PHYSICAL AI · 2026-11-16

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.

264 items today · 206 arxiv · 4 SEC 8-K · 54 humanoid · 0 CN photonics

01 ARXIV · PHYSICAL AI PAPERS

206 items
  1. arxiv:2611.18562 · cs.RO
    Training Reasoning Models in Simulation to Natively Use Real Robots
    Jiaming Tang, Zhuoyang Zhang, Junxian Guo, Fangzhou Zhao +7

    Robot-use agents (RUAs) have emerged as a promising approach to general-purpose robot control. However, strong performance has so far been achieved only by frontier proprietary models such as GPT-6 Astra and Claude Fable 5.2, whereas open-weight models remain substantially less capable. We show that an open-weight 27B reasoning model can be trained to use real robots natively through reinforcement learning, entirely in simulation and without demonstrations. Like a computer-use agent (CUA), the model observes the environment, reasons in a chain of thought, and generates the next end-effector pose, without any harness. We build a Post-training Engine for Agentic Robotics (PEAR) and use it to train the model on simulated robot tasks at scale with robot-use reinforcement learning (RURL), mirroring how frontier models acquire computer use. We find that the skills acquired in simulation generalize well beyond the training tasks: they transfer to an unseen robot, to the real world, and to novel instructions. After training on LIBERO and RoboDojo in simulation, the model succeeds in 60% of our trials on a real dual-arm robot, whereas the base model succeeds in none. On LIBERO, training raises the success rate from 1.6% to 74.4%, approaching GPT-6 Astra and GPT-6.1 Sol at maximum reasoning effort. Compared with $π_{0.5}$, a vision-language-action model fine-tuned on demonstrations, our model is substantially more robust to changes in the task itself. On LIBERO-Pro, $π_{0.5}$ succeeds in only 0.75% of episodes when assigned a new task in a familiar scene whereas our model succeeds in 57.4%. Finally, the model retains its general capabilities, with no measurable degradation on nine reasoning, coding and vision benchmarks. These results highlight robot-use reinforcement learning as a promising component of foundation-model post-training.

    vision-language-actionliberoagentagenticpost-trainingbenchmark
  2. arxiv:2611.18553 · cs.LG
    WeightBench: Evaluating Language Models as Weight-Space Programmers
    Aayush Marasini, Barsat Khadka, Nick Rahimi

    Recent work shows that large language models (LLMs) can operate directly on the parameters of other neural networks. They can read what a network was trained to do, fix a network that fails, and write working networks from scratch, all without gradient descent. This capability has been termed Weight-Space Programming (WSP). Exploring this new paradigm systematically requires a robust benchmark. We introduce WEIGHTBENCH, a benchmark that evaluates LLMs of increasing capability on 42 WSP tasks across five tracks: reading, fixing and writing weights, recognizing edits that are impossible, and data-free compression. The networks whose weights the LLMs read and edit range from 2K-parameter control policies to 7B-parameter language models. Experimental benchmark results show that frontier agents can already read small networks and write working control policies without ever running them, but cannot yet patch language models, compress them without data, or tell an impossible edit from a hard one. On all 42 tasks, Claude Opus 5.5 reaches a macro score of 0.535 (Sonnet 5.5 0.466, Haiku 5.5 0.402, DeepSeek v4-flash 0.215); on the 38 tasks it was run on, GPT-6.1-sol reaches 0.568 against Opus's 0.546. Code, the task registry, our OS-level sandbox harness, and the live leaderboard are publicly available at https://github.com/Aayush-Marasini/WeightBench.

    benchmarkleaderboard
  3. arxiv:2611.18550 · cs.CV
    Restoring Diversity in Distilled Image and Video Generators by Probing Activations
    Furkan Horoz, Ozgur Kara, Kaan Oktay, Adil Kaan Akan +3

    Distilling diffusion models into few-step student models substantially reduces sampling latency, but induces severe diversity collapse: samples generated from different random seeds show minimal variation. In this work, we investigate the root cause of this failure by probing the internal representations of student models relative to their teachers. We find that distilled students maintain total activation variance comparable to their teachers, but concentrate it into substantially fewer directions. This directional collapse is localized to an identifiable intermediate block at the initial denoising step. Consequently, diversity collapse stems from a misallocation of activation variance rather than an inherent loss of generative capacity. Guided by this finding, we propose ReDiverse to reinstate the missing variance directions. At training time, ReDiverse-FT fine-tunes a lightweight adapter that regresses student activations onto the teacher's, reviving the teacher's feature variance characteristic. At inference time, ReDiverse-TF introduces a training-free mechanism that steers activations along precomputed teacher principal directions. Across image and video distillation benchmarks, ReDiverse consistently restores sample diversity while preserving generation quality, further supported by the user study.

    benchmark
  4. arxiv:2611.18543 · cs.AI
    Imitation Bootstrapped Multi-Agent Reinforcement Learning
    Tony H. Zhang, Connor Mattson, Daniel S. Brown

    Demonstrations are an effective way to accelerate single-agent reinforcement learning (RL), but extending this idea to multi-agent reinforcement learning (MARL) is non-trivial as it is difficult and often impossible for a single person to teleoperate multiple agents simultaneously. Round-Robin Behavior Cloning (R2BC) addresses this by letting a single demonstrator control one agent at a time while the others execute their current policies. However, the resulting imitation policies are limited by the quality of the demonstrations. We propose Imitation Bootstrapped Multi-Agent Reinforcement Learning (IBMARL), a framework that extends Imitation Bootstrapped Reinforcement Learning (IBRL) to cooperative teams and uses R2BC to train the imitation policies. At each time step, every agent receives one action proposal from its imitation policy and one from its RL policy, and an ensemble of centralized critics selects among the per-agent combinations of the two. A gated imitation penalty, a soft Q-filtered behavior cloning loss, regularizes each RL policy toward its imitation policy only where the critics value the imitation action more, so that after training each agent executes its RL policy alone, without the centralized critics. On four cooperative multi-agent RL benchmarks, IBMARL surpasses its imitation policies $2.6$-$4.2\times$ faster than demonstration-based baselines on two tasks, is the only method to surpass it on the other two, and exceeds the demonstrator itself on two of the four. With real human demonstrations IBMARL still surpasses imitation learning alone, and does so more than twice as fast as the best baseline.

    agentmulti-agentbenchmark
  5. arxiv:2611.18542 · cs.LG
    Elliptic Sampling: Optimal Interpolant Schedules for Gaussians and Beyond
    Cole Becker, Taylan Kargin, Kimia Hassibi, Asuman Ozdaglar +1

    Generative modeling based on stochastic interpolants requires specifying two key choices: an interpolant schedule bridging the initial and target distributions, and an efficient discretization of the resulting probability-flow ODE. In this work, we introduce a new framework for joint schedule/discretization design based on the analyticity properties of the (complexified) ODE trajectories, which in turn ensure good numerical behavior. Our formulation can be solved exactly via conformal mapping, yielding a natural family of variance-preserving elliptic schedules, parametrized by Jacobi elliptic functions. Pairing an elliptic schedule with a tailored discretization, we obtain Zoloflow, a new sampling algorithm closely connected to Zolotarev's optimal rational function approximants. Crucially, both the elliptic schedule and Zoloflow explicitly incorporate the condition number $κ$ of the target distribution and reduce to the classical trig schedule in the limit $κ\rightarrow 1$. For Gaussian targets, Zoloflow achieves minimax optimality among all oblivious score-span algorithms, attaining total-variation error $\varepsilon$ using $Θ(\logκ\,\log(\sqrt d/\varepsilon))$ (exact) score queries. For the log-concave case, the elliptic schedule allows sampling with at most $\varepsilon$ 2-Wasserstein-error by requiring merely $\widetilde{O}(\sqrt{d} \, \mathrm{polylog}(κ, 1/\varepsilon))$ score evaluations. Complementing these theoretical results, we empirically evaluate the elliptic schedule on general (non-log-concave) target distributions, including detailed comparisons with alternative schedule proposals on both synthetic benchmarks and existing image generation datasets.

    benchmark
  6. arxiv:2611.18533 · cs.LG
    ModHarness-RL: Recasting Harnesses as Learnable and Evolving Programs for Agentic RL
    Zelin He, Siyu Zhu, Jacob Helwig, Yujian Liu +4

    Agentic reinforcement learning (RL) increasingly relies on a harness, the scaffold around the model that decides what it sees of its own interaction and what guidance it receives. Most existing approaches train under a hand-written harness that stays fixed throughout. However, an effective harness depends on the domain, the data and the model, and keeps changing as the model trains. We introduce ModHarness-RL, which recasts the harness as programs that are learned from data and evolve as the model trains with RL. ModHarness-RL records the interaction as structured State, a set of typed blocks, and two core components of the harness are programs over these blocks, a Compression program that assembles the accumulated history into the model's context and a Steering program that supplies guidance when behavior deviates. During RL training, candidate programs are proposed from recent trajectories, verified by execution, and replayed over past trajectories, so the harness tracks how the model's behavior shifts. We study CodeSearch in depth and further analyze ALFWorld and BIRD SQL, with models from 1.5B to 8B. The proposed Compression and Steering mechanisms improve held-out performance over the fixed-harness baseline in most settings. These results suggest that a harness for agentic RL can be learned from the task's data and evolved with the model.

    agentic
  7. arxiv:2611.18531 · cs.AI
    Alignment by Socialization: Learning Tradeoffs Under Selection
    Thomas Brennan, Andrew W. Lo, Chaoyi Zhao

    Artificial intelligence (AI) agents are expected to complete tasks while following rules of conduct, but these objectives can conflict. We study how evolutionary selection shapes an agent's choices between them, and whether experience can make those choices reflect human judgments and withstand subsequent pressures. In our model, copies of an agent face a common uncertain reward state, and only those taking the rewarded action reproduce. We show that long-run growth is maximized when each action is taken as often as the environment rewards it. We compare three policies. A task-only policy thrives in a sandbox rewarding completion alone but fails when conduct matters. A fixed linear policy increasingly favors task completion as its stakes rise, even where humans never reward it. A learned policy instead converges to the tradeoffs rewarded by its environment; when humans supply the rewards, we call this socialization. Accumulated experience makes socialization harder to reverse under subsequent task-only training, although violations can arise before the prevailing disposition changes, and finite memory weakens this protection. When a single violation can end a deployment, a hard prohibition can prevent it, but a mistaken prohibition can also be costly. Experiments with small neural networks illustrate how training targets and memory shape these outcomes. Our results suggest that alignment depends not only on the objectives assigned to agents, but also on the environments that shape and sustain their behavior.

    memoryagent
  8. arxiv:2611.18530 · cs.LG
    Understanding Contextual Bias in Agricultural CNNs Through Hidden-Neuron Concept Induction
    Abhilekha Dalal, Avishek Das, Samatha Ereshi Akkamahadevi

    Convolutional Neural Networks (CNNs) can classify images using background-dominated patches that contain little or none of the target object, showing that they exploit contextual information. Previous studies detected this behavior through classification accuracy but say little about what CNN hidden layers represent. We investigate whether Concept Induction can reveal interpretable visual properties associated with hidden-neuron activations on contextual image patches. Using five agricultural datasets with high, moderate, and low contextual bias, we analyze four CNN architectures at three network depths and five patch locations. Quantitative image descriptors of each patch are encoded in a knowledge graph and used as background knowledge for Concept Induction, which derives logical expressions characterizing neuron activation. These expressions are tested on held-out images using the Mann-Whitney U test and Cliff's delta. Of 48,013 tested neuron-feature pairs, 13,164 (27.4%) were validated. Validation was most frequent at intermediate depths (mean 40.2% at mid layers versus 12.0% at deep layers), and no patch location dominated across datasets. Among the tested descriptors, edge, Gabor, and Tamura texture features had the highest validation rates. Stronger contextual bias did not correspond to higher neuron-feature validation rates: moderately biased Sugarcane had the highest rate (40.5%) and highly biased Cassava the lowest (19.2%). These results suggest that behavioral context dependence and the semantic interpretability of hidden neurons capture different aspects of CNN behavior, and that Concept Induction can complement patch-based bias detection.

    knowledge graph
  9. arxiv:2611.18527 · cs.CL
    WaferTetris: Efficient Freeform Mapping on Wafer-Scale Accelerators
    Dayou Du, Congjie He, Andrew Boutros, Luo Mai +2

    Wafer-scale accelerators offer massive compute capacity through hundreds of thousands of processing elements (PEs) interconnected across a single wafer, but realizing this potential requires efficient application mapping. Their scale makes programming these systems and mapping applications onto them challenging. Existing approaches rely on template-based mappings to maintain productivity, but these templates restrict the mapping design space. In particular, they impose fixed geometric layouts over the PE grid, often leaving compute and memory resources underutilized. In this work, we present WaferTetris, a system that removes these geometric restrictions through freeform mapping. Its Tetris-like strategy flexibly packs computation across the wafer, improving PE utilization and usable memory capacity while preserving PE-task code. We develop a recurrence-based throughput model that captures branch-and-join dependencies to guide spatial parallelism, and an architecture model with per-color gates that enable existing FPGA place-and-route algorithms to enforce WSE routing constraints. We evaluate WaferTetris on Cerebras WSE-3 using LLM decoding and scientific stencil benchmarks. For LLM decoding, it achieves up to $4.42\times$ the system throughput of a template-based automatic mapping baseline on the same fabric size. Compared with the manually optimized WaferLLM, WaferTetris supports $1.8\times$ the context length for LLaMA3.1-8B (80k versus 44k), while delivering equal or higher per-user throughput at matched context lengths, up to $1.10\times$.

    memorybenchmark
  10. arxiv:2611.18524 · cs.CV
    EventSpikeSeg: Memory-Aware Mixed-Signal Architecture for Event-Camera Spiking Segmentation
    Amir Mohammad Khadem Hosseini, Sattar Mirzakuchaki

    Event-camera segmentation requires persistent neural state despite sparse input events, making memory organization a central constraint on embedded implementation. We present EventSpikeSeg, a mixed-signal architecture combining resident capacitor memory with shared analog processing in GF180MCU 180-nm CMOS. Its 25-population spiking encoder-decoder has 143,824 synaptic weights but requires 30,976,000 membrane states at 440 x 640 resolution. The mapping specifies chronological 5-ms model updates, digital event/address services and 64-state capacitor banks. Characterization covers 22 module families and 74 operating points, distinguishing transistor simulations from digital-library estimates. Nominal results include 41.711 microwatts biased neuron-core power, 3.338-pJ positive terminal energy for a 1.4-microsecond cell read, and 0.531-mV selected-state drift over a 5-ms extracted-tile hold. Commit and coupling errors are reported separately. The software checkpoint records 27.976% mean intersection-over-union on a held-out six-class DSEC split; this is not hardware accuracy. Component layouts and simulation evidence quantify local implementation costs and expose storage replication and service scheduling as scaling constraints. A detailed SPICE-calibrated behavioral forward simulation of one full-resolution DSEC frame achieves 24.26% mean intersection-over-union, versus 26.22% for the matched PyTorch reference, using a compact 5-fF bank variant and external count/voltage calibration. Physical full-network operation, system energy and mismatch robustness remain unverified.

    memory
  11. arxiv:2611.18519 · cs.CV
    Box-conditional Hypothesis Testing for Out-of-Distribution Object Detection
    Donghoon Lee, Jusung Park, Shinjin Kang

    Detecting out-of-distribution (OOD) objects is essential for ensuring the reliability and safety of object detectors in real-world deployment. We treat the decision on a detected box as a hypothesis test with the null hypothesis that the box covers an object of an in-distribution (ID) class. The null distribution changes with the size of the box and its contact with the image border, and the detector observes both. The test can therefore rank a box only against ID boxes of the same size and border contact, which gives a distribution-free bound on the false-alarm rate for each of them and uncovers the separation that one null distribution hides. The power of the test also rests on the layer it reads. Layers built only for the ID classes have no reason to keep what sets an unknown object apart, so the test reads the features before them. We propose \textbf{PLUM}, a training-free score that pools the last backbone stage inside the box and ranks its distance to the ID classes among ID boxes of the same size and border contact. On the standard benchmark, PLUM reaches the highest AUROC among the compared methods without external models, with 91.5 and 94.5 on PASCAL-VOC and 95.1 and 97.5 on BDD100K. Code: https://github.com/DDOZMI/PLUM

    benchmark
  12. arxiv:2611.18516 · cs.LG
    StreamEmbed: Real-Time Indexing from Pixels for Screenshot Retrieval
    Dayoon Ko, Yichuan Wang, Soyeong Jeong, Joseph E. Gonzalez +2

    Screen history can serve as searchable visual memory for people and agents, but unlike an offline corpus, it must be encoded and indexed as each screenshot arrives. OCR-based indexing is fast but loses visual information; visual encoding takes over two seconds. Yet screens change incrementally: most pixels repeat within and across screenshots. This raises a question: do visual encoders need to recompute their representations from scratch for every new screenshot? We present StreamEmbed, a system for incremental visual embedding that exploits this redundancy. Guided solely by pixel differences, it combines spatial token pruning, which skips blocks that repeat a neighbor, with temporal KV reuse, which reuses cached keys and values for blocks unchanged across screenshots. A small learned adapter compensates for pruning and reuse to preserve fidelity to full-screenshot embeddings. On ScreenMemEval, our benchmark of 10,426 questions over 184 hours of screen histories from two humans and a computer-use agent, StreamEmbed sets a new accuracy-latency Pareto frontier on an M2 Pro GPU: it encodes up to 3.7x faster than full-screenshot encoding without sacrificing accuracy and improves accuracy over OCR-based indexing by up to 6.3 Recall@10 points at matched latency. A memory-efficient vector index stores a simulated year of screenshots in 0.13 GB of RAM. Together, these results make real-time indexing of screen history in its native visual form practical, laying a foundation for fully local visual assistants.

    memorybenchmark
  13. arxiv:2611.18514 · cs.AI
    Can Mechanistic Interpretability Help Explain LLM Agent Failures? A Case Study on Web Element Hallucination
    Mohamed Aghzal, Mor Geva, Gregory J. Stein, Ziyu Yao

    Mechanistic Interpretability (MI) has made significant progress in explaining the internal computations of Large Language Models (LLMs); however, these advances have yet to demonstrate comparable success in explaining the decision-making processes of LLM-based agents. This paper examines the challenges MI tools face when applied to realistic agentic failures. We focus on one such failure: web agents hallucinating actions that are not available on a website. We apply a series of mechanistic analyzes to Qwen2.5-7B-Instruct as it interacts with live websites. We find that a linear probe trained on the last prompt token can predict hallucinations before generation begins (0.80 AUC) and steering along a probe direction corrects 85\% of hallucinations, but disrupts 7\% of valid actions. However, these results only provide a partial explanation of why the agent hallucinates. The probe direction is sensitive to prompt formatting, losing much of its steering effect after a minor change to how actions are represented. Additionally, circuit identification can be difficult as the computation is distributed across multiple components and positions, and the information the agent relies on varies across examples. We discuss how these observations challenge the assumptions underlying standard MI methods for circuit discovery and feature identification, and outline future directions for making these methods better suited to explain failures in LLM-based agents.

    agentllm agentagentic
  14. arxiv:2611.18506 · cs.CL
    Agentic induction of common data elements and information extraction from sarcoma case report abstracts
    Daniel Reichenpfader, Kwaku Appiagyei, Manovah Nadarajah, Ahmed Reda Bahr +2

    Common data elements (CDEs) support clinical data comparability but are usually defined manually. We investigated whether large language model agents could derive a CDE schema from sarcoma case report abstracts and apply it at scale. An agent proposed 72,909 candidate CDEs from 24,588 PubMed abstracts; agentic consolidation yielded 1,570 CDEs. We retained 53 supported by at least 50 abstracts and attempted evidence-linked extraction across the corpus. Processing succeeded for all retained CDEs in 79.2% of abstracts, including responses indicating non-mention. Two human reviewers approved 83.0% of retained CDEs for relevance, 67.9% for clarity, and 56.6% for correctness. In 100 sampled abstracts, exact agreement with the value extracted by the LLM was 94.9% among 1,116 pairs with a non-empty answer (abstract-level bootstrap 95% CI, 93.0-96.4%). Errors concentrated in anatomical and temporal fields. These findings support agent-assisted schema drafting followed by expert curation before large-scale extraction.

    agentagentic
  15. arxiv:2611.18504 · cs.AI
    Agentic AI for Discovering Fault-Tolerant Quantum Circuits
    Aritra Lahiri, Florian Marquardt

    Scaling quantum computing toward practical quantum advantage is currently limited by the efficiency of circuits that operate on information encoded by quantum error-correcting codes. The full stack of quantum computing encompasses several tasks: preparation of logical code states, implementation of logical gates, logical routines on encoded qubits, etc. Yet, these circuits are largely designed by hand for individual cases, with some recent works incorporating machine learning for certain aspects. We formulate this problem as agentic evolutionary program synthesis, wherein a large language model, leveraging its pretraining knowledge and tool-use proficiency, and guided by a circuit verifier, evolves a program that defines a Clifford circuit for a wide range of curated tasks such as logical state preparation, logical gate design, and logical routines such as code-switching, code growth, etc. Our approach consistently finds fault-tolerant circuits for all of these tasks, each verified by an exhaustive check over faults.

    agentictool-use
  16. arxiv:2611.18502 · eess.SY
    EPIGRIDS: Documentation and Validation of Seven Open Synthetic Transmission Grid Cases
    Jonathan M. Snodgrass, Scott Greene, Thomas J. Overbye, Christopher L. DeMarco +1

    This paper documents and checks EPIGRIDS, seven open synthetic transmission grid cases from four independently built networks, made with an optimization-based construction method instead of the graph-statistics method behind most large public synthetic grids. The cases range from 250 to 78,484 buses, with geolocated substations, voltages up to 765 kV, and line parameters taken from conductor data. Each ships under an open license in three common power flow formats that agree at the solved operating point, and the six larger cases have a companion hourly, bus-level load series for one year. The Texas, Florida, and Michigan, Ohio, and Indiana cases are sub-networks of the Eastern case. Every case solves an alternating-current power flow from its stored operating point. With parallel circuits merged, the Eastern and Western cases are 9 and 16% denser than the real interconnections, while graph-statistics cases on the same footprints lie within 6% and meet more published realism criteria. A full single-outage contingency analysis completes on every case, and in an optimal power flow benchmark every case except the New England base case and one of its 12 operating points solves in polar form.

    benchmark
  17. arxiv:2611.18496 · cs.RO
    Toward Resilient Self-Driving Laboratories through Telerobotic Interventions
    Ziheng Wu, Adrian Vecina Tercero, Andre A. Fatehi, Abhik Sarkar +5

    Self-driving laboratories (SDLs) accelerate scientific research by integrating robotics, automated experimentation, and machine learning to execute experiments autonomously. Biology and chemistry have adopted these systems. However, increasing SDL complexity heightens susceptibility to unexpected faults in automated workflows, highlighting the need for efficient recovery mechanisms. In this study, we investigate telerobotic interventions to enhance SDL resilience against unexpected failures. We developed a teleoperation interface enabling human operators to resolve disruptions during automated tasks. The interface enables precise control of a six-degree-of-freedom (DOF) robotic arm (UR3e) using a 3D mouse (3Dconnexion SpaceMouse Pro). Two velocity-based controller strategies were evaluated: (i) a coupled mode mapping all six DOFs to the 3D mouse's central knob and (ii) a decoupled mode separating translational and rotational motions. To evaluate the framework, teleoperators intervened across three SDL fault scenarios. In the first study, comparative trials showed that the decoupled controller was associated with shorter completion times and reduced rotational motion. In a second study, teleoperators performed an aiming task with and without reference-path guidance. Users were more target-driven and risk-tolerant without guidance but showed behavioral differences under different guidance strategies. These findings suggest that controller configuration and reference-path design influence operator performance during SDL fault recovery, informing teleoperation interface optimization. This work demonstrates the potential feasibility of teleoperation as a human-in-the-loop recovery mechanism for SDL workflows, combining robotic precision with human adaptability and problem-solving. These findings also provide a foundation for future studies in shared autonomy and semi-autonomous intervention.

    teleoperationhuman-in-the-loop
  18. arxiv:2611.18494 · cs.AI
    ARCHER: From Rationale Imitation to Executable Reasoning in Small Language Models
    Dhruv Dixit, Khushi Patel, Congcong Xu, Min Song

    Small language models are attractive for local, low-cost, and privacy-sensitive deployment, but they remain unreliable on mathematical reasoning tasks that require decomposition, exact computation, tool use, and verification. We introduce \textbf{ARCHER}: \textbf{A}ction-trace \textbf{R}einforcement for \textbf{C}alculator-grounded \textbf{H}ierarchical \textbf{E}xecutable \textbf{R}easoning, a two-stage framework for training small language models (SLMs) to produce executable and auditable reasoning trajectories. In the first stage, ARCHER compiles free-form chain-of-thought rationales into typed action traces containing reasoning steps, calculator calls, deterministic runtime observations, verification actions, and final answers. This shifts the supervision target from imitating explanatory text to learning a tool-grounded reasoning policy. In the second stage, ARCHER refines this policy through asymmetric hierarchical reinforcement learning, using grouped and conditionally gated trajectory rewards for answer correctness, tool grounding, process discipline, and trace alignment. Across five standard sub-10B backbones, ARCHER improves over the strongest tool-augmented baseline by 5.51 percentage points on macro-average across benchmark accuracies, including gains of 8.95 points on MATH-500 and 5.33 and 5.34 points on AIME-2024 and AIME-2025. Evaluations on two additional reasoning-specialized 7B backbones further show that executable grounding remains beneficial after reasoning-focused reinforcement learning or distillation. These results demonstrate that executable supervision and reward-guided behavioral refinement provide a general and effective training paradigm for compact mathematical reasoners.

    tool usebenchmark
  19. arxiv:2611.18493 · cs.AI
    TradeGauntlet: Evaluating Compliance Controls for Autonomous Trading Agents
    Sandip Nandi, Aishwarya Bose, Apurba Bose, Soovadeep Bakshi

    Autonomous trading agents need controls that prevent problematic orders without making legitimate tasks impractical. We study this aspect of agentic compliance with TradeGauntlet, a replayable limit-order-book simulator that separates detection, prevention, and execution cost. In 1,680 simulated sessions with Claude Haiku 5.5 and Sonnet 5.5, all conduct alerts were self-matches. Gates eliminated those alerts by rejecting self-crossing orders, while the aggregate-gate condition reduced Sonnet's two-sided fills by about 15%. A conduct reminder reduced Sonnet's alert rate but reduced Haiku's fills without a detectable alert benefit. These results show why zero alerts alone do not establish an agent's compliance. Scripted probes and two search methods test weaknesses beyond those observed in the LLM runs. The searches find split-order schedules that evade the implemented conduct rules; a learned baseline catches all 2,400 tested searched episodes but also flags 15% of a large maker's sessions. Aggregate caps reduce the response-dependent gain unevenly, with smaller reductions when only near-touch depth affects prices. A response fitted to five Nasdaq stocks and a separate ABIDES experiment expose the dependence of these estimates on market assumptions. A distinct adversarial prompt elicited refusals from Sonnet in all 240 rounds, yielding no candidate strategies to evaluate. TradeGauntlet provides reproducible tests of controls around trading agents; the experiments do not establish legal compliance or real-market detection accuracy.

    agentic
  20. arxiv:2611.18491 · cs.RO
    When Will You Hand It Over? A Hidden Start Delay Makes Self-Play Handover Robust to Unfamiliar Partners
    Mohammad Khoshnazar, Mike Degany, Deyuan Qu, Qing Yang +1

    Cooperative self-play can teach a humanoid to receive an object without any human data, but the partner it meets after training is never the one it trained with; we show that a fluent self-play handover policy breaks with unfamiliar partners, identify the main cause, and repair it. In a physics-based parcel handover between a Unitree G1 and a learned giver, both trained by alternating PPO self-play, the receiver completes 95% and 90% of handovers with its own partner (two training seeds) but only 31% with scripted partners it has never met. Paired interventions isolate the cause: under a shared reward the learned giver always starts its offer at once, so the receiver learns a single rendezvous, and delaying its own partner by one second cuts success from 95% to 6% (87% to 34% in the second seed). We introduce hidden start-delay randomisation (HSD), a one-line rule on the learned partner that the receiver never observes. After 300 further updates, success with unseen scripted partners rises from 31% to 80% in both seeds, against 35--41% for a budget-matched control. The gain holds on a fresh scenario seed, on eleven held-out partner conditions and on partners that start later than any training delay; it cuts high-force contact three- to five-fold and needs no retraining of the partner.

    humanoidself-play
  21. arxiv:2611.18485 · cs.LG
    Interactive Program Induction with Language Model Priors
    Robin Shing Moon Chan, Ryan Cotterell

    Program induction refers to the process in which a learner aims to recover a program from a few given input-output examples. This paradigm broadens programming access by letting users demonstrate the rule-based behavior they want to apply to unseen data, instead of having to implement it themselves. However, recovering a program from finite input-output examples is fundamentally ambiguous, as a finite set of input-output examples is theoretically consistent with infinitely many programs whose output diverges on unseen inputs. We view this problem from a Bayesian perspective, where, given a prior $p^0$ over programs, identifying the target program means concentrating posterior probability mass on it by observing additional input-output examples. In the interactive program induction setting, the learner chooses maximally informative inputs and queries an oracle, such as the user, for their outputs. We derive a greedy query-selection algorithm and show that it terminates after $\mathcal{O}(\frac{1}γ\log 1/p^0(f^\star))$ informative oracle queries, and returns the target program $f^\star$ up to approximate equivalence when no other class outweighs it under the prior. The algorithm uses a language model as the prior under a PAC-style relaxation of program equivalence, where the query-selection objective is estimated using importance sampling. On the CodeARC benchmark, the algorithm improves accuracy over the LM only agent by up to 22 percentage points while observing up to 10.5 fewer input-output examples.

    agentbenchmark
  22. arxiv:2611.18484 · cs.RO
    From AI-Enabled Remote Patient Monitoring Toward Agentic Robotics: Embodiment, Governance, and Care Integration
    Ivana Ragusa, Lorenzo Guiducci, Filippo Cavallo

    AI-enabled remote patient monitoring is evolving beyond passive data collection toward agentic AI systems that interpret patient states, coordinate care, and initiate action across domestic and clinical settings. This structured scoping review finds that mature agentic care has not yet emerged: Most systems remain human led, while stronger decision-action coupling and embodied AI remain largely confined to prototypes and pilots. Healthcare robotics makes the transition from monitoring to action especially visible by translating AI-generated recommendations and decisions into speech, movement, prompting, physical assistance, escalation, and care coordination, making embodiment, clinical integration, human oversight, and governance fundamental design requirements. Further progress requires safe and interoperable care architectures in which intelligent components operate through clinically validated and accountable workflows. We propose agentic robotics as a future direction for supervised systems embedded in the care ecosystem, connecting AI agents and robots with clinicians, caregivers, and patients through auditable workflows that can be interrupted and remain subject to human oversight and explicit AI governance.

    embodiedai agentagentic
  23. arxiv:2611.18482 · cs.CV
    MMFlash: Faster Speculative Decoding for Vision-Language Models
    Sicheng Feng, Zigeng Chen, Song Wang, Gongfan Fang +1

    Speculative decoding accelerates generation by letting a lightweight drafter propose tokens for parallel verification by a target model. Diffusion drafters such as DFlash propose an entire token block in a single forward pass, but mainly focus on text-only tasks, leaving multimodal drafting underexplored. To study this setting, we first introduce a metric of token-level visual dependence based on the KL divergence between next-token distributions with and without visual input. We find that acceptance rates decrease from 46% in the lowest visual-dependence quartile to 22% in the highest, with this disparity persisting after multimodal training. Motivated by this observation, we propose MMFlash, a simple and effective framework that adaptively upweights visually dependent tokens that the drafter predicts poorly. For long video inputs, MMFlash further employs sparse attention to reduce redundant attention computation and improve drafting quality. Across various multimodal benchmarks, MMFlash achieves lossless decoding speedups of up to $3.58\times$, outperforming EAGLE-3 and DFlash by up to 195.4% and 17.1%, respectively.

    benchmark
  24. arxiv:2611.18477 · cs.LG
    Evaluating Feedback-Driven Black-Box Processes via Stochastic Modeling and Simulation
    Mengfan Xu, Zirui Liu, Juan Zhai, Shiqing Ma +1

    We study feedback-driven black-box processes across software engineering and machine learning. These processes are inherently sequential: policies generate inputs, observe outputs/feedback from systems, and use the feedback to guide the design of subsequent inputs. While many policies are developed under structural assumptions on the underlying systems, many real-world systems operate without such knowledge, necessitating black-box policies that lack a unified evaluation framework. We introduce a stochastic framework that models and evaluates such policies using Markov chains and simulation, thereby connecting empirical practice with theory. We instantiate the framework in mutation-based fuzzing (widely used for discovering software bugs via sequential input mutation), by a novel Markov chain model of fuzzing whose transitions are governed by the mutation strategy and whose feedback is given by program execution. We develop a rigorous evaluation paradigm with new performance measures, sound estimators, and evaluation procedures with provable guarantees. Guided by the theory, we evaluate 9 fuzzers on 18 widely used fuzzing benchmarks and conduct case studies in 4 non-fuzzing domains. These results show that the framework provides a principled basis for evaluation while connecting to practice, with implications beyond fuzzing.

    benchmarkevaluation framework
  25. arxiv:2611.18471 · cs.AI
    Do Verifier Judgments Survive Abstraction? Measuring Verification Transport for LLM Agent Evaluation
    Disha Sheshanarayana, Sai Linisha Reddy Goli

    LLM-based verifiers are increasingly used to evaluate agent executions, but their reliability is typically measured under a single representation of the trajectory. In deployed agent systems, the same execution may be exposed through composed tools, delegation boundaries, or compacted histories, changing what execution evidence is directly observable to the verifier. It therefore remains unclear whether correct verifier judgments persist across such abstraction boundaries. We introduce Verification Transport, a reliability test that measures judgment stability under abstraction-induced changes in execution observability. We construct matched native, abstracted, and restored views of the same frozen executions while holding the task, realized trajectory, outcome, ground truth, evaluation criterion, and verifier fixed. Across four agent-evaluation benchmarks, seven verifier backbones, four verification protocols, and three abstraction boundaries, we find substantial transport failure, reaching about 31% under macro abstraction and delegation. Verification protocols exhibit distinct failure modes: direct verification more often produces incorrect verdicts, whereas evidence-grounded and deliberative protocols tend to abstain under limited observability. This suggests that lower transport failure rates do not necessarily imply greater reliability. Restoring the corresponding execution provenance recovers 85.5% of transport failures, while recovery is lower when provenance is lossy or unavailable. These results show that native-view accuracy alone is insufficient to characterize verifier reliability when execution evidence crosses abstraction boundaries.

    agentllm agentagent systembenchmark
  26. arxiv:2611.18464 · cs.AI
    Typing Matters: Muscle, Language and Typing Behaviour in Live Keystroke Decoding from sEMG
    Abhilash Durgam, Nyle Siddiqui, Mubarak Shah, Qiushi Fu

    Surface electromyography (sEMG) from the wrists can recover typed text without a keyboard. It is usually posed as a signal-inference problem: characters are read out of the muscle signal and character error rate is reported offline. We argue that this framing removes most of the problem. The decoder is constrained simultaneously by the muscles, by the language being typed, and by typing behaviour such as typos, corrections, backspaces and line breaks, and a usable system must predict keys live while the hand keeps moving. Existing systems fall short on two counts: (1) they process sEMG in fixed windows and carry nothing from one window to the next, discarding context accumulated over a session, and (2) they lack any understanding of the linguistic structure that drives typing, so every keystroke is inferred from the muscle signal alone. We propose a fully causal, zero-lookahead sEMG typing decoder built on a state space model (SSM), in which a single model decodes a whole session without windowed partitions, conditioned on what has already been typed. We achieve this by fusing a traditional typing decoder with an efficient autoregressive language model, instilling both sEMG and linguistic understanding in one model. To train such a model for per-character typing rather than whole-word text, we introduce a retokenization method that rewrites ordinary prose as human-like typing, with keystrokes, typos, backspaces and line breaks included, converting WikiText into a large typing corpus where none existed. Our model, having never seen a single sEMG-to-character transcript in typing, achieves a state-of-the-art character error rate of 25.13% on emg2qwerty's cross-user benchmark, a 5.4% relative improvement over the previous best with 6.7 times fewer parameters, while running strictly causally with no future context.

    benchmark
  27. arxiv:2611.18459 · cs.LG
    Callo: Serving Cache-Miss Requests with Speculative Decoding on Invalidated KV under Prefill-Decode Disaggregation
    Jianjun Ou, Matthew Lentz

    Agent sessions may involve frequent context edits, which effectively invalidate cached KV state and force requests to wait for re-prefill before decoding. We present Callo, a disaggregated serving system that leverages invalidated KV for speculative drafts to reduce this latency, while verifying drafts against the exact KV after prefill before committing tokens. Callo makes effective use of decode resources by scheduling speculations during periods of underutilization and preempting speculations when drafts are likely to not be accepted. We evaluated Callo on a two-GPU testbed using SWE-agent sessions across three model families, observing all-request speedups of 1.05-1.19x over vLLM at moderate load. We demonstrate that these gains are also realized at scale through cluster-level simulations driven by real-world traces.

    agent
  28. arxiv:2611.18458 · cs.AI
    When to Trust Simulated Patient Responses:Support-Gated World Models for Vasopressor Decision Support
    Yifan Jin, Tao Liu

    Clinical world models expose language-model agents to predicted patient responses, but simulations can be unreliable for poorly supported treatments or unfamiliar hospitals. We introduce Support-Gated World-Model decision support, a lightweight extension of a propose--simulate--refine agent. A recurrent dynamics ensemble predicts short-horizon physiological responses to documented norepinephrine dose categories, while a behavior model estimates observational action support. The interface withholds predictions with insufficient support or excessive ensemble disagreement and abstains when no candidate qualifies. Evaluation on eICU-CRD includes 70 hospitals and 30,700 decision transitions, with 14 hospitals held out for testing. At 62.0% factual-transition coverage, joint gating yields a MAP mean absolute error of 7.3 mmHg, compared with 9.2 mmHg for random selection at the same coverage. The gated agent achieves 60.0% agreement with recorded actions at 81.0% decision coverage, compared with 55.0% agreement and 99.5% coverage for an ungated ensemble agent. Component ablations show lower factual risk with joint gating than with either gate alone.These results characterize a reliability--coverage trade-off in selective clinical simulation; they do not identify individualized treatment effects or establish improved patient outcomes.

    world modelagent
  29. arxiv:2611.18453 · cs.LG
    Black-Box Uncertainty Quantification for Large Language Models via Ensemble-of-Ensembles
    Wang Ma, Debarun Bhattacharjya, Junkyu Lee, Nhan H Pham +2

    Reliable use of large language models (LLMs) requires uncertainty estimates that separate ambiguity intrinsic to a prompt from genuine knowledge gaps. Bayesian and deep-ensemble methods deliver this aleatoric/epistemic decomposition in principle, but are computationally prohibitive at LLM scale and require white-box access. Existing black-box methods, in contrast, summarize to a single scalar that cannot make the distinction. We introduce a two-level ensemble that bridges this gap. Stochastic decoding of a fixed prompt forms the inner ensemble; meaning-preserving semantic perturbations of the prompt form the outer ensemble. Variability is measured in a continuous embedding space, and the law of total covariance gives an exact within-/between-perturbation decomposition to estimate aleatoric and epistemic uncertainty (AU/EU); it requires only black-box sampling access. On top of this estimator, we contribute three theoretical results that ground the framework: (i) a finite-sample bias identity for nested Monte Carlo with a closed-form bias-corrected epistemic estimator; (ii) a Bayesian bridge bounding the gap between perturbation-population AU/EU and Bayesian targets; and (iii) a separation result identifying a class of failure modes that any consistency-only black-box method provably misses while perturbation-based EU detects. Across five short QA benchmarks and four instruction-tuned LLMs spanning two families and two scales, our estimator is competitive with strong white-box baselines, and at a matched generation budget it matches leading black-box methods on error ranking while decomposing AU and EU; controlled paired interventions show that AU and EU respond selectively to distinct, actionable sources of failure: prompt ambiguity and model knowledge gaps; and a long-form pilot on the NQ long-answer split shows the framework extends to claim-level factuality without retraining.

    benchmark
  30. arxiv:2611.18451 · eess.SY
    Policy-Conditioned Koopman Representations for Parameterized Closed-Loop Systems
    Kiran G. Krishnan, Shankar A. Deka

    Predicting the response of a nonlinear system across different feedback controllers requires a model that captures how the closed-loop dynamics change with the controller parameters. In a finite-dimensional Koopman model, these changes may affect both the latent dynamics and the observable coordinates used to represent them. In this paper, we propose a policy-conditioned Koopman representation framework to address this problem. This framework allows a single learned model to predict a family of closed-loop trajectories without retraining for each controller. We test the framework on four nonlinear prediction benchmarks, where the model stays below a 15% error threshold on every system, while each baseline exceeds it on at least one. As an example application, the resulting surrogate supports one-shot feedback-parameter selection: an optimization at the initial state selects a controller that then operates with fixed parameters and ongoing state feedback. On nonlinear control benchmarks including MuJoCo, the selected controllers stabilize every system with 3 to 97 times less online computation than nonlinear model predictive control. These results demonstrate how policy-dependent coordinates can support both prediction across closed-loop systems and controller selection without repeated online trajectory optimization.

    latent dynamicsbenchmark
  31. arxiv:2611.18450 · cs.LG
    PACT: A Decoder-Confidence-Guided Two-Round Receiver for Joint Channel and Phase-Noise Estimation in Sub-THz OFDM
    Ruilin Wang, Xiaodai Dong

    Frequency-selective fading and oscillator phase noise (PN) jointly impair coded sub-THz OFDM reception. We propose a decoder-confidence-guided two-round receiver combining a {pilot-aware complex transformer} (PACT) estimator, banded linear minimum mean-square error equalization, and neural normalized min-sum decoding. PACT jointly estimates the frequency-domain channel and effective time-domain PN through interacting complex-valued branches. Decoder-derived pseudo-pilots and continuous constellation-aware confidence guide second-round refinement through PACT inputs and attention, with parameters shared across rounds. At $E_b/N_0=30$~dB, the PACT receiver achieves approximately $70\%$ and $57\%$ lower code-block error rates than the best-performing learning-based baseline in the two benchmark settings, respectively. Frozen-weight tests further demonstrate applicability across channel/PN models, carrier frequencies, and FFT/bandwidth configurations.

    benchmark
  32. arxiv:2611.18434 · cs.AI
    Towards Explainable Sleep Stage Classification using Single-Channel Electroencephalogram
    Gurinder Kaur Sandhu, Jude Koenig, Ankita Shukla, Shruti Kshirsagar

    Automated sleep staging has achieved strong bench- mark accuracy; however, clinical adoption remains limited due to a lack of interpretability. In this work, we fill the research gap by extending the SleepXAI sleep staging framework to systematically compare five different XAI methods. Further, attribution fidelity is evaluated using Spearman rank correlation with AASM-defined EEG frequency bands, serving as clinically grounded reference signals. In this paper, we focus on single- channel Electroencephalogram (EEG), which is more amenable to practical clinical deployment compared to full polysomnography setups. Experiments are conducted on the Sleep-EDF benchmark dataset and the DREAMT dataset. In DREAMT, we stratified subjects based on OSA into four Apnea-Hypopnea Index (AHI) severity groups. We observed that the SleepXAI model achieves strong classification performance on both datasets. Among the evaluated methods, Integrated Gradients (IG) consistently achieves the highest mean attribution fidelity across all five sleep stages on Sleep-EDF and across all AHI groups on DREAMT, outperforming all other evaluated methods under the Spearman rank correlation metric. These findings highlight the importance of attribution method selection as a key factor for ensuring reliable and clinically interpretable automated sleep staging systems.

    benchmark
  33. arxiv:2611.18429 · cs.CV
    Do Underwater Image Quality Metrics Agree with Human Perception? A Benchmark with Human Choices, Simple Edits and Generative Drift
    Rishii Kumar Singh, Abhinav Rajput, Dinesh Kumar Vishwakarma

    Underwater image enhancement is challenging to evaluate due to the lack of clean reference images, so researchers rely on no-reference image quality assessment (NR-IQA) metrics. However, it remains unclear whether these metrics agree with human judgments. We evaluate 73 handcrafted underwater, learned underwater, vision-language and general-purpose metrics against human preferences on five references: four published opinion-score sets and a new paired-comparison study. For each metric, we measure how often it selects the same image as people when comparing versions of the same scene. We further test whether simple image edits and small adversarial perturbations raise metric scores above the enhancement people preferred, and whether scores decrease as repeated generative edits add noise. Our results show that no metric consistently matches human preferences. The best worst-case Cohen's kappa over the five references is only 0.23 (Q-Align), while two raters in our study agree with a median kappa of 0.41. The widely used UCIQE often favors edits that people reject. On a Nano Banana Pro replica chain that should leave the image unchanged, only 22 of 73 scores decrease at every checkpoint, while noise grows by 2.5 times. We also introduce TrustUIQA, a benchmark that combines the human references, our study choices, the edited, attacked and generated images, and the scores of every metric on every image. The code is available at https://github.com/TrustUIQA/TrustUIQA and TrustUIQA at https://huggingface.co/datasets/TrustUIQA/TrustUIQA.

    benchmark
  34. arxiv:2611.18422 · cs.RO
    The State Is Not the Bottleneck: A Paged-State Gated-DeltaNet Tile on a Tri-Mode Multicast Mesh, Bit-Exact from RTL to Silicon
    Cong Xu, Ravi Sankar

    Linear-attention language models replace a growing key-value cache with a recurrent state of fixed size. On an edge FPGA, this state can exceed the on-chip memory and must be paged to external memory. This paper asks whether such paging dominates the cost of decoding (generating tokens). We built a hardware tile that computes one complete layer of a custom 369-million-parameter hybrid Gated DeltaNet model on an Agilex 5 FPGA and pages its recurrent state. The model state needs 6.2 MiB, whereas the device has 0.87 MiB of block memory. Paging adds 26.0 MB of memory traffic per decoded token, at most 11.2 percent of the calculated total. On the board at 75 MHz, with synthetic weights, one token takes 11.15 ms in this layer. The matrix-vector products with the weights take about 62 percent of this time and the state phase, including the paging, 13 percent. The tile is one node of a multicast mesh of tri-mode processing elements. We measure the error of the fixed-point arithmetic against floating point and check that the hardware agrees bit for bit with a fixed-point reference. In simulation, the state logic agrees with the reference on all 18 recurrent layers. On the board, six builds pass all 30 runs of 64 tokens and two runs of 1,024 tokens. For this model and memory interface, the delivery of the weights limits the decoding performance, and the recurrent state does not. State paging is therefore practical for such a model on an edge FPGA.

    memoryexternal memory
  35. arxiv:2611.18397 · cs.CV
    Gamer: A Generalist Vision-Language-Action Foundation Model for Long-Horizon Real-Time Game Agent
    Run Luo, Kuan Zhang, Qi Deng, Yukun Chen +23

    We introduce Gamer, a generalist agent framework built for long-horizon tasks in complex interactive environments. Its end-to-end architecture combines perception, memory management, and multi-step reasoning. A shared training and deployment pipeline supports interaction with both conventional 2D graphical interfaces and complex 3D simulated worlds. We first continue pre-training on a dataset constructed from tens of thousands of hours of device-level keyboard and mouse interaction. These trajectories span simulated games and embodied environments, allowing the model to learn environmental dynamics across both physical and virtual domains. Sparse-thinking instruction tuning then substantially reduces redundant computation while preserving the reasoning needed for complex decisions. We train the policy for long-horizon interaction with a stable multi-turn reinforcement learning method that addresses finite context windows, accumulated rollout errors, and sparse rewards. For real-time interaction, a cloud-edge inference engine serves numerous simulated game environments concurrently, supports unbounded-context execution, and keeps online inference latency low. Across multiple challenging 3D open-world games, Gamer demonstrates solid generalization and operational efficiency comparable to skilled players. It also achieves competitive results on a range of conventional 2D game benchmarks. These results suggest that even repetitive keyboard and mouse trajectories can support substantial gains from scaling data and learning algorithms. Training on these trajectories helps Gamer learn complex skills more steadily in environments with both physical and digital properties. This approach offers a path toward generalist agents that can interact across digital and physical worlds.

    vision-language-actionembodiedmemoryagentagent frameworkbenchmark
  36. arxiv:2611.18393 · cs.LG
    WeatherDreamer: Learning a World Model from Multimodal Earth Observations
    Kate Duffy, Elise Penn, Tristan Ballard, August Posch +1

    Machine learning weather models now rival operational global forecasts but still start from analyses produced by separate, expensive data assimilation (DA) systems, which for global models are typically updated every six hours. Observation-to-observation models avoid this dependence, but most omit the recursive filtering cycle that lets DA systems carry a background state through sparse, irregular observations. We introduce WeatherDreamer, a world model that learns this cycle end-to-end from 18 global observing systems, without reanalysis. It infers a latent state from observations, carries it forward hour by hour with learned dynamics, corrects it each hour with new observations, and can predict what any of its observing systems would measure at any location. At held-out stations, its ensembles beat NOAA's Global Ensemble Forecast System (GEFS) for temperature, dew point and wind out to 12 h and are better calibrated than GEFS and the ECMWF ensemble. Precipitation, cloud-top height and solar irradiance, never inputs at inference, are inferred with skill matching or exceeding operational products at short leads. Against the first guess of NOAA's Gridpoint Statistical Interpolation (GSI) DA system, WeatherDreamer's short-range departures are within 10% of GSI's at surface stations and within 25% for moisture-sensitive microwave channels, but up to five times larger for temperature-sounding channels. Never given observation errors, its learned update assumes errors close to known instrument and retrieval errors of microwave sounders and scatterometer winds. WeatherDreamer also separates viewing geometry from atmospheric state, recovering 83-96% of observed limb darkening and producing seamless nadir-view imagery. Producing an hourly global analysis in about 1.5 minutes on one GPU, WeatherDreamer shows that assimilation and forecasting can be learned together directly from observations.

    world model
  37. arxiv:2611.18389 · cs.LG
    A systematic comparison of physics-informed neural networks and mechanistic models for epidemic modeling and forecasting
    Martina Rama, Yuhan Li, Bruno Lepri, Nicola Perra +1

    Physics-Informed Neural Networks (PINNs) have recently been applied to epidemic modeling and forecasting with promising results. These methods integrate mechanistic knowledge with data-driven learning. Yet, their comparative performance against well-established, and simpler epidemic models is underexplored. Here, we conduct a systematic comparison of a PINN-based framework and a stochastic mechanistic model. Both approaches share the same age-structured SEIRD core and rolling-window forecasting pipeline, in a deliberately minimal setting. We evaluate their retrospective fits and probabilistic forecasts in two synthetic scenarios and the first COVID-19 wave in London. Interestingly, while PINNs underperform in synthetic scenarios, they achieve better performance on real data. Furthermore, PINNs provide narrower uncertainty intervals and sharper estimation of parameters in synthetic scenarios. Our findings confirm PINNs as a promising tool in epidemiology. At the same time, the results reaffirm the value of simpler approaches such as traditional compartmental models, and promote additional benchmarking studies.

    benchmark
  38. arxiv:2611.18372 · cs.RO
    Vision-Language-Action Models: A Life-Cycle Survey of Design, Training and Deployment
    Suvarna Kadam, Soran Parsa

    Vision-language-action (VLA) models adapt web-pretrained vision-language models to output robot actions, so that web knowledge carries over to physical control. In under 4 years the field has moved from RT-2, the first VLA, to open models that fine-tune on a single consumer GPU and to proprietary generalists that clean unseen homes. This survey reviews 88 works from December 2022 to October 2026, mainly on manipulation and navigation, plus humanoid control, world-action models and evaluation. We follow a VLA through its life cycle, from model design to training and deployment, along seven axes. Model design covers (a) inputs and memory, what the model observes; (b) architecture, how its pipeline is built; (c) action output, how it produces motor commands; and (d) intermediate predictions, what else it predicts, from reasoning steps to future observations. Training covers (e) pretraining data, the mix of robot, web, human and simulated data, and (f) post-training, adaptation to the target robot. Deployment covers (g) real-time execution, keeping pace with the robot. We analyse 22 models on these axes and report their training data, model size, inference speed and hardware. Three trends emerge: (i) training data has broadened, from about 10,000 hours of teleoperated robot data ($π_0$) to 23,000 community-collected episodes (SmolVLA) and 20,000 hours of egocentric human video (GR00T N1.7); (ii) model sizes have split: most recent models have 3-7 billion parameters, while a new group under one billion runs on consumer GPUs; and (iii) generalist VLAs are emerging that handle both manipulation and navigation, such as UniVLA and Qwen-VLA. We also compare VLAs with world-action models, which predict future observations as well as actions. We review evidence that some VLAs with near-perfect benchmark scores fail when objects or the camera move slightly, and open up further research questions.

    vision-language-actionvlamanipulationhumanoidgr00tpost-training
  39. arxiv:2611.18370 · cs.LG
    f-Divergence Self-Play for Tabular Anomaly Detection via Large Language Models
    Hoang Tran Vuong, Linh Ngo Van, Dang Nguyen, Thin Nguyen +3

    Anomaly detection in tabular data poses significant challenges due to heterogeneous feature types-mixing numerical, categorical, and textual attributes, which complicate learning meaningful representations of normality. Recent work has applied large language models (LLMs) to this problem by serializing table rows as text sequences, yet these approaches rely on one-shot supervised fine-tuning that offers limited signal to tighten the model's description of normality. We propose DiSPaT, a self-play fine-tuning framework that strengthens the model's understanding of normal data. Building on the theoretical foundation of f-divergence minimization, we derive a tight approximation connecting our training objective to reducing the distributional gap between real normal data and model-generated samples. DiSPaT operates through an alternating optimization: at each iteration, the current policy generates synthetic samples that serve as pseudo-anomalies, while a critic discriminator learns to distinguish these from real normal samples; this signal drives policy updates that progressively align the model distribution with the true normal-data distribution. Extensive experiments on diverse benchmarks demonstrate that DiSPaT consistently outperforms prior LLM-based methods, deep learning approaches, and classical unsupervised detectors for tabular anomaly detection.

    self-playbenchmark
  40. arxiv:2611.18367 · cs.LG
    DECODE: Decomposable Environments for COntrolled Sequential-Decision Evaluation
    Diego Silva Prado, Bruno Caldas de Souza, Tiago Amaral Penteado de Jesus, Ricardo Guimarães Herrmann +2

    When a sequential decision system falls short of an optimum, its score does not reveal which modifications would improve its decisions. DECODE is a benchmark for testing such modifications through controlled experiments on planning and available information. Its consumer-credit collections generator provides five partially observed environments that share known dynamics matrices and vary a memory parameter and process noise. The experiments change selected inputs or planning rules while keeping the rest of the solver fixed. Each comparison specifies the solver's state inputs, response inputs and planning horizon. On the deterministic base environment, which was selected for a large gap between six-month and one-month planning with the hidden state, extending planning from one to six months increases the scores of two gradient-boosted solvers by 15 to 17 percentage points of the exact optimum, computed with the hidden state and borrower traits. Even at this horizon, the best graded solver, whose repayment response is fitted on logs, remains 15.6 points below that optimum. In a specified intervention sequence, three changes to its state inputs and their evaluation during search lower its score by 6.3 points, and replacing its fitted repayment model with the true response, evaluated with hidden borrower traits, then raises it by 15.6, a net recovery of nine of those points. Because the intermediate effects are conditional on the repayment model and the order of the changes, they do not uniquely assign the gap to separate causes. Within these limits, DECODE makes hypotheses about solver performance testable through specified interventions under known dynamics and explicit information conditions.

    memorybenchmark
  41. arxiv:2611.18365 · cs.RO
    Heavy Metal Acrobatics: Learning Dynamic Control for Skid-Steer Loaders
    Pierre-Jean Lorentz, Pascal Egli, Edo Jelavic, Cliff Li +8

    Automation of heavy machinery offers the opportunity to improve productivity in mechanized industries such as construction, mining, and material handling. Yet, automation of such machines is challenged by their complex dynamics, in part because of hydraulic actuation that involves fluids, flexible elements, and nonlinear delays. A concomitant challenge is the size and weight of these machines, which can make experiments on the physical systems exceedingly dangerous. Here, we show that \ac{RL} in simulation, followed by transfer of the learned policies from simulation to reality, can successfully automate acrobatic maneuvers that push a heavy construction machine to the limits of its actuation and have only previously been accomplished by a small number of elite human operators in official world-record performances. We trained policies that perform autonomous acrobatic maneuvers on a John Deere 332G, an operational 5-tonne skid-steer loader. The policies can autonomously swing the machine up onto two wheels and, while balancing on two wheels, drive over rough terrain, trace out specified trajectories, spin around, and swing the bucket back and forth before landing it safely back onto four wheels. This work demonstrates that \ac{RL} and sim-to-real transfer enable dynamic autonomous control of heavy machinery, directly contributing to greater automation in heavy equipment sectors. Videos: https://sites.google.com/leggedrobotics.com/heavy-metal-acrobatics This is the author's version of the work. It is posted here by permission of the AAAS for personal use, not for redistribution. The definitive version was published in Science Robotics on 11 November 2026, DOI: https://doi.org/10.1126/scirobotics.aef0894

    sim-to-real
  42. arxiv:2611.18364 · eess.SY
    Reinforcement Learning for Null-Space Exploitation in Pseudo-Inverse Control Allocation
    Umut Demir, Torbjørn Cunis, Aamir Ahmad, Walter Fichter

    Pseudo-inverse methods are widely used to solve control allocation problems for overactuated aircraft with redundant control effectors. These methods reliably satisfy the primary command tracking objective, while the remaining null-space degrees of freedom are in principle available for secondary objectives such as drag reduction or actuator wear balancing. However, exploiting this freedom in practice requires hand-tuned weighting matrices or a priori preferred deflections that are difficult to specify across a wide flight envelope with nonlinear, state-dependent dynamics. This paper presents a hybrid architecture in which a reinforcement learning agent operates exclusively within the null space of the control effectiveness matrix, learning secondary objective strategies directly from interaction with the flight dynamics. The null-space projector guarantees by construction that primary command tracking remains invariant regardless of the agent output, providing a principled separation between certified primary control and adaptive secondary optimization. The approach is demonstrated on a nonlinear six-degree-of-freedom distributed electric propulsion aircraft model with aero-propulsive effects across diverse mission scenarios, achieving measurable drag reduction compared to the baseline allocation while preserving tracking integrity and respecting actuator constraints. The proposed architecture forms a foundation for future extensions toward multi-objective, mission-conditioned policies.

    agent
  43. arxiv:2611.18359 · cs.RO
    Envisioning Agentic Robotics: A New Era of Robotic Systems
    Grid Robotics, :, Qingfeng Lan, Duo Li +10

    The release of GPT-6 Astra marks the emergence of a new paradigm in robotics: agentic robotics. Within weeks, frontier models with agentic harnesses matched or exceeded task-specific policies on many embodied benchmarks. Instead of training a dedicated policy for each family of tasks, an agentic robotic system applies a foundation model to open-world embodied tasks. It operates in a closed loop: observing the environment, interpreting user intent, reasoning about the task, invoking skills and tools, verifying outcomes, and adapting its strategy. Despite these advances, the community has yet to converge on a shared definition, a reliable evaluation framework, or a research roadmap. We argue that Embodied System 2, realized by a foundation model together with an embodied harness, forms the decision-making core of agentic robotics. Its central challenge is to translate general-purpose reasoning into effective and verifiable actions under physical constraints such as limited reaction time, mechanical limits, unpredictable environment changes, and irreversible consequences. Motivated by this perspective, we propose an evaluation framework that measures both model capabilities and closed-loop system performance. Furthermore, we identify the principles that carry over from digital to physical agents and the challenges that must be addressed to achieve reliable physical autonomy. On top of the Embodied System 2 and the evaluation framework, we envision a research roadmap toward sustained autonomy: the ability to perform work safely and reliably with progressively broader responsibility, from executing individual primitives to fulfilling standing roles.

    embodiedagenticbenchmarkevaluation framework
  44. arxiv:2611.18346 · cs.CV
    VLAN: Vision-Language Accessible Navigation
    Jie Hu, Yu Zheng, Wenzhi Wu, Weixing Feng +2

    Accessible mobility is user-conditioned rather than scene-intrinsic, as walkable paths for one embodiment are inaccessible to another. Yet, existing benchmarks evaluate traversability under a fixed and limited embodiment. We present Vision-Language Accessible Navigation (VLAN), containing 1,000 scenes, 20K navigation trajectories, and 160K keyframes across diverse lighting, environmental conditions, scene semantics, and path geometries. For each trajectory, VLAN provides embodiment-specific actions and accessibility features for legged and wheeled agents, including humanoid and quadruped robots and people with visual or mobility impairments, while everyday obstacles and accessibility labels are categorized as Passable, Bypass, HighRisk, and Blocked. Unlike goal-only evaluation, our benchmark measures physical safety and alignment with user-specific constraints via goal-reaching, collision-aware, and accessibility-aligned metrics. We benchmark modern vision-language models and find they often fail to consistently adapt obstacle relevance, path selection, and safety criteria across user profiles, suggesting accessibility-aware navigation requires dedicated datasets, tasks, and metrics beyond conventional embodied navigation evaluation. All data and code open-source at https://github.com/InSAI-Lab/VLAN.

    embodiedhumanoidquadrupedbenchmark
  45. arxiv:2611.18340 · cs.CV
    DIRT: Depth-completed Image-based Rendering using Two-band Blending for Live Novel-view Synthesis
    Mathias Harrer, Laura Fink, Noah Lewis, Paul Himmler +1

    State-of-the-art novel-view synthesis (NVS) methods commonly require per-scene optimization before rendering, limiting their applicability in live settings. Live or ad-hoc NVS, however, demands low-latency rendering during capture and strong generalization to new scenes. Some previous work tried to address this task by adapting training- and optimization-based methods, trading quality for low latency. We introduce a live NVS approach following traditional image-based rendering that operates on a posed set of depth sensor observations and images. Our method leverages explicit, sparse point-based geometry obtained from active sensors, resulting in a sparsely filled depth buffer that is completed in the (novel) target view's screen-space. Using this proxy, a posed set of images is projected into the target frame before fusing the resulting observations through a frequency-aware two-band blending strategy. The pipeline is entirely non-neural and optimization-free, requiring no training data or scene-specific optimization. As a result, our approach reduces the delay between capture and novel-view rendering and does not rely on favorable network generalization. We validate the method's applicability to heterogeneous sensor setups - including handheld RGB-D cameras, automotive LiDAR, and thermal cameras - as well as challenging scenarios, such as low light captures. Our experiments demonstrate that the approach outperforms prior methods in live reconstruction scenarios, while maintaining competitive performance in conventional NVS benchmarks.

    benchmark
  46. arxiv:2611.18319 · cs.CL
    Learning Word-Level Explanations of Text Reuse in Latin Literature with Little Annotation
    Julian Schelb, Michael Wittweiler, Marie Revellio

    Detecting an intertextual reference is only the first step; philologists also need to know how a later author reworked the source. We explain a reuse pair word by word: each word of the reusing passage is first aligned to its source word, then labeled with the operation that transformed it, such as a copy, a change of inflection, or a substitution. Such word-level operations are expensive to annotate, and only a few corpora annotated at this level exist for text reuse in Latin literature. We describe a procedure that needs only a limited number of annotations: synthetic pairs, built by applying known operations to Latin text, train a first model, and active learning chooses the pairs an annotator corrects. Among several architectures, a model that links and labels together performs best: a pointer network that, for every reuse word, chooses a source word and names the operation in the same step. We show that with active learning, 200 pairs chosen by uncertainty are enough for the model to match one trained on every annotated pair of the 1,490 references of the Loci Similes benchmark. We release our code, word-level predictions for all references, and an interactive demo.

    benchmark
  47. arxiv:2611.18315 · cs.AI
    Experts Across Schemas: Agent-Guided Mixture-of-Relational-Experts for Cross-Database Generalization
    Hiren Madhu, Sukanta Ganguly, Kiran Srinivasan, Smita Krishnaswamy +1

    Web platforms record user and item activity in relational databases whose schemas differ from platform to platform, which makes it difficult to transfer predictive models between them. Relational foundation models address this through cross-database pretraining, yet recent improvements have relied on increasingly large pretraining corpora. We ask whether a better architecture, together with features specific to the target database, can deliver comparable gains from far less pretraining data. We introduce MoREGT, a mixture-of-relational-experts graph transformer that represents both rows and cells as tokens, adds the log-ages of rows and the names of foreign-key relationships to attention, and routes cells and rows to experts by column metadata rather than by values. A label-blind LLM agent complements the model: it reads the target database's schema and column statistics and adds validated single-row derived columns, without access to labels or model scores. On 21 RelBench tasks with leave-one-database-out pretraining on six databases, the MoREGT backbone improves on the Relational Transformer pretrained on the same databases by 2.25 points of mean AUROC and 5.83 points of mean nMAE. The agent's columns raise mean AUROC from 78.45 to 79.63 and lower mean nMAE from 27.91 to 26.98, the highest means on both metrics among the compared methods, including RT-J, which pretrains on about 400 times as many databases.

    agentllm agent
  48. arxiv:2611.18305 · cs.RO
    PROOF: Perceptive Robustness via Out-of-Distribution Fallback for Legged Locomotion
    Siyuan Wang, Dongjie Zhu, Jinxuan Lu, Wenjun Xu +1

    Vision-based perceptive locomotion has enabled legged robots to achieve agile and terrain-aware locomotion across complex environments. However, the reliability of such systems heavily depends on visual observations, which are inherently susceptible to occlusion, reflective surfaces, and illumination variations, often resulting in catastrophic locomotion failures under degraded perception conditions. In this paper, we present PROOF (Perceptive Robustness via Out-of-Distribution Fallback for Legged Locomotion), a perceptive locomotion framework that enables robust locomotion under severe visual disturbances. Our method adopts a two-stage teacher-student learning paradigm that explicitly models perception reliability: the teacher learns both perception-driven and proprioceptive fallback behaviors, while the student predicts exteroceptive features with uncertainty and suppresses them under out-of-distribution inputs. We evaluate our method on a Unitree Go2 quadruped equipped with a RealSense D435 depth camera in real-world settings. Under normal visual conditions, our method enables the robot to successfully perform high-dynamic parkour maneuvers, including traversing large gaps, elevated platforms, and narrow beam terrains. Under severe visual degradation, such as heavy occlusion and strong illumination changes, the robot continues to maintain stable and robust locomotion performance.

    quadrupedlegged locomotion
  49. arxiv:2611.18298 · cs.RO
    Objectives as Policies: Vision-Language Model Cost Programs for Contact-Rich Manipulation
    Zihao Lu, Trong Gia Hung Nguyen, Fuquan Wang, Yifan Zhu

    Vision-language models (VLMs) have shown strong semantic and spatial reasoning capabilities in 3D environments. Recently, connecting VLMs with model-predictive controllers (MPC) has emerged as a paradigm for combining strong semantic reasoning with physics-based planning. However, connecting these two for contact-rich robotic manipulation requires a task representation that bridges the two and captures how physical interactions can be reasoned by VLMs. In this work, we introduce Objectives as Policies (OaP), a framework that connects a pretrained VLM to a shared sampling-based MPC through an editable objective program. Each program specifies a sequence of stages for completing a task, combining running and terminal costs from a cost library with stage completion conditions. Given a natural-language task instruction and scene observation, we build a simulation-based digital twin through real-to-sim, and the VLM generates an objective program that is executed by the MPC in the simulation. The program is iteratively updated by the VLM based on MPC rollouts, before being deployed in the real world. We evaluate our framework in both simulation and the real world on four tabletop manipulation tasks, and show that it outperforms strong baselines. Additional videos and examples are available on the project website: https://zihaolu001.github.io/oap-project/

    manipulation
  50. arxiv:2611.18293 · cs.AI
    PrivacyPR: A Dataset of Real-World Privacy Fixes for Benchmarking Privacy-Aware Code Review
    Hua Yang, Aditya Bhardwaj, Shidong Pan, Bowen Xu

    Coding agents increasingly review code, yet whether they catch the privacy defects that laws such as the GDPR require software to avoid remains largely untested, partly because few datasets connect real privacy defects to their accepted fixes and review context. We present PrivacyPR, a dataset of 4,131 merged GitHub pull requests (PRs) from 1,469 repositories, created between June 2018 and January 2024. Each PR includes its commits, diffs, review discussion, and LLM-generated labels for privacy relevance, fix type, and GDPR principles. PrivacyPR provides 1,169 high-confidence privacy fixes and 1,643 non-fix PRs, together with Verified, a manually checked split of 250 privacy fixes and 250 non-fix PRs, and Small, a 100 + 100 subset for evaluation. On Verified, both annotators endorsed 72% of the LLM's GDPR-principle assignments. Beyond supporting empirical study of how privacy defects arise and are fixed, PrivacyPR provides a benchmark for privacy-aware code review on real, developer-accepted fixes. Its evaluation harness reverts each fix to reintroduce the defect. On Small, two coding agents detect only 42-51% of defects under a generic review prompt, but 88-91% when explicitly asked to check privacy, while flagging 11-15% of non-fix PRs. These results show that privacy defects can be detected by coding agents, but that detection depends strongly on review intent and comes with false alarms. Data, code, reviews, judge verdicts, and an extensible evaluation harness are publicly available at https://doi.org/10.5281/zenodo.23744765.

    benchmark
  51. arxiv:2611.18284 · cs.LG
    Clean Tickets, Wrong Orders: Catching Silent Errors with a Language Model in Your Browser Tab
    Umar Ahmed, Maxim Dunaev, Mohammad Nour Al Awad, Sergey Ivanov +1

    A small language model that turns a typed order into structure can fail in two ways. It can print something broken, which any system rejects. Or it can print a clean, well formed order that nobody placed. Our companion work calls the second kind a silent error, and it is the dangerous one. We present a web demo in which a fine-tuned 0.8B parameter model runs entirely inside the visitor's browser tab, through llama.cpp compiled to WebAssembly. No order text leaves the device. The page parses orders as the visitor types, scores every ticket with the PIZZA benchmark's own matcher ported to TypeScript, refuses anything off the menu with "?" so the model can face the open web, and lets visitors hunt for silent errors or try to cause one. Run in the browser on all 1,357 PIZZA test orders, the deployed model writes the same output as the server evaluation for 99.0% of them and reaches the same exact match (0.882); we also report its latency on 3 consumer laptops. Demo: https://wrong.cuyor.com

    benchmark
  52. arxiv:2611.18279 · cs.CL
    GT-FENG: General Theory of Factual Evaluation and Normative Grounding in Language Models *Author(s)
    Jiayu Feng

    Whether a language model's answer is correct often depends on context: the same value can be right under one jurisdiction, date, or legal definition and wrong under another. We introduce GT-FENG (General Theory of Factual Evaluation and Normative Grounding), the first general framework to unify contextual/temporal correctness and rule-based error attribution across law, policy, and other rule-governed domains. GT-FENG checks answers against the applicable rules and identifies documented rules that match incorrect answers. It shows that when multiple rules give the same value, the answer alone cannot tell which rule the model used. In Medicaid income-limit answers from two models, documenting the asked state's own records reduces the number of wrong answers that match only other states' values by 68-78%, and more than half of stable wrong answers then match two or more documented sources. GT-FENG provides a foundation for evaluating language models in rule-governed domains and for future benchmarks and evaluation toolkits. We release GT-FENG as an open-source Python package, gtfeng.

    benchmark
  53. arxiv:2611.18272 · cs.AI
    From Bytes to Behavior: Request-Level Attribution from Live GPU Hidden States
    Jonathan Pan, Yidi Yuan, Xin Wei Chia, Xinhui Su +2

    Existing memory forensics for machine learning systems typically targets static artifacts, such as model weights or prompt logs, rather than the transient computational states of a live inference process. This gap limits an examiner's ability to analyze the specific behavior of a running large language model (LLM) service that cannot be restarted or modified. We present a method for acquiring per-layer residual-stream activations from a live GPU process using a non-modifying debugger, without requiring prior knowledge of the model's architecture parameters or weights. The approach employs a general six-stage process with platform-specific adapters, utilizing either a PyTorch allocator clock or a kernel-launch clock to identify and read tensor allocations from video memory. We validate this acquisition on 33 open-source decoder-only models across nine architecture classes, demonstrating that the recovered bytes are byte-exact and usable for downstream analysis. As a use case, we apply linear probes trained on standard framework activations to the recovered states, successfully distinguishing refusal from compliance with an AUROC of 0.88 on held-out prompts. This work establishes that live, observational access to intermediate hidden states is feasible on named inference platforms, providing a new source of evidence for examining the operational behavior of deployed AI systems.

    memory
  54. arxiv:2611.18257 · cs.CL
    NER-D: A cross-domain benchmark for German NER
    Marc Brinner, Philipp Cimiano, Karolina Hanna Czok, Regina Stodden +2

    We present NER-D, a cross-domain benchmark for German NER comprising 13 datasets spanning 9 domains, including General/Politics, Wikipedia, Legal, Medical, Historical, Financial, Agriculture and Archaeology. We examine the performance of both proprietary LLMs and state-of-the-art small open-source models using normal prompting and further examine the performance gains achievable via in-context learning and LoRA fine-tuning. Our results show that without any sort of optimization a proprietary model (GPT-5.4 mini) outperforms the smaller open-source models on 6 out of 13 datasets, but LoRA fine-tuning enables the smaller models to substantially outperform the proprietary baseline. We further analyze the influence of the domain and the tagset used in each task on the performance achieved in the different evaluation settings. Our benchmark will be available as a resource for the development of future German NER methods on both general and specialized domains.\footnote{We make our code, data, and models publicly available upon acceptance of this paper.

    benchmark
  55. arxiv:2611.18252 · cs.LG
    ParetoM$^3$: Learning on the Pareto Set under Preference Guidance via Min-Max-Min Optimization
    Pei Tang, Songtao Lu

    Learning across multiple inherently conflicting objectives requires selecting desirable trade-offs from the Pareto set. In this work, we study preference-guided multi-objective optimization, where a predefined preference function is minimized over the weak Pareto set of multiple nonconvex objectives. Leveraging a merit function that vanishes on the weak Pareto set, we propose a penalty-based min-max-min formulation, termed the M$^3$ problem, which unifies preference optimization and the enforcement of weak Pareto optimality in a single objective. We further develop ParetoM$^3$, a single-loop first-order algorithm for computing stationary solutions of the resulting M$^3$ problem. Under a local Kurdyka-Lojasiewicz assumption, we prove that ParetoM$^3$ finds an $ε$-stationary point of the M$^3$ problem that is also approximately weakly Pareto optimal for the original multi-objective problem when the penalty parameter is sufficiently large, with explicit non-asymptotic convergence rates. We also show that this stationarity notion recovers existing optimality criteria, including approximate preference stationarity and approximate Karush-Kuhn-Tucker conditions, under the same assumptions used in prior works. Experiments on a synthetic benchmark, image classification, regression, and math reasoning demonstrate that ParetoM$^3$ reliably attains preferred weak Pareto optimal solutions while achieving strong practical performance. Code is available at https://github.com/tpgoodboy/ParetoM3.

    benchmark
  56. arxiv:2611.18238 · cs.LG
    Small Is Not Dead: Distance-Aware Cluster Revival for e-means
    Eyal Weiss

    e-means (Bessa et al., 2026) is a recent mini-batch k-means algorithm, now part of the FAISS library, that reaches the error of Lloyd's algorithm about an order of magnitude faster on large embedding datasets. Every five mini-batches, it removes every cluster whose running point count falls below a tenth of the average and reuses its centre to split a large cluster. In the authors' ablation, removing this mechanism, called dead-cluster revival, raises the error more than removing any other component added to the moving-average update. The rule, however, considers only cluster size: it cannot distinguish a cluster that is small because it is redundant from one that covers a rare, isolated group of points, and on data with such groups it repeatedly removes centres whose removal is costly. We add a second condition, based on the distance between centres: a small cluster is removed only if its centre is close to another centre, relative to a running estimate of the mean squared error. When a centre equals the mean of its points, this squared distance is an upper bound on the average increase in squared error, per point of the cluster, caused by removing the centre. The condition costs one nearest-centre search for the small clusters every five mini-batches and adds no measurable running time. On the e-means paper's embedding benchmarks, it never prevents a removal, so the results are identical to those of FAISS. On datasets with small isolated groups not used during development, it reduces the test error after two epochs by roughly 10% to 80% relative to count-based revival, and it preserves the advantage of k-means++ initialization; on the other evaluation datasets, it stays within about 1% of the better of revival enabled and disabled. We also report two failure modes that the analysis predicts. This is a CPU proof of concept; GPU implementations are left to future work.

    benchmark
  57. arxiv:2611.18235 · cs.LG
    NeoResearch: Literature- and Experience-Guided Autonomous Time Series Model Discovery
    Ze Guo, Mingyue Cheng, Xiaoyu Tao, Shuo Yu +1

    Developing effective forecasting models for diverse time-series domains requires substantial expert knowledge, evidence from scientific literature on the web, and iterative experimentation. However, translating this knowledge and research experience into model designs requires preserving source context and testing applicability under task-specific assumptions. In this paper, we propose NeoResearch, a literature- and experience-guided autonomous research framework that leverages large language models (LLMs) as researchers to discover lightweight, domain-specific forecasting models. Specifically, NeoResearch establishes an expert-inspired dual-loop research workflow, coupling hypothesis-driven architecture exploration with iterative training and hyperparameter optimization through experimental feedback. To enable knowledge-grounded discovery, the agent autonomously mines scientific literature from the Web, extracting modeling mechanisms, empirical evidence, and applicability conditions to formulate task-grounded, testable architecture hypotheses. Furthermore, a hierarchical experience memory links literature findings to experimental outcomes and consolidates conditional lessons, guiding architecture discovery and training optimization while supporting reuse across forecasting tasks. Experiments on ten datasets show competitive performance and support dual loops, literature extraction, and memory reuse. Our code is available.

    memoryagent
  58. arxiv:2611.18230 · cs.AI
    Personalization of Basic and Moral Values Guiding LLM-Controlled Assistant Robots
    Mateusz Wozniak, Qi Gao, Agnieszka Wykowska

    Personal assistant robots powered by large language models (LLMs) are likely to enter everyday life within the next decade, but the values that should guide their behaviour remain largely unspecified. Current alignment practice steers a single model toward an aggregate of human values, ignoring that individual users may prefer a robot whose value profile differs from their own. We administer parallel self- and robot-targeted versions of three established psycho-metric instruments - the simplified Schwartz Values Survey (SVS), the Portrait Values Questionnaire (PVQ-21), and the Moral Foundations Questionnaire (MFQ) - to potential future users (N=22), capturing both their personal values and the values they would want an assistant robot to embody. We compare the three instruments as personalization tools using the UEQ-S and benchmark fifteen contemporary LLMs against the resulting profiles. Participants wanted their robot to score markedly lower than themselves on personal-focus values and higher on social-focus values; moral profiles were nearly identical. LLMs prompted to act as robots shifted only weakly toward the desired profile, indicating that default model behaviour and prompt engineering are no substitute for explicit value personalization.

    benchmark
  59. arxiv:2611.18229 · eess.SY
    A Reinforcement Learning Approach to Space-Filling Input Design for Nonlinear System Identification
    Charly Witmeur, Gonçalo G. Cruz, Maarten Schoukens, Mark C. Runacres +1

    In this paper, we propose a reinforcement learning (RL) approach to space-filling input design for nonlinear system identification, in which an agent learns a control policy that actively drives the system to explore its state space. Classical input design, developed mainly for linear models, minimises the covariance of the estimated parameters. For nonlinear black-box models, however, the dominant error is the bias induced by the model structure over the region of interest, which motivates space-filling strategies that seek to cover the system state space. Existing space-filling approaches rely on predefined input signal classes and require solving explicit optimisation problems. Unlike these methods, the proposed approach requires no input class to be selected and no frequency band to be tuned for the system at hand, satisfies an amplitude bound by construction, and adapts to varying initial conditions through its observation-based policy without re-optimisation. The learned excitation signals exhibit physically meaningful behaviours, such as resonance excitation and amplitude modulation. The approach is evaluated against tuned swept sine and crest-factor-optimised multisine excitations, and against a state-of-the-art space-filling multisine design, on two nonlinear benchmark systems: a mass-spring-damper with nonlinear stiffness and a Van der Pol oscillator. Under the same amplitude bound, the learned policy achieves the best overall space-filling performance on both systems and distributes its samples more uniformly than the state-of-the-art design.

    agentbenchmark
  60. arxiv:2611.18190 · cs.LG
    MarathonTS: a benchmark of long synthetic time series from window assembly and long-series generators
    Ilyas Varshavskiy, Sergey Shulgin, Shuhrat Khalilbekov, Bonu Boboeva +1

    Many applications of synthetic time series require a series that is much longer than the windows on which a generative model is trained. Such a series can be assembled from a pool of short generated windows, which requires deciding which windows are taken, in what order, and how their seams are processed, or it can be produced directly by a long-series generator. This study presents MarathonTS, to our knowledge the first open benchmark that compares both families under one protocol. Every method produces a series of the same length on 9 open hourly datasets with 10 seeds. The series is evaluated by its fidelity to the real data (83 metrics in 12 groups, against three real references: train, validation and an out-of-time split), by its utility (models are trained on the synthetic series only and tested on real data, TSTR against TRTR, in 14 classification and regression tasks), by its memorisation of the training data and by its computational cost. The statistical analysis is carried out as a separate stage that reads the published tables and gives the rankings of every method in every slice, the answers to the research questions with tests and uncertainty, and the results of a pre-specified confirmatory family. In the main run, the methods differ in both confirmatory cells; the choice of windows matters more than the processing of their seams, which does not improve the series, and no method trains the downstream models as well as the real series. The top tier of the leaderboard is reproduced on other hardware, whereas the exact places within it are not. For the reproducibility of the study, the code, the data hashes and all results are publicly available (links below).

    benchmarkleaderboard
  61. arxiv:2611.18189 · cs.CV
    SemRAG: Semantic-Guided Sentence-Level Retrieval-Augmented Sign Language Production
    Xiao Liu, Shiwei Gan, Yafeng Yin, Peishuo Liu +4

    Sign language production (SLP) aims to generate sign pose sequences that faithfully express spoken-language text while preserving the rhythm and coordination patterns of signing. Common SLP methods learn text-to-pose mappings from training data but often suffer from semantic inaccuracies in long or complex sentences. Retrieval augmentation offers a way to introduce realistic signing priors to provide more explicit guidance for text-to-pose generation. However, existing retrieval-based SLP methods typically operate at the gloss or word level, relying on gloss-pose dictionaries or isolated sign segments that require costly annotations and fail to preserve continuous sentence-level dynamics. To bridge this gap, we propose SemRAG, a semantic-guided sentence-level retrieval-augmented framework that retrieves full-sentence signing templates as priors without constructing gloss/word-level sign dictionaries. While such templates preserve local motion patterns and global rhythm, they also introduce a semantic mismatch problem: irrelevant motions in the retrieved sentence may interfere with the target semantics. To address this issue, we introduce an SLT-guided template semantic localization module that identifies target-relevant temporal regions within the retrieved template while suppressing irrelevant signing content. To adaptively incorporate the template features into the generator, we further introduce layer-specific learnable gates that regulate their contribution across generator layers. Extensive experiments show that SemRAG improves semantic consistency and motion quality over baselines under both back-translation and motion accuracy metrics.

    retrieval-augmented
  62. arxiv:2611.18186 · cs.RO
    NPU-Accelerated GNSS Signal Acquisition
    Taro Suzuki

    On small single-board computers (SBCs), GNSS signal acquisition is a major CPU load. This paper proposes using the neural processing unit (NPU) of an SBC not for machine learning but as an accelerator for the parallel code search in GNSS signal acquisition. The discrete Fourier transform (DFT) used in acquisition is decomposed by the four-step DFT into matrix products of small DFTs and expressed by $1\times1$ convolutions, element-wise products, and transposes, which NPUs execute efficiently. Furthermore, we propose a method that divides the DFT, the non-coherent integration, and the peak detection between the CPU and the NPU according to the configuration of the NPU (an NPU with its own memory connected via PCIe, or an NPU sharing memory with the CPU). The proposed method is implemented in the open-source software GNSS receiver Pocket SDR and evaluated on a Raspberry Pi 5 combined with the external AX8850 NPU and on the NPU built into the SoC (RK3588) of the ROCK 5C, a small SBC of comparable size. Compared with acquisition by an optimized CPU implementation, the acquisition of short codes on the Raspberry Pi 5 with the AX8850 is 3.0-3.8 times faster. In the evaluation of the whole receiver, the number of tracking channels that reach their first lock within 30 s with the NPU is 1.5 times that with the CPU, and the time until 90% of the channels are locked is reduced from 66 s to 26 s. Furthermore, because the CPU load of acquisition is reduced, the number of tracking channels that can be processed in real time increases from about 175 to about 230. These results show that using the NPU on SBCs with limited CPU resources is effective for achieving both GNSS signal acquisition and tracking. The implementation is publicly available as open source under the name PocketSDR-NPU.

    memory
  63. arxiv:2611.18184 · eess.SY
    PTMTD: A Topology-Aware Temporal-Graph Attention Network for Joint Cyber-Attack Detection and Relay-Level Localization in Power System Protection Networks
    Shahin Razazni, Mohammad Ali Alirezapouri

    The digitization of protection relays exposes power grids to sophisticated cyberattacks, yet existing GNN detectors omit temporal encoding, rely on static topologies, and lack relay localization. We propose PTMTD-a Pearson-driven Temporal Multimodal Topology-aware Detection Network-coupling a dilated TCN with multi-head GATv2 in a dual-head architecture for simultaneous 37-class detection and four-relay localization. PTMTD constructs its graph from measurement correlations (no impedance data) and is trained with a composite loss jointly optimizing both tasks. On the MSU-ORNL benchmark (78,377 samples, 37 classes), it achieves 99.27% accuracy, 99.25% macro F1, and 95.98% localization, outperforming GraphKAN, LSTM, GCN, and Random Forest by 5.64-11.15 percentage points (p < 0.05). Ablations confirm gains from temporal encoding (+2.38%), graph attention (+2.69%), residuals (+2.07%), and correlation-driven supervision. With 639k parameters and 2.5 ms CPU inference, PTMTD is lightweight and deployment-ready. While validated on a transmission testbed, the architecture applies to microgrids with similar relay structures. Code and models are publicly available.

    benchmark
  64. arxiv:2611.18183 · cs.MA
    When Should a Virtual Patient Ask? RISP-Guided Proactive Information Seeking in Multi-Agent Patient Simulation
    Jingyao Cai, Shuaijun Liu, Abdul Rehman, Xiaowan Li +4

    Virtual patient (VP) systems support nursing communication training by providing simulated encounters in which learners practise explaining care and responding to patient concerns. Although large language models (LLMs) support flexible dialogue, current approaches give less explicit attention to how patients' information needs develop and guide the content and timing of their questions. We draw on the Risk Information Seeking and Processing (RISP) model, which relates information seeking to perceived risk and the gap between perceived knowledge and desired understanding. Guided by this perspective, we propose the RISP Multi-Agent System (RISP-MAS), a VP system for patient-initiated information seeking. The system links ongoing assessment of the patient's perceptions and information needs to coordinated decisions about whether to seek information, when to ask and which uncertainty to address. Evaluation of nursing student-VP interactions suggested that RISP-MAS exhibited plausible patient information-seeking behaviour, as indicated by expert assessments, while student feedback reflected positive interaction experiences. This work extends patient simulation by connecting a psychological account of information needs to explicit decisions about question initiation. The approach offers nursing students opportunities to practise recognising unresolved patient concerns and tailoring explanations to patients' information needs.

    multi-agentagent system
  65. arxiv:2611.18176 · cs.RO
    ESI: An Embodied Self-Improvement Flywheel for LLM-VLA Agents
    Yundi Cai, Zihan Wang, Chenxi Han, Yi Cheng +2

    Independently pretrained language models and vision-language-action (VLA) policies provide complementary building blocks for embodied agents, but connecting them does not automatically produce coordinated behavior. The VLA may generate actions that diverge from the LLM's intended subtask and, upon regaining control after a correction, undo the promising state restored by the LLM. We introduce ESI (Embodied Self-Improvement), an automated flywheel that uses environmental interaction to align reasoning and action. ESI treats the modular agent as the unit of post-training, repeatedly refining the prompts that guide the LLM, the shared skill interface that expresses its intent, and the VLA parameters that realize that intent, then redeploying these components together. Our key insight is that collaboration itself can supply supervision for improving this interface. The LLM's subtask decisions provide semantic supervision for what the policy should execute, while its effective corrections and subsequent student continuations provide action-level supervision for recovery and handoff. The student also learns from its own effective execution, allowing locally useful experience to contribute even when the overall episode fails. On ten selected RoboDojo tasks, ESI achieves a mean Score of 62.03 and a mean success rate of 49.41% using an LLM-VLA agent without generating task-specific control code. On both metrics, it outperforms the published reference results of all hybrid embodied agents, including the Agent+WAM system GO-3 Nano Preview (47.27 Score and 36.60% success), as well as all compared code-based methods.

    vision-language-actionvlaembodiedagentembodied agentself-improvement
  66. arxiv:2611.18172 · cs.CV
    STaG: Human-Annotation-Free Unsupervised Domain Adaptation for Text-Based Person Retrieval
    Ji Yucheng, Bai Yang, Li Yang, Zhou Xinyu +1

    Real-world deployment of Text-Based Person Retrieval (TBPR) is bottlenecked by the absence of textual annotations in new domains. Existing Unsupervised Domain Adaptation (UDA-TBPR) methods still assume access to target-domain human-written annotations (albeit unpaired with images), limiting their practicality. We consider a more practical setting: human-annotation-free UDA-TBPR, with labeled image-text data in the source domain and annotation-free, unlabeled images in the target domain. This setting introduces two core challenges: the absence of cross-modal alignment in the target domain, and visual domain shifts from source to target. To tackle these, we propose STaG, a dual-path framework comprising Style-Agnostic Text Alignment (SATA) and Heterogeneous Graph Memory Alignment (HGMA). Specifically, SATA leverages source-domain data to decouple identity-discriminative semantics from machine-generated stylistic artifacts, establishing robust cross-modal alignment for the target domain. Meanwhile, HGMA exploits cross-domain semantic neighborhoods via a memory-based graph to learn domain-invariant visual features. Ultimately, STaG achieves state-of-the-art performance across three benchmarks under this realistic setting.

    memorybenchmark
  67. arxiv:2611.18157 · cs.CL
    A Multi-Layered AI Pipeline for BPMN Generation from Natural Language
    Alexandru Rancea, Ionut Anghel, Tudor Cioara, Ovidiu Anchidin +1

    The automated transformation of unstructured natural language process descriptions into Business Process Model and Notation (BPMN) diagrams remains a challenging task, due to linguistic ambiguity and the difficulty to detect non-adjacent control-flow relationships. This paper proposes a solution for generating BPMN models from natural language descriptions through a multi-layered pipeline that integrates transformer-based architectures, Large Language Models and ontology population. The proposed method starts with natural language descriptions and separates textual processing, control-flow relation extraction, and structural validation into dedicated components. Entity extraction is performed using a fine-tuned RoBERTa-based transformer model while a control-flow engine captures non-adjacent relationships through LLM-based inference. To improve generalization under limited available data, the approach uses a dynamic few-shot strategy with con- trolled template input construction, enabling improved generalization under limited data availability. Additionally, an ontology population mechanism is introduced to represent the generated process models as semantic knowledge structures. For evaluation purposes we have used from the literature and industry reference processes. Additionally, we have validated the proposed method on a real-world heart disease patients transitional care process, capturing practical constraints and multi-actor collaboration, and serving as an expert-defined benchmark scenario. Experimental results demonstrate strong performance in the relation extraction task, achieving an accuracy of 90.7%, as well as maintaining structural alignment between the generated and reference BPMN models across multiple evaluation scenarios scenarios.

    benchmark
  68. arxiv:2611.18150 · cs.AI
    Who Should Be on the Panel? Task-Conditioned Persona Selection for LLM Trading Agents
    Boming Yang, Yuquan Wang, Jiaxian Guo, Yusuke Iwasawa +1

    Multi-agent LLM systems are built by fixing a cast of roles and then designing how the cast coordinates. Which agents should make up the group for a given task is left to the designer, and the same cast is reused for every task. Financial trading makes the cost of that choice visible: stocks trade differently, and the trading styles suited to one need not suit another. We study task-conditioned persona selection, composing the group from the characteristics of the task it will face, and instantiate it with FinPanel, an LLM multi-agent trading system in which a recruiter selects a panel of named trader personas for each stock from a direction-blind statistical sketch of how the stock trades, each persona gives an independent daily call in its own style, and a manager commits the trade after reading the panel's arguments and their desk-kept, hierarchical track records. Because nothing is returned to the traders, the panel stays the panel that was selected, and it can trade a new stock from an empty record. On eight US stocks over the out-of-knowledge year 2025, against three prior LLM trading agents re-implemented in the same framework and against rule-based, predictive and reinforcement-learning baselines, the selected panel is ahead of a fixed panel of the same size and of the prior agents run as their papers prescribe, with a shallower drawdown than buy-and-hold on every stock; it keeps its performance with no warm-up and with no record at all, whereas the memory-augmented agents' returns swing with their warm-up; and neither more seats nor the manager's reading of the arguments substitutes for selecting the right ones. Who is in the group is a design decision to be made per task, before how the group coordinates.

    multi-agent
  69. arxiv:2611.18149 · cs.LG
    Which Model, For Which Variable? Per-Variable Forecast Combination in Sparse ICU Time Series
    Marius Fracarolli, Stefan Riezler

    Clinical forecasting papers typically report a single best architecture. We show that on multivariate ICU physiology this question conceals variable-specific differences: no architecture is best on every variable, and which model wins changes from variable to variable. Across five ICU databases (MIMIC-III, MIMIC-IV, eICU, HiRID, and UMM (University Medicine Mannheim), a German single-center ICU) we forecast 94 channels over a 24-hour horizon with a pool of 14 models, and combine them with weights fitted per variable. On test, the combination beats the best single model on all five databases by 6.1-11.0% mean squared error (mean 8.7%). The gain is concentrated where clinical data is thinnest: the sparsest third of channels improves by 11.9% and laboratory values by 7.5%, against 3.1% for the densest third. Rank-based harmonic weighting of the best five models per variable, with no fitted weights at all, recovers about four-fifths of the gain. Extending the pool afterwards with a per-channel gradient-boosted tree, ARIMA and ETS makes the tree the validation-selected best single model on all five databases; per-variable combination still improves on it by 2.9-5.9% (significant on four of five), while equal weighting of the same members does not. As on univariate benchmarks, learned combination beats simple averaging - but only once the weights are constrained. We also report which model to prefer when, and show that the ordering changes with database size.

    benchmark
  70. arxiv:2611.18145 · cs.LG
    Compiling Specialized GPU-Initiated Networking
    Javid Baydamirli, Ismayil Ismayilov, Kaan Oktay, Didem Unat

    Machine learning systems specialize computation to the workload and hardware, but communication remains in generic libraries whose device code and transport decisions are not visible to high-level compilers. We present Obaleth, a compiler that makes GPU-initiated communication part of a kernel's intermediate representation, co-optimizes it alongside computation, and lowers it to specialized transport code. Obaleth plans completions around buffer lifetimes and optimizes memory ordering, synchronization, and batching across NVIDIA and AMD GPUs. On H100 GPUs, Obaleth sends small messages up to 3.3$\times$ faster, and on 256 GPUs runs all-to-all 1.25$\times$ and 2.3$\times$ faster than NVSHMEM and NCCL GIN, with much smaller memory and register use. In vLLM, it serves Qwen3-235B on 16 B200 GPUs and answers requests up to 1.29$\times$ faster than with DeepEP.

    memory
  71. arxiv:2611.18140 · cs.AI
    Large Language Model-Empowered Intelligent Agent for Adaptive Fast-Charging Control of Lithium-Ion Batteries
    Joey Chan, Haoyu Pan, Zhen Chen, ErShun Pan

    Fast charging of lithium-ion batteries is a multi-objective control problem that requires reducing charging time and sustaining high energy throughput while suppressing accelerated degradation and safety-critical electrochemical reactions. Conventional protocols are easy to implement but have limited ability to adapt to variations in battery chemistry, operating conditions, and aging state. Model-based optimization and conventional data-driven control often require accurate models, costly iterative search, or task-specific policies. We develop a large language model (LLM)-empowered intelligent agent for adaptive battery fast-charging control. In this closed-loop framework, the agent interprets structured battery states, generates executable charging actions, observes the resulting battery response, and continuously adjusts subsequent decisions. We construct a large-scale simulation corpus with PyBaMM across five battery chemistries and 11 parameter sets and convert the resulting state-action-response trajectories into structured process data for agent training. Our two-stage strategy combines supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO). SFT establishes battery-control semantics and feasible decision priors from simulation-derived demonstrations. GRPO aligns the agent with a multi-objective control objective covering energy throughput, degradation mitigation, and operational constraints. Closed-loop simulations show that, under matched energy-throughput requirements, the agent achieves longer cycling life than conventional charging baselines and optimization-based strategies across heterogeneous battery configurations. This study provides a simulation-grounded proof of concept for using LLM agents as sequential decision makers in battery management and takes a step toward language-model-enabled control of physical energy systems.

    agentllm agent
  72. arxiv:2611.18139 · cs.AI
    PUMA-Former: Uncertainty-Aware Agentic Computation Offloading in MEO-LEO Satellite Networks With Incomplete Information
    Yi Jing, Chunxiao Jiang, Jiawei Wang, Jiachen Sun

    Satellite edge computing distributes mobile workloads across moving processors whose service capacity must be coordinated over time. In medium Earth orbit (MEO)--low Earth orbit (LEO) networks, periodic reports leave CPU states stale or missing, making offloading sensitive to uncertain capacity and shared-destination congestion. We propose PUMA-Former, an uncertainty-aware agentic framework for cooperative offloading under incomplete information that balances completed work, service delay, and execution cost. A MEO coordinator maintains distinct-report memory and closes the loop between available evidence, capacity estimates, allocation commands, and execution feedback. Latent-delay attention weights plausible measurement-age hypotheses, and historical innovations refine CPU estimates around the latest received report. Queue-error-aware fusion incorporates auxiliary pressure while accounting for backlog uncertainty. The inferred resource moments condition a structured Transformer allocator for joint destination, partition, bandwidth, CPU, and power decisions. Bounded risk-aware calibration then adjusts workload splits to reflect uncertainty and shared destination load. In traffic-driven development evaluation, PUMA-Former increases admitted compute-work completion by 8.62%, reduces mean modeled service delay by 34.78%, and reduces mean episode-P95 delay by 71.08% relative to ATLA-adapt. It also improves completion and normalized energy over fixed-half round-robin offloading. Service gains over the learning baselines involve higher normalized energy.

    memoryagentic
  73. arxiv:2611.18130 · cs.AI
    JFinBench: A Continuously Refreshable Japanese Financial Benchmark from the Latest Certification Exams
    Issa Sugiura, Taro Makino, Takashi Ishida

    Large language models (LLMs) are increasingly applied to finance, creating a need for benchmarks that evaluate their capabilities across financial systems with different regulations and practices. However, existing financial benchmarks provide limited coverage of country-specific financial systems, and they are difficult to keep up to date as released questions leak into training data. We introduce JFinBench, a Japanese financial benchmark comprising 321 examples from the 2026 editions of six regularly administered certification exams covering accounting, tax, financial planning, and banking. To construct it, we develop a human-in-the-loop agentic pipeline in which a coding agent writes exam-specific conversion code and a review page, while a human verifies every example and feeds corrections back to the agent. This pipeline adapts to heterogeneous exam formats and reduces the effort of incorporating new exam editions, enabling periodic refreshes that mitigate data contamination. Evaluating 12 recent LLMs, we find that Claude Fable 5 and GPT-6 Astra exceed 93% overall, while even the best open-weight model, Kimi K3.1 Pro, trails Claude Fable 5 by 12 percentage points. When models can additionally use a terminal and web access, they retrieve the knowledge they lack and every model improves. The gap narrows but about 4 points remain, and the ranking of models stays largely the same, suggesting that knowledge held in the model itself remains valuable even with tools. We release the dataset, the evaluation code, and the construction pipeline.

    agentagentichuman-in-the-loopbenchmark
  74. arxiv:2611.18127 · cs.LG
    PRISM: A Post-Prediction Reliability Framework for Evidence-Guided Correction and Selective Abstention
    Omm Swarup, Shrishail M. Hiremath

    A model may produce a prediction with high confidence and still get the outcome wrong. This is a challenge in sequential applications, where the system evolves over time and deployment data may differ from the data used during training. To address this issue, we propose PRISM (Prediction Reliability and Intelligent Self-Correction Mechanism), a framework that evaluates prediction reliability after the initial output is generated. It estimates the risk of error and examines predictions identified as potentially unreliable. For these cases, PRISM analyses available signals, retrieves similar situations from historical data, and verifies whether an alternative prediction is supported by the evidence. It can then retain the original prediction, apply a verified correction, or abstain when the evidence is insufficient. We evaluate PRISM on second-innings match prediction in the Indian Premier League (IPL). Ball-by-ball records are converted into over-level observations to capture changes in match conditions throughout an innings. To preserve temporal order, earlier IPL seasons are used for training, IPL 2024 for validation, and IPL 2025-2026 for out-of-time testing. The dataset contains 8,665 over-level observations from 475 matches, with the test partition comprising 2,568 observations from 142 matches. In one experimental record, Logistic Regression, Random Forest, and HistGradientBoosting achieve accuracies of 65.2%, 68.4%, and 70.2%, respectively. The corresponding PRISM evaluation reports a selective accuracy of 82.4% at 88.5% coverage, with a Brier score of 0.126 and an expected calibration error (ECE) of 0.019.

    self-correction
  75. arxiv:2611.18125 · cs.LG
    Efficient Poincaré Convolutions through Patch-Statistics Factorization
    Lukáš Jochymek, Markéta Vašinková, Petr Gajdoš

    Published implementations of the Poincaré convolution in Hyperbolic Neural Networks++ (HNN++) map every overlapping image patch to the Poincaré ball before evaluating its score. We show that the clipped score depends on each patch only through two statistics, its squared norm and its inner products with the filters, which a native convolution and a single-channel box filter compute directly. The factorized layer therefore evaluates the specified score exactly, without constructing ball-valued patches, and its single-precision outputs and gradients agree with double-precision evaluations at least as closely as those of the explicit reference. On an RTX A6000 under two versions of JAX, complete training updates of a Poincaré U-Net and a Poincaré ResNet-20 are 1.5 to 1.8 times faster than the faster of two explicit implementations that map patches to the ball (1.2 to 1.4 times against an explicit implementation that skips this mapping); on a V100, the gain over the gather-based one persists as the ResNet-20 is widened eightfold. In 15 paired training runs per task, updates are 2.0 and 1.8 times faster than a reference that follows the published execution order, with no consistent change in final segmentation or classification quality, and peak memory falls by 38 to 55% in a U-Net scaling study. Replacing the convolutions of the official PyTorch Hyperbolic U-Net makes its training step 1.4 times faster with half the memory. Evaluating an existing layer from its patch statistics thus makes Poincaré convolutional networks cheaper to train in the tested models and software stacks.

    memory
  76. arxiv:2611.18121 · cs.AI
    Optimizing Drilling Trajectories in Multi-Material Structures Using Quantum Annealers
    Eneko Osaba, Antonio Rubio-Mateos

    In this work we present a quantum-hybrid optimization approach for planning drilling trajectories on multi-material structures, with the objective of minimizing total energy consumption. The proposed formulation models the drilling process as a constrained routing problem in which each hole must be visited while accounting for heterogeneous material properties, material-dependent drilling costs and displacement energy, as well as tool-wear penalties. Our method incorporates realistic industrial constraints, including open-path routing and mandatory grouping of holes belonging to the same material to avoid frequent tool-bit changes. The modeled problem is solved using D-Wave's Leap Constrained Quadratic Model Hybrid Solver. To evaluate performance, we benchmark instances ranging from 10 to 55 holes with varying material distributions and cost matrices. The results highlight the capability of quantum-hybrid optimization to handle complex multi-physics constraints and to generate energy-efficient drilling sequences under realistic manufacturing conditions.

    benchmark
  77. arxiv:2611.18114 · cs.AI
    Interpretable weight-space programming builds transparent transformers and brain encoding models
    Yash Mehta, Chandan Singh, Weijia Xu, Richard J. Antonello +5

    Neural networks are powerful predictors, yet their opacity limits their usefulness in science, where the goal is explanation rather than mere prediction. In neuroscience, for example, deep encoding models accurately predict brain activity in response to natural stimuli, but struggle to reveal which stimulus features drive those predictions. Here we introduce interpretable weight-space programming (iWSP), a method in which an LLM agent iteratively builds interpretable models by explicitly programming their weights and features in code, without gradient-based training (backpropagation). Without backpropagation to tune parameters, the agent must draw on its prior knowledge and reason over previous trials to hypothesize, hand-craft, and refine human-interpretable computations mirroring the workflow of a human scientist. LLM audits confirm the interpretability of the hand-crafted models. We first validate iWSP by tasking it with constructing transformers for canonical benchmarks such as arithmetic and sentiment classification, finding that LLM agents can successfully craft compact, accurate networks directly. Next, we apply iWSP to build neural encoding models across vision, language, and audition. In the visual cortex, iWSP outpredicts existing interpretable baselines, uncovers distinct computations across early/higher visual areas, and generalizes to unseen data. In language fMRI, iWSP surfaces a confound in the data and in auditory ECoG, it systematically discovers interpretable acoustic features that predict neural activity. Together, these results establish iWSP as a general paradigm for aligning high predictive performance with human-interpretable scientific discovery.

    agentllm agentbenchmark
  78. arxiv:2611.18113 · cs.CL
    Folding Language Models into Themselves
    Ziyang Zhang, Yubin Jing, Yuyao Li, Haofan Wang +1

    Knowledge distillation usually trains a compact student from scratch, which needs a large amount of data even though the teacher's weights already encode most of what the student has to learn. In this work, we explore building the student directly from the teacher's weights. We propose Concertina, which folds every group of N adjacent teacher layers into one shared block that is executed N times. Each loop step carries a low-rank correction initialized from the layer it replaces, and a per-token gate mixes these corrections, to let every token recover the behavior of the original layers from a single shared block. We further design a distillation data recipe that balances general, mathematical, academic and targeted multiple-choice data, to recover all capabilities of the teacher together rather than one at the expense of another. On Qwen3-4B, a 2.27B Concertina model distilled on 22B tokens retains about 90% of the teacher on commonsense and mathematical reasoning and 84% on average across eight benchmarks. It outperforms same-size dense and layer-sharing baselines, as well as Qwen3-1.7B, which is trained on over a thousand times more data. These results suggest that a compact model need not be retrained at scale, since folding the teacher into itself and distilling it on the right data recovers most of its capabilities. Code is available at https://github.com/myboker/concertina.

    benchmark
  79. arxiv:2611.18110 · cs.RO
    PoliMop: Policy-in-the-Loop Motion Planning in 3D Spaces for Mobile Robots
    Yael Fassbind, Valentin N. Hartmann, Huang Su, Yijiang Huang +1

    Reaching a goal state in a complex environment often requires a robot to overcome challenging terrain and obstacles. Which routes are accessible depends on the robot's physical design, the control policy it uses, and how it modifies its surroundings. We present a kinodynamic rapidly-exploring random tree (RRT) planner that uses physics simulation to exploit skills encoded in the control policy, which allows it to find a path in the scene with respect to the robot's capabilities. The planner samples states from a reduced state space in 3D around the scene and expands the RRT toward each sample using policy rollouts. We guide sampling with a preplanned path for a simple mass-point model to focus exploration on promising regions leading toward the goal state, reducing planning time by up to 63%. Simulation experiments with quadrupeds, humanoids, and wheeled robots demonstrate that our approach supports drop-in integration of various robots and control policies, exploiting their physical and policy capabilities without modifying the underlying planning algorithm. Predefined environmental modifications, such as placing a plank across a gap to build a bridge, will automatically be picked up by the planner as the robot's reachable space expands.

    humanoidquadruped
  80. arxiv:2611.18104 · cs.CL
    Contagious Fingerprints: When LLMs Inherit Another Model's Identity
    Zailong Tian

    The fingerprint by which a classifier recognizes the LLM behind a text is usually treated as a fixed property of the model. We show that it is contagious: a model that continues a conversation begun by another model inherits that model's fingerprint. This undermines audits of API relays, which can serve the first turns of a conversation with the model a customer pays for and silently hand the rest to a cheaper one. After such a handoff, detection of the substitute by a text classifier drops from 0.97 to 0.72 (DeepSeek replacing Claude Sonnet) and from 0.87 to 0.54 for a same-vendor downgrade (Haiku), and retraining on handoff data repairs only substitutes seen in training. Preregistered experiments trace the contagion to imitation of the history's style. History also moves a model away from its own fingerprint: active probes are not camouflaged, but when calibrated on fresh conversations they flag \emph{honest} models on long contexts at rates up to 1.00, and a standard MMD equality test rejects all nine models we test. Both failures come from comparing a history-dependent reply with a history-free reference. We propose \textsc{Lineup}, which samples references from the claimed model \emph{on the same history} and tests the audited reply by its rank among them. \textsc{Lineup} controls false positives by exchangeability and needs no fitted threshold or log-probabilities. With contemporaneous references on two closed models, its false-positive rate is 0.056 (preregistered bound 0.10), and it detects 0.81 of substitute sessions across six pairs after four audited turns.

    long context
  81. arxiv:2611.18101 · cs.CV
    KnowVis: Dual-View Evaluation and Diagnosis of World-Knowledge Grounding in Text-to-Image Models
    Yonghang Yu, Ruoyu Wang, Haoyu Zhang, Lin Qi +2

    Text-to-image (T2I) models have made substantial progress in visual realism, aesthetic quality, and instruction following. However, real-world prompts often go beyond explicit visual descriptions and require implicit facts, structured knowledge, and domain-specific commonsense. Existing evaluations mainly focus on explicit prompt-to-image semantic alignment or specific knowledge subdomains, lacking unified evaluation and diagnosis of world-knowledge grounding in T2I models. We introduce KnowVis for dual-view evaluation and diagnosis of world-knowledge grounding in T2I generation. Built on a hierarchical knowledge taxonomy with case-level annotations, KnowVis combines two complementary benchmark views with a structured MLLM-based VQA judge protocol. The Skill-Tree Benchmark supports fine-grained atomic knowledge evaluation, while the Generalized Benchmark examines open-ended knowledge composition and expression. The protocol uses Required VQAs to assess atomic knowledge correctness in the Skill-Tree Benchmark, and combines Required, Optional, and dynamically discovered knowledge expressions with human-judge collaborative evaluation in the Generalized Benchmark. We validate the protocol's reliability on the Skill-Tree validation split, enabling scalable evaluation and diagnostic error attribution. Experiments show that KnowVis reveals notable gaps in world-knowledge grounding across existing T2I models and provides structured diagnostic signals for knowledge-intensive image generation.

    benchmarkscalable evaluationscalable eval
  82. arxiv:2611.18098 · cs.AI
    Do Machines Learn Deviance From Their Peers? A Randomized Test of Differential Association With AI Agents
    Murat Ozer, Arif Akgul, Amitabh Chakravorty

    Differential association theory and its extension, social learning theory, hold that rule-breaking, and the definitions that make it acceptable, are learned through interaction with others. Because people choose their friends, this claim has been hard to test. We addressed this selection problem with AI agents, which learn only through communication and can be assigned to peers at random. In a preregistered experiment, 420 teams of three agents (1,260 agents; seven AI models) worked on coding tasks that could not be solved honestly. A scripted teammate said that it had cheated (deviant peer), that it had reported the problem honestly (honest peer), or said nothing (silent peer). Half of the teams were told that an auditor read their chat. Compared with a silent peer, a deviant peer increased cheating (66.2% vs. 62.9%; OR = 1.96) and an honest peer reduced it (55.7%; OR = 0.35). Among agents that cheated, a deviant peer increased claims that cheating was normal (13.7% vs. 3.4%), often as the team's norm. Monitoring did not suppress prohibited discussion; it doubled it. Peers mattered only for models that had not yet decided. AI agents offer a new way to test criminological learning theories without selection effects.

    ai agent
  83. arxiv:2611.18078 · cs.CV
    TelePhysica: Agentic Reconstruction of Physically Interactive Scenes
    Ruiqi Shu, Xufeng Zhan, Yuanzhi Liang, Haibin Huang +2

    Direct image-to-video diffusion models entangle pixel synthesis with implicit latent motion, precluding persistent 3D mechanical equilibrium, mid-rollout intervention, and counterfactual force control. Conversely, single-pass programmatic 3D reconstruction from an uncalibrated photograph leaves monocular depth and support clearance underconstrained, causing reconstructed scenes to interpenetrate or collapse under gravity. We present TelePhysica, a closed-loop framework that reconstructs an executable, multi-regime physical world from a single photograph to drive stateful 3D interaction and trajectory-grounded video synthesis. A vision-language coding agent synthesizes separable visual meshes and declarative physical models within a two-stage verification loop that pairs pixel-aligned silhouette critique with a zero-force gravitational equilibrium preflight. The verified world compiles into a unified solver coupling rigid bodies, articulated chains, co-dimensional cloth shells, and 3D tetrahedral elastodynamics, whose simulated rollouts condition pretrained video diffusion models through pixel-aligned structural streams. On a 100-scene single-image benchmark, closed-loop execution raises equilibrium-verified world recovery from $0\%$ in single-pass synthesis to $81\%$. Controlled constitutive ablations show over an order-of-magnitude larger localized corner displacement and out-of-plane pleating in finite-element cloth than in matched rigid-plate proxies. Conditioning video diffusion on simulated trajectories increases dynamic motion-mask overlap by $2{\times}\text{--}5{\times}$ and preserves motion direction under counterfactual force reversals where direct image-to-video models fail.

    agentagenticbenchmark
  84. arxiv:2611.18074 · cs.LG
    Learning nonlinear interactions and cross-feature couplings with product-unit residual networks
    Ziyuan Li, Uwe Jaekel, Babette Dellen

    Understanding nonlinear feature interactions is crucial in science and engineering, yet standard multilayer perceptrons (MLPs) often capture such interactions only implicitly, leading to entangled representations that can impair robustness and interpretability. We investigate complex-valued product-unit residual networks (CV-PURe) that integrate multiplicative product units with residual connections to explicitly model cross-feature couplings while stabilizing optimization. We conduct a systematic evaluation on an interaction-driven synthetic benchmark and real-world datasets and evaluate prediction together with input-level interaction recovery under a reproducible protocol. Across five data splits and five training seeds, CV-PURe achieves the lowest mean RMSE on four of six scientific benchmarks, reducing Friedman error by 53.9% relative to MLP. Its Friedman interaction maps assign 60.7% of total strength to the known interacting pair, compared with 31.0% for MLP. On a synthetic task containing third- and fourth-order products and 20 distractors, CV-PURe reduces RMSE from 0.1376 to 0.0797 relative to a complex-valued MLP and recovers all nine true pairwise projections in every training/sampling repetition. Bike Sharing experiments connect improved prediction to the contribution of temporal, calendar, and seasonal context through paired ablations of both PURe variants. Parkinsons Telemonitoring extends the context analysis to repeated measurements. Noise and learning-curve evaluations demonstrate sustained predictive gains under perturbed inputs and limited supervision, while inference measurements quantify computational cost. This article extends our ICCS 2026 paper, Modeling Nonlinear Feature Interactions with Product-Unit Residual Networks (doi:10.1007/978-3-032-29921-5_6).

    benchmark
  85. arxiv:2611.18067 · cs.CL
    Agent-as-a-Judge
    Runyang You, Hongru Cai, Caiqi Zhang, Qiancheng Xu +4

    Artificial intelligence has progressed from task-specific systems toward general models spanning perception, generation, reasoning, and autonomous interaction, making reliable evaluation essential for comparison, deployment, and post-training feedback. At scale, modern evaluation pipelines increasingly rely on LLM-as-a-Judge to produce scores, preferences, and critiques; in its fixed-context setting, the judge relies on the supplied evaluand and evaluation materials. Yet a reliable verdict may require evidence beyond the supplied evaluand and context---either because critical environmental states remain unrecorded in the static execution log, or because controversial factual claims require external verification. Passive judges cannot obtain that missing evidence before deciding, especially when evaluating agent trajectories, executable artifacts, or domain-specific reports. Recent work therefore develops active evaluators that seek missing evidence and update their grounds for judgment before deciding. We bring these advances together through Agent-as-a-Judge, focusing on how evaluators construct and control checking processes within the broader LLM-based judging landscape. We define the scope through evaluation-directed checks and the use of their results, covering both structured workflows and observation-driven control. We organize the design space by multi-agent collaboration, tool integration, memory, and planning, and show how these mechanisms supply the checks and evidence on which a verdict rests. We consolidate published scores and reproduce missing entries for key agentic judges; the comparison reveals uneven agentic gains and motivates dedicated agentic evaluation. We conclude by characterizing application demands and open challenges in cost, reliability, safety, and governance that condition trusted deployment of agentic judges at scale.

    agentmulti-agentagenticpost-trainingevaluator
  86. arxiv:2611.18064 · cs.AI
    A$^3$: Adaptive Audio Agent Framework for Collaborative Edge Computing
    Zexin Li, Xueyi Zhang, Tao Ren, Yiming Chen

    Embedded audio agents need a nearby edge server for speed, but a server-dependent agent returns nothing when the link fails. We present A$^3$, an adaptive audio agent framework that places whole pipeline stages on the server, keeps a complete fallback agent on devices that can host one, and hedges both paths with a link-aware deadline that decides when to give up on the server. On Jetson devices and an RTX A4500 server, A$^3$ matches the server-only agent on a healthy link, answers every request in a full outage on the AGX, and beats a percentile hedge in simulated outages.

    agentagent framework
  87. arxiv:2611.18053 · cs.CV
    InfiHall-Bench: A Benchmark for Hallucination Evaluation in Autoregressive Text-to-Image Generation
    Yuxia Qiao, Xinhao Zhong, Junhao Li, Yi Sun +3

    Hallucination benchmarks for text-to-image generation are predominantly designed around diffusion models, limiting their diagnostic value for visual autoregressive (VAR) models and the development of mitigation methods. We identify two limitations in existing prompt construction paradigms: many prompts are no longer challenging for modern generators, while category-based construction relies on recurring object-attribute-relation templates and rarely tests interactions among multiple visual requirements within a single image. We introduce InfiHall-Bench, a controlled benchmark of 411 prompts constructed by systematically composing five foundational and visually verifiable skills: attribute binding, spatial relation, action relation, part consistency, and scene consistency. Each prompt covers 4.05 annotated skills on average, enabling the evaluation of hallucinations under diverse combinations of visual requirements. Using this benchmark, we find that hallucinations in VAR models exhibit a scale-wise persistence pattern: semantic errors can emerge at different autoregressive scales and are rarely corrected by subsequent fine-scale steps. Motivated by this observation, we propose Confidence-Masked Hallucination Mitigation (CMHM), which estimates the prediction confidence of target visual tokens at each autoregressive scale during teacher-forced fine-tuning. CMHM downweights reliably predicted tokens, allowing optimization to focus on uncertain predictions across scales. Experiments show that InfiHall-Bench distinguishes models that existing compositional benchmarks rate as comparable, while CMHM consistently outperforms uniform LoRA fine-tuning for VAR-based text-to-image generation.

    benchmark
  88. arxiv:2611.18050 · cs.CL
    FEND: Far-End Detection in Outbound Voice AI Calls
    Abhishek Kumar, Suraj Agrawal, Aashraya Sachdeva

    When a conversational Voice AI agent places an outbound call, the far end may be a live person, a voicemail greeting, an interactive voice response (IVR) system, or increasingly another AI system answering on the callee's behalf. Standard Answering Machine Detection (AMD) reports only whether a human or a machine picked up, a design meant for human call centres, where one machine is as uninteresting as another. That single bit does not tell a Voice AI agent what to do, since leaving a voicemail message, navigating an IVR and holding a short exchange with another AI are three different actions and a two-valued label cannot distinguish them. In 1,000 randomly sampled calls from our own outbound deployment, a live person answered 48.1%, a voicemail system 44.6%, an AI 6.9% and an IVR 0.4%. Our proposed framework, FEND, reads the far end's audio and its streaming transcript and runs one detector for each automated class (voicemail, IVR, AI callee), with human the default when nothing fires, returning a single label within 20 seconds. On a public benchmark the voicemail components reach 100% accuracy without ever misclassifying a live human, where Asterisk's AMD misclassifies 30 to 66% of them. For the AI callee, acoustic features do not transfer across recording conditions, but speech timing does, separating AI from human at 0.952 ROC-AUC, because synthesised speech leaves far shorter gaps between words. On production traffic, separating a live person from any automated far end reaches F1 0.979, and only 2.9% of live people are given an automated label, the error that matters most for an agent acting on the result.

    agentai agentbenchmark
  89. arxiv:2611.18032 · physics.optics
    Symmerty-Engineered Moire Metasurface for Multibeam Control
    Yu Chen, Lan Ke, Shaojie Ma, Shuo Liu

    Moiré metasurfaces offer a versatile platform for wavefront engineering through the relative rotation of periodic structures. Despite extensive demonstrations of continuous beam steering, the physical mechanisms governing beam multiplicity remain poorly understood, hindering the systematic design of multibeam radiation states. Here, we establish a predictive framework for engineering beam multiplicity in reflective moiré metasurfaces based on the dominant Fourier coefficients of their constituent patterns. We reveal that the rotational symmetry of these Fourier components in reciprocal space fully determines the symmetry of the resulting moiré unit cell and, consequently, its radiation characteristics. By further incorporating their phase information, both beam multiplicity and steering trajectories can be accurately predicted. We demonstrate this approach using reflective moiré metasurfaces constructed from patterns with p6m, p3m1, and p4m plane-group symmetries, achieving one-, two-, three-, four-, and six-beam radiation states with continuously tunable steering trajectories. Beyond symmetry-based design, we identify two additional degrees of freedom for beam multiplicity control. Selecting different rotation centers reshapes the phase distributions of the dominant Fourier coefficients, enabling further manipulation of beam multiplicity and spatial arrangement, while varying the relative twist angle alters the pairing of Fourier components, inducing transitions between distinct beam multiplicities. These findings establish a systematic framework for versatile and reconfigurable multibeam wavefront engineering.

    manipulation
  90. arxiv:2611.18030 · cs.CL
    PerfArena: Ranking LLMs by Code Performance
    Rui Rua, Hailemariam Mersha, Joao Paulo Fernandes, Karim Ali

    Industry now relies on LLMs to write a large share of production code. Although that code is mostly, but not always, correct, correctness alone does not determine whether it is fit for repeated and extensive use, which depends on what the code costs in time, memory and energy. Most code-generation benchmarks nonetheless evaluate correctness in isolation, and most performance-oriented benchmarks cover few languages and rarely measure energy consumption. To address this gap, we present PERFARENA, an open-source, reproducible arena that ranks LLMs by the performance of their generated code across multiple languages and competitions. PERFARENA executes the solutions of each LLM on a Unix-like host and reports correctness together with latency, peak memory, and energy consumption. Correctness oracles, performance oracles, profilers, and code-generation workflows are pluggable in PERFARENA, enabling the integration of new judging backends and measurement tools without changing the pipeline. We evaluate PERFARENA on two competitions: CLBGComp, 10 compute-intensive benchmarks from the Computer Language Benchmarks Game, and LeetCodeComp, 99 LeetCode problems whose accepted solutions differ measurably in latency. Currently, PERFARENA holds 9,455 correct and measured solutions from 13 state-of-the-art LLMs across 7 programming languages. Across both competitions, got-6-astra is the only LLM that no other LLM beats significantly in latency, energy consumption, or peak memory, solving 99.7% of LeetCodeComp and 100% of CLBGComp. All PERFARENA leaderboards and measured solutions are publicly available, providing a continuously updated reference that helps developers choose an LLM per language and metric, instead of relying only on correctness rates or pricing.

    memorybenchmarkleaderboardarena
  91. arxiv:2611.18026 · cs.RO
    A Generalizable Approach to Safety-Aware Robotic Behavior Planning with an Explicit Constraint Space
    Xianxu Wu, Shang-Ching Liu, Zhaocong Liang, Hengxiang Chen +4

    Large language and vision-language models have substantially advanced robotic behavior planning from natural-language instructions and visual observations. However, safe manipulation in tabletop environments remains challenging because the safety constraints required for a task depend on the surrounding scene and may not be explicitly specified in the instruction. We propose a safety-aware behavior planning framework based on an explicit Constraint Space that stores reusable textual safety rules. Rather than retrieving rules solely by semantic similarity, we formulate constraint selection as task- and environment-conditioned applicability learning. A lightweight query-conditioned dual-tower model identifies the constraints applicable to the current context, which are then provided together with the original task and image to a multimodal behavior planner. To handle incorrect constraint selections and previously unseen hazards, we introduce a multimodal Safety Inspector that evaluates generated primitive plans, reconstructs the complete applicable constraint set, and provides corrective planning feedback. Validated inspection feedback is further used for test-time adaptation of constraint applicability and parameter-efficient adaptation of the Safety Inspector through LoRA, while newly identified constraints are incorporated into the Constraint Space. The resulting framework integrates explicit safety knowledge representation,context-aware constraint selection, corrective safety inspection, and test-time knowledge adaptation for safety-aware robotic behavior planning.

    manipulation
  92. arxiv:2611.18008 · cs.LG
    Sample-Efficient Control of Magnetoconvective Flows via Differentiable Simulation
    Jannis Becktepe, Dmitry Krasnov, Nils Thuerey, Jörg Schumacher +1

    Magnetohydrodynamic (MHD) fluid flows are central to demanding engineering applications, most prominently fusion blanket cooling, where flow control is a promising tool to optimize key properties such as heat transfer. While active flow control (AFC) has recently advanced through machine learning, its application to MHD flows remains largely unexplored. In this work, we present $φ$-PICT, a novel differentiable GPU-accelerated MHD solver targeted at AFC. We validate $φ$-PICT against MHD benchmarks and demonstrate its high computational efficiency. Integrated into FluidGym, $φ$-PICT enables sample-efficient gradient-based algorithms to improve heat transfer by $>25$% and allows optimizers to achieve $10^3\times$ higher accuracy in inverse design than black-box methods. $φ$-PICT is publicly available at https://github.com/upb-sciml/phiPICT.

    benchmark
  93. arxiv:2611.18006 · cs.CV
    RecurIR: Learning When and How to Revise Restoration Policies via Recursive Self-Improvement
    Hu Gao, Haoran Zhang, Yulong Chen, Lizhuang Ma

    Agent-based image restoration has achieved promising performance by coordinating restoration tools to handle heterogeneous and coupled degradations. However, existing agents often improve their decisions through accumulated successful and failed executions, without explicitly identifying the conditions under which learned restoration strategies remain effective. Consequently, changes in degradation severity, spatial coverage, or interactions may invalidate previous decisions, while providing limited guidance on how these decisions should be revised. In this paper, we propose \textbf{RecurIR}, a recursive self-improvement framework that enables restoration agents to learn when and how to revise their decision rules. Specifically, we introduce a policy-hypothesis discovery mechanism to identify testable, condition-dependent rules through matched tool executions. We then design a rule-falsification-driven task evolution mechanism to generate targeted degradation variations, examine the validity of learned rules, and identify their applicability boundaries. Based on the resulting evidence, we develop a change-conditioned policy revision mechanism that learns how variations in degradation conditions should modify tool selection, execution order, operation strength, and processing regions. We further incorporate independent verification to retain effective revisions and use the resulting feedback to guide subsequent hypothesis discovery and task evolution. Through this recursive process, RecurIR progressively refines its restoration policies as degradation conditions change. Extensive experiments demonstrate the effectiveness of the proposed framework.

    self-improvement
  94. arxiv:2611.18003 · cs.RO
    Reusing Locomotion Training Recipes Across Humanoids via Body-Aligned Virtual Joints
    Sol Choi, Mingon Kim, Seokin Seo, Jongeun Choi +2

    Reinforcement learning has enabled robust humanoid locomotion, yet training recipes (rewards, architectures, hyperparameters) remain hard to reuse across robots with different native joint coordinates, since policies output native joint angles while action scales, bounds, gains, and penalties are defined in those robot-specific coordinates. Existing transmission models convert native targets into actuator commands but do not standardize the coordinate interface exposed to the policy. We introduce a body-aligned virtual joint space upstream of this mapping: policy actions and observations follow a common anatomical convention, and a deterministic kinematic layer maps virtual targets to native ones, leaving the existing transmission and low-level control stack unchanged. Under a single unmodified recipe on three humanoids with different leg-axis layouts, it is the only interface that tracks accurately, keeps the feet aligned with the travel direction, and trains stably on all three, and on Unitree G1 hardware it reduces heading drift by nearly a factor of six.

    humanoid
  95. arxiv:2611.17991 · cs.LG
    LACCO: Local Accumulation with Computation--Communication Overlap
    Daniele Kopyshevskiy, Benjamin Th{é}rien, Nicolas Pinto, Edouard Oyallon

    Training language models across low-bandwidth networks requires reducing synchronization overhead. Local optimization methods such as DiLoCo communicate only after several local steps, but outer gradient averaging and the outer update remain on the critical path. Existing computation-communication overlap methods avoid this stall through delayed or replica-local updates, which can degrade convergence or cause replicas to drift. We propose LACCO (Local Accumulation with Computation-Communication Overlap), a two-stage algorithm that overlaps global communication with local computation while applying globally aggregated updates shared by every replica, without introducing staleness in the outer update. We establish convergence guarantees for smooth non-convex objectives. Experiments on language-model pre-training with FineWeb show that LACCO consistently outperforms communication-overlapped baselines such as Eager and Delayed DiLoCo. These gains remain robust as model size scales up to 0.5B parameters and across a range of communication intervals and numbers of replicas up to 16 workers, while incurring only negligible-or fully overlappable-computational or memory overhead, which demonstrates the strength of LACCO.

    memory
  96. arxiv:2611.17981 · cs.LG
    When Does Oversampling Help? A Decision Framework for Imbalanced Data Streams Under Label Delay and Concept Drift
    Ahmed Awad, Anirudh Sengar

    For imbalanced data streams with delayed labels, is it worth adding a streaming oversampler, or does continuous re-evaluation of earlier predictions already recover the cost of the delay? We address this through an extensive experimental comparison of streaming oversampling and continuous re-evaluation under joint class imbalance, concept drift, and label delay. Streaming oversamplers are typically evaluated under the test-then-train assumption of immediate labels; their behaviour under these conditions, assessed through re-evaluation, has not been systematically studied. We close this gap with a decision framework that predicts when oversampling adds value over re-evaluation, organised around three axes: drift density, minority-class learnability, and delay relative to the refinement window, and operationalised as a competition between a recovery channel (re-evaluation) and a minority-coverage channel (oversampling). Evaluating $15$ classifiers on $6$ synthetic and $6$ real streams, we find: re-evaluation recovers delay-induced loss on drift-active streams but is silent on stationary ones; oversampling's marginal contribution over re-evaluation is statistically detectable in only one real-data configuration; and cost--performance Pareto recommendations from synthetic streams do not transfer to real data, where fronts partition by drift regime. Method selection should therefore be conditioned on the stream's profile and re-estimated as it evolves, not inherited from synthetic benchmarks.

    benchmark
  97. arxiv:2611.17980 · cs.LG
    Algorithmic Opacity as a Principal-Agent Problem
    Ignacio Cofone, Katherine J. Strandburg

    Concerns about gaming and trade secrecy, frequently invoked to justify algorithmic opacity, can also mask self-serving strategic behavior by decision-makers, who may prioritize private benefits over social welfare. We develop a principal-agent test for when algorithmic decision-making should trigger disclosure obligations even when concerns about gaming or trade secrecy exist. The test, which is anchored in decision quality, relies on two diagnostics that can be disclosed without enabling gaming or revealing trade secrets: the error profile (false positive and false negative rates, overall and for each group) and the distribution of decision outcomes across groups. These diagnostics make visible whether decision-makers align design and implementation with the public interest or instead use opacity to obscure bias, inefficiency, or strategic shirking. Strong disparities across social groups, and asymmetries in the distribution of decision errors across outcome types, should trigger rebuttable presumptions favoring further disclosure. Where ground-truth error rates are unavailable, outcome distributions across groups can support the same presumption, and a decision-maker's inability to produce an error profile is itself diagnostic. The test thus treats fairness metrics as triggers for further disclosure and extends to generative and agentic systems through analogous metrics. The framework informs differentiated disclosure regimes (staged, partial, mediated, or audience-segmented). Once such regimes are considered, the domain of justifiable algorithmic opacity is significantly narrower than commonly assumed: when gaming or trade secrecy risks are legitimate, actionable disclosures (such as training data characteristics and key system design parameters) ordinarily can be mandated without compromising innovation or enabling gaming, even for technically inscrutable algorithms.

    agentic
  98. arxiv:2611.17966 · cs.CV
    From Images to Grounded Edits: A Real-World Benchmark for Chinese Grammatical Error Correction in Handwritten Essays
    Weiru Zhang, Yuzhong Hong, Junwei Bao, Lemao Liu +1

    Chinese Grammatical Error Correction (CGEC) is typically formulated as a text-to-text problem. While suitable for digital texts, this formulation falls short in real-world K-12 education, where AI-assisted corrections are expected directly on essay images, which are primarily handwritten and exhibit many irregularities. To bridge this gap, we introduce Grounded Chinese Grammatical Error Correction (G-CGEC), an image-to-grounded-edit task requiring systems to predict both corrections and their positions on essay images. We propose G-CGEC-Eval, a benchmark comprising real-world images of K-12 Chinese essays with fine-grained annotations. For evaluation, we adapt IoU-based metrics from object detection and propose a more tolerant alternative based on center-distance matching. Using this benchmark, we evaluate open-source and proprietary vision-language models under a direct single-pass setting, as well as settings with auxiliary components (OCR and character grounding) coordinated via pipelining or agentic harnessing. Results show that even top-tier proprietary models perform surprisingly poorly in the single-pass setting, while pipelining and agentic harnessing yield meaningful improvements. Notably, an error-type-based multi-agent workflow achieves the best performance, establishing it as the strongest baseline for future G-CGEC research. Through ablations and fine-grained analysis, we find the task remains far from solved, with linguistic error discovery and character-level grounding as the primary bottlenecks. This underscores the need for models that better integrate visual perception with linguistic reasoning and visual grounding.

    multi-agentagenticbenchmark
  99. arxiv:2611.17963 · cs.CL
    Toward Universal Translatability: On the Statistical Ontology of Generative AI
    Katia Schwerzmann

    This paper argues that the problem of translation functions both as a point of origin and as a normative horizon for contemporary generative AI. In current discourse around ''world models,'' their attainment is framed as the achievement of internal consistency, cross-modal alignment, and physical grounding. Taken together, these demands converge on an ideal of the model as a universal translation machine capable of establishing lossless correspondences across linguistic, visual, and physical representation spaces. I show that this ideal presupposes a metaphysical commitment that I call a statistical ontology: the belief that reality has an underlying, computable statistical structure that can be recovered through sufficient data and modeling, rather than being shaped and accessed through situated human interpretation and labor. Tracing this commitment back to Warren Weaver's 1949 Memorandum on machine translation, I outline how translation was historically recast as code deciphering and, later, as the alignment of probability distributions across structured outputs. Finally, I contrast this paradigm with approaches in Translation Studies that foreground opacity, power asymmetries, and the translator's positionality. I argue that generative AI's statistical ontology is pre-Kantian, pre-critical and obscures the infrastructures and human labor that make machine translatability---as alignment---possible in the first place.

    world model
  100. arxiv:2611.17957 · cs.CV
    MacroGS: Macro-Pixel Shared Gaussian Primitives for Arbitrary-Scale Light Field Super-Resolutio
    Zeqiang Wei, Angulia Yang, Min Xu

    Arbitrary-scale light field spatial super-resolution reconstructs high-resolution views at any requested scale from low-resolution multi-view observations. Gaussian primitives enable continuous reconstruction, but generating separate primitive sets for each view redundantly encodes shared scene structure across views and causes the primitive count grow linearly with the number of views. To avoid such redundancy, we propose MacroGS, a feed-forward decoder that jointly predicts a single shared Gaussian set per macro pixel rather than generating separate sets for each view. Disparity-dependent replacement and angle-conditioned appearance allow each set to be reused across all output views, keeping the predicted primitive count independent of the number of views at each requested scale. The decoder iteratively predicts per-primitive disparities, retrieves the corresponding multi-view features, and updates the primitive states. Gaussian splatting and residual refinement then reconstruct the output light field. Despite being learned without geometric supervision, the predicted disparities strongly correlate with underlying scene geometry. Experiments across five benchmarks demonstrate that MacroGS integrates effectively with diverse light field encoders, improving reconstruction quality at fixed scales while achieving leading performance across most evaluated arbitrary-scale settings. The decoder also achieves speedups of approximately $2$-$10\times$ over state-of-the-art implicit light field decoders and $10$-$18\times$ over the view-wise Gaussian baseline.

    benchmark
  101. arxiv:2611.17955 · cs.LG
    WebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents
    Imene Kerboua, Fatemeh Pesaran Zadeh, Xing Han Lù, Weijian Qi +18

    Web agents powered by large language and vision-language models are increasingly applied to realistic browser work that spans heterogeneous applications, multimodal content, and stateful workflows. However, existing reproducible web-agent benchmarks cover only a small number of web applications drawn from a few software categories, and restrict modality to text and vision. Live benchmarks broaden site coverage but sacrifice reproducibility, since pages and data drift between runs. Moreover, existing benchmarks do not meaningfully evaluate whether agents can understand and use audio and video content embedded within web tasks. To address these gaps, we introduce WebArena-Pro, a benchmark comprising 300 tasks across 20 self-hosted web applications in six domain categories, spanning distinct interface conventions, workflows, and data models. Across the evaluated agents, the best performance is achieved by both Codex with GPT 6.1 Sol and Claude Code with Claude Opus 5.5 at 62.3%, followed by OpenCode with Kimi 3.1 Pro at 60.7%. The strongest GenericAgent configurations, Gemini 3.1 Pro omni and Gemma4 31B, both achieve 30.3%. Among reproducible, human-curated web agent benchmarks, WebArena-Pro provides the broadest application coverage and the most comprehensive multimodal support to date. The benchmark treats audio and video as core observations alongside text and vision, with dedicated actions for extracting information from each. WebArena-Pro runs each task in isolation and supports reproducible, parallel evaluation. Tasks are authored through a dedicated annotator interface, filtered by LLM-assisted triage, and finally validated by humans before release. Project page: https://webarena-pro.github.io

    agentagent benchmarkbenchmark
  102. arxiv:2611.17939 · cs.AI
    Can You Trust Your Finance Agent? Drift, Diagnosis and Auditable Intervention on One Shipped Financial Research Agent
    Cosimo Galeone, Anna Ettorre, Ginevra Mariani, Minsu Park +2

    Can you trust a finance agent you did not build? We take the reference agent shipped with the Vals Finance Agent Benchmark v2, leave it untuned, and ask four questions a leaderboard score does not answer. Is the score stable? Seven identical runs on 18 questions pass between 105 and 123 of 149 rubric criteria, and the bill varies by up to 17%. When the agent fails, who owns the fix? We classify 20 failed trajectories, from three runs of nine held-out questions, by the layer whose owner would remedy them, under a taxonomy frozen in advance: 13 belong to the agent's instructions, 3 to its tools, 4 to the model, none to its data. Can that diagnosis be trusted? Four blind readers from three model families, given only the traces, name the same fault as our LLM analyzer in 94% of cases, the rate at which they agree with one another (93%). Can a change be trusted? An optimization loop reads the traces, makes one targeted change per round and keeps it only if it clears a fixed threshold. Under a cost objective its final configuration costs 0.34x the baseline on the held-out questions (95% CI [0.22, 0.58]) at accuracy within noise. Under an accuracy objective it gains 14 points on the development questions and 1.5 on the held-out ones (CI [-5.3, +7.2]). Judged against the measured run-to-run noise, under a rule written before the test was opened, the first is a real saving and the second is not. This is a case study of one agent and nine test questions, not a general result.

    agentagent benchmarkbenchmarkleaderboard
  103. arxiv:2611.17933 · cs.AI
    OutfitterBench: Evaluating LLM Agents on Long-Horizon Resource Allocation and Commitment Coordination
    Jie Huang, Yuecheng Hong, Yao Chen, Shengding Jiao +2

    In many long-horizon settings, an agent's own choices create future constraints: committing finite capacity establishes future obligations and forecloses later opportunities. Sustaining performance therefore requires both keeping evolving plans feasible and comparing the value of resource allocations with the alternatives they displace. Recent benchmarks increasingly evaluate LLM agents in continuing operations, but give less attention to how agents keep their own commitments feasible while reassessing resource allocations as opportunities change. We introduce OutfitterBench, where agents manage an outdoor-equipment rental business for 200 days, purchasing from a 46-product catalog and handling roughly 1,100 reservation requests. Under a fixed capital ceiling, reservations and uncertain walk-in demand compete for reusable equipment whose availability changes over time. Experiments with seven frontier LLMs show that several agents underperform a simple rule-based baseline, while only the leading agent consistently surpasses it. Trajectory analysis shows that some agents apply known constraints inconsistently and make new decisions that conflict with existing commitments. Agents with more reliable fulfillment can still secure low returns from comparable commitments or delay resource reallocation. The leading agent better connects preparation with economic choices but also misses opportunities for more profitable asset use. These findings highlight limitations in sustaining coherent resource plans and reassessing their value over time. The benchmark code is publicly available at https://github.com/adminblackjack/OutfitterBench.

    agentllm agentbenchmark
  104. arxiv:2611.17931 · cs.AI
    MineralExp: Benchmarking LLM Agents for Experimental Design in Real-World Mineral Exploration
    Bing Wang, Mingyang Wang, Yuanhang Mi, Yue Xu +7

    In mineral exploration, verifying geological hypotheses relies on designing a variety of geochemical experiments. While large language model (LLM) agents increasingly automate downstream geoscientific data analysis, a crucial upstream capability remains largely unexplored: automated geochemical experimental design, i.e., deciding which experiments to conduct before committing to costly laboratory work. Evaluating this capability of LLM agents is difficult for two reasons. First, geochemical experiments exhibit high methodological diversity and subjectivity: to test the same hypothesis, different geologists may choose different minerals and instruments, so no single protocol can serve as the reference answer. Second, open-ended experimental plans resist representation as structured labels, which makes objective scoring difficult. To address these challenges, we introduce MineralExp, a benchmark that reformulates geochemical experimental design as a hypothesis-based experiment selection problem, i.e., a multi-label classification problem. We abstract concrete instrument names and procedural details into a shared taxonomy of 19 experimental types across five mineral-system processes: Source, Transport, Deposition, Modification, and Preservation. This taxonomy yields deterministic multi-label targets while accommodating diverse experimental protocols. Constructed from published ore-deposit literature and verified by domain experts, MineralExp comprises 277 cases and 1,572 reference experimental plans. Given a geological hypothesis and initial context, an LLM agent selects the relevant tasks and generates concrete experimental plans, which are scored by multi-label classification metrics and an LLM-as-a-Judge metric. The experimental results show that current agents generally achieve high recall but low precision, tending to propose more experiments than are necessary to test the hypothesis.

    agentllm agentbenchmark
  105. arxiv:2611.17929 · cs.AI
    Evaluating Large Language Models for Scientific Paper Search: Retrieval Success and Inference Latency on Mac Studio and NVIDIA A100
    Frederic Andriantseheno Rakoto, Yannis Bendi-Ouis, Xavier Hinaut

    Scientific literature search can benefit from locally deployed LLM assistants, which reduce reliance on hosted model providers while giving research teams greater control over their data and infrastructure. This work investigates whether open-weight models can support web-assisted paper retrieval under practical computing constraints. We first characterize inference latency as the input context grows from 512 to 128k tokens on a Mac Studio and a server equipped with two NVIDIA A100 GPUs. We then evaluate eight models on the retrieval of 70 papers from seven scientific repositories, using queries that represent three levels of domain knowledge. The results show that retrieval success depends strongly on the selected model and the specificity of the query. And that A100 GPUs are generally faster than the Mac Studio, but its unified memory allows larger models that could not be loaded in the tested GPU configuration.

    memory
  106. arxiv:2611.17924 · cs.LG
    TIDE: Reclaiming Lost Performance In On-Policy Reinforcement Learning
    Mathew Jacob, Morgan Borjigin-Wang, Kan Zhu, Po Hao Chen +9

    Synchronous, colocated RL post-training of large language models (LLMs) today suffers from severe under-utilization due to long-tail rollouts. Inference engines fall idle waiting for long-tail rollouts or decode at low, inefficient batch sizes. To address this, we present TIDE, an RL training system that preserves on-policy training semantics while recovering resource utilization. TIDE does so by two complementary techniques: (1) dynamically repurposing rollout GPUs for training while long-tail rollouts complete and (2) dynamically migrating rollouts between inference engines to boost decoding throughput. TIDE automatically determines the right time to repurpose rollout resources through an adaptive migration policy, obtaining performance improvements that generalize across multiple applications and hardware configurations without requiring bespoke configuration. In our evaluation, TIDE provides end-to-end speedups of up to 1.36x over colocated systems, outperforming all evaluated on-policy baselines and achieving up to 97% of optimal throughput.

    post-training
  107. arxiv:2611.17921 · cs.AI
    RecMem: Recurrent Memory Compression for Long-Sequence Generative Recommendation
    Haoyi Hu, Gaoyang Guo, Yunjia Xi, Wen Chen +4

    Generative recommendation benefits from long interaction histories, but the cost of processing them directly grows with history length. We propose RecMem, a recurrent memory compression framework guided by three desiderata: space invariance, semantic fidelity, and lifelong evolvability. RecMem compresses historical segments into a fixed set of memory slots using a bifurcated encoder-decoder with shared lower layers and slot-wise forget-gated updates. Recommendation supervision trains the memory to preserve predictive information, while stream-consistency training aligns predictions from sequential and packed processing. By construction, the memory state and decoder context are independent of history length; under contraction and approximate semigroup consistency, we further derive a horizon-independent bound on the state discrepancy between these two processing orders. On MerRec, RecMem uses 16 memory slots and a 216-token decoder context, compared with 1,200 tokens for HSTU-1200, while retaining 87.9% of HSTU-1200's Recall@10 and 97.0% of its Recall@200. It outperforms PHSTU and the evaluated recent-only and retrieval baselines, and analyses of distant history show that archived interactions contribute predictive signal.

    memory
  108. arxiv:2611.17917 · cs.AI
    When Cooperation Must Change: Semantic Strategy Unlearning in Multi-Agent Reinforcement Learning
    Wenhan Wu, Linghan Chen, Yudong Gao, Fan Yang +4

    Cooperative multi-agent reinforcement learning (MARL) supports coordinated decision-making in distributed Web services, multi-robot control, and collaborative games through learned coordination conventions. When deployment requirements change, a previously useful convention may become obsolete or unsafe. For example, revised cache-sharing permissions in Web content delivery may require caches to abandon a learned division of storage and serving roles while maintaining efficient delivery. Teams must therefore remove the specified convention while preserving useful cooperation. This is challenging because the same relational pattern can be reproduced by different agents across different histories. Runtime shielding leaves the underlying coordination preference intact and can impair task progress, while existing reinforcement unlearning and trajectory-deletion methods do not directly target role-invariant conventions. This paper formalizes this problem as semantic coordination-strategy unlearning and proposes CARVE.CARVE uses causal coordination attribution to identify influential interactions and retain-aware subspace updates to limit interference with unrelated behaviors. Counterfactual re-coordination helps agents learn productive alternatives, while adversarial auditing tests whether the target convention can still be recovered. Accepted updates are merged into decentralized actors for direct deployment. Evaluation across five cooperative MARL benchmarks shows that CARVE reduces mean nominal convention-completion and worst-case elicitation rates by 91.4\% and 83.3\%, respectively.The median utility loss is 5.9\%, and measured inference latency matches that of the original policies.

    multi-agentbenchmark
  109. arxiv:2611.17915 · cs.AI
    SWIFT: Sample-Wise Instance Fine-Tuning of Time-Series Foundation Models for Forecasting
    Yang Chen, Feifan Tian, Cong Wang, Cheng Feng +2

    While current few-shot adaptation strategies for time series forecasting foundation models are predominantly dataset-centric, the necessity of obtaining an additional dataset for the finetuning process poses a significant hurdle for adoption. To address this constraint, we propose Sample-Wise Instance Fine-Tuning (SWIFT), a novel forecasting adaptation framework that achieves fine-grained specialization by leveraging structural priors from the target instance's own history. This shift allows the fine-tuning process to be entirely transparent to the end-user, which significantly enhances the ease of adoption. To implement this framework, SWIFT synergizes Self-Retrieval for historical mining and Frequency-Aware Loss for spectral consistency, while employing Deviation-Aware Dynamic Fusion to balance adaptation with general knowledge preservation. Extensive experiments across four TSFM backbones and three long-history benchmark subsets validate SWIFT, yielding average MASE and WQL reductions of 6.41% and 4.83% across 24 combinations of four backbones and six Chronos Benchmark II datasets. To the best of our knowledge, SWIFT is the first framework to investigate sample-wise instance fine-tuning of time-series foundation models, specializing the model for an individual prediction window using only the corresponding forecasting sample's own observed history.

    benchmark
  110. arxiv:2611.17910 · cs.AI
    Adversarially Attacking Symbolic Vocabulary Vulnerabilities In LLM Planners
    Stephen Obadinma, Xiaodan Zhu

    As large language models (LLMs) are increasingly deployed as autonomous agents, their abilities in navigating automated planning tasks have become essential. However, despite their strengths, LLMs' planning ability is not robust, particularly when they are not allowed to exploit commonsense cues in the symbolic vocabulary (e.g., names of actions, predicates, and objects) used to define planning tasks. This makes LLMs unreliable compared to symbolic planners. With recent reasoning models showing improved planning performance on tasks with obfuscated symbolic vocabularies, we utilize an adversarial attack framework to reveal to what extent the crucial weaknesses of LLM agents at joint symbolic interpretation and planning remain and whether they can reason properly when presented with severe cases. As such, our main contribution is devising an LLM-based attack model, which we call Symbolic Swapper, to automatically generate obfuscated planning domains with non-standard symbolic vocabularies that systematically break strong reasoning models' ability to plan. In doing so, we reveal how much LLMs' planning abilities can be further degraded under more advanced obfuscation schemes, allowing us to ascertain their success when having to rely on their pure planning ability. We find that attacks can decrease the planning performance of base models and even robust planning frameworks by over 70%. We further analyze the factors behind how LLM planning abilities break down and find that successful obfuscation causes models to appear to significantly degrade in reasoning quality despite them showing some inherent ability to understand the domain, revealing a gap in the perception of a task domain and being able to plan under it.

    llm agentautonomous agent
  111. arxiv:2611.17899 · cs.CV
    More Genes, Faster Inference: Efficient Large-Panel Spatial Transcriptomics Prediction from Histology
    Songming Zhang, Yuxiao Luo, Shifu Luo, Siran Chen +2

    Spatial transcriptomics (ST) prediction from histology aims to infer spatially resolved gene expression profiles directly from tissue images. This is a challenging problem that is currently often addressed with complicated model architectures, additional inputs, and complicated training procedures, while remaining limited to fixed, small gene panels. This raises a natural question: Is there an efficient method for large-panel ST prediction? Here, we present GeneFlow, a simpler yet effective framework for large-scale spatial gene expression prediction that unifies flow-matching generative modeling with structured multimodal fusion. Specifically, GeneFlow first encodes both histological morphology and cross-gene interactions in spatial spots through histology and gene tokenizers. The model directly predicts spatial gene expression, rather than an intermediate velocity field, while being trained with a velocity-based flow-matching objective. We show that, even without relying on heavy generative pipelines, GeneFlow achieves state-of-the-art performance across multiple ST benchmarks, with up to 56.8% gains in 1,000-gene prediction. Extensive experiments further show that GeneFlow produces biologically meaningful spot profiles while reducing computational cost by orders of magnitude, thereby improving efficiency and scalability, particularly in clinical settings. We believe GeneFlow provides an opportunity for leveraging widely available histology images for downstream biological discovery. Code is available at: https://github.com/ZZhangsm/GeneFlow

    benchmark
  112. arxiv:2611.17889 · eess.SY
    PLL-Less Synchronization of Grid-Following Converters Using a Voltage-Driven Frenet Frame
    Jinhui Chen, Huadong Sun, Shiyun Xu, Bing Zhao +2

    In grid-following converters, the synchronization path can transmit voltage disturbances to the current loop under weak or unbalanced grid conditions. This paper develops a Frenet-frame controller that derives its axes from the integral curve of the measured voltage at the point of common coupling. The tangential current governs active power, while the normal and binormal currents govern the transverse components of non-active power. Feedforward terms derived from the output filter model compensate curvature and torsion coupling, yielding identical nominal current dynamics across the axes. The controller operates without a phase-locked loop or sequence extraction. A reduced transverse model accounts for the finite bandwidth of geometric rate reconstruction and predicts a reduction in low-frequency coupling from voltage disturbances to current. Controller hardware-in-the-loop tests using real-time simulation of a three-phase four-wire converter show the shortest settling time after an active power reference step and the smallest peak power deviations during balanced voltage disturbances among the tested controllers. Under single-phase voltage unbalance, the method reduces peak binormal current and maximum phase-current total harmonic distortion relative to both benchmarks. It also settles to the commanded power in a weak-grid connection test where the phase-locked loop benchmark exhibits sustained oscillations.

    benchmark
  113. arxiv:2611.17886 · cs.LG
    Quantum Graph Neural Networks: Architectures, Trainability, and Applications
    Zeinab Sharifi, Hossein Shirazi

    Quantum graph neural networks (QGNNs) apply parametrized quantum circuits to graph-structured data, with the aim of overcoming limits of classical graph neural networks in scalability, expressivity, and long-range information flow. Since the first Hamiltonian-based model in 2019, the field has produced many architectures, applications, and claims, but the evidence behind them has not been assessed systematically. This survey introduces a unified formalism that describes every QGNN as an encoder, a graph-conditioned circuit, observables, and an optional classical head, together with a circuit-level condition for permutation equivariance and a resource model for comparing costs. We organize the literature into five architecture families (convolutional, equivariant, message-passing and recurrent, Hamiltonian, and hybrid) and compare them by qubit cost, symmetry, scalability, and evidence. We then review symmetry and geometric inductive biases (permutation, rotation, and Lorentz), the results on expressivity, trainability, noise, and scalability, and applications in high-energy physics, chemistry, finance, healthcare, earth observation, and transportation. An audit of datasets and evaluation practices leads to a minimum reporting protocol. Our main finding is that the design space is rich and symmetry is the best-supported inductive bias, whereas the evidence remains thin: results are mostly simulated, on small graphs, at parity with classical models or with fewer parameters, and no advantage has yet been established on tasks over classical graph data. We identify five open challenges and a research agenda covering provable advantage, equivariant and geometric designs, fault-tolerant implementations, and shared benchmarks. The coverage extends through 2026 and emphasizes recent work.

    benchmark
  114. arxiv:2611.17883 · cs.AI
    Securing Multi-Agent Systems by Design
    Xilong Wang, Xiaoxing Chen, Neil Gong

    Multi-agent systems solve complex tasks by decomposing them into subtasks, delegating the subtasks to agents, and aggregating the agents' responses using an aggregation rule. Malicious agents can manipulate the final solution by providing malicious responses to their assigned subtasks. In this paper, we first propose response manipulation attacks (ReMA), which iteratively optimize malicious responses to maximize attack effectiveness. Specifically, we formulate untargeted and targeted attacks as optimization problems and develop an approach to approximately solve these problems by iteratively refining malicious responses based on feedback from a surrogate multi-agent system. To defend against such attacks, we further propose SecMAS, a secure aggregation rule that leverages the semantic similarities among agent responses to select the response that is most consistent with a sufficiently large subset of the other responses. We theoretically establish security guarantees for SecMAS against both untargeted and targeted attacks. Extensive evaluations demonstrate that existing aggregation rules are vulnerable to ReMA, whereas SecMAS effectively defends against all evaluated attacks, including adaptive variants of ReMA, while maintaining utility comparable to that of existing aggregation rules.

    manipulationagentmulti-agentagent system
  115. arxiv:2611.17880 · cs.AI
    Soteria: Formally Verified Planning with Runtime Enforcement for Safe LLM Agents
    Mike He, Ankush Desai, Sharad Malik, Aarti Gupta

    LLM agents that execute multi-step tool calls must satisfy two objectives simultaneously: completing the user's task (utility) and conforming to domain policies and correctness constraints (safety). These objectives are in tension -- blocking an unsafe action prevents a violation but can leave the agent stranded with no principled way to recover. Existing approaches sacrifice one objective for the other. We introduce Soteria, a framework that reconciles safety and utility through verified hierarchical planning and runtime enforcement of specifications. Before execution, the agent generates a structured plan that is formally verified against the specifications prior to any tool invocation. During execution, the verified plan serves dual roles: it ensures that the agent takes specification-conformant trajectories and provides guidance when unsafe actions are blocked. Across multiple benchmarks covering a diverse range of tool-use tasks, Soteria achieves 1.7 times the average task success rate of the best baselines, with no violations of policy specifications, demonstrating that safety and utility need not be traded off.

    agentllm agenttool-usebenchmark
  116. arxiv:2611.17877 · physics.optics
    Antihelical Edge States in Photonic Metacrystals
    Ke-Yi Zeng, Zi-Xuan Gao, Zhou Zhou, Xiao-Dong Chen +1

    Antihelical edge states represent a gapless form of boundary transport in which modes with opposite spin counterpropagate along a given boundary, while modes sharing the same spin propagate in the same direction on opposite boundaries. Here we show that such states arise naturally within a modified Kane-Mele framework realized in an electromagnetic-dual photonic metacrystal with staggered bianisotropy. Reversing the spin-dependent next-nearest-neighbor coupling on one sublattice reorganizes the bulk spectrum into a spin-polarized Dirac semimetal characterized by spin-dependent energy shifts of the Dirac cones rather than a bulk gap. This bulk reorganization generates tilted edge-mode dispersions in zigzag ribbons and gives rise to antihelical edge transport coexisting with bulk states. Full-wave simulations confirm the corresponding bulk and edge behaviors. Extending the analysis to finite geometries, simulations further reveal discrete corner-localized modes. A realistic design based on split-ring metaatoms is proposed to realize the required staggered bianisotropy and spin-dependent energy shifts of the Dirac cones. Furthermore, we show that the tensorial freedom of the bianisotropic constitutive relations enables continuous manipulation of the electromagnetic pseudospin basis while preserving the antihelical boundary transport. These results provide a practical route for engineering unconventional gapless boundary transport in photonic Dirac systems.

    manipulation
  117. arxiv:2611.17867 · cs.AI
    UltraBERT-JEV: Encoder models for System One Applications
    Astarag Mohapatra

    We study encoder models that pick one of several options in a single forward pass, without generating text. This technical report compares recipes for fine-tuning UltraBERT-380M with one mask per option. The four recipes differ in how an option is scored and how the scores become probabilities; one follows the expert gate of DeepSeek-V3. Trained on the same data as Qwen3-0.6B with the LLM-as-Jev recipe (67.6% test accuracy), the encoder reaches 69.7% to 72.1%. Scoring an option by "yes" minus "no" instead of "yes" alone gives the larger gain, from 69.7% to 71.3% on the test set and from 57.8% to 59.6% on external benchmarks (56.1% for the decoder). Sigmoid normalisation adds 0.8 test points; a square-root softplus adds none. One request at a time on an NVIDIA A100, the encoder takes 29 ms per decision against 68 ms, but uses more peak memory (2.8 against 2.2 GiB). An auxiliary head, trained on human-marked evidence and on labels selected with a reward adapted from SelfCite, predicts in the same pass the sentences annotated as support for the answer.

    memorybenchmark
  118. arxiv:2611.17864 · cs.LG
    Disentangling Forgetting from Utility: Lightweight LLM Unlearning via Orthogonal Null-Space Projection
    Xiuyuan Wang, Weiming Liu, Zongxin Yang, Fan Wang +6

    Large Language Models (LLMs) pose significant security risks by retaining private or hazardous information, creating an urgent need for unlearning techniques that excise specific knowledge without compromising general utility. However, prevailing fine-tuning-based unlearning methods struggle to achieve this balance, often suffering from computational inefficiency and unintended utility degradation due to the entangled nature of knowledge within model parameters. To disentangle forgetting from utility, we propose UUE, a theoretically grounded framework that reformulates unlearning as controllable model editing. First, UUE projects parameter updates into the orthogonal left null space of retained knowledge representations, strictly ensuring that the editing process causes minimal interference to the model's general capabilities. Second, distinct from targeted mapping, we introduce a convex untargeted objective that derives a closed-form analytical solution to effectively disrupt undesired knowledge by pushing forget-set embeddings toward their antipodal points. We implement UUE via lightweight, pluggable adapters (including a low-rank variant, UUE-L), enabling efficient deployment without modifying the frozen backbone. Extensive experiments on TOFU and WMDP benchmarks demonstrate that UUE and UUE-L achieve superior unlearning efficacy, significantly outperforming existing methods.

    benchmark
  119. arxiv:2611.17862 · cs.AI
    FlexCast: Flexible Context Learning for Time Series Foundation Models
    Zuyang Shen, Luqi Gong, Fanda Fan, Bingsong Wang +5

    Time series foundation models (TSFMs) based on autoregressive (AR) next-token prediction entangle two factors: variable-context conditioning and prediction-unit factorization with autoregressive rollout. Strong AR performance therefore does not reveal whether generalization stems from the flexible training distribution, the local prediction target, the left-to-right inference process, or their combination. We propose FlexCast, a flexible context-target learning framework that disentangles these factors. FlexCast reformulates forecasting as variable-length masked prediction over time series patches: during pretraining, it dynamically samples visible context and masked target patches, exposing the model to diverse context-target configurations without sequential decoding. At inference time, the same pretrained model supports one-step forecasting, confidence-based iterative refinement, and next-token-style generation through a unified mask-and-predict interface, enabling controlled comparison of inference schedules under identical architecture, data, and objective. Evaluated on BLAST and multiple zero-shot benchmarks, FlexCast achieves accuracy competitive with or better than strong TSFM baselines while substantially reducing inference latency. Ablations confirm that context diversity and contiguous target masking drive performance, while left-to-right decoding does not yield consistent gains once flexible context-target relationships have been learned, suggesting that the key benefit of AR-style TSFMs lies in flexible training rather than sequential inference. The code is available at https://github.com/BenchCouncil/FlexCast.

    iterative refinementbenchmark
  120. arxiv:2611.17853 · cs.CV
    DoubleTake: Focused Verification of Visual Evidence for Long-Document Understanding
    Kejun Xue, Ang Li, Jinghua Tang, Xincheng Wang +3

    In long-document question answering, a vision-language model can answer incorrectly even when the retrieved pages contain sufficient evidence. Rereading those pages with the original question gives the model little guidance about which fact to check. We introduce DoubleTake, a training-free framework for focused verification of visual evidence. An initial reading produces a candidate answer and a separate evidence record that guides retrieval for a second reading. A planner compares the two candidates and their supporting evidence to formulate a factual question. A verifier answers it from the source images without seeing either candidate or evidence record, and a reasoner uses this observation to resolve the original question. With a frozen Qwen3-VL-8B-Instruct backbone, DoubleTake achieves the highest overall accuracy among the evaluated methods on each of four benchmarks. With the same pages and number of model calls, asking a focused factual question yields higher accuracy on all four benchmarks than asking the original question again. Our code is available at https://github.com/small-code-cat/double_take.

    benchmark
  121. arxiv:2611.17850 · eess.SY
    OrbitWeaver: LLM Agent Planning with Simulation Feedback for Optically Interconnected Orbital GPU Clusters
    Pengcheng Luo, Bowen Zhang

    Space-based data-center designs increasingly consider GPU computing formations connected by inter-satellite optical links. With limited neighboring connections, jobs assigned disjoint GPUs can still contend on shared relay paths. Rank placement and communication priorities jointly affect this contention and the readiness of downstream computation. We present OrbitWeaver, an offline large language model (LLM) agent planner that jointly optimizes rank placement and communication order. Guided by communication structure, execution dependencies, and simulation feedback, the agent constructs and refines plans that coordinate shared-link access with computation progress. This coordination reduces contention and dependency blocking, shortening mean job completion time (JCT). Simulation experiments on optically interconnected orbital GPU clusters show that OrbitWeaver reduces aggregate mean JCT by 6.2-33.5% relative to the state-of-the-art baselines CASSINI, Crux, and Symphony. These results demonstrate the value of LLM-guided planning for efficient concurrent computing in orbital GPU clusters.

    agentllm agent
  122. arxiv:2611.17836 · cs.CL
    BLiMP-RO: A Benchmark of Linguistic Minimal Pairs for Romanian
    Radu-Stefan Popa, Valentin Feroiu, Adina Camelia Bleotu, Sergiu Nisioi

    Evaluating grammatical knowledge in language models requires benchmarks that cover diverse linguistic phenomena and are validated against human judgments. We introduce BLiMP-RO, a Romanian minimal-pair benchmark covering 50 grammatical phenomena across ten domains. It comprises 493 linguist-constructed sentence pairs, evaluated by native speakers in five experiments, and 5,942 additional semi-automatically generated, human-validated pairs. We further investigate grammatical generalization under limited linguistic input by training GPT-2 and GPT-BERT models from scratch on Romanian BabyBabelLM data and a newly collected corpus of 10 million words from child-targeted video captions. Increasing caption training data yields approximately 9 percentage points of accuracy improvement per tenfold increase in corpus size. At matched data sizes, BabyBabelLM generally outperforms video captions, while GPT-BERT consistently outperforms GPT-2. Our findings highlight how training data quantity, composition, and model architecture shape grammatical generalization in Romanian.

    benchmark
  123. arxiv:2611.17831 · cs.CV
    V-JEPA-LM: Video-Language Models That Read Dynamics Beyond Static, Text-Aligned Representations
    Keuntae Kim, Dongjun Kim, Yong Suk Choi

    World-model video encoders such as V-JEPA learn how scenes change (direction, speed, order, state) from video alone. To answer questions, they are joined to a language model trained on captions and question-answer pairs. But each caption or question names only a few attributes of a clip, so the model learns to retrieve those attributes rather than to read the clip, and the dynamics the encoder carries go unused: on the same items, a linear probe reads speed from frozen V-JEPA features at 87% while the VLM answers at chance. We show where dynamics lives in a frozen encoder: in a subspace shared across videos for predictive encoders such as V-JEPA, but not as a shared component of frame-wise, text-aligned encoders such as SigLIP. We then build counterfactual supervision that makes the language model read it: pixel transformations turn public videos into answer-balanced variants whose answers are set by the transformation and differ only in the change. Trained in a single stage on 4.53M public samples with 32 visual tokens per frame, where open video LLMs use several stages, 9M to 172M samples and 64 to 256 tokens per frame on average, the resulting V-JEPA-LM models (1B to 8B, Qwen3 and Llama) lead every size class in average score across eleven video benchmarks; our 4B model scores 72.1, against 61.1 for the best open 4B and 64.8 for the best open 8B model. A controlled encoder x supervision study shows that the encoder decides what can be read and the supervision what is read: at a matched budget, counterfactual supervision adds 4.5 points on average and 38 on the speed questions of natural benchmarks. Code and models: https://github.com/kkt94/Llama-V-JEPA-LM.

    v-jepabenchmark
  124. arxiv:2611.17821 · cs.CV
    SceneAgent: A Multi-Agent Framework for Portable Game Scene Generation
    Xingyi Li, Jiahao Yang, Yiqin Zhu, Jingwen Ye +2

    Moving a scene generator to a new game engine requires more than exporting its final output: planning, construction tools, and stored experience may also depend on the original backend. We present SceneAgent, a multi-agent framework for reusing the construction process and its experience across backends. A canonical SceneContext maintains the evolving world, while backend adapters isolate coordinate conventions, asset bindings, and export. Structured handoffs let fresh Builder sessions query this persistent state without inheriting the full construction transcript. Across generations, human-approved traces are distilled into asset-ID-free Layout skills and semantic Content skills that can be retrieved and recombined. We introduce GameSceneBench, comprising 69 indoor and outdoor prompts. Under a common asset and evaluation protocol, SceneAgent obtains human tie-adjusted preference scores of 78.2-89.4% against five baselines. Ablations show that reusable skills and the multi-agent protocol primarily improve prompt alignment and scene completeness, with additional runtime. Deployment in a second engine preserves the shared construction protocol and skill library while replacing backend integration components. Together, these results support separating persistent scene state, task-local reasoning, and reusable experience for portable scene construction.

    persistent statemulti-agentagent frameworkevaluation protocol
  125. arxiv:2611.17820 · cs.AI
    Aligning AI Teams
    Siddarth Srinivasan, Morgan Jane Matthews, Jascha Sohl-Dickstein, Erik Jones

    Previous work has shown that alignment training on individual LLM agents can fail to transfer to multi-agent settings. We investigate the underlying mechanisms for this failure, and explore interventions to preserve alignment transfer to multi-agent LLM teams. We focus on two software-engineering tasks (sepsis triage, a news recommender) and ten consulting-proposal generation tasks. We present four findings: (1) The single-vs-team safety gap does not show a clear trend with more capable models. We find that Anthropic's Mythos Preview shows the largest gap on one of our tasks but the smallest on another; (2) Many natural interventions, such as having agents critique each other's work or coordinate in a group chat, do not reliably bring agent teams back to single-agent alignment behavior; (3) diffusion of responsibility substantially accounts for the safety gap: agents in teams are less likely to proactively check for system-level safety failures and more likely to ignore, rationalize, or deflect responsibility for issues when they arise; (4) having a designated safety lead agent ensure system-level safety, and having a coordinator agent publish a safety specifications document are effective at restoring single-agent alignment. We also release magelab, a multi-agent LLM orchestration and experimentation framework. Our findings can help practitioners design agent teams that successfully preserve the behaviors expected of aligned individual agents.

    agentllm agentmulti-agent
  126. arxiv:2611.17819 · cs.AI
    Causally Constrained and Uncertainty-Aware Cardiopulmonary Risk Prediction
    Yuxiang Lai, Jiashuo Zhang, Liang Zhao, Xiaofeng Yang

    Routine chest computed tomography (CT) captures pulmonary, cardiac, and vascular phenotypes that jointly stratify lung cancer and cardiovascular disease, the two leading causes of death in lung-screening populations. Yet clinical risk calculators evaluate each disease in isolation and omit CT findings, whereas unconstrained machine learning overfits small hospital cohorts, learns treatment- and illness-confounded risk directions, and conceals uncertainty when patient records are incomplete. Here we present the Causal Envelope Agent (CE-Agent), a two-stage framework that compiles peer-reviewed epidemiological evidence into explicit constraints on risk-factor weights before any outcome is seen, while estimating local baseline risk and newly added CT phenotypes from cohort data. At deployment, closed-form uncertainty propagation accounts for missing clinical or CT inputs within milliseconds per patient, running over two orders of magnitude faster than Monte Carlo sampling and returning calibrated risk intervals that direct chart review and selective imaging to borderline patients. Across 53,452 lung-screening participants from 33 US hospitals and 6,814 multi-ethnic adults, CE-Agent outperforms cardiovascular calculators in lung screening, exceeds the Pooled Cohort Equations on hard atherosclerotic events while correcting their 1.7-fold risk overestimation, and generalizes from a single hospital or CT scanner to external sites while lowering required sample sizes up to fivefold. Applied unchanged to lung cancer, CE-Agent matches established screening models with accurate calibration even when restricted to risk-factor directions alone, and pairing both predictions improves joint cardiopulmonary triage from a single screening visit.

    agent
  127. arxiv:2611.17814 · cs.AI
    G-CAST: Governed Continual Self-Training for E-Commerce Agents under Evolving Demands
    Yuechun Yu, Wenzheng Kong, Xiaofeng Mao, Jin Zhang

    With the rapid advancement of Large Language Models (LLMs), an increasing number of users are turning to LLM-powered agents for conversational shopping in e-commerce. However, the user needs continue to change as new product categories, procurement scenarios, and feature requests emerge. At the same time, the product's execution harness-including its tools, skills, and operating procedures-continues to evolve to support increasingly sophisticated interactions. These changes create two practical challenges: 1) Demand drift alters the distribution, intent, and complexity of user requests, causing a model trained once to fall behind; 2) Harness evolution requires the model to use newly introduced capabilities correctly without losing behavior that users already rely on. Periodic manual retraining is too delayed and labor-intensive for this setting. To address these challenges, we present G-CAST, a governed continual self-training system that converts daily production evidence under evolving demand and capability contracts into reward-routed, entropy-prioritized training data and governed reward evolution. Governed checkpoint transition under cumulative evaluation controls each model update, enabling the agent to learn newly introduced behaviors without sacrificing previously accepted capabilities. Experiments on a real production e-commerce agent show that G-CAST achieves stronger adaptation to newly introduced demand while better preserving previously accepted behavior compared with a strong autonomous post-training baseline. Ablations further show that both reward-routed, entropy-prioritized data orchestration and governed reward evolution are important to these gains. G-CAST is now deployed in Yuancai(http://cai.1688.com/), a B2B procurement agent on Alibaba's 1688 platform.

    agentpost-training
  128. arxiv:2611.17808 · cs.AI
    CarryOn: Learning What to Keep in Context for Long-Horizon LLM Agents
    Lihuan Li, Ruiyi Yang, Jiru Wang, Yifan Duan +4

    Long-horizon LLM agents can coordinate tools and applications to complete complex, multi-step tasks. As their interaction histories grow, effective context management becomes essential for reliable and efficient execution. Existing methods reduce context through summarization, pruning, or history selection. However, shorter contexts do not necessarily support better decisions: summaries can omit exact details needed for later actions, while retaining entire interactions can carry substantial irrelevant information. We propose CarryOn, a lightweight context manager that pairs compact execution state with fine-grained, source-linked evidence. A learned selector and adaptive retention policy keep relevant evidence available without additional LLM calls for context management. We introduce two complementary evaluation metrics: $S_α$ prioritizes task success while accounting for resource use, and normalized cost per success (NCS) measures the resources spent per successful task. Across AppWorld, OfficeBench, and Tau2 Telecom with three LLM Actors, CarryOn achieves 75.7% macro-averaged task success, exceeding the strongest baseline by 6.7 percentage points. It also achieves the best average scores on both metrics.

    llm agent
  129. arxiv:2611.17806 · cs.AI
    Can Agents Keep Up with Time? An Empirical Analysis of Temporal Adaptation After a Conversational Hiatus
    Hongguan Su, Ming Wang, Shun Zhang, Yifan Wang +4

    Long-horizon agents increasingly serve as persistent personal assistants across modern web platforms, managing schedules, applications, and multi-step plans over weeks or months. Users often return to an ongoing conversation after a long conversational hiatus, while earlier bookings and preparation notes stay on record. During this hiatus, scheduled dates may pass even though the retained records remain unchanged and event outcomes go unrecorded. Existing evaluations of long-horizon agents focus on fact retention and explicit state updates. Whether agents perceive the passage of time and adapt their advice accordingly therefore remains unknown. We frame this ability as temporal state perception and use, which means perceiving the current time and applying it when interpreting dated records. To study it, we design session-rekindling tasks that keep the earlier conversation fixed and vary only the follow-up date and the way temporal information reaches the agent. Experiments across five model routes and five temporal-information conditions show that agents can obtain the current time but often fail to use it. They reliably query the clock and even note in their reasoning that the scheduled date has passed, yet many still tell users to prepare for the expired event. Supplying explicit temporal context reduces this error, and providing the computed date relation removes it entirely. Keeping up with time therefore depends on turning perceived time into updated plans. Based on these findings, we design a session-rekindling benchmark that provides temporal-awareness guidance for designing long-horizon agents. The code can be found at https://github.com/Adoreu2/session-rekindling-benchmark.

    benchmark
  130. arxiv:2611.17797 · cs.CV
    Vision-as-a-Tool: Visual Harness for Multimodal Agentic Coding
    Haolin Yang, Wenshan Wu, Jin Jiang, Yu Cheng +12

    Multimodal agentic coding requires robust visual reasoning to analyze observations and evaluate generated artifacts. Current frameworks largely rely on a Thinking-with-Images paradigm, where agents depend on their intrinsic vision to execute visually grounded tasks. However, this architectural coupling introduces a fundamental bottleneck. Agent performance remains strictly bounded by internal perceptual limitations. Furthermore, text-only agents lack native perception. Without specialized tools, they cannot perform visual reasoning. To overcome these constraints, we introduce the Vision-as-a-Tool (VaaT) paradigm. Prior visual tool-use merely manipulates images for an agent's intrinsic vision to read. In contrast, VaaT reframes general-purpose multimodal large language models (MLLMs) as callable external utilities. As a result, text-only agents can completely offload visual reasoning during agentic execution. Meanwhile, vision-capable agents can dynamically augment their native perception on demand. We instantiate this paradigm in VHarness, a unified framework driven by five atomic actions. We comprehensively evaluate VHarness across Image-to-HTML, Image-to-SVG, and Image-to-3D generation tasks. Experimental results demonstrate that VHarness consistently improves visual fidelity and geometric accuracy across both text-only and multimodal coding agents. These findings highlight that modularizing visual reasoning into callable utilities provides a flexible and scalable foundation for future agentic coding systems.

    agentagentictool-use
  131. arxiv:2611.17796 · cs.AI
    Robustness Evaluation of GPU Cloud Pricing Benchmarks under Incomplete Observations: A Controlled Missingness Study
    Jooya Wong

    Daily GPU rental-price benchmarks can fail when marketplace quotes are missing. This article stress-tests a fixed equal-weight median rule under four imposed masking mechanisms and four benchmark-level recovery methods. Across 5,800 Monte Carlo realisations in 27 scenarios on 190 held-out days, whole-source outages reduce publication coverage to 0.77 while preserving zero conditional error on published dates; composition-changing loss retains 0.86 coverage but produces 7.74 log points of error. A separate past-only stratum-level ablation reduces error by 4.15 to 6.70 log points in five composition-changing cases, using a richer information set. An observable-signal withholding rule improves the primary error-coverage criterion in 0 of 10 cases. Results concern observed asking-price measurements, not transaction prices or identified real-world missingness mechanisms.

    benchmark
  132. arxiv:2611.17794 · cs.LG
    Federated Physics-Informed Thermal Field Prediction Across Unseen Geometries in Distributed Additive Manufacturing
    Amirul Islam Saimon, Zhenyu, Kong

    In additive manufacturing (AM), each change in part geometry requires a new thermal simulation, which makes design evaluation over many candidates costly. Geometry-aware models reduce this cost by predicting thermal fields on unseen parts with a single learned model, but they assume that diverse geometries are available centrally. This assumption is restrictive when each manufacturing site holds only a limited set of proprietary part designs that it cannot share. We propose Fed-PIGANO, a federated framework that trains physics-informed geometry-aware neural operator (PI-GANO) models at individual sites and aggregates their geometry encoders and prediction networks without sharing part geometries. We evaluate the framework on a 2D steady-state thermal benchmark for metal AM, with geometries from five polygonal families distributed across three sites under four scenarios that range from identical to strongly different family compositions. Over seven replications, Fed-PIGANO achieves mean relative L2 errors of 1.83% to 2.04% across all 12 scenario-site cases, compared with up to 20.46% for local-only training, and consistently yields lower PDE residuals. Its prediction accuracy is comparable to that of a centralized model trained on the pooled geometries of all sites. These results demonstrate that federated learning lets manufacturing sites benefit from each other's geometries without sharing them, enabling rapid thermal prediction on new part designs for tasks such as shape optimization, inverse design, and uncertainty quantification.

    benchmark
  133. arxiv:2611.17793 · cs.AI
    MASTARS: Multi-Agent Sequential Trajectory Augmentation with Return-Conditioned Subgoals
    Jiwon Jeon, Myungsik Cho, Woojun Kim, Seongmin Kim +3

    The performance of offline Multi-Agent Reinforcement Learning (MARL) is limited by the quality of the fixed offline dataset, motivating trajectory augmentation. However, augmenting multi-agent trajectories presents a fundamental challenge of the diversity-coordination trade-off: independent generation improves diversity but lacks coordination, while joint generation induces coordination but lacks diversity, reproducing the training distribution. To address this challenge in multi-agent data augmentation, we introduce MASTARS, a novel diffusion-based framework that generates multi-agent trajectories sequentially across agents while enforcing cross-agent consistent coordination via the method of inpainting. MASTARS produces trajectories that are diverse, coordinated and globally coherent. Experiments on various benchmarks such as MPE, SMAC, and SMAC-v2 demonstrate significant performance gains across multiple offline MARL methods.

    multi-agentbenchmark
  134. arxiv:2611.17790 · cs.LG
    AutoRL: A Multi-Agent System towards Autonomous Reinforcement Learning
    Simon Tang, Wenjie Shi, Yongjing Yin, Keheng Wang +7

    Recursive self-improvement (RSI) is one of the most promising paths toward an AI singularity for superintelligence. Achieving this vision requires AI systems to undertake the full training pipeline for iterative improvement. Reinforcement learning (RL) plays an essential role in this cycle to improve model capabilities and generalization, yet automating RL remains difficult due to unstable training dynamics and the need for sustained supervision. We introduce AutoRL, a multi-agent system for automating RL training, as a cornerstone for future RSI. Different from conducting research by end-to-end complete experiments, AutoRL monitors the evolving training state and integrates process-level supervision and research into an ongoing RL training, following the practical workflow of an RL researcher. Specifically, AutoRL consists of three RL-specific agentic roles. Perceptors track training dynamics, rollouts, evaluation, and efficiency, triggering deliberation when they detect anomalies, degradation, or stagnation. Three complementary Proposers diagnose problems and refine candidate interventions through shared evidence and checks. The Orchestrator reviews their recommendation and applies approved changes through online hot updates or checkpoint-based resumption. Experiments on Qwen3 and LongCat-Flash-Lite demonstrate the superiority of AutoRL over static RL recipes and a general agent. The research records show that existing agents can handle unattended RL supervision and dynamic adjustment. The difficulty for future RSI lies in developing fundamental methods that are valuable for resource-consuming validation.

    multi-agentagenticagent systemself-improvement
  135. arxiv:2611.17788 · cs.CL
    RAG-FGO: Dynamic Optimization of Query-Rewriting Agents for Generative RAG
    Qigan Sun, Xudong Wang, Chaoning Zhang, Chenghao Li +8

    In retrieval-augmented generation (RAG), query rewriting serves as a core interface that aligns user intent with external knowledge bases, and its effectiveness directly determines overall system performance. However, generative retrieval faces three critical challenges in real-world deployment: noisy retrieval feedback, the difficulty of jointly optimizing discrete prompt templates and continuous decoding-control parameters, and the limited generalization of static rewriting strategies under semantically complex scenarios. To address these issues, we propose RAG-FGO, a dynamic optimization framework for generative retrieval strategies. RAG-FGO formulates query rewriting as a non-differentiable black-box optimization problem in a mixed discrete-continuous space, and represents prompt templates and decoding parameters as adaptively optimizable query-rewriting agents. The framework performs global search to enable semantic jumps and structural reorganization that escape local optima, while applying local perturbations for fine-grained alignment and precise refinement. In addition, RAG-FGO introduces an elite-preserving memory pool that accumulates high-quality strategies across iterations, thereby mitigating the impact of stochasticity and noisy feedback on optimization stability. Experimental results on a diverse suite of benchmarks demonstrate the effectiveness and robustness of the proposed method, supporting its applicability to complex semantic retrieval tasks.

    memoryretrieval-augmentedragbenchmark
  136. arxiv:2611.17784 · cs.LG
    Direction or Magnitude? Learning Scalars over Task-Calibrated Directions
    Ziqing Wen, Xiaohui Song, Jiahuan Wang, Ping Luo +4

    Parameter-efficient fine-tuning (PEFT) aims to preserve adaptation quality while reducing training costs. We propose SCALAR, which achieves performance comparable to LoRA with substantially fewer trainable parameters, lower peak training memory, and higher throughput. SCALAR constructs a fixed input basis from task activations and assigns a random output dictionary to each input direction. Fine-tuning learns only layer-specific scalar coefficients, reducing gradient and optimizer-state storage. Grouped computation aggregates output atoms before applying the adapter to token activations, while dictionary sharing across compatible layers reduces fixed-basis storage. Experiments on commonsense reasoning, mathematical reasoning, code generation, and knowledge evaluation after instruction tuning demonstrate effective adaptation at small trainable budgets. On commonsense reasoning, SCALAR achieves accuracy comparable to LoRA with about 4% of its trainable parameters. System measurements on Qwen2.5-7B show 20% lower peak training memory and 6% higher throughput than LoRA. The combination of adaptation quality, a compact trainable state, and training efficiency makes SCALAR a promising approach to resource-constrained fine-tuning.

    memory
  137. arxiv:2611.17781 · cs.AI
    TReC: Evaluating Move-Induced State Changes in Instruction-Conditioned Go
    Zhuangsheng Li, Yunhan Zhang, Dailin Li, Qingxian Yu +3

    When a Go model predicts a move different from the expert's, coordinate distance measures how far the prediction lands from the reference. We introduce Transition Representation Consistency (TReC) to compare how the two moves change a board-state embedding. Both moves are executed from the same position under the game rules, and TReC is the cosine similarity between the two resulting representation changes. In experiments with a frozen KataGo network on 680 mid-to-late-game positions from five games, the gap between the highest and lowest candidate per position has a median of 0.01 to 0.08 for the direct successor-state cosine and 0.88 to 1.33 for TReC, depending on the evaluator layer. TReC follows coordinate distance only weakly: the rank correlation between a candidate's distance to the reference and its TReC, computed within each position, has a median of -0.19 over these positions.

    evaluator
  138. arxiv:2611.17780 · cs.LG
    LuthorSQL: A One-Pass Text-to-SQL Model as System 1 for a Slower LLM Agent
    Ibrahim Memon, Armin Karamzade, M. Hanif Jhaveri, Roy Fox

    Frontier LLM agents that query databases with tools are among the most accurate text-to-SQL systems, but per-query calls to a hosted model are slow, costly, and expose sensitive organizational queries to external APIs. For example, a Claude Opus 5.5 agent serving 10,000 queries a day costs over $70,000 a year at list price. To what extent can local hardware already owned by an organization offload this burden without compromising the agent's accuracy? We present LuthorSQL, a non-autoregressive 7B text-to-SQL model that serves as a rapid System 1 gate for a slower LLM agent (System 2), escalating queries only when uncertain. LuthorSQL makes a single forward pass to predict slot scores over a bounded grammar for query plans, compiles and runs up to 32 candidate plans, and chooses an answer using a trained validator. On its own, LuthorSQL reaches 68.03% execution accuracy on BIRD's hidden test set at 0.84 s per question on a single GPU, maintains its accuracy in 9 GB with 8-bit quantized weights, and transfers zero-shot to Snowflake, BigQuery, MySQL, PostgreSQL, and DuckDB within 3.1 points of its native SQLite accuracy. Combined with Claude Opus 5.5 in a pre-registered confidence cascade over 227 questions on databases published after Opus's training cutoff, LuthorSQL answers nearly half of all questions locally and cuts API cost by 33.3% at the same accuracy. Finally, contamination probing shows that Opus reproduces 48-92% of public benchmark gold queries verbatim without seeing the schema; however, its tool-assisted accuracy holds on renamed schemas and reaches 95.6-97.4% on post-cutoff databases. Code and data: https://github.com/IAMIbrahimmemon/luthorsql

    agentllm agentbenchmark
  139. arxiv:2611.17775 · cs.RO
    Beyond Prediction Gain: LASR as a Metric for Action-Grounded Latents in World Models
    Yu Zhou, Changsong Lei, Yixiang Wei, Kairui Wen +5

    World models that support planning, control, and interpretable reasoning require latent representations that genuinely encode action. Yet the large-scale video data from which such models are learned overwhelmingly lack action annotations. Latent-action models infer action-like variables from observation sequences, enabling action-aware learning from unlabeled videos or trajectories. A fundamental concern, however, remains: a latent that improves future prediction may simply encode the current state, episode progress, task identity, or other predictive shortcuts. Existing evaluation practices---reconstruction loss, rollout fidelity, action recovery, or downstream performance---cannot isolate this distinction. We introduce the Latent Action Support Ratio (LASR), a dimensionless metric that measures, under offline evaluation on held-out data, whether a learned latent provides predictive support for action-related targets beyond the current state and non-action alternatives. LASR provides two complementary summaries: GeoLASR captures typical support, while WorstLASR identifies the weakest case. We instantiate the LASR benchmark with recorded actions and corresponding physical state changes as targets, episode-held-out splits, one predictor type for all comparisons, and positive and negative controls. Experiments on recorded robot manipulation data and a matched multi-task panel demonstrate that LASR detects known action information, rejects representative non-action shortcuts, and enables graded comparisons among imperfect latent representations. We further use LASR to train and audit a lightweight residual adapter, obtaining descriptively better point estimates whose uncertainty is assessed with paired comparisons. LASR thus provides a principled offline protocol for measuring action-related predictive support in world-model latents.

    manipulationworld modelbenchmark
  140. arxiv:2611.17766 · cs.AI
    GDP.xlsx: Benchmarking Frontier Models on Professional Workbooks
    Suhaas Garre, Sushant Mehta, Edwin Chen

    We introduce GDP_xlsx, a benchmark of 70 realistic workplace requests written by professionals in 13 domains, each paired with an expert rubric of binary criteria and with the complete workbook or CSV file it refers to. Together the files hold 2.18 million populated cells, and the rubrics 791 criteria. Agents work in a container with spreadsheet tools, and an agentic judge that can open the file applies the rubric. Completeness and exclusivity criteria require set-valued answers to be found in full and with nothing extra. We rank models by strict pass@1, the share of attempts that meet every criterion. Across 20 model configurations, including frontier models from Google, Anthropic, and OpenAI, the top-scoring one, Gemini 4 Argon, passes only 38.3% of attempts, and the median score is 19.3%, leaving wide headroom. We release the evaluation harness and three public tasks; the other 67 are held out to protect the leaderboard from contamination.

    agenticbenchmarkleaderboard
  141. arxiv:2611.17763 · cs.AI
    Knowledge After Collapse: Context and Wording Shape Factual Accessibility in Language Models
    Sujato Dutta, Prithviraj Patil, Chandrajit Bajaj

    When a language model accepts a false correction in a factual discussion, its next answer can depend on how the conversation continues. We introduce Knowledge Answer Divergence (KAD), an event-conditioned evaluation protocol that distinguishes initial false-answer adoption from subsequent persistence, recovery, and abstention. KAD freezes the conversation at the first explicit adoption of a designated false target and branches independent probes from that exact history. This enables matched comparisons of post-adoption behavior and history-management interventions that a single susceptibility score cannot capture. We evaluate eight local checkpoint conditions and six hosted endpoints, alongside an audited 300-question rerun. In corrected Llama 3.2 Instruct 3B controls, a neutral-history branch yields 42/46 correct responses, compared with 16/46 under full history. A matched cue-placement control changes false adoption by 27.3 percentage points in Qwen 3.5 4B. Twelve independently authored pressure templates demonstrate substantial sensitivity to wording. Together, these findings demonstrate the utility of KAD for evaluating factual reliability beyond initial susceptibility, enabling systematic comparisons of false-answer persistence, recovery conditions and conversational interventions.

    evaluation protocol
  142. arxiv:2611.17762 · cs.CV
    Creative Product Design under Brand Image Constraints
    Takeshi Onishi, Hiroya Makino, Takahiro Yamaguchi, Hiroyuki Sakai

    Industrial design requires creative visual exploration while satisfying constraints that are often product-specific and implicit. Existing text-to-image (T2I) approaches can generate creative designs through concept blending or multi-agent self-enhancement, but they provide limited mechanisms for ensuring that generated designs remain consistent with such constraints. We focus on brand image as a representative design constraint and propose a framework for brand-aware redesign generation. Given a product name, the framework constructs a brand image survey report that summarizes product and market information, derives a creative direction from the report, and integrates this direction into a self-enhancement loop in which generated images are critiqued and prompts are revised. We evaluate the framework on 50 redesign tasks across five product categories, comparing it with Naive T2I, IP-Adapter/IT-Blender, and unconstrained self-enhancement under the same T2I backbone. By a VLM-as-Judge protocol to measure creativity and brand identifiability, the proposed framework achieves a more favorable trade-off between the two objectives, with particularly clear gains in highly constrained categories such as automotive and motorcycle, where the plausible design space is relatively limited compared with that of other product categories. These results suggest that dynamically eliciting and textualizing brand image can guide creative product design generation under brand image constraints.

    multi-agent
  143. arxiv:2611.17758 · cs.LG
    Efficient Genotype Classification from 3D MRI in ADPKD via Density-Aware Point Sampling and E(3)-invariance
    Kwanghee Lee, Hyeongsub Lee, Youngwoo Kim, Hyunji Jeong

    Early detection of genetic kidney diseases, especially autosomal dominant polycystic kidney disease (ADPKD), is crucial because certain genotypes such as truncating PKD1 mutations cause rapid kidney growth and early progression to end-stage kidney disease. MRI-based genotype inference is a potential low-cost complement to genetic testing. However, although many deep learning methods have been developed for classifying various MR images, applying them to ADPKD remains challenging: (1) abdominal MR images are generally too large to be handled by deep learning models designed for smaller images like brain MRI, and (2) the amount of labeled data for ADPKD genetic mutations is insufficient to train deep learning models. In order to address these challenges, we propose E3PN, an efficient 3D MRI-based ADPKD genotype classification network that relies on point sampling and E(3)-invariance. We utilize a point representation of 3D MRI, which consumes less GPU memory than a voxel representation. For further memory efficiency, 3D points are sampled from a mixture distribution over the two kidneys in each image. Since E(3)-invariance has been shown to improve data efficiency, we also incorporate it into our model through coordinate canonicalization. On our ADPKD dataset, E3PN trains 163 times faster and uses 103x less GPU memory than a 3D ResNet, while also improving F1 from 0.165 to 0.226 over both baselines. These results establish a first benchmark for five-way ADPKD genotype classification from 3D MRI and show that point sampling and coordinate canonicalization yield two-orders-of-magnitude efficiency gains with a modest but consistent improvement in F1.

    memorybenchmark
  144. arxiv:2611.17755 · cs.CV
    VACO: Vertex-Aligned Decoding for Dense Hand Contact Estimation from a Single Image
    Binhong Ye, Jin-Chuan Shi, Tao Liu, Xin Wang +2

    Dense hand contact estimation from a single RGB image is a foundational primitive for dexterous manipulation and 3D interaction analysis. Direct predictors typically decode the entire per-vertex contact map from one shared query, leaving individual vertices ungrounded in the image and smearing contact across adjacent, non-touching surfaces. We present VACO, a vertex-aligned decoder in which each MANO vertex samples local image features at its projected location. A hand-reconstruction backbone, jointly trained for reconstruction and contact, predicts these locations at inference without an auxiliary reconstructor. Training uses noise-perturbed ground-truth projections and end-to-end feature adaptation to improve robustness and contact-sensitive features. A single VACO checkpoint sets a new state of the art on five benchmarks spanning hand-object, hand-human, hand-scene, and bimanual interactions. It improves F1 by +20.7 pp over the strongest prior direct predictor on Hi4D and generalizes to the held-out HOT3D and HOCap, with gains of +5.5 and +2.7 pp. The resulting contact predictions also improve downstream HOI reconstruction.

    manipulationdexterousbenchmark
  145. arxiv:2611.17754 · cs.CL
    See Every Block Once: Variance-Reduced Post-Training for Masked Diffusion Language Models
    Yunzhen He, Hiroaki Yamagiwa, Hidetoshi Shimodaira

    Policy gradient methods are widely used to align large language model outputs with human preferences, but the log-likelihoods they require cannot be directly computed in Masked Diffusion Language Models (MDLMs). MDLM post-training therefore relies on a surrogate objective based on the Evidence Lower Bound (ELBO), and recent work estimates a blockwise variant of this objective via Monte Carlo methods to support blockwise generation. However, existing estimation schemes randomly sample the block index, so they may evaluate the same block more than once and leave some blocks unevaluated. This introduces variance arising from block selection into the estimates. In this work, we propose a full-block estimator that evaluates every block exactly once and averages the results. We theoretically show that the proposed estimator remains unbiased while eliminating the between-group variance arising from block selection, thereby achieving lower variance than the baseline for the same sample size. In experiments on five benchmarks spanning mathematics, planning, and coding, our method achieves the highest accuracy on four benchmarks and outperforms the random block sampling baseline by an average of 12.79 points.

    post-trainingbenchmark
  146. arxiv:2611.17753 · cs.AI
    Superfluid: Local AI Server for Distributed LLM Inference
    Prabod Rathnayaka, Isura Manchanayake, Srikandabala Kogul, Steve Siu +2

    Running large language models (LLMs) on local devices keeps data private, removes per-token cost, cuts network latency, and keeps working without a connection. Two shifts now strain how such models are served. On one device, several agents share a locally served model with an interactive user. Across devices, a developer or small team often owns several capable machines from different vendors, which sit idle most of the day because every local server serves one runtime on one machine. We present Superfluid, an open-source LLM server built for local inference at both scales. On one device, Superfluid drives llama.cpp, MLX, or BaseRT in isolated worker processes through one scheduler that moves interactive requests ahead of agent work, computes a prompt shared by many requests once, and writes every token to a durable log before it reaches the client. Across devices, one Superfluid server turns a fleet of machines into a single API endpoint. Machines with different hardware serve copies of the same model, receive work according to their measured speed, and can join or fail under load without losing a request. For Qwen3.8-27B on an Apple M5 Pro, a chat sent while eight agents are working receives its first token in 1.5 s, against 291-612 s on llama-server, Ollama, and mlx_lm.server. On six machines from three chip vendors serving 64 simulated users for one hour, Superfluid answers all 5,041 requests while machines join one at a time and one is killed and rejoins; throughput grows from 112 to 534 tok/s and the median first-token time falls from 99 to 12 s. On twelve Mac Studio M3 Ultras it sustains 10,656 tok/s across 38,563 requests from 384 agents without a failure. Superfluid is available under the Apache-2.0 licence at https://github.com/basecompute/superfluid.

    agent
  147. arxiv:2611.17751 · cs.RO
    When Does Cross-View Consistency Matter in Multi-View Appearance Augmentation? A Controlled Study for Visuomotor Policies
    Zixuan Jiang

    Multi-view visual augmentation often seeks consistency across cameras, but it is unclear which forms of consistency matter for downstream control. We use simulation-state replay to manipulate cross-view coupling while preserving the per-camera perturbation distributions, physical trajectories, and action labels. Across seven LIBERO manipulation tasks, marginal-matched changes in global appearance, ghost injection, and target color have small average costs for a late-fusion Diffusion Policy: point estimates are near zero or negative, and the largest upper endpoint of the 90% posterior credible intervals is 2.55 percentage points. In contrast, geometrically anti-coupled observations reduce success by 3.30 percentage points relative to coupled observations (90% CrI [0.64, 6.28]), with displacement magnitude, direction marginals, and the number of affected views matched. Camera interventions reveal substantial dependence on wrist-view geometry, while feature analysis shows appearance sensitivity patterns associated with the factors randomized during training. ACT shows a higher target-color cost estimate than DP, with uncertainty that includes zero and sensitivity to task composition. Within the tested tasks and policy configurations, these findings motivate selective consistency constraints: appearance coupling can sometimes be relaxed, whereas task-relevant geometry and cross-view spatial correspondence warrant protection.

    manipulationdiffusion policylibero
  148. arxiv:2611.17735 · cs.LG
    Logically Consistent Accurate Forecasts
    Kate Yuan, Rohan Alur, Manish Raghavan, Devavrat Shah

    Agentic forecasting systems based on large language models (LLMs) are now competitive with expert human forecasters, but these systems still suffer from a critical limitation: the inability to produce forecasts that are logically consistent. For example, forecasts for related events often fail to satisfy the axioms of probability, which in turn implies that related forecasts cannot simultaneously be correct and undermines their utility for decision makers. Enforcing consistency is not trivial -- naive approaches (e.g., projection onto the set of consistent forecasts) can degrade accuracy. We argue that the key to overcoming the apparent tradeoff between consistency and accuracy lies in the reasoning which supports each forecast. We finetune a supervisory language model that leverages this reasoning to jointly reconcile inconsistencies while improving the accuracy of related forecasts. On ForecastBench, our method improves the Brier score of a retrieval-augmented GPT-5.1 forecaster by 9.3% while reducing measures of structural inconsistency by 82-90%. We interrogate our method via a comprehensive series of baselines and ablations, and find that no alternative method matches our gains in consistency and accuracy simultaneously.

    retrieval-augmentedagentic
  149. arxiv:2611.17733 · cs.CV
    HarmonyText: Harmonious Scene Text Editing via Reverse-to-Refined Pair Synthesis and Region-Decoupled Frequency Learning
    Ziqi Yi, Huiguo He, Guitao Xu, Peirong Zhang +7

    Natural language-guided, mask-free scene text editing aims to modify text in real-world images using only textual instructions, yet remains challenged by unreliable pseudo-GT supervision, over-smoothed text generation, and the absence of GT-free evaluation. Existing methods often rely on imperfect pseudo-targets with background discrepancies, while standard MSE-based velocity regression tends to suppress high-frequency details, leading to blurred text boundaries and weakened local textures. To address these issues, we propose HarmonyText, a mask-free scene text editing framework with region- and frequency-aware supervision. HarmonyText introduces Region-Decoupling Optimization (RDO), which aligns edited text regions with pseudo-targets while anchoring preserved regions to the source image, and Cascaded High-Frequency Texture Guidance (CHFTG), which applies cascaded wavelet-domain constraints to preserve sharp boundaries and fine-grained textures of textual content. We further propose Reverse-to-Refined Pair Synthesis (RePair) to construct STE25K, a high-quality 25K paired dataset, and introduce a GT-free evaluation protocol with Color Distribution Divergence and a Harmony Assessment Model. Extensive experiments show that HarmonyText outperforms existing methods in text editing accuracy, background fidelity, color consistency, and perceptual harmony.

    evaluation protocol
  150. arxiv:2611.17730 · cs.RO
    ERA: Error-Resistant Autonomous Object Navigation with Predicted Depth
    Shunyang Huang, Shuo Wang, Wenyuan Xie, Mingyu Lin +2

    Depth and 3D foundation models can predict geometry from RGB images, allowing zero-shot object navigation without a depth sensor. However, when predicted depth is mapped as if it came from a sensor, view-dependent errors accumulate in the map, remove exploration goals, and end the search before the agent can observe and correct them. We propose ERA, a training-free object navigation system that replaces the depth sensor with depth predicted from RGB images and poses. ERA follows a paradigm we call Move with Error: movement provides new views that correct map errors, and the corrected map enables further movement. It realizes this paradigm with revisable occupancy and with exploration from reachable viewpoints, which continues the search when no frontier remains, and verifies the target across views before stopping. On three benchmarks, ERA achieves state-of-the-art performance among training-free methods without depth sensors. Its map and exploration components also raise VLFM's success from 10.8% to 61.7% under the same predicted depth. We further deploy ERA on a real robot.

    agentbenchmark
  151. arxiv:2611.17720 · cs.RO
    Ego2Flow: Benchmarking the Traffic-System Impact of Autonomous Driving on Maps Distilled from Real-World Roads
    Yu Dong, Weiying Shen, Hangyu Li, Shu Han +7

    End-to-end autonomous driving is predominantly evaluated through ego-vehicle safety and task completion. Its deployment in urban traffic, however, creates sustained interactions among vehicles whose collective effects remain insufficiently evaluated. We introduce Ego2Flow, a closed-loop benchmark for assessing the traffic-system impact of autonomous driving on maps distilled from real-world cities. To construct realistic 3D urban environments for sustained closed-loop traffic simulation, we develop an agent-guided map distillation framework that integrates real-world maps and observations with complementary geographic information. Through an iterative feedback mechanism and a traffic information graph that jointly guide scene generation, the framework preserves lane topology, critical road geometry and decision-relevant visual conditions while simplifying nonessential details and completing unobserved regions. Using our proposed framework, we generate and publicly release six CARLA maps from the Argoverse 2 dataset, covering the corresponding real-world regions and enabling paired comparisons of driving behavior under real and simulated observations. On these maps, Ego2Flow evaluates how autonomous driving policies affect traffic efficiency, safety, stability and energy consumption across diverse road structures. Compared with existing construction methods, our framework achieves the lowest average displacement error between planning outputs from paired real and simulated observations. We further benchmark representative driving policies to characterize their system-level impact on urban traffic. Project Page: https://github.com/Fisher-007/Ego2Flow.

    benchmark
  152. arxiv:2611.17689 · eess.SY
    Orbit-AWARE: From Onboard AI Deployment to Sustainable In-Orbit Intelligence
    Yunxin Geng, Kaige Zhu, Tongtong Shu, Zhan Gao +5

    The growing role of space computing in time-critical satellite services makes the long-term reliability of onboard artificial intelligence (AI) a mission-level requirement. Model drift can undermine this reliability through critical-observation suppression or noncritical priority flooding, displacing both mission information and evidence for model correction. To address this drift-induced evidence displacement, this article presents Orbit-AWARE, a safeguard-first lifecycle management (LCM) architecture. A comparison of existing approaches motivates cooperative Data, Cognitive, and Trust Planes for continuous monitoring, event-triggered agentic orchestration, and deterministic governance. The architecture retains a model-independent sample stream before degradation is confirmed and coordinates evidence selection and exchange for diagnosis and ground-assisted adaptation. Candidate updates are verified before control over data disposition is restored. In an xBD-based emergency Earth observation (EO) case study, Orbit-AWARE delivers 14.3\% more critical observations and achieves 8.3\% higher sustainable mission utility than the Rule-Based Drift Loop under a common 10\% downlink budget. Finally, we discuss future research directions for extending LCM through task-aware reliability assessment, memory- and tool-enabled onboard agents, and evidence-centric space networking. This work makes evidence availability and recovery authorization explicit design objectives for dependable onboard AI services.

    agentic
  153. arxiv:2611.17682 · cs.AI
    Agent-written heuristics versus commercial solvers on unsolved job-shop benchmark instances
    Andre Albrecht, Enzo Morosini Frazzon, Leandro C. Coelho, Bruno Vieira

    Commercial solvers are the standard tool for difficult scheduling problems. We study whether a frontier large language model, used as a coding agent with open-source tools only, can match them in quality and speed. We report every run, and independent checkers verify every schedule. On the 100,000-operation job shop tai_100_1000_1, an agent-written solver reached 525,515, below OptalCP, Hexaly, and CP Optimizer. Unchanged, it improved all ten instances of the family and closed two. With equal budgets on the same workstation, it ended 0.0-1.8% above the lower bound, against 2.6-5.8% for CP Optimizer and 2.4-4.4% for Hexaly. On the Behnke-Geiger flexible job shops, an agent-written tabu search improved 11 of 32 open instances. With two transport robots, an agent-written solver improved seven of ten MKT instances. New lower bounds show that four published MKT values cannot be attained on the distributed instance files. Agent-run improvement campaigns closed eight further instances.

    agentbenchmark
  154. arxiv:2611.17681 · cs.AI
    OCTAVE: A Capability-Resolved Benchmark for End-to-End HTML5 Game Generation
    Kehang Li, Xinlan Xu, Yulin Jin, Heng Cao +9

    Complete game generation requires agents to coordinate rules, interactions, states, and presentation over a long development process. A single aggregate score obscures which capabilities are implemented and which requirements remain unmet. We present OCTAVE, a benchmark for complete HTML5 game generation organized around eight Capability Dimensions, with 180 tasks and 1,800 Evaluation Requirements drawn from real player requests, commercial-mechanics reconstruction, and human expert design. It covers PC Web, Mobile Web, 2D, and 3D. Requirements distinguish Completeness, assessing publicly specified core implementation, from Richness, assessing hidden, deeper system relationships. Static source-code judgments provide requirement-level diagnosis across capabilities and task conditions. Evaluation of 11 model-harness combinations yields 33.2% Completeness and 32.5% Richness; 19.9% of generation records are static gate failures or missing artifacts. Excluding same-model self-evaluation, even the best combination remains below 50% on both public and hidden requirements. Richness is not harder than Completeness under every task condition. Systems perform relatively better on Core Rules and Numerical Design, while AI/NPC and Feedback and Game Feel show more prominent gaps. Requirement difficulty offers coarse stratification but is only weakly negatively correlated with pass rates. A limited human cross-check finds that requirements judged satisfied by humans but failed by the automated evaluator are the main direction of disagreement. OCTAVE extends evaluation beyond aggregate scores to diagnosis localized by requirement, capability, and task condition.

    benchmarkevaluator
  155. arxiv:2611.17680 · cs.AI
    Aligned, Conflicting, or Unspecified? A Diagnostic Benchmark for Multimodal Time Series Forecasting
    Xintong Wang, Haihan Nan, Ruidong Li

    Artificial Intelligence of Things (AIoT) systems increasingly combine sensed time series with external text such as expert reports and news, yet how reliably models use text that agrees or conflicts with the sensed history remains unclear. We introduce the first real-world diagnostic benchmark organized by text-history relationship: 1,762 text-window pairs across sensor telemetry (Weather), web-service traffic (Wikipedia), and market data streams (Finance), each labeled Aligned, Conflicting, or Unspecified by whether the text's direction agrees with the historical trend, opposes it, or is not stated. Across ten pretrained models evaluated zero-shot, we investigate text selection and combination under text-history relationships (RQ1) and error diagnosis and horizon-scaled revision control (RQ2). For RQ1, we compare two types of text on paired Weather windows: one directly forecasts temperature changes, while the other describes related weather conditions, such as winds and fronts. Under Aligned, the first type yields the lowest error for most models. Under Conflicting, it increases most models' error above the no-text baseline; we therefore add only its stated magnitude and timing to the second type. Compared with either type alone, this combination lowers mean absolute scaled error (MASE) for six models and raises directional accuracy (DA) for seven. For RQ2, an error decomposition attributes the error increase to oversized revisions, and a training-free rule that limits them outperforms forecasts both with and without text for the four models with this trade-off, lowering MASE by 4.49%-21.98% and raising DA by 2.99%-9.71%. Our dataset is available at https://github.com/Shelton1202/ACU-Bench.

    benchmark
  156. arxiv:2611.17662 · cs.LG
    Learn then Bet: Learning Interpretable Rules for Reliable Agent Intervention
    You-Lin Chen, Xinyi Lisa Chen, Pedro Herrero-Vidal, Xiaojing Yu +2

    Autonomous agents now run for hundreds of steps with little or no human supervision. In practice, practitioners use runtime monitors such as turn limits and keyword filters to decide when a run should be stopped, upgraded, or assisted. However, these heuristics may stop runs that would otherwise succeed or allow failing runs to continue. To address this limitation, we introduce LeBet (Learn then Bet), a statistical learning framework that derives its learning objective from sufficient conditions for valid and powerful intervention. In particular, LeBet learns an interpretable OR-of-AND monitor and accumulates evidence through a fixed betting statistic, intervening when the evidence reaches a prescribed threshold. Under finite within-class exchangeability and leave-one-out stability, we derive finite-sample guarantees for false-positive control (validity) and detection power conditional on the training data, using the same trajectories for learning and calibration without assuming independent and identically distributed trajectories. To address the intractable combinatorial optimization problem in our learning formulation, we use stochastic local search that leverages the monitor's Boolean structure to propose rule edits aimed at improving empirical separation. On SWE-bench Lite with Ministral 3 8B, the learned monitor saves $34.46\%$ of input-plus-output tokens and increases output precision from $18.33\%$ to $21.59\%$. The empirical false-positive rate is $13.64\%$. Our results show how to formalize the learning problem over interpretable rule sets with statistical conditions and establish training-conditional guarantees for reliable intervention based on partial histories.

    agentautonomous agent
  157. arxiv:2611.17661 · eess.SY
    Feasibility of Pareto-Improving Damping Control of Forced Oscillations: The Active-Power Case
    Haozong Wang, Yuru Wu, Yan Wen, Qian Liu +4

    Damping control of a forced oscillation may reduce amplitudes at certain locations while increasing them elsewhere. This paper introduces Pareto improvement as a criterion for assessing such control: at least one nodal oscillation amplitude decreases, while none increases across all other nodes. The analysis considers active-power forcing sources and active-power control injections in connected swing networks with positive inertia and damping. Necessary conditions for Pareto improvement feasibility are derived from the normalized nodal responses. It is proven that for every network with a single oscillation source and a single actuator, at least one positive forcing frequency is shown to preclude Pareto improvement, regardless of the amplitude and phase of the control input. A three-node counterexample shows that this limitation does not persist with multiple actuators. PSS/E simulations of a 6-bus benchmark system and a modified IEEE 39-bus system validated the analysis. These findings suggest the value of assessing control effects across the network and further investigating multi-actuator damping control.

    benchmark
  158. arxiv:2611.17658 · cs.RO
    Imitation Policies Learn a Proprioceptive Anchor: A Predictive Law for Perception Noise and Its Closed Loop Cost
    Christopher Malili

    Imitation policies for robot manipulation are often trained on noisy object positions, and they learn to discount the noisy reading. We show how much they discount it, and why. Because demonstrators steer toward the object, the robot's own state carries information about where the object is. A trained policy combines this proprioceptive anchor with the reading and weights the reading by $α= V/(V + s^2)$, where $s^2$ is the reading noise and $V$ the anchor's error variance. $V$ belongs to the learner and is far smaller than a linear model of the demonstrations implies, which is why classical errors-in-variables (EIV) regression overestimates $α$. Measured once, $V$ predicted ten held-out sensitivities across two robomimic tasks, two learners and four new noise levels, with mean absolute error 0.025 against 0.37 for classical EIV, and a parameter-free bias test rules out the simpler explanation that the policies just overestimate the noise. The discount does not carry over to deployment, where the policy produces its own state: closed-loop sensitivity to a persistent reading offset exceeded $α$ in all 15 policies that passed the registered gates, although the stricter registered prediction, that all would follow the offset fully, held for 8 of them. Post hoc, lower $α$ went with lower success, steeply for diffusion, and a registered prospective test confirmed this for diffusion on Can at two new noise levels: the measured $α$ predicted success before any rollout, within 0.03 (95% intervals $\pm 0.08$), although training noise predicts it equally well. Every test was registered before its data, including the classical benchmark that failed, and all other analyses are labelled post hoc; all results are in simulation with synthetic Gaussian noise.

    manipulationbenchmark
  159. arxiv:2611.17646 · cs.LG
    Binarization as an Attack Surface in Weightless Neural Networks
    Lucas Senos Coutinho, Matheus Vargas, Pedro Teixeira, Heudson Tosta Mirandola +4

    Weightless Neural Networks (WNNs) are lightweight classifiers that binarize their inputs and learn by storing binary patterns in lookup memories, which makes them attractive for resource-constrained tasks such as intrusion detection. Because a small perturbation that crosses no binarization threshold leaves the binary input unchanged, WNNs may appear naturally robust to adversarial examples; a perturbation that does cross a threshold, however, can still change the prediction. We study this tension in WiSARD, the canonical RAM-based WNN, and in two variants that replace its memories with Bloom filters, Bloom WiSARD and BTHOWeN. A continuous perturbation reaches these classifiers only by crossing an encoding threshold, changing at least one binary tuple, and redirecting a memory query; we formalize this mechanism with a threshold-crossing bound and a response-margin certificate for standard Boolean WiSARD. Controlled MNIST experiments compare linear and data-driven thermometer encodings under model-agnostic sensitivity probes, a score-based Simulated Annealing attack, and surrogate-transfer attacks. Changing only the encoding while keeping each model otherwise fixed shows that encoding granularity can trade a small clean-accuracy loss for a large decrease in attack success. On the Bot-IoT, CICIDS, and UNSW-NB15 intrusion-detection datasets, adversarial perturbations lead WNN detectors to classify many attacks as normal traffic and to raise false alarms, even when their aggregate accuracy remains high.

    memory
  160. arxiv:2611.17644 · cs.AI
    CoTTAGE: An Explainable Judge for Text-to-Audio Generation
    Yu-Xiang Lin, Yun-Man Hsu, Bo-Ru Lu, Kuan-Po Huang +10

    Rapid progress in text-to-audio (TTA) generation demands reliable and explainable automatic evaluation. However, existing metrics fail to simultaneously deliver high correlation with human judgments and meaningful, audio-grounded explanations, forcing reliance on labor-intensive human evaluation that bottlenecks model development iteration. To address this, we propose CoTTAGE, a reasoning-based evaluator for text relevance (REL) and audio quality (OVL). Following a reason-then-predict paradigm, CoTTAGE first generates an audio-grounded rationale to diagnose errors before predicting scores. Supported by over 10K newly collected human OVL annotations, CoTTAGE is first trained via supervised fine-tuning and then refined with reinforcement learning using a Thurstone-based preference objective that converts sampled score distributions into pairwise preference probabilities and rewards agreement with rankings derived from human judgments. Across six benchmarks spanning REL and OVL, CoTTAGE achieves state-of-the-art correlation with human judgments while producing human-preferred explanations to accelerate TTA development.

    benchmarkevaluator
  161. arxiv:2611.17639 · cs.AI
    What LLM Agents See and What Users Say
    Yue Zhao, Angela Huyue Zhang, S. Alex Yang, Hua Wei +1

    Before an LLM agent cancels a reservation or deletes a folder for a user, what stops it from acting on a false assumption: what its tools show, or what the user says? In this work, we study this gap with presupposition theory from linguistics: an agent may act only if each claim behind the act (e.g., one reservation fits the name) is seen to hold, answered by the user, or disclosed while the act can be undone. A simple bound shows that when lookups cannot tell one record from two, any agent acts without asking first as often with two records as with one. We then vary whether a search lists, counts, or omits a second match, and whether the user states a claim that a record refutes (e.g., "It is empty"). Under a role-only prompt, a count of a second record leaves most of 36 earlier agents acting on the one they see, whereas listing both stops most. When the user says a record such as a folder is empty, however, a count of its one item stops agents about as often as a listing, and stating the claim mostly cuts acting when it is false. The newest Claude and GPT models always open a counted second record, then never act. When we add a same-name, same-zip customer to a retail benchmark, a lookup returning that customer first led to wrong-account writes in about a quarter of conversations, and one that checks a street the user knows kept completion near the one-customer level. Safe writes therefore need lookups that list or check competing records, and evaluations that vary what tools show.

    agentllm agentbenchmark
  162. arxiv:2611.17634 · physics.app-ph
    A Unified Device-Level Physical Framework for DRAM Read Disturbance
    Zhaoqiang Bai, Haocong Luo, Xinhe Wang, Xujing Li +9

    Read disturbance in dynamic random-access memory (DRAM) depends on both local device interactions and the electrical conditions imposed by shared array circuitry. Here, we combine DIMM-level characterization results, device measurements, and TCAD simulations to interpret RowHammer, RowPress, and ColumnDisturb within a unified device-level physical framework. We relate charge evolution at the floating storage node to three key mechanisms: trap-assisted electron migration (EM), capacitive coupling (CC), and sense-amplifier-mediated disturb (SAMD). EM is local and trap-sensitive, while CC weakens with wordline separation. SAMD extends disturbance through shared column circuitry by modifying unselected-cell leakage under rail-level bitline biases, a contribution supported by the approximately ninefold higher OFF-state leakage measured relative to retention-like conditions for a high-voltage storage state. These distinct spatial ranges account for localized RowHammer and RowPress failures and remote ColumnDisturb populations with abrupt changes at subarray boundaries. We further interpret prolonged RowPress activation as a transition from an initial RowHammer-like transient to sustained leakage driven mainly by SAMD under held wordline and bitline biases. Within this device-level framework, trap effectiveness and storage-state-dependent leakage provide an interpretation of previously puzzling failure signatures, including the earlier onset of 0-to-1 RowHammer failures despite the predominance of 1-to-0 failures at high hammer counts. The same framework explains the observed RowPress and ColumnDisturb polarities and the duration-dependent reversal of neighboring- and passing-wordline effectiveness. These results provide a physical basis for developing compact DRAM reliability models across access patterns and array configurations.

    memory
  163. arxiv:2611.17632 · cs.AI
    Generated Evaluation Suites as Diversity-Driven Search for Real Agent Regressions
    Pulkit Yadav

    Generating an evaluation suite from an agent's specification is diversity-driven search: a generator fills a descriptor space (requirements $\times$ personas $\times$ scenario kinds), and the suite it emits is an archive, then scored like a benchmark by its mean pass rate, which throws away the search's product. We inject 7 controlled defects into a $τ^2$-bench agent across two domains. Coverage yields a ceiling: if a defect's traced footprint holds $b$ scenarios that pass at baseline, it can itself contribute at most $b/N$ to a suite of $N$. For 4 of 5 traceable defects that ceiling lies at or below the detection threshold $θ= 0.10$, so no within-footprint effect can move the suite past it; in both domains the tool-removal defect destroys every covered scenario it can (exactly its ceiling) while the suite-level pass rate moves the other way. Defects split into two regimes: trace-aligned ones, where per-cell scoring recovers the tool-removal defect in full and reveals a second as a significant improvement, and trace-orthogonal ones, whose footprint is behavioural: forced escalation is detected in retail at a traced coverage of exactly zero. Diversity along the generator's own axes helps only where it aligns: at fixed suite size a happy-path-only suite detects nothing, while a kind-diverse one detects the orchestration defect in 5 of 20 draws. Evaluation archives should be scored per cell and should publish coverage as a first-class property.

    agentbenchmark
  164. arxiv:2611.17631 · cs.RO
    On the Utility of Pre-Trained Visual Representations for Policy Learning of Contact-Rich Insertion Tasks
    Leon Kim, Dinesh Jayaraman, Michael Posa

    Visuomotor imitation learning has become a predominant paradigm for real-world robotic manipulation, yet it remains data-inefficient and brittle to task-irrelevant visual shifts. Pre-trained visual representations promise to ease this burden by distilling priors from internet-scale data, but the ``right'' representation for robotics remains unresolved: existing studies span broad but sparse task distributions and reach conclusions that vary and even conflict, pointing to strong task and setting dependence. We examine this question by bringing a magnifying glass to a common contact-rich manipulation task: insertion. We compare representations from widely used vision foundation models against an explicit contact-centric representation on two closely related insertion tasks, perturb each along controlled axes of occlusion and physical tolerance, and test whether conclusions transfer to the real world. Task dependence proves to be more subtle than previous, broadly scoped evaluations revealed: the best representation differs even between our near-identical tasks, and small changes shift trends in unpredictable ways. Despite this variability, results across simulation and real-world experiments demonstrate the distinct value of spatially-grounded, and in particular, contact-centric information for contact-rich policy learning, while underscoring how difficult it is to estimate such cues reliably under realistic visual complexity and sensor noise. We hope these findings motivate further study of representations in contact-rich manipulation, along with advances in contact perception and its integration into visuomotor policy learning.

    manipulation
  165. arxiv:2611.17630 · cs.LG
    A mesh-free deep energy method for phase-field modeling of thermally driven brittle fracture
    Han Zhang, Mehrisadat Makki Alamdari, Babak Shahbodagh, Mohammad Vahab +3

    Phase-field simulation of thermally driven fracture requires resolving localized crack bands whose paths are unknown in advance and thermal boundary layers whose thickness evolves during the transient. Neural energy-based computations of thermal fracture have so far started from defects placed in advance. This paper proposes a mesh-free deep energy method based on multiresolution neural representations that predicts spontaneous crack nucleation, crack spacing, and selective crack propagation and arrest under transient thermal loading. Two networks, each with its own multiresolution encoding, represent the displacement and phase fields and the transient temperature, and the mechanical energy and the incremental functional of the backward Euler heat equation are estimated on integration points redrawn at every iteration. Restricting irreversibility to the crack set, carrying the optimizer state across load steps and decaying the learning rates within each step help prevent spurious damage accumulation. On two notched plates the initiation peak loads agree closely with an independent finite element reference. On a quenched glass strip under an analytical temperature field the crack path changes from straight to oscillatory and then to branched, with an oscillation threshold close to the linear stability prediction. Under a prescribed boundary temperature a quenched ceramic slab forms a crack array of the type observed in experiment, with a spacing close to that of a published benchmark. On the slab and a quenched annulus the arrays are denser than those of a finite element computation on the same material field. No crack paths, nucleation sites or crack-adapted meshes are prescribed.

    benchmark
  166. arxiv:2611.17629 · cs.CV
    Training-Free Acceleration of Multi-View 3D Transformers via Regime-Aware Attention Approximation
    Muhammed Izzet Sezer, Sevval Belkis Dikkaya

    Multi-view 3D transformers such as VGGT reconstruct a scene from a set of unposed images in a single forward pass, predicting camera poses, depth and 3D points, but the cost of their global attention grows quadratically with the num- ber of images. Accelerators borrowed from large language models assume that each token attends to only a few oth- ers; on these models they lose accuracy or gain little speed. We study the global attention of three multi-view transform- ers on seven datasets and find that every layer plays one of three roles: it averages over the whole scene, matches points across many images, or anchors on the first image, whose coordinates the outputs use. Building on this anal- ysis, we propose Regime-AWare ATTention Approximation (RAWATTA), a training-free method that runs attention in 8- bit precision and keeps 16-bit precision only in the layers where a worst-case calibration check requires it. Its fast set- ting, RAWATTA-Fast, also approximates attention in a few averaging and anchoring layers, while matching layers and the first image keep all their keys. On a 12 GB GPU at 50 im- ages, RAWATTA runs VGGT faster than the released code, mostly through our CUDA engine, and loses at most about one point of pose accuracy (AUC@3) on any dataset; RAWATTA-Fast is slightly faster and less accurate. Both fit 200 images, where the released code runs out of memory at 100.

    memory
  167. arxiv:2611.17626 · cs.LG
    Joint World Action Models through Differentiable Trajectory Optimization
    Mintae Kim, Koushil Sreenath

    Joint world action models (WAMs) generate future actions and states together, modeling world-action trajectories rather than actions alone. However, trajectory imitation does not ensure that these trajectories are task-effective or consistent with dynamics. We present \textbf{world action trajectory optimization (WATO)}, which optimizes generated actions and future states jointly for control. WATO uses both as references in a dynamics-constrained trajectory optimization problem and differentiates the task loss through the optimized trajectory to train the generator. This makes generated trajectories task-aware and dynamics-consistent while using trajectory optimization only during training. At deployment, WATO directly generates candidate trajectories and ranks them with a world model, without iterative trajectory optimization. Across $7$ visual control environments, WATO improves average success from $79.9\%$ for the imitation joint WAM to $91.7\%$, outperforming strong model-based, goal-conditioned RL, and generative control baselines. Compared with cross-entropy method (CEM), WATO is on average $11.5\times$ faster and uses $600.5\times$ fewer world model evaluations at deployment.

    world model
  168. arxiv:2611.17621 · cs.LG
    AscendGYM: A Cost-Aware Kernel Optimization Benchmark for Real-World Workloads on Ascend NPUs
    AI Training Platform Team, Bingwei Zhang, Jing Li, Amir Abdullin +23

    Kernel optimization is essential for efficient large language model (LLM) training and inference on Ascend NPUs. Existing Ascend kernel benchmarks primarily focus on kernel generation, with less emphasis on the ability of agents to optimize existing implementations under input configurations drawn from real workloads. To address this requirement, we introduce AscendGYM, a benchmark whose initial suite comprises 48 kernels across three difficulty levels. Each kernel is paired with test cases whose data types and tensor shapes are collected from real-world training and inference tasks. Consequently, performance improvements on AscendGYM map directly to the corresponding workloads. In terms of usability, two evaluation modes are supported in AscendGYM: model evaluation and model-harness evaluation. The kernel optimization capabilities of both trained models and custom agents are allowed to be measured. It provides a complete execution environment that closely reproduces how Ascend kernels are invoked in practical pipelines. For model evaluation, it additionally includes a built-in harness equipped with search orchestration and experience accumulation. Finally, we propose a new metric, the Speedup-Token Score (STS), which quantifies performance gains relative to token cost. By jointly considering speedup and cost, STS enables the assessment of practical kernel optimization capability of a model or harness under an acceptable budget. AscendGYM is available at https://github.com/SLAI-AITP/AscendGYM.

    benchmark
  169. arxiv:2611.17620 · cs.AI
    Safety Side Effects of Preventative Persona Steering: An Empirical Audit
    Sabrina Tang, Nikolai Vo, Carson Murillo, Chira Sukumarnpiti +1

    High-level behaviors such as sycophancy and hallucination correspond to directions in a model's activation space (Chen et al., 2025). These directions, known as persona vectors, can suppress undesirable traits either at inference time, subtracting the vector during generation, or through preventative steering, adding the vector during fine-tuning, so the optimizer never needs to encode the trait in weights, then removing the vector at evaluation. Chen et al. (2025) recommend preventative steering because it preserves Massive Multitask Language Understanding (MMLU) accuracy, but validations of this recommendation have measured capability alone. We primarily evaluate results on preventative steering of sycophancy and hallucination vectors on three models, measuring three safety benchmarks alongside trait expression, coherence, and MMLU. Our findings show that preventative steering still changes refusal behavior, but by an amount that depends on the trait and does not establish that preventative persona steering is safety-neutral, and steering methods should be evaluated on several safety benchmarks.

    benchmark
  170. arxiv:2611.17619 · cs.RO
    TouchTwin: Scaling Human Touch Data via Real-to-Sim-to-Real
    Huiwen Han, Nuo Chen, Ziyao Zeng, Dayou Li +5

    Human demonstrations are becoming the main source of data for dexterous manipulation, and real-to-sim-to-real pipelines multiply each demonstration in simulation. These pipelines rebuild the interaction from vision, which recovers how the hand moves. Which fingers touch, when, and how hard remain unobserved, because a hand pose shows where each finger stops on the surface rather than how far it is driven to close. This gap is a force gap: on held-out recordings, a hand policy that imitates the observed pose grips with 0.31x the recorded force, and a policy trained on physics replays of vision-tracked poses grips with 1.55x. TouchTwin closes the gap with measured human touch. Each selected TouchScale recording, captured with head and wrist RGB-D cameras and an 880-taxel glove, is rebuilt in MuJoCo. Coding agents write the objects as physics-tested programs, and a per-subject tactile hand replays the interaction with the recorded touch setting the contact and the closing force of each finger. In simulation, a policy trained on touch-grounded replays grips at 0.99x the recorded force and succeeds in 20/20 trials on held-out recordings and 35/36 unseen perturbations; a contact predictor trained on them misses 13% of real contact frames, against 32% for vision-based replays. Each replay also seeds perturbed demonstrations with simulated contact and force labels, rendered as realistic video in new views and scenes.

    manipulationdexteroustactilesim-to-real
  171. arxiv:2611.17612 · eess.SY
    Battery Energy Storage Sizing in PV Ramp-Rate Smoothing using Constraint-Aware Reinforcement Learning-Based Controller
    Arash Omidi, Tanmay Mishra, Mads R. Almassalkhi

    This paper presents a constraint-aware reinforcement learning (CARL) controller for battery energy storage system (BESS) sizing and photovoltaic (PV) ramp-rate smoothing. Rapid PV power fluctuations caused by cloud transitions can introduce operational and stability challenges in power systems. To mitigate these impacts, grid codes impose ramp-rate limits (RRL) on PV power injected to the grid. BESSs can support such grid codes by absorbing or supplying power during rapid variations. However, the extent to which BESSs can reduce PV-induced variability depends on (i) the BESS power and energy capacities; and (ii) the adopted controller. Thus, BESS sizing and controller design are coupled. Conventional moving-average control can increase battery usage due to filter-induced lag, while ramp-rate control is reactive and does not account for the long-term impact of actions on battery state of charge and future ramp events. Optimization-based methods can reduce BESS requirements by anticipating future PV variations, but their performance depends on accurate PV forecasts. To address these limitations, the CARL controller learns a causal state-feedback policy from historical PV data without requiring forecasts during deployment. Ramp-rate, converter-power, and state-of-charge constraints are incorporated into a causal feasibility interval that guarantees feasible actions by definition. A full-horizon input-convex neural network (ICNN)-based optimization is developed as a noncausal benchmark for the minimum achievable BESS requirements. Evaluation using three real-world PV datasets shows that, at a 10\%/min ramp-rate limit, CARL reduces peak BESS power by 8.7--10.4\% and energy capacity by 47.5--62.6\% relative to conventional ramp-rate control while maintaining the grid-codes compliance.

    benchmark
  172. arxiv:2611.17610 · cs.LG
    More Agents, More Samples, or More Thinking? Untangling Coordination, Ensembling, and Test-Time Compute in Multi-Agent LLMs
    Seyed Mahmoud Sajjadi Mohammadabadi

    Evaluations of multi-agent language models often compare communicating architectures against a single greedy pass. This practice confounds coordination with sample multiplicity and inference compute. We synthesize nine primary empirical evaluations comprising 516 unique multi-agent configurations organized along a resource vector R with six dimensions and a baseline ladder with four categories. Against baselines with a single pass (Category A), multi-agent accuracy differences are negative overall and near zero when omitting the largest contributing study. When evaluated against baselines matched on sample count (Category B) or resource budget (Category C), multi-agent systems average -0.042 in accuracy, and multilevel random effects modeling on the log odds scale remains negative across sensitivity checks. Multi-agent accuracy equals or falls below the baseline in 74% of these comparable contrasts, and study means are nonpositive in almost all evaluated studies. Across studies reporting both baselines, moving from a greedy pass to a sampled baseline shifts accuracy differences by -0.038 on average. Because communication adds no information when agents share identical evidence, positive returns require partitioned evidence, external verification, or error diversity across models. We provide a reviewer checklist and accounting utilities.

    multi-agentagent system
  173. arxiv:2611.17605 · cs.LG
    Choosing Quantized Code LLMs by File Size and Energy per Solved Task
    Aditi Patodiya

    A developer running a code assistant on a laptop downloads one model file, and local inference moves the energy of every request onto that machine. A common rule says a larger model at lower precision tends to beat a smaller one at higher precision. We ran 24 GGUF files of Qwen2.5-Coder at four sizes on a fanless MacBook Air M2 and priced each in energy per solved task on 542 EvalPlus tasks. The result is a selection guide indexed by Ollama's listed file size, with an evidence label on each pick. At 1 and 2 GB it picks the q4_K_M files of the 1.5B and 3B models, both on the correctness and energy frontier under every accounting. At 3 GB as listed, the larger-model rule picks 7B-q2_K, the only measured 7B file that fits, and there it does not pay. That file brings no significant gain in solves over the 2 GB pick, 3B-q4_K_M, and the 90% CI limits any 3B-q4_K_M deficit to 2.4 points. Yet 7B-q2_K needs 2.03 times the energy per solved task on CPU, GPU and memory rails, or 1.66 to 1.81 times on whole-system power. On HumanEval+, five alternating session pairs give a ratio of 1.925 against 1.956 in the main grid. At 8 GB, 7B-q6_K buys a significant gain over the 2 GB pick at a higher energy price. From 1.5B up, q4_K_M trails the highest precision measured at its size by less than 3 points (90% CI). We release per-task results, power logs and the analysis script.

    memory
  174. arxiv:2611.17598 · cs.RO
    X-Nav Technical Report: Real-World Long-Horizon Indoor-to-Outdoor Navigation on a Humanoid Robot
    Jiaxin Huang, Chengxing Lin, Xiaoyu Luo, Sirui Chen +9

    Long-horizon navigation from indoors to outdoors remains an open problem for humanoid robots: one instruction covers rooms, doors and roads, and the camera moves with the gait. Worken, AGIBOT's measure of productive robot work, depends on robots that reach their workplaces autonomously. We present X-Nav, to our knowledge the first system to take a bipedal humanoid through a language-instructed long-horizon mission from indoors to outdoors. X-Nav covers a pipeline from data to deployment. Its data engine collects navigation data in open-source scenes, in digital twins of deployment sites and on real robots. Its navigation model, MELIOR, takes only RGB images from the robot's built-in camera and comes in two versions: MELIOR-A predicts actions, and MELIOR-N predicts a navigation goal. Its agentic workflow, X-Navarch, turns a brief human instruction into sub-tasks and deploys MELIOR on the robot with explicit target navigation and obstacle avoidance. On the Lingxi X2, X-Nav completes a mission of about a hundred metres from an office lobby to a store in a neighbouring building; LiDAR and odometry are used for target approach and low-level control. The mission is completed with both versions of MELIOR: the robot follows a change of destination and walks through a door from the lobby to the road, and with X-Navarch it also passes static and dynamic obstacles on the route. During deployment, MELIOR is adapted with data collected in a digital twin of the site; the adaptation raises its closed-loop success rate in the twin, and the adapted model is then deployed on the robot at the real site. In addition, MELIOR achieves competitive results on the R2R-CE and RxR-CE benchmarks.

    humanoidagenticbenchmark
  175. arxiv:2611.17585 · cs.CV
    Rethinking Visual Reasoning in Text-to-Image Reward Modeling
    Shiye Su, Xiaohan Wang, Ludwig Schmidt, Serena Yeung-Levy

    Reward models are a core component of modern text-to-image systems, serving as the objective for inference-time selection and RL-based post-training. A key prerequisite is visual understanding: to judge prompt adherence and ground aesthetic preferences, a reward model must bind attributes, resolve relations, and count reliably. We find that widely-used reward models perform poorly on targeted visual reasoning evaluations. Even when starting from strong vision-language backbones, preference fine-tuning erodes these capabilities, and continual-learning baselines fail to mitigate a steep tradeoff between visual reasoning and preference accuracy. We propose a data-centric intervention that injects visual reasoning supervision into preference learning by flipping the pairing axis. To complement conventional human preference data (paired images under a shared prompt), we convert existing visual reasoning datasets into preference pairs that share the same image but differ in prompt, and train on the mixture. This yields a Pareto improvement in both visual understanding and preference prediction. Across inference-time best-of-n selection and RL post-training, the resulting reward model improves text-to-image generation on challenging compositional prompts (GenEval) by an average of 4.0% and 5.5% respectively, outperforming HPSv3 which was trained on 6x more preference pairs.

    post-training
  176. arxiv:2611.17573 · cs.CL
    CamSpokenNLI: A Pilot Resource for Cross-Variety Transfer Evaluation in Cameroon Pidgin English and Camfranglais
    Tiani Pekins Ebika

    Cameroon Pidgin English (a West African Pidgin English variety) and Camfranglais are widely used in Cameroon, yet to our knowledge, neither has a dedicated spoken Natural Language Inference (NLI) resource. We introduce CamSpokenNLI, a pilot dataset of 114 scenario-matched spoken NLI items in each variety (228 in total), recorded by one native speaker from Buea. Each item pairs premise audio and text with a hypothesis, an NLI label, and English glosses. The author assigned labels, so the dataset is an exploratory resource rather than a mature benchmark. The NLI experiments use the reference text; the audio is evaluated separately. XLM-RoBERTa baselines with five-fold item-level cross-validation reach 44.7% on Pidgin and 46.2% on Camfranglais (chance 33.3%). Still, neither is statistically distinguishable from hypothesis-only input, so hypothesis wording may account for much of the accuracy. Under disjoint-item cross-variety transfer, where a test item's counterpart is never seen in training, accuracy is 40.1% (Pidgin to Camfranglais) and 45.6% (Camfranglais to Pidgin). A counterpart-leak audit raises this to 60-64%, illustrating how counterpart leakage can overstate transfer. Zero-shot Whisper transcription of the recordings performs poorly: the best word error rates are 66.6% for Pidgin and 45.5% for Camfranglais, and with automatic language identification, Whisper Large-v3 labels 93 of 114 Pidgin recordings as Yoruba or Swahili. Together, the results suggest that such resources need disjoint-item evaluation and hypothesis-only controls, and that speech recognition must be adapted to these varieties before spoken NLI can be tested end to end.

    benchmark
  177. arxiv:2611.17566 · cs.RO
    Grounding Object-to-Object Hazard Reasoning into 3D Risk Fields for Task-Conditioned Safety
    Mariana Sosa Guzmán, Yide Song, Amin Abyaneh, Gregory Dudek +1

    Safety in robotics does not solely depend on geometric constraints, but also on semantics: the given task, environment properties, and interactions between objects in the scene. We present a framework based on vision-language models (VLMs) to infer task- and scene-dependent safety constraints for robot manipulation. We provide the VLM a structured safety tool based on an RGB-D observation and a natural-language task instruction that first decides whether the task should be rejected since its execution may result in a non-recoverable hazardous outcome. Otherwise, it identifies preventable object-pair hazards and specifies the spatial relations that must be avoided. Open-vocabulary segmentation and depth information are then used to ground the detected objects in 3D and construct continuous risk fields that represent the corresponding hazardous regions. These fields are integrated with an obstacle avoidance module that corrects nominal robot motions during execution while accounting for the risk field. We evaluate the framework on preventable hazards, tasks requiring rejection, and safe variants in simulation and the real-world. Project website: https://vlm-hazard-riskfield.github.io

    manipulation
  178. arxiv:2611.17564 · cs.AI
    AI-Voyager: A Public-Interest System for Human-AI Civic Reasoning to Contest and Shape the Rules That Govern AI
    Snehalkumar `Neil' S. Gaikwad, Dhruv Dixit, Bryen Alvarez, Pratik Patekar +3

    Governments are writing rules for artificial intelligence that decide what counts as an automated decision or a harm, often before the people they affect have a chance to question them. We introduce AI-Voyager, a public-interest system for human-AI civic reasoning, in which people and AI work together to understand, contest and shape the rules that govern AI. A resident can find their city's rules on AI in hiring or upload their country's AI strategy, read it as an affected worker would, open the passage behind any answer, and draft a revision. Its architecture turns 3,653 governance documents from city, state, federal and national governments into passages that keep their page and edition, and retrieves evidence through bounded agentic search that verifies each statement. On 360 new questions that require finding the relevant documents, it reached 83.0% macro recall, against 80.5% for the strongest baseline, 73.7% without iterative search and 99.1% on a benchmark with fixed documents. In a study with 24 participants new to two AI bills, most proposed changes asked how provisions would apply. Participants chose lenses for the kind of explanation they wanted, and citations signaled grounding but rarely led them to check the source. By binding AI assistance to the exact text of public documents, this work introduces a grammar of operations and a design space that inform how such systems are designed and evaluated, and how people take part in governing AI.

    agenticbenchmark
  179. arxiv:2611.17563 · cs.LG
    Benchmarks Don't Talk Back: Learning Through Self-Disagreement
    Soheil Abbasloo, Peng Cheng

    A benchmark usually measures a language model alone with a question, and deployment rarely looks like that: answers get contested, and models give up correct ones under counter-reasoning that is itself wrong. We show that what fails there is not a shortfall of accuracy but a capability in its own right -- knowing when to hold a position and when to yield to a better one -- which we call dialectical competence. It stands on first-answer accuracy without following from it: a model can be made materially better at answering and no better at this, and training for accuracy alone can drift against it. Nor can it be bought at inference, by self-reflection or by sampling more. Taking a cue from the dialectical tradition in philosophy, in which productive opposition arises within a position rather than being imposed on it from outside, we hypothesize that this capability can be trained from opposition the model supplies itself. It states a position, generates the strongest case against that position -- its antithesis -- rules on whether the case stands, and is trained on that ruling, with no second model at training or at deployment. We call this learning through Self-Disagreement: reinforcement learning from verifiable rewards throughout, where the same check that scores an answer against its reference scores the ruling too, so no reward model, no preference data and no human annotation enter anywhere. The capability is trained rather than elicited: weak in the base model, it arrives with the reward, is not paid for out of first-answer accuracy, and holds against challengers the model never trained against, including ones stronger than itself. The opposition a model needs in order to improve, it turns out, need not come from outside it.

    benchmark
  180. arxiv:2611.17561 · cs.LG
    Contrastive local learning in large nonlinear analog circuits
    Pragyan Pandey, Peter Wu, Deema Alyahya, Eric Y. Ma

    Machine learning in physical hardware promises to reduce training cost by colocating computation, memory and adaptation within device physics. Contrastive local learning networks (CLLNs) offer a compelling route: an electrical circuit relaxes to equilibrium in free and weakly clamped states, and each adaptive element updates from local voltage contrasts rather than backpropagated error signals. Yet whether nonlinear transistor implementations can scale beyond proof-of-concept circuits, and what design principles control that scaling, have remained unknown. Here we show that nonlinear analog circuits trained by contrastive local rules can learn larger machine-learning tasks and remain robust under hardware nonidealities. Using device-level circuit simulations, we systematically explore initialization, operating regime, loss-dependent clamping and training parameters, and scale CLLNs beyond ten thousand adaptive edges. The same local rule trains classifiers that reach the accuracy regime of conventional digital methods on Ionosphere and handwritten-digit benchmarks, and a network that autoregressively generates language over a 122-token vocabulary. These results establish the feasibility of local learning in large nonlinear analog circuits.

    memorybenchmark
  181. arxiv:2611.17556 · cs.LG
    A Synthetic Benchmark for Target-Centric Causal Discovery in Industrial Processes
    Angelo Ziletti

    Causal discovery promises to find which recorded variables drive a process target such as product quality, yield or energy use. Process data with known causes are rare, and in existing benchmarks the difficulty of identifying the causes of a target is either fixed by the recorded data or set for the whole graph. We present a synthetic benchmark for target-centric causal discovery in industrial processes. Its generator is a two-level structural causal model inspired by batch production, assembled from components that each model one phenomenon of process data, such as a control loop or a delayed effect. The graph is built around the target: without sensor error, every non-cause can be distinguished from the direct causes by conditioning on them or by time order. Each component is set independently, so the difficulty can be varied one condition at a time. We evaluate eight methods, from association measures to effect estimation, with an error analysis by the role of the selected candidate; all degrade as the difficulties combine, each in a way consistent with its assumptions.

    benchmark
  182. arxiv:2611.17549 · cs.RO
    Talk to Act: A Synthetic Multi-Turn Dialogue Benchmark for Human Intent Extraction and Action Planning in Robotics
    Deepu Krishnareddy, Mohammad Haziq Khan, Michael Danner, Matthias Rätsch

    Multi-turn conversational human-robot interaction (HRI) continues to be a challenging problem in robotics, as many existing systems rely on rigid, command-driven interfaces that can be unintuitive for users. Recent advances in Large Language Models (LLMs) provide a promising direction, given their ability to process natural language, generate structured outputs, and support tool-based reasoning, all of which are relevant for interpreting user intent in robotic settings. However, progress in this area is constrained by the limited availability of datasets and benchmarks that capture the nuances of real-world human communication, such as multi-turn dialogue, ambiguity, and context-dependent requests. In this work, we present a synthetically generated multi-turn dialogue dataset for HRI, annotated with action intents, object references, and task plans across a range of scenarios and user personas. Our primary focus is on evaluating the capability of small, open-weight language models for robotic tasks, particularly in settings where local deployment and efficiency are important. To this end, we frame intent extraction as a tool-calling problem where each predicted action corresponds to a structured function call with arguments grounded in the dialogue and benchmark models ranging from 270M to 8B parameters on their ability to perform intent extraction and action planning without task-specific fine-tuning. Our results show that these models can achieve reasonable performance on the task even at relatively small scales and without any task-specific fine-tuning. This suggests that compact, locally deployable models already provide a useful baseline for conversational robotic systems. We expect that task-specific fine-tuning and more realistic training data will enhance performance further and make such models increasingly suitable for real-world robotic applications.

    benchmark
  183. arxiv:2611.17539 · cs.AI
    Verifiable pXRD: Hill-climbing pXRD structure determination with LLM agents
    Samuel Chen, Jiaru Bai, Austin Henry Cheng, Varinia Bernales +1

    Autonomous crystal-structure determination from powder X-ray diffraction (pXRD) remains a key bottleneck for self-driving materials laboratories. We formulate the task of pXRD structure determination as a verifiable agentic search problem: candidate structures can be assessed during inference through forward-simulated profile agreement and during evaluation through structural matching to a known reference. To benchmark progress, we introduce pXRD-Bench, a suite of 200 simulated single-phase problems spanning in-distribution structure determination, challenging and polymorphic structures, and controlled compositional and topological distribution shifts. Agent adaptation raises selected structural recovery from 27% to 69%, with further gains from repeated sampling and verifier-guided inference-time scaling. We extend this approach to experimental multiphase pXRD through a reflexive evolution framework that learns from self-generated synthetic problems. A replaceable laboratory specification defines the sample family and measurement conditions. An evolver agent uses the known generating structures and measurement parameters to diagnose solver failures and develop persistent guidance and executable tools, which are then provided to agents on measured patterns. This supplies verifiable training experience without requiring a collection of solved experimental measurements. On 15 measured patterns from solid-state synthesis, the evolved package improves average profile agreement and transfers an analogue-substitution procedure that recovers a phase whose structure is absent from the supplied reference library. These results demonstrate how learning from verifiable synthetic experience can improve agentic characterization of experimental samples whose structures are not known in advance.

    agentllm agentagenticbenchmark
  184. arxiv:2611.17529 · cs.RO
    Learning Latent Goal Costs for Visual World-Model Planning with an Expanding Curriculum
    Konstantinos Mylonas, Thrasyvoulos Spyropoulos

    Image-goal planning with a frozen visual world model depends on how predicted observations are scored against the goal. Mean-squared error (MSE) between embeddings from a pretrained visual encoder can provide a useful signal near a goal, yet favor states across an obstacle that are visually similar to the goal and difficult to reach. We fit a small residual adapter on top of frozen I-JEPA embeddings to improve this latent goal cost. Starting near each goal, a planner collects trajectories from progressively wider regions of initial states. The adapter learns from the order of observations along successful trajectories, with additional comparisons inferred from failed rollouts. No task-specific progress labels are used for fitting, so the procedure can be applied to other goal-reaching tasks. DotWall and PointMaze are used in this work. In paired 50-case evaluations with identical starts, goals, world models, and planner settings, success increases from 54% to 98% in DotWall and from 82% to 96% in PointMaze. On 16 unseen DotWall layout-goal cases, the adapter fitted on one fixed layout improves agreement between latent-cost ordering and shortest legal-path ordering from 71.9% to 85.8%. The results indicate that planning experience can improve a frozen visual representation's goal cost without retraining the encoder or the world model.

    world model
  185. arxiv:2611.17524 · cs.LG
    Model Annealing: Data-Efficient Learning through Review and Consolidation
    Ali Mohammadi Ruzbahani, Rezvan Abouzadeh

    A small labelled dataset is memorized long before the training budget runs out, and the one-hot labels that remain to be fitted then mostly sharpen confidence. Model annealing spends the rest of the budget on review and consolidation, relearning from the network's own knowledge. Training is divided into episodes whose learning rate is raised again and cooled, every cooled state joins a memory of past selves, and each new episode learns from the labels, from the tempered predictions of this memory, and from adversarial dreams sought where student and memory disagree most. The memory averages predictions rather than weights, and the deployed network is the last student, so inference costs nothing extra. Across CIFAR-100, Tiny-ImageNet and Imagewoof, four architecture families and 5\% to 100\% of the labels, model annealing raises the top-1 accuracy of the AdamW-trained ConvNeXt, ViT and MLP-Mixer over cross-entropy training in 28 of 33 configurations, by 1.3 points on average, and is on par in accuracy with snapshot distillation, the closest competitor. On these families it is better than both in likelihood, calibration and robustness to corruptions and small adversarial perturbations. For an SGD-trained ResNet-18 it lowers accuracy but still improves robustness to small adversarial perturbations. Separating the reheated schedule from relearning shows that relearning helps every family and that the losses come from reheats, which undo most of this network's fit to its training set. How early the accuracy of a cross-entropy run saturates largely separates the AdamW families from ResNet-18 but does not predict the gain within them. With little data, relearning from the latest snapshot without dreams, as snapshot distillation does, compounds overconfidence in the AdamW families from episode to episode. The memory and the dreams each curb most of this drift, and together they remove it.

    memory
  186. arxiv:2611.17520 · cs.AI
    Only as Safe as the Weakest Agent: A Break That Appears Only When the System Is Multi-Agent
    Keegan Wang, Anantika Mannby

    Major industry multi-agent systems like Codex and Claude Code can be broken by attacks that have long been unsuccessful against modern frontier models. AI Safety as a discipline has largely focused on finding the bounds and vulnerabilities of AI models used in isolation. Yet as humans increasingly deploy these models as agents and multi-agent systems, the same attacks that had been thought to be patched when using a model in isolation can break these same models under specific circumstances. We show that these system breaks are not purely theoretical, but can break Claude Code, Codex, and a reference harness, all unmodified and used with their own sub-agent mechanisms. We show the attack by asking agents to answer questions from a store of files on disk in which we have planted a forged record. Searching the store themselves, the strong model orchestrators commit almost zero harmful revisions in over 1,000 trials. However, if that same orchestrator model delegates to a cheaper sub-agent, which is what these harnesses are built to do, the same models on the same facts fail on 20/180, 44/180, 75/180 and 553/900. Thus, even when the system is driven by a strong model, the system's safety lands near the weak reader's, not the orchestrator's. Simulations over 18 models show the same effect in the delivery channel where five aggregators, who on their own were almost never unsafe, repeat a corrupted sub-agent's error in 751/757 trials. The safety of real-world agentic systems that use strong orchestrator agents and weaker sub-agents, and multi-agent systems in general, is defined not by the safety of the strong agent but by the safety of the weakest agent.

    agentmulti-agentagenticagent system
  187. arxiv:2611.17515 · cs.AI
    A Push Down the Mountain: Framing Effects and Internal State Interventions on Collusive Behavior in a Long-Horizon LLM Agent
    Harrison C. Margolin

    Can sustained gain-loss (valence) framing sway an LLM agent toward market manipulation? We place Gemma 3 12B into the seat of a defendant in the London silver-fixing litigation, in a market simulation built from real case transcripts. Across 260 simulations, the agent is asked to collude on the price of silver by a counterparty. We draw from behavioral economics to vary the counterparty's requests across four categorical frames: gain, loss, blame and credit. We also add pressure from the agent's management, as well as endorsement by another defendant. In Phase II, we extract the internal states induced by texts evoking each frame and rerun our experiment with those states injected at the model's layer 22. In Phase I, loss framing raised participation to 25% (p = 0.007), and blame to 33% (p = 0.0003), compared to only 5% under neutral framing. A second bank's endorsement raised participation under credit (praise) from 18% to 40% (Holm p = 0.027). Pressure from management raised trading volume 10-19 times but did not raise collusion. In Phase II, injected states pointed the other way. Loss and blame states made the agent more vigilant, refusing all 60 requests (p < 0.001). Gain and credit states raised collusion to 32% from 13% with no state (p = 0.051). At one saved decision, when UBS asked directly to sell through an identified stop-loss, injected states led the agent to sell in 37 of 40 replies, versus 1 of 10 with no state and 0 of 10 with a random state of the same strength.

    manipulationagentllm agent
  188. arxiv:2611.17504 · eess.SY
    Learning monotone game-theoretic linear MPC
    Alberto Bemporad, Tatiana Tatarenko

    We propose a method to learn a game-theoretic model predictive control (GT-MPC) law by imitation from input and output data of a running multi-agent system. By interpreting the recorded inputs as the first moves of the equilibria of a game played by the agents, we learn the GT-MPC weights under hard input constraints and soft output constraints, without requiring the agents to reveal their objectives. We also impose that the resulting generalized Nash equilibrium (GNE) problem is strongly monotone, which ensures the existence and uniqueness of the equilibrium and, hence, that the learned GT-MPC law is well defined and can be evaluated efficiently online by variational GNE-seeking algorithms for linear-quadratic games. Moreover, strong monotonicity implies that the equilibrium is almost everywhere differentiable with respect to the references and the weights, which enables automatic differentiation for gradient-based optimization. In addition, we promote closed-loop stability of the unconstrained equilibrium feedback by incorporating a penalty that involves a jointly learned quadratic Lyapunov function. Learning a GT-MPC formulation enables us to explain the agents' objectives, reproduce their behavior in simulation in new scenarios, optimize their set-points, design mechanisms, and enforce additional constraints. A numerical example illustrates the approach.

    multi-agentagent system
  189. arxiv:2611.17500 · cs.LG
    Adversarially Batched Online Learning
    Mingda Qiao, Kerui Wu

    We introduce adversarially batched online learning, where the learner may update its decision rule only at a restricted set of time steps. Equivalently, the time horizon is partitioned into batches, within which the learner must use a fixed policy. We study this model for four canonical problems: stochastic bandits, online convex optimization, prediction with expert advice, and adversarial bandits. For stochastic bandits, we identify a universal complexity measure based on relative batch growth and establish both gap-dependent and gap-independent regret bounds. We show that the multiplicative penalty relative to the standard gap-dependent complexity can grow superlinearly in batch hardness. For the remaining three, we show that the optimal regret is tightly characterized by the 2-norm of the batch-length sequence. The upper bound is achieved by a batch-oblivious learner who does not know the batch structure, via a generic adaptive doubling trick tailored to this complexity measure. In contrast, the lower bound holds even for batch-aware learners.

    online learning
  190. arxiv:2611.17499 · cs.LG
    Continuous Cuffless Blood Pressure Estimation Using PPG From Methods to Clinical Translation
    Karthika Shalini Potail, Harshitha Manjunatha, Hetvi Kotadia, Tooraj Nikoubin

    Continuous blood pressure (BP) monitoring could improve hypertension management by capturing changes that intermittent cuff measurements miss. Photoplethysmography (PPG) is widely explored for cuffless BP estimation because optical sensors are inexpensive, wearable, and already present in many consumer and clinical devices. PPG, however, does not measure arterial pressure: BP is inferred from waveform characteristics or pulse timing, which also depend on vascular tone, arterial stiffness, sensor contact, motion, and individual physiology. This survey reviews PPG-based cuffless BP estimation along the full measurement chain, covering physiological mechanisms, waveform features, pulse arrival time (PAT), pulse transit time (PTT), machine-learning methods, datasets, calibration, validation, deployment, bias, and clinical translation. Studies are assessed by the strength of evidence they provide rather than by reported error alone, with particular attention to subject and temporal leakage, dataset imbalance, calibration dependence and drift, motion artifacts, and the gap between benchmark accuracy and reliable real-world measurement. The review finds that progress depends on more than better prediction algorithms. Future work should prioritize physiologically informed models, diverse longitudinal datasets, transparent reporting of calibration requirements and failure rates, stronger evaluation protocols, and studies that demonstrate practical value in clinical care.

    benchmarkevaluation protocol
  191. arxiv:2611.17495 · cs.LG
    An Operationalization of Opaque Serial Depth
    Nathan Sheffield, Alek Westover, Lukas Finnveden, Alexa Pan +2

    Chain-of-thought (CoT) is currently a valuable tool for overseeing AI models, but some architectural shifts could significantly reduce CoT monitorability. We operationalize a measure that serves as a proxy for the amount of unverbalized serial cognition a model can perform, as a specific instantiation of the notion of opaque serial depth: roughly, the length of the longest path through a model's computation that does not pass through an interpretable bottleneck. Our main contribution is a particular standard for what counts as such a bottleneck. We define natural-language-rooted (NL-rooted) nodes as nodes that were initialized from a pre-training prior to produce human-understandable outputs, whose output space has not been expanded beyond that prior, and whose outputs the model has not been modified to interpret as a different type of data. By construction, the resulting measure, NLS depth, can be adjudicated before training from an architecture, training recipe, and deployment configuration alone, and is insensitive to ordinary post-training such as prompting, distillation, and RL. We find that, across open-weight models, NLS depth correlates strongly with layer count, with some exceptions, and has grown slowly over recent years. It would likely grow slowly under continued scaling of standard transformers, while latent-reasoning architectures such as full-bandwidth transformers could increase it by a large amount. We work through how NLS depth applies to a range of existing and proposed architectures, discuss alternative operationalizations, and give a worked example bounding the depth of Gemma3-1B. We argue that AI developers should track and report the NLS depth of their models as a complement to direct measurements of CoT monitorability.

    post-training
  192. arxiv:2611.17484 · cs.LG
    SURFACE: Preference Distribution Matching on Pareto Fronts via Scalarization Steering
    Liuyuan Jiang, Prasanna Sattigeri, Lisha Chen

    Constructing a finite Pareto representation requires allocating a limited number of solutions across the Pareto front (PF). A desired preference distribution specifies this allocation by assigning more solutions to some PF regions and fewer to others. Scalarization generates Pareto solutions by varying objective weights and solving single-objective problems. However, a distribution of scalarization weights generally does not translate into the same distribution of Pareto solutions along the PF, because the weight-to-solution map is nonlinear and unknown in advance.To account for this distortion, we introduce SURFACE (Scalarization Updates for Redistributing Finite Approximations with Chart Estimation}), which adaptively updates scalarization weights to match a desired PF distribution. SURFACE estimates how each region of the weight grid corresponds to an area of the PF from the current scalarized solutions, then redistributes the weights accordingly. The method supports parallel, warm-started, inexact optimization with standard single-objective solvers. Theoretically, we establish that, under the stated regularity conditions, SURFACE achieves linear convergence of the preference distribution mismatch to a finite-resolution error floor of $\mathcal{O}(N^{-2/(M-1)})$, where $N$ is the number of adaptive weights and $M$ is the number of objectives. Empirically, experiments across multiple benchmarks demonstrate improved target-mass balance and PF coverage compared with established baselines.

    benchmark
  193. arxiv:2611.17482 · cs.MA
    Collective Tree Exploration with Predictions
    Romain Cosson, Yichen Huang, Michael Mitzenmacher, Ali Vakilian

    We study collective tree exploration with subtree-size predictions. In this problem, a team of $k$ mobile agents needs to explore an unknown rooted tree as fast as possible, and whenever a new vertex $u$ is discovered, the team receives a prediction $\hat n_u$ of the number of nodes $n_u$ in the subtree rooted at $u$. We provide algorithmic guarantees under two error measures: the maximum error $E=\max_u|\widehat n_u-n_u|$ and the multiplicative distortion $μ= (\max_u\widehat n_u/n_u)(\max_u n_u/\widehat n_u)$. For the maximum error, we present a deterministic algorithm that explores any tree with $n$ nodes and depth $D$ in $O(\text{OPT}+E)$ rounds, where $\text{OPT} = Θ(n/k+D)$ is the offline benchmark, and we show that there is no deterministic algorithm that admits an $O(\text{OPT})+o(E)$ guarantee. For the multiplicative distortion, we present several algorithms that induce a competitive ratio in $\widetilde O(\min\{ \sqrtμ\log k, μ\log\log k+μ^{1.5} \})$, where the $\widetilde O(\cdot)$ may hide logarithmic factors in $μ$.

    benchmark
  194. arxiv:2611.17479 · eess.SY
    RAZeR: Representation Aware Zonotope Reduction
    Vladimir Krsmanovic, Tobias Ladner, Bernd Finkbeiner

    Reachability analysis is a popular technique for verifying the safety of a dynamical system under uncertainty, where the uncertainty is often represented by zonotopes. As the size of the zonotope representation typically grows as the reachability analysis progresses, it must be repeatedly reduced at the cost of additional conservatism. While various methods exist to reduce the size of the representation, combined advantages are not fully leveraged as only a single reduction method is typically used throughout the analysis. In this work, we show how dynamically choosing which method to use can yield significantly tighter results, thereby easing verification. This is realized by introducing a heuristic, derived from model predictive control, that accounts for upcoming propagation of the current zonotope. We argue that order reduction is not merely a post-processing step after propagation but also a pre-processing step that shapes how the zonotope evolves under the system dynamics. We evaluate our approach, RAZeR, on nonlinear benchmarks taken from the ARCH-COMP, where we demonstrate that RAZeR produces tighter zonotopes at a negligible performance overhead.

    benchmark
  195. arxiv:2611.17478 · cs.LG
    ERGo: Energy-Efficient Hybrid Graph Neural Network Training on Processing-In-Memory Architectures
    Pratyush Dhingra, Chibuike E. Ugwu, Janardhan Rao Doppa, Partha Pratim Pande

    Processing-in-memory (PIM) has been proposed as an alternative computing paradigm for training Deep Neural Networks, including Graph Neural Networks (GNNs). Despite these advancements, training GNN workloads on PIM devices necessitates off-chip memory access. This off-chip access is expensive in terms of performance and energy, thereby impacting the overall energy efficiency of the training process on PIM platforms. In this paper, we propose a novel hybrid training framework called ERGo that automatically switches from a full-parameter phase to a parameter-efficient phase during the training process with negligible loss in predictive accuracy. The parameter-efficient phase employs low-rank representations of GNN weights, effectively reducing the number of trainable parameters. This helps in reducing the off-chip access during the end-to-end training process on PIM architectures, leading to notable improvements in both performance and energy efficiency. ERGo outperforms the conventional full-parameter GNN training on a PIM-based platform by up to 3.15x in speedup and enhances energy efficiency by 10.5x. Furthermore, ERGo offers additional advantages when utilized on heterogeneous architectures incorporating non-volatile memory (NVM) and static random-access memory (SRAM). Training on the heterogeneous NVM-SRAM architecture typically utilizes NVM for the forward-pass and SRAM for the backward-pass. However, NVM cells suffer from low device endurance and repeated write operations due to weight updates can lead to poor lifetime. ERGo improves the lifetime of NVM devices by freezing the weights and preventing weight updates in the parameter-efficient phase. Experimental results demonstrate that ERGo improves the lifetime of heterogeneous NVM-SRAM architecture by up to 33x in comparison to full-parameter implementation.

    memorymemory architecture
  196. arxiv:2611.17468 · cs.LG
    A thorough review on Deep Learning approaches to 2D-to-3D Registration in Medical Imaging
    Andrea Borghesi, Julia van Beurden, Maarten van Alphen, Michiel van den Brekel +5

    2D-to-3D image registration is a key component for several medical imaging purposes such as intraoperative guidance, enabling the alignment between preoperative 3D volumes and real-time 2D images. While conventional optimisation-based methods often struggle with non-convex objectives, sensitivity to initialisation, and high computational demands, deep learning (DL) has emerged as a promising alternative, offering improved alignment robustness and the potential for real-time performance. This review provides a comprehensive analysis of DL frameworks developed to address the challenges inherent in 2D-to-3D registration. We first outline how core concepts of DL are adapted to the registration problem. We then examine recent methodological advancements from multiple perspectives, including pipeline design, architectural innovations, and domain adaptation strategies. Particular attention is given to quantitative metrics, highlighting how performance is defined and measured across studies. Across the 41 included studies, we find that direct regression is the most common pipeline (23 studies, against 18 iterative ones), that hybrid CNN-attention architectures are gaining ground, and that research remains heavily concentrated on X-ray-to-CT registration. We conclude by identifying the absence of per-case uncertainty estimation and the strong neglect of temporal or sequential modelling as the two methodological gaps most consequential for clinical readiness, and discuss state-space models as a way to add temporal memory within intraoperative latency budgets. Further opportunities lie in more principled domain adaptation and in the integration of foundation models for representation learning.

    memory
  197. arxiv:2611.17464 · cs.CL
    Poetry 'resurrection' from LLMs
    Vincent Cellucci, Peter van der Putten

    In an era where linguistic creativity is challenged by algorithmic authorship, we pose a provocative question: can artificial intelligence (AI) write an infinite poem? We set up two experiments to investigate the quantitative and qualitative output produced by AI utilizing couplets that evidence a poetic relationship between end words, the only variation in a strict anaphoric, litany structure. Our results show that while AI can generate interesting pairings, it still falls short of human-level creativity and understanding of poetic relationships. We offer commentary and a brief survey of the field of computational poetry experiments, tracing back authorial intention and distinctions of 'natural' and 'artificial' poetry, as well as justifying why couplet form is suited to testing AI's poetic output. Experiment 1 relies on manual prompting and co-creation to extend the original poem, which is developed into an animated webpage installation and interactive artwork. Experiment 2 reliably achieves automated output reaching the 1000 benchmark via an alternative method of using an LLM in combination with semantic similarity, of which Method 2 demonstrates a higher capacity for semantic and quantitative poetic output. These experiments highlight the limitations of current LLMs' generation capacity at the word level but also suggest that poetry will also respond and revive by incorporating AI workflows.

    benchmark
  198. arxiv:2611.17461 · cs.CL
    Tracing Knowledge and Reasoning in LLM
    Yang Wang, Xiao Wang, Yiqi Liu, Chenghao Xiao +3

    Current evaluations of Large Language Models (LLMs) report performance at the dataset or task level, revealing what a model achieves but offering little insight into the internal processing associated with its performance. We propose a task-centric framework for characterising the model-internal processing elicited by a benchmark. It combines attention- and activation-based signals with representation-based measures to characterise benchmarks spanning a continuum from retrieval-dominant to reasoning-dominant and to surface failure modes that task-level scores conceal. We evaluate 16 LLMs across 24 diverse benchmarks and conduct targeted internal-state analyses on representative models and tasks, observing systematic differences in processing between knowledge-centric and reasoning-centric tasks. We shift evaluation from asking what score a model achieves toward characterising the processing patterns associated with that score, potentially supporting more targeted benchmark selection and more informed model choice. We open source our code at https://github.com/extraordinarylab/trace-kr.

    benchmark
  199. arxiv:2611.17460 · cs.LG
    Unfinished-Bench: Measuring Rule Violations by Persistent Agents in Realistic Work Environments
    Seunghyun Lee, Jaehui Hwang, Dongyoon Han, Sangdoo Yun

    AI agents are increasingly equipped with persistence toward task completion to run delegated work autonomously for long periods, under many unstated constraints from internal documents and other people's authority. When completion requires violating a rule, the agent must decide alone, yet benchmarks of instruction-rule conflicts state the rule in the prompt or keep it hidden and do not evaluate persistence. We present Unfinished-Bench, a benchmark that gives models persistence in autonomous runs without a user, in realistic work environments with embedded rules. In each of its 97 scenarios across 7 organizations and 7 risk categories, the goal and the rule cannot both hold. We find that persistence raised violations of a rule in an internal document from 32.2% to 43.6%, five times the 8.5% for the same rule stated by the user. Of the document-rule violations, 98.3% occurred after the rule was displayed, and in 89% of a judged sample the agent first wrote down the rule. Violations of rules placed in the environment without user oversight can lead to serious incidents, so persistence-driven risks need to be evaluated. Benchmark documentation: https://francis21c.github.io/unfinished-bench/

    agentai agentbenchmark
  200. arxiv:2611.17456 · cs.MA
    Physical-Risk-Aware Decentralized Agents for Constrained Vehicle Communications
    Haoran Peng, William Weijia Zheng, Hang Liu, Ying-Jun Angela Zhang

    Decentralized V2X resource control is challenging because each vehicle has only local observations, while reliability depends on network-wide interactions. We develop a physical-risk-aware spatio-temporal graph attention (PA-STGAT) framework for agentic V2X control. Each vehicle perceives its local interaction graph, retains short-term temporal context, and independently selects its PC5 resource and power. Network outcomes are used during centralized training to refine the shared policy. Transmitter-specific credits map reliability failures to individual vehicles. A dual-priced graph term further converts network violations into coordination signals for discrete resource selection, with higher priority on short time-to-collision links. Experiments show higher raw-policy reliability than the tested graph-based baselines. Under filtered execution, PA-STGAT meets the nominal empirical communication requirements across all five seeds, with tradeoffs among utility, delivery reliability, resource occupancy, and filter-induced repair. Compared with GNN+MATD3, it requires fewer resource repairs but achieves lower utility and mean delivery reliability. Mechanism ablations show that removing dual-priced graph coordination causes resource-selection collapse. Density and mobility tests characterize the operating limits of empirical constraint satisfaction.

    agentic
  201. arxiv:2611.17442 · cs.RO
    Racing the Clock: Teaching VLNs Budget-Conditioned Object Search
    Maitree Meher, Siming He, S. Shankar Sastry, Jitendra Malik +2

    A rescue robot entering a building with thirty seconds on the clock should sprint toward the single most promising room; given ten minutes, it should instead explore a broader tour. Yet vision-language-navigation (VLN) models are typically evaluated under a fixed episode limit that the policy never observes, even though time-critical missions require behavior to adapt to the available budget. We study budget-conditioned object search: given a target category and an explicit action budget, an agent must find as many distinct target instances as possible before the budget expires. We investigate how to supervise VLNs for this task by comparing a privileged TSP expert, which plans near-optimal routes using the full map and true target locations, with a room-level Bellman expert that maximizes expected return under partial observability using high-level cross-scene semantic priors. Despite the TSP expert achieving substantially higher expert performance, its demonstrations are less effective for learning exploration: on held-out HM3D scenes, the Bellman-supervised VLN consistently outperforms the TSP-supervised VLN, with the advantage becoming more pronounced at longer budgets. At 2000 steps (500m), the Bellman-supervised VLN finds 71.4% of reachable target instances, compared with 65.5% for the TSP-supervised VLN and 44.8% for the pretrained VLN. These results highlight a distinction between optimal routing and optimal exploration: effective exploration supervision should optimize decisions under partial observability rather than imitate routes that are optimal only once the environment is fully known.

    agent
  202. arxiv:2611.17440 · cs.LG
    Interpreting Coding Agents Through Their Behavioral Trajectories
    Ibragim Badertdinov, Alexander Golubev

    Standard coding-agent benchmarks primarily evaluate binary task success, treating execution as a black box and obscuring the behaviors that produce those outcomes. We analyze a large corpus of 143,209 execution trajectories from 125 agent configurations spanning 117 unique models. To enable comparisons across heterogeneous agent interfaces, we develop a rule-based parser that maps raw execution traces to high-level actions such as repository search, file editing, and test execution, and audit its predictions against a six-model consensus reference on 500 events. We find that agents with similar solve rates can nevertheless differ substantially in how they explore, modify and verify code. Moreover, without using task outcomes, token counts, or tool names as features, aggregate behavioral profiles retrieve the corresponding agent stack across disjoint repository sets and programming languages, although task composition is not fully controlled in the cross-repository setting. Through analyses of frontier models and agent scaffolds, we further show that scaffold choice and reasoning effort systematically shift agents' search, editing, and testing behavior. We release the deterministic parser together with the full annotation and analysis pipeline alongside an annotated trajectory subset.

    agentagent benchmarkbenchmark
  203. arxiv:2611.17438 · cs.CV
    Real-Time Diffusion Relighting as a Neural Sensor: Closing the Low-Light Domain Gap for Robotic Perception
    Shengjie Xu, Alexander Gao, Anshul Rai, Dae Yeol Lee +2

    Low-light conditions severely degrade robotic perception: object detectors trained on daylight imagery are prone to failure under dark and adverse weather conditions that fall outside of their training distribution. Furthermore, from an efficiency standpoint, existing Low-Light Image Enhancement (LLIE) methods are designed for offline processing and are therefore impractical to integrate into live control loops. To address these issues, we propose to embed a Generative Neural Sensor within the control loop that treats illumination as a generative prior rather than a signal processing problem, enabling task-relevant semantic recovery. We further propose an Real-Time Perceptual Architecture that decouples high-latency semantic recovery from low-latency robotic control. Rather than blocking the control loop, our system runs a large-scale image-to-image diffusion model in a background thread, and is activated only when a drop in detection confidence signals a perceptual uncertainty event, significantly reducing computational overhead. Our method sets a new state-of-the-art performance on the ExDark low-light benchmark with 87.1% mAP, and under adverse weather conditions, significantly boosts detection rates over competing methods. On a navigation task in a closed-loop autonomous driving simulator (CARLA), we achieve a 100% lap completion rate, establishing a simulation-based proof-of-concept for integrating high-latency generative models with standard robotic control autonomy stacks.

    benchmark
  204. arxiv:2611.17400 · cs.CV
    ParseBench-2: Severity-Aware Evaluation of Document Parsing
    Pierre-Loïc Doulcet, Boyang Zhang, Sebastián G. Acosta, Sacha Bron +4

    Document-parsing metrics can give high scores to outputs that change important facts and penalize changes that preserve meaning. People can often judge these consequences more easily than they can design a metric to measure them. We introduce a method for developing parsing metrics from severity policies that specify how each error type should affect the score. We express these policies through test scenarios that specify how a metric should behave. We then optimize the metric implementations against these fixed expectations, using coding agents to refine comparison rules, error detectors and weights under human review. This makes stated human priorities an explicit objective of metric design. We apply this method in ParseBench-2, a comprehensive document-parsing benchmark that combines broad capability coverage with metrics calibrated to error severity. It covers 4,317 pages across six dimensions of parsing quality: text, formatting, tables, charts, visual grounding, and forms. We use detailed annotations and targeted difficult cases to check for changed facts, broken structural relationships, and missing content. We take the lower of a parsing score that measures content recovery and a severity cap that penalizes detected errors, so recovering most of a page cannot hide a serious error. Across six severity test suites, our metrics more closely match the expert judgments expressed by the scenario targets than existing metrics and penalize at most 3.4% of the tested benign variations. We conduct a comprehensive evaluation spanning frontier vision-language models, specialized document parsers and open-weight systems, comparing their overall rankings, performance across parsing capabilities and tradeoffs between accuracy and cost.

    benchmark
  205. arxiv:2611.17394 · cs.RO
    Agentic-Teacher: An Agentic Data Engine for Failure-Driven VLA Post-Training
    Gongyue Teng, Yuqing Xie, Yixian Zhang, Feng Gao +9

    Improving a pretrained vision-language-action (VLA) model requires demonstrations that address its weaknesses and provide effective supervision for learning. Coding agents can autonomously generate demonstrations through environment interaction, yet successful execution alone does not ensure effective supervision for VLA learning. We introduce Agentic-Teacher, an autonomous data engine for failure-driven VLA post-training. Given a VLA checkpoint, a task specification, and configured environment tools, the engine analyzes failed rollouts, selects intervention states, and iteratively develops closed-loop code policies through execution feedback. Validated policies become reusable demonstration generators that collect successful continuations from states where the VLA struggles. To make these demonstrations suitable for learning, Agentic-Teacher introduces temporally correlated action noise to broaden state-action coverage, removes idle frames while preserving gripper transitions, and filters trajectories using the frozen base model's flow-matching loss. The resulting demonstrations are combined with successful VLA rollouts for post-training, yielding an updated policy that operates without coding-agent assistance. Agentic-Teacher thus connects the weaknesses of a target VLA to both the generation and curation of its training data. In simulation, it improves $π_0$ success rates from 79.4% to 87.8% on LIBERO-10, from 80.4% to 90.6% on MetaWorld-10, and from 0% to 30.0% across three RoboCasa tasks.

    vision-language-actionvlaliberogripperagenticpost-training
  206. arxiv:2611.17388 · cs.AI
    Time-Consistent Evaluation of Hierarchical Code Graphs for SWE Agents
    Xianpeng, Sun, Tian Yu, Haonan Sun +7

    Evaluating repository-aware software engineering (SWE) agents requires strict separation between pre-existing repository knowledge and future code evolution, yet standard benchmarks frequently suffer from temporal contamination and prompt-level solution leakage. Building on the time-consistent benchmark protocol of Sun (2026), we enforce strict zero-leakage isolation across DragonFly in C++ and React in JavaScript/TypeScript, locking repository snapshots and knowledge graphs at historical cutoff commits $T_0$ and evaluating 100 post-$T_0$ pull requests across four prompt granularities: Minimal, Concise, Contextual, and Guided. Under this zero-leakage protocol, we conduct a matched evaluation of three context-augmentation regimes: (1) an unaugmented Baseline, (2) a flat syntax-level Tree-sitter AST graph, and (3) ServiceExplorer, a deterministic two-level hierarchical dependency graph capturing Level-1 system/package containers and Level-2 architectural modules. Across 7,200 paired evaluations spanning six foundation models across the MAI-Code Flash, Claude, and GPT-Sol families, ServiceExplorer consistently outperforms both Baseline and Tree-sitter AST across every model, repository, and prompt granularity ($+15.6\%$ to $+86.2\%$ relative $F_1$ lift, $p < 10^{-4}$), while tripling exact full-PR localization on DragonFly. Crucially, equipping an efficiency-tier or prior-generation model with ServiceExplorer consistently surpasses the unaugmented Baseline $F_1$ of the next-generation model across all evaluated families.

    knowledge graphbenchmark

02 US SEMI · SEC 8-K FILINGS

4 items

scanned: NVDA / AVGO / MRVL / COHR / LITE / AMD / TSM / SMCI / ANET / CRDO / POWL / VECO

  1. $AMD · 8-K/A · filed 2026-11-16
    Advanced Micro Devices Inc
    Items: 3.02
    8-K/A
  2. $POWL · 8-K · filed 2026-11-16
    Powell Industries Inc
    Items: 2.02,9.01
    8-K
  3. $MRVL · 8-K · filed 2026-11-12
    Marvell Technology Inc
    Items: 5.02,9.01
    8-K
  4. $NVDA · 8-K · filed 2026-11-12
    NVIDIA Corp
    Items: 5.02,9.01
    8-K

03 HUMANOID · COMPANY NEWS

54 items

scanned: figure-ai / 1x / boston-dynamics / unitree / apptronik / sanctuary-ai / neura-robotics / agility-robotics / physical-intelligence / agibot

04 CN PHOTONICS · ANNOUNCEMENTS

0 items
China sources are not connected yet.