambiguity recognition in Singapore law. SG-LegalClarify contains 1,100 lawyer-authored source questions spanning 10 legal domains, including 340 fully specified and 760 underspecified questions. Each underspecified question has a fixed hidden-fact assignment and a fully specified counterpart, yielding 1,860 evaluation instances. Responses are graded
benchmark - arxiv:2611.19636 · cs.AIROUNDS: Source-Framed Diagnostic Revision after Multi-Turn Clinical EvidenceJinrui Fang, Runhan Chen, Yining Wu, Jiawei Xu +5
Large language models (LLMs) achieve high scores on medical benchmarks, but does this performance translate into reliable diagnosis as a conversation unfolds? In diagnostic conversations, users can provide clinical findings across multiple turns, ask for a diagnosis without supplying answer options, and introduce opinions from others. Models must therefore form diagnoses from the available evidence and decide how to respond to subsequent advice. We introduce ROUNDS, a framework that evaluates these abilities after a fixed sequence of clinical evidence delivered over multiple turns. ROUNDS compares diagnosis with and without answer options, presents the same incorrect diagnosis under clinician and lay framing, and varies support for competing diagnoses in written five-clinician panels. Across seven models and 1,035 cases from five datasets, average accuracy was 76.9% with answer options and 57.0% without them. Clinician-framed misleading advice produced lower final accuracy than lay-fra
benchmark - arxiv:2611.19634 · cs.LGMulti-Objective Causal Bandits: Minimal Intervention Space and Policy-Level LearningMuhammad Qasim Elahi, Mahsa Ghasemi, Murat Kocaoglu
Many decision-making problems involve multiple objectives, where actions affect several performance metrics, fairness criteria, or safety constraints. We study multi-objective causal bandits with multiple reward nodes, where actions correspond to interventions on subsets of variables in a causal graph and performance is evaluated through Pareto optimality. Existing approaches to multi-objective bandits often rely on pairwise Pareto dominance or optimize a fixed scalarization. However, pairwise comparisons alone can fail to identify actions dominated by randomized policies, whose expected reward vectors lie in the convex hull of action-level rewards. Thus, an intervention may be suboptimal even if no single intervention dominates it, because it can be dominated by a randomized policy over other interventions. We formalize this policy-level phenomenon and show that it has consequences for both causal action-space reduction and online learning. First, we provide a necessary-and-sufficient
online learning - arxiv:2611.19625 · cs.LGRobust Well-Log Reconstruction Across Point Gaps and Whole Wells via Boundary-Gated Soft-Thresholding Kalman FusionYongan Zhang, Zhihan Qiu, Yizheng Li, Youzhuang Sun
Missing wireline logs hinder subsurface characterization. Deep reconstruction networks suffer from gap-boundary steps and cross-basin shifts, while interpolation and Gaussian smoothers struggle with long intervals and outliers. We propose boundary-gated soft-thresholding Kalman fusion (ST-KF), unifying multiresolution wavelet shrinkage, distance-gated residual prior learning, and closed-form robust Kalman smoothing with learning--inference decoupling. Across two public benchmarks totaling 263 wells in the Norwegian North Sea and New Zealand's Taranaki Basin, ST-KF ranks first over 0.15--100-m anchored gaps with 0.09% parameter overhead. Against the strongest of nine baselines, pooled RMSE falls by 5.0%--8.3% for bulk density and 6.4%--15.6% for compressional slowness; against the four strongest baselines, reductions are 0.7%--4.1% for neutron porosity and 1.3%--2.9% for photoelectric factor. Ablations and sensitivity analyses confirm operator synergy and stable performance: removing st
benchmark - arxiv:2611.19624 · cs.LGWell-Log Reconstruction with Pre-trained Geological Knowledge: A Knowledge-Conditioned Large-Language-Model Generative AgentYongan Zhang, Yitao Yu, Yizheng Li, An Gong +1
Missing well-log intervals make subsurface reconstruction ill-posed. Existing numerical regression models lack geological prior knowledge to constrain ambiguous inverse mappings. We propose GeoLog-Agent, a knowledge-conditioned generative agent using pre-trained large language models (LLMs) as a source of geological priors. A universal Geo-Harness translates LLM stratigraphic reasoning into formation-anchored petrophysical trajectories, feature-wise linear modulation, and conditional adversarial constraints. On the Norwegian North Sea benchmark, GeoLog-Agent achieves a mean $R^2$ of 0.9307 and a peak $R^2$ of 0.9713. It exceeds the nine-baseline average $R^2$ of 0.8387 by 0.0920, with a 30.97% average RMSE reduction, and exceeds the strongest baseline by 0.0274--0.0498 in $R^2$, with 16.25%--24.05% lower RMSE. Ablations confirm the effectiveness and synergy of geological prior reasoning, multi-scale dilated generation, and conditional adversarial modules. Transferring the coupling mech
agentbenchmark - arxiv:2611.19623 · cs.AIData-Driven Heuristic-Guided Business Process RedesignNan Sai, Karolin Winter, Remco Dijkman
Business process redesign (BPR) has a long-standing catalog of best-practice redesign heuristics, yet, applying them remains a mostly manual task that requires extensive domain knowledge. Despite recent advances using Large Language Models (LLMs) for conversational process model redesign, limitations remain. For example, redesign is user-driven instead of data-driven and is not guided by best practices such as the redesign heuristics. Moreover, only one change can be handled at a time. To address these limitations, given an event log and a BPMN model, we present the DH-BPR framework to realize data-driven heuristic-guided BPR using LLMs. To design DH-BPR, we first conducted an analysis of which redesign heuristics can be automatically detected and applied given information contained in an event log such as the fraction of cases that bypass an activity. Based on this, DH-BPR detects and validates heuristic candidates through a combination of rule-based techniques and an LLM. After that,
benchmark - arxiv:2611.19621 · cs.ROToward Planning Representations That Separate Task Structure from Visual ContentAarav Srivastava, Vishwas Sathish, Rajesh P. N. Rao
Effective planning requires representations that preserve task-relevant geometry and object identity while ignoring visual variation that does not alter the task. Existing evaluations of visual generalization often change several factors simultaneously, making the source of a failure difficult to identify. We introduce a controlled evaluation of object-goal navigation (ObjectNav), the task of reaching an instance of a named object category, in procedurally generated ProcTHOR houses. Training and evaluation houses share floor plans, object locations, goals, and start states, while visual factors are changed individually or in combination. Across model-free reinforcement learning (RL), model-based RL, and agents using frozen pretrained visual representations, our results show that appearance changes impair navigation, but their effects differ sharply by factor. Surface material changes account for most of the average change in success, whereas clutter and lighting have little effect. The
agent - arxiv:2611.19616 · cs.ROTowards Probabilistically Safe Reinforcement Learning: Theory and Deployment on Humanoid and Aerial RobotsPol Mestres, Varun Gabbita, Jason H. K. Siu, Sergio A. Esteban +3
Reinforcement learning (RL) enables robots to acquire complex skills, but standard RL formulations do not guarantee that learned policies satisfy safety constraints, which limits their use in safety-critical settings. Existing safe RL methods with provable guarantees typically support only restricted classes of constraints and require unbiased estimators of the objective and constraint functions, which prevents their direct integration with state-of-the-art algorithms such as proximal policy optimization (PPO). In this paper, we extend the recently introduced robust safe gradient flow-based reinforcement learning (RSGF-RL) framework to overcome these limitations. Our formulation accommodates a general class of trajectory-level safety constraints, including chance constraints such as collision avoidance with a prescribed probability, and our analysis allows for biased value and gradient estimators. We show that, with high probability, the constraint violation throughout training is boun
humanoidwhole-body controlsim-to-real - arxiv:2611.19614 · cs.AIA Hierarchical Statistical Framework for Evaluating Agent ReliabilityMirco Milletarì, Tuhin Kundu
Agents operating in complex scenarios often face uncertainty due to incomplete information about the environment they operate in, ambiguous user queries, complex business rules and the inherently stochastic dynamics of the governing policy. Several new benchmarks have been proposed to evaluate LLM-based agents on complex tasks in multi-tool environments, including settings with policy constraints and varying levels of intent ambiguity. These conditions make it important to evaluate agentic systems systematically and quantify their performance under different conditions to assess their reliability in production. This need is evident in the recently introduced Thinkingbox benchmark, where even the strongest models struggle to perform consistently across diverse business scenarios. This paper presents a statistical framework for evaluating uncertainty and reliability in agentic systems. We define a hierarchy of scenarios, instances, and samples to assess model performance and variability
agentagenticbenchmark - arxiv:2611.19609 · cs.LGBefore the Word: Name Information in Description-Derived States of Small Language Models, With and Without Naming FailureFernando Cardenas Piepereit
The tip-of-the-tongue (TOT) state, knowing an object but not its name, motivates this study as an analogy only. We ask whether small language models form a name-predictive state from a name-free description, and whether it differs when naming fails but description selection succeeds. In an internally pre-registered study (a frozen document, no external registry) of Qwen2-0.5B, SmolLM2-1.7B and Phi-3-mini with 200 items, we combine a similarity read-out, a Patchscopes-style read-out and activation patching with matched controls. H1 (name information in the state) was supported in all nine model-by-description-type combinations; the agreement rule was met by the similarity measure together with patching, while Patchscopes stayed near chance. Post-hoc robustness checks after internal review (logit lens, single-position patch, manipulation check) temper this reading: name information in middle layers is clearest under the multi-position patch and the similarity measure, and weak under sing
manipulation - arxiv:2611.19603 · cs.LGNAP-Wave: A Neural Adaptive Plane-Wave Architecture for High-Frequency Helmholtz ProblemsHamza Kamil, Mohammad Mahdi Abedi, David Pardo
Training neural networks to simulate highly oscillatory wavefields presents optimization challenges, often hindered by spectral bias and poorly conditioned physics-based loss landscapes. To address these challenges and enhance convergence, we propose NAP-Wave, a joint neural architecture and training strategy that expands the target scattered wavefield in an adaptive neural plane-wave basis. The basis functions are extracted from all hidden layers and modulated by trainable plane-wave carriers, embedding the high-frequency oscillatory physics directly into the network's trial space. NAP-Wave is trained on a physics-informed loss built from the Lippmann-Schwinger integral equation, so no differentiation of the network with respect to its input is required. We develop a decoupled optimization strategy in which Adam updates the parameters of the neural plane-wave basis, while the linear combination coefficients are computed exactly by a least-squares (LS) solver. We couple this with an ad
benchmark - arxiv:2611.19598 · cs.MAProtecting the System, Stranding the User: The Limits of Multi-Agent AuthorizationGurpriya Bhatia, Clifford Neuman
Current multi-agent authorization architectures prioritize system integrity over human outcomes. While recent models successfully prevent privilege escalation, they externalize the cost of failure onto the human operator. Re-centering authorization around human impact exposes four structural vulnerabilities in current paradigms. First, calling an agent's understanding of user intent verified is misleading. Intent has no formal definition, so no check against it can be definitive. Systems should instead report an openly probabilistic estimate and scale their caution with the risk of the action. Second, we argue that dynamic trust scores do not survive multi-agent handoffs; transmitting algorithmic judgment across a company boundary without underlying evidence risks violating Contextual Integrity. Third, treating a safely halted workflow as a successful refusal is a category error if it strands the user; compensating actions must be bound directly to the capability token, not only to the
multi-agent - arxiv:2611.19597 · cs.LGSynthetic Neurosymbolic Task Generation with Controlled Constraint ComplexityMykhailo Buleshnyi, Zsolt Zombori, Michael Benedikt
We introduce NeSySy, a synthetic generator for neurosymbolic learning tasks that jointly produces logical constraints and consistent datasets. Unlike existing benchmarks, NeSySy provides fine-grained control over key properties such as constraint structure, tightness, input dependence, and data generation and alignment strategies. In particular, it allows the use of rich constraints over infinite domains, and uses solvers to ensure that the constraints are meaningful and that generated samples satisfy the constraints. We design NeSySy to support both learning with declarative constraints and learning from distant supervision. Across a range of constraint families, we provide evidence that its parameters induce systematic and largely independent effects on task difficulty. Using NeSySy, we study constraint-aware learning methods, identifying regimes where explicit constraint enforcement is beneficial and where expressive unconstrained models can instead learn constraint structure direct
benchmark - arxiv:2611.19586 · cs.LGSLiDE: Structured Linear Dynamics for Forecasting with Exogenous InputsSebastian Pütz, Theodore Glavas, Benjamin Schäfer, Boris N. Oreshkin +1
Accurate time series forecasting with exogenous inputs is critical across domains, including energy and retail, yet modern deep learning models often overfit to correlations that do not generalize under shifts in these inputs. We propose SLiDE, a Koopman-inspired architecture for forecasting with exogenous inputs that imposes structure on latent temporal dynamics. SLiDE approximates nonlinear system evolution using a learned linear operator in a latent space, combining (i) a history encoder that reconstructs the current latent state from past targets and inputs with (ii) a shared linear rollout driven by future exogenous variables. By enforcing a common linear recurrence across encoding and prediction, SLiDE aligns how past information is encoded with how future states evolve, reducing parameterization and preventing reliance on spurious correlations in the input history. Empirically, SLiDE achieves state-of-the-art accuracy on real-world benchmarks, including electricity price forecas
benchmark - arxiv:2611.19578 · cs.CVEndoComparer: Comparative Surgical Video Question Answering with Pseudo-Siamese Visual Adaptation and Perceive-then-Compare ReasoningShi Li, Lalith Sharan, Joël L. Lavanchy, Kun Yuan +7
Purpose: Surgical training benefits from explicit comparison between cases: in patient condition, in technique and in skill proficiency. Surgical vision-language models (VLMs) offer a promising and flexible interface for this purpose. Yet current VLMs typically see one video at a time, and they often fall short when asked to compare videos. Methods: We propose EndoComparer, a VLM for comparative surgical video question answering with two components. (a) A Pseudo-Siamese Visual LoRA (PSVL) gives each video slot its own low-rank adapter over a shared frozen vision encoder, separating the two videos in the visual pathway at the cost of a single encoder pass. (b) A Perceive-then-Compare chain of thought (PtC-CoT) trains the model to first state what each video shows, then derive the answer from these statements. We further curate EndoDiffVQA, a benchmark of nearly 11K multiple-choice questions on video-pair over cholecystectomy and gastric bypass procedures, covering anatomy, tool, action
benchmark - arxiv:2611.19569 · cs.AILearning from Mutants: Teaching Small Language Models to Synthesise Loop InvariantsDaragh King, Vasileios Koutavas
Large language models (LLMs) have proven promising for synthesising loop invariants, a long-standing obstacle to automated program verification. Yet, frontier LLMs are typically accessible only through third-party APIs, hampering reproducibility and local deployment. Small language models (SLMs) sidestep these obstacles but require task-specific fine-tuning, for which high-quality training data is needed. We present a failure-driven data curation method for fine-tuning language models to synthesise loop invariants. Inspired by mutation testing, our approach applies semantic mutation operators to ground-truth invariants, producing semantic mutants that induce verification failure. Each mutant is paired with verifier feedback and its ground-truth counterpart, yielding training examples for feedback-guided invariant refinement. Whereas the vast space of mutants is a long-standing burden in mutation testing, it becomes an asset when generating training data, multiplying the examples obtain
iterative refinement - arxiv:2611.19564 · cs.AIBehavioral Assurance for Socially Deployed AI AgentsNicole H Brennan, Jiaee Cheong
AI agents are increasingly deployed in socially consequential settings, yet existing evaluations provide limited assurance that they behave consistently with their intended characteristics across diverse interactions. We introduce an evaluation framework for measuring the behavioral consistency of socially deployed AI agents. The framework operationalizes agent behavior along five measurable axes and uses a standardized evaluation battery spanning six socially salient interaction settings to assess behavioral differentiation, consistency, and drift. Controlled persona configurations serve as experimental anchors, enabling systematic testing of whether intended behavioral differences persist across contexts rather than appearing only as stylistic variation. Preliminary analyses demonstrate measurable differentiation across configurations and suggest associations between behavioral profiles and interaction outcomes. By shifting the focus from persona design to behavioral assurance, this
agentai agentevaluation framework - arxiv:2611.19559 · cs.ROResWAM: Learning What Changes, Not What Stays, for World-Action ModelsChongyu Fan, Gaowen Liu, Ramana Rao Kompella, Jason Corso +1
World-action models (WAMs) augment vision-language-action policies (VLAs) with future prediction so that the policy can anticipate how its actions will affect the physical world. Existing WAMs supervise complete future observations. However, observations typically change little over an action chunk, so the future loss is dominated by copying static background content rather than learning task-relevant robot and object motion, limiting policy learning. Moreover, existing WAMs rely on large video generation models or separate world models, incurring heavy training and inference compute that hinders closed-loop deployment. To improve policy performance, we propose visual residual learning, supervising only the feature change from the current frame to the future frame in a frozen DINOv3 space. Compared with complete future features, this residual learning target yields a higher feature rank with richer and more diverse representations of scene changes, and spatially concentrates on the rob
vision-language-actionliberorobotwinworld modelbenchmark - arxiv:2611.19554 · cs.AIInquest: Log-Guided Root-Cause Code Localization with a Staged AgentYuvraj Sehgal, Sneh Patel, Mahsa Panahandeh, Naser Ezzati-Jivan
When a production issue is reported, engineers get a description and some logs, but no failing test and no way to reproduce the fault. The logs show where the failure was noticed, not where it was caused: in 49% of the real reports we study, the text names none of the files that the fix later changed. We built a benchmark for this setting from two public datasets and 27 open-source projects (Apache, Red Hat, MongoDB, MariaDB, Qt and Sakai). It holds 38,001 resolved issue reports with their pasted logs, the code as it was while the bug was still present, and the files the fix changed. We present INQUEST, an agent that finds the faulty code using a 27B open-weights model. Code, not the model, controls the procedure. It first gathers the evidence, including the files fixed in similar past issues that were resolved before the report was filed. The model then makes only three bounded decisions: which files to open, what role each file plays in the failure, and how to rank them. On 26,544 ev
agentbenchmark - arxiv:2611.19549 · cs.AIBeyond Prompts: Runtime Contracts for Coding AgentsReda Heddad
Coding agents act on repositories through shell commands and file edits, and teams steer them with written engineering rules. Industrial settings expect such rules to be enforced and their application to be evidenced, and a rule given as prompt text provides neither. We test whether executable contracts enforced while the agent works reduce rule violations compared with the same rules in the prompt, retrieved as guidance, or checked only after generation. Violations are scored by an oracle whose methods differ from the in-loop checks, and the oracle is validated by two human reviewers and a model reviewer. Across three studies on two model families (240, 112 and 108 runs), enforcement during execution produced no oracle-detected violation in 104 paired runs, against 21 to 55% of runs without rules, and no loss of functional correctness in the two confirmatory studies. In the confirmatory study on tasks aimed at known failure situations (Study 3, protocol hashed before the first run), e
agent - arxiv:2611.19542 · cs.LGFunc-JEPA: Learning a Sobolev Space of Functions as Latent VectorsAlejandro Moreno Arcas
Pretraining an encoder once and reusing it across tasks is a widespread approach in vision and language. However, neural symbolic mathematics still trains a dedicated sequence-to-sequence model for each task, so that no representation of a function is shared. In this work, we propose the Function Joint-Embedding Predictive Architecture (Func-JEPA, abbreviated f-JEPA): an encoder of mathematical functions, pretrained on their symbolic form, whose latent space imitates the Sobolev space H^1 and in which differentiation is a learned latent operator. The norm of a latent is computed exactly from the function, whereas its direction is learned to match Sobolev angles. The frozen encoder is then reused by lightweight modules for decoding, composition, integration, and regression from samples. The representation and four downstream tasks are evaluated within and outside the pretraining distribution. f-JEPA recovers exact formulas from noisy samples more often than the compared symbolic-regress
benchmark - arxiv:2611.19536 · cs.AIAgent Authenticator: Passkey-Grade, Hardware-Bound Authentication for AI AgentsHarish Santhanalakshmi Ganesan
AI coding agents now push code, publish packages, call model APIs and operate cloud resources on behalf of developers. They authenticate to these services with the same bearer secrets humans use, personal access tokens, API keys and package-registry tokens, stored where the model, its subprocesses and any co-resident malware can read them. Incidents in 2025 and 2026, including prompt-injection attacks on coding agents and self-propagating supply-chain worms, show that this arrangement fails in predictable ways. Passkeys solved the analogous problem for humans, but they require a human gesture on every use and cannot be operated by an autonomous process. We present the Agent Authenticator, a design that gives agents passkey-grade authentication properties: per-service hardware-bound keys, no shared secrets, origin binding enforced by the runtime rather than the model, normalized attestation of the execution environment, human-signed scoped mandates, and cross-device human approval for h
agentai agent - arxiv:2611.19530 · cs.AILLM Trading Agents Earn Beta, Not Alpha: A Look-Ahead-Free BenchmarkJackson Wang
Large language model (LLM) trading agents now trade stocks for their users, and benchmarks rank models by their agents' returns. Yet a single run's return says little about skill. In our study, two runs of the same model that differed only in the random seed ended 5.2 points apart on average, so telling skill from luck takes many backtests across markets, models and seeds. However, live arenas trade each day only once, and backtests either use the few months after a model's training cutoff or run on years the model may remember, often with news joined to prices by calendar date. In this paper, we present rta-bench-1, a backtest over three US markets (a rally, a bear market and a crash). It shows each headline only after its publication minute, plays each period blind and named, and checks every run for look-ahead. It also splits each return into factor exposure and alpha and scores every decision against random choices of the same day and sector. In 91 runs of six models from four vend
benchmark - arxiv:2611.19527 · cs.CVPathLoop: Self-Evolving Multi-Agent Reasoning for Evidence-Grounded Whole-Slide Image AnalysisJincheng Li, Zhengyi Liu, Zelin Liu, Yihui Zhan +11
Analyzing whole-slide images (WSIs) requires combining evidence across tissue regions and spatial scales. Multi-agent systems offer a promising approach to this task. However, existing systems make limited use of molecular evidence, often select visually prominent but less representative regions while overlooking global context, and struggle to reuse experience from previous cases. To address these challenges, we present PathLoop, a self-evolving multi-agent framework that integrates omics data for evidence-grounded WSI analysis. Specifically, PathLoop coordinates visual and omics interpreters across slide, patch, and pixel levels, using molecular signals to guide region selection and combining local tissue details with global slide features. A judgment agent then checks the consistency of the collected evidence against Pathology-Wiki, a knowledge base built from 48 entries across 24 organ systems, and triggers re-examination when the evidence is inconsistent. To reuse past experience,
memoryagentmulti-agentagent frameworkagent systemself-evolving - arxiv:2611.19525 · cs.AIObjectify, Don't Delete: Complaint-Derived Evaluation of Ambient Clinical NotesNikita Gupta, Narayanan Unny Edakunni, Vikram Okaly, Suranjana Datta +3
Evaluating the quality of AI-generated clinical notes is hard: standard metrics like ROUGE measure lexical similarity to reference notes, while expert checklists encode what committees believe matters. Neither captures what clinicians actually complain about. Recent work has shown that evaluation criteria can be mined from user feedback and filtered to criteria that LLM judges can reliably enforce. We argue that this is where much of the value is lost. The criteria that fail enforceability are disproportionately the ones clinicians raise most often, because high-frequency complaints are stated as several questions glued together. No yes/no judgment can grade such composites as posed. We introduce an objectification step that rewrites such criteria into sets of scoped, determinate decisions. We demonstrate the approach through an evaluation pipeline that mines criteria from complaints, objectifies them, and deploys LLM judges against both planted defects and production notes. Across ele
benchmark - arxiv:2611.19512 · cs.CLSada-ASR: A Production Multidialectal Arabic Speech Recognition SystemMahmoud Khaled, Mohammed Abd-Elnaser, Mahmoud Mohsen, Hazem Ahmed +1
Callers to Arabic customer-service lines speak their own dialect, while speech recognizers are built and evaluated mainly on Modern Standard Arabic (MSA). A recognizer that mishears the dialect breaks the call before anything else can help. We present Sada-ASR, the speech recognizer of EchoNova's production Arabic voice agent. On a 120-hour test set of MSA and eight dialects, it has the lowest word and character error rates of twelve systems, against eleven from Microsoft, ElevenLabs, Google, OpenAI, Speechmatics, IFM, Cohere, NVIDIA and Audar: 11.69% word error rate against 13.51% for the strongest, Microsoft's MAI-Transcribe-2 (a lead of 0.74 points on clips unaffected by that system's fixed number and script conventions), and 15.52% against 16.96% averaged over varieties. Its largest margins are on Omani (4.30 points), Egyptian (3.78) and Saudi (2.19) speech; it is level on Syrian and loses on Lebanese (24.96% against 23.05%), mostly because it writes Lebanese words in MSA forms and
leaderboard - arxiv:2611.19495 · cs.CVVegetation Health Classification for Low-Data and Edge Hardware DeploymentYifei Cheng, Haoyu Li
Unmanned aerial vehicles (UAVs) equipped with cameras are popular, viable tools for vegetation and environmental health monitoring. However, existing image classification methods are incompatible with UAV implementation for independent practitioners in remote, resource-constrained environments due to large dataset requirements, GPU hardware, inflexible retraining, and low interpretability. This paper presents a lightweight, interpretable vegetation classification algorithm operating with as few as five reference images per class, without GPUs or heavy machine learning frameworks. It uses a 2D hue-saturation histogram feature representation and Earth Mover's Distance or Quadratic Form Distance, paired with a log-likelihood probability score, to classify images into three vegetation health categories. We evaluated six systems: three custom pipelines and three established models across low-data scenarios ($n = 5, 10, 25$) and a full-data regime of 2,172 drone images from 28 South American
memory - arxiv:2611.19493 · cs.CVBeyond Paired Inputs: Learning to Leverage RGB-T Cross-Modal Relations in Multimodal Large Language ModelsTianyi Zhao, Qihao Xu, Jiawen Xi, Duanni Meng +4
RGB and thermal infrared (TIR) sensors provide complementary observations, yet their distinct sensing characteristics make effective cross-modal coordination challenging. Although existing multimodal large language models (MLLMs) can process paired RGB-T inputs, they often fail to translate cross-modal relations into effective task-specific decisions, leaving the potential of these relations to improve task performance and robustness largely untapped. Motivated by this observation, we introduce RelMLLM, an RGB-T MLLM that explicitly learns to exploit cross-modal relations in task-specific decision making. It represents modality reliability, the cross-modal relation, and the dominant modality as a verifiable decision state. We derive task-grounded supervision for this state by comparing RGB-only, TIR-only, and paired RGB-T task outcomes. A hierarchical relation-guided reward supervises modality-specific evidence, relation-state prediction, and the final task output, while encouraging co
benchmark - arxiv:2611.19485 · cs.CVPALIN: Pixel-Aligned LiDAR-NIR Multimodal 3D Object Detection with BenchmarkYanan Zhang, Amit Vishnia, Chengbo Ai, Eleni Christofa +5
3D object detection is critical for intelligent transportation systems. LiDAR has been widely adopted for accurate geometric measurements; however, purely geometric representations remain ambiguous under sparse observations and occlusions, while cross-sensor multimodal fusion suffers from calibration and synchronization errors. To overcome these challenges, we leverage an Ouster LiDAR to construct PALIN (Pixel-Aligned LiDAR-NIR), a large-scale intrinsically pixel-aligned multimodal dataset jointly capturing geometry and NIR intensity, comprising approximately 8,000 annotated frames from static urban intersections and diverse mobile suburban scenes. Building upon this dataset, we propose CMambaFusion, a novel fusion framework consisting of a four-directional Cross-Mamba module for bidirectional cross-modal interaction and a Gated Concat-Mamba module for joint feature refinement. Extensive experiments demonstrate that PALIN enables effective multimodal learning, and CMambaFusion achieves
benchmark - arxiv:2611.19475 · cs.CVStyleJudge: A Human-Validated, In-Context-Learning Vision-Language Evaluator for Artistic Style TransferLokesh, Pramukto Mandal, Somanshu Juneja, Jashnoor Singh +5
Automatic evaluation of artistic style transfer remains an open problem: existing metrics score an image pair in isolation, without seeing an example of the intended style transformation, and correlate poorly with human judgment. We introduce StyleJudge, a compact vision-language evaluator conditioned on a small set of reference source-to-stylized pairs before scoring a query pair, mirroring how a human rater calibrates to a style before judging a candidate. On a human-annotated benchmark of 878 rated pairs across 12 artistic styles (728 from 10 training/validation styles, 150 from 2 unseen styles), our efficient 0.8B-parameter StyleJudge reaches a Pearson correlation of 0.291 (Spearman 0.325) with human ratings on seen styles and 0.381 on unseen ones. On seen styles, it achieves the highest positive correlation among tested metrics: Gram-matrix style loss and CLIP similarity correlate weakly ($r=0.041$, $r=0.168$), while LPIPS (scored higher when closer to the source) correlates negat
memorybenchmarkevaluator - arxiv:2611.19471 · cs.LGInterpreting Random Forests: Comparing Surrogate Models and Rule Extraction MethodsLea L. Mairhöfer, Jonathan B. Krause, Björn-Hergen Laabs, Silke Szymczak
Random forests (RFs) often achieve high prediction performance, but their ensemble structure can be difficult to interpret. Surrogate models and rule extraction methods offer interpretable approximations, yet guidance on selecting an appropriate method for a specific research question remains limited. We compared 12 methods for interpreting RFs, with a decision tree as a prediction-oriented reference, across 11 simulation settings and 67 benchmark datasets for regression and classification. Evaluation addressed prediction and structural similarity to the RF, model complexity, prediction performance, and runtime. Simulations additionally assessed explainability using known effect and noise variables. No method performed best across all objectives. For regression, SLIM provided the strongest general performance for prediction similarity or performance on test data, while a linear model offered a simpler global interpretation of linear effects when performance was comparable. For classifi
benchmark - arxiv:2611.19466 · cs.ROSuccessive Convexification in Learned World Models for Safe Offline PlanningSamuel Younger
World models learned from offline data allow an agent to plan without further interaction with the environment, but the sampling-based planners used with them handle safety constraints only by scoring or rejecting sampled action sequences. We instead pose safe planning in the latent space of a learned world model as a successive convex program (SCP), in which a learned discrete-time control barrier function is imposed as a constraint at every step of the horizon. SCP requires smooth dynamics, so we build the networks from smoothed MaxMin activations and spectrally normalised layers, and bound the resulting linearisation error in terms of the Lipschitz bound and the smoothing width. As SCP is a local method, we start it from a sampled plan. On two Safety-Gymnasium navigation tasks with offline data, this planner has fewer violations than MPPI and CEM drawing 128 times as many samples at equal wall-clock time in our main comparison, though not significantly, and earns greater return at l
world modelagent - arxiv:2611.19465 · cs.AIVAST: Compiling Grounded Vulnerability Scenarios into Verifiable, Multi-Step Environments for Evaluating LLM Red-Team AgentsZhen Wan
Evaluating Large Language Model (LLM) security agents requires reproducible environments whose workflow difficulty can be varied and whose failures can be diagnosed. We present VAST (Verifiable Adversarial Scenario Training & Evaluation), a declarative range compiler separating vulnerability scenarios from a compiler-introduced workflow-hardening layer. Following a post-hoc code and AST audit across all topics, the main benchmark retains 109 seed edges (45 with strict upstream CVE annotations), 72 family identifiers (62 under shared-HMAC relay grouping), 81 topic configurations (76 core plus 5 bounded execution models), and 1,387 instances (359 canonical four-gate-revalidated specs plus 1,028 seed variants) in a 42:39 partition. On the TopicServer compiler stratum, a budget-matched 270-episode prompt contrast across 10 compiled mechanism-family clusters (30 P2/P3/P4 instances, 40 turns, temperature 0.0) shows that glm-5.3, grok-4.7, and deepseek-v4-pro solve 12/90 instances under contr
benchmark - arxiv:2611.19461 · cs.AIProgram Distillation: Transferring LLM Capability from System 2 to System 0Pujun Zheng, Wanying Ren, Guoxiu He
Recent studies leverage large language models (LLMs) to construct executable programs through environment interaction, feedback analysis, and code revision. However, prior work has mainly treated LLMs as optimizers, overlooking their ability to transfer knowledge across system levels. We propose Program Distillation (PD), a framework that formalizes the transfer of abilities from System 2 to System 0. During learning, the LLM provides task knowledge and draft strategies; at deployment, a student program executes the resulting policy. We show that this approach can improve performance while reducing inference time. We introduce PDBench, a benchmark comprising 27 discrete- and continuous-control tasks with fixed initial policies, and PDMetaHarness, a model-harness-independent meta-system for monitoring and guiding the learning process. The strongest model achieves a normalized mean score of 0.728, surpassing PPO (0.656) and Jev-like decision models (0.239), while running exclusively on C
benchmark - arxiv:2611.19455 · cs.CLText Speeding Itself Up: Recency and Restraint in Session-Local DraftingAntal van den Bosch
Speculative decoding lets a cheap draft model propose tokens that a large language model verifies in a single forward pass. When the verifier keeps a KV cache, verification is cheap and drafting must be nearly free. Session-local drafters, which predict from the prompt and the text generated so far, meet that bar without training. We compare a recency-weighted, backing-off, confidence-gated session memory against prompt lookup decoding in its reference form, which copies from the earliest earlier match, and in a variant that copies from the most recent one, with paired per-prompt bootstrap intervals throughout. On general, medical, and legal text with a Mistral-7B verifier, recency accounts for the speed: most-recent lookup gains up to 0.09 in speedup over earliest-match lookup on 500-token generations, and session memory's weighting and backoff add nothing measurable on top, at 30 to 40 times the drafting cost. Energy tells a different story. On medical and legal text, unselective dra
memory - arxiv:2611.19448 · cs.AICLEW: Log-Anchored Backward Tracing for Root-Cause Code LocalizationWenji Bai, Muhammad Waseem, Jaakko Peltonen, Pekka Abrahamsson
Failures in industrial software systems may involve multiple components and be difficult to reproduce outside the production environment. Root-cause code localization therefore often relies on issue descriptions and runtime logs. However, log evidence may be noisy or incomplete, and the code reporting a symptom may differ from the code containing the underlying defect. These conditions make it challenging to identify relevant log evidence and determine which functions could have caused the reported failure. To address this gap, we introduce CLEW, an agentic approach that combines rule-based log-to-code anchoring with LLM-guided backward code tracing. CLEW organizes selected observations into diagnostic states, guides source exploration from anchored symptom functions, and returns ranked candidate functions, an affected module, and an explanation comprising a proposed mechanism path and associated log events. For evaluation, we construct LogLoc, a benchmark comprising 192 real-world iss
agenticbenchmark - arxiv:2611.19446 · cs.LGArchitectures and KV Replay for Depth-Recurrent Language ModelsLucia Licakova, Niklas Schmidinger, Lukas Hauzenberger, Sebastian Böck +1
Depth recurrence increases effective model depth by repeatedly applying a fixed computation block. However, sharing attention parameters does not share the key-value (KV) states generated at different recurrence iterations, which causes memory growth with recurrence depth during autoregressive decoding. We study how to reduce this memory cost through architectural and cache-sharing approaches within a common depth-recurrent framework. We first study a looped hybrid that repeats a stack of Transformer layers, with recurrent mLSTM or Gated DeltaNet blocks replacing most attention layers and periodic Transformer attention anchors retained. We then restrict recurrence to a dedicated recurrent workspace in a Parcae-style architecture and test its stabilising adapter as a wrapper for the looped hybrid. Finally, we train a looped Transformer with recurrent KV replay, where only selected recurrence iterations write KV states that later iterations reuse. This trains the model to operate with th
memory - arxiv:2611.19440 · cs.CVFew-Shot Visual Concept Extraction for Steering Diffusion TransformersNabyl Quignon, Antitza Dantcheva
Image steering aims to provide fine-grained control over concepts in generated images, enabling targeted and continuous manipulation without broadly altering the generation process. Recent steering methods for diffusion models primarily focus on the text-conditioning space or rely on costly auxiliary training. Deviating from that, we here propose a novel steering approach that extracts visual concept directions directly from the image-side activations of Diffusion Transformers. Specifically, we showcase few-shot concept extraction in the activation space using a simple difference-of-means estimator over intermediate residuals. We observe (a) scene-level concepts such as lighting, weather, and style are distributed across image tokens, as well as (b) local concepts such as object colors, textures, materials, and facial attributes are concentrated in specific regions. Towards controlling both concepts, we introduce a new unified training-free framework that uses global difference-of-mean
manipulation - arxiv:2611.19439 · cs.AIDo Rule-Based Tools Help LLMs Translate Endangered Uralic Languages in the Prompt or over MCP?Iaroslav Chelombitko, Mika Hämäläinen
Komi-Zyrian, Moksha, Erzya and Udmurt have mature rule-based resources, among them finite-state analysers, constraint grammars and bilingual dictionaries, which language models do not use. We ask whether giving a model access to these resources improves machine translation from the four languages into English, and whether it matters how the access is provided: as analyses computed in advance and written into the prompt, or as tools that the model calls itself through the Model Context Protocol (MCP). We evaluate manually: an expert linguist read every English output of the strongest model under three conditions, 816 sentences in all, accepted or rejected each one, and named the main error in every rejected translation. With the analyses in the prompt, the number of accepted translations rises from 13 to 43 out of 272; with the tools attached it rises to 55. The gain comes from the bilingual dictionary. For Udmurt, whose dictionary has no English side, the prompt-based condition brings
tool calling - arxiv:2611.19418 · cs.CLReal-World End-to-End Product Entity ResolutionDimitrios Karapiperis, George Papadakis, Manolis Koubarakis
Product Entity Resolution (PER) in e-commerce is challenging due to unstructured text, formatting inconsistencies, promotional noise, and fine-grained product differences. We present an end-to-end evaluation of four zero-shot PER pipelines---Lexical Cascade, Semantic Cascade, Hybrid Fusion Cascade, and Generative-Lexical Cascade---across established benchmarks and introduce two real-world datasets characterized by cross-lingual noise and dense brand collisions. We evaluate lexical and dense blocking with cross-encoder, tensor-search, and Large Language Model verification, considering accuracy, runtime, and memory consumption.
memorybenchmark - arxiv:2611.19413 · cs.LGPMR: MoE Routing with Shared Prototypes and Congestion PricingWending Wang, Binqi Huang, Ju'an Li, Yu Guo +3
Conventional linear routers in sparse mixture-of-experts models select experts directly from token-expert matching scores, coupling input matching with expert allocation. We propose Prototype-Mediated Routing (PMR), a two-stage router with shared prototypes. PMR first computes each token's distribution over prototypes, then uses it to mix prototype-conditioned expert distributions according to the law of total probability. This formulation models input matching and expert allocation separately, explicitly represents many-to-many associations between prototypes and experts, and provides separate interfaces for context adaptation and load regulation. Within this framework, short-term states summarize preceding tokens to adapt the prototypes, while congestion prices updated from dispatch feedback achieve loss-free load balancing by adjusting the conditional expert distributions. We use a two-pass blockwise approximation to accelerate short-term state processing, achieving over 20 times th
benchmark - arxiv:2611.19394 · cs.CVDetecting Small Space Objects for Space Situational Awareness with Collaborative Edge AIXingyu Qu, Peng Hu
With the rapid growth of space activities, near-Earth space is becoming increasingly congested, leading to a higher risk of spacecraft collisions and space debris generation. Recent close-proximity encounters and collision incidents necessitate efficient and cost-effective onboard systems for space object detection (SOD) and collision avoidance. While deep learning (DL) enabled Edge AI using low-cost vision sensors on single or multiple satellites has shown promises for SOD, significant challenges remain, including limited datasets, small SOD, and the lack of comprehensive performance assessments. This paper presents a collaborative Edge AI framework for multiview SOD and systematically benchmarks DL models and fusion strategies. Results show that higher image resolution significantly improves small SOD. For multiple views, image-level fusion reduces the combined communication and inference latency by 30.13% and 32.80% on average compared with feature-level and attention-based fusion a
benchmark - arxiv:2611.19388 · cs.AISARGO: Agreement-Gated Reinforcement and Suppression via On-Policy Self-Guidance for Visual Hallucination MitigationJinyu Li, Zhiming Zhang, Hao Fang, Hongyao Yu +3
Vision-language models remain susceptible to visual hallucinations, often mixing grounded information with unsupported claims in the same response. Response-level reinforcement learning assigns a shared advantage to all tokens in a response, causing negative updates to suppress correct information alongside unsupported claims. To reduce hallucination while preserving useful visual information, we introduce SARGO (Self-guided Advantage-Routed Grounding Optimization), a post-training framework for selective reinforcement and suppression through on-policy self-guidance. Direct and reciprocal weighting gives better-supported tokens stronger reinforcement on positive updates and weaker penalties on negative updates. Agreement between overall and hallucination-specific advantages selects when to apply this guidance. At the response level, a reference-calibrated reward couples factuality with the retention and expansion of grounded information, discouraging lower hallucination scores obtained
post-training - arxiv:2611.19387 · cs.ROLong-Horizon Causal Inertial Odometry in Multistory BuildingsTianzi Zhang, Yan Wang, Hao Qiao, Jian Kuang +1
In indoor 3D localization inside multistory buildings, an elevator or an escalator accelerates for a few seconds and then cruises at constant velocity for tens of seconds. Gait excitation disappears, the specific force degenerates into a constant gravity reaction, and instantaneous velocity becomes unobservable within a 1-2 s window. The output of short-window learned inertial odometry consequently degrades toward the conditional mean of the training set, and vertical displacement is systematically underestimated. We propose Local-Token Causal Attention (LTCA), a long-horizon architecture that compresses each 1 s inertial window into a 1 Hz motion token and sustains, over a bounded 60 s causal cache, the motion state established at the onset event. We further introduce a pluggable barometer branch that, when online, supplies a direct elevation constraint to the vertical channel through local differential height observations, and we train it with paired online/fallback conditional train
benchmark - arxiv:2611.19385 · cs.AIDoes a completed simulation run mean a usable urban wind result for AI agents?Zhongke Qu, Yunwei Zhang, Junwei Su, Yabin Zhang +4
AI agents based on large language models (LLMs) can now configure and run computational fluid dynamics (CFD) solvers, but a completed run does not ensure that delivered metrics can be recomputed from raw outputs. This gap was examined for urban wind simulation, where multi-case comparisons must hold non-target conditions fixed and reported values must be traceable to raw samples. UrbanWind-AgentBench, comprising 200 tasks and 367 cases in 10 engineering task families, was constructed. Public task contracts specify variable factors, fixed conditions, numerical thresholds and deliverable requirements, and two-layer scoring separates a structural score for deliverables, run provenance and raw evidence from a numerical score obtained by independent recomputation against references. WindPilot was then proposed as an execution framework that incorporates contract parsing, execution provenance, deterministic verification and bounded acceptance review into the LLM execution loop. On 30 OpenFOA
agentai agent - arxiv:2611.19384 · physics.opticsPorosification of the dielectric layer in alternating-current electroluminescence devices enables high-resolution tactile imagingMingwei Gu, Licheng Li, Zeyuan Zhou, Ying Yang +9
Vision-based tactile sensing enables high-resolution contact perception, yet bulky optical architectures consume valuable space and constrain the functional design of dexterous robotic hands. Here we introduce a self-emissive tactile imaging strategy based on alternating-current electroluminescence (ACEL) devices, in which a compressible porous dielectric encodes contact pressure into spatial emission patterns without external illumination or discrete sensing pixels. We identify a nonlinear capacitance-dependent emission window that provides a basis for tunable pressure-to-light conversion. Pore-size engineering further suppresses local emission deficiencies, reducing spatial intensity variation by nearly ninefold and enabling faithful imaging of contact features down to 100 μm. The continuous electroluminescent interface also reproduces complex contact patterns without computational reconstruction of electrical fields. Direct lamination of the ACEL device onto a complementary metal-ox
dexteroustactile - arxiv:2611.19375 · cs.CLCausalGloss: Multi-Evidence Action Control for a Simultaneous Mandarin Speech-to-ASL Gloss CascadeMingze Li, Satoshi Nakamura
Simultaneous Mandarin speech-to-American Sign Language (ASL) gloss translation requires deciding when an evolving utterance supports committing each gloss. Fixed Wait-k cannot adapt to source-gloss reordering, while training only on reference histories causes exposure bias. We present CausalGloss, an English-prefix-to-ASL-gloss action policy for a Mandarin-speech cascade. It combines lexical and semantic evidence, bounded delay, negation constraints, burst WRITE actions, and incomplete-history recovery. On a frozen 500-example How2Sign-derived benchmark with weak LLM-generated references, CausalGloss improves the fixed Wait-k frontier and nearly matches a three-seed TransLLaMA-style learned adaptive baseline at k = 5 in Token F1 (0.7047 versus 0.7067), with 54.4% lower LAAL. Against CLERC Epee human-gloss references, it achieves 0.649 Token F1 and Pareto-dominates three-seed TransLLaMA-style k = 5 and fixed k = 5, 7, reducing LAAL by 58.2-69.9%. Ablations separate timing from retention
benchmark - arxiv:2611.19364 · cs.CVXLSAR-Bench: Towards Extremely Large SAR Imagery Understanding of MLLMsLa Jiang, Lingfeng Chen, Jiaqi Zhu, Fengrong Guo +4
The rapid development of multimodal large language models (MLLMs) has highlighted the need for comprehensive benchmarks to assess their capabilities and limitations in remote sensing (RS). However, evaluating MLLMs in large-size synthetic aperture radar (SAR) scenarios remains challenging, as understanding complete scenes requires both global semantic interpretation and fine-grained target perception. Existing SAR benchmarks predominantly rely on small image patches, limiting their ability to assess these capabilities jointly. To address these limitations, we present XLSAR-Bench, a comprehensive benchmark for evaluating the perception and reasoning capabilities of MLLMs in large-scene SAR imagery. XLSAR-Bench comprises 405 images with an average size of 7153$\times$7153 pixels, ten tasks spanning visual question answering, visual grounding, and detailed image captioning, with all evaluation samples meticulously annotated manually, assisted by a semi-automatic captioner. Experimental re
benchmark - arxiv:2611.19359 · cs.AIOne Run Is Not Enough: Reminding a small Gemma 4 coding agent of its last finding made no measurable difference in a paired testPatrick Passos
When a coding agent runs out of room, its earlier findings are cut and it forgets them. We tested a simple fix: automatically putting the agent's last finding back in front of it. We fixed the rules in advance and ran 98 paired coding tasks with and without this reminder, using Gemma 4 31B (QAT, 4-bit) in the Gemma 4 Developer Agent harness. The reminder made no measurable difference: across all 98 tasks, 47 were solved with it and 48 without (-1.0 points, 95% interval [-8.3, +6.3], exact McNemar p=1.0). The primary outcome (a documented deviation from our preregistration), the first 48 pairs, was 29/48 with memory against 25/48 without (+8.3 points, [-2.1, +18.3], p=0.22); the extension of 50 tasks, decided after that result, pointed the other way (18/50 against 23/50, -10.0 points, p=0.125), so the sign did not replicate. With 19% of tasks flipping between two identical no-memory runs, the combined interval excludes harms larger than about 8 points and gains larger than about 6, and
memoryagent - arxiv:2611.19358 · cs.CVRETR: A Structure-Preserving RGB-Event Transformer for Robust 3D Lane DetectionJingtao Dong, Hao Zhuang, Hao Yang, Liyuan Pan +1
Robust 3D lane detection requires accurate metric road geometry recovery from visual inputs, yet conventional RGB-based approaches suffer from fragile lane evidence under adverse illumination, motion blur, and long-range perspective compression. Event cameras offer complementary high-temporal-resolution, high dynamic-range structural cues that are robust to these challenges, but their sparse, motion-dependent responses cannot be directly fused into consistent 3D geometry. In this paper, we present the first attempt to introduce event cameras into 3D lane detection. To enable systematic research, we first build two multimodal benchmarks with metric 3D lane annotations: DSEC-3DLD (real-world sequences) and Ev-OpenLane (large-scale simulated sequences). We further propose RETR, a structure preserving transformer that converts complementary RGB and event observations into coherent 3D lane geometry. RETR first aligns modality consistent evidence through Reciprocal Context Flow Fusion, then
benchmarkevent camera - arxiv:2611.19356 · cs.LGA Comparative Evaluation of BERT-Based and LLM-Based Models for Indic Sentence SimilarityAishwarya Gunjal, Kanchan Khemnar, Pawan Poojary, Pranav Kulkarni +1
The relative effectiveness of BERT-based encoders and large language model (LLM)-based embeddings for Indic semantic textual similarity remains insufficiently characterized. We compare eight BERT-based and five LLM-based models on the 1,500-pair STS Benchmark development set translated into ten Indic languages and on the native Marathi MahaSTS test set. IndicSBERT achieves the highest Pearson correlation in all ten translated language sets, with a macro-average of 0.843, followed by BGE-M3 at 0.792. Harrier-270M is the strongest LLM-based model on translated STS-B, with a macro-average of 0.751. On MahaSTS, IndicSBERT and Bhasha-Embed-v0 achieve Pearson correlations of 0.901 and 0.900, respectively, while EmbeddingGemma leads the LLM-based models at 0.871. The highest-scoring BERT-based model thus exceeds the highest-scoring LLM-based model in both settings, although the magnitude of the difference varies across benchmarks. These findings indicate that LLM-derived embeddings do not uni
benchmark - arxiv:2611.19349 · cs.LGSMELL: Sparsity-aware Memory Efficient Federated Fine-tuning for Long-context LLMsYin Xu, Yongbo Yang, He Sun, Miao Sun +1
Federated fine-tuning enables Large Language Models (LLMs) to collaboratively adapt to distributed private data. However, long-context fine-tuning on resource-constrained devices faces substantial challenges arising from excessive activation memory, heterogeneous token-importance patterns across clients, and communication overhead. To address these challenges, we propose SMELL, a Sparsity-aware and Memory-Efficient federated fine-tuning framework for Long-context LLMs. SMELL integrates three complementary mechanisms: Local Sparse-Forward Adaptation (LSFA), Consensus-Anchored Sparsity Coordination (CASC), and Communication-Efficient Progressive Update Recovery (CPUR). These mechanisms jointly enable memory-efficient local adaptation, heterogeneity-aware sparsity coordination, and lightweight model-update transmission. Experiments on OPT-350M and LLaMA2-7B demonstrate that SMELL reduces peak memory consumption by up to 88.35% compared with conventional fine-tuning and 13.25% compared wit
memorylong-context - arxiv:2611.19345 · cs.LGInformation Limits of Replay-Free Recurrent State MigrationKhizar Qureshi
When a recurrent model is replaced, its old state may not determine what the new model would compute from the same history. We characterize the information that must be retained before the replacement is known. For a time-invariant affine linear family, the minimum continuous causal memory for order-$k$ local migration equals the reachable dimension of a lifted coefficient system minus the information in the anchor state. A controllable $d$-state model with scalar input needs at most $kd$ additional real coordinates, even when arbitrary matrix replacement directions are revealed only after streaming. A single rank-one direction can already require all $kd$. Conditional singular values and auxiliary-bit bounds distinguish this local information law from recovery at a fixed tolerance. We also give invariant-subspace and future-output constructions. Controlled experiments compare the resulting representations. In a small language model, equal-memory cubic interpolation improves mean state
memory - arxiv:2611.19344 · cs.LGDelay-Aware Reinforcement Learning for O-RAN Control under Stochastic Reward FeedbackXingqi Wu, Junaid Farooq, Tao Li, Juntao Chen +1
Reinforcement learning (RL) is increasingly employed for real-time control in Open Radio Access Networks (O-RANs). However, the asynchronous nature of O-RAN feedback introduces stochastic delays between actions and their observable rewards, leading to misaligned credit assignment and unstable training. This letter formulates the O-RAN control problem as a Markov decision process (MDP) with delayed and stochastic rewards, where the true reward corresponding to an action is received after an unknown delay. To address this, we propose a delay-aware deep RL framework that integrates a long short-term memory (LSTM)-based delay predictor and a temporal reward realignment module within an off-policy learning pipeline. Using empirical delay traces collected from the Open AI Cellular (OAIC) testbed, we demonstrate that the proposed framework achieves faster convergence and significantly reduces latency violations and packet drops compared to conventional RL methods that ignore feedback delay. T
memory - arxiv:2611.19332 · cs.AINI-ACT: A Neural Intent Interface for Brain-Embodied ManipulationYinghao Li, WeiHsien Lee, Zeyu Zhang, Canyang Zhao +3
Intracortical brain-computer interfaces can drive robotic arms, but they keep the user inside the fast sensorimotor loop: every approach, correction, and grasp must be neurally specified. Whether intracortical population activity can instead serve as a task-level condition for a learned embodied policy, without symbolic commands or hand-designed skill libraries, remains unknown. NI-ACT addresses this question with a hierarchical brain-embodied framework: a brief motor-cortical signal is read once as a persistent intent, and a visuomotor policy then closes the perception-action loop from vision and proprioception. In a counterfactual grasping task where the scene cannot specify the target, the neural intent alone determined which object the robot grasped; replacing the intent changed the behavior accordingly. Passing the full population latent as a dedicated intent token, rather than a one-hot decoded direction, raised grasp completion from 69% to 81% over 132 held-out scenes, whereas a
embodiedmanipulationgraspembodied agent - arxiv:2611.19330 · cs.LGBenchmarking the State of AI for Crop Yield Forecasting: A Global Assessment Across Modeling ParadigmsMichiel Kallenberg, Philip Janz, Christoph Jörges, Vageesh Saxena +3
Accurate crop yield forecasting is essential for food security, agricultural policy, and commodity markets. Although statistical, process-based, and machine learning approaches have all been proposed, their relative strengths remain poorly understood due to the lack of large-scale, standardized comparisons. We address this gap through a comprehensive benchmark of five modeling paradigms -- statistical baselines, process-based models, conventional feature-engineered machine learning, deep sequence models, and tabular foundation models -- using the CY-Bench benchmark spanning 63 country-crop datasets (>12,000 administrative regions) across six continents. Beyond conventional pooled accuracy metrics, we introduce a multi-dimensional evaluation framework that separately quantifies overall, spatial, temporal, and anomaly prediction skill. Across multiple crops and countries, feature-engineered machine learning and tabular foundation models achieve the strongest overall predictive performanc
benchmarkevaluation framework - arxiv:2611.19329 · cs.LGGradient-Based Co-Design of Systems and Reinforcement Learning Policies with Differentiable SimulatorsAkhil S. Anand, Xun Shen, Marc Peter Deisenroth, Sebastien Gros
Co-design is the problem of jointly optimizing the design of a system and the policy that operates it. Neither can be chosen independently: a system design is only as good as the policy trained for it, and the optimal policy changes with the design. When that policy is learned by reinforcement learning (RL), evaluating a candidate design requires a full policy training run, and the design is usually optimized by black-box search such as Bayesian optimization (BO) or evolution strategies (ES). These methods scale poorly with the design dimension and ignore the differentiable structure of modern simulators and, in particular, the sensitivity of the optimal policy with respect to the design. We instead derive a gradient-based co-design method, where the design gradient decomposes into a direct effect of changing the design at a fixed policy and an indirect effect through the design-dependent optimal policy. The indirect effect requires the sensitivity of the converged policy to the design
benchmark - arxiv:2611.19324 · cs.ROMARS: Dynamic Routing of Reinforcement-Learned Atomic Skills for Multi-Stage Robotic ManipulationZequan Zeng, Junqin Yang, Kai Wang
Vision-Language-Action (VLA) models have demonstrated strong capabilities in multi-task robotic manipulation. However, they still face limitations in control precision and execution stability during critical manipulation stages, such as object grasping, alignment and insertion. To combine general-purpose manipulation capabilities with task-specific fine-grained control, while enabling effective coordination between generalist and specialized policies throughout multi-stage manipulation tasks, this paper proposes MARS (Modular Atomic-skill Routing System), a dynamic routing framework for reinforcement learning-based atomic skills. MARS integrates a frozen generalist VLA policy with multiple independently trained atomic skill policies. Using a shared RL Token state representation and the reference action of VLA, specialized control policies are optimized through reinforcement learning for different cr`itical manipulation stages. For high-level policy selection, we develop a dynamic route
vision-language-actionvlavla policymanipulationliberograsp - arxiv:2611.19323 · cs.AIThe Data Debt of Stockouts: Partial Identification and Minimax-Regret Inventory ControlJinbiao Chen, Shuang Jin, Hanzhang Qin
Problem definition: Inventory decisions select which demand enters sales logs. Unrecorded forecasts can leave demand frequencies unidentified even when stocking levels are recorded. We study which multistage ordering decisions these logs can justify and what finite samples can guarantee. Methodology/results: Nested stockout thresholds yield exact sets of compatible demand laws and finite representations for policy evaluation. We choose one deterministic inventory rule to minimize its largest expected cost gap from the informed optimum under the same law. This population value is data debt for the specified information and policy class. Under independent demand, carryover can create positive debt despite common optimal actions in every isolated period. Finite samples add learning error above debt. The required sample count grows linearly with the selection bound, with upper and lower orders matching up to logarithmic factors. This rate requires fixed demand and cost bounds, sufficient r
policy evaluation - arxiv:2611.19319 · cs.AIHardening Agentic Web Populations Through Multi-Criterion Competitive Co-Evolution: Lessons from Wargame Strategy OptimizationShashank Raj, Kalyanmoy Deb
Agents on the open web will face adversaries that keep adapting, so a defense that is only tested once will eventually be found and broken. We propose treating jailbreak attacks and safety guardrails as two populations that evolve against each other, rather than as a fixed attacker probing a fixed defense. This idea comes directly from wargame strategy optimization, where two-population competitive co-evolution with alternating evolution cycles and cross-play evaluation has already been shown to produce strategies that are robust to a moving opponent, and where a companion line of work has shown that the same machinery can measure how much competitive pressure a population must absorb. However, real-world applications can never rely on a single criterion as a key performance indicator, rather must use multiple conflicting criteria, such as, cost and efficiency. We describe how this mechanism transfers to language model safety, argue why the resulting multi-criteria, competitive and coe
agentic - arxiv:2611.19318 · cs.ROResidual Reinforcement Learning for Visual Servo Control with Zero-Shot Sim-to-Real TransferMahdi Nobar, Alisa Rupenyan, John Lygeros, Jürg Keller
We address tracking a moving object on a conveyor with a Franka Emika Panda robot arm, despite an imprecise digital twin, measurement errors, and noise. Our novel approach augments a loop-shaped inverse-Jacobian PI controller with a residual policy learned via Soft Actor-Critic in the digital twin. A delay-compensated Kalman filter fuses out-of-sequence camera measurements to provide latency-aware moving target position estimates. We compare the achieved performance to a performance bound for a class of model-based controllers derived from a frequency-domain sensitivity analysis, quantifying the minimum achievable error imposed by uncertain kinematic model mismatch. Simulation and hardware experiments indicate that our residual policy drives tracking error below the bound. Across conveyor speeds (27.9-41.8 mm/s), our method reduces mean tracking errors from 1.1-1.4 mm (baseline inverse-Jacobian PI) to 0.3-0.7 mm. These results show that learning residual corrections in a limited-fideli
sim-to-realfranka - arxiv:2611.19311 · cs.LGMacro Control and Capability Evaluation of Black-Box World Models: A Two-Regime Framework for LLM-Agent MarketsZhiyang Niu, Chuanjie Li, Huayuan Shen, Mengmeng Zhao +1
Large language model (LLM) agents can be organized into multi-agent markets that act as world models, mapping macro information to emergent macro dynamics while the evaluator observes only macro inputs and outputs. This paper separates a model's capability into two stages that a single aggregate score cannot separate: information simulation, from macro information to micro behavior, and macro emergence, from micro behavior to the macro outcome. We introduce a soft macro-control method that announces a public reference path day by day without touching individual decisions, orders, or matching, and a two-regime evaluation framework that measures both stages from macro observations alone. In the reference-only regime the symmetric true and flipped references share a common level R, which measures realization under the announced reference, while their paired gap B measures structural bias; in the information-injection regime external information dilutes the reference, so only the true-flip
world modelmulti-agentevaluatorevaluation framework - arxiv:2611.19299 · cs.LGStaged Semantic Distillation: Preserving Free-Form Semantics in a Low-Compute Non-LLM Surrogate of a Staged LLM AgentZhiyang Niu, Huayuan Shen, Chuanjie Li, Mengmeng Zhao +1
Staged LLM agents make market and social simulators realistic, but one 102-day run of a 200-agent, ten-index market simulator issues 203,100 language-model calls and returns one realization. Conventional low-compute distillation drops their free-form intermediate state. We present staged semantic distillation, a low-compute method that preserves non-structured semantics in a non-LLM student trained from one recorded trace. A frozen encoder maps each teacher output to a 1024-d vector; the student sees typed stage-tagged vectors with stage-specific visibility, is jointly supervised, and trains with scheduled self-history. One TwinMarket trace yields a 13.2M-parameter deterministic student that trains in 1,658 seconds on one 8-GB GPU and runs with zero teacher or decoder calls and no text generation (0.14 ms per decision). Its analysis stays close to the teacher (cosine 0.943), and swapping semantic inputs between agents moves 64.7% of decisions against 12.8% for a matched donor, so the c
agentllm agent - arxiv:2611.19298 · cs.LGNetBreaker: Benchmarking Network Dismantling across Objectives, Scales, and ToplogoiesWenchuan Yang, Tianci Bu, Suoyi Tan, Changjun Fan +2
Network dismantling seeks to fragment a network with as few node removals as possible, which is widely applied in vulnerability analysis, epidemic containment, and infrastructure protection. Although learning-based methods have become increasingly prominent, their progress remains difficult to assess because existing studies differ in method coverage, datasets, evaluation criteria, and experimental protocols. Thus, we introduce NetBreaker, a comprehensive benchmark designed for the learning era of network dismantling. NetBreaker organizes 30 methods (18 non-learning and 12 learning-based) under a common taxonomy and evaluates them on 185 networks from eight unified domains, ranging from 14 to 1.69 million nodes. To enable fair comparison, heterogeneous method outputs are converted into node removal sequences and assessed under a unified execution protocol. The evaluation covers four complementary dimensions: target-based effectiveness, trajectory-based effectiveness, structural diagnos
benchmark - arxiv:2611.19297 · cs.LGWhere Mixture Recognition Fails: An Ingredient-Level Audit of Unseen Combinations in SmellNetZirui Liu, Zheming Zhu
An electronic nose must name every ingredient of a mixture, including combinations it has never measured. SmellNet, a public benchmark for this task, scores ingredients with a hit rate that is told how many each sample contains. We audit what this protocol hides. We hold out ingredient combinations from the SmellNet training data, evaluate four architectures (MLP, CNN, LSTM and a Transformer encoder) without revealing the ingredient count, and repeat the evaluation once on the official unseen split. Under the count-informed protocol, 90-96% of mixtures receive an ingredient that is absent, on our folds and on the official split alike. On our folds, every model identifies one ingredient reliably but the second rarely, and it completes the pair with a familiar training partner about twice as often as chance. On the official split, even the first-ranked ingredient is correct in only 29-46% of binary mixtures, often because a frequent training ingredient that is absent from every test pair
benchmark - arxiv:2611.19292 · cs.AIZero-Shot Weight-Space Programming for Prediction and Classification of Target Variables in Structured DatasetsEduardo C. Garrido-Merchán
Weight-space programming fixes every parameter of a feed-forward network by hand, with no gradient descent anywhere in the construction. Every demonstration so far starts from a target function handed to the model as a rule system, making it a compilation problem. A supervised dataset withholds that specification, so the procedure must do the statistical inference itself and then express the result in the parameterisation it was given. We ask whether a large language model can do both. It is shown a small dataset in full, with the architecture of a feed-forward network, and writes out every weight and bias as a literal number, with no optimizer, no gradient, no code execution, and no chance to evaluate it first. Each construction is sealed and pre-registers its expected score. The graded suite runs from two toy problems, through nine synthetic problems disguised by a hidden rotation, to the classical benchmarks and one real engineering regression. The frozen network matches the closed-
benchmark - arxiv:2611.19285 · cs.CVGenerate, Align, and Restore: MLLM-Empowered Diffusion for Generalizable Low-light Image EnhancementKaisheng Pang, Feng Li, Chen Zhang, Junjun Jiang +3
Low-light image enhancement (LLIE) is essential to numerous vision applications operating under adverse illumination. Despite substantial advances, real-world nighttime enhancement remains challenging due to the scarcity of strictly aligned day-night pairs and complex degradations involving extreme darkness and heterogeneous artificial light sources, leading to fragile generalization. To overcome these challenges, we propose MrLIE (MLLM-empowered reference Low-light Image Enhancement), a novel reference-based unsupervised LLIE framework that synergizes the generative and reasoning capabilities of multimodal large language models (MLLMs) with conditional diffusion. MrLIE employs the MLLM as a high-fidelity reference synthesizer and diagnostic critic, iteratively generating, diagnosing, and correcting normal-light references directly from degraded nighttime observations. Based on this, our method governs the enhancement process through two complementary mechanisms, in which a confidence
benchmark - arxiv:2611.19284 · cs.CVLC-ISP Benchmark: Learning Advanced ISP with Live and Capture StreamsKai Chen, Junyu Lou, Jinhua Zhang, Weiyi You +1
High-quality paired RAW-to-sRGB datasets are essential for data-driven ISP, yet existing datasets often rely on cross-device collection, causing misalignment and failing to faithfully reflect real capture conditions. We therefore introduce LC-ISP, a large-scale benchmark built from the intrinsic gap between a camera's live preview and offline capture streams. LC-ISP contains 53.9k spatially aligned groups, each comprising live RAW/live sRGB and multi-frame fused RGB/capture sRGB, and covers day, night, indoor, outdoor, and HDR scenes. It also provides capture-stream intermediates for modular supervision. We further develop a compact two-stage baseline that uses these intermediates as auxiliary targets. Under a 2.5-GFLOP budget for 448$\times$448 RAW inputs, it outperforms the strongest single-frame baseline by over 2 dB PSNR while improving texture, denoising, and color reproduction. LC-ISP supports realistic benchmarking, the design of supervision strategies, and the development of li
benchmark - arxiv:2611.19275 · cs.AIBeyond Clumsy Agentic Tool Calling: How Cueing and Spotting Make the DifferenceXindi Yang, Yan Liu, Jiajun Sun
LLM-based task-oriented agents depend on tools to interact with their environments, but may overlook execution feedback, repeat ineffective calls, or persist with unproductive strategies. These behaviors can hinder agents' ability to use tools autonomously and sustain task progress. We characterize tool-use symptoms at two temporal granularities: local clumsy tool-use behaviors and tool-use detours spanning multiple interactions. Towards greater autonomy in tool use, we present ToolPulse, an online middleware supporting autonomous tool use through cueing and spotting. ToolPulse tracks execution state and temporal interaction patterns, monitoring proposed calls before dispatch and inspecting execution feedback prior to the agent resumes reasoning. It combines state-based rules, semantic sequence analysis, and interaction-feature similarity to identify intervention opportunities and provide verbal reminders, structured feedback, or context-specific advice from auxiliary models. Our proce
agentautonomous agentagentictool usetool-usetool calling - arxiv:2611.19274 · cs.ROHospitalBench: Diagnosing Where and Why Embodied Agents Fail on Long Hospital Task ChainsGaytri Jena, Kapil Wanaskar, Aman Chadha, Vasu Sharma +2
Long-horizon embodied tasks fail in ways that end-task success cannot see. HospitalBench is a diagnostic benchmark of ten hospital chores in NVIDIA Isaac Sim, each two unscored warm-up steps followed by a causal chain of five scored sub-tasks in one continuous simulation: the state one step leaves (an object pose, a choice, a fact seen once) determines what the next step must do, unlike benchmarks that place sub-tasks side by side. Vision-language agents act through skill-level tools from camera images. A scripted robot that makes only the calls a model gets is the solvability check of a chore. From each trajectory the benchmark reads where a chain breaks, what the agent did after a failed call, and how far its own position estimate drifts from the true one. This version reports first results only: 15 of 60 planned runs (three models at high effort, one run each on chores T1 and T10 from both start states and T3 from one), with scripted-robot proofs on three of the ten chores. Of the 1
embodiedagentembodied agentbenchmark - arxiv:2611.19272 · cs.RODistributed Shaping of Robot Swarms Through Spatial Moments of Arbitrary OrderJin Chen, Hector Garcia de Marina
This paper shapes a swarm into a target pattern, such as a figure taken from an image, by controlling its spatial moments. The image is sampled into a reference configuration whose centroid and covariance set the position and principal spreads of the formation, while its higher-order moments, expressed in covariance-normalized (whitened) coordinates, encode the remaining shape. In contrast with geometric formation control, which prescribes distances, bearings, or relative positions between specific agents, this specification concerns the swarm as a whole and assigns no agent to a particular place. The key mechanism is orthogonalization in the swarm's own empirical inner product: projecting negative-gradient velocities onto the complement of low-degree polynomials sampled at the agents makes every higher-order correction preserve all lower-order moments exactly. The resulting hierarchical controller has lower triangular error dynamics and converges locally exponentially for any positive
agent - arxiv:2611.19254 · physics.opticsBringing coherent closer to compute: coherent-lite multidimensional on-chip interconnect beyond 100 terabits per secondYongzhu Hu, Aolong Sun, Huayuan Qin, An Yan +14
The continued scaling of artificial intelligence (AI) demands optical interconnects that deliver higher bandwidth density within stringent electrical-bandwidth and energy budgets. Coherent optics offers a route beyond the per-lane bandwidth constraints of intensity-modulation direct-detection links, yet its receiver complexity and power overhead have largely confined its use to longer reaches. Here we bring coherent optics closer to compute by demonstrating a coherent-lite multidimensional on-chip interconnect that combines wavelength- and mode-division multiplexing with self-homodyne detection. The architecture co-transmits the local oscillator (LO) and parallel data channels within a single silicon waveguide, using polarization-division routing to preserve LO purity and support high-order modulation. Based on this principle, we develop an inverse-designed dual-polarization mode multiplexer that provides over 37 dB inter-polarization isolation across the C band in an $8 \times 6~μ$m$^
optical interconnect - arxiv:2611.19252 · eess.SYExploration as Feedback in Model Predictive Path Integral ControlJunsei Ito, Kohei Honda, Yasuaki Wasa
This paper investigates how exploration affects closed-loop stability in model predictive path integral (MPPI) control. With a warm start, MPPI optimizes a control sequence by sampling candidates around the previous solution, and their covariance, the exploration covariance, is typically tuned to balance solution quality and computational cost. Small exploration is attractive because most samples then contribute to the solution, and existing stability analyses treat reduced exploration only as a bias toward the previous solution. Whether insufficient exploration can destabilize the closed loop has therefore remained open. We show that, for linear quadratic (LQ) problems with infinitely many samples, the exploration covariance also determines the feedback gain, i.e., how strongly MPPI updates the previous solution in response to the current state. With small exploration at a fixed temperature, MPPI nearly repeats the previous solution and fails to stabilize an exponentially unstable pla
quadruped - arxiv:2611.19251 · cs.ROPhysics-Informed Causal Reasoning for Nonstationary Robot DynamicsJungjin Lee, Se Jin Kim, Sung Kwon On, Insoon Yang
Reasoning about what an action will cause requires internalizing how a system behaves. As conditions seldom stay fixed, this must often happen within a few interactions. Yet systems that look identical in a snapshot can respond differently to the same action, depending on physical conditions the snapshot hides. Context-based models capture this separation by summarizing recent transitions into a latent context, but do not specify how that context should change the equations of motion. Physics-informed neural networks build those equations into the model, but describe only a single system and cannot recognize when it has changed. We present Physics-Informed Causal Reasoning (PICR), which bridges the two. Only a context, re-inferred from recent interactions, is updated; it conditions a fixed physics-informed network that determines the robot's inertia and potential energy, so the robot adapts to new physical conditions from its own experience without retraining or a hand-specified robot
franka - arxiv:2611.19247 · cs.CVGameWM: Toward Next-Generation Video Games with Interaction-Centric World ModelsBing He, Runze Wang, Fei Zha, Zhongyuan Wu +4
For more than half a century, video games have pursued better graphics, more natural interaction, and more engaging gameplay. Generative video models (VMs) now bear on all three. Neural networks already draw most of the pixels in shipped games, and recent world models let a lightweight game engine keep the world state while a VM turns the engine's coarse proxy video into final frames. These systems, however, treat the VM as a renderer that only restyles motion the engine has already authored, which leaves most of its generative capacity unused and limits the diversity a game can offer. We present GameWM, a design for next-generation games built on a clear division of labor. The GameWM-engine keeps what must be exact: rules, state, camera, and the layout of the world, which also spares the video model from having to remember it. The GameWM-VM generates what is expensive to author by hand: the look of the world and the detail of the player's interactions. One game mechanic can then be re
world model - arxiv:2611.19244 · cs.AIAgentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct PatternsDeyang Kong, Zhaoyi Li, Sijian Ren, Yuan Wei +4
On-policy distillation (OPD) transfers specialized capabilities to language-model agents, but its cross-domain effects on interaction remain unclear. We investigate how OPD reshapes ReAct interaction patterns across mathematics, ALFWorld, and $τ^2$-bench, focusing on reasoning, actions, and observation acquisition. General-reasoning supervision reduces mathematical tool attempts from 70.4% to 14.4% while increasing accuracy from 16.7% to 38.0%; code-interpreter supervision improves household actions and retail queries. In contrast, ALFWorld supervision reduces $τ^2$-bench accuracy from 41.5% to 6.6%, accompanied by frequent tool calls that violate user-information dependencies. To address negative transfer using only source-domain training trajectories, we explore selective OPD guided by teacher preference changes before and after domain-specific RL. Retaining the top 95% of positions raises $τ^2$ accuracy from 11.5% to 29.9% with a source-learning trade-off.
agentic - arxiv:2611.19237 · cs.AIAuditable Authorization for LLM Agents: Empirical Evidence Across Model, Harness, and Workflow LayersBangjie Deng, Xiaoxuan Lu, Wenjun Chen, Chengge Wen +7
Human approval provides effective control only if rejected work stops and authorized work completes. We formalize refusal safety and bounded authorization completeness, and audit three execution boundaries: the model selects a call, the harness schedules and resumes it, and the workflow routes the decision to a protected effect. Paired decisions, exact traces, and native business-state witnesses support experiments with three models, five prompt arms, four runtime approval designs, two additional framework replays, and 100 public workflow sources. The paired model panel leaves seven of 300 approvals unfinished despite zero denied effects; parallel confirmation loses an approved branch in three tasks per model, restoration omits 336/672 CrewAI approvals, and four of 59 comparable n8n workflows execute after Reject. Comparing intent, generated calls, dispatch, and native effects identifies the first observable divergence and its repair target. This yields a practical audit contract that
llm agent - arxiv:2611.19230 · cs.AICompoundingKernel: Self-Improving from Zero to FirstQian Qu, Hongyang Wang, Wenbo Chen, Yichuan Li +8
We present CompoundingKernel, a self-improving agent system that writes GPU kernels by search. As of 2026-11-11, its generated kernels rank #1 on the public L1, L2 and Quantization boards of SOL-ExecBench (NVIDIA B200). They have held #1 on L1 since 2026-10-27, albeit a short loss on 2026-11-07. All 235 problems have a site-verified correct kernel; on the 209 public-board problems they are 4.40x faster than the benchmark's reference baseline. The search starts from zero: agents get each problem's PyTorch definition but no reference kernels or official scores. A per-problem archive outside the model holds every correct kernel and its lineage. An LLM controller reads it, picks a parent or a new design, and gives each parallel child agent one direction. Checks outside the model decide correctness. At equal wall time, it beat a coded UCB style tree on 35 of 38 problems by making 2.4x as many correct attempts. Over a thousand agents ran at once, and one warm restart fix raised their evaluat
agentagent systemself-improvingbenchmark - arxiv:2611.19227 · cs.AICan an Agent Rethink What It Knows? Selective Scene Memory Updates in a Changing WorldZhekai Ren, Jiaming Wu
Task configurations in factories, warehouses, and everyday physical environments change over time. Physical AI systems therefore need to incorporate new facts into scene memory while preserving valid information about equipment, objects, and operating constraints. We study updates to this parametric semantic-memory layer and present a joint evaluation of correction and preservation under finite update budgets. At fixed current fact count and memory parameter count, the protocol combines independent single-fact updates, budget trajectories, old-to-new answer diagnostics, and first stable success. Using a frozen 4B language model with late-layer LoRA, we evaluate seven preparation conditions and six update budgets on four-field synthetic semantic configurations. In two exposure-matched history conditions, increasing the budget from eight to 32 updates raises joint success from 6.25% to 94.79% and 91.15%, respectively, while exact obsolete-value outputs fall from 174/192 and 166/192 to ze
memoryagent - arxiv:2611.19220 · cs.AIRevisiting What Makes a Good Context Subset: A Sparse Latent Basis PerspectiveCen-Jhih Li, Siddhant Chaudhary, Aditya Bhaskara
Selecting context subsets is increasingly important in modern LLM inference, yet what makes a subset effective remains unclear. We study this question in the controlled setting of few-shot demonstration selection. We examine a hypothesis that demonstrations should collectively provide task-relevant information needed by a target query. Assuming that the query representation provides a useful proxy for these requirements, we formulate selection as sparse reconstruction and instantiate it with Residual Explanation Search (RES), a Matching-Pursuit-style diagnostic procedure. Across seven benchmarks and eight open-weight LLMs, no retrieval objective is universally best. RES outperforms MMR in 34 of 56 model--dataset pairs, with its clearest advantage on sentiment tasks, suggesting that query-conditioned collective reconstruction may capture useful complementarity that pairwise dissimilarity misses. Relevance-based retrieval remains strong on template-heavy reasoning benchmarks, while all r
benchmark - arxiv:2611.19218 · cs.LGNeural Dual Bounds: Valid-by-Construction JGLP Warm-Starts for MAP and Constrained MAPAkshay Vyas, Shivvrat Arya, Brij Malhotra, Tahrima Rahman +2
We present Neural Dual Bounds, a calibration-free framework that produces deterministic upper bounds for MAP inference in probabilistic graphical models using a single neural-network forward pass. The framework trains a network to output the edge messages of a mini-bucket style upper bound on a given join-graph decomposition in a self-supervised manner by directly minimizing a smooth dual objective. The output of this network can be used at test time to warm-start a standard JGLP optimizer. Across diverse benchmarks, this warm start yields dual gaps that are orders of magnitude tighter than those from zero initialization, with the largest gains in the early-iteration regime. We also apply our framework for the constrained MAP (CMPE) task, where neural warm starts with test-time adaptation are comparable to the adaptive baseline on Bayesian networks and achieve the tightest dual gaps, by one to two orders of magnitude, on the grid and segmentation families.
benchmark - arxiv:2611.19215 · cs.AIMem2Fair: Online Fairness Control for Text-to-Video Generation via Fair MemoryPengyu Chen, Jiaming Zhang, Xiaohua Feng, Yuyuan Li +1
Text-to-video (T2V) generation models can produce systematically unfair demographic outcomes when prompts leave protected attributes unspecified. For example, prompts describing nurses without specifying gender may disproportionately generate videos depicting female nurses. Mitigating unfairness in T2V generation is particularly challenging: training-based methods that modify model internals require architecture-specific access and risk disrupting temporal consistency, while sampling-based methods often incur costly pre-deployment video generation. We present the first conditional online approachability formulation of attribute-level fairness for T2V generation. Building on this formulation, we introduce Mem2Fair, a plug-and-play framework that maintains concept-conditioned Fair Memory. Using generated videos at inference time as feedback, Mem2Fair selects prompt-level attribute interventions based on the deviation from each concept's prescribed fairness target, without training or sep
memory - arxiv:2611.19214 · cs.AISTaSBO: LLM-Guided Strategy Planning and Configuration Selection for Bayesian OptimizationYoungseok Yoon, Peng Li
Bayesian optimization (BO) enables sample-efficient search for expensive black-box problems, but its performance depends strongly on surrogate-acquisition configurations whose suitability varies across problems and optimization states. We take a significant step toward LLM-driven strategic management of BO with STaSBO, a task-agnostic meta-controller that plans search behavior and jointly selects surrogate-acquisition configurations, while numerical BO retains responsibility for probabilistic modeling and candidate generation. STaSBO comprises two coordinated stages: strategy planning and strategy-guided configuration selection. A structured memory system connects prior plans, configuration selections, and candidate behaviors to optimization states, allowing the controller to revise its trajectory summary and behavioral plans during planning. To navigate the large joint configuration space, the selection stage shortlists configurations using surrogate fit evidence. Prospective BO-actio
memory - arxiv:2611.19213 · cs.CVIncremental Learning of Vision-Language Models via Task Subspace Projection and Dynamic LoRAHao Fu, Qian Feng, Hanbin Zhao, Jiahua Dong +4
Recent pre-trained vision-language models often encounter a Multi-Domain Task-Incremental Learning (MTIL) scenario in practice, where a set of classes of multi-modal tasks arrive incrementally. Due to privacy concerns and memory constraints, MTIL with pre-trained models encounters forgetting of knowledge from old tasks, degradation of zero-shot transfer capability, and underfitting of new-task knowledge. To overcome these challenges, previous MTIL methods attempt to learn a discriminative cross-task identification (CTI) module and an effective new-task adaptation (NTA) module. However, current CTI modules suffer from severe task confusion between seen and unseen tasks, and NTA modules cannot adaptively balance the performance and parameter cost while incorporating task-specific knowledge. To alleviate the above dilemmas, we propose an effective and efficient TSP-DLoRA method for MTIL, which consists of Task Subspace Projection (TSP) and Dynamic Low Rank Adapter (DLoRA) modules. Specifi
memory - arxiv:2611.19212 · eess.SYSite-Specific Cell Shaping: An Agentic FrameworkZhuo Xu, Zhaolin Wang, Yuanwei Liu
An agentic framework is proposed for cell shaping in future sixth-generation (6G) wireless network planning. Cell shaping controls the coverage and interferences of base stations (BSs) through antenna parameters configuration, providing an important means of improving coverage. In contrast to conventional approaches that reply on costly performance evaluations, the proposed framework achieves intelligent and efficient antenna configuration through a perception-reasoning-reflection loop. Specifically, a Configuration Agent combines scene information, evaluation history, and reflection feedback to generate antenna configuration proposals. Then, a ray tracing(RT)-based evaluator assesses these proposals and returns a structured performance summary. Next, a Reflection Agent interprets these results to diagnose site-specific performance deficiencies, recommend feasible configuration adjustments, and accumulate conditional experience. Through this iterative process, the framework accumulates
agentagentichuman-in-the-loopbenchmarkevaluator - arxiv:2611.19207 · cs.LGOn-policy distillation as a Backdoor Propagation ChannelZane Fu, Yanzhao Zheng, Ziting Qin, Qianshan Wei +7
On-policy distillation (OPD) is emerging as an important paradigm for the post training of large language models, improving student models by leveraging supervision generated online by teacher models. However, whether this process also transfers latent security risks from teacher models to student models remains insufficiently studied. In this work, we reveal a new security risk in which an attacker first implants a conditional backdoor into the teacher model through supervised fine tuning (SFT), after which OPD can transfer the backdoor to the student model with only an extremely small proportion of trigger prompt exposure. More concerningly, we find that such backdoor transfer is substantially stronger under OPD than under conventional distillation methods, suggesting that OPD may provide an efficient channel for propagating backdoors across models. Our mechanistic analysis shows that the backdoor signal is primarily transmitted through anomalous differences in the conditional probab
post training - arxiv:2611.19202 · cs.CLOmni-Workspace: Benchmarking Omni-Modal AI Assistants on Multimodal Workspace TasksYuhao Wang, Yucheng Zhao, Yiran Zhong, Feida Zhu +4
Omni-modal agents are expected to move beyond multimodal understanding toward completing complex tasks in realistic multimodal environments. In these settings, information is distributed across heterogeneous files and modalities, with cross-file and cross-modal dependencies requiring agents to continually coordinate perception, reasoning, and tool use as newly gathered evidence informs subsequent actions. However, existing omni-modal benchmarks primarily focus on multimodal perception and understanding, leaving complex task completion through sustained interaction with multimodal workspaces underexplored. We introduce Omni-Workspace, a benchmark comprising 380 deep-research-style tasks across four multimodal workspace types and two difficulty levels. These tasks require agents to inspect and process workspace files, retrieve external information when needed, integrate evidence across files and modalities, and produce artifacts satisfying explicit requirements. Evaluating 18 representat
tool usebenchmark - arxiv:2611.19200 · cs.LGLogT: Logically Think with Images for Visual SearchYanjun Fu, Quanzeng You, Jiadong Guo, Yujie Lu +5
Thinking with images empowers vision-language models (VLMs) to use an external Zoom-in tool for fine-grained visual search. Recent works employ supervised fine-tuning (SFT) and reinforcement learning (RL) with tool-specific designs to enable such capabilities. However, achieving logical reasoning alongside accurate and efficient tool use remains a significant challenge. To address this, we shift the focus away from increasing RL complexity and instead identify SFT data quality as a key bottleneck. We then propose LogT (short for Logically Think with Images), a fully automated SFT data engine that explicitly enforces logic-chain-guided reasoning. LogT generates an SFT dataset featuring challenging, visually grounded questions and structured tool-use traces with human-like logic. We validate our approach by fine-tuning two base models, Qwen2.5-VL-7B and InternVL3.5-8B, on the dataset and evaluating them across seven benchmark test sets. We show that LogT-SFT based on Qwen2.5-VL-7B outper
tool usetool-usebenchmark - arxiv:2611.19199 · cs.AIWhat Moves a LoCoMo Score? Memory Context, Answer Prompts and JudgesNikhil Kumar Jha, Amitoj Cheema
LoCoMo is a widely reported benchmark for long-term conversational memory, but published scores mix memory quality with choices of answer prompt, judge and baseline layout that are rarely reported together. We measure how much these choices move a score, and evaluate Becos, a team memory system for coding agents, as a black box under all of them: ten conversations (1,540 questions), two answer models (gpt-4o-mini, gpt-4.1-mini), two answer prompts and three LLM judges, against a whole-conversation baseline rerun under identical conditions. Changing only the judge moves a score by up to 7.2 points and changing only the answer prompt by up to 9.3. A blind human check of 150 answers finds the judges reliable when they agree, and the primary judge's errors on disputed answers working against Becos rather than for it. Becos answers from a context with 90% fewer characters than the whole conversation. Its configuration was developed on LoCoMo itself, so there is no held-out set. With gpt-4o-
memorybenchmark - arxiv:2611.19198 · cs.ROSelf-Calibrating Vision-Agent-Action: Deploying Pick-and-Place to New Robot Cells without Demonstrations or Fine-TuningJae Hun Hwang, Yechan Seo, Hong-In Won
In simulation, Vision-Agent-Action (VAA) performs pick-and-place in robot cells with gripper or base-frame changes without robot demonstrations or fine-tuning. It uses pre-trained vision-language and language models for interpretation, measures the command map and camera geometry from its own motion and fingertips, estimates lens distortion, and plans grasps from measured object depth. A controlled study of eight public LIBERO-fine-tuned vision-language-action models (VLAs) varies robot hardware, camera geometry, appearance, and base frame. Gripper and frame changes dominate their losses in the tested fractional design. Across 18 cells with these changes, frozen VAA achieves 230/360 successes (64%), the refined VAA 268/360 (74%), and the strongest released VLA on this subset 76/360 (21%). Paired ablations support the self-calibration bundle, lens estimation under compounded changes, and the measured-depth pipeline. On a two-step LIBERO-Goal chain with object state carried between instr
vision-language-actionvlaliberogrippergrasp - arxiv:2611.19196 · cs.ROEgo2Sim2Humanoid: Agentic Real-to-Sim for Learning Loco-Manipulation VLAs without Robot DemonstrationsChen Liu, Nikolaus Correll
Imitation learning for visuomotor policies typically relies on teleoperated robot demonstrations, which are costly because every episode requires a trained operator controlling a physical robot in real time through teleoperation interfaces and wearable body-tracking hardware. Egocentric glasses capture human demonstrations without a robot or handheld grippers, yet they track only the head and in-view human hands, and directly copying human motion to a robot with different stature, arm reach and hand morphology fails to produce feasible contacts. \method{} converts egocentric glasses recordings into physically verified robot training trajectories. An agentic real-to-sim stage reconstructs collision and textured visual assets from the recorded RGB-D and head streams, completes unseen surfaces when needed, and predicts target contact bands on the manipulated objects. During motion adaptation, a grasp planner turns each contact band into a feasible robot wrist pose and finger closure, bile
manipulationhumanoidteleoperationgr00tgrippergrasp - arxiv:2611.19194 · cs.LGGeneralization Testing of Grid Foundation Models for Power Flow and Optimal Power FlowAyushi Jolotia, Parikshit Pareek
There is growing interest in foundation models (FMs) for power grids, particularly for power flow and optimal power flow. Their usefulness depends on generalization accuracy, which existing works test on their own benchmarks. We propose a three-axis framework to test their generalization limits: changed admittance matrices of training grids, load levels, and unseen grids. We test GENCO, LUMINA and GridSFM against the DC baseline, alongside their reported limitations, and discuss theoretical reasons for their failures. Experiments reveal that adding a line degrades all FMs more than removing one, and longer lines eventually make LUMINA and fine-tuned GENCO worse than DC. All three degrade beyond their training load range, where the loading-voltage curve steepens. On unseen grids, LUMINA degrades sharply even inside its training range, and GridSFM fails on the grid farthest from it. Thus, beyond training conditions, FMs behave more like meta-models requiring test-time fine-tuning data an
benchmark - arxiv:2611.19190 · cs.RODistilling Safety Filters for Filter-free Multi-agent RLSrikar Babu Gadipudi, Somayeh Sojoudi, Claire J. Tomlin
Reinforcement learning (RL) can teach robot teams to navigate shared spaces, but trial and error causes collisions that real robots cannot afford. Control barrier function (CBF) safety filters prevent these collisions by correcting unsafe actions, yet in a team the filter becomes a centralized optimization whose number of constraints grows quadratically with the team size. We propose an approach that builds the filter's safety into the RL policy and examine how such a system behaves during filter-free deployment. However, the filter that keeps training safe also hides what the policy must learn: it masks the cost of unsafe actions, which actor-critic methods then overvalue, and it keeps near-collision states out of the training data. Our approach therefore learns only from executed safe actions with implicit Q-learning and distills the filter into the policy at synthetic near-collision states. We prove that filter distillation loss bounds the policy's constraint violation and hence how
multi-agent - arxiv:2611.19187 · eess.SYA Unified Control Framework, Mathematical Model, and Five-layer Closed-loop Case Study for Embodied Intelligence Control SystemsYiwei Zhou, Zhongcheng Lei, Xiaoran Dai, Wenshan Hu +1
Embodied intelligence holds that an agent develops cognition and behavior through real-time body-environment interaction, so its control system must tightly couple perception, decision making, learning, optimization, constraints, and physical execution. This report asks whether a unified framework exists, how its mathematical models unify, and how the framework is realized computably. Four conclusions emerge. (1) No single framework is accepted, but the field converges on a common skeleton: from operational-space and whole-body control to MPC, reinforcement learning, active inference, and diffusion/flow policies, all reduce to constrained sequential decision making on the configuration manifold. (2) Two languages describe it-deterministic optimal control (Euler-Lagrange plus constrained OCP, reduced in real time to a QP) and probabilistic inference (MDP/POMDP plus free-energy minimization)-rigorously bridged by control as inference and KL/path-integral control. (3) Engineering unificat
embodiedmanipulatorwhole-body controlworld modelagent - arxiv:2611.19184 · cs.CVDo VLMs Really Look at the Road? A Diagnostic Evaluation for Autonomous DrivingShankar Gangisetty, Muskan Singh, C V Jawahar
Vision-language models (VLMs) are increasingly being evaluated for autonomous driving through question-answering benchmarks, yet high benchmark accuracy does not necessarily reflect visually grounded reasoning. We present a diagnostic evaluation framework that systematically probes whether VLMs base safety-critical driving decisions on visual evidence or instead rely on language priors, answer biases, and dataset regularities. Our benchmark comprises two task categories, Adverse Weather and Junctions \& Intersections, with a nuScenes spatial reasoning extension, and evaluates open-source (general, driving, and spatial-reasoning) VLMs along four diagnostic protocols: image versus no-image evaluation, counterfactual weather sensitivity, linguistic robustness under paraphrasing and negation, and spatial reasoning. All ground-truth labels are derived programmatically from BDD100K and DriveLM annotations, enabling reproducible evaluation without manual labeling. Across all models, we find t
benchmarkevaluation framework - arxiv:2611.19183 · cs.CVWA-JEPA: Goal-Aware Predictive Feedback Action Refinement for a World-Action ModelTao An, Weihang Liang, Tianqi Wang, Chuqing Cao +6
JEPA-based world models predict action consequences in latent space, but representative planning approaches rely on candidate-action search, requiring repeated evaluation of action sequences and incurring substantial online computational cost. Direct action generation avoids this search, yet how to leverage latent predictions to guide action generation given task goals remains a key challenge in translating learned physical dynamics into effective decisions. We introduce WA-JEPA, a world-action model that couples action generation, action-conditioned latent prediction, and goal-aware feedback. The model first generates an action proposal and predicts the resulting future scenes and proprioceptive states, then refines the proposal through two complementary feedback paths, using the task goal as a reference. Goal-discrepancy residual feedback (GDRF) uses differences between the latent representations of predicted scenes and the task goal to guide chunk-wise action refinement. Future-scen
world modelaction-conditioned - arxiv:2611.19179 · cs.CVIsolating Perception for Multi-View Planning under Occlusion-Induced Partial ObservabilityKevin Nguyen, Ceyhun Efe Kayan, Feng Liu, Li Zhang
Vision-language models (VLMs) are beginning to saturate visual planning benchmarks, yet they fail to resolve depth ambiguity and degrade sharply with problem size under occlusion-induced partial observability (PO). We introduce BW-PO based on the Blocksworld domain, a controlled benchmark that reduces existing VLM planners and formalizers to near-zero performance, and propose VLM-as-perceiver, a paradigm that narrows the VLM's role to local perception while delegating cross-view evidence reconciliation and plan generation to external solvers. Across model-specific challenging subsets of BW-PO, this raises plan success from 1% to 97% for Gemma-4-31B, 0% to 95% for Qwen3.5-122B-A10B, and 3% to 84% for Gemini-3.1-Pro. Isolating perception in this way separates two bottlenecks that are otherwise entangled: reconciliation under depth ambiguity, and completeness of visual extraction under scene density.
benchmark - arxiv:2611.19172 · cs.CVEviRef: Evidence-Specific Memories for Logical Anomaly Detection across Capture ConditionsYaoting Huang, Wenqi Cai, Yijin Wei, Jianjian Qin +3
Logical anomalies arise when every component looks normal but their combination violates the scene rules. In practical inspection, these rules must hold under varying backgrounds, illumination, and imaging quality, with only a few dozen normal images per scene, whereas existing methods couple normality with appearance and merge different constraints into one reference. We model logical normality as decomposable constraints and propose EviRef, a training-free method that builds a separate reference memory for each family of evidence. Component prototypes built once per scene parse images under any capture condition into one label space, on which occupancy and relation constraints are verified. Semantic constraints are verified with whole-image embeddings independent of parsing, and additional evidence such as text is admitted through a task-level gate without modifying existing memories. On VID-AD, EviRef attains a mean image-level area under the receiver operating characteristic curve
memory - arxiv:2611.19165 · cs.AIFrom Sensors to Silicon: Causal Cross-Layer Self-Healing Edge AI for Secure and Energy-Adaptive IoT SystemsMostafa Darvishi
The migration of artificial intelligence from centralized cloud infrastructure to resource-constrained Internet-of-Things (IoT) devices reduces communication latency and improves data locality, but it also tightly couples sensing quality, inference behavior, microarchitectural activity, hardware reliability, security, and energy management within the same physical node. Existing edge-intelligence solutions predominantly address these dimensions in isolation, making it difficult to determine whether an observed degradation originates from sensor malfunction, data-distribution shift, malicious execution, hardware faults, or resource and thermal stress. This paper presents a causal cross-layer self-healing Edge-AI framework that continuously observes synchronized telemetry spanning sensor health, inference uncertainty, processor hardware-performance counters, accelerator and memory activity, power and thermal state, and communication conditions. A temporally informed causal reasoning engi
memory - arxiv:2611.19157 · cs.AIFundingClaw: Automated Grant Proposal Writing Agents SystemRuizhi Liu, Siru Ge, Siran Wang, Qianli Ma +1
Grant proposal writing remains a manual, labor-intensive bottleneck: proposals must satisfy rigid, agency-specific templates, and the same intellectual content must be repeatedly re-encoded into incompatible formats across funding organizations. We introduce FundingClaw, a multi-agent framework built on template-driven skill selection, where a Template Parser ingests any agency's specification and selects the appropriate writing skills for each stage, so that one research brief can produce compliant proposals for NSFC, NSF, KAKENHI, or ERC without modifying generation logic. A blueprint-skeleton strategy drafts a structured outline before any prose is written, mitigating the semantic drift that plagues long-form generation. Three sequential stages produce the rationale, research content, and research foundation sections, with multi-source verification of the PI's record and optional PDF-grounded extraction of representative works. Optional human feedback checkpoints and a closed-loop s
multi-agentagent framework - arxiv:2611.19153 · cs.ROImpedanceMimic: Learned Impedance as Action Space for Force-Accurate Motion ImitationMatthias Jammot, Pembe Gizem Özdil
Recovering the forces behind a motion is ill-posed: different controllers can reproduce the same movement while generating different contact forces. Physics-based motion imitation resolves this ambiguity through simulation, but relies on fixed-gain proportional-derivative (PD) control. We show that fixed gains impose a trade-off between motion and force tracking, as contact and swing place conflicting demands on joint stiffness and damping. We introduce ImpedanceMimic, a reinforcement-learning policy that modulates per-joint stiffness, damping, and position setpoints from motion and force measurements, but requires only motion at inference. Across 129 participants and six activities, ImpedanceMimic outperforms matched fixed-gain PD under every integration scheme, direct-torque, muscle-actuated and existing imitation controllers in kinematic and force fidelity, approaching a dedicated motion-to-force regressor. By expressing gains as multiples of inertia-scaled priors, a single policy c
humanoid - arxiv:2611.19140 · cs.AIWhen the Mythical Man-Month Meets Agentic Web Systems: Structured Coordination for Scalable DevelopmentZhiyao Wu, Zi Liang, Haibo Hu
As agentic systems are increasingly deployed to support complex Web tasks, multi-agent architectures offer greater parallelism, specialization, and task coverage. However, increasing the number of agents does not necessarily improve system performance, as communication, synchronization, redundant work, and coordination overhead may offset the benefits of additional agents. This presents a fundamental scaling challenge for agentic Web systems: how can agent populations be organized to balance task performance and coordination costs? To address this challenge, we systematically investigate the relationship between agent population size, per-agent token consumption, and coordination overhead in collaborative agentic systems. Experiments on complex Web-oriented software development tasks reveal a non-linear scaling pattern: per-agent token consumption initially decreases as the agent population grows, but eventually exhibits diminishing returns. Motivated by these findings, we introduce th
agentmulti-agentagentic - arxiv:2611.19138 · cs.ROSuperGlance: Driving at a Glance with a Vision-Language ModelJieyuan Pei, Xin Yan, Mingwei Xu, Haiyi Zhang +8
Autonomous driving is turning to vision-language-action models (VLAs). They answer and plan by decoding, one step at a time. Each step is another forward pass, so a plan can take over a second of GPU time. Embodied AI often pairs such a slow, reasoning System 2 with a fast System 1 that acts. Human drivers plan at a glance, with System 1. Faster designs usually put System 1 outside the vision-language model (VLM), in a planner without its abilities. Yet before a VLM writes its first token, it has already read the scene into one hidden state. We show that, once trained, this hidden state is enough to drive and to answer questions about what the model sees. SuperGlance reads it to turn a decoder-only VLM into an encoder-style driving policy. Its linear head scores a fixed set of 5,273 candidate trajectories. With the set fixed, rewards are computed offline, so supervised targets can be reward-weighted and reinforcement learning maximizes the exact expected reward. Before and during polic
vision-language-actionembodied - arxiv:2611.19131 · cs.ROWireless Edge-Assisted Adaptive Latent Reasoning for Vision-Language-Action Models in Physical AIFatih Temiz, Melike Erol-Kantarci
Vision-language-action (VLA) models enable embodied artificial intelligence (AI) by translating visual observations and language instructions into robotic actions. Emerging use cases in physical AI are expected to drive the need for efficient inference and communication, as the computational demands of the VLA models pose challenges for real-time control with limited onboard resources. Test-time scaling through latent reasoning allows these models to refine hidden representations without generating explicit chain-of-thought (CoT) tokens, but introduces a trade-off between task success, inference latency, and energy consumption. In this paper, we address the efficient collaborative inference under a realistic communication scenario. We propose a wireless edge-assisted framework for adaptive latent reasoning (WEALR) that augments onboard computation with edge computing. A decision-making agent jointly selects the execution location and latent-reasoning convergence threshold. We formulate
vision-language-actionvlavla modelembodiedmanipulationlibero - arxiv:2611.19127 · cs.LGEmpowering Masked Diffusion Models to Self-Correct with Leave-One-Out TransformersSofian Zalouk, Vincent Counathe, Paul Jünger, Daniel Cao +3
Masked diffusion models (MDMs) have emerged as a promising paradigm for language modeling, enabling parallel token generation with competitive performance. Yet standard MDMs suffer from two fundamental limitations: (1) inability to revise tokens, and (2) sparse training signal. We revisit hollow transformers, which address both limitations by predicting each token from its surrounding context, independently of the input at that position. We introduce the Leave-One-Out Transformer (LOOT), a more efficient and expressive architecture that advances hollow transformers through two complementary advances. First, we redesign the merging mechanism to enable frequent interaction between left and right contexts, substantially improving predictive performance over prior hollow transformers. Second, we derive a lower-variance estimator of the MDM NELBO that supervises all tokens and improves validation perplexity. Together, these advances yield a competitive standalone diffusion language model ca
self-correction - arxiv:2611.19124 · cs.LGANCHOR: A Port-Hamiltonian Model for Forecasting Team Dynamics in Human-in-the-Loop SystemsDiana Romero, Salma Elmalaki
Cyber-physical-human systems close a loop around people by sensing human behavior, predict it and act on it. Human teams are a natural target, because collaboration breaks down in ways that a timely prompt could prevent. Collaborative mixed reality makes this loop practical, since each headset senses who speaks, where its wearer looks and who stands together, and it can deliver a light prompt. In this loop, the headsets are sensors, the prompts are actuators and the team is the plant. Closing the loop needs a plant model that is small enough to learn from few teams, safe to iterate and readable by a designer. We propose ANCHOR, which models a team as a ball in an energy bowl. The bottom of the bowl is where the team rests, and it is set by the team's recent behavior, its own history and other teams. Formally, ANCHOR is a switched port-Hamiltonian system with 52 parameters, and each part stands for a testable claim from team psychology. Because of this structure, for any fitted paramete
human-in-the-loop - arxiv:2611.19123 · cs.AIBeyond Plain MCP: Do Decision Models and Code Mode Make Tool Use Cheaper and Faster?Abhishek Rai
Model Context Protocol (MCP) gateways manage tool catalogs for language model agents using two primary paradigms: code mode where agents write scripts against tool stubs and decision layers which are cheap routers that preselect tool subsets. Using a controlled harness across 12 servers, up to 1,958 distractor tools, and three open-weight agents including Qwen, Gemma, and Nemotron, we present a preregistered evaluation on a sealed test split to determine when each approach succeeds. Code mode substantially outperforms plain MCP, cutting token usage by 70 to 90 percent while significantly boosting utility. Declaring return-shape hints in code mode stubs is a first-order variable, swinging success by up to 39.5 points for lower-performing agents, an effect larger than adding a router. The central hypothesis that a decision layer in front of code mode beats code mode alone is not confirmed, showing minor, statistically non-significant improvements for weaker agents and a slight cost for t
agenttool use - arxiv:2611.19122 · cs.AIAgent-Guided Process Verification for Mathematical Self-CorrectionRui Zou, Mengqi Wei, Xin Zhao, Ji-Rong Wen
Large language models have made substantial progress in mathematical reasoning, yet in complex multi-step reasoning, an error at any step may propagate along an incorrect path, and the resulting solution may omit necessary branches. Existing natural-language reflection methods often struggle to locate critical errors precisely, while outcome verification typically checks only whether the available candidate answers satisfy the problem conditions and can therefore overlook necessary branches. To address these limitations, we propose Agent-Guided Process Verification (AGPV) for mathematical self-correction. AGPV decomposes the current solution into an ordered sequence of key claims, verifies them individually while invoking a shared mathematical toolbox when needed, and stops upon detecting the first error. A summarization agent converts the failure evidence into a diagnosis that guides a correction agent to re-derive the subsequent solution from that point; AGPV then adapts the next ver
agentself-correction - arxiv:2611.19121 · cs.CVDecoEdit: Beyond Flat Control for 3D-Aware Unified Image EditingXiaotang Gai, Jinbo Xing, Zijie Meng, Jiayi Yi +5
While recent advances in multimodal image editing have achieved impressive results, current frameworks remain largely confined to ``flat" 2D controls (e.g., text prompts, bounding boxes, or masks). Precise 3D-aware geometric manipulation -- such as coordinating an object's location, physical scale, and 3D orientation while strictly preserving its fine-grained identity -- remains an elusive challenge, typically suffering from severe detail degradation or requiring expensive per-instance fine-tuning. In this paper, we present DecoEdit, an all-in-one image editing framework that goes beyond flat control by seamlessly integrating 1D textual, 2D spatial, and 3D volumetric inputs under a unified architecture. First, to power this unified model, we establish a data curation pipeline and curate a large-scale dataset, namely DecoDataset, containing 1.6 million samples, encompassing diverse multi-dimensional (1D/2D/3D) tasks and their complex compositions. Furthermore, we establish a comprehensi
manipulationbenchmark - arxiv:2611.19118 · cs.AILatency, Speech Recognition and Turn-Taking in Voice Agents for Spanish Phone Negotiation: An Externally Measured Comparison of Pipecat, LiveKit and GPT-LiveJovan Popoca Moran, Eduardo de Avila-Armenta, Raúl Alejandro Aquino-Ibarra, Jorge Octavio Nicolás-Díaz +1
A voice agent that negotiates by phone must answer quickly, understand spoken amounts and cope with a caller who talks over it. We compared three architectures for such an agent, negotiating freight rates in Spanish: two cascades (speech recognition, language model and speech synthesis) orchestrated by Pipecat and by LiveKit Agents, and GPT-Live, an OpenAI speech model that listens and speaks without separate stages. A program played the carrier with the same 47 recordings of one speaker for all agents, without the telephone network, in 77 calls per architecture over five negotiation scenarios, and measured latency on the listener's side with a validated instrument. Median latency to the content response was 2,616.0 ms (Pipecat), 3,090.0 ms (LiveKit) and 2,444.8 ms (GPT-Live); the confidence interval of GPT-Live's 171.2 ms advantage over Pipecat includes zero. According to its own timestamps, GPT-Live began to answer 800 ms after the end of the line, close to the 0.8 s published by Ope
agent - arxiv:2611.19116 · cs.AIA Hybrid State Space-Attention Architecture for Forecasting the Trajectories of Battery DegradationShrey Patel, Bhargav Srinivas Rao, Samet Ayhan
Early-life battery degradation trajectory forecasting aims to predict a battery cell's full state-of-health (SOH) trajectory from its initial charge-discharge cycles, potentially reducing cell qualification and evaluation time from months to days. Despite recent advances, accurate forecasting remains challenging for emerging battery chemistries, data-scarce domains, and cells exhibiting degradation behaviors beyond those observed during training. Existing state-of-the-art approaches increasingly rely on model scaling, large language model-derived aging-condition embeddings, and extensive fold-specific hyperparameter tuning, raising concerns about computational efficiency, reproducibility, and generalization. This work introduces DecayFormer, a compact hybrid state-space and attention architecture designed for early-life degradation trajectory forecasting. DecayFormer combines cycle-level selective recurrence, using a single scalar state per channel, with an attention mechanism queried
benchmark - arxiv:2611.19115 · cs.AILearning inadequacy of Bayesian neural operators for PDE-constrained inference and designPratyush Kumar Singh, Lianghao Cao, E. Bruce Pitman, Danial Faghihi
Partial differential equations (PDEs) underpin predictive modeling across science, engineering and medicine. Their use in digital twins, parameter inference and optimal design requires repeated solutions across parameters and scenarios. For nonlinear, transient and multiphysics systems, these calculations can make numerical simulation prohibitively expensive. Here we introduce the residual-informed Bayesian neural operator (RI-BNO), which learns operator inadequacy using a posteriori error estimates from the governing PDEs and incorporates its correction and uncertainty into Bayesian training and prediction. We establish local error bounds for the residual-informed estimates under stability and smoothness assumptions. We demonstrate RI-BNO on a non-linear reaction-diffusion benchmark with DeepONet and latent-subspace architectures, Markov chain Monte Carlo inference of glioma parameter fields from magnetic resonance imaging, and Monte Carlo risk-averse optimization of three-dimensional
benchmark - arxiv:2611.19110 · cs.AICLEAR: Legible and Steerable Recommendation via Grounded Natural-Language InterestsQi Guo, Reza Mirghaderi, Iordanis Fostiropoulos, David Fan +50
Modern recommenders keep learned preferences in representations people cannot inspect or change. We present CLEAR (Controllable and Legible End-to-end Architecture for Recommendation), an end-to-end system deployed at Meta and built around an editable natural-language interest layer. CLEAR makes a ranked subset of inferred interests readable and stores accepted additions and suppressions as authoritative preference state. Behavioral grounding maps each inferred interest back to supporting events in the user's own history. The resulting interest object is also a serving address: its label and profile-conditioned rewrites drive lexical retrieval, its supporting events seed behavioral retrieval, its projection queries an embedding index, and its stable identity supports ranking, attribution, and exposure control. A fresh user declaration may precede behavioral support, so it enters through the same serving identity and initially draws candidates from topical retrieval. Ranking makes those
benchmarkevaluation protocol - arxiv:2611.19109 · cs.LGST-GAT: A Leakage-Free Spatio-Temporal Graph Attention Framework for Robust Network Intrusion DetectionNikhil Parkar, Sanskar Pandey, Siddhi Prajapati, Suganeshwari G +1
Intrusion detection systems are a central defence in modern networks. Deep learning models report strong anomaly-detection results, but evaluations on standard benchmarks such as CSE-CIC-IDS-2018 are often inflated by two methodological problems: temporal leakage from overlapping sliding windows, and contamination from preprocessing fitted before the train/test split. Per-flow classifiers also ignore the spatio-temporal dependencies across sequences of network flows. We propose a Spatio-Temporal Graph Attention Network (ST-GAT) that represents traffic as a sequence of graph snapshots. It builds k-Nearest Neighbour (k-NN) edges within each window, applies multi-head Graph Attention Network (GAT) layers for spatial aggregation, and uses a Bidirectional Long Short-Term Memory (BiLSTM) network with temporal attention for sequence modelling. To prevent leakage, we use non-overlapping temporal windows (stride equal to window length) and fit all preprocessing on the training split only. We co
memorybenchmark - arxiv:2611.19108 · cs.AIGuiding AI Agents to Generate Conjectures by Encoding How Theorems Build on One AnotherHita Kambhamettu, Daniel Krashen, Lyle Ungar
Recently, LLMs have been making headlines, and headway, in mathematical research. Much of this work focuses on proving existing open conjectures. Deciding what to prove is harder and less studied, because of the lack of a clear criterion for what makes a conjecture worth pursuing. We present a method for conjecture generation that builds on the observation that mathematical theorems build on one another: new theorems can modify the setting or hypotheses of earlier theorems, strengthen their conclusions, or extend their scope. We encode these relationships in a new kind of graph, the Theorem Evolution Graph (TEG), which we instantiate as an online search method for LLM agents. An agent builds a TEG in two steps: (1) identifying relevant theorems from papers found by following citations backward and forward from a seed paper; and (2) comparing those theorems to record how one builds on another. LLM judges and an expert mathematician preferred TEG-generated conjectures over those from an
agentai agentllm agent - arxiv:2611.19105 · cs.AICan Agentic LLMs Write Task-Based Parallel Code? A Study of OpenMP Tasking, HPX, StarPU and TTGLinus Bantel, Anna-Lena Roth, Tim Thüring, Jonas Posner +1
Task-based programming offers more expressive power than loop-level parallelism, but the runtimes providing it differ widely in API style, maturity, and the amount of public code to learn from. This paper asks how well LLM agents write C++ for four of them: OpenMP Tasking, HPX, StarPU, and TTG. Instead of single one-shot prompts, we use an agentic workflow in which the model edits files, compiles, and runs its kernels until it reproduces a reference output. Five locally hosted open-weight LLM implement seven HPC kernels, including matrix multiplication, Cholesky and LU factorization, in all four frameworks and a serial baseline to verify their understanding of the problem in general. Every kernel is evaluated independently for correctness at each concurrency level, strong scaling, and peak memory. Model capability dominates the outcome: three models produce correct, genuinely task-parallel kernels in 57-82% of the cases, while two models rarely compile their own code and almost never s
llm agentagentic - arxiv:2611.19101 · cs.LGA predictive, per-engine energy roofline of the Apple Neural EngineSpencer H. Bryngelson
The Apple Neural Engine (ANE) is in over a billion devices, but is characterized indirectly via CoreML. Using a CoreML-free fused-graph dispatch path, we measure the ANE, along with the Apple GPU and CPU on the same device, with per-engine energy accounting. The consumer-available roofline does not predict the ANE in practice. A critical correction is that the ANE sees two effective memory bandwidths: a slow activation path and a fast weight-stream path. Two roofline ridges follow, and the weight-stream ridge is below that of the GPU. The ANE single-operand compute peak decreases instead of plateauing. Its weight stream is byte-bound in fp16 but limited by the dequantizer's element rate in int8. As a result, int8 compression saves energy at unchanged throughput and an int4 lookup-table stream raises throughput and efficiency at a cost of 2.55 points of ImageNet top-1. The corrected roofline identifies the lower-energy ANE paths and bounds it from below. A constructed device map reprodu
memory - arxiv:2611.19100 · cs.ROLeaning Over to Reach Far-Away ObjectsAllen Devaraj, William Xie, Nikolaus Correll
We study supported reaching to extend a humanoid robot's workspace by bracing against the environment for stabilization using a straightforward yet capable whole-body model predictive control (MPC) approach, exclusively with local sensing and fiducial markers. While typical solutions for whole body humanoid control aim at avoiding or accommodating contact, contact is explicitly required to reach objects placed on surfaces far out of normal reach. We employ sampling-based MPC for whole-body leaning, bracing, reaching, and recovery on a Unitree H1-2 humanoid. The controller runs in real-time at 34 Hz on a laptop, transfers sim-to-real, extends reach by about 50 cm beyond the unbraced standing workspace, and sustains the loaded posture for over four minutes, before pose estimation drifts. The controller reaches the commanded target in 89% of trials and completes the full reach-and-recovery cycle in 81%, across 54 trials spanning six targets and three table heights. We further characterize
manipulationhumanoidsim-to-real - arxiv:2611.19095 · cs.CVDEFDVO: Deformable Deep Visual Odometry for Equirectangular ImagesBurak Sevsay, Egor Bondarev, Erkut Akdag
Omnidirectional ($360^\circ$) cameras capture the full surrounding scene in every frame, providing a strong advantage for visual odometry (VO) over pinhole sensors with a limited field of view. However, their equirectangular projection (ERP) introduces latitude-dependent distortion, an artificial seam at the image border, and a sampling bias toward the poles. Existing $360^\circ$ VO methods address these challenges only partially. DEFDVO closes this gap with a patch-based deep VO method that operates directly on ERP imagery, combining spherically initialized deformable feature extraction, end-to-end seam-aware estimation, and spherical-gradient patch selection. The deformable feature extractor combines spherical geometric priors with learned sampling offsets, while seam-aware processing preserves ERP continuity, and spherical-gradient selection reduces polar sampling bias. Trained exclusively on synthetic TartanAirV2 ERP imagery, DEFDVO improves the best prior average absolute trajecto
benchmark - arxiv:2611.19094 · cs.CLWhat Transfers Across Cuneiform? Script-Level Substrate and Language-Specific Regularities via Segmentation and RepresentationChuanjun Zhou, Adam G Anderson
Cuneiform recorded typologically unrelated languages for more than three millennia, raising a natural question: which of its statistical regularities are properties of the shared writing system, and which belong to individual languages? We answer this with two probe pairs targeting the same script-versus-language decomposition. The first compares two unsupervised, dictionary-free segmentation methods: transitional probability (TP), grounded in infant statistical-learning research, and Morfessor. To our knowledge, this is the first systematic evaluation of dictionary-free segmentation on cuneiform; the only prior published baseline relies on a hand-built dictionary (Max-Match, F1 = 0.74). Both our methods reach F1 > 0.97 within language (Morfessor reaches F1 > 0.99 on Akkadian and Sumerian), exceeding MaxMatch by more than 23 F1 points, but diverge sharply in cross-language transfer: TP holds F1 > 0.95 across all six language pairs, while Morfessor collapses to F1 = 0.27 from the 13K-to
benchmark - arxiv:2611.19081 · cs.RORoboCoEvo: Expanding Robot Capabilities through Agent-Policy Co-EvolutionRan Qi, Yixin Xiao, Dawei Wang, Zhiyuan Li +6
Integrating MLLM-based agents with robotic foundation models has attracted increasing attention as a means of enhancing robotic capabilities. However, the improvement of such integrated systems remains largely decoupled across their agent and model components, with advances in one component not necessarily benefiting the other, substantially limiting the capabilities of the combined system. As both MLLMs and robotic foundation models become increasingly capable, fully realizing their combined potential calls for co-evolution, an iterative process in which improvements in agents and robotic foundation models reinforce one another. We propose RoboCoEvo, a failure-guided co-evolution framework that couples the refinement of the agent's explicit and reusable executable knowledge with parameter learning in robotic foundation models, such as vision-language-action (VLA) models. Through a code-as-policy formulation, the agent develops and refines executable failure-detection and recovery prog
vision-language-actionvlarobotwinbehavior-1kagentcode-as-policy - arxiv:2611.19078 · cs.AIAssume/Guarantee Contracts for the Compositional Multi-Agent Reinforcement Learning of Heterogeneous TeamsDominic Klukas, Cyrus Neary
Multi-agent reinforcement learning (MARL) problems for heterogeneous teams require agents to collaborate to complete team tasks while making the best use of diverse agent capabilities. We introduce assume/guarantee contracts for heterogeneous multi-agent systems (AGC-MAS) and Iterative Heterogeneous Assignment for MARL (IHA-MARL). AGC-MAS contracts are compositional objects that encode the assumptions provided to, and the guarantees required of, each subtask contributing to the overall team task. We formulate contracts as tuples that specify a subtask, an automaton, timeout and success-probability requirements, and dispatch behavior. The automaton tracks an agent's high-level progress through the subtask associated with its contract and provides a natural mechanism for synchronization across contracts. Our abstraction enables robust interaction across heterogeneous low-level subtask implementations, allows heterogeneous agents to substitute for one another in satisfying contracts, and
agentmulti-agentagent systembenchmark - arxiv:2611.19074 · cs.LGTokenFormer V2: Breaking the Resolution-Complexity Dilemma in Unified RecommendationKaihui Wu, Guoming Li, Liang Guo, Yifeng Zhou +7
Industrial recommendation faces a resolution--complexity dilemma: dense interactions make detailed modeling of long behavior histories expensive, while compact history summaries restrict the evidence available to prediction. A unified backbone must preserve rich behavioral details within strict serving budgets. We introduce \textbf{TokenFormer V2}, a unified architecture that explicitly decouples \emph{representation resolution} from \emph{interaction density}. Its group-wise token layout jointly encodes behavior segments with non-sequential context, expands interaction ranges via hierarchical group merging, and coordinates distributed groups through \emph{NS Fusion} and \emph{Global Sync} modules. Through this \emph{local--global decomposition} paradigm, computation is concentrated on structured interactions while preserving fine-grained evidence, ensuring that every encoded event remains directly accessible to the \emph{Full-Resolution Readout} module. On the Tencent Advertising Algo
benchmark - arxiv:2611.19061 · cs.AILarge Language Models for AI InfrastructureBolin Shen, Shaoyi Huang, Yushun Dong
Large language models (LLMs) are increasingly used to design, optimize, and operate the infrastructure behind AI training and serving, yet the literature remains fragmented across hardware design, machine learning systems, electronic design automation, and high-performance computing. We present a comprehensive survey of LLM-assisted AI infrastructure across its lifecycle, from accelerator design and kernel engineering to compilation, deployment, operations, data management, and energy optimization. We organize the literature by the primary resource the LLM manages: artifact-centric methods for computational artifacts, operational-centric methods for deployment context and system state, and resource-centric methods for data and energy. These categories span the lifecycle and cover AI-specific concerns such as distributed-training configuration, inference serving, GPU-cluster scheduling, and data loading. Building on this taxonomy, we conduct a cross-study quantitative synthesis of repor
benchmark - arxiv:2611.19047 · cs.ROCertify, Then Select: Viability-Aware Runtime Steering of Frozen Diffusion PoliciesAngelo Moroncelli, Loris Roveda
Diffusion policies, which generate short action sequences by denoising random noise, are a leading way to learn robot skills from demonstrations. Once deployed, they meet constraints the demonstrations never contained, and retraining for each is impractical. A safety filter edits the policy's action to satisfy them, but the edit ignores the task, and a large edit can leave the policy where it cannot continue. We propose CAST (Certified Action Selection at Test time): draw a few action chunks from the frozen policy, certify each with a barrier condition, and execute the certified chunk that a learned value ensemble ranks highest. We prove that a certified step is safe whatever the critic does; that, whenever a certified sample exists, the executed chunk stays within log N nats of the policy conditioned on feasibility, while a safety-filter projection can be singular to it; and that an exact closed-loop identity splits the gain of ranking into availability, spread and the critic's target
manipulationhumanoid - arxiv:2611.19044 · cs.LGHard or Mislabeled? Failure Screens Enrich Vision Benchmarks for Label ErrorsHaolin Guo
Hard vision benchmarks often keep images that strong classifiers get wrong. Since a classifier is also marked wrong when the label is wrong, such failure screens may collect label errors. We measure this on ImageNet's twice-relabeled validation set with 103 classifiers. Among the 1,892 images that all 66 supervised classifiers get wrong, 61% have a wrong label, versus 11.1% overall. The fewer images a screen lets through, the more of them are mislabeled, along approximately one curve. Fit on one model family, this curve predicts the screens of two held-out model families with a mean error of 2.4 percentage points, and replays of ImageNet-D's and ImageNet-Hard's screens within 0.1 and 6.3 points. On new pools, ImageNet-V2 and CIFAR-10, enrichment at the strictest screens is stronger than the curve predicts: on ImageNet-V2, by human relabeling, 67% of the images that all 66 classifiers get wrong are mislabeled, against 46% predicted. Screens that let models choose among only some of the
benchmark - arxiv:2611.19043 · cs.AIMore Context, Mixed Results: Evaluating LLMs for Identifier Recovery and Code DeobfuscationSimon Queyrut, Valerio Schiavoni, Giuseppe Antonio Di Luna, Daniele Cono D'Elia +2
Source code is often obfuscated through identifier renaming, incentivizing reverse engineers to develop techniques for recovering meaningful names. To assess how LLMs compare with these techniques, we evaluate 4 traditional baselines and 3 LLMs (Qwen, OLMo and DeepSeek) on identifier detection and semantic recovery across 51 recent Java projects, 3 obfuscators, and 6 context/retrieval settings. LLMs generally outperform traditional techniques on both tasks, but benefit from different forms of context and settings: agentic retrieval yields the strongest detection results, reaching 91.56% $F_1$, while enclosing-class context generally improves semantic recovery at a higher token cost. Despite these gains, exact recovery peaks at 16.42% overall (under one-hop context) and 32.93% for field names (under repository retrieval), which are the easiest to recover. Class identifiers are the hardest to detect, with a mean detection $F_1$ of 48.48% across models, settings and obfuscators. More cont
agentic - arxiv:2611.19038 · physics.opticsSpacer-Engineered Spectral Selectivity in Metal-Carbon Solar-Thermal AbsorbersX. Romain, A. P. Vafa, M. Florescu
Spectrally selective solar-thermal absorbers must capture sunlight while limiting emission at operating temperatures where the Planck spectrum overlaps the solar band. We connect this trade-off to a controlled optical mechanism in a metal-carbon metasurface. A radiative-limit calculation using AM1.5D illumination identifies how cutoff wavelength, infrared absorptivity and concentration together constrain equilibrium temperature, while the metasurface calculations reveal the additional role of a sharp spectral transition. We then separate the dielectric gap from the absorbing carbon layer in a gold--carbon architecture. Frequency-domain fields and material-resolved dissipation show that the spacer controls the overlap of a gap-localised mode with carbon, extending the high-absorptivity band and sharpening its infrared roll-off. A thickness sweep demonstrates that the spectrum with the strongest visible absorption need not give the highest radiative-limit equilibrium temperature. Polar-a
benchmark - arxiv:2611.19037 · cs.LGThe Critic Can Hurt: On the Safety of Critics in LLM-Guided AutoML SearchGaoyuan Du, Xiaoyang Liu, Rex Zhou, Dengdu Jiang +3
LLM critics increasingly steer iterative search in agentic AutoML, prompt optimization, and self-refinement--evaluated for whether they help, but rarely for whether they can hurt. We isolate the critic's marginal effect through a four-arm audit and characterize the loop's tolerance through controlled fault injection. We show a naive 8B critic drives search below the no-critic random-search floor on 7 of 8 tabular datasets, losing up to 0.245 AUC in the worst case; the magnitude varies strongly with dataset, frontier, and budget. When: stronger critics attenuate the loss on the evaluated suite, without establishing a universal capability threshold. Why: controlled fault injection shows that non-directional errors are nearly costless under best-so-far search, whereas directional errors consume finite search opportunities; an idealized effective-sample calculation yields a necessary condition, $q(m-1) > β$, not a deployment certificate. How: a simple observable validation gate and a learn
agenticself-refinementbenchmark - arxiv:2611.19034 · cs.RORefinement vs. Replanning: Inference Compute Allocation in Robot PoliciesAlba Maria Tellez Fernandez
A frozen chunked robot policy can spend deployment-time inference compute in two coupled ways: more work within each policy call to refine a predicted action chunk, or more frequent policy calls to replan from fresh observations. We study this allocation without updating model weights. Our evaluation procedure reports success with both nominal compute rate and complete attempt cost, selects comparisons on discovery tasks, and tests prespecified comparisons on held-out tasks. Comparisons added after inspecting the held-out results only summarize observed patterns. We demonstrate the procedure with one frozen SmolVLA checkpoint in synchronous LIBERO simulation, using denoising count as the refinement control and fixing chunk length at K=50. Arithmetic and latency characterize cost differently: at d=10, 82.1% of counted arithmetic is independent of denoising, whereas a descriptive affine fit to fixed-input host time assigns 17.7% to its intercept. At h=2, d=2 succeeds on 139/160 attempts
robot policylibero - arxiv:2611.19029 · cs.AISurveyWeave: An Agentic Framework for Automated Survey GenerationFranklin Zhang, Xiao Yang, Xin Luna Dong, Juhua Hu
Surveys help researchers keep up with fast-moving fields, but they take weeks to months to write. Automated survey generation (AutoSurvey) can alleviate this situation, but it is challenging to draw on hundreds to thousands of precisely scoped papers, choose among many valid structures, and stay faithful to its sources while offering insights. We present SurveyWeave, an agentic architecture that decomposes the problem into four tasks coordinated by three agents, each reviewing others' work and conducting revisions based on feedback. We also introduce WeaveEval, a five-dimension evaluation grounded in prior surveys. A comprehensive comparison shows that SurveyWeave outperforms Codex by 8 points and the strongest AutoSurvey baseline by 6 points.
agentic - arxiv:2611.19027 · cs.LGHeteroWear: Sensor-Aware Self-Supervised Learning for Heterogeneous Wearable Activity RecognitionYasaman Mohammadpour, Pouya Shaeri
Self-supervised learning for wearable sensing is usually confined to a single dataset or device family, leaving open whether representations transfer across heterogeneous sensor configurations and whether gains come from pretraining rather than architecture or richer sensor input. We present HeteroWear, a sensor-aware self-supervised framework pretrained on an eight-dataset wearable corpus (173 subjects, 358,261 windows) spanning inertial and physiological signals across body locations, devices, and sampling rates. HeteroWear separates within-sensor temporal modeling from cross-sensor relational modeling using sensor-context embeddings, temporal and cross-sensor Transformers, and latent cross-attention fusion. Training combines structured sensor dropout, cross-configuration consistency, and per-axis masked sensor modeling; missing sensors are omitted rather than reconstructed. Under leave-one-dataset-out pretraining and subject-disjoint evaluation on PAMAP2, WISDM, HHAR, HARTH, and UCI
benchmark - arxiv:2611.19020 · cs.LGAdvancing the Calibration-Compute Frontier of LLM Classifiers via Asymmetric DuosBerke Akturk, Ali Can Sahin, Rasit Cem Solatay, Emanuel Sommer +2
The importance of reliable uncertainty estimation in large language models (LLMs) has become increasingly pronounced, yet modern models remain poorly calibrated and frequently overconfident. In this study, we address this by proposing an efficient method for LLM classifier calibration using the Asymmetric Duos framework. Our approach combines a strong base model with a smaller sidekick model via logit-space fusion, utilizing global scaling parameters optimized on a validation set. Across diverse benchmarks, this Asymmetric LLM Duo consistently achieves best or second-best performance on expected calibration error and negative log-likelihood, without degrading predictive accuracy. More importantly, these gains are realized at a fraction of the computational cost of more complex baselines, placing our method on the calibration-compute Pareto frontier. We further demonstrate that our method acts orthogonal to various established calibration-aware techniques, meaning it can be easily combi
benchmark - arxiv:2611.19013 · cs.AIMemoryLane: Supporting Reminiscence Conversations Between Older Adults with Mild Dementia and Lay Caregivers Using a Simulated AI AgentSoumay Agarwal, Aaryan Gajula, Lingyao Li, Renkai Ma +4
Reminiscence Therapy (RT) is a non-pharmacological intervention that invites patients to discuss and review past life experiences and memories to improve their emotional and mental well-being. In practice, many reminiscence conversations are led by lay caregivers, such as spouses and adult children. While traditional RT often relies on personal photographs, curating them can be tedious and difficult. In addition, caregivers who lack shared life experiences with the patient may struggle to facilitate engaging reminiscence conversations. This paper presents MemoryLane, an AI-integrated tool that helps lay caregivers facilitate reminiscence conversations without curated photographs. MemoryLane supports a triadic conversation among the patient, the caregiver, and a simulated AI agent that acts as the patient's peer. A shared interface shows the agent and AI-generated images of the memories being discussed, while a private smartphone interface lets the caregiver control the behavior of the
agentai agent - arxiv:2611.19011 · cs.LGTaking Band-Limited Maximum Likelihood to High Dimensions: Linear Memory, the Parametric Rate, and a Self-Tuning Cut-offRahul Agarwal, Ankit Agarwal
Band-limited maximum likelihood (BLML) estimates a density as the square of a band-limited function. The estimate is non-negative, and when the square root of the true density is band-limited its error falls at the parametric rate $n^{-1}$. Its algorithm bins the data, which is fast only in one or two dimensions: from three on every point has its own bin, the kernel system is dense, and memory grows as $n^2$, 40 GB at $10^5$ points. Whether the rate survives in higher dimension, where kernel estimators slow down, had not been tested. We study dense data, densities that fill their space, not data near a low-dimensional manifold. We show that the binned equations are the stationarity condition of a strictly convex objective and minimise it by block coordinate descent without forming the kernel matrix: memory is linear in $n$, a sweep costs $O(n^2 d)$ and a fit takes about ten, and one A100 GPU fits $10^6$ points in eight dimensions in about ten minutes. On band-limited truths the error t
memorybenchmark - arxiv:2611.19009 · cs.CVRole-Separated Diffusion Transformers: Global Rule Isolation and Deterministic Sampling Manifold CompressionZhiguang Liu, Yi Shang
We present Role-Separated Diffusion Transformers (RoSeDiT), a new diffusion transformer architecture that isolates the global rule from the image tokens. RoSeViT showed in visual reasoning that a vanilla ViT spreads a shared rule over all $L$ tokens at an entropy of order $Ld$, whereas an enforced controller route reduces it to order $d$. In class-conditional DiT and SiT, the class and the noise level are explicitly conditioned by adaLN, so a RoSeViT-style controller channel on the ViT side is insufficient to establish the full role separation. By introducing the adaLN listening mechanism, which makes adaLN compute under the controller's command, RoSeDiT brings the entropy theorem and role separation into the latent diffusion domain. Under the same SiT training and evaluation protocol at the same level of parameter count, RoSeDiT surpasses SiT at all four model scales at 400K steps, and RoSeDiT-L/2 even surpasses SiT-XL/2's 17.2 FID with 15.34 and 32% fewer parameters. RoSeDiT-XL/2 sur
evaluation protocol - arxiv:2611.18999 · cs.AIAgent-ZTA: A chain-Aware Zero Trust Model for Bidirectional Security-State Propagation and Runtime Containment in Multi-Agent DelegationAsadullah Abdullah Khan, Toqeer Ali Syed
Multi-agent systems increasingly delegate tasks and permissions across chains of autonomous agents that invoke protected tools. Although agent identities and delegated credentials may remain valid throughout a workflow, the security condition of any agent along the chain can deteriorate during runtime execution-either through upstream compromise or downstream ingestion of untrusted inputs. Authorization models based solely on the immediate caller and its static delegated scope fail to account for these mid-execution state changes. This paper develops Agent-ZTA, a chain-aware Zero Trust authorization and containment model that links runtime security-state changes across multi-agent delegation graphs. The model distinguishes an agent's path-specific delegated authority bound from the effective permissions currently exercisable under trusted policy. It defines an architecture featuring independently enforced policy decisions, authenticated delegation provenance, a dynamic chain-state view
agentautonomous agentmulti-agentagent system - arxiv:2611.18998 · cs.AIModel Maps: A Documentation Framework for Tracing Foundation Model Ancestry and Understanding AI EntanglementRachel Hong, Madeleine Grunde-McLaughlin, Sarah Cen, Jevan Hutson +2
The development of foundation models incorporates pre-existing AI artifacts like models, datasets, and benchmarks. These relationships, as part of the broader AI supply chain, produce "entangled" webs in which properties of artifacts upstream can influence the outputs of a foundation model downstream - a phenomenon known as AI entanglement. Without visibility into these connections, a black-box model can inherit upstream properties that result in security vulnerabilities, hallucinations, or biased predictions, yet developers are left with no recourse to identify or mitigate the root issue. To address these concerns, we introduce model maps as a conceptual documentation framework for tracing model ancestry that builds on previous documentation frameworks like ecosystem graphs, datasheets, and model cards. A model map for a given foundation model is a detailed knowledge graph that represents the relationships between artifacts that influence the resulting foundation model. Our framework
knowledge graphbenchmark - arxiv:2611.18996 · cs.LGRevealing the Biases of Large Language Models By Inferring Prior DistributionsJoshua Ryo Cheng, Thomas L. Griffiths, Jake C. Snell
Large Language Models (LLMs) have been known to exhibit a wide range of social biases including name-nationality associations, occupation-gender associations and racial stereotypes. Existing approaches to evaluating these biases typically involve measuring surface-level behavior where a model is fed a dataset of prompts and its outputs are scored relative to ground-truth answers. Inspired by prior elicitation in cognitive science, we propose a different perspective that reconstructs latent bias as a prior underlying the model's behavior. We formulate the problem as a Bayesian inverse problem where, given prompt-response observations on a bias benchmark dataset, we infer a low-dimensional latent distribution that summarizes the model's stereotyping tendencies. This paper develops the mathematical foundation, a practical reconstruction algorithm, and experiments using the Bias Benchmark for QA (BBQ) benchmark to infer the prior distributions of LLMs over responses in ambiguous social con
benchmark - arxiv:2611.18994 · cs.LGLarge language models as oncology world modelsJulien Vibert, Jakob Nikolas Kather
Large language models (LLMs) can synthesize cancer patient cases, but their internal representation of oncology, or world model, has not been explored systematically. By only specifying tumor type, we show that LLMs tend to generate similar patients per cancer, with on average 81.0% to 93.2% of a cancer's patients sharing the most frequent value of each baseline feature. By systematically varying patient features in 12 clinical scenarios, we show that they follow biomarker and treatment logic in 29 of 36 model-scenario assessments, against 1 of 12 when previous treatment or frailty matters. Continuing real patient trajectories, an LLM predicted 89.9% 12-month survival, and 79.2% with similar patients as examples, against 72.2% observed. To represent these trajectories, we developed the Clinical State Space, a dynamic representation framework that we also applied to real patients. Despite the limits we show, improved world models may support digital twins, which our framework can evalua
world model - arxiv:2611.18982 · cs.RORobot Demonstration Videos Should Include Transparency DisclosuresNozomi Nakajima
This paper argues that the robotics and AI community should adopt disclosure standards for demonstration videos. As physical and financial constraints limit direct public access to embodied AI, public dissemination of progress in robotics is mainly through curated demonstration videos rather than direct experience with the technology. In a large-scale (N=9,341), pre-registered experiment across six countries, we find that the general public struggles to discern both the visual authenticity (e.g., distinguishing real footage from CGI) and behavioral authenticity (e.g., distinguishing autonomy from teleoperation) of robot demonstration videos. The omission of these details distorts public perception, leading people to overestimate the ability of robots to perform tasks and navigate real world environments. Using evidence from behavioral science, we design a disclosure intervention and demonstrate that the inclusion of clear disclosures in robot demonstration videos significantly improves
embodiedteleoperation - arxiv:2611.18981 · cs.CLGPBam: A Retrieval-Aware German Public Law Exam BenchmarkAbdullah Al Zubaer, Lorenz Wendlinger, Simon Alexander Nonn, Michael Granitzer +1
Legal reasoning in civil-law systems is not rule-lookup: the German state examination asks for a structured opinion (Gutachten) that names the governing norms, applies them to the facts, and argues to a conclusion. We introduce GPBam (German Public Law Bar-style Exam), 81 public-law examination cases with reference solutions, paired with a recitation task over 200 statutory provisions. Drawn from state examination preparation, the cases are broad and rarely repeat, and are not publicly available - we find no evidence of exact-solution memorisation in the models we evaluate, and keeping them private limits contamination. Because each reference solution's cited norms form a retrieval gold set over the very instances the essays are graded on, GPBam scores retrieval and long-form generation jointly, which makes it a benchmark for legal retrieval as well as for legal writing and lets us ask how much retrieval quality a legal opinion actually needs. Across 32 open-weight and proprietary mode
benchmark - arxiv:2611.18979 · cs.CVProbing Where Surprise Happens: Localized Violation-of-Expectation Evaluation of Video World ModelsYanbin Xu, Ming bo Cai
Violation-of-expectation (VOE) benchmarks evaluate intuitive physics in video world models by testing whether physically impossible events produce greater prediction errors than matched possible events. However, conventional VOE evaluations aggregate prediction errors across space and time, making it unclear whether successful discrimination reflects model's understanding of physical violations at the locations and time where they occur. To address this limitation, we introduce localized VOE evaluation, a framework that measures prediction error differences between possible and impossible events within spatial regions and temporal intervals associated with physical violations. We compare global, temporally localized, and spatiotemporally localized evaluations of V-JEPA models pretrained on developmental egocentric videos or large-scale natural video datasets. Our results reveal that global and localized judgments can diverge in both directions: global aggregation can either obscure cor
world modelv-jepabenchmark - arxiv:2611.18976 · cs.ROTMA: Sim-to-Real Forceful Humanoid Manipulation via Tactile-Driven Motor AdaptationYiting Chen, Kelin Yu, Sarthak Kaingade, Bryan Trinh +4
Recent humanoid loco-manipulation in articulated environments has largely focused on geometry-dominant interactions, where success is primarily determined by geometric alignment and motion execution. In contrast, forceful multi-finger manipulation requires the robot to adapt to real-world uncertainties in contact and interaction dynamics. We present Tactile-Driven Motor Adaptation (TMA) and an integrated whole-body system for zero-shot sim-to-real forceful humanoid manipulation, and study round-doorknob opening as a challenging benchmark requiring stable multi-finger frictional grasping, sufficient torque generation without slip, force-regulated pushing, and whole-body stability. TMA structures policy learning around interactions observable through tactile feedback, uses cone geometry as an inductive bias for multi-finger contact representation, and adapts motor behavior online from tactile history. An upper-body manipulation policy is trained alongside a pretrained general locomotion
manipulationdexteroushumanoidtactilesim-to-realgrasp - arxiv:2611.18974 · cs.LGBilateral Log-Linear AttentionKazi Ashik Islam, Guojing Cong
Sequence models represent past context at different resolutions, using token-level keys and values, fixed-size recurrent states, or hierarchies of compressed summaries. Full attention supports strong associative recall but requires quadratic work during training and prefill and a linearly growing key-value cache during inference. Linear attention and state-space models reduce these costs through fixed-size recurrent memory, but this compression can limit recall. Log-linear attention offers a middle ground by maintaining logarithmically many state matrices. However, its hierarchy favors finer resolution near the current query while compressing earlier context into coarser summaries. To complement this allocation of resolution, we introduce bilateral log-linear attention, which pairs the existing hierarchy with a second hierarchy anchored at the sequence origin. By combining their reads, the two branches provide finer access to both early and recent context while remaining causal. We der
persistent state - arxiv:2611.18963 · cs.LGSpecRelay: Amortize or Avoid the Round Trip in Edge-Cloud Speculative DecodingJai Jain, Chihyo Ahn, Yong Liu, Charles Chen +9
Edge-cloud speculative decoding moves autoregressive drafting to an edge device while a larger cloud model verifies draft tokens in parallel, so that edge drafting and cloud verification overlap under a synchronous draft-verify schedule. Beyond low-latency local networks, however, the next verification request of that schedule can be sent only after the previous reply arrives, so every round still waits out a full round trip, and throughput falls as the network RTT grows. To recover this loss, we present SpecRelay, an edge-cloud LLM inference system that addresses this cost at two levels: it amortizes round trips when cloud verification is necessary and avoids them when local reasoning suffices. For cloud-bound queries, an asynchronous verification pipeline keeps fixed-budget draft trees in flight while earlier replies are still outstanding, and the cloud rejects stale requests before the target-model forward pass (the asynchronous schedule). At the query level, an RTT-conditioned rout
benchmark - arxiv:2611.18956 · cs.LGSupercharging Olmo-core for Efficient and Scalable MoE TrainingTianhua Tao, Akshita Bhagia, Dirk Groeneveld, Pete Walsh +7
Mixture-of-Experts (MoE) models add parameters without proportionally increasing per-token compute. However, training cost also depends on bytes moved, memory held, and kernels launched, and many of those costs follow total parameter size rather than active compute. On infrastructure built for dense models, this mismatch is costly: in early baselines, throughput fell as stored expert parameter size grew, even though the useful work per token stayed nearly fixed. To close this gap, we present a redesigned training stack in Olmo-core, built for efficient, scalable MoE training. We replace the Fully Sharded Data Parallel (FSDP) of earlier Olmo-core versions with an architecture built on Distributed Data Parallel (DDP), and add Expert Parallelism (EP), Pipeline Parallelism (PP), a distributed optimizer, and activation recompute. We propose NVSHMEM-based rowwise EP, which writes each token directly into its expert's buffer, avoiding dispatch and combine permutation overhead. We make EP full
memory - arxiv:2611.18935 · cs.ROOptimal Operating-Point Selection for Single-Linearization Trajectory Planning of Nonlinear Robot ManipulatorsBhanuka Dayawansa, Rohan Munasinghe
Single-point linearization provides a computationally efficient basis for optimization-based trajectory planning of nonlinear robot manipulators. However, planning performance depends on the selected linearization point. This paper presents an offline performance-driven approach that treats the location of a single linearization as an optimization variable. Candidate operating points are sampled from a common cubic seed trajectory, and each corresponding affine model is used to solve an identical constrained finite-horizon quadratic program (QP). The resulting control sequences are executed on the original nonlinear dynamics, and the operating point minimizing the nonlinear trajectory objective is selected through grid search and bounded refinement. Evaluation on four point-to-point motions of a three-degree-of-freedom UR5 manipulator yields optimal normalized operating-point locations between 0.325 and 0.935, demonstrating trajectory-dependent selection. Compared with the best-perform
manipulator - arxiv:2611.18930 · cs.AIBetween Rigor and Reality: How AI Safety Benchmark Developers Understand Benchmark Use and MaintenanceMegan Li, Ningjing Tang, Christopher Stewart, Hoda Heidari +1
Safety benchmarks are cited as the primary tool to understand the risks posed by generative AI (GenAI) systems, yet growing evidence suggests they often fail to meet the needs of real-world safety evaluation. We present findings from five interviews with AI safety benchmark developers in academia and industry. We find a disconnect between perceptions of the intended usefulness of safety benchmarks and their practical usefulness in safety evaluations of deployed systems. Our participants offered perspectives on two challenges contributing to this disconnect: (1) the difficulty of achieving inter-rater reliability on safety constructs, and (2) a lack of clarity regarding how to address persistent threats to external validity. Based on our analysis, we argue that safety benchmarks must not only be grounded in deployment contexts, but actively integrated with them. To facilitate this, we outline pathways for more transparent communication between academic and industry AI safety stakeholder
benchmark - arxiv:2611.18919 · cs.ROCorrecting Persistent Failures in Point-Cloud Grasping Policies using Human Demonstrations in Virtual RealityAchyuthan Unni Krishnan, Karthik Rengarajan, Bharath Narasimhan, Chris von Csefalvay
Robots in homes, warehouses and farms must pick up objects they have never seen, usually without a 3D model or a category label. A grasping policy that works directly on the point cloud from a depth sensor needs only the geometry the robot observes, so it applies to any new object. Offline grasp generators such as Contact-GraspNet work this way, but they rank grasps by learned plausibility rather than by whether the grasp actually lifts the object. Reinforcement learning (RL) can optimize for that directly and make a point-cloud grasping policy much better on average, yet it leaves specific grasps persistently unsolved. We present a human-in-the-loop failure loop that targets those grasps: situations an RL policy keeps failing are mined automatically, sent to a virtual-reality application where a person picks the object up with their own hands, and returned to training as two grasp points. The person contributes only where to grasp; the robot chooses its own approach. In simulation, RL
grasphuman-in-the-loop - arxiv:2611.18916 · cs.LGSimplify, Agree, Distill: Label-Free Self-Distillation with Stepping StonesDingyang Chen, Jinfeng Xiao, Zhenyu Zhang, William Hu +7
Can a language model teach itself to reason without labels? On-policy self-distillation (OPSD) uses the model as its own teacher, but strengthens that teacher with a ground-truth solution. The natural label-free substitute, voting over the model's attempts, fails precisely on hard problems: attempts disagree, and their most common answer is often wrong. We propose STONE, a label-free alternative that moves agreement to easier related problems. The model writes these stepping stones, keeps those it solves consistently, and shows the teacher their worked solutions, with final answers removed; the student sees only the original problem. No step reads a label, and inference needs no extra context or model calls. Across five math benchmarks, the proposed method achieves the highest average accuracy for Qwen3-4B and Qwen3-8B in thinking mode. Without thinking mode, it gains 13.2 and 22.7 points, 2.5 and 2.7 times the gains of OPSD with ground-truth solutions.
benchmark - arxiv:2611.18914 · cs.LGHow Much Do Foundation Models Add to Gradient Boosting on Financial Event Sequences?Carlos Eduardo Martins Relvas, Gustavo di Giovanni Bernardo
Foundation models pretrained on raw transaction sequences are proposed for credit, fraud and churn modeling, with reported gains over gradient boosting, but the data and baselines behind these gains are rarely public. We measure how much the pretraining and adaptation recipes of these models add to a baseline of point-in-time window aggregates and LightGBM, and we propose TBS (Temporal Boosting-Sequence ensemble), which averages the ranks of a boosting model enriched with label-free event-transition features and of a pretrained sequence network fine-tuned jointly with the aggregates. On a synthetic credit-card dataset, the public MBD-mini banking dataset and two proprietary datasets of trade credit and B2B churn, we compare 27 methods at three label budgets with five seeds and paired bootstrap intervals, using sequence models of about 0.4 million parameters pretrained per dataset. Frozen self-supervised embeddings rarely improve the baseline, and a fine-tuned sequence network alone cle
benchmark - arxiv:2611.18908 · cs.AIMemeEconomy: Do LLM Agents Trade Ethics for Survival?Syed Nazmus Sakib, Nafiul Haque, Ahnaf Tahmid Manan, M M Mahabub Morshed +1
Alignment-trained large language model (LLM) agents are increasingly deployed in autonomous, multi-round settings where ethical compliance can come into direct tension with task success. While safety literature extensively documents failure modes like sycophancy and reward hacking, it largely overlooks a critical vulnerability: how agents behave under simulated existential economic pressure. We introduce MemeEconomy, a multimodal agentic market simulation in which LLM agents act as meme investors. Agents select content from a five-tier harm taxonomy and target communities under low-pressure and high-pressure (survival) environments. Each decision is logged in a Belief-Desire-Intention (BDI) schema that we use as a diagnostic instrument, externalizing the agent's event understanding, declared priorities, and harm acknowledgment. Across ten models from four frontier model families, harmful selection rates remain consistently high and rise by +15 percentage points on average under an econ
llm agentagenticbenchmark - arxiv:2611.18894 · cs.CVCISyn: Human Mesh Recovery in Close Interactions with Physics-Simulated Synthetic DataYufan Deng, Lorenza Prospero, Christian Rupprecht
3D human mesh recovery (HMR) in close-interaction scenarios is hindered by the lack of large-scale, accurately annotated training data: real capture is expensive, and existing synthetic datasets rarely include physical contact. We introduce CISyn, a large-scale synthetic dataset purpose-built for close human-human interactions, generated via physics simulation and Blender-based rendering. By simulating collision dynamics between SMPL-X-rigged humanoids, CISyn naturally produces the contact, mutual occlusion, and pose entanglement that characterize close-interaction scenes, with no manual annotation effort. The released set provides 25,000 sequences with accurate 3D annotations. We further propose an uncertainty-aware loss that allows models to learn more effectively from the heavily occluded examples in our data. Augmenting state-of-the-art HMR training with CISyn and replacing the regression loss with our uncertainty-aware variant yields consistent improvements on close-interaction be
humanoidbenchmark - arxiv:2611.18891 · cs.CVBring Back Language to Actions: Restoring VLM capabilities in VLA Models via Parameter MergingMohamed Salim Aissi, Olivier Sigaud, Laure Soulier, Nicolas Thome
Vision-Language-Action models (VLAs) are obtained by fine-tuning pre-trained Vision-Language Models (VLMs) on robot demonstration data, enabling action prediction from visual observations and language instructions. However, this fine-tuning induces catastrophic forgetting, substantially degrading the semantic and visual grounding capabilities of the original VLM. We address this by merging the parameters of a VLM and its VLA derivative without retraining either model. Concretely, we extract the task vector (the difference between VLA and VLM weights) and decompose it via Singular Value Decomposition. For each component, we learn a scalar mixing coefficient controlling how much of the VLA adaptation is retained, optimised to jointly preserve action prediction and vision-language capabilities. Experiments show that our merged model performs well on both axes simultaneously, consistently outperforming established merging baselines.
vision-language-actionvlavla model - arxiv:2611.18890 · cs.AIBenchmark Factory: Agentic Benchmark Creation via Dataset Curation and Label-Free Quality AnchorsShervin Shahidi, James Poon, Hang Xie, Kenneth O. Xu +4
Teams deploying coding agents need benchmarks built from their own code, but curating and validating them by hand is resource intensive, and an unvalidated benchmark can mislead silently. We present Benchmark Factory, an autonomous meta-agent loop with human checkpoints: an LLM agent writes sourcing queries over landed production code changes, runs models of known relative strength on each candidate, diagnoses why a candidate fails, and repairs it by editing its queries, prompts, and configuration until it clears quality anchors that need no per-task human labels. We used Benchmark Factory to build benchmarks for five developer cohorts at our organization. Engineering raters scored a 5% sample 2.3 for task quality and 2.6 for grader verdicts on a 1-to-3 scale. On configurations never used to build them, the benchmarks ranked Claude Opus~4.7 at high reasoning effort above low effort (60.6% vs. 54.4% of tasks passed, $p=0.002$) and GPT-5.5 above GPT-5.4 (65.8% vs. 60.5%, $p=0.004$); 9 of
agentllm agentagenticbenchmark - arxiv:2611.18887 · cs.AIWe Have All the Cards: Securing Agent Cards at the Operational Trust Boundary of the Agent2Agent ProtocolMiguel A. Guirao Aguilera, Satyajayant Misra, Efren Lopez-Morales
AI agent interoperability protocols enable orchestrators to discover and invoke external agents using machine-readable metadata. In the Agent2Agent (A2A) protocol, Agent Cards contain attacker-controllable natural-language fields that may influence agent selection before a provider is invoked, creating a pre-invocation trust problem. We characterize this discovery-time attack surface through a taxonomy of ten Agent Card manipulation attacks and present SECCARD, a layered validation framework combining protocol checks, deterministic analysis, and LLM-based semantic reasoning. We measure the emerging public A2A ecosystem by probing 36.1 million Internet domains and collecting 310 deployed Agent Cards. Two annotators independently label the corpus with 96.8% agreement. We then construct a 660-card real-card perturbation benchmark with matched benign controls and a 260-card synthetic adversarial benchmark spanning all ten attack categories. On controlled attacks injected into real Agent Ca
manipulationagentai agentbenchmark - arxiv:2611.18883 · cs.AISPLICE: Structured Prompt Local Iterative Combinatorial EvolutionAnish Acharya, Phillip Studans, Amit Dhanda, Sina Amini Niaki +3
Natural language optimization (NLO) improves an agent by having a model critique and rewrite its natural language policy parameters: system prompts, tool instructions, reasoning directives, few-shot demonstrations. These artifacts are multi-section documents in which some sections encode contracts (safety clauses, tool schemas, output formats) that must not drift, while others are legitimate optimization targets. Most iterative NLO methods treat the artifact as a single editable string, producing three recurrent failure modes: unsafe global rewrites, beam collapse onto a single lineage, and unchecked bloat, none addressed by formal guarantees. We introduce SPLICE, a unified NLO framework that poses optimization as constrained combinatorial search over an edit graph of block-structured, selectively-mutable states. We map each failure mode to a matched guarantee: (i) per-block mutability preserves contract text verbatim, with a price-of-freezing bound; (ii) elitist beam search with diver
agentbenchmark - arxiv:2611.18880 · cs.CVRAVEN-Bench: A Paired EO-IR Video QA Benchmark for Aerial Multimodal UnderstandingYu Hu, Jianyang Gu, Yue Cao, Hao Liu +4
Multimodal large language models (MLLMs) have advanced rapidly in image and video understanding, yet their ability to reason over paired aerial electro-optical and infrared (EO-IR) videos remains poorly understood. Such video poses a harder test than standard RGB video QA: targets are often small, evidence may appear only briefly, viewpoints change with the platform, and useful cues may appear in only one stream. We introduce RAVEN-Bench (Reasoning over Aerial Videos with EO-IR aligNment), a paired EO-IR video QA benchmark for low-altitude aerial multimodal understanding. RAVEN-Bench contains 100 temporally aligned EO-IR video pairs and 1,200 human-authored four-way multiple-choice questions, each independently audited for answer keys, modality requirements, evidence intervals, and group validity. It supports EO-only, IR-only, and paired EO+IR evaluation, and uses hierarchical capability levels and structured question groups to diagnose paired-input gain beyond the stronger unimodal co
benchmark - arxiv:2611.18869 · cs.CVUnSPAD: Uncertainty-aware Single Photon Burst PhotographySachin Shah, Kalyan Bhetwal, Levi Burner, Christopher A. Metzler
We introduce UnSPAD, an uncertainty-aware burst photography framework capable of fusing and denoising arbitrarily long sequences of binary single photon avalanche diode (SPAD) measurements. Unlike prior methods that require fixed burst lengths or rely on powerful generative priors, UnSPAD combines learned motion estimation with uncertainty-informed measurement fusion and denoising to produce temporally consistent, hallucination-free reconstructions. UnSPAD first forms an initial scene estimate by averaging flow-aligned binary frames. UnSPAD then uses per-pixel uncertainty maps to perform uncertainty-aware denoising; the denoiser is informed about which regions accumulated the fewest observations (and thus contain the most relative shot noise) due to motion, occlusion, or alignment errors. This two-step reconstruction pipeline helps mitigate hallucinations---outside of the structure implicitly introduced by the flow estimates, the initial scenes estimates are entirely prior-free. Experi
benchmark - arxiv:2611.18868 · cs.LGSLICES: A Benchmark and Evaluation Framework for Self-Supervised Learning on Sparse ICU Time SeriesHannes Ill, Dimitrios Proios, Leo Anthony Celi, Peter J. Schüffler
Self-supervised learning (SSL) results for intensive care unit (ICU) time series are hard to compare when data processing, encoders, and adaptation vary together. We present SLICES, a benchmark and evaluation framework for comparing and extending SSL methods under a shared clinical setup. It fixes 24-hour value and observation-mask inputs, a Transformer encoder, patient-level splits, four endpoints, two adaptation protocols, and five matched seeds across MIMIC-IV, eICU-CRD, and their pooled union. Masked autoencoding (MAE), joint-embedding prediction (JEPA), and contrastive learning provide reference implementations. Among these objectives, MAE leads mean AUPRC in all nine frozen classification comparisons, also under converged probes, and in seven of nine fine-tuning comparisons. Fine-tuned MAE leads hospital mortality at full labels and scores higher than a supervised Transformer of the same architecture at every label fraction, and SSL fine-tuning gives the lowest length-of-stay err
benchmarkevaluation framework - arxiv:2611.18866 · cs.AIDécoupage: Directorial Audio-Visual Control for Long Multi-Shot Video GenerationYifan Wang, Hao Ouyang, Qiuyu Wang, Yihao Meng +8
In cinematic storytelling, sound and visual transitions do not always coincide in lockstep: dialogue frequently bridges across cuts to reveal reactions or introduces new scenes via J- and L-cuts, while deliberate silences pace the scene. Existing audio-visual generative frameworks, however, lack explicit mechanisms to coordinate independent visual and acoustic timing over extended horizons. As narrative duration and shot complexity scale, these systems struggle to maintain long-range dialogue pacing, often hallucinating speech during intended silences or erroneously driving mouth movements on silent listeners. To enable fine-grained cinematic control, we present DECOUPAGE, a screenplay-driven framework for long-form, multi-shot, and multi-character joint audio-video generation. Rather than coupling modality boundaries by default, DECOUPAGE uncouples generation into independently bounded visual and acoustic tracks governed by persistent character identities. DECOUPAGE builds on a joint
benchmark - arxiv:2611.18865 · cs.CVFiReIQ: Field-Level Confidence Estimation with Reasoning and Image Quality in Document ExtractionXiao Liu, Yage Wang, Qi Yu, Giri Iyengar
Vision-Language Models (VLMs) have demonstrated strong performance in structured document extraction, yet their deployment in high-stakes applications requires reliable confidence estimates for individual fields. Existing uncertainty estimators often aggregate signals across an entire generation, obscuring field-specific reasoning uncertainty while overlooking visual degradations that may lead to confident extraction errors. To address these limitations, we introduce FiReIQ, a field-level confidence estimation framework that combines field-aligned reasoning signals from a single VLM generation with local and global image-quality features through quality-conditioned visual fusion. To facilitate systematic evaluation, we further introduce FiReIQ-Bench, a benchmark spanning financial receipts and legal contracts. Experiments with two reasoning VLMs show that FiReIQ consistently improves confidence discrimination, selective prediction, and probabilistic accuracy over existing estimators. T
benchmark - arxiv:2611.18859 · cs.LGExstill: The Most Efficient Single-Chip Processor for the Next TokenValavan Manohararajah, Vithursan Thangarasa
Generating a token with a language model requires a fixed amount of arithmetic, whose energy sets a floor beneath any hardware. For Qwen3-1.7B at a 16K-token context, gate-level measurements at 5 nm place this floor at 0.89 mJ per token, yet an NVIDIA B300 spends 391-1,120x that. The excess is the price of generality, software layers, programmable hardware and moving weights and key-value (KV) caches from external memory, and it widens with each process generation. We present a co-design approach that brings token generation close to this floor and then lowers the floor itself, optimizing silicon and model together in a gate-level search loop. Fixing one model architecture in the datapath, with weights still loadable, and holding weights and every conversation's KV cache on-die, a pre-silicon 5 nm design brings an unchanged Qwen3-1.7B token to 3.4x its floor. Co-designing the model with the circuits at fixed quality lowers the floor to 0.29 mJ and the token to 0.52 mJ, 1.8x the new flo
external memoryagentbenchmark - arxiv:2611.18856 · cs.LGSelf-Improvement Memory Enables Continual Recursive Self-ImprovementXin Yu, Fei Xia, Jiaqi Pan, Youbao Tang +6
Harness optimization is often studied with a fixed policy model and benchmark. When either changes, the optimizer must decide which parts of an evolved harness to reuse or revise, but the harness alone may not explain why earlier edits were made or how they performed. In this work, we study Continual Recursive Self-Improvement (CRSI), a setting for continuing existing harness-optimization methods across compatible benchmarks or policy models. We propose Self-Improvement Memory (SIM) for the harness optimizer, rather than the task-solving policy. SIM groups edits by recurring harness problem and records their rationale, predicted and observed outcomes, and evaluation context. New attempts extend this history as optimization continues. In our experiments across four domains, SIM achieves higher final scores than harness-only and root-cause-only continuation in cross-dataset and pairwise cross-model comparisons. Across ten independent source runs in coding, paired continuations with SIM i
memoryself-improvementbenchmark - arxiv:2611.18851 · cs.AIKUDEX: A Zero-Shot Decompositional Framework for Logical Analysis in Natural LanguageZixu Zhao, Tianhang Zhang, Tong He, Zheng Zhang
While Large Language Models (LLMs) excel at generating fluent text, they often struggle with multi-step logical reasoning with inferential connections among statements. We introduce KUDEX (Knowledge-Upward-Downward Reasoning Extension), a novel zero-shot, decompositional framework for analyzing and verifying logical relationships in natural language. Unlike traditional approaches that rely on surface-level entailment checks, KUDEX leverages LLMs to decompose reasoning into atomic primitives and constructs transparent reasoning graphs that serve both as execution traces and audit logs. Our framework includes a general prompt adaptation pipeline that adapts to diverse reasoning tasks without additional fine-tuning or domain-specific pipelines. Through comprehensive experiments on benchmarks across four domains, we demonstrate that KUDEX provides superior performance compared to existing methods, offering a more granular and robust framework for assessing logical relationships between sta
benchmark - arxiv:2611.18849 · cs.CVScaling Channel-Adaptive Self-Supervised LearningAlice V. De Lorenci, Seungeun Yi, Theo Moutakanni, Piotr Bojanowski +3
Recent advances in self-supervised pre-training of foundation models for natural images have made them a popular choice for various visual systems and applications. Self-supervised strategies are also promising in non-RGB scientific imaging domains such as in biology, medical and satellite imagery, but their broader application is hampered by heterogeneity in channel composition and semantics between relevant datasets: two datasets may contain different numbers of channels, and these may reveal distinct aspects of an object or scene. Recent works on channel adaptive strategies report substantial advantages for those that account for variable channel compositions without sacrificing the ability to jointly encode channels; yet, how these strategies behave at scale remains unclear. We here show that, surprisingly, trained across large-scale datasets, independent-encoding of channels outperforms joint-encoding methods by a substantial margin. We validate this result along an extensive set
benchmark - arxiv:2611.18822 · cs.CVLongCat-Video-Avatar 1.6 Technical ReportMeituan LongCat Team, Xunliang Cai, Yiqi Chen, Feng Gao +7
Recent advances in audio-driven avatar video generation have enabled increasingly realistic synthesis, yet achieving precise lip-audio synchronization, natural, human-like motion, and visual consistency over long sequences remains challenging. We present LongCat-Video-Avatar 1.6, which builds on LongCat-Video-Avatar 1.5 to improve expressive naturalness and reduce error accumulation during extended generation, e.g., exaggerated lip movements and overly red, temporally unstable lip coloration, enhance the naturalness of facial expressions and coordinated head and body movements. These improvements are driven by a refined post-training pipeline that integrates reinforcement learning, improved classifier-free guidance strategies for DMD-based few-step distillation, long-video rollouts, and on-policy distillation. Together, these refinements improve expressive realism and enable the generation of videos exceeding 10 minutes while maintaining visual quality and temporal consistency. Arena e
post-trainingarena - arxiv:2611.18818 · cs.LGWhat to Practice Next? Latent Profile-Guided Task Selection for Skill-Evolving LLM AgentsQijin Zhang, Hanzhen Wang, Yan Huang, Zhongxiang Dai +1
Task selection shapes the experiences available for skill evolution in LLM agents, yet fixed practice schedules overlook differences in skill requirements and evolving task value. Most existing methods, however, rely on fixed schedules or scalar measures of progress, leaving nonstationary skill demands unaddressed. We propose SPG-Select, a framework for allocating limited practice budget using a multidimensional profile inferred from binary outcomes via MIRT. The framework predicts surrogate progress, defined as the difference between consecutive profile estimates, and prioritizes progress in dimensions with larger estimated proficiency gaps. Our SPG-Select-R policy combines these predictions with uncertainty-based exploration and periodic restarting to accommodate nonstationarity. We evaluate SPG-Select on ALFWorld and WebShop with three skill-evolution backends. In the reported runs, SPG-Select-R achieves higher mean held-out terminal success than uniform sampling in all six backend-
llm agentbenchmark - arxiv:2611.18811 · cs.CVCARET: Clinical Assertion-Aware Residual Calibration for Medical Image-Text AlignmentYifan Jin, Tao Liu
Medical vision--language models align radiographs with free-text reports, but radiology language distinguishes explicit findings, negation, uncertainty, and omission. Treating these report states as interchangeable during feature adaptation can change image predictions without supporting visual evidence. We introduce CARET, a parameter-efficient image-side adaptation method that retains pretrained image and text encoders and uses paired positive and negative clinical statements to define a low-dimensional semantic correction space. A bounded residual adjusts image features along these directions, while assertion-aware supervision distinguishes explicitly present and absent findings from uncertain or unmentioned ones. At inference, the adapter requires only an image; textual references remain fixed. In zero-shot classification, CARET achieves 67.80\% accuracy on CheXpert 5$\times$200 and 65.20\% on RSNA. These scores exceed the published CLEFT-LoRA results by 1.06 and 0.27 percentage po
evaluation protocol - arxiv:2611.18809 · cs.LGLEAP-LLM: Preserving lead-time structure for question-conditioned ECG interpretationAakash Agarwal, Yidan Shen, Xin Fu, Renjie Hu
The electrocardiogram (ECG) records cardiac electrical activity across multiple leads and is used for detecting cardiac abnormalities. Recent ECG-language models enable large language models (LLMs) to interpret cardiac signals and answer clinically relevant questions. However, existing ECG-language models typically encode the ECG independently of the clinical question, forcing the ECG representation to retain broad and potentially irrelevant signal information. Moreover, these models often aggregate information across leads which can obscure lead-specific details needed for a particular question. To tackle these limitations, we introduce LEAP-LLM, an ECG-LLM architecture for Lead-aware ECG Alignment with Prompt conditioning. LEAP-LLM uses the clinical question to guide ECG feature extraction while preserving explicit lead and temporal structure. Task-trained instruction representations distill information from the question and guide two complementary branches. The Lead branch captures
benchmark - arxiv:2611.18807 · cs.LGRecovered or Memorized? Investigating Dynamical Consistency and Faithfulness in World Model LatentsJerome Francisco, Kyle Deng, George Flint, Kevin Zhu
World models develop internal simulators that can learn the structure and dynamics of an environment. It has been demonstrated that their internal representations are both meaningful and causal, but are these representations faithful to the governing dynamics of the environment they are learning? We introduce Anchored Recovery Certification (ARC), a procedure that utilizes a discrete Fréchet distance calculation with ground truth to evaluate dynamical consistency on candidate world models up to an affine gauge. We use the procedure on TD-MPC2 (decoder-free world model) that has learned the pendulum swing on an observation lift under varying hyperparameters at k = 5, 10, 25 steps. The model is then steered within a PCA-derived intervention subspace and compared against a ground-truth simulator using relative error to investigate dynamical faithfulness. ARC sufficiency held for TD-MPC2 most consistently at k = 10, with six of seven constraints statistically significant after Holm-Bonferr
world modeldreamerv3 - arxiv:2611.18803 · cs.LGClustering-Based Artificial Tree Ensembles for Random Forest ApproximationLea L. Kronziel, Maren S. Wieder, Silke Szymczak, Björn-Hergen Laabs
Random Forests (RFs) offer strong prediction performance but are difficult to interpret. Most representative trees (MRTs) and artificial representative trees (ARTs) provide surrogate explanations, but a single tree may miss patterns within a forest. We investigate benefits and limitations of clustering-based representative tree ensembles for RF surrogate modeling and assess the contribution of similarity-based clustering relative to random tree partitions. Using linear and nonlinear simulations and regression benchmarks, we evaluate fidelity, prediction accuracy, complexity, and prediction stability against equally sized ensembles of top-ranked MRTs and randomly selected trees. Increasing ensemble size generally improves fidelity, accuracy, and prediction stability. ART ensembles show higher fidelity in simulations and often greater prediction stability for small ensembles. Compared with random tree partitions, similarity-based clustering generally improves ART fidelity in simulations
benchmark - arxiv:2611.18802 · cs.LGActuarial-Informed Neural Networks: Constrained Deep Learning for Sex-Specific Multi-Population Longevity ForecastingDavide Rindori
We introduce Actuarial-Informed Neural Networks (AINN), a constrained deep learning framework for sex-specific multi-population longevity forecasting that embeds actuarial knowledge into LSTM training through differentiable penalty terms. Applied to a high-longevity 6-country cluster (Switzerland, Sweden, Norway, West Germany, Netherlands, Japan; 1956-2020), the framework is a challenger to the Li-Lee benchmark and unconstrained neural approaches. Our central finding is that monotonicity constraint intensity lambda functions as a continuous governance dial with a quantifiable accuracy-stability trade-off. At lambda = 0.01, 30-year forecast confidence intervals narrow by 6.5% at zero RMSE cost. At lambda = 0.1, they narrow by 25% at only +0.84% RMSE cost. This trade-off is invisible to standard hyperparameter optimizers, which select lambda by minimizing one-step RMSE, the wrong metric for structural constraints in recursive forecasting. The framework adds three contributions: (i) resid
benchmark - arxiv:2611.18796 · cs.AIAn LLM-assisted theoretical model for aeroengine installation noise scatteringXun Huang
A theoretical model is developed for the canonical problem of aeroengine fan noise impinging on, and scattering from, a nearby semi-infinite rigid plate. The model chains two Wiener--Hopf problems in different coordinate systems. The first stage solves the duct-radiation problem analytically in axisymmetric cylindrical coordinates. The second stage expands the field incident on the plate into a double angular spectrum of plane waves: the axial spectrum is inherited from the first stage, while the spanwise spectrum is generated by a Weyl--Sommerfeld expansion of the cylindrical harmonics. Each spectral component is then scattered by the plate. The mathematical crux is the coordinate transformation between the axisymmetric duct frame and the Cartesian plate frame. The core physical insight--that the Weyl--Sommerfeld identity provides an exact spectral bridge between these systems--originates entirely with the authors. The heavy algebra of translating between the two spectral representati
manipulation - arxiv:2611.18792 · cs.LGHybrid Compression for Memory-Efficient Matrix-Vector MultiplicationFelipe A. Louza, Giovanni Manzini, Guilherme P. Telles
Iterative machine-learning algorithms repeatedly access large real-valued matrices through matrix-vector products, and their performance is limited by memory bandwidth rather than arithmetic throughput. Lossless compression schemes that support computation directly over the compressed data reduce both the data moved and the memory footprint of these workloads. The mm-RePair framework [11] follows this approach by applying the RePair grammar compressor to the Compressed Sparse Row/Value (CSRV) representation of a matrix and performing matrix-vector multiplications directly on the compressed data. Its effectiveness, however, degrades on matrices that are dense or contain many distinct values. In this paper we present a hybrid strategy for the mm-RePair framework that combines grammar and entropy compression. The CSRV sequence is partitioned into a grammar-compressible component, formed by the entries that mm-RePair can compress effectively, and an entropy-compressible component holding t
memory - arxiv:2611.18791 · cs.CVFrom Retouched Images to Paired Videos: A Synthetic Benchmark for Face Video RetouchingPeiyuan Zeng, Fengchuang Xing, Shiming Wu, Hongwei Yue +1
Paired unretouched/retouched face videos are costly to produce because professional editing must remain consistent over time, while releasing identifiable face footage raises privacy, consent, and governance concerns. We present DeepVFR, an initial controlled synthetic benchmark-not a new retouching architecture-for paired face video retouching. Starting from paired synthetic DeepFRR images, Wan2.2 generates mild-motion source clips, and backend-specific face transfer constructs synchronized proxy retouching targets, followed by automatic validity screening and manual visual inspection. DeepVFR contains 1,024 pairs of 150-frame clips at 30 fps, evaluated under an identity-disjoint protocol with explicit frame correspondence; release metadata will record the construction route. Within-benchmark baselines, external checkpoints applied to DeepVFR, qualitative analysis, and a 17-participant ranking study establish reference behavior in the controlled domain. The constructed targets exhibit
benchmark - arxiv:2611.18790 · cs.CVSame Evidence, Different Verdicts: Delivery of CT Images and DICOM Metadata Changes VLM Quality-Control DecisionsNicolas Avelar
Vision-language models (VLMs) are being proposed as automated checks on clinical imaging inputs, where a scan and its DICOM metadata can disagree; a truncated CT series wrongly admitted to volumetric follow-up, such as lung-nodule growth assessment in screening, biases the measurement. A natural safeguard is to let the model judge the image before it sees the metadata. We find this backfires. On 60 NLST chest CT pairs with coverage defects whose ground truth is computed from scan geometry, a false header that Claude Haiku 5.5 resisted when shown with the image (2% of its detections overruled) overturned 91% of them when it arrived after the model's answer (Nemotron-3.5-VL 43% to 100%). With the order reversed, the image overturned 98-100% of the header's errors (Claude models): whichever source came second won. This is not generic sycophancy: a neutral "please confirm" reversed 0 of 1,072 correct refusals, and the effect persisted when the header arrived as a tool result, as in agentic
agentic
02 US SEMI · SEC 8-K FILINGS
5 itemsscanned: NVDA / AVGO / MRVL / COHR / LITE / AMD / TSM / SMCI / ANET / CRDO / POWL / VECO
- $NVDA · 8-K · filed 2026-11-17NVIDIA CorpItems: 2.02,9.018-K
- $AMD · 8-K/A · filed 2026-11-16Advanced Micro Devices IncItems: 3.028-K/A
- $POWL · 8-K · filed 2026-11-16Powell Industries IncItems: 2.02,9.018-K
- $MRVL · 8-K · filed 2026-11-12Marvell Technology IncItems: 5.02,9.018-K
- $NVDA · 8-K · filed 2026-11-12NVIDIA CorpItems: 5.02,9.018-K
03 HUMANOID · COMPANY NEWS
54 itemsscanned: figure-ai / 1x / boston-dynamics / unitree / apptronik / sanctuary-ai / neura-robotics / agility-robotics / physical-intelligence / agibot
Figure AI (10)
- Figure AISeptember 30, 2026F.02 Decommission
- Figure AISeptember 17, 2026Helix 2.5: Zero-Shot 30-Home Generalization
- Figure AIOctober 09, 2025Introducing Figure 03
- Figure AISeptember 03, 2026Figure and Nscale Sign Strategic Partnership For Up to 100,000 GPUs on the NVIDIA Vera Rubin Platform
- Figure AIAugust 25, 2026Introducing Index: Building The World’s Largest and Most Diverse Physical Dataset
Boston Dynamics (10)
Unitree 宇树 (10)
- Unitree 宇树Components
- Unitree 宇树Kung Fu Meets Spring, Unitree SFG Robots Present "Cyber Real Kung Fu" in the Year of the Horse2026-05-31Media Coverage
- Unitree 宇树Welcoming Myanmar President Min Aung Hlaing to Unitree2026-08-05Media Coverage
- Unitree 宇树Unitree founder Wang Xingxing graces the cover of Time magazine2026-08-05Media Coverage
- Unitree 宇树Unitree Announces H2 Plus, an NVIDIA Isaac GR00T Reference Humanoid Robot for Academic Research2026-06-01Media Coverage
Apptronik (1)
Sanctuary AI (6)
- Sanctuary AIBlog PostTIGER: Task-Space Imitation Guidance for Efficient Reinforcement LearningDemonstrations as a compass, not a scriptRead More
- Sanctuary AIPress ReleaseProduct UpdatesSanctuary AI Expands Physical AI Strategy to Industrial Robotics, Demonstrating Production-Ready AI PerformanceRead More
- Sanctuary AICorporate NewsDaniel Friedmann Appointed CEO of Sanctuary AIRead More
- Sanctuary AIPress ReleaseZeon Invests in Sanctuary AI and Partners to Advance Specialized Materials for Dexterous RoboticsRead More
- Sanctuary AIThought LeadershipWeb Summit Reflections: Canada’s Physical AI Moment Can’t WaitRead More
Agility Robotics (10)
- Agility RoboticsAgility’s Humanoid Deployment ProcessPras VelagapudiAugust 04, 2026
- Agility RoboticsAgility Robotics Files Customary Registration Statements Following Completion of Business CombinationIn the NewsNovember 10, 2026
- Agility RoboticsJonathan Hurst answers your questions about building Digit 5InsightOctober 23, 2026
- Agility RoboticsThe Realistic Pathway to HomeInsightMay 26, 2026
- Agility RoboticsAgility and AIInsightMarch 16, 2026
智元 AgiBot (7)
- 智元 AgiBotSix AGIBOT G2 Robots Complete 1,730 Tasks in 24-Hour Shanghai Convenience Store Trial2026-11-06
- 智元 AgiBotAGIBOT Unveils GO-3 Preview, Takes Top Spot on Three Embodied AI Benchmarks2026-10-29
- 智元 AgiBotHow Much Work Do Robots Really Deliver? The Thinking Behind Worken2026-10-29
- 智元 AgiBotAGIBOT Introduces Worken to Measure Productive Robot Work at AGIBOT Innovation Day 20262026-10-28
- 智元 AgiBotAGIBOT Launches Genie Studio 2.0 to Bring Robots into Real Workplaces2026-10-26