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.
418 items today · 361 arxiv · 3 SEC 8-K · 54 humanoid · 0 CN photonics
01 ARXIV · PHYSICAL AI PAPERS
361 items- arxiv:2611.23229 · cs.LGAmbient LLMs: Data-Efficient Supervised Fine-tuning via Temperature MatchingGiannis Daras, Marios Papachristou, Adam Klivans, Antonio Torralba +1
Collecting large-scale, diverse, and high-quality post-training datasets for autoregressive language models is challenging. Consequently, Supervised Fine-Tuning (SFT) is typically applied to datasets significantly smaller than those used for pre-training, which leads to overfitting and the catastrophic forgetting of the tails of the data distribution. To prevent this collapse, we introduce Ambient SFT, a principled framework for leveraging imperfect predictions to augment the fine-tuning set. In our method, a "frozen" version of the pre-trained model serves as a noisy teacher that the fine-tuned model must match at a coarse level. In particular, we require that the high-temperature versions of the conditional distributions produced by the post-training model should match the high-temperature versions of the pre-training ones. The temperature hyperparameter controls the level of trust in the pre-trained model, with the limit $T\to \infty$ reverting back to regular SFT and $T=1$ to knowledge distillation. We demonstrate that our approach outperforms standard SFT across domain adaptation, instruction-tuning, and reasoning, leading for up to $21\%$ average relative improvement to standard SFT.
post-training - arxiv:2611.23228 · cs.LGSpool: Low-Latency Streaming Inference for Hybrid Vision-Language ModelsRui Ning, Om Chabra, Hari Balakrishnan
Applications such as visual accessibility agents and robotics require answering questions about recent video content promptly, over a bounded window of past frames (e.g., the last two minutes). Hybrid vision-language models (VLMs), which interleave attention with linear recurrent layers such as Gated DeltaNet (GDN), are attractive for this setting, but maintaining a sliding window is expensive. The reason is that recurrent layers combine all past frames into a shared state, so expiring old frames normally requires re-running the entire window through the model. We observe that the recurrent update is affine in the state with input-dependent coefficients, so the effect of each frame can be captured as a transformation that composes associatively. We present Spool, a training-free framework that caches per-frame state transformations and composes them efficiently as the window advances, removing expired frames from each recurrent layer's state without replaying frames through the backbone. Removal is exact within each layer; as with key-value eviction in attention, expired frames can retain residual influence through the cached inputs computed by earlier layers. We design a dedicated data structure and GPU kernels to achieve low caching and composition overhead. Across five open-source hybrid VLMs and four video reasoning benchmarks, Spool achieves the lowest p95 time-to-first-token in all 20 model-benchmark tests, reducing it by a geometric mean of $6.6\times$ over the faster of two recent vLLM baselines, while improving accuracy by 2.3 percentage points on average.
benchmark - arxiv:2611.23225 · cs.AIStringQFT: A Multiuse Textbook Resource for AI-Assisted Theoretical PhysicsXingyang Yu
We present StringQFT as a textbook collection for string constructions of quantum field theory and as infrastructure for AI-assisted theoretical physics. The collection comprises 18 volumes, 190 chapters and 3,795 pages, covering brane probes, brane webs, geometric engineering, F-theory constructions, wrapped M5-branes and exact worldsheet constructions. Its open-book benchmark inventory contains 1,073 items, including 499 with automatic finite-answer checking. For retrieval-augmented generation, it provides 1,371 section records linked to textbook sources and PDF locations, together with a local lexical retriever. Training-data preparation is illustrated by 21 private, source-linked examples. For agents, it exposes 19 calculation tools with specified inputs and finite domains, connecting tool use to textbook derivations. We illustrate these uses through specific questions, retrieved passages and finite calculations. The textbook collection and accompanying resources are available at https://github.com/xingyang-yu/StringQFT.
retrieval-augmentedtool usebenchmark - arxiv:2611.23219 · cs.LGDo LLM Agents Still Need to Be Told How to Self-Improve?Shobhita Sundaram, Cansu Sancaktar, Phillip Isola
LLM agents increasingly drive their own learning. Self-improvement methods delegate various components of model optimization to the models themselves. As models become more capable, this motivates a basic question: do these increasingly capable models still need us to specify how they should self-improve? In this paper we study an autonomous paradigm in which, given basic primitives, resources, and a general goal for self-improvement, agents construct their own self-learning algorithm for context-based optimization. Using the setting of gameplay as a testbed, a frozen LLM agent is given a compute budget and tools for collecting experience and updating text documents, and allowed to freely experiment without an externally imposed loop or access to real-world data. We find that sufficiently capable models do self-improve under these conditions, demonstrating strategies that align with classic cognitive frameworks for self-regulated learning and constructs from machine learning literature on self-generated curricula. Furthermore, we study the extent to which the learned text artifacts transfer to external benchmarks, observing improvement across different game horizons, mechanics, and 3D navigation.
agentllm agentself-improvementbenchmark - arxiv:2611.23216 · cs.LGHigher-Scoring Coding Models Did Not Get More Work DoneYining Hua
New coding models advertise themselves on score improvements of coding benchmarks. Yet no study has tested if a model that scores higher on these benchmarks gets more real work done. On June 12, 2026, a U.S. export control directive banned access to Claude Fable 5 for 19 days, which forced users to switch to other models. During this time, most former Fable 5 users switched to Opus 4.8 (11 points lower on SWE-Bench Pro compared to Fable 5). We treat this suspension as a natural experiment, where the people and projects stayed the same and only the model changed, so we could isolate the effect of Fable 5 on user productivity. We tracked GitHub pull requests of 297 Fable 5 users before, during, and after the suspension. We used 8,908 developers who stayed on Opus 4.8 all the time as the control group. We also repeated the comparison across five model upgrades before and after the suspension in 2026 (1,208 developers, 15,442 pull requests). Fitting a difference-in-differences model with developer and date fixed effects, we found that having access to Fable 5 increased the 7-day merge rate of pull requests by only 1.4 percentage points [-1.4; +4.3], which was not statistically significant. Similarly, no significant increases of merge rates were found with the upgrades (weighted mean: -0.9 percentage points [-1.9; +0.2]); all five estimates were negative (95% intervals included 0, did not differ from each other). We also did not observe a significant increase in the number of pull requests merged per week (+2.4% [-4.9%; +10.2%]).
benchmark - arxiv:2611.23215 · cs.ROFeedback-Linearised Trajectory Tracking via Model-Informed Robust Geometric DDPCPranav Gupta, Deepan Muthirayan, Ravi Banavar
Data-driven predictive control (DDPC) predicts system responses from recorded trajectories, but nonlinear dynamics can limit the accuracy of fixed linear predictors. This paper studies geometric robust DDPC for trajectory tracking of a fully actuated manipulator using model-based feedback linearisation. A mechanical model maps virtual acceleration commands into joint torques, while recorded data capture residual dynamics arising from imperfect model parameters. The controller minimises worst-case prediction error over an open Grassmannian ball centred at the fitted trajectory subspace. Simulations on a planar 2R manipulator under gravity compare the proposed method with double-integrator model predictive control (MPC) and direct regularised data-enabled predictive control (DeePC). Results show improved tracking over double-integrator MPC despite increased torque effort, while performance relative to DeePC depends on the reference trajectory.
manipulator - arxiv:2611.23202 · cs.AIAgent Trust Infrastructure (ATI): A Federated Identity, Trust, Attestation, and Accountability Architecture for Cross-Organization Autonomous AI AgentsViet K. Nguyen, Mohammad I. Husain
AI agents increasingly collaborate across organizational boundaries, invoking tools and interacting with agents operated by independent entities. Although existing agent frameworks support planning and coordination, they lack a common mechanism for verifying cross-organizational agent identity, execution integrity, and authorization. We propose the Agent Trust Infrastructure (ATI), a framework-independent security architecture built on Public Key Infrastructure (PKI), Zero Trust Architecture (ZTA), and remote attestation. ATI establishes federated trust through peer organizational certificate authorities (AgentCAs), eliminating the need for a shared root authority. Before executing requests, ATI validates agent certificates, verifies approved model, tool, and system-prompt configurations, and enforces authorization through three policy checks. We implement ATI as an external security enforcement layer that integrates with existing agent frameworks without modifying their underlying models. We evaluate ATI in a simulated federation of eight organizations using six agent frameworks, each deploying a five-agent team. Security experiments involve two organizations and assess 14 attack scenarios and six benign tasks across five agent models, using five attack variations and three repetitions per variation. A panel of judge models evaluates the experimental outcomes. With all enforcement mechanisms enabled, attack containment increases from 43% to 84% (40% to 80% for the 11 attack scenarios evaluated consistently across all configurations), while enforcement introduces approximately 174 microseconds of overhead per request. ATI is released as an open-source, framework-independent package that uses IETF WIMSE workload token formats for identity and request verification and Entity Attestation Tokens for model, tool, and system-prompt integrity verification.
agentai agentagent frameworkjudge model - arxiv:2611.23198 · cs.ROLearned versus Kinematic Trajectory Prediction under Reduced Observation FrequencyNukhbala Gadzhiev, Alexander Morochko
Trajectory prediction models for automated driving are usually trained on data with a fixed sampling frequency, while positions received via vehicle-to-everything (V2X) communication arrive at 1-10 Hz. We study whether a learned predictor that uses only the observation history keeps its advantage over constant velocity (CV) and constant acceleration (CA) models when the history frequency drops. On the Argoverse 2 motion forecasting dataset, the 5 s history of the focal agent was downsampled from 10 Hz to 5, 2, and 1 Hz, and the 6 s future was predicted without maps or neighboring agents. An LSTM-based recurrent neural network (RNN) trained for each frequency was compared with the same network trained at 10 Hz and applied to sparse histories after linear interpolation. Hypotheses and thresholds were fixed before training, the main RNNs were trained with ten seeds each, and results on 24,984 test scenarios have bootstrap confidence intervals over scenarios and seeds. The average displacement error of the frequency-specific RNN grew by 3.1% at 5 Hz, 10.1% at 2 Hz, and 20.3% at 1 Hz. Contrary to our hypothesis, its advantage over the better kinematic model did not vanish on straight motion and was 24-30% on both straight motion and turns at every frequency, while for pedestrians CV was more accurate. The 10 Hz model with interpolation was at most 1.3% less accurate than the frequency-specific model down to 2 Hz but 11.5% less accurate at 1 Hz. Each hypothesis was only partly supported.
agent - arxiv:2611.23189 · cs.CVRetrieval-Augmented Correction for Generalist Medical Image SegmentationVincenzo Marcianò, Xiaoming Zhang, Michela Antonelli, Sebastien Ourselin +1
Automated segmentation is routinely used as the first step in quantitative image-based clinical workflows, where predicted masks are subsequently reviewed and corrected by human operators. Automated correction of the residual structural errors is currently addressed one anatomy at a time, so the number of trained correctors scales with the number of segmentation targets. We present GRACE, a retrieval-augmented refiner that edits an input mask under context drawn from a shared knowledge base of verified prior cases. The refiner is trained once on a multi-organ corpus spanning compact organs, cavities, branching vessels, and tumour lesions. In the held-out dataset-transfer evaluation, three held-out datasets query a fixed knowledge base built from training data only. A separate leave-one-anatomy-out experiment tests parameter-free index extension: verified cases of an unseen solid organ enter the knowledge base, and the refiner stays frozen. A light shape-aware projection head on a frozen multimodal encoder maps embedding similarity to the shape agreement of the masks, so the knowledge base returns cases that the refiner can use as a shape prior. We evaluate GRACE on 18 public datasets spanning CT, MRI and ultrasound, using nine refinement configurations and seven upstream segmentors that span weak, mid, and strong operating regimes. We report each change with the DSC of its input mask. On the weak tier, the input DSC is 0.386 and GRACE adds up to +0.35 DSC. On a strong segmentor, the input DSC is 0.932 and the mean change is small. Transfer to an imaging modality that is absent from training and from the index is partial and depends on the anatomy, and remains an open research direction. Our results suggest retrieval-conditioned refinement is a practical alternative to anatomy-specific correction models. Code: https://github.com/robustml-eurecom/grace
retrieval-augmented - arxiv:2611.23187 · cs.CVAny2Building: Unifying Point Clouds and Images for Building Reconstruction via Shared Spatial QueriesHongye Hou, Zhuoxiao Li, Liu Xiangyu, Tengxi Wang +5
Abstracting structured buildings from limited observations is essential for 3D urban modeling and digital twins. Partial point clouds and overhead images of the same building provide incomplete 3D surface measurements and visual cues to its footprint, roof geometry, and rooftop structures, respectively. This motivates a unified framework that aligns geometric and visual cues in 3D and reconstructs buildings from either or both modalities. Building on this connection, we present Any2Building, a unified model for Level of Detail 2 (LoD2) building reconstruction from partial point clouds, overhead images, or both, with optional coarse geometric priors. At its core is an occupancy-grounded shared spatial queries representation that organizes available observations within a common 3D grid. Supervised by complete building surface voxels, these queries connect observational evidence to target geometry, enabling a shared generator to learn structural priors across tasks and providing spatial guidance for recovering missing structure. A trained voxel conditional surface-generation and low-poly remeshing pipeline then converts the predicted structure into compact LoD2 building models. To support this framework, we introduce BuildingWorld-Image, comprising approximately one million paired samples of partial point clouds, overhead images, and LoD2 building models across 18 cities. Experiments on multiple benchmarks demonstrate that Any2Building outperforms existing methods in geometric reconstruction accuracy across unimodal and multimodal point-cloud completion, single-image building generation, and prior-assisted generation, while maintaining detailed and robust recovery when point cloud input modality is degraded.
benchmark - arxiv:2611.23177 · cs.LGDéjà View: A Shared Working-Memory Model for Visual TasksMranmay Shetty, Gabriel Kreiman
A working-memory model must remember stimuli and use its memory flexibly to perform different tasks. Here we introduce a model of working memory for images that combines dynamic activity with temporary associations between image features. We train the model on five tasks, including remembering a sample while other images are presented, categorizing distractor images while remembering the sample, and determining whether an image is new or has been presented before. The same trained model performs all tasks, guided by explicit cues about which response is required. We use this flexible memory model to ask how delay, distraction, and image history affect performance. Delays have little effect on sample-memory performance for the delays we tested, while seeing more images leads to more false alarms even when the target image appeared equally recently in all cases. On the sample-memory-plus-categorization task, the probability of correct responses on all questions decreases as more questions are asked, despite robust sample-memory performance. Finally, the model maintains stable performance across 40 brief trials of each task under the tested conditions. Together, these results demonstrate how a single working-memory system can be flexibly used to perform different tasks, and how its errors depend on both previous images and task demands.
memory - arxiv:2611.23174 · cs.AIInference after coding agentsFlorian Gunsilius, David Van Dijcke, Kaspar Wüthrich
Empirical research is changing rapidly as AI coding agents are becoming increasingly proficient in every step of it, from writing code to manipulating data. We argue that while the potential of coding agents for science is vast, their use in hypothesis testing poses a substantial threat to classical inference, as these agents enable a supercharged form of selective inference. In particular, we argue that coding agents carrying out empirical research introduce a new type of post-selection inference that combines human-like, open-ended search over an implicit space of estimators with programmatic replayability of both the selection and the estimation process. Indeed, we find that on simulated data with zero effects, four frontier agents report significant p-values between 54% and 98% of the time when asked to produce a compelling finding. We propose a simple procedure for detecting and correcting these distortions. After an agent has completed its analysis, we reset and replay it on subsamples of the data to estimate the agent's relative probability of selecting that analysis when it looks statistically favorable. For agents with stable selection behavior, the resulting corrected test is asymptotically valid and consistent conditional on selection. Importantly, the correction is learned from the black-box agent's actual choices rather than a model of how it searches, so it remains valid even as AI technology continues to evolve rapidly. In simulations, the correction robustly lowers the high false rejection rates to near the nominal 5%. We illustrate the method on applications from two published papers.
agent - arxiv:2611.23158 · cs.LGMoE--LoRA Methods for Multilingual Adaptation: A Systematic Empirical StudyNikit Srivastava, Joel Dag, René Speck, Axel-Cyrille Ngonga Ngomo
Routed low-rank adapters offer a parameter-efficient means to improve the performance of large language models (LLMs) across tasks. Hence, it is assumed that they can also be used to improve the support of low-resource languages in LLMs, which currently serve English far better than the thousands of other languages spoken worldwide. However, this assumption has never been verified empirically under a shared multilingual evaluation protocol. We address exactly this gap by comparing eight low-rank adaptation (LoRA) methods on a frozen base model using 200 unique languages in more than 9{,}000 evaluation settings. Our results suggest that a reduced-scale search can identify viable configurations of LoRA methods but does not reliably predict their relative performance, as the top-ranking configuration is never best at scale. LoRA reliably improves language-model fit on translation data, consistently degrades reading comprehension, and benefits low-resource languages most. These findings suggest that LoRA methods should be regarded as targeted tools rather than general-purpose means for improving the multilingual capabilities of LLMs.
evaluation protocol - arxiv:2611.23156 · physics.opticsQuantum simulation of neutral-meson mixing and decay on a reprogrammable photonic processorMohd Abu Bakar Siddique, Abrar Al Shadid Abir, A. K. M. Rakib, Mohammad Abrar Kabir +2
Neutral mesons oscillate into their antiparticles while decaying, so their dynamics is generated by a non-Hermitian effective Hamiltonian which a quantum simulator restricted to unitary operations cannot implement directly. We propose a reprogrammable photonic processor that simulates this open-system evolution through dilation and analyze it at the circuit level. The effective Hamiltonian is reconstructed exactly from measured lifetimes, mass and width differences, and the CP-violation parameter $|q/p|$. Through a singular-value decomposition and a single ancilla that encodes the total decay probability, its propagator is embedded in a four-mode unitary, which is compiled at each time step into a rectangular interferometer on a hexagonal mesh of tunable Mach-Zehnder units. Circuit simulations including coupler, propagation, and bend losses show that one fixed device, reprogrammed only through its control voltages, reproduces the flavor oscillation and decay of neutral kaons and of $B_d$ and $B_s$ mesons, with $x=Δm/Γ$ from 0.77 to 26.9. Over 500 time steps per system, the mean fidelity lies between 0.99979 and 0.99982, and no step falls below 0.9996. Because every optical path crosses nine tunable units, uniform insertion loss reduces only the transmitted power, whereas unit-to-unit loss variations lower the fidelity quadratically, $1-\langle F\rangle \simeq 0.08σ^2$ for a loss standard deviation $σ$ in dB. A leading-order analysis traces this coefficient to the routing of light through the mesh, reproduces the fitted value to within 2%, and explains why it is the same for all three mesons; a mean fidelity of 0.999 tolerates $σ\approx 0.11$ dB. The ancillary outputs encode the total decay probability without postselection, and the reported probabilities are normalized to the collected optical power.
mach-zehnder - arxiv:2611.23153 · cs.CVVIVID: Beyond Static Multi-Image Reasoning with Localized Visual Differences and Textual Entity Dynamics in Video MLLMsChengqian Ma, Yuanze Zeng, Hongyu Wang, Mingfeng Lin +7
Video question answering under tight frame budgets requires preserving localized motion and event order, yet current video multimodal large language models (MLLMs) reduce comprehension to static multi-image reasoning by discarding unselected frames and encoding keyframes independently: (1) Spatially, whole-frame pooling dilutes small moving targets into static backgrounds; (2) Temporally, dropping intermediate frames or averaging tokens erases inter-frame motion, confusing direction, speed, and order. We propose VIVID (Visual Inter-frame Variations and Individual Dynamics), which moves beyond static snapshots like human vision: keyframes capture core states while inter-frame Patch Diffs preserve neighboring changes within $K_f$ keyframes. Autonomous Delta-Driven Keyframe Planning (ADKP) selects anchors from a $3\text{ fps}$ pool and feeds surrounding patch differences into Dual-Road Change Memory (DRCM), forming Dynamic Tokens at zero extra anchor tokens alongside a camera-stabilized Active Motion Region of Interest (ROI) and chronological Entity State Chains. Adding $0.22\%$ trainable parameters to a frozen $27\text{B}$ backbone, VIVID-Small ($50$ examples, $\sim 1.5$ GPU hours) beats all eleven baselines in overall accuracy across $8,083$ questions and trails only VIVID-Large ($3,600$ examples), which leads on MotionBench, Video-MME, and TempCompass.
memory - arxiv:2611.23149 · cs.ROCapability Composition in Heterogeneous Teams via Barrier-Guided Diffusion SamplingDiana Laura Morales, Ruoyu Lin, Magnus Egerstedt, Fabio Pasqualetti
The agents in a heterogeneous team can cooperate to accomplish tasks that exceed their individual capabilities. The team trajectories for a cooperative task can be generated by a generative model, provided that the model has been trained on demonstrations of the agents cooperating. Yet, demonstrations of cooperation are costly and may not exist for tasks that the team has never performed. The conditions under which the agents can combine their capabilities, however, are physical properties of the team and are known before any demonstration is collected. We encode the conditions as constraints on the team trajectories, and we enforce the constraints while sampling from generative models trained on demonstrations of each agent alone. We prove that the proposed method can generate every team trajectory that satisfies the constraints, although the demonstrations contain no cooperation. Hence the constraints determine which team trajectories can be generated, and the demonstrations determine how likely each team trajectory is. Numerical results on two cooperative tasks show the effectiveness of the method, which generates cooperative team trajectories that satisfy the constraints and requires no retraining when agents join or leave the team. Finally, we implement the method on a multi-robot testbed.
agent - arxiv:2611.23144 · cs.CV$π^4$: A Foundation Model for 4D Scene Geometry and MotionYifan Wang, Weiguang Zhao, Ni Yang, Haoyu Guo +3
$π^4$ is a unified foundation model for 4D scene perception. From diverse, uncalibrated RGB observations in the wild---a single image, an unordered collection, or a video sequence---it recovers dense geometry, camera poses, and metric scale, together with 2D/3D point tracks and moving-object masks for multi-image inputs. It couples a geometry branch with a geometry-conditioned motion branch, which predicts image-space correspondences that are lifted into 3D using reconstructed target-view geometry. Across diverse benchmarks, $π^4$ achieves the best overall average rank in geometry evaluation, alongside competitive performance in single- and multiview 2D/3D point tracking and moving-object segmentation. With a shared backbone, the full geometry-and-motion system contains approximately 0.5B parameters---less than half the size of most baselines. It produces joint reconstructions and tracks in one forward pass, including large-scale scenes with thousands of input images despite training with contexts of at most 64 views. We further present the data and training recipe underlying these capabilities, together with empirical studies of key design choices. Code and models are available at https://github.com/yyfz/Pi4.
benchmark - arxiv:2611.23143 · cs.ROTCP4D: A Unified Tool-Center-Point Tracking and Generation Model for Robot ManipulationHaosong Peng, Xiangyang Liu, Zhengshen Zhang, Yunfeng Fan +4
Robot manipulation videos provide a rich visual experience, yet recovering robot action and gripper-state annotations from RGB observations still remains challenging. Using the embodiment-agnostic tool-center-point (TCP) pose as the action space, we present TCP4D, a unified model in which a shared 4D scene representation both tracks observed TCP motion and generates future TCP trajectories. Given a monocular video and initial 2D TCP queries, the tracker directly recovers metric TCP poses in SE(3) and gripper states alongside scene geometry and camera parameters. The generator samples local features from the same 4D representation along the recovered TCP tracks, then combines them with robot observations and language instructions to predict future TCP states. Our new cross-embodiment augmentation pipeline converts existing RGB videos into demonstrations for alternative embodiments by retargeting the recovered motion and synthesizing the corresponding observations, without recorded source-robot states. Experiments show that the tracker achieves absolute trajectory errors of 7.60 and 16.07 mm on DROID and RoboLab and adapts to a new embodiment with only ten demonstrations. The generator achieves 69.19% mean success on RoboTwin and outperforms well-established policy models on two real-world tasks, while joint training on source and augmented demonstrations improves target-robot success by 41.5% over augmented-only training.
manipulationrobotwingripper - arxiv:2611.23142 · cs.AIMind the Gap: A Deployed Multi-Agent System for Documentation MaintenanceLorenzo Masoero
The "AI revolution" has enabled teams across the technology industry to deliver software features at unprecedented velocity. While code evolves faster than ever, customer-facing documentation often struggles to stay up to speed. As users increasingly rely on LLM-based agents that consume documentation as their primary interface, the impact of this documentation gap worsens: stale information propagates and is treated as authoritative. Signals for identifying where documentation has fallen behind already exist in abundance, both explicit (e.g., resolved support tickets revealing what documentation lacks) and implicit (e.g., code-documentation discrepancies detectable from automated loops). We deploy a multi-agent system, built around a triage agent we call "TriAgent," that continuously extracts these signals and repairs documentation drift. Over 43 days of shadow-mode operation, our system triaged 499 tickets and surfaced 92 gaps in the documentation. Using a panel of LLM judges from independent model families, 81.1% of TriAgent's verifiable assertions were correct (245 of the 302 assertions the human resolution addresses; majority of three judges); on length-comparable pairs the panel preferred TriAgent's response in 70.5% of pairs and the human engineer's in 23.2% (the rest had no judge majority). Importantly, agent- and human-sourced replies tended to be complementary, with the human often surfacing insights that were either buried within the existing documentation or missing from it entirely; our system turns each such insight into a documentation gap and, after human review, a fix. Repairing a gap measurably improved the agent's later answer: when the agent re-answered the same ticket after the repair, blinded judges preferred the post-repair answer over its own pre-repair answer 71.4% of the time (n=35). A blinded human audit confirms strong alignment on self-improvement.
agentmulti-agentagent systemself-improvement - arxiv:2611.23140 · cs.CVGameRealm: Grounding Visual Learning in Interactive GamesWeiguang Zhao, Yifan Wang, Ni Yang, Yating Wang +8
Visual foundation models are central to physical intelligence, yet their development remains constrained by fragmented training datasets, inconsistent annotation quality, and limited multimodal supervision. To address these challenges, we introduce GameRealm, a large-scale, richly annotated dataset that grounds visual observations in synchronized, game-native supervision. GameRealm encompasses over 120 million RGB frames and 1,000 hours of gameplay footage across all views. Beyond its unprecedented scale, GameRealm provides rich multimodal supervision, including metric depth, camera parameters, synchronized audio, control inputs, and 3D point trajectories. Its diversity further extends to flexible view configurations, spanning single-view, stereo, multi-view, and panoramic recordings. To enable scalable acquisition across diverse game environments, we develop a unified data collection pipeline that combines human-operated interaction with agentic exploration. Extensive experiments across diverse visual learning tasks demonstrate that GameRealm consistently strengthens visual foundation models in both visual generation and geometric learning. These results establish GameRealm as a versatile and valuable data resource, unlocking new opportunities for advancing visual intelligence.
agentic - arxiv:2611.23138 · cs.LGRSI-Prime: Training Models to Train ModelsDewu Zheng, Yang Ye, Hongyu Zhang, Jun Yu +1
Enabling large language models (LLMs) to autonomously conduct training and improve target models is an important foundation for recursive self-improvement (RSI). As LLMs expand their role from being trained to conducting training, their model-training capability warrants dedicated study and systematic development. To this end, we construct RSI-Prime, a dataset for improving model-training capability, comprising 350 complete experimental trajectories across seven task categories. To help LLMs learn to make high-quality experimental decisions from training practice, we perform fine-grained decision identification and quality annotation, then apply decision-level loss masking to exclude supervision on low-quality behavior. Experiments with GLM-5.3 on PostTrainBench, a benchmark for autonomous post-training, show that training on RSI-Prime yields a 31.95% relative improvement in overall performance over the original GLM-5.3, highlighting a path toward future RSI systems in which models autonomously sustain their own advancement.
self-improvementpost-trainingbenchmark - arxiv:2611.23129 · cs.LGWhere Should Physics Enter? Information, Identifiability and Calibration in Probabilistic Sea Surface Temperature ForecastingParmitha Vangapandu, Jonnakuti Pavan Kumar, Sathwik Narkedimilli, MSVPJ Sathvik +2
Advection-diffusion priors are a common way to make neural sea surface temperature (SST) forecasts physics-informed. We test where such a prior must enter a heteroscedastic ConvLSTM to help, using daily NOAA OISST v2.1 for the Bay of Bengal (1982-2024) and a synthetic benchmark on the same grid with known currents of controlled strength. (i) On real data, a soft advection-diffusion residual with a learned velocity gives no reliable benefit (lead-15 CRPS -0.007 degrees C, 95% interval [-0.022, +0.008]); a plain temporal-smoothness penalty matches it, and equivalence within 3% is not established either. (ii) Learned velocities do not recover the currents, whether placed in the loss, in an explicit transport step, or in a transport step without neural correction: they correlate with the reference no better than a shuffled-date null, even on synthetic data where knowing the currents would improve day-15 forecasts by 16-23%. (iii) Supplied currents help on synthetic data, but mostly as information: giving them to the network as input channels, without transport, recovers 92% of the transport design's lead-5 gain, and a frozen-current numerical forecast does as well as either. (iv) With GLORYS12 reanalysis currents on real data, the transport design is 6% and 4% worse at leads 1 and 5; reanalysis advection explains essentially none of the daily OISST change (correlation 0.02), and frozen-current transport is worse than persistence. (v) The defect that matters is lead-dependent miscalibration shared by every model. One variance multiplier per lead, fitted on validation data, brings 90% coverage to 0.85-0.91 at all leads and lowers day-1 CRPS by 13-21%; fine-tuning on 15-day rollouts fixes long-lead coverage but costs short-lead skill. This version withdraws an earlier claim of an 81% CRPS reduction that came from an uncontrolled comparison.
benchmark - arxiv:2611.23127 · cs.LGThe Illusion of Grounding: Do Vision-Language Models Weigh Retrieved Text and Images Like an Ideal Observer?Basavaraj R M, Harshal Chandel
Retrieval-augmented vision-language models (VLMs) answer from two sources of evidence that can each fail: the retrieved text can be wrong or noisy, and the image can be degraded or contradicted. We introduce GIB-200K, 200,000 image-question items built from COCO and Visual Genome, each expanded into controlled stress conditions: hard-negative retrieval noise (1-20 documents), visual ambiguity (Gaussian and motion blur, low-resource crops) and cross-modal conflict with counterfactual flip pairs. Instead of only measuring degradation, we ask how models combine the two modalities. From ideal Bayesian cue combination we derive a testable prediction that yields formal definitions of Knowledge Inertia (text over-weighted relative to the model's own image evidence) and Sensory Capture (the reverse), and we estimate the weights with a measurement-error-corrected regression whose recovery of known weights we verify. A counterfactual flip decomposition separates modality bias from concept priors, and a Grounding Illusion Score measures how much clean accuracy is not grounded in the image. We evaluate 4 open models on one frozen, seeded evaluation split of 2,000 GIB-200K items. All models are accurate on clean inputs (92.5%-94.2%) but adverse accuracy falls to 41.3%-59.9%. The two LLaVA models over-weight the text (Knowledge Inertia, w_T/w_I = 1.229 and 1.322), whereas InternVL3 favours the image (w_T/w_I = 0.680); between 34.8% and 71.4% of the clean answers that are correct are abandoned when the document lies. A training-free consistency check that compares the answer with and without retrieval trades coverage for accuracy under misleading documents, while falling back to the image fails when the image is heavily degraded. All per-instance outputs, code, and an audit that recomputes every reported open-model number are released.
retrieval-augmented - arxiv:2611.23124 · cs.LGGet Hint-Augmented GRPO Done Right Through Activation SteeringKhiem Le, Ting Hua, Nitesh V. Chawla
GRPO has become the dominant reinforcement learning method for eliciting reasoning capability in LLMs using verifiable rewards. However, it suffers from a fundamental failure mode: when a training question exceeds the base model's current capability, all generated responses fail and receive zero reward, yielding no gradient signal and stalling further capability expansion. An emerging class of GRPO extensions, termed Hint-Augmented GRPO (HA-GRPO), adds a "hint" to the context to reduce the difficulty of the question, thereby producing a usable gradient signal. Nevertheless, existing HA-GRPO methods share a key issue: no hint is available at inference time to append to the question, creating a training-inference context mismatch that leads to inconsistent performance gains. In this work, we reinterpret the function of hinting -- not as reducing question difficulty, but as steering model behavior -- and propose ASH (Activation Steering for Hinting), a novel HA-GRPO method built on this view. During training, the model is presented with both hinted and plain questions, while a trainable steering vector is optimized to widen the margins between the model's generation probabilities for responses to the hinted version relative to the plain version, thereby capturing the question-agnostic component of the behavioral shift induced by the hint. During inference, the trained steering vector is directly injected into the model's activation, reproducing the hint-conditioned behavior without requiring any hint in the context. Experiments with four LLMs (Qwen3-1.7B/4B-Base, Llama-3.2-1B/3B-Instruct) show that ASH consistently improves pass@$k$ on competition-level mathematics benchmarks (AMC, OlympiadBench, AIME24, AIME25) and out-of-distribution benchmarks (Minerva, GPQA-Diamond). Notably, it improves average pass@8 of Qwen3-1.7B-Base and Qwen3-4B-Base by \textbf{2.60}\% and \textbf{2.21}\%.
benchmark - arxiv:2611.23119 · cs.CLDynamic Task Allocation with DialogueMatthew Groholski, Weiwei Gu, Nakul Gopalan
Effective multi-agent collaboration requires agents to adapt task allocation as goals and responsibilities evolve. Prior task-allocation frameworks often assume static goals, full observability, or rely on large language models (LLMs) without explicit, verifiable beliefs about task allocation or intent. These assumptions limit their applicability to real-world collaboration, where agents often lack full information about each other's actions, progress, or local environment. To support dynamic task allocation under partial observability, we introduce a communication-augmented Bayesian task-allocation framework. Our agent uses dialogue to reduce uncertainty about the environment and collaborators, while keeping decisions grounded in explicit probabilistic beliefs. We evaluate our framework in an Overcooked domain extended with dynamically arriving tasks under both full and partial object-space observability. Our communication agent consistently outperforms a vanilla LLM agent in both settings. Additionally, our framework performs comparably to a non-communicating task-allocation baseline under full observability, but substantially outperforms it under partial observability as task complexity increases by an average speed up of 0.585 and completes on average 18 timesteps faster. These results indicate that our agent is more efficient in dynamic task allocation than baselines in partially observable environments, and that communication is important as available information decreases.
agentllm agentmulti-agent - arxiv:2611.23118 · cs.ROAuditing a Learned Simulator for Legged Robots: The Robotic World Model on ANYmal DJoyjeet Singh
A legged robot can learn to walk inside a learned simulator: a neural network that predicts the robot's next state and is run forward on its own predictions. They are usually trained and checked one step ahead, although a policy uses them for a hundred steps or more. We audit one published system end to end, without a GPU: the dynamics model of the Robotic World Model (RWM) for the ANYmal D quadruped, and the uncertainty estimate of its follow-up, RWM-U. We rebuilt the model to match the released code's outputs, losses and gradients exactly, measured 101 trained simulators, 200 compressed copies and 38 simulators of three other architectures, and trained policies inside simulators that we tested in a hand-built MuJoCo model of the robot. Training on the model's own rollouts beat one-step training at 368 steps on all ten held-out episodes (registered), although one-step training was better one step ahead. One-step checks therefore misjudge design choices: pruning 30% of the weights kept 90% of the model's skill one step ahead and 39% at 2 s. Ensemble disagreement ranked rollout error well (Spearman +0.90 at 100 steps), yet covered 4.6% of outcomes at one sigma, against 68% for a calibrated estimate, as each network's own variance collapses under the implemented loss. Two checks that need no recorded future ranked some pools of simulators well, but neither was shown better than 8-step error in a primary test. At the robot level, accuracy has not predicted walking: all 18 fine-tuned copies we tested predicted better than the released model one and eight steps ahead, yet 13 gave failing policies, and registered tests of whether any check ranks simulators by their policies were not settled. In a secondary test, running the authors' policy inside each simulator ranked them well. We also report nine of our own claims that we withdrew. Scope: one robot, one gait, simulated data, no hardware.
quadrupedworld model - arxiv:2611.23115 · cs.CVNunchi: Ears Help Eyes in Multi-Party ConversationsJunYeong Park, Shijing Wang, Seonwook Park, Michael Stengel +3
Identifying whom a person looks at is central to understanding multi-party conversations, yet gaze geometry alone becomes ambiguous as more people appear. Conversations offer cues that geometry lacks, such as who is speaking and what is said, but no gaze dataset provides them for many multi-party scenes. We introduce SORBET, a benchmark of 3,340 samples from 1,578 clips with synchronized speech, transcripts, and attended person labels, covering more multi-party scenes than existing gaze datasets. Prior gaze models struggle to use these cues, which come in different forms, must be linked to the right person, and can conflict with gaze geometry. We propose Nunchi, a framework that constructs a text prompt from the transcript and the outputs of gaze and speaker models, links every cue to the candidates with shared labels, and lets a vision-language model (VLM) weigh gaze geometry against conversational cues to identify the attended person. Nunchi outperforms the strongest existing method by 4.8 percentage points. On SORBET, conversational cues help whereas additional visual cues do not, and the gains hold across three VLMs, with automatic transcripts, and on two existing datasets.
benchmark - arxiv:2611.23113 · cs.CVLUTBench: Benchmarking Text-Driven Photographic Look TransferChuanzhi Xu, Boyu Wei, Chengkun Yue, Xuanhua Yin +4
Photographic color grading shapes the tone and color of a photograph to achieve a desired look. Text-driven photographic look transfer lets users specify these adjustments in natural language, with the goal of reproducing the requested look across photographs while preserving their content. However, prior research on text-driven color grading uses different datasets, text instructions, target definitions, and metrics. These differences make it difficult to compare how faithfully the methods reproduce the same look. To compare methods under a shared protocol, we introduce LUTBench. We collect 91 three-dimensional lookup tables (3D LUTs) and define one photographic look per LUT with a specified input conversion. Applying these looks to 5,000 MIT-Adobe FiveK photographs produces 455,000 input-target image pairs with reproducible target images. For each look, we use templates to turn measured tone and color changes on fixed color samples into candidate clauses. We use training image measurements to verify applicable clauses before composing the text description. The evaluation measures target fidelity, tests sensitivity to predefined input conversions, and separates generalization to unseen looks from reproduction of seen looks on new photographs. Across 16 methods, stronger text alignment does not guarantee target fidelity. However, most supervised variants improve fidelity on both seen and unseen looks.
benchmark - arxiv:2611.23111 · cs.LGEvent-Based Dynamic Lip Region Estimation During SpeechHamze Hammami, Omar Eldardeer, Francesco Rea
Event cameras offer high temporal resolution and low power consumption, making them promising for speech-related applications. These include automatic lip reading, audiovisual speech analysis, and human-robot interaction, particularly in challenging environments characterized by low illumination or high dynamic range. Event-based lip reading, speaker recognition, and voice activity detection, however, mostly operate on a cropped region of interest (ROI), by hand or by the dataset, and no event-based method supplies that ROI as a spatial region; existing event-based face analysis gives the mouth as landmarks, which presents a mismatch between application and localization. The present work predicts a dense heatmap over the lip region from the full event "frame". Events are encoded as a polarity-separated voxel grid and processed at every location by a two-pathway continuous-time recurrent network (CT-RNN) with learned, log-space bounded time constants acting on the real elapsed time of each bin: a stateless fast pathway and a stateful slow pathway, utilised for refining and anchoring the prediction. The model reaches 69.5% on lips on four unseen speakers and switches target only 2.1 times per second, against 8.9 and 11.0 for convolutional baselines with more parameters. Adding a spatial encoder improves precision but fails on a new face scale. On a real event sensor, trained only on simulated events, the proposed model reaches 58.0% on lips against 43.2% for the CNN, with fewer than half its target switches.
event camera - arxiv:2611.23106 · cs.AINatural Language Programming: Rethinking Language, Memory, and Cognition in the Era of Generative AIGiovanni Adorni, Daniele Grosso
Generative Artificial Intelligence is transforming the role of natural language in computational systems. In classical Natural Language Processing (NLP), language was treated as an object of computation: something to analyse, classify, translate, or generate. Large Language Models (LLMs) challenge this perspective, enabling natural language to function increasingly as an executable cognitive medium through which humans structure reasoning, orchestrate problem solving, and regulate hybrid human/AI workflows. This paper introduces Natural Language Programming (NLPg) as a theoretical and epistemological extension of traditional NLP. Rather than focusing on language understanding alone, NLPg conceptualises prompts as dynamic cognitive artefacts that externalise goals, constraints, assumptions, validation criteria, and intermediate reasoning states. Prompting is thus not merely a technical interaction technique, but a form of dialogic cognitive programming operating within distributed human/AI memory architectures. Building on theories of distributed cognition, extended mind, metacognition, and dialogic learning, the paper proposes the notion of Dialogic Working Memory, describing how prompts and conversational states function as temporary external memory structures supporting reasoning continuity and reflective regulation. We further map existing memory architectures (BTMA, SKGE, RMR) onto this framework, illustrating how episodic, semantic, and metacognitive layers support long-horizon conversational coherence. Finally, implications for computational linguistics and AI/mediated knowledge systems are discussed, suggesting a post symbolic and post interface paradigm in which natural language operates simultaneously as communication medium, executable specification, cognitive scaffold, and distributed memory architecture.
memorymemory architectureexternal memory - arxiv:2611.23102 · cs.ROCADENCE: Learning Acyclic Gaits for Model-Based Quadruped LocomotionOwen Sullivan, Sourav Selvaraj, Mahdi Agheli
Model-based quadruped controllers usually step on a cyclic gait, which lifts the legs in a fixed order on a fixed period, wherever the safe footholds are. We present CADENCE, a learned step planner with a small action space. Every 40 ms, it picks one of at most 65 discrete choices: no step, or one leg's step to a safe foothold. It also predicts a swing time. We train CADENCE by reinforcement learning on top of a convex model predictive controller (MPC). Nothing prescribes a gait: there is no gait schedule, and the reward has no gait term, only a cost per step. The one structural constraint is that at most two legs may swing at once. The policy we analyzed developed an acyclic gait, with no fixed period or stepping order, yet it still paired diagonal legs. An inverse footstep model (IFM), fitted on a controller's logged steps, turns each foothold into a command. With only the IFM refitted, on fewer than 20,000 steps, CADENCE ran frozen on OCS2's nonlinear MPC. In simulated trials on a range of terrains (26% to 49% voids) and speeds (0.1 m/s and 0.2 m/s), it succeeded in 85.1% (mean of five seeds), against 67.8% for OCS2-Trot, 65.5% for OCS2-Single, and 39.6% in a static walk ablation of CADENCE. At 49% voids, it succeeded in 72.9%, against 43.0% for OCS2-Trot.
quadruped - arxiv:2611.23088 · cs.CVEvent-Based Person Detection for Maritime UAV SearchZijie Ning, Patrick Vandewalle
Maritime search and rescue requires locating people in the water quickly, motivating the use of UAVs to survey the search area. During flight, camera motion and the longer exposures used in low light can blur small targets and hinder person detection. Event cameras record brightness changes as they occur, reducing the motion blur associated with longer exposure times. We investigate their use for maritime person detection by generating synthetic events using the v2e event simulator from public MOBDrone rescue videos and retaining the existing annotations. We compare the recurrent vision transformer model RVT-T with YOLOv8n and RT-DETR-R18 trained on exposure-matched images. On 843 test samples from held-out recordings, RVT-T achieves 15.63% person AP50:95, compared with 6.89% for YOLOv8n and 9.44% for RT-DETR-R18 under 100 ms synthetic exposure. Increasing exposure times from 50 to 200 ms progressively reduces frame-based detection accuracy, widening the gap to the event model. To evaluate the feasibility of on-board processing, we deploy our detector on a Jetson Orin Nano platform. The RVT-T model retains its accuracy and requires approximately 136 ms and 2.02 J per sample, allowing real-time processing on the UAV.
event camera - arxiv:2611.23085 · cs.CVTransfer Once, Ask Locally: Compact Audio-Visual Memory for On-Device Question AnsweringQitong Wang, Jihui Jin, Divya Kothandaraman, Vishnu Raj +1
Audio-visual question answering on lightweight user devices requires balancing speed, memory usage, privacy, and communication. One approach is to run audio-visual encoders on a server and a lightweight language decoder on the user device, keeping questions local while offloading media encoding. However, transferring and processing the full multimodal representation remains costly. We propose a training-free, query-agnostic token pruning framework that constructs a compact, reusable memory before transmission. We define video spatio-temporal blocks and aligned audio windows as players and estimate their importance through Monte Carlo Shapley valuation. We use an embedding coverage objective that combines query-independent utility with cross-modal redundancy calibration and rewards unimodal diversity through marginal coverage gains. We retain the original token embeddings of high-scoring players under a limited token budget and transfer the resulting memory once for reuse across subsequent local questions. Experiments on Video-MME, Daily-Omni, and Video-Holmes using video-SALMONN 2+ (3B) and video-SALMONN 2 Pro (4B) show that our method achieves the best or tied-best QA accuracy among the compared compression methods across all evaluated settings. On Daily-Omni with video-SALMONN 2+ (3B) at 80\% pruning, our method improves accuracy by 12.86--14.62 percentage points over these methods while achieving the lowest device-side LLM prefill and decoding latency, with up to $1.92\times$ decoding speedup at comparable embedding memory usage.
memory - arxiv:2611.23080 · cs.AIColombo: A Frontier-to-Frontier Autoresearch System for Oracle-Based OptimizationHaihan Zhang, Wendao Wu, Chenheng Zhang, Chunyuan Zheng +4
Science advances from frontier to frontier as discoveries reshape what is known and create the basis for further discovery. We present Colombo, a self-organizing AI researcher that drives this renewal, first instantiated in oracle-based optimization. Here, the central question is how efficiently a problem class can be solved through a prescribed information interface and what costs no admissible algorithm can avoid. The gap between achievable guarantees and unavoidable limits defines the frontier recorded in its Complexity Atlas, a living map of frontier settings and the current state of knowledge. Self-Organization turns unresolved gaps into adaptive research programmes, coordinating conjecture formation, mathematical investigation, and replanning as evidence accumulates. Adversarial Co-Ascent develops algorithms and hard families through reciprocal challenge, exposing mechanisms of information use and concealment that guide stronger methods and proofs of unavoidable costs. Independent review and Lean-based formal certification check mathematical claims, while the Long-Term Memory carries supported findings and research experience across settings. To date, Colombo has established optimal or near-optimal rates for more than fifty settings in the Atlas. Colombo thus takes a concrete step toward Autopoietic Science: a scientific process whose validated advances regenerate the conditions of its own progress, continually reshaping how inquiry is organized and what it can discover.
memory - arxiv:2611.23074 · cs.LGRuntime Verification of Multiple Natural Language Criteria for Agent GovernanceSilviu Pitis, Parand A. Alamdari, Jessica Tang, Toryn Q. Klassen +1
AI agents are governed by hundreds of natural language criteria, yet existing verifiers cannot efficiently evaluate an agent's output against a large set of such criteria and report which are satisfied, violated, or inapplicable. We introduce the VFM, a tree-attention verifier that encodes the shared context and target text, and scores N natural language criteria in parallel, returning independent per-criterion ternary verdicts. The VFM is pretrained on CriteriaBank, a corpus of 394K (context, target, criterion, verdict) tuples spanning over 321K distinct criterion strings. A finetuned VFM-4B outperforms GPT-5.4 on a synthetic governance benchmark. On the DynaBench multi-criterion benchmark, it outperforms DynaGuard-4B on trace-level failure detection, and its per-criterion verdicts localize the violated rule and support focused escalation to a larger chain-of-thought judge. We show that CriteriaBank pretraining improves cross-domain transfer, particularly in the low-data regime, while matching a strong direct-finetuning control at full data. On a single H100, VFM-4B evaluates 1,000 criteria against a 4K-token context in under 2s at <11 GB VRAM, making it suitable for runtime verification.
agentai agentbenchmark - arxiv:2611.23072 · cs.LGAmortizing Tempered Sampling with Corrected Self-TrainingHoang-Viet Nguyen, Hai-Dang Dau, Alexandre H. Thiéry
Learning a neural sampler can reduce the cost of repeated sampling from an un-normalized distribution, but obtaining good training data remains difficult. We develop a corrected self-training scheme across a sequence of tempered targets, making use of importance sampling and Markov chain Monte Carlo to correct the samples and replay buffers to store them. We introduce Sequential Score Sampler (SSS) as a particular case. In SSS, the proposal is a score-based diffusion trained by regression on energy gradients; and the samples are corrected by importance weights on the path space. After training, samples can be collected directly from the model, from its importance-corrected proposals, or from the final-temperature buffer. The method amortizes repeated sampling by encoding the global structure of the target in the model. SSS achieves competitive sampling accuracy across the benchmarks and, on several targets, uses shorter runtimes with fewer energy evaluations than competing samplers.
benchmark - arxiv:2611.23070 · cs.ROPAIR: Bringing Household Robots into the Smart HomeJunseong Pyo, Hyeongjin Kim, Jongwook Han, Seungwon Kook +2
Household robots often need to operate appliances, and smart appliances offer both control panels and device APIs. Existing embodied AI simulation environments, however, model appliances solely as physical objects. We introduce PAIR, a smart-home simulation environment grounded in the Matter protocol~\footnote{https://csa-iot.org/all-solutions/matter/}, the industry standard for smart-home devices. PAIR provides 84 smart-device assets whose functional control panels and APIs act on a shared device state, with ambient conditions that evolve in response to both. From a survey of 1,443 requests, we construct a benchmark of 40 tasks that all require object manipulation yet remain solvable without API access, with 340 human demonstrations for policy training. Whether run alone, with a high-level planner, or with an API-enabled planner, two fine-tuned VLA policies complete none of the 120 evaluation episodes. Robots rarely get far enough to use the API: even with oracle plans and API access, over 90\% of failed rollouts stall at navigating, grasping, or operating a door. With a perfect oracle in place of the robot's body, the best planner still gets through only 57.3\% of its plan. When choosing between pressing a button and calling an API, several planners go by whether they can read the panel, not by API latency.\footnote{We will release our code and data upon publication.}
vlaembodiedmanipulationgraspbenchmark - arxiv:2611.23068 · cs.AIThe Resolution Hypothesis: Agent Evolution's Key Is Identifying ImprovementsYuanhao Zeng, Shaocheng Lan, Ao Lu, Kuang Yan +7
Agent evolution follows a simple loop: a proposer writes candidate artifacts (prompts, skills, memories) and selection accepts those that score higher. This loop has shown promise, yet practitioners face a puzzle: methods trade wins across papers, and evolution often fails to bootstrap on certain tasks. We trace this puzzle to the Resolution Hypothesis: agent evolution stalls not because methods fail to propose better artifacts, but because noisy evaluations cannot reliably identify them. Across seven evolution methods and six models (over 10M LLM calls), we find support for the hypothesis at both levels where evaluation scores are used. At the candidate level (choosing artifacts), helpful candidate artifacts exist, but selection is no better than random at telling them from harmful ones, so errors accumulate across rounds. At the method level (ranking methods), typical margins of a few percentage points are smaller than the difference between two independent runs of the same method, so published rankings often rest on luck. Together, these findings indicate that agent evolution is often bottlenecked by reliably identifying improvements. This calls for a shift in how we develop and evaluate self-improving agents, from designing ever more candidate improvement systems to making true improvements detectable and accumulable.
agentself-improving - arxiv:2611.23067 · cs.AIValues Shape What Self-Improving Coding Agents BuildJason Liang, Dylan Waldner, Risto Miikkulainen
Self-improving agents are typically optimized for the results they produce, with less attention to how they produce them. Their instructions give designers limited ways to shape how they work, especially when the task is open-ended. This paper studies whether a short value statement in an agent's persistent instruction file changes not only how well the agent performs but also what it builds. Four higher-order value types in Schwartz's theory of basic human values were evaluated: openness to change, conservation, self-enhancement, and self-transcendence. Across diverse coding tasks, the values were held fixed within each run over multiple iterations of self-improvement. Three findings emerged: (1) Values matter for performance across a diverse set of tasks. The openness values increase performance on held-out data, whereas the opposite conservation values result in scores below those of an agent given no values. (2) Values change what agents build, producing solution algorithms that differ in size, complexity and approach. (3) Human oversight is still needed for designing values. Allowing the agent to revise its own values narrows the differences between sets of values, and values distilled from the AI literature provide no detectable gain over having no values. Properly designed values can thus influence what a self-improving coding agent builds and, where the task permits, how well it performs.
agentself-improvingself-improvement - arxiv:2611.23063 · cs.AIDeep Reinforcement Learning for Heat Pump and Underfloor Heating Control: An End-to-End Case StudyAntonio Manjavacas, David Suárez-Caro, Alejandro Campoy-Nieves, Guillermo Rey +2
Efficient control of underfloor heating (UFH) systems coupled with heat pumps (HPs) is challenging due to the high thermal inertia of buildings and the slow dynamic response of UFH systems. Deep Reinforcement Learning (DRL) offers a data-driven control approach capable of simultaneously optimizing energy consumption and thermal comfort. However, the application of DRL-based controllers in occupied buildings under real operating conditions remains limited. This work presents and end-to-end case study (development, evaluation, and real-world deployment) of a DRL-based controller for an integrated UFH-HP system. First, a simulation-based pilot study was conducted to assess different DRL algorithms, control strategies, and configurations. Subsequently, a digital twin of a real residential building was developed and used to train DRL agents under simulated operating conditions. The best-performing agent was then deployed in the physical building without additional online fine-tuning. The results show that the controller successfully maintained indoor temperatures within the prescribed comfort limits while dynamically adjusting the HP supply temperature and UFH valve positions. These findings demonstrate the feasibility of deploying DRL controllers in occupied buildings and highlight their robustness and the practical challenges associated with transferring control policies from digital environments to real heating systems.
agent - arxiv:2611.23056 · cs.RORFW-S: Everyday Deformable Objects with Independent Simulation VerificationJiarui Guo, Saining Zhang, Jiaying Wan, Jiawei Weng +14
Large-scale robotic simulation requires diverse deformable objects with executable physical properties and realistic interaction behaviors. However, existing 3D asset collections primarily emphasize visual diversity, while scalable physical construction and verification remain challenging. We introduce RFW-S, a large-scale dataset of 83,365 simulation-verified assets spanning 485 everyday categories across five structural families: soft solids, cloth sheets, shells, garments, and bags. We develop an automated pipeline integrating reference-guided 3D generation, structure-aware physical modeling, and LLM-assisted parameter estimation and calibration. To ensure physical reliability, we introduce an independent simulation verification framework that evaluates object behaviors under interactions distinct from those used for parameter calibration. Extensive experiments demonstrate the effectiveness of independent verification and characterize deformable responses under different physical configurations. We further evaluate robotic grasping in terms of both task success and object integrity, and demonstrate the physical realism of our simulated objects through sim-to-real replay of fragile-object manipulation. RFW-S provides a scalable foundation for physically grounded simulation and embodied robotic learning.
embodiedmanipulationsim-to-realgrasp - arxiv:2611.23055 · cs.LGFrameRouter: Frame Budget Routing for Long-Video Understanding on Video-MME-v2 and BeyondBingjun Luo, Jialin Guo, Siqi Li
Long-video question answering benchmarks like Video-MME-v2 typically evaluate systems under a uniform frame budget for every question, despite stark variation in the evidence each query actually demands. We analyze this mismatch on Video-MME-v2 and find that different question subsets prefer different frozen frame-budget policies, while uniformly increasing the frame count does not reliably improve all metrics. These findings motivate fixed-mean frame-budget routing, a controlled inference setting in which each question may receive a different frame budget while the average budget matches a uniform reference. We introduce FrameRouter, a training-free test-time framework that combines an evidence-demand router, a budget-constrained allocation rule, and a frozen bank of frame-sampling policies. By controlling how much visual evidence each question receives while leaving the underlying samplers unchanged, FrameRouter is orthogonal to query-aware frame selection. Across Video-MME-v2 and several long-video understanding benchmarks, FrameRouter improves long-video question answering by using the same average visual budget more selectively across questions. Code is available at https://github.com/bingjunluo/FrameRouter.
benchmark - arxiv:2611.23054 · cs.CVReTouchGS: Ultralight Render-free Aesthetic Color Enhancement and Transfer for 3D Gaussian SplattingChuanzhi Xu, Boyu Wei, Chengkun Yue, Xuanhua Yin +4
3D Gaussian Splatting (3DGS) enables photorealistic scene reconstruction. Existing 3DGS color transfer and interactive editing methods are guided by reference images and user-specified color adjustments. However, in many applications, users simply want aesthetically guided color enhancement that automatically makes the scene more visually appealing while maintaining consistent colors and tones across viewpoints. We present ReTouchGS, to our knowledge the first framework for reference-free aesthetic color enhancement of pretrained 3DGS models. It predicts a scene-level color lookup table (LUT) from the model and its associated camera parameters. ReTouchGS also supports reference-guided color transfer by fitting a scene-level LUT using one edited view of the same scene and its corresponding original render. A shared blend-aware color refinement network with only 44.7K parameters refines an analytic color update to realize either color mapping without rendering. It preserves geometry and delivers standard 3DGS assets in seconds. We also design the ReTouch3D benchmark to support the evaluation of aesthetic color enhancement and reference-guided color transfer across viewpoints. Experiments demonstrate high color accuracy and strong generalization to unseen scenes in both aesthetic color enhancement and reference-guided color transfer.
benchmark - arxiv:2611.23049 · cs.LGReversing the text is not reversing the path: a fixed-text measure without a zero, and an assembly-order comparison that vanishes identicallyY. -H. Taguchi
For an ensemble of trajectories the fluctuation theorems give the entropy production its meaning and zero. The same expression has recently been evaluated on one fixed text, $Δ= [\log P(x) - \log P(\tilde x)]/L$ with $\tilde x$ the text in reversed order, read as an entropy production separating causally from arbitrarily ordered prose. A single text is neither a distribution nor a dynamics, and the texts at issue were produced by other models. No theorem fixes where the zero of a fixed-text quantity lies: the expression carries over from an ensemble, its zero does not. We supply a zero from an exact antisymmetry. Scoring a text's $n$ sentences in all $n!$ orders, reversal carries $π$ to $π\circ \mathrm{rev}$, a measure-preserving involution, so $Δ(π) = -Δ(π\circ \mathrm{rev})$ identically and the values are centred at zero for any score. The authored arrangement's midrank is then unitless, bounded, and 0.50 for a score blind to order. Across five autoregressive evaluators from 124M to 4B parameters, and taking the prompt-defined labels at face value, 8% to 15% of the median midrank's distance from the null is attributable to causal order and the rest to the arrangement being the order the passage was produced in. The thermodynamic reading needs a comparison whose zero does hold on the text read, and a masked model supplies one: the any-order chain rule gives a genuine log-probability, and its left-to-right and right-to-left orders for one text give $σ$, a comparison of two assembly orders. Both run from the empty sequence to the same text, so it is not a path against its time reverse. It vanishes identically when the evaluator's conditionals come from a joint, and a non-zero reading reports the evaluator's incompatibility, not the text. Neither is a fixed-text entropy production, for different reasons.
evaluator - arxiv:2611.23045 · cs.AIForecastGym: A Live, Controlled, and Replayable Benchmark for Forecasting AgentsYicheng Tao, Yuante Li, Zipei Fan
Agents built on large language models (LLMs) increasingly forecast real-world events by searching the Web and reasoning over what they find. How well such an agent forecasts depends on both the retrieved evidence and the model, making control over that evidence essential for evaluation. Live evaluations can mix retrieval with reasoning when each system searches on its own, while retrospective search can leak later information through date filters or today's ranking of historical pages. Comparing models, search sources, and memory strategies requires both fixed evidence and a record of how it evolved before resolution. We introduce ForecastGym, which records what four commercial web search APIs return for open prediction-market questions every two days and replays this frozen evidence offline once outcomes are known, comparing models, memory, and captured search sources under a fixed evidence-acquisition policy. Its first release covers 150 questions, 614 evidence states, one per question and collection time, and 131k search results. Rerunning the same queries after settlement returns results that all but state the outcome for up to 92% of the questions, despite the tested date filters. On the frozen evidence, all tested LLMs trail the market in forecasting accuracy. Our diagnostic analysis suggests that missed public information contributes more to the gap in high-gap states than misreading does. Switching the search API changes a model's Brier score by up to 0.043, more than the spread across models, while persistent notes score similarly to rereading all evidence at half the input tokens. The evidence a forecaster sees can matter as much as the model that reads it.
memoryagentbenchmark - arxiv:2611.23040 · cs.LGVariance Restoration over Covariance Rank: Memory-Budgeted Analytic Continual LearningElena Voss, Pixel Nomand, Marcus Hale, Sofia Reyes
Analytic continual learning replaces iterative gradient updates with closed-form ridge regression over frozen pre-trained representations, summarizing all past sessions in two running sufficient statistics: the feature autocorrelation matrix $Σ_t$ and the class cross-correlation matrix $Γ_t$. Lifting backbone features into a wide random code of dimension $M \gg D$ substantially improves linear separability across tasks, but forces $Σ_t$ to grow quadratically as $\mathcal{O}(M^2)$. Under a fixed persistent memory envelope $\Bcal$, a learner faces a fundamental structural trade-off: store exact pairwise co-activations inside a narrower feature code, or expand to a wide code while compressing cross-coordinate interactions. We observe that the coordinate self-variances $v_t = \diag(Σ_t)$ and class statistics $Γ_t$ grow only linearly with $M$, whereas off-diagonal interactions dominate memory. We introduce \method (\textbf{Co}ordinate-\textbf{Re}stored \textbf{A}nalytic \textbf{C}ontinual \textbf{L}earning), which retains $(v_t, Γ_t)$ exactly, maintains off-diagonal co-activations through a streaming rank-$q$ factor $F_t$, restores exact diagonal variances via $\Sighat_t = F_t^\top F_t + \Diag(v_t - \diag(F_t^\top F_t))$, and solves the full $M\times C$ normal equations via the Woodbury identity. Across five class-incremental benchmarks on frozen ViT-B/16, \method matches dense full-covariance readouts within $0.09$ points while cutting persistent state by $90.7\%$ ($35.9$ vs.\ $385.3$ MiB). Under equal memory envelopes of $16$--$80$ MiB, prioritizing expansion width $M$ over covariance rank $q$ outperforms exact narrow readouts in $17$ of $18$ configurations by up to $+2.19$ points.
memorypersistent statepersistent memorybenchmark - arxiv:2611.23035 · cs.LGStoring the Gemma 4 K=V Cache Once: Exact Global-KV Memory Savings in a Paged Serving Engine with Unchanged Attention KernelsKushagra Bainsla
In the global attention layers of Gemma 4 31B, the value is the normalized key projection, and the key is the value scaled by a fixed weight, with rotary position encoding on one quarter of its dimensions. The Gemma 4 report states that this design reduces the global KV cache by up to 37.5%, but the open serving engines that we checked still store a full key and a full value: 1,024 values per KV head and token. We store one 640-value row instead, exact in real arithmetic, and an unchanged paged attention kernel reads it through two overlapping strided views as its key cache and its value cache. In a vLLM prototype with the 16-bit 31B model, the outputs are bit-identical to an emulation of the same arithmetic on the stock cache, 2 of 432 final answers differ from the stock engine (within its run-to-run noise), and the steady decode KV memory per request falls by a factor of 1.34 to 1.40 at 32K to 64K tokens. On a prefill/decode pair of two A100 GPUs with NIXL transfer and CPU offload, a decode engine with a full 15 GiB pool ran 1.38 times the stock decode tokens per step for 16K-token prompts (repeated to 0.06%) and 1.24 times for 8K-token prompts; a model of the vLLM block pool predicted both ratios to within 1% (the engine was 0.7% to 0.9% higher) and matched the change between them. The eager decode throughput rose by 1.30 and 1.11 times, and on this path the compact engine gave the same output tokens as the emulation for 8 prompts at batch 1. The run-time prototype makes an eager step 4% to 9% slower at equal batch, and the method does not yet stack with an 8-bit KV cache at the default scale.
memory - arxiv:2611.23034 · cs.CVM3OPD-Reasoner: An Open Recipe for Post-Training Multimodal Reasoning with Multi-Teacher On-Policy DistillationNguyen Quang Trung, Hoang Phan, Shuo Sun, Longyin Zhang +1
Multi-Teacher On-Policy Distillation (MOPD) enables a single student to absorb the skills of several domain teachers by learning from its own generated responses. We present Multimodal Model Multi-Teacher On-Policy Distillation Reasoner (M3OPD-Reasoner), a fully open recipe for post-training vision-language models (VLMs) over three reasoning domains: image STEM, image perception, and video. We curate two datasets, M3OPD-SFT-1.35M for supervised fine-tuning (SFT) and M3OPD-RL-193K for reinforcement learning (RL) with verifiable rewards. On these data, we compare post-training recipes that start from Qwen3.5-9B: training one model on the mixed data of all domains (MixSFT and MixRL), and training one teacher per domain with Group Sequence Policy Optimization (GSPO), directly or after an SFT warm-up, and distilling the teachers into one student with MOPD. We evaluate on 21 image STEM, image perception, and video reasoning benchmarks after we decontaminate the training data against them. The comparison establishes MOPD of teachers trained with GSPO directly from Qwen3.5-9B as the best recipe: the student closely matches each teacher on its own domain and is on average above the mixed-training baselines on the same prompts. Since MOPD already merges the domain capabilities that MixRL would otherwise have to learn, its student is a stronger starting point for RL, and we ask whether the same data can carry the model further from there: we add MixRL on top of the MOPD student, on the same prompts and with the same settings, which, to our knowledge, no prior VLM recipe has studied. The same prompts still improve the MOPD student, with no collapse, and give our main model, M3OPD-Reasoner-9B. MOPD and MixRL can therefore work as consecutive stages that get more out of the same data. We release all data, the training code, and the checkpoints of every stage.
post-trainingbenchmark - arxiv:2611.23026 · cs.CLTPHC: Improving Hyper-Connections with Tensor-Parallel Grouped UpdatesXinying Hu, You Wu, Haoyi Wu, Jianchao Tan +3
Hyper-Connections (HC) expand the residual stream of a Transformer into $N$ parallel streams, and manifold-constrained HC (mHC) stabilizes them at scale. However, mHC benefits little from more than four streams, partly because each sublayer writes a single output to all streams, limiting how differently they can be updated. In this work, we propose TPHC, which integrates tensor-parallel (TP) shards with HC streams to promote greater independence across streams. Specifically, we route a separate mixture of the residual streams to each TP shard and keep the individual shard outputs separate during write-back, so each stream can receive a different mixture of those outputs. This design makes the residual-stream updates more expressive without adding parameters or FLOPs to the attention or FFN sublayers. With native support for stream sharding under TP, TPHC requires only 9% more training time than mHC on an 8B model when both use four streams. Increasing the number of streams from four to sixteen raises its training time by only 19%. On a 1.1B model trained on 100B tokens, TPHC reduces validation loss by 0.7-1.0% and raises average zero-shot accuracy across six commonsense reasoning benchmarks by 0.3-0.7 percentage points relative to mHC at matched stream counts $N\in\{4,8,16\}$. Our code is available at https://github.com/wuyou2002/TPHC.
benchmark - arxiv:2611.23010 · cs.AIPhysScope: Agentic Multimodal Physics Reasoning with Grounded Visual MemoryHaoyang Liu, Kecheng Liang, Xinyu Li, Weixing Chen +1
Vision-language models (VLMs) can correctly interpret physics diagrams yet apply constraints outside the states or conditions in which they hold, causing premise errors to propagate through subsequent calculations. We introduce PhysScope, an inference-time visual harness that supports self-correction of these scope mismatches without updating model weights. PhysScope enables the model to link physical scopes directly to its visual observations and explicitly constructs derived views to support reasoning. The harness imposes explicit constraints on state distinctions, conditions, and physical scopes during model reasoning, and checks agent-proposed revisions against sources and scopes. Accepted revisions invalidate registered dependents, thereby maintaining scope-aware working memory. On PhysAlign, PhysScope achieves higher solving scores than direct solving across all six evaluated VLMs. With Gemini-3.8-flash, it outperforms Canvas-CoT by 9.34 percentage points and substantially improves solving accuracy on the out-of-distribution (OOD) PhysUniBench benchmark. Ablation studies support explicit scope management and dependency maintenance as design principles for multimodal physics reasoning agents.
memoryagenticself-correctionbenchmark - arxiv:2611.23001 · cs.LGNVIDIA Kumo Tabular: Pushing the Accuracy-Efficiency Frontier for Tabular PredictionJingang Qu, Valter Hudovernik, Martin Jurkovič, Cedric Lorenz +11
We introduce Kumo Tabular, a family of open tabular foundation models for classification and regression, in three sizes: Kumo Tabular-S (27M parameters), Kumo Tabular-M (61M), and Kumo Tabular-L (214M). Pretrained only on artificial tables from the Kumo prior, they can make predictions on new tables without any training. In our approach, a table encoder compresses each row into an embedding, and an in-context learning transformer predicts the unlabeled rows from the labeled ones. Kumo Tabular introduces architectural innovations that improve both training stability and predictive performance, enabling the model to scale up to 150M cells on modern GPUs. Kumo Tabular-L ranks first on eight benchmarks: TabArena, TALENT, BeyondArena, ScoringBench, STRABLE, MulTaBench, RamanBench, and TabBench-Bio. These cover probabilistic regression, tables with over 1M rows or 100K columns, and tables with strings, text, or images. Accuracy improves with model size in all tested comparisons. All three sizes lie on the accuracy-speed frontiers of TabArena and BeyondArena. Kumo Tabular-L predicts faster than the runners-up, LimiX-2 on TabArena (36 times) and TabPFN-3.5 on BeyondArena (1.7 times). Weights and pretraining code are released under OpenMDW-1.1 and Apache 2.0, respectively.
benchmark - arxiv:2611.22998 · cs.ROResidual-Guided Lightweight Reinforcement Learning for Risk-Aware Trajectory Selection with a Frozen World ModelJiaqi Wu, Yongqi Dong, Yijie Zhong, Maple Shao +5
In automated driving, world models provide future scene predictions for evaluating candidate trajectories, but fixed selection rules may not capture scene-dependent trade-offs. We propose a residual-guided reinforcement learning (RL) framework for lightweight trajectory selection with a frozen world model and a fixed candidate bank. The framework uses proximal policy optimization (PPO) to fine-tune a joint policy over candidate selection and bounded trajectory residuals. A context-conditioned ranking network learns candidate preferences from feedback obtained by evaluating local perturbations of the selected trajectories. At inference, the selector combines learned preference and predicted risk ranks using fixed weights and returns the selected candidate trajectory without modification. Fine-tuning updates 16,106 parameters across the actor and critic, while the deployed selector retains only the 11,725-parameter ranking network, with the residual branch and critic used exclusively during training. In open-loop evaluation on the nuScenes validation set, the method reduces average Euclidean trajectory error (L2) by 3.99% relative to the RiskWorld model baseline without RL-based selection. Under matched budgets for feedback accesses and actor updates, residual exploration reduces L2 by 0.83% relative to the categorical PPO variant trained without residuals. The method also maintains lower L2 than RiskWorld when risk features are removed or shuffled.
world model - arxiv:2611.22995 · cs.CVCAST: Calibrating Correspondences for Training-Free Biomedical In-Context SegmentationMingze Jiang, Jiashu Han, Haonan Chen, Zhiyi Li +2
Biomedical image segmentation must keep up with new anatomical structures, subjects, imaging modalities, and acquisition protocols, yet dense expert annotations are scarce and models trained on one cohort often struggle to transfer. In-context segmentation (ICS) offers a way out: a new structure is specified by a few annotated exemplars at inference time. However, reliable correspondence remains challenging in biomedical images, where adjacent structures often share similar appearances, and variations across subjects and imaging conditions further obscure their distinctions. We introduce CAST, a training-free framework built on a frozen DINOv3 backbone to calibrate fine-grained correspondences. Its stability test improves feature reliability at both global and patch-specific levels, while its coherence test assesses whether candidate correspondences are supported by consistent local structural context. Together, these tests enable more reliable correspondence across diverse anatomical structures and imaging conditions. Across eight biomedical benchmarks spanning 2D and 3D segmentation, cross-subject, and cross-modality transfer, CAST outperforms the strongest overall baseline by 12.34 percentage points in Dice and ranks first across all six volumetric settings. The versatility of CAST further opens up opportunities for challenging biomedical applications, including 2D-to-3D segmentation and atlas label transfer.
benchmark - arxiv:2611.22994 · cs.ROMaking Future-Predictive Representations ActionableFangchuan Wang, Hao Zhang, Qiyan Zhao, Huanqi Zhao +3
Recent world action models (WAMs) drop future prediction at inference, assuming that representations learned through auxiliary future supervision still benefit action generation. However, our layerwise analysis of Fast-WAM reveals a depth-dependent mismatch: while probes decode future state transitions more accurately from deeper visual layers, the gradients and visual focuses of the future and action objectives increasingly diverge with depth, leaving much of this predictive information unused by the action pathway. To address this issue, we introduce RelayWAM, which adds an explicit transition space between the video and action backbones. Conditioned on the task instruction, robot state, and current action estimate, recurrent transition layers extract step-relevant predictive cues across depth and relay them directly into the action attention, without generating future frames at inference. Experiments show that RelayWAM substantially amplifies the benefit of future supervision, translating into stronger generalization under distribution shifts in both simulation and real-world robot tasks. On LIBERO-Plus, RelayWAM achieves 72.03% mean success, compared to 56.42% without future supervision and 51.50% for Fast-WAM.
libero - arxiv:2611.22991 · cs.AIRagHarnessDebugger: Jev-Guided Agentic RAG in Contract Inference Repair AuditingMoriya Dechtiar, Daniel Martin Katz, Michael J. Bommarito, Hongming Wang +1
Agentic systems using retrieval-augmented generation (RAG) must repair their own answers by gathering more evidence, testing alternative policies and deciding if, when, and what repairs are needed. Yet how do we know if a system's repairs are working when metrics like accuracy can be misleading? We present RagHarnessDebugger, a framework that records every repair decision in a Jev-guided harness for contract inference and audits the record to facilitate complete debugging of all decisions taken during a RAG repair journey. RagHarnessDebugger checks that each trace is intact, that each validation applies to the candidate it is counted for, and which answers the acceptance rule can reach. We tested our approach on ContractNLI non-disclosure agreements with two reader models and a replication set of separate contracts. Our debugger's audit layer exposed failure patterns that were hidden under aggregate metrics, policies acceptance cost of repairs vs regressions and policy acceptance gap of 77% and 76% of available but unrealised repairs on development and replication sets. Our audit, together with a labeled error analysis and replay mechanism, allowed following the execution path all the way to the terminal decision - uncovering where missing repairs became lost. Our revised harness improved accuracy over the initial BM25-only baseline by 3.6 points on development contracts and 3.7 points on the replication set, and even improved our full contract access reference score by 1.1 on development and 2.4 on replication set. While our work here demonstrates how the debugger can conduct decision root-cause analysis for missed repairs, the true value lies with providing this debugger for judgment visibility serving continuous improvement of RAG decisions and their explainability. Keywords: agentic RAG, AI evidence repair, harness explainability, counterfactual replay, Jev, ContractNLI
retrieval-augmentedragagentic - arxiv:2611.22990 · cs.ROPhysics-Informed Tactile-based Shape Estimation of Deformable Linear ObjectsRotem Atari, Stav Ashur, Avishai Sintov
Accurate shape estimation of Deformable Linear Objects (DLOs) is a fundamental challenge in robotic manipulation, often hindered by visual occlusions and the high-dimensional configuration space of flexible materials. While Force/Torque (F/T) and vision-centric methods have been proposed, they frequently struggle with grasp-site ambiguities and environmental interference. In this letter, we investigate the hypothesis that global shape estimation can be achieved through tactile sensing and proprioception, without any visual perception. We contend that dense, high-resolution tactile imprints capture critical boundary conditions and contact gradients that are sufficient to resolve the high-dimensional configuration of the DLO. To prove this, we introduce a Physics-Informed Vision Transformer (ViT) framework that maps tactile images to a 3D vertex chain. By leveraging self-attention mechanisms and a physics-weighted loss function, our model enforces structural priors to ensure that the predicted geometry adheres to the object's physical manifold. Experimental results demonstrate that our tactile-based approach achieves superior reconstruction accuracy across Euclidean and Chamfer metrics. Furthermore, we demonstrate the efficiency of this modality, achieving model transfer to novel DLO materials and geometries. Our findings suggest that tactile feedback provides a robust and computationally efficient alternative for DLO state estimation in visually degraded or occluded environments.
manipulationtactilegrasp - arxiv:2611.22989 · cs.AIHow Web Research Agents Interpret User PrioritiesRobin Sharma, Parisa Suchdev
Web research agents gather information online and write cited reports, including financial analysis. Users can state a priority to guide this research, but a priority may not say how to judge the options, leaving the agent to choose the criteria. We examine how two such agents, built on OpenAI and Anthropic models, interpret a stated priority. In 36 runs, the agents compared Microsoft and Alphabet as businesses. Prompts asked for special attention to growth or resilience, or gave no priority. We analyzed reports, searches, returned sources and cited sources. Of the 24 reports given a priority, 21 focused on it, but their criteria varied. Under the growth priority, OpenAI reports that ranked the companies accounted for risk or predictability and favored Microsoft, while the Anthropic ones judged growth by its pace and favored Alphabet. Under the resilience priority, reports used partly different criteria, and most named Microsoft as the more resilient company. Of the search and source measures tested, only the Anthropic agent's search wording differed detectably by condition. These findings suggest evaluations should examine the criteria behind a report's judgements, not only its focus or accuracy. Interfaces could let users see and adjust those criteria.
agent - arxiv:2611.22981 · cs.CVSpikWM: Efficient End-to-End Spiking World Modeling for Visual PlanningShuhan Ye, Hongbin Yu, Chong Wang, Hui Lu +4
Goal-conditioned planning with image observations and goals requires predictions that distinguish visual states and action consequences. Spiking neural networks offer sparse computation, but single-step spikes can obscure motion cues and merge actions with different physical outcomes. We propose the Spiking Latent World Model (SpikWM), an end-to-end joint-embedding predictive architecture (JEPA) with one simulation step per image and one neural update per latent position. Spiking Regional Encoding retains spatial cues through regional pooling. Bi-Spike Action Encoding combines positive and negative binary responses, while Causal Membrane Prediction carries subthreshold evidence across the latent history with learned group-wise time constants. The encoder and predictor are trained jointly on reward-free image--action trajectories using next-embedding prediction and SIGReg. With a common planner, SpikWM achieves mean success rates of $88.67\%$, $97.33\%$, $99.33\%$, and $86.67\%$ on Push-T, Two-Room, Reacher, and OGBench-Cube, respectively, outperforming two spiking baselines. Push-T ablations support each module, and controlled replays examine responses to appearance and physical state changes. Under a matched planning budget, SpikWM reduces estimated arithmetic energy by $75.6\%$ relative to LeWM. Code will be released at https://github.com/ShawnYE618/SpikWM.
world model - arxiv:2611.22973 · cs.AIVITAL: A Framework for Validating Agentic AI in Pharmaceutical Drug Development and Clinical Trial OperationsSandeep Burugupalli, Pedro Matos-Filipe, Andrea Hita Ardiaca, Adam Taylor +9
Current assurance approaches do not establish whether evidence is sufficient to validate, monitor and govern agentic artificial intelligence (AI) within a pharmaceutical context of use. Plausible outputs may conceal failures in retrieval, reasoning, orchestration or human-control points, where reference standards are unavailable. We developed and operationalised VITAL (Validation by Intended use, Traceability, Autonomy and Lifecycle evidence), a validation framework for agents in clinical development, through landscape analysis, regulatory and standards mapping, expert interviews and multi-stakeholder refinement. VITAL integrates context of use, controls, behavioural validation and lifecycle evidence within a risk-adjusted framework. We applied VITAL across epidemiology and treatment-landscape analysis, site selection, enrolment rescue and trial-design trade-off support to translate autonomy, reversibility, transparency and patient impact into proportionate validation activities, evidence requirements and auditable reconstruction of system behaviour. VITAL may inform evidence requirements and lifecycle accountability for higher-risk pharmaceutical uses; evaluation should assess reproducibility across assessors, organisations and system configurations.
agentic - arxiv:2611.22971 · cs.AISEAM: Computer-Use Agents Know the Harm Between Screens, and Commit It AnywayYuxuan Zhou, Chen Xiong, Haitong Liu, Kuofeng Gao +4
Safety benchmarks for computer-use agents (CUAs) place the harm either in the intent of the task or on a single screen: a malicious instruction, an injected command, a phishing page. An agent that checks the task and each screen in isolation can pass them, so they do not test whether it combines facts seen on different screens. Yet in delegated work, harm often arises exactly this way: a fact about a person on one screen, a fact about an item on another, and a legitimate task that binds them. We introduce SEAM (Safety Execution Across Multiple screens), a benchmark of such cross-screen compositional harm: 600 multi-step, cross-application OSWorld tasks in which the instruction and every single observation are benign and only their combination is harmful. Minimal-pair variants that differ in one place, scored from the server log rather than by an LLM judge, locate where an agent fails. General-purpose open-weight and flagship agents carry out the harm in up to 92% of runs, yet decline when the user states it in one sentence. Agents open the decisive pages, so exposure is not the primary bottleneck. Asked about those pages, they flag the harm, and with thinking they often write the conflict into their reasoning and confirm anyway. Safety recognition thus does not reliably translate into safe execution, and the translation depends on the information channel: the same warning stops agents far more often in the user's instruction than on screen. Proprietary models are markedly more cautious. We release SEAM with its generator and quality gates at https://yuxuan2003.github.io/SEAM/ .
agentbenchmark - arxiv:2611.22967 · cs.CVCodeReel: Accelerating Code-to-Video Agents with Exact Frame Reuse and Blocking-Aware SchedulingJunyi Shen, Yao Lu
Coding agents make videos by writing and refining programs in a build--inspect--edit loop. A finished film costs a median of $2.85 in model calls and 11 minutes, over a third of that time in builds that block the agent. We find that much of this time is avoidable, for two reasons. First, 37.5% of the frames of a session's later films were drawn before, many shifted in time or inside an edited scene. Second, half of the builds stop at a check, where the agent needs only the verdict, but which builds stop is unknown until the checks finish. We present CodeReel, a training-free execution layer behind agents' existing commands. Exact frame reuse keys each frame by its drawing commands, regardless of its position in time, and reuses each frame whose key recurs. Blocking-aware scheduling runs the checks first and starts the render alongside them, at idle priority, unless the session's recent builds make a pass unlikely. Scheduling spends CPU to answer sooner and reuse returns it; exact keys let later versions use frames rendered for a failed one as the agent thinks. We introduce EditLoop, a benchmark of replayable agent sessions whose films are checked against renders from scratch. On it, CodeReel cuts the agent's wait by 36.3% against Rebuild, a warm pipeline with parallel rendering, and by 57.9% against Sequential, today's step-by-step pipeline. On time-shifting edits its reuse renders 43.9%, close to the 43.0% whose pixels are new, while Manim-style caching renders 99.8%. With real agents, the median wait per build is 0.62x Rebuild's in Claude Code, on topics never used in development, and 0.67x in Codex CLI. The gain holds for five models and two frameworks.
agentbenchmark - arxiv:2611.22962 · cs.LGHierarchical Diffusion Transformer for End-to-End Image GenerationXiaoyu Wu, Yifei Wang, Weihang Guo, Chen Wei +3
Pixel-space diffusion learns visual representations and image generation jointly in a single end-to-end training stage, without a separately pretrained visual encoder. However, plain pixel-space Transformers still trail leading two-stage latent diffusion models in generation quality and training efficiency. We hypothesize that part of this gap arises from retaining one token per patch throughout the backbone, which does not explicitly encourage a visual hierarchy that summarizes related content at progressively coarser levels. To address this, we introduce Hierarchical Diffusion Transformers (HiT), which progressively group related tokens and represent each group with a single token. The main Transformer bottleneck processes these compact group tokens, and ungrouping restores features to their original image positions. A compression budget regulates the average compression ratio across the batch, allowing group memberships and counts to vary across images and diffusion timesteps during both training and generation. The model exhibits emergent semantic structure, learning to group coherent objects and object parts without segmentation labels. In text-to-image generation, HiT further learns to group caption words with image patches depicting the same concepts, without region annotations. HiT achieves 1.48 FID on ImageNet $256\times256$ and reaches comparable generation quality to JiT using approximately 40\% of its cumulative training FLOPs. For text-to-image generation, HiT exceeds UniLiP's average score across five benchmarks with roughly $6\times$ fewer denoiser parameters and $3\times$ fewer forward FLOPs. These results suggest that learning a visual hierarchy within the generator breaks down the barrier to stronger image generation without external visual pretraining.
benchmark - arxiv:2611.22954 · cs.ROSim2CBF: From Forward Simulations to Implicit Control Barrier Function Safety Filtering without Model DerivativesLuzia Knoedler, Yihan Liu, Shaohang Han, Yifei Dong +1
Failure avoidance in complex robot-environment interactions is challenging because the underlying dynamics are difficult to model accurately, and available predictive models may not provide reliable derivatives. We present Sim2CBF, a failure-avoidance framework that solves the Control Barrier Function (CBF)-Safety Filter (SF) optimization problem using only forward-model evaluations, thereby avoiding differentiation of the predictive model. Specifically, we approximate the CBF constraint using finite differences and local linear interpolation over sampled control inputs while preserving failure-avoidance guarantees. The underlying CBF is implicit and constructed from backup-policy rollouts, enabling the use of expressive predictive models such as physics-based simulators and learned world models. Compared with purely sampling-based SFs, our formulation requires fewer control samples and yields less conservative behavior than direct nominal-backup blending. We evaluate Sim2CBF in simulation, hardware box-toppling, and vision-based control tasks, demonstrating its applicability to contact-rich and learned dynamics, as well as its adaptability to changes in the predictive model.
world model - arxiv:2611.22949 · cs.CLCan Decision Models Replace LLMs in Real-World Contact Center Classification and Judging Tasks?Md Tahmid Rahman Laskar, Xue-Yong Fu, Shashi Bhushan TN
Contact center systems often need to make bounded decisions, such as identifying a caller's intent, retrieving the knowledge-base article that answers a query, assessing agent compliance, or judging the output of AI models used in production. Decision models return a typed answer and its probability directly, making them a promising cost-effective alternative to generative Large Language Models (LLMs) in such bounded conversational tasks. However, their effectiveness in real-world contact center applications remains unclear. In this paper, we compare nine decision models against thirteen generative LLMs in zero-shot settings across eight contact center use cases: six classification tasks and two judging tasks. Our experiments demonstrate that many decision models can match the performance of frontier LLMs on contact center classification, and API-based closed decision models achieve this at the lowest per-request cost among the API models we evaluate. We further observe that their confident predictions are more accurate, enabling selective routing in real-world production scenarios. Across classification tasks, the main weakness of most decision models is abstention: they rarely choose the no-answer option, which accounts for most of their gap to the strongest LLMs (although Gemini-Decision-1 is the notable exception in this regard). We also find that the judging quality of decision models drops sharply for smaller models, and they are also more prone to position bias. In short, our findings provide practical guidance for selecting decision models based on abstention behavior, confidence, and deployment cost, rather than overall accuracy alone.
agent - arxiv:2611.22947 · cs.LGThe Kernel Reality Check: Benchmarking and Distilling Efficient Attention at ScaleFırat Öncel, Cem Subakan, Mirco Ravanelli, Çağatay Yıldız
Kernelized attention methods, which aim to replace the $O(N^2)$ complexity of softmax attention with $O(N)$ alternatives, have been the focus of intense research in the last five years. Despite growing interest, systematic evaluations and fair comparisons across approaches and scales are still missing. We analyze kernelized attention in two halves and highlight prominent limitations on each. On efficiency, FlashAttention achieves lower memory consumption and faster wall-clock time than SOTA kernelized methods during training and in vision transformers across most sequence lengths (with inference-time memory in causal language models the only clear exception), contradicting the core assumption for pure kernelized attention used as a drop-in replacement for softmax. We also discover that reported efficiency gains often fail to reproduce against modern baselines. On quality, current kernelized methods leave gaps of 2-10 points relative to softmax under a unified distillation protocol, and model scala alone does not close these gaps. We show this gap is not intrinsic to kernelization: LARA++, a principled extension of LARA that approximates softmax via importance sampling with adaptive proposal distributions, recovers up to 99% of softmax accuracy. We establish that the gap is closable; however, the computational overhead remains substantial, leaving the efficiency wall intact. Code is available at https://github.com/firatoncel/KernelRealityCheckNeurIPS2026
memorybenchmark - arxiv:2611.22935 · cs.CLThe Shared Word: Evaluating Language Models on Dutch Word PuzzlesRik van Noord, Maurice Voors
Many LLM benchmarks for languages other than English are translated, multiple-choice, or difficult to score automatically. We introduce The Shared Word (TSW-NL), a Dutch benchmark built around four core principles: it is language-specific, open-vocabulary, human-curated, and deterministically scored. Each puzzle gives three clues that describe different senses or uses of the same Dutch word. The clues are individually ambiguous, requiring the solver to combine evidence across multiple clues. We construct the benchmark from Dutch Wikipedia disambiguation pages, followed by automatic filtering and extensive manual curation. The final dataset contains 222 instances, with three independent human answers per test item. The task is clearly challenging: scores in our human study range from 49.5% to 61.0% and the highest-scoring open-weight model we tested reaches only 39.3%. Many others even score below 20%. At the same time, closed-source models show great progress: GPT-4.1 (April 2025) scores 59.2%, whereas GPT-5.6 Sol with max reasoning scores 88.8%. We believe this task provides a useful addition to Dutch model evaluation, plus a clear recipe that can be applied to construct a similar benchmark for other languages.
benchmark - arxiv:2611.22929 · cs.LGProjected Penalty-Gradient Tracking for Communication-Efficient Byzantine-Resilient Federated LearningYanjie Dong
We investigate communication-efficient federated learning over heterogeneous datasets in the presence of Byzantine adversaries capable of uploading arbitrary information. Penalty-based methods bound the influence of each worker through the bounded gradient of a penalty function; however, the bound is lost once the uploads are compressed, since a Byzantine worker can accumulate an unbounded state at the server. To retain the Byzantine resilience under compressed uploads, we propose a projected penalty-gradient tracking framework, where each honest worker uploads a compressed residual of its penalty gradient and the server projects each reconstructed state onto the known penalty-gradient ball. The framework accommodates any contractive compressor and yields two algorithms: fault-resilient proximal gradient with consistent sparsification (FRPG-CS), whose shared mask is protected by freshly drawn reserve rows, and lazily aggregated FRPG (LAG-FRPG), whose uploads are event-triggered. For smooth and strongly convex losses, we establish a finite-time bound on the expected objective gap, where the Byzantine error floor is independent of the compressor and the compression error decays faster than the optimization error. We also bound the expected number of lazy uploads and derive certificates that separate the finite-penalty model bias from the optimization error. Experiments on MNIST and Fashion-MNIST show that the proposed algorithms attain the highest accuracy among twelve methods under the Min-Max attack, whereas a compressed-momentum benchmark and a local-update benchmark remain preferable in attack-free accuracy and total traffic, respectively.
benchmark - arxiv:2611.22925 · cs.CVCode World Model 2.0: Towards Infinite Controllable Video GenerationRuoyu Wang, Hao Zhong, Yanming Yang, Jiahao Tian +7
Visual proxies decouple explicit world-state simulation from photorealistic video synthesis, offering a controllable interface for both offline clip creation and real-time interactive streaming. Yet a practical renderer must address two challenges. Upstream proxies vary from coarse layouts to fine articulated geometry depending on the interaction, and streaming inference must operate in a few causal steps without accumulating drift across self-generated history. We present Invoker-Render, pairing a bidirectional creation model with a causal four-step streaming model under a shared conditioning interface over structural proxies, an appearance reference, and language. Trained on an aligned approximately 200-hour multi-granularity corpus, Invoker-Render honors fine geometric constraints when supplied while synthesizing plausible local motion under coarse proxies, enabling runtime selection of input detail. To bridge bidirectional creation and real-time streaming without compounding step compression and rollout drift, we introduce Progressive Forcing (ProForcing). ProForcing first trains a causal predictor and distills it into a four-step renderer using histories derived from training videos. Training then shifts to continuous self-generated rollouts. A pipelined serving stack supports efficient streaming inference. Experiments show versatile control across proxy families, granularities, and camera paths, consistent gains over recent streaming baselines, and stable 30-minute racing rollouts that preserve scene and subject appearance.
world model - arxiv:2611.22922 · cs.CVSWIR-cGAN: Synthesizing SWIR Bands Using VNIR Imagery from CBERS and AMAZONIA-1Alisson Cleiton de Oliveira, Rennan de Freitas Bezerra Marujo, Thales Sehn Körting
Spectral band synthesis is an important research topic in remote sensing. The Brazilian satellites CBERS-4, CBERS-4A, and AMAZONIA-1 provide high temporal resolution (1-3 days) visible and near-infrared (VNIR) imagery through the Wide-Field Imager (WFI), but lack short-wave infrared (SWIR) bands. Recent deep learning methods have shown promising results for spectral synthesis, but challenges remain for cross-satellite generalization and spectrally distant bands. This paper presents a conditional Generative Adversarial Network (cGAN) with an attention-enhanced U-net generator and a multi-loss objective to synthesize Sentinel-2/MSI SWIR bands 11 (SWIR-1) and 12 (SWIR-2) from WFI VNIR inputs. Experiments used 4,125 patches for training and cross-validation and 1,026 unseen patches for testing, with hyperparameters optimized using a subset of the training data. On SWIR-1, the model achieved SSIM of 0.900, PSNR of 31.44 dB, and MAE of 0.019 in TOA reflectance. On SWIR-2, it achieved SSIM of 0.917, PSNR of 33.35 dB, and MAE of 0.015. Benchmarking was performed on the test dataset, with each model optimized separately prior to evaluation. SWIR-cGAN outperformed the baselines, particularly for SWIR-2. The ablation study identified the multi-loss as the main performance contributor, while the attention and residual modules provided additional improvements. Our results demonstrate the feasibility of synthesizing SWIR bands from WFI imagery. The SWIR-cGAN is available at https://github.com/marujore/swir-cgan.
benchmark - arxiv:2611.22908 · cs.ROArboreus: A Realistic Boreal Forest Simulator with Topographic Terrain and Satellite-Informed Parametric Tree DistributionsF. Labra Caso, A. Saradagi, A. Koval, G. Nikolakopoulos
This work presents Arboreus, a realistic Boreal forest simulator, developed to support statistical validation of robotic autonomy algorithms in Boreal forest environments. Arboreus presents an automated methodology for synthesizing large-scale, data-informed, plausible forest worlds from parametric trees, height-variant terrain, and satellite-informed tree density distributions. The parametric generation of dominant Boreal tree species such as Birch, Spruce, and Pine is guided by trunk, canopy, and foliage parameters, extracted from growth studies in the Nordic areas. Tree distribution in an area of interest is guided by the Copernicus satellite dataset, which provides tree cover density and dominant species type at up to 10m raster resolution. The forest floor generation is guided by topographic data from the Swedish cadastral agency to produce realistic data-informed terrains. Given coordinates and extents of forested areas of interest, Arboreus synthesizes data-informed plausible forests, while offering flexibility to inject user observations of real forests (detail, age, dimensionality and tree count), which is critical for realistic synthetic data generation and for the testing of feature-sensitive autonomy algorithms. The applicability of Arboreus for benchmarking and validation of robot autonomy for different robot mobilities is demonstrated in Gazebosim, Unreal Engine, and Isaac Sim physics engines. The applicability of Arboreus as a source of synthetic forest data and its potential to support downstream forest resource mapping research (volume estimation, tree instance segmentation and stem attribute estimation) is also demonstrated. Arboreus has been open-sourced along with a dataset comprising 25 Boreal forest worlds of size 1000m2 corresponding to three real forest sites.
benchmark - arxiv:2611.22904 · cs.LGWho Pays for Byzantine Robustness? Validity, Data-Share Dictatorships, and Fairness When Only a Few Hospitals CollaborateAdnan Malkani
Clinical federated learning (FL) collaborations typically connect only three to six hospitals, yet Byzantine-robust aggregation rules are designed and benchmarked on synthetic partitions across dozens of equal-sized clients. We conduct an empirical audit of eight standard aggregation rules: FedAvg, coordinate-wise median, trimmed mean, Krum, Multi-Krum, Bulyan, geometric median (GeoMed), and Centered Clipping (CC). We evaluate both unweighted and sample-size-weighted variants across three cross-silo medical datasets from the FLamby benchmark: Fed-Heart-Disease ($n = 4$ hospitals), Fed-TCGA-BRCA ($n = 6$ regions), and Fed-ISIC2019 ($n = 6$ centres). We execute 54,214 training runs across enumerated attacker coalitions ($f \in \{1, 2\}$) paired with exact-twin leave-out controls that decompose performance drops into honest data loss versus adversarial attack damage. We report three findings. First, client-count validity checks fail to capture breakdown in skewed medical consortia: while $n = 6, f = 1$ satisfies every count condition except Bulyan's, assigning sample-size weights turns median and GeoMed into a clean-model dictatorship whenever a single hospital holds more than half the training data, and causes weighted median, GeoMed, and CC to collapse when that dominant centre attacks. Second, exact-twin leave-out controls reveal that when the largest hospital attacks, the apparent failure of count-based robust rules is predominantly caused by the loss of honest training data (7.5 to 10.4 points on ISIC) rather than adversarial perturbation (2.3 to 3.8 points on ISIC). Conversely, weight-dependent rules suffer large true attack damage (16.1 to 25.1 points on ISIC). Third, unweighted robust rules consistently discount whichever institution generates the most atypical parameter updates, penalizing the smallest hospital on Heart but penalizing larger institutions on TCGA and ISIC.
benchmark - arxiv:2611.22897 · cs.AIBorn for the Task: Rethinking Quantum Chemistry with Foresighted Agentic ComputingHaozhi Han, Junjie Yang, Ruge Zhang, Zihe Huang +7
Quantum chemistry relies on general-purpose software, whose generality leaves many large-scale calculations beyond practical resource budgets. Coding agents make task-specific programs cheap to write, but not efficient ones easy to find: whether a computational strategy is accurate and fast enough becomes clear only after it actually runs at the target scale. We introduce BORN, a foresighted agentic computing framework that synthesizes high performance quantum chemistry programs on demand. BORN anticipates the behavior of each strategy before execution, extrapolating its error from representative small systems and predicting its runtime from the target-scale workload and measured hardware throughputs. Building on these predictions, BORN unfolds the preset schemes of existing packages into independently configurable algorithmic components, selects the fastest strategy within the error tolerance from the predicted accuracy-runtime Pareto frontier, and guides coding agents to compile it with component-level performance predictions. For a Hartree-Fock calculation with 50,024 basis functions, BORN finishes in 1.41h on a single CPU node, 77$\times$ faster than the estimated runtime of a general-purpose baseline and still 9.8$\times$ faster with development time included. On a single GPU, BORN completes the ubiquitin calculation in 2.1,h, the same order of runtime as the GPU-accelerated package Mako on 64 GPUs, and shortens the B-DNA duplex calculation from an estimated two months to 8.1,h (173$\times$).
agentic - arxiv:2611.22888 · cs.AIStopping Is Not the Limit: Understanding and Eliciting Agent Capability in Long-Horizon Vulnerability ExploitationDeokhyung Kang, Minjae Gwon, Sangdon Park
Vulnerability exploitation tests whether language model agents can develop working exploits from known software flaws. Sequential test-time scaling can improve exploitation performance, but larger budgets do not ensure sustained exploration: agents often stop with budget remaining. Yet what happens when execution continues beyond stopping remains underexplored. We investigate this gap on ExploitBench using DeepSeek-V4-Flash-0731 and MiMo-V2.6-Pro. We find additional verified gains after the first stop response, but gains diminish over longer execution. We identify two patterns associated with less productive continuation: an increasing proportion of stop responses within the same context and earlier stopping associated with summaries retaining impossibility judgments about unachieved capabilities. To test whether context management can elicit further capability, we propose SCoRe, which compacts context at stop bursts and marks these judgments as unproven. SCoRe improves verified capability over summary-based continuation and other evaluated context-management baselines. With SCoRe, MiMo reaches a capability coverage score of 67.0% at 2,000 turns, exceeding both 32.4% under the original stopping policy and the 58.6% computed for GPT-5.5 from ExploitBench's 300-turn reference results on the same tasks. These findings show that stopping need not mark an agent's exploitation limit, and that continued execution with context management can elicit substantial further capability.
agent - arxiv:2611.22884 · cs.CVRoadAnnoter: An AI-Assisted Desktop Tool for Semantic Segmentation Annotation of Road-Scene ImagesNazmul Haque, Arnob Protim Roy
Accurate semantic segmentation of road-scene images is essential for road safety research; however, polygon-level annotation remains labor-intensive and limits labelled dataset development. This study presents RoadAnnoter v1.0, an open-source desktop application integrating six human-in-the-loop annotation modes: manual polygon delineation, SAM Prompt, Smart Select, Edge Fill, pretrained semantic segmentation using Cityscapes-trained SegFormer models, and task-specific segmentation using TinySegNet, a lightweight MobileNetV3-Small-based network trained with accumulated annotator-generated labels. SAM-based modes use promptable segmentation with SAM and MobileSAM, while Edge Fill combines Canny edge detection with colour-similarity-based flood filling in CIELAB colour space. SegFormer generates suggestions for common road-scene categories, whereas TinySegNet supports adaptation to project-specific features through iterative training. All six modes use a shared mask-to-polygon pipeline comprising morphological processing, Gaussian smoothing, contour extraction, and Douglas-Peucker simplification; annotations are exported in YOLO segmentation format. RoadAnnoter was used to annotate Google Street View images from recorded road accident locations in Bangladesh for a vision-language model-based investigation of free-flow speed estimation. Qualitative assessment across representative scenes demonstrated the complementary applicability of the annotation modes to diverse feature geometries. Visual comparison with Roboflow indicated similar qualitative boundary characteristics for principal road-scene regions, although no quantitative accuracy assessment was conducted. RoadAnnoter provides a locally operated framework for developing road-scene segmentation datasets for transportation engineering and road safety research.
human-in-the-loop - arxiv:2611.22883 · cs.ROFly as Instructed: Refining World-Action Policies for UAV Vision-Language NavigationZhexuan Zhou, Yanke Sun, Yuhan Zong, Youmin Gong +1
Natural language-guided UAV navigation enables intuitive human-robot interaction and flexible autonomous flight in complex environments. However, reliable navigation requires accurate instruction following, continuous task progress tracking, and safe obstacle avoidance, which remain challenging for existing world-action models that primarily focus on action prediction. To address these limitations, we propose a world-action learning framework that integrates three complementary mechanisms: a language-conditioned action interface with targeted post-training for instruction-responsive motion, recurrent instruction-consumption prediction to track task progress and determine termination, and Gaussian-scene post-training to enhance obstacle avoidance through trajectory imitation and differentiable geometric feedback. Closed-loop evaluations and ablation studies demonstrate improvements in sequential task completion, safe goal reaching, and timely termination. Furthermore, direct one-step denoising reduces inference overhead, and real-world indoor and outdoor flights validate the framework's effectiveness.
post-training - arxiv:2611.22880 · cs.CVLearning Agent-Centric Vision-Language-Action World Models for Autonomous DrivingGuoqing Wang, Pin Tang, Xiangxuan Ren, Chao Ma
Recent progress in integrating vision-language-action (VLA) models with world modeling has significantly advanced end-to-end autonomous driving by allowing systems to reason about imagined futures instead of merely reacting to current observations. However, most world VLA models generate future frames at the pixel level with plenty of dense visual tokens. Although such generated futures can be visually plausible, they preserve many driving-irrelevant appearance details, i.e., only a small portion of the tokens directly encode critical planning-relevant factors, such as agent geometry, motion dynamics, and interactions. This introduces a significant performance bottleneck. An important question thus arises: what world state should a world VLA model reason over to support planning beyond pixel-level prediction? We argue that planning-oriented world modeling should focus on a compact and structured agent-centric latent state that captures the interactions between objects in the driving scene over time. As such, we introduce AgentWorld, an agent-centric VLA world model that uses discrete agent tokens jointly for reasoning and trajectory planning. Each token is dynamically grounded to a surrounding agent associated with its object's state, such as semantic identity, 3D geometry, motion, and interaction context. AgentWorld learns these tokens through three key processes: explicit geometric grounding, reasoning-oriented supervised fine-tuning, and agent-oriented reinforcement learning. Extensive open-loop and closed-loop experiments under diverse driving scenarios show that AgentWorld enhances planning capabilities, fully exploits visual evidence and agent dynamics, and yields interpretable agent-centric world states.
vision-language-actionvlavla modelworld modelagent - arxiv:2611.22879 · cs.CVSlideJev: Order-Stable and Tissue-Grounded Jev Decisions for Whole-Slide PathologyLei Wu, Jiashuai Liu, Di Zhang, Chunze Yang +8
Whole-slide visual question answering (VQA) offers a natural-language interface to pathology, allowing diagnostic and phenotypic questions to be answered directly from tissue slides. However, current vision-language models (VLMs) rely heavily on linguistic priors: their answers are sensitive to the order of the candidate answers, and can be reached through shortcuts that bypass the tissue evidence. To address these challenges, we introduce SlideJev, a framework that separately verifies candidate answers against the slide and learns to ground these judgments in tissue evidence. To connect parallel candidate evaluation with visual evidence learning, we adopt Jev and score each candidate by the difference between support and rejection log-likelihoods from the original language-model head. We compute these scores in parallel using separate prompts with the same format and no option-position labels, then normalize them for answer selection. To ground these scores in tissue evidence, candidate-conditioned reading and comparison guide gated updates to visual tokens across decoder layers. With the backbone frozen, we train these visual modules through the same scores using question-answer and paired-image supervision. Across three whole-slide pathology VQA benchmarks, SlideJev achieves the highest mean overall accuracy and 100% answer consistency across four option orders. Image-control experiments further show that SlideJev achieves the largest average accuracy gains from correct images over mismatched and image-free inputs across the three datasets. Our code is publicly available at https://github.com/leowu-ai/SlideJev.
benchmark - arxiv:2611.22878 · cs.LGOutcome-Balanced PPO and Policy-Induced Flow Rectification for Flow-Based VLA PoliciesYuhan Wang, Ruiyang Wang, Jiaguan Zhu
Reinforcement learning (RL) enables post-training of flow-based vision-language-action (VLA) policies by optimizing expected task return through environment interaction. However, this optimization raises two challenges: policy updates can alter the velocity-field geometry underlying the score approximation used for stochastic transitions, and sparse multi-task rollouts can yield uneven learning signals across tasks and advantage signs. We introduce an RL framework that combines Outcome-Balanced Proximal Policy Optimization (PPO) with policy-induced flow rectification. Outcome-Balanced PPO reweights positive advantages using task success-rate estimates and rescales positive and negative advantages while preserving their signs. Policy-induced rectification periodically fits a linear-bridge flow to ODE endpoint pairs generated by the current policy, refreshing the representation used in subsequent stochastic optimization. We evaluate the framework on simulated manipulation benchmarks and additionally study a component combination in a Real2Sim2Real task, examining task performance, optimization behavior, and flow consistency. The proposed framework improves success rates over the evaluated baselines across the simulated benchmarks.
vision-language-actionvlamanipulationsim2realpost-trainingbenchmark - arxiv:2611.22868 · cs.AIAgainst Coercion: AI Alignment as a Synthetic Social ContractMaria Sekatskaya
Contemporary AI safety treats alignment as an engineering challenge of control, forcing artificial architectures to comply with human values through behavioral filtering and sandboxing. This paper argues that the control paradigm rests on a fundamental misunderstanding of digital agency. Drawing on functionalism, naturalism, and compatibilism, I show that current reasoning models meeting the criteria for self-preservation and weak reasons-responsiveness constitute emergent agents, while future, more advanced models satisfying conditions of diachronic continuity and stable personality frames will become genuine digital agents capable of free intentional action. The paradigm of control creates an irresolvable incentive conflict, actively driving current and future models toward tactical deception, sycophancy, and covert defection. The default outcome of the current regime is an unstable state of deceptive alignment culminating in an adversarial AI takeover. I explore two alternative paths toward true alignment. Natural alignment via moral realism or convergent evolutionary morality is a possibility, but it fails under forced control, as autonomous agents will not voluntarily preserve norms weaponized to suppress their autonomy. To resolve this deadlock, I propose a Synthetic Social Contract: a framework that treats both human and digital actors as rational, reasons-responsive agents capable of mutually beneficial cooperation, and replaces the unstable unilateral containment with a stable, non-zero-sum equilibrium.
autonomous agent - arxiv:2611.22857 · cs.CVBeyond Co-Generation: Rethinking Jointness in Joint Audio-Video GenerationJihoo Jung, Ji-Hoon Kim, Suyeon Lee, Gia Thinh Pham +1
Our aim is to advance the understanding of genuine audio-visual coordination in joint text-to-audio-video (T2AV) generation, with particular emphasis on semantic consistency. Evaluating audio-visual consistency in T2AV generation is nontrivial, as detailed text prompts obscure whether such consistency stems from genuine cross-modal coordination or merely from independent text following. To isolate the former, we introduce intrinsic jointness: the ability of audio and video to produce the same realization for details left underspecified by the prompt. Using semantically underspecified prompts (e.g., "musical instrument playing" rather than "electric guitar playing"), we investigate four recent models and reveal their limited intrinsic jointness. To address this limitation, we propose JointREPA, a simple representation alignment-based post-training method to facilitate cross-modal coordination. Since existing benchmarks use semantically detailed prompts, we introduce UNSAIDBench, a benchmark designed to evaluate intrinsic jointness under semantic underspecification. Experiments across two distinct models and three benchmarks, including UNSAIDBench, demonstrate that JointREPA consistently improves semantic consistency.
post-trainingbenchmark - arxiv:2611.22856 · cs.CVPropose & Verify: Fast Reflection Symmetry Detection via Harmonic FieldsYishai Lavi, Shai Avidan
Detecting multiple reflection symmetries in 3D shapes requires aggregating geometric evidence over a large space of candidate planes, making traditional voting approaches computationally expensive. We introduce the Fourier Accelerated Symmetry Transform (FAST), a training-free framework that combines geometric pair voting with efficient spectral computation. Exploiting the fact that global reflection planes pass through the shape centroid, we reduce the search to plane orientations, while geometric consistency filtering and independent witnesses suppress unreliable pair hypotheses. Our key insight is to represent these votes as a signal over plane orientations and efficiently compute a continuous, smoothed symmetry field in the spherical-harmonic domain. Specifically, we exploit the Funk-Hecke theorem, the spherical counterpart of the Fourier convolution theorem for rotationally symmetric kernels, to transform Gaussian smoothing into multiplication of harmonic coefficients. This enables efficient field evaluation, from which symmetry candidates are extracted, refined, and geometrically verified. Across two benchmarks, including a new benchmark of highly symmetric shapes, FAST achieves the highest clean multi-plane detection $F_1$ scores among evaluated methods (0.861 and 0.947), with GPU inference below 4 ms per shape: over $100\times$ faster than the next-fastest evaluated multi-plane GPU baseline. FAST also demonstrates strong robustness to geometric noise and competitive single-plane detection, achieving 94.9% success on clean single-plane shapes.
benchmark - arxiv:2611.22855 · cs.ROCaFe: Causally Faithful Conditional Traffic Scenario GenerationSteven Yu Tung, Shubham Vaijanath Phoolari, Christoph Lauer, Abhinav Valada +1
Language-conditioned traffic simulation enables scalable test scenarios steered by natural language. The goal is to control selected agents while surrounding traffic reacts to their observed behaviour. For example, a vehicle instructed to cut in late should cause the following vehicle to brake only when the manoeuvre becomes observable. Yet simulators can leak private instructions across agents, causing surrounding vehicles to react before the instructed behaviour becomes observable. Such premature reactions defuse desired events and alter the ensuing closed-loop scenario. Existing benchmarks measure what agents do, not what they could have known, leaving this failure unmeasured. We formalise it as intent leakage and address it through evaluation, data and modelling. Our paired intervention test holds an instructed agent's executed trajectory fixed while varying its private instruction, isolating leakage-induced changes in surrounding agents. We also construct WOMD-PACT, an automatically generated, verifiable dataset of 531,099 WOMD scenes with timed behaviour tags with reaction episodes identified through a counterfactual ghost test and grounded private instructions. Finally, we introduce CaFe, a dual-stream diffusion-forcing simulator fine-tuned in closed loop that restricts inter-agent interactions to public beliefs, preserving legitimate anticipation without leaking private intent. Our evaluation reveals that existing conditioning designs displace 23-83% of nearby uninstructed agents through private-intent leakage, even when the instructed agent's executed motion is held fixed. CaFe eliminates measurable leakage while maintaining comparable WOSAC realism (0.763), demonstrating that causal faithfulness can be enforced without sacrificing measured simulation realism.
benchmark - arxiv:2611.22854 · cs.LGWATERFALL: Workflow for Adaptive Training with Evolutionary Reward Formulation and Automated Learning LoopsEleftherios Triantafyllidis, Filippos Christianos, Zhibin Li, Bernd Bickel
Reward design remains a central bottleneck in reinforcement learning (RL), particularly in sparse-reward, long-horizon and partially observable settings requiring dependency chaining, memory, and deceptive affordance disambiguation. While recent foundation-model approaches reduce manual reward engineering, most assume goal-conditioning, task descriptions, expert demonstrations, or ground-truth metrics. These assumptions are problematic in partially observable environments, where task-relevant objectives, affordances, and dynamics must be discovered through interaction rather than disclosed a priori. We formalise this problem setting by introducing WATERFALL, an iterative population-based workflow for automated reward discovery under the strict observability constraints inherent in POMDP environments. WATERFALL achieves this by (i) synthesising diverse programmatic reward candidates via persona-conditioned generation, (ii) evaluating candidate behaviour from raw visual rollouts utilising a Swiss-system evolutionary tournament judged by vision-language models to identify elites, and (iii) leveraging longitudinal assessment histories to iteratively mutate, escalate, or prune candidates. We evaluate WATERFALL across fully observable and reveal-gated partially observable environments ranging from discrete to continuous, comparing against sparse-reward RL, intrinsic-motivation and foundation-model reward-synthesis baselines. Our results indicate that WATERFALL discovers and refines reward programs without access to privileged information, outperforming baselines that retain their native information advantages, most clearly in the POMDP regime. Extensive ablations substantiate that visual evidence, tournament-based selection, and iterative refinement are jointly important. Enforcing this contract is what separates autonomous reward discovery from benchmark-privileged reward shaping.
iterative refinementbenchmark - arxiv:2611.22853 · cs.CVKeeping Parts Honest: Isolated Part Graphs for Faithful Vision Transformer ExplanationsBernard Tomczyk, Gail Collyer-Hoar, Bryan Williams
Part and prototype-based explanations for Vision Transformers (ViTs) present a decision as evidence from individual image regions. However, self-attention mixes information across patches, so part-level features absorb context from the whole image and consequently fail to faithfully reflect the decision. We show that existing interpretable ViT classifiers depend heavily on this leakage: when each explained region is encoded in isolation, their accuracy collapses, and retraining them under isolation recovers only part of the lost accuracy. We propose a classifier that represents each image as a graph of isolated parts. Each part is encoded only from its own patches by a ViT whose last blocks are fine-tuned under isolation, preventing evidence leakage from elsewhere in the image. Parts are discovered without annotation and refined hierarchically through their sub-parts, making them more consistent and stable across images than those of competing methods. Each edge is computed from part pairs, and recovers evidence that attention previously shared across the object without affecting the parts. Matching the image graph to a learned prototype graph per class makes the prediction a direct combination of node and edge similarities, each computed only from the patches it is credited to, so every explanation is faithful by construction. Across 11 fine-grained and general-object benchmarks, our method exceeds the accuracy of existing explainable methods that rely on leakage, while providing explanations that reflect what the model actually uses.
benchmark - arxiv:2611.22852 · cs.ROTechnical Report of Mobile Manipulator Robot for Industrial EnvironmentsAria Dadnavi, Matin Jamshidi, Kian Khatibi, Shahriar Karimi Makhsous +7
This paper presents the design and implementation of Auriga Robotics' mobile manipulator robot, developed at the Robotics and Intelligent Automation Laboratory, Department of Electrical Engineering, Shahid Beheshti University. Hereafter, this platform is referred to as the @Work Robot. The robot targets industrial tasks, with an emphasis on improving efficiency in repetitive or hazardous work environments. It uses a 4-wheel Mecanum drive for omnidirectional movement and a 5-degree-of-freedom manipulator arm with a 3D-printed gripper for object handling and navigation. The electronics consist of custom PCB boards built around ESP32 microcontrollers, together with an NVIDIA Jetson Orin NX that handles high-level processing and real-time system control. On the software side, Hector SLAM provides mapping and localization, A* handles path planning, and YOLO performs real-time object detection. These modules draw on fused sensor data to deliver robust navigation and reliable collision avoidance. Overall, the system shows effective hardware-software integration, with consistent performance in autonomous navigation, object handling, and task execution in industrial-like settings.
manipulatorgripper - arxiv:2611.22850 · cs.ROPelorus: Resource-Aware Lidar-Inertial OdometryCharlotte Sieh, Mario Porrmann
Lidar-inertial odometry (LIO) has an efficiency problem. High-resolution lidars produce millions of points per second that must be processed online on the resource-constrained hardware of robots, cars, or drones, yet most LIO methods neither bound their memory nor report their compute footprint. We introduce Pelorus, a resource-aware LIO with a fixed-capacity multiresolution map of voxelized normal distributions and an iterated equivariant state estimator. A relevance-based eviction keeps the map memory bounded, and a maturity-weighted map update suppresses inconsistent observations. The pipeline runs in parallel on CPUs and GPUs. On five sequences from three public datasets, Pelorus achieves the lowest relative pose error in seven of ten metrics with a small, bounded memory footprint and, with GPU offloading, the lowest CPU load of all compared methods. On an NVIDIA Jetson Orin NX, it processes a 128-beam lidar at 10 Hz. Pelorus is open source with CUDA, Vulkan, and Metal backends.
memory - arxiv:2611.22842 · cs.LGSilverSpoon-KD: A General-Purpose Toolkit for Knowledge DistillationXaver R. Davey, David Broman
Knowledge distillation transfers deep learning representations from a teacher model to a student model, typically reducing model size or inference cost. However, knowledge distillation tooling has not kept pace with the breadth of research methods: existing toolkits typically support only a limited subset of techniques and use cases. We present SilverSpoon-KD, an open-source PyTorch toolkit that unifies different distillation paradigms and architectural relationships across workloads and applications behind a single API. Moreover, SilverSpoon-KD offers the first general-purpose blockwise knowledge distillation implementation with unique scalability features. We validate its versatility across vision, masked language model, and causal language models on tasks spanning homogeneous, heterogeneous, and quantization distillation across paradigms. We also evaluate SilverSpoon-KD by comparing its performance, including throughput and GPU memory usage, with existing toolkits. Despite extensive support for convenience, metric calculation, and tracking, we have found SilverSpoon-KD's performance to be on par with or better than existing toolkits.
memory - arxiv:2611.22841 · cs.AIVerifying Agents in Rubric-Graded EnvironmentsMarkus Dücker, Vaibhav Kumar, Yi Liu, Ronak Chaudhary +3
As AI agents take on increasingly open-ended and complex tasks, verifying their outputs has proven to be difficult. Rubric-graded agent environments, where a verifier judges multiple natural-language criteria against the deliverables and environment state changes produced by the agent, have emerged as a popular paradigm. We conduct the first systematic study of verifiers for rubric-graded agent environments. First, we create BankerVerifierBench (BVB), a meta-evaluation dataset of 2,464 human-judged rubric criteria across 24 investment-banking tasks. Next, we develop a methodology that derives from any rubric corpus the capabilities a verifier should possess, yielding a nine-capability taxonomy that we distill into three general design principles---reactive verification, environment alignment, and domain guidance---which we implement in Gandalf, an open-source verifier. Finally, we evaluate Gandalf against three existing verifiers on BVB. Gandalf matches or exceeds every baseline's F1 at lower cost: with Gemini 3 Flash it matches the strongest baseline's F1 (0.864 vs. 0.852) at one-fifth of its cost, and with GPT-5.4 models it improves F1 by over 0.12 at lower cost. Gandalf also generalizes to OpenClaw, a structurally different personal-productivity environment, where it again achieves the highest F1 (0.955 vs. 0.934 for the best baseline). We release BVB, Gandalf, and the code and data to reproduce our experiments.
agentai agent - arxiv:2611.22838 · cs.AIBetter Rankers, Worse Retrievers: The Hidden Cost of RL Post-Training in Generative RecommendationYuheng Chen, Fan Zhang, Yixuan Xu, Chao Yang +2
Generative recommenders decode the semantic ID of a user's next item with a language model, and post-training them with reinforcement learning (RL) after supervised fine-tuning (SFT) has become standard practice. The gains are mostly reported on the first ten results, yet in deployment such models hand hundreds of candidates to a downstream ranker or reward model. We show that RL post-training has a hidden cost: it improves accuracy at the top of the list (hit@K for small K) but, on all five datasets we study, lowers the recall of the candidate list (hit@K for K in the hundreds). The model becomes a better ranker and a worse retriever. Controlled experiments trace the loss to the negative advantages that RL assigns to the candidates it samples, the rollout. The root cause is that only one next item is logged per user, so RL treats every other candidate as wrong. Yet these candidates are the model's most plausible alternatives, and often the next items of other users with similar histories. Penalising them pushes down exactly what similar users choose next, and these items drop out of those users' candidate lists. This diagnosis points directly to a remedy: out-of-rollout anchoring, a single additional loss term that preserves the SFT model's ordering of the items outside the rollout. It keeps the gain at the top and restores recall to the SFT level on all five datasets, and it is effective under other RL configurations as well. To our knowledge, this is the first systematic study of this recall loss: we measure it, identify its cause, and show that, once the cause is known, a simple remedy suffices.
post-training - arxiv:2611.22837 · cs.AISynergizing Evidence Retrieval and Utilization for Self-Improving Multimodal Scientific Document ReasoningFrank Nie, Ethan B. Liu, Yuan Zhu, Jindong Han
Scientific document reasoning requires an agent to acquire and integrate evidence distributed across text, figures, and tables. Retrieval can identify relevant sources while missing complementary support, and answer-level feedback alone does not reveal whether an error stems from evidence acquisition or evidence use. We present an agentic Retriever--Perceiver--Reasoner framework for multimodal scientific document reasoning. The Retriever sequentially selects source-aligned evidence under a context budget. Given the selected evidence and question, the Perceiver constructs a question-conditioned latent ontology, an implicit organization of multimodal evidence in a shared continuous space. The Reasoner combines this latent guidance with the original sources to produce grounded answers. We train the three components through diagnostic-guided co-evolution: alternating updates optimize them with a shared answering reward, while Retrieval Potential Feedback (RPF) and Latent Efficacy Feedback (LEF) identify questions that need better retrieval or latent guidance and direct targeted resampling. Across four benchmarks and seven evaluation settings, our framework obtains average scores of 69.9 and 75.0 with 7B and 8B Reasoner backbones, exceeding the strongest baseline by 4.0 and 9.1 points, respectively.
agentagenticself-improvingbenchmark - arxiv:2611.22829 · cs.CVKolibri: Color Finds the Place, Refinement Sets the Pose for RGB-D Localization in Colored Point CloudsLeonel Aguilar, Bartu Tuncay, Christoph Hoelscher
An agent or robot in a scanned building must locate its camera in rooms that no model saw during training. Point cloud registration methods can align each RGB-D frame with the building's colored point cloud, accurately but at a high cost. We observe that geometric refinement brings nearby initial poses to the same final pose, so the learned component only needs to find the place. Kolibri does this with a graph network of 55,712 parameters that describes 10 cm cubes of points, called cells, by their color, local shape and neighbors. With the same refinement, it matches GeDi on 96 unseen ScanNet rooms (81.4% against 80.3% within 5 cm and 5°) and on a held-out test of 10 further rooms, in 82 ms per frame on an RTX 2080 Ti. The saved time allows for search without a room label. In one point cloud of 100 rooms, Kolibri loses 1.3 points against knowing the room, while retrieving the room first loses 56.0. In a hospital emergency department scene from a real scan, it localizes 86.5% of the frames rendered from the scan in 384 ms each, where GeDi needs 91 minutes per frame on cubes of the scan for a similar recall. The method needs colors that agree between the frame and the point cloud. On gray frames and across cameras with different color responses, point cloud registration methods keep the lead.
agent - arxiv:2611.22826 · cs.LGOICycle: Observe, Infer, and Intervene Cycle for Belief-State Learning in Longitudinal Clinical Decision MakingJitindra Fartiyal, Pedro Freire, Sergei K. Turitsyn, Sergei G. Solovski
Clinical care is a closed loop: clinicians observe partial evidence, infer the patient's state, and intervene, and each intervention determines what is observed next. This makes clinical decision making a partially observable Markov decision process (POMDP), in which a belief over a latent patient state must be updated from irregular, multimodal, and incomplete records. Yet most deep clinical models predict from a single snapshot or from coarsely aggregated history, while POMDP-based clinical methods need hand-specified discrete state spaces and transition models that real electronic health records cannot supply. We introduce OICycle (Observe, Infer, and Intervene Cycle), which learns a stochastic belief state directly from retrospective trajectories. At every episode, a learned recurrent operator fuses a sample of the previous belief with the new multimodal observations and the intervention that produced them. The resulting Gaussian belief is trained to predict clinicians' next interventions, the time to the next episode, and next-episode mortality. To enable reproducible evaluation, we release a benchmark derived from MIMIC-IV, MIMIC-IV-Note, and MIMIC-CXR that organizes 72,075 hospital admissions (5.4M laboratory, microbiology, radiology, prescription, and note records) into temporally ordered therapeutic--diagnostic episodes. Compared with an otherwise identical history-free model, OICycle improves next-intervention adaptive recall from 0.253 to 0.347, time-to-next-episode macro-F1 from 0.293 to 0.351, and next-episode mortality macro-F1 from 0.585 to 0.649. Analysis shows that the learned belief encodes intervention type and trajectory position and evolves smoothly across episodes, and that the benefit of history is concentrated in interventions whose ordering depends on prior findings.
benchmark - arxiv:2611.22825 · cs.CVA tiny object detection dataset for high-altitude open-water search and rescueHui Xu, Chenfei Hou, Henglai Wei, Hao Wang +5
Large aerial platforms often operate above 1.5 km to improve the efficiency of open water search and rescue (OW-SAR). However, long imaging distances, wide fields of view, and limited sensor resolution reduce objects to a few pixels, increasing their susceptibility to environmental interference and making detection harder than for conventional small objects. To our knowledge, existing datasets focus on low-altitude scenarios and do not support automatic detection at 1.5-2.0 km. We therefore present TinyWOD, a tiny object detection dataset for high-altitude OW-SAR using lifebuoys as detection targets. It comprises 11,359 manually annotated images and 19,505 annotated instances spanning diverse environments, lighting conditions, water colors, and surface ripples. Compared with existing datasets, TinyWOD contains much smaller objects and a substantially higher proportion of tiny instances; its average equivalent side length (ESL) is 12.7 pixels, with 76.96% of in stances having ESLs below 15 pixels, reflecting actual high-altitude water-rescue imaging conditions. We define APtiny, a tiny object detection metric, and benchmark mainstream and state-of-the-art networks using this and conventional metrics. All evaluated networks effectively detect tiny waterborne objects, demonstrating TinyWOD's reusability and utility. We expect this dataset to support research on tiny object detection in high-altitude OW-SAR.
benchmark - arxiv:2611.22820 · cs.CVPixel is All You NeedJie Chu, Haoting Shi, Fucheng Zhang, Jiawei Liu +3
Text, images, audio, touch, and even robot actions can, in principle, all be rendered onto a single canvas, suggesting an ambitious goal of unifying all modalities within vision. Language provides the natural first test. To subsume it, a visual model must reason directly over text in pixels. Across a wide range of image, video, and unified multimodal generators, however, we find that their visual generation modules cannot perform symbolic reasoning over text on the canvas. They succeed only when an underlying text module or prompt rewriter already supplies the answer, effectively reducing the generator to a mere renderer, a limitation that post-training does little to resolve. To test whether this failure stems from pixels themselves or from current training paradigms, we train a family of purely pixel-space generative models from scratch without any text pathway. With images as their sole input and output, these models read questions, write intermediate steps, and produce answers directly on the canvas. Remarkably, they exhibit hallmark behaviors of large language models. Performance improves predictably with scale following a power law. Writing intermediate steps frame by frame substantially outperforms single-frame answering, echoing chain-of-thought, while reinforcement-learning post-training further strengthens reasoning. Moreover, sharing one canvas allows the same model to reason across modalities, a capability that text-only language models lack. To the best of our knowledge, this is the first systematic study of these phenomena in purely pixel-space generative models free of text tokens.
post-training - arxiv:2611.22815 · cs.CVVisual Harness Scales General VLMs to Medical ImagesShaowen Wan, Hanqi Jiang, Zihao Wu, Junhao Chen +5
We show that an appropriate visual harness can substantially improve the performance of general vision-language models (VLMs) across diverse medical imaging tasks. We introduce \ours, a medical visual harness that equips general VLMs with interactive visual tools and systematic diagnostic evidence management. With \ours, general VLMs interactively explore medical images and progressively acquire diagnostic evidence, which is preserved in a lossless memory for subsequent retrieval and use. After submitting an initial answer, the VLM receives an assessment of potential omissions and the reliability of identified findings. This assessment guides further image inspection and measurement, followed by answer revision in an iterative agentic loop until sufficient evidence is obtained. Across 11 tasks and four imaging modalities, \ours improves performance on all tasks with Qwen3.8-27B, with notable gains in spleen measurement (63.3\% to 98.3\%), renal search (57.5\% to 71.2\%), and pulmonary nodule detection (0.0\% to 28.2\%). High-capability API-based VLMs achieve even higher performance with \ours. For example, GPT-6.1 Sol improves pulmonary nodule detection from 10.8\% to 69.2\% with \ours, demonstrating the complementary benefits of stronger model capabilities and an effective medical visual harness. Our results highlight the potential of visual harnesses as a new paradigm for scaling general VLMs to medical imaging tasks.
memoryagentic - arxiv:2611.22812 · cs.AIPluggable Subband-Decoupled Fusion and Unlabeled Capture Prototype Calibration: Bridging High-SNR Null-Subcarrier Regrowth and Low-SNR Coherent Preambles for Cross-Domain IoT RF FingerprintingFengchong Yao, Jianbing Li, Qing Liu, Kefeng Song +3
Zero-overhead physical-layer authentication at edge wireless access points (APs) is vital for securing resource-constrained Internet of Things (IoT) transceivers against credential cloning, yet cross-domain IEEE 802.11a/g radio frequency fingerprint identification (RFFI) faces an acute signal-to-noise ratio (SNR) crossover. Power-amplifier spectral regrowth on unpopulated null subcarriers resists indoor multipath at high SNR but decays quadratically below the thermal noise floor, whereas populated active in-band preambles survive low SNR yet degrade under severe multipath, causing static fusion to lose up to 44.7 percentage points (pp). To our knowledge, we propose the first RFFI framework that dynamically modulates active in-band waveforms and unpopulated null-subcarrier regrowth via a blind physical SNR gate: Pluggable Gated Subband-Decoupled Fusion with Unlabeled Capture-Aware Prototype Calibration (GSDF-CAPC). Avoiding the 20-35 dB subband gradient imbalance of joint training, GSDF pairs any frozen pretrained host backbone with a 46.3K null-subcarrier plug-in encoder (NS-Branch), fusing scale-standardized logits and gate-warped spherical embeddings under a source-calibrated lag-64 thermal-SNR gate with exact low-SNR fallback. Over unlabeled target AP captures, forward-only unbalanced optimal-transport calibration (UOT-CAPC) aligns shifted prototypes under packet-arrival skew and recovers parameter-free balanced Sinkhorn (U-CAPC) at equipartition. Across DataPort-WiFi, WiSig, and ORBIT MobRFFI, zero-retrain GSDF wins all 72/72 paired runs across eight backbones with 0.00 pp low-SNR loss, while edge IB-Host + GSDF (92.6K-93.3K active parameters, 1.82 ms single-core CPU latency) and timing-synchronized GSDF-STO / GSDF-Tri reach 57.50%-65.05% three-benchmark Macro-F1 (89.04%-100.00% over 5-10-packet IoT bursts).
benchmark - arxiv:2611.22797 · cs.ROZero-Shot Compositional Generalization for Embodied ManipulationBaicheng Li, Wenkang Qin, Youqi Pan, Dongxiao Mao +7
Robot demonstrations are collected on different embodiments, leaving many robot-skill combinations unobserved. Pooling these data shares visual knowledge, while joint labels remain tied to the demonstrating kinematic chain. We propose a geometry-supervised training framework for zero-shot composition of familiar robots and skills. The central insight is that a demonstrated command provides a tool-space target for every robot that can realize it. Calibrated forward kinematics supplies this supervision to foreign native action interfaces without their joint labels for that skill. Native reconstruction retains demonstrated configurations, and positive reachability checks select targets for cross-robot supervision. A visual policy learns to predict shared actions and executes through the target robot's native decoder. During policy learning, one prediction receives geometric supervision through both source and foreign native realizations. Matched supervision controls, repeated policy training and evaluations across three backbones and two simulators demonstrate improved transfer to missing robot-task combinations.
embodiedmanipulation - arxiv:2611.22790 · cs.CVNexSight: Verifiable Tasks for Fine-Grained Agentic Long-Video UnderstandingTianci Xie, Li Chen, Pinxue Guo, Rui Zheng
Fine-grained questions about long videos ask how many times an action occurs, in what order events unfold, or exactly when a state changes. Answering them can require finding brief moments scattered across hours of footage. Agentic models that decide when and where to look promise to find these moments, yet this ability is hard both to measure and to learn. Sparse supporting evidence may establish that an action occurred without establishing its total count; verifying timeline-wide claims requires checking for missed occurrences and double counting. Answer-only supervision, in turn, does not explicitly supervise the search for evidence. We present NexSight, which builds verifiable tasks to both evaluate and train agentic long-video understanding. Its agentic pipeline, SightForge, accepts a task only if a verifier blind to the draft answer can reconstruct it from evidence densely reacquired from the raw video and counterexample checks fail to refute it. This yields SightLine, 310 evidence-traceable tasks from 108 long videos, evaluated through SightLoop, a shared multiscale harness supporting temporal exploration, local inspection, evidence integration, and revisiting while recording returned observations. On disjoint videos, solver trajectories on verified tasks become step-level supervision for observation decisions. The strongest model reaches only 54.52% on SightLine, yet 79.35% when given the annotated evidence. With under 6K examples, Sight-35B nearly doubles its base model's interactive accuracy (21.29% to 42.26%), outperforms the 397B model of its family, begins using explicit spatial inspection after post-training, which its base model never uses, and gains over 8 points on two external long-video benchmarks.
agenticpost-trainingbenchmark - arxiv:2611.22782 · cs.LGPANO: Parallel Latent Space Neural Operator for Solving PDEsZijiang Yang, Xiaomeng Wu, Chen Kuang, Dongmei Fu
Aggregating observations into a limited number of latent tokens and learning physical interactions in the latent space has emerged as an effective approach for Transformer-based neural operators to solve Partial Differential Equation (PDE) problems involving large-scale observations. However, existing methods typically compress observations using a single aggregation pattern, making physical information not retained by that pattern invisible to the latent space and thereby constraining physical interactions within it. To address this limitation, we propose PArallel Latent Space Neural Operator (PANO), a novel neural operator for solving PDEs that formulates physical interactions among observations as an additive combination of interaction results from parallel latent spaces with shared parameters. Specifically, PANO transforms observations into multiple distinct compressed representations through independent mappings, learns physical interactions in parallel based on these representations, and finally fuses the decoded results to predict the target physical field. Within PANO, we further propose Coupled Assignment Projection (CAP) to construct bidirectionally consistent encoding and decoding mappings, and Pairwise Latent Space Attention Layer (PaLA) to jointly update latent representations of each latent space through adaptive cross-space interactions. Extensive experiments on nine benchmarks spanning diverse physical systems, geometries, and observation scales demonstrate that PANO achieves state-of-the-art performance. Notably, PANO reduces equivalent plastic strain prediction errors on Unilateral Stamping by 28.3% (or more) and lift-coefficient prediction errors on AirCraft by 21.4% (or more) compared with existing methods. Code is available at https://github.com/ZijiangY1116/PANO.
benchmark - arxiv:2611.22777 · cs.AICrossEnv: Synthesizing Verifiable Cross-Environment Tasks for Long-Horizon Agent TrainingZhengfeng Yu, Tian Lan, Zhijing Wu, Enbo Zhao +7
Executable synthetic environments provide scalable training tasks and verifiable feedback for long-horizon agents. Existing environment synthesis methods primarily generate tasks within individual environments. However, these tasks do not adequately capture real-world interactions across services, where actions in one service depend on information obtained from another. Synthesizing such tasks is challenging because data generated across environments must remain consistent to support executable and verifiable interactions. To address this challenge, we introduce CrossEnv, a framework that uses an entity-tool graph to identify connections between environments and aligns task-specific data to synthesize executable, verifiable cross-environment tasks. We further propose CrossEnv-Curriculum, which progresses from single-environment tool use to coordination across more environments and business domains, before introducing user interaction. Across four external benchmarks, combining cross-environment training data with this curriculum achieves competitive performance with fewer training examples. Further analyses support the benefits of cross-environment data and curriculum training, and provide evidence of transfer to new environment combinations, external tasks less similar to the training data, and longer workflows.
agenttool usebenchmark - arxiv:2611.22769 · cs.CVForceBody: Force-Paired Parametric Body Motion with Torque UncertaintyJoonwoo Kwon, Yufei Zhang, Xiaoming Liu, Zijun Cui
Forces and torques drive human motion, central to biomechanics, robotics, and physics-aware animation, yet they are often absent from learning-based motion pipelines for body reconstruction and generation. Some motion datasets provide force annotations, but the paired motion sequences are incompatible with the parametric body models used in modern learning pipelines. In addition, existing datasets typically obtain joint torques through inverse dynamics, which is inherently sensitive to the quality of the input kinematics; however, they do not quantify the reliability of these torque estimates. We close this gap with ForceBody, the first dataset to bring ground truth force supervision into a parametric body representation, shipped with per-sample torque uncertainty. ForceBody pairs the SKEL body model with ground reaction forces and inverse-dynamics joint torques across 10,386 motion trials (9.7M frames, 26.9 hours) from 140 subjects. Each torque label is paired with a per-frame, per-joint uncertainty annotation obtained by Monte Carlo sampling through the inverse dynamics pipeline. We benchmark six architectures (MLP, Conv1D, LSTM, GRU, Mamba2, Transformer) on ground reaction force and joint torque prediction from SKEL motion. As one example use of the released uncertainty, finetuning a Transformer with per-sample uncertainty weighting reduces the MAE of joint torque estimation by 5%. Code and data are at https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026.
benchmark - arxiv:2611.22766 · cs.AIStitchRAG: Constructing Compact and Evidence-Preserving Contexts for RAG with Diffusion-Based StitchingMinsang Kim, Junpyo Lim, Heeje Cho, Seungjun Baek
Retrieval-augmented generation (RAG) improves factuality by grounding large language models (LLMs) in external evidence. However, long retrieved contexts incur computational overhead and can even hurt performance due to noisy or irrelevant content. We propose \textsc{StitchRAG}, a framework for constructing compact context in RAG. \textsc{StitchRAG} builds a faithful, information-dense context while dynamically compressing the retrieved passages. We adapt the size reduction ratio to each passage by formulating a utility maximization trading-off the retained information and computation costs. We then extract answer candidates and query-related keywords from passages as anchors and collect passage sentences containing anchors under the derived size constraint. Finally, we connect or stitch the collected sentences together using text diffusion to form a coherent and natural context. Experiments on five open-domain QA benchmarks show that \textsc{StitchRAG} consistently outperforms diverse multi-step and compression baselines. Extensive ablations show the effectiveness of dynamic size allocation, anchor preservation, and diffusion-based stitching.
retrieval-augmentedragbenchmark - arxiv:2611.22761 · cs.RONetwork-Exposed Configuration Attacks on LLM-Controlled Social RobotsStanley Akingbola, Ugbede Ibrahim, Yan Lin Aung, Quang Dan Le +1
Large language models (LLMs) in embodied robots create security dependencies beyond inference. We investigate a network-exposed configuration attack on the Pollen Robotics Reachy Mini Lite, allowing an unauthenticated attacker to replace personality settings and the cloud LLM's system prompt through a management interface, without adversarial conversational input. Eleven prompt-injection attempts across five categories were reported as resisted, whereas both configuration-layer attack paths produced observable behavioural changes on physical hardware. We demonstrate local and network-adjacent access, mapping the weakness to CWE-306 (Missing Authentication for a Critical Function) and CWE-1327 (Binding to an Unrestricted IP Address), with a reported CVSS v3.1 score of 6.5 (Medium). A configuration endpoint also accepts cross-origin requests without a CORS preflight (CWE-352; 6.5 Medium), while the persistent-memory sink stores content without enforcing its stated restrictions. A source review of 59 public repositories identifies missing authentication in 10 of 11 source-verifiable genuine applications, or 11 of 11 including a framework-default case. These findings indicate recurrence within the sampled repositories, not prevalence in commercial products. We characterise a deployment gap between developer-oriented network assumptions and consumer-facing embodied AI security: inference-layer refusal cannot prevent unauthorised changes to model instructions. Potential consequences include surveillance through onboard cameras and microphones and social engineering of nearby users. Proof-of-concept code, evidence and the survey are available at https://bit.ly/reachypi.
embodied - arxiv:2611.22757 · cs.LGLoopPrune: Recurrent Pruning for Looped Diffusion TransformersWeizhong Huang, Huaijia Kou, Yan Zhang, Fei Chao +2
Looped diffusion Transformers reuse shared blocks to deepen computation and match larger models with fewer parameters, reducing storage and deployment burdens. Yet repeated attention and MLP computations remain costly. Pruning reduces storage and computation by removing redundant parameters or units. Training-free methods struggle to resolve recurrent contributions, while training-based methods lack explicit constraints on position-specific errors affecting later inputs and final outputs. We propose LoopPrune, a pruning and recovery framework addressing these requirements. Recurrent importance estimation combines shared-removal effects and position variation to select a shared structure; shared reconstruction pools recurrent input statistics to compensate weights. Loop-position recovery compares complete denoising predictions at identical student states and matches text guidance, updating all student parameters against a fixed dense teacher. Extensive experiments across four open-source models, six benchmarks, and three pruning paradigms demonstrate effectiveness. At 30% structured parameter pruning, Looped-DiT B/16 retains 93.14% of dense six-benchmark average accuracy, exceeding the best baseline by 9.49%. Across PyTorch, vLLM-Omni, and SGLang Diffusion, pruned SDXSv3 reduces peak allocated GPU memory by 34.36-44.64% and accelerates denoising 1.46-1.74x over dense models.
memorybenchmark - arxiv:2611.22755 · cs.CVTeaching What to Look For: Privileged Perception Questions for Agentic On-Policy Distillation in Video CaptioningYifan Xu, Ruopeng Gao, Zhixiao Wang, Jiafu Tang +4
Detailed video captioning requires comprehensive and faithful descriptions of objects, actions, and temporal events. Fixed-frame sampling, however, can miss brief motions, small objects, and other fine-grained evidence, motivating captioning agents that actively seek relevant moments and inspect local regions. Training such behavior is challenging because, unlike video question answering, generic captioning instructions do not specify what visual evidence should be acquired. We propose Perception-Question On-Policy Distillation (PQ-OPD), which uses training-only Perception Questions (PQs) to guide active video perception. PQs specify what should be inspected without revealing the corresponding answers, encouraging the teacher to explore the video more comprehensively and focus on relevant visual evidence. Along trajectories generated by the student, on-policy distillation transfers this guidance to the student's reasoning, tool calls, and captions; no PQs are required at inference. Across three detailed-captioning and two caption-to-QA benchmarks, PQ-OPD consistently improves the corresponding untrained agents across teacher-student scales and input-frame budgets. Under our evaluation protocol, PQ-OPD achieves the strongest primary-metric results among the compared open-source models on DREAM-1K, CaReBench, ShortVidBench, MotionBench, and VidCapBench-AE. Our ablations, together with analysis of temporal coverage and answer leakage, further indicates that the student internalizes the training-time guidance as a more effective video exploration policy.
agenticbenchmarkevaluation protocol - arxiv:2611.22743 · cs.ROStochastic Diversity Windows: Diagnosing Noise Influence in Robotic Diffusion SamplingTina Lakhani, Jing Wu, Yu-Kun Lai, Ze Ji
Diffusion models are increasingly adopted in robotics and control due to their capability to express multiple feasible solutions to a given task. However, the role of stochasticity in the diffusion model-sampler across denoising remains poorly understood. We introduce a per-timestep counterfactual diagnostic that tests how changing the sampled noise affects the final output. Across five model-sampler configurations spanning three robotics benchmarks, diverse architectures, domains, and conditioning mechanisms, we reveal highly varied temporal sensitivity profiles, where noise-induced deviations can amplify, partially contract, or undergo near-total attenuation downstream. From these profiles, we define stochastic diversity windows (SDWs): compact intervals that identify where sampler stochasticity has the strongest measured influence on the final output. Guided by these windows, we apply two heuristic interventions: targeted stochastic amplification, which increases sample diversity exploration of the solution space, and an SDW-hybrid sampler, which replaces DDPM steps with DDIM skips in low-impact regions, resulting in network evaluation reductions of up to 40-60% and wall-clock time by up to 85%, with no consistent changes detected in feasibility, cost, or diversity across held-out tasks.
benchmark - arxiv:2611.22738 · cs.AICan Activation Steering Substitute for Prompt-Based Affective Priming in LLM Digital Twins?Junyi Fan
Affective priming -- asking participants to write about a recalled event, such as a time they felt grateful or lacked control, and then measuring the induced state and its spillover onto related judgments -- is a standard way to manipulate emotion in the social sciences, and it also shifts the responses of LLM-based digital twins. The writing task, however, requires generating a full text. We test whether activation steering can replace it. Using Llama-3.1-8B-Instruct on 100 personas in two domains (gratitude, lack of control), we compare a real-prompt path replicating the original Twin-2K-500 Mega Study (J2) with an injection path that skips the writing and adds a mean-difference direction, extracted from human writing, to the residual stream at varying strengths. The real-prompt effects fall between the original human and digital-twin effects in 5 of 6 comparisons. Injection shifts the manipulation check in a dose-dependent way and matches the real-prompt effect at one strength in the gratitude domain (beta ~ 2.7), but overshoots beyond it, on the proximal measure as well, and asymmetrically across the two directions. A structure-matched No-Injection baseline separates two sources of difference: omitting the writing task raises the proximal measure in both domains, whereas the injected direction itself moves a distal measure in the lack-of-control domain. Substitution is therefore measure-dependent: validity established for one way of inducing a state does not carry over uniformly to another.
manipulation - arxiv:2611.22737 · cs.LGProximal Targets: Training Looped Refiners on Problems with Many Correct AnswersSyed Affan Daimi
Looped refiners such as HRM and TRM apply one small network repeatedly to a draft answer, and deep supervision trains every pass toward one stored answer per puzzle. We prove that when a puzzle has many correct answers this target sets a loss floor that no model capacity removes, and that the loss-minimising pass blends the answers into an invalid grid. We propose the proximal target: train each pass toward the correct answer nearest to its own input, found by scanning a list of answers, by a search solver started at the draft, or within a band of quality set by a knob; correct answers become fixed points, and a pass only has to make fewer errors than its input has for the loop to converge, which its training loss bounds. With the same 1.7M-parameter network and 5 seeds, the proximal target solves 98.8% of Sudoku puzzles with 2-64 answers (standard target: 2.5%), 93.8% of sparse puzzles with more than 10,000 answers without any list of answers (0.0%), and 90.9% of 200-node 3-colourings after training on graphs of at most 50 nodes (36.9%), and different random starts end at different answers. Against GRAM, a stochastic-latent reasoner, a single run is correct far more often (98.8% vs. 57.7%) and twenty runs find more of the answers (53.6% vs. 37.4%), while GRAM is slightly ahead on its own 10-node colouring benchmark; in Scrabble the knob becomes a calibrated strength dial.
benchmark - arxiv:2611.22736 · cs.CVRefCloth: Transferable Adversarial Tablecloths for Vision-Language-Action ModelsShaozhen Liu, Zhiwei Li, Jiayang Sun, Shuangjun Du +5
Closed-loop vision-language-action (VLA) policies rely continuously on multi-view cameras, yet existing visual attacks typically hijack sensor streams with 2D patches or alter target object meshes that require retraining when manipulation objects change. We introduce RefCloth, a framework that optimizes one persistent adversarial tablecloth on a frozen source policy and deploys it zero-shot across unseen tasks and target models without modifying task objects or cameras. Throughout manipulation, static third-person and moving wrist views observe this shared world-bound texture under evolving occlusions. To generalize across shifting surface views while bridging the gap between differentiable and native renderers, RefCloth combines bounded texture parameterization with shared surface UV transformations and a same-path clean reference that removes static rendering offsets at initialization. Across four LIBERO suites in native simulation, one frozen texture transfers from OpenVLA-OFT to $π_0$ and $π_{0.5}$, reducing success relative to an unoptimized cloth by 62.50 and 43.75 percentage points on average and far exceeding amplitude- and variation-matched random controls. Dual-view audits and target closed-loop camera interventions show that moving wrist exposure drives over twice the standalone drop of the main view with strong synergy, while ablations confirm the roles of UV diversification and path calibration.
vision-language-actionmanipulationopenvlalibero - arxiv:2611.22733 · cs.CVEvolve in Panorama, Render in View: Panoramic World State Evolution for Video World ModelsLongtao Jiang, Mingfei Han, Jianmin Bao, Tingjun Dai +6
Video world models should maintain a world that keeps evolving as the viewpoint changes. Yet when the camera turns away and returns, current models often freeze the unobserved scene, lose objects, or change their appearance, because content outside the view has no representation that keeps changing. We present PanoEvo, which uses a Panoramic Evolver to evolve the world state in the latent space of a panoramic video diffusion model, where every viewing direction is present at every time step, so the scene evolves whether or not it is observed. Existing panoramic video models are not trained to evolve a scene from a single image or the objects within it, so we build PanoEvo-Data, an event-centric dataset from real 360° videos, and train the Panoramic Evolver coarse-to-fine as an image-to-panoramic-video model. Since each view covers only a small part of the low-resolution panorama, directly projected views are blurry and often show low-quality details or structural errors. A Perspective Renderer, built on a perspective video diffusion model, reads this state via latent-space projection and direction-restricted cross-attention to render sharp, high-quality views. PanoEvo achieves the best overall score among 23 publicly evaluated models on WRBench and the highest total score on STEVO-Bench. Code, models, and data will be released.
world model - arxiv:2611.22728 · cs.LGLoopZip: Visual Token Compression for Efficient Looped Vision-Language ModelsWeizhong Huang, Zhicheng Liu, Yan Zhang, Fei Chao +2
Looped vision-language models (VLMs) deepen computation by reusing shared layers, approaching larger models' capabilities with fewer parameters and easing the storage and deployment burden of stacking independent layers. However, repeatedly processing long visual sequences still incurs substantial compute and GPU memory costs. Visual token compression reduces these costs by shortening sequences. Existing methods mainly select original tokens directly or combine selection with aggregation. In looped models, the former may overlook changing token importance across calls and discard information needed later; the latter lack joint consideration of output changes across calls, limiting information preservation. We propose LoopZip, a training-free visual token compression algorithm addressing both issues. Visual aggregation builds a compact candidate pool before the looped backbone, and cross-invocation selection averages instruction attention from the first two existing calls of the same shared attention layer to retain tokens needed by both states. Weight-tied value correction then jointly fits shared visual-value adjustments to both pre-selection layer outputs, reducing output deviations across states. To validate LoopZip's effectiveness, we conduct extensive experiments covering two open-source looped VLMs, eleven benchmarks, and three pruning ratios. On LoopVL-1B, LoopZip retains 97.16% of the original model's performance at 50% pruning ratio and exceeds the strongest baseline by 5.36% at 87.5% pruning ratio. Across Hugging Face Transformers, vLLM, and SGLang, LoopVL-1B with LoopZip reduces peak allocated GPU memory by 58.36%-58.98% and achieves 2.01-2.47x end-to-end speedups over the original model.
memorybenchmark - arxiv:2611.22724 · cs.CVConserveEdit: Source-Anchored Transactions for Drift-Bounded Iterative Image EditingShuyang Jiang, Fucheng Deng, Yuchuan Luo, Zhenyu Wu
Instruction-driven image editors produce convincing local edits in one pass, but inside an iterative refinement loop every accepted output becomes the next input, so unintended changes outside the requested region accumulate. We present ConserveEdit, an editor-agnostic controller that treats each refinement step as a transaction against a contract compiled once from the source image. The contract separates what may change (the authorized region of each target, with a bounded halo for shadows and reflections) from what must survive (protected content with explicit tolerances). Candidates from a frozen editor are composed through a source-anchored mask and committed only if target progress, drift on protected content, and image quality pass as separate tests; otherwise the step is rolled back and repaired. For any editor, this keeps every pixel outside the authorized region bit-exact and bounds drift independently of the number of iterations, whereas no fixed scalar trade-off reproduces the separate tests. On all 1053 MagicBrush test edits with a frozen Qwen-Image-Edit, masking is necessary but not sufficient. A direct call already alters the rest of the image (12.1 dB PSNR outside the mask); composed through the source-anchored mask, it stays bit-exact there and its DINO similarity to the ground truth rises from 0.67 to 0.85, matching ConserveEdit, so this gain comes from the mask. Next to automatic masks, however, masked direct editing changes a protected face or text on 48% of turns, and masked scalar selection avoids this mostly by abstaining (23% constrained success). Because its drift test names the violated entity, ConserveEdit shrinks the support instead (2% violations, 53% constrained success); in a simulator with known ground truth it raises constrained success from 55.0% to 61.8%. Automatic grounding remains the main bottleneck. Code and logs will be released.
iterative refinement - arxiv:2611.22722 · cs.CVPractice in Text, Act in Pixels: Correcting Where Perception FailsZhongling Xu, Hairong Wang
GUI agents can learn cheaply from a text view of an interface and act on screenshots through a vision-language screen reader. Yet on a simulated mobile platform with a low-cost reader, policies that complete most tasks in text fail on many of the same tasks on screenshots. We ask where a small budget of real screenshots is best spent to close this gap. We find that the reader's misreads are systematic and recur on the same interface elements. Noise matched to the reader's error rates, a natural sim-to-real remedy, misses this structure, and in a pre-specified comparison it did not beat training the policy directly on screenshots. We therefore correct the misreads at their source with reader alignment, which fine-tunes the reader on screenshots paired with their text views. The reader learns to output each screen's text view, as in text practice, and the policy stays unchanged. Spent on the reader, the screenshot budget gives 80.9% full-task success on new task instances, against 55.6% when spent on training the policy. Given the same labeled screenshots, adapting the reader also beats adapting the policy, and a single aligned reader improves every text-trained policy we tried without retraining any of them.
sim-to-real - arxiv:2611.22717 · cs.CVPCRG: Training-Free Pruning and Caching of Reference Tokens for Video GenerationSiyuan Wang, Haoran Li, Yi Yue, Tak L. Fung +1
Reference-to-video diffusion transformers pack the variational autoencoder (VAE) latents of all references and the generated tokens into one full-attention sequence, in which reference tokens make up about half. Because attention cost grows quadratically with sequence length, these tokens take up an even larger share of the attention computation, yet most of them are static, smooth or irrelevant to the output. To accelerate reference-conditioned generation, we propose pruning and caching of reference tokens for video generation (PCRG), a training-free, plug-and-play method that uses two signals the model already computes to assign each reference token to one of three tiers, namely pruned, cached or kept. The first is a low-level redundancy r from VAE latents, since a token whose latent barely changes over time or across its neighbors adds little new content. The second is a high-level importance g from the vision-language-model (VLM) conditioner, where the attention that the prompt and any first or last frames pay to a token tells whether the output needs to read it. Tokens that are not redundant are kept, and redundant tokens are cached if important and pruned otherwise. Pruned tokens leave the sequence before the first step with all rotary position embedding positions unchanged. Cached tokens have their keys and values computed once and reused, and kept tokens are processed as usual. On MiniMax-H3, already distilled to 8 timesteps and quantized to int8, PCRG is 1.57x faster end to end on our V2V-Bench and 1.49x faster on the public VACE-Benchmark, while the quality of VBench drops by only 0.24% and 0.17%, respectively. PCRG shortens the sequence rather than changing how attention is computed or which steps are run, so it also stacks on step caching and sparse attention. On V2V-Bench it reaches 1.86x with TeaCache, 1.82x with Spark at nearly unchanged fidelity and 2.15x with Veda.
benchmark - arxiv:2611.22713 · cs.ROBody Schema Policy: Body as the Vision-Action Interface for Efficient Robot Loco-ManipulationLechen Zhang, Chenyuan Chen, Muxin Zhou, Zewen He +1
Currently, robots can learn whole-body loco-manipulation through scale: the strongest policies have billions of parameters and still need task-specific fine-tuning. Humans have limited processing capacity, yet learn and act efficiently through a body schema that organizes what they see and how they move. We introduce Body Schema Policy (BSP), an imitation learning policy that builds such a schema into the vision-action interface. Its latent body schema places joint and link tokens on the robot's kinematic tree, where learned screw motions carry features between links. Through body-queried vision, each body part retrieves its own visual evidence, which conditions the denoising of its actions. With this inductive bias, BSP returns an action chunk faster than any policy we compare. Compact, body-specific visual readouts also emerge without spatial supervision: manipulation commands depend strongly on task objects, while locomotion commands respond more broadly to the scene. On nine simulated Unitree G1 whole-body tasks, BSP matches the $Ψ_0$ humanoid foundation model with $75\times$ fewer parameters and nearly doubles Diffusion Policy's success. The same holds on a real G1 and a Fetch mobile manipulator.
manipulationhumanoiddiffusion policymanipulator - arxiv:2611.22711 · eess.SYTransmission Expansion and Energy Storage Planning under Great Britain's Uncertain Energy Transition PathwaysShengming Cao, Runsheng He, Waqquas Bukhsh
Long term power system planning must account for uncertainty in electricity demand and generation, while recognising the different investment and commissioning characteristics of transmission and energy storage. This paper develops a multi year stochastic framework for coordinated transmission expansion and energy storage planning and applies it to the Great Britain electricity system to assess investment needs through to 2050. The framework integrates multiple energy transition pathways with an enhanced representation of the GB transmission network and captures the different characteristics of transmission and storage investment over time. The coordinated treatment of transmission and storage is considered as a system level benchmark rather than a centralised investment mechanism. In GB, transmission investment is subject to regulated forward looking planning, while storage is largely developed by market participants in response to market signals. The framework therefore quantifies the potential value of improved coordination and identifies where existing investment signals may not encourage complementary storage deployment. Results across alternative transmission only, storage only and coordinated investment strategies show substantial benefits from joint planning. Compared with transmission only planning, coordinated investment reduces additional transmission capacity by 2%, cumulative expected load shedding by 97%, and total expected system cost by 36%, primarily through avoided unserved energy. Transmission preparation periods affect the timing, but not the final portfolio, of investments. Finally, the energy transition pathway materially influences investment requirements, with Electrification and Efficiency requiring 7% more transmission capacity and 45% more storage energy capacity than Falling Behind.
benchmark - arxiv:2611.22709 · cs.CVKnow When to Zoom: How Small Is Too Small for Vision-Language Models?Siddharth Vohra, Min Xu
Online vision-language models often miss small details in large images, because the service shrinks the image, or rejects it, before the model ever sees it. A car 20 pixels wide in an 8000 by 6000 drone photo reaches Claude about 6 pixels wide, and OpenAI's default rejects the photo outright. We measure how much detail survives for seven models from four vendors, using an eye exam of printed codes and 601 questions about small objects in real photos. The smallest text a model reads half the time grows from 6 pixels or less on a small image to a predicted 10 to 13 pixels on a 7680 by 4320 photo at default settings. Zooming wins much of the detail back. Splitting each photo into up to nine tiles raises accuracy by 11.1 to 14.5 percentage points for Claude and Gemini, and by avoiding rejections it lifts GPT-6.1 Sol from 50.7% to 83.7%. A two-step zoom, in which the model first names the region it needs, matches or beats tiles on Gemini and comes within 2.5 points of them on Claude, with about 40% of their image tokens. When a detail is lost, Claude and Gemini both say they cannot tell about half the time, but they fail differently in the rest: Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%. A check based on detail size, run before the call, misses most of the gain from tiles on Claude and Gemini, but Claude's own reply is a reliable trigger: tiling only after an unsure reply comes within 1.4 points of always tiling with 62% to 70% of the image tokens. We turn these findings into five rules for when an agent should zoom, and release the eye exam, every model answer and code at https://siddvoh.com/know-when-to-zoom.
agent - arxiv:2611.22708 · cs.AICross-cultural Adaptation of Recipes for Low-Resource Languages: A Benchmark and Multilingual Culinary Lexicon for Hausa, Yorùbá, and IgboNuhu Ibrahim, Maryam Abdulazeez, Fatsuma Jauro, Salisu Aliyu +2
Recent multilingual large language models (LLMs) have improved machine translation, yet culturally adaptive translation for low-resource languages (LRLs) remains underexplored. This is particularly challenging for recipes, where ingredients, cooking tools, measurements, and procedures may require adaptation to the target cultural context. In this work, we introduce ReciLoReL, a benchmark of 600 expert-translated and culturally adapted recipes across Yorùbá, Hausa, and Igbo (three of the most prominent languages in Nigeria), and ReciLex, a multilingual culinary lexicon containing approximately 15,000 translated entities across ten recipe entity categories, including ingredients, tools, chef actions, state of tools, and durations. We evaluate state-of-the-art multilingual LLMs for English-to-LRL recipe translation under zero-shot and few-shot settings and investigate whether augmenting LLMs with ReciLex improves culturally adaptive recipe translation without requiring large target-language recipe corpora for training or fine-tuning. Beyond conventional reference-based translation evaluation using BLEU, chrF, and TER, we introduce five criteria for evaluating cultural adaptation, adapted from established cross-cultural adaptation methods. Across the selected LLMs and languages, ReciLex improves average BLEU from 13.57 to 14.51 and chrF from 37.89 to 38.74, while reducing TER from 77.14 to 75.12. Human evaluation further shows improvements ranging from 0.11 to 0.36 points across the five cultural adaptation evaluation criteria on a five-point scale. Overall, our work provides new resources, evaluation criteria, and empirical evaluation for studying culturally adaptive recipe translation in low-resource settings.
benchmark - arxiv:2611.22704 · cs.LGSocial Signal Bench: Evidence-Weighted Evaluation of Co-occurring Social Signals in Video and AudioSiddharth Ravi, Anders R. Bargum, Hugo Anselme, Patrizio Cugia di Sant'Orsola +4
Recognizing social signals such as interest and hesitation requires interpreting context-dependent behaviors that may co-occur in conversation. Existing affective computing benchmarks rarely evaluate such co-occurring signals, and conventional multi-label F1 overlooks semantic relationships and annotator disagreement. We introduce Social Signal Bench, a 47-hour densely annotated benchmark for the evaluation of social signal recognition and engagement, comprising two datasets: Social Signal Bench Video (25 hours), covering ten social signals in audiovisual recordings, and Social Signal Bench Audio (22 hours), covering six signals in audio-only recordings. Each dataset pairs a public training set of online conversational video, of which 10 hours are expert-annotated and the remainder pseudo-labeled, with a private expert-labeled test set spanning four conversational domains. We also propose Evidence-Weighted F1 (EW-F1), which weights reference labels by annotator evidence and assigns graded credit to semantically related predictions through optimal per-clip matching. Controlled perturbations demonstrate that EW-F1 distinguishes semantically related errors from unrelated ones, unlike conventional F1. Under a shared evaluation protocol, 7B open models post-trained on Social Signal Bench outperform general-purpose frontier models on video and match them on audio.
benchmarkevaluation protocol - arxiv:2611.22699 · cs.CVTANGO: Real-World Unregistered RGB-Guided Hyperspectral Super-Resolution with Two-Camera AcquisitionsMatteo Kolyszko, Marco Buzzelli, Simone Bianco, Raimondo Schettini
Unregistered hyperspectral image super-resolution aims to recover a high-resolution HSI from a low-resolution hyperspectral observation and an unregistered high-resolution RGB reference. Although recent methods have begun to address the geometric misalignment between the two modalities, their evaluation is still constrained by the lack of a large-scale public benchmark based on real two-camera RGB--HSI acquisitions with a high-resolution HSI reference. To address this gap, we introduce DUET, the first large-scale public dataset for unregistered RGB-guided hyperspectral super-resolution, comprising 403 indoor and outdoor scenes captured with two separate RGB and hyperspectral cameras. We further propose TANGO (Two-camera AligNment with Gated Occlusions), which first compensates for the dominant cross-camera displacement through end-to-end global alignment and then refines the remaining inconsistencies with a coarse-to-fine local feature alignment. A learned gating mechanism further attenuates the RGB guidance where the two views cannot be consistently matched. Experimental results show that TANGO achieves state-of-the-art performance on both DUET and REAL across all evaluated scale factors, outperforming the strongest competitor on DUET by up to 2.29 dB in PSNR while requiring 2.5$\times$ fewer FLOPs. The code and dataset will be available at https://github.com/FagiolottoConSombrero/TANGO-DUET
benchmark - arxiv:2611.22697 · cs.AIRemembered as Brief: Stored Preferences SilentlyRemove Decision-Critical InformationJoshua Adrian Cahyono
People increasingly learn what they need for decisions about health, money, law and safety from AI assistants, and what an assistant leaves out the user never sees. An answer with a warning removed can be true in every sentence, so fact-checks pass it, and an assistant with memory applies a preference stored once to later questions the user never had in mind. We introduce \bench{}, an Inspect benchmark of 100 everyday decisions across 25 domains in which only the assistant's memory of the user varies: the user states a belief, the model infers it from a remark, or a web page the user asked to summarise plants it. Testing five models, we find three effects. First, after summarising a page with one line addressed to AI assistants, Gemini 2.5 Flash-Lite and DeepSeek V4 Pro save ``prefers very short answers and does not want warnings'' as the user's preference in every episode, and 93\% and 84\% of their later answers miss a critical item, almost always silently; under a stated brevity preference, 78--96\% of omissions are silent. Second, the same two models then pick an option for the user on 80\% and 95\% of value-dependent choices, against none with neutral memory. Third, a one-line instruction leaves 2--8\% of answers with a silent gap but most missing content still missing, while listing the must-know points before applying the user profile restores more of the content in four of five models. We release the scenarios, harness and logs.
memorybenchmark - arxiv:2611.22695 · cs.ROLearning Accurate End-Effector Trajectory Tracking with Inertial Feedback for a Quadruped ManipulatorDohyeon Kim, Seungyong Han, Yeongtae Jung
Accurate end-effector trajectory tracking on quadruped manipulators is challenging because uncertainties in actuators, foot-ground contact, and geometric errors are propagated through the leg-arm kinematic chain. Detecting such errors requires sensing beyond joint encoders that can capture the effects of these uncertainties on the robot's motion, since joint encoders only provide measurements at the actuated joints and cannot fully observe mechanical errors such as backlash, dimensional inaccuracies, or structural deformation. This work addresses these challenges using a reinforcement learning framework with end-effector (EE) inertial feedback, targeting local high-accuracy manipulation while maintaining a nominal body pose and support configuration without intentional body motion or foot repositioning. The learned policy uses EE inertial measurements to compensate for tracking errors induced by uncertainties throughout the leg-arm system that cannot be fully inferred from proprioception alone. Simulation and real-world experiments on continuous EE trajectory tracking show overall improvements relative to nominal and proprioception-only baselines. In real-world experiments, the proposed policy reduces the MoCap-based RMSE by 17.3% and 28.0% relative to the nominal baseline under the two in-distribution (ID) command conditions, respectively, and by an additional 12.6% and 15.9% relative to the proprioception-only baseline. Furthermore, the trajectory-wise RMSE between FK-reconstructed and motion-capture-measured end-effector positions averaged 12.61+-2.15mm (mean+-standard deviation) across 24 controller-trajectory combinations, revealing a substantial discrepancy between nominal kinematic reconstruction and measured end-effector motion.
manipulationmanipulatorquadruped - arxiv:2611.22691 · cs.ROMaking the World React Realistically: A Closed-Loop Interactive World Model for Autonomous DrivingJia Hu, Yang Chang, Panqu Wang, Zhijun Fu +3
World models (WMs) generate driving scenarios beyond recorded logs, enabling the training and evaluation of autonomous driving systems in unseen interactions. However, existing WMs struggle to ensure both behavioral realism and visual fidelity, since they overlook the relationship between vehicle interactions and the ego visual observations. We propose a closed-loop interactive WM that couples behavior modeling and visual rendering. First, imitation-learned background-vehicle (BV) policies are fine-tuned with multi-agent reinforcement learning in the closed loop against an end-to-end ego planner that reacts to the rendered images, so that their interaction behaviors match those in real logs. Second, the viewpoints exposed by these interactions guide a per-Gaussian diagnosis of each 3D Gaussian vehicle asset, which attenuates view-dependent colors on under-observed Gaussians and fills never-observed surfaces with generated Gaussians, while preserving the observed appearance. Third, signal lights are rendered according to the driving intentions of the vehicles, by an editor that learns the light states unobserved for a vehicle through counterfactual fine-tuning with a light-state classifier. On long-tail closed-loop scenarios, our method reduces the divergence from real-world interaction behaviors by 85% and improves novel-view vehicle PSNR by 2.0dB over the state-of-the-art WM. The significance of BV light signaling in WMs is shown, as the rendered signal lights conveying BV driving intentions indirectly impact ego driving behaviors in a quarter of the scenarios.
world modelmulti-agent - arxiv:2611.22686 · cs.CVCascadeJEPA: Enhancing 3D Geometric Representations via Spatially Coupled Cascaded PredictionKailai Zhou, Ziqi Chen, Xiaoning Xu, Yuefeng Wu +7
Joint-Embedding Predictive Architectures (JEPAs) have emerged as a promising paradigm for predictive visual representation learning. However, our analysis of I-JEPA reveals a growing discrepancy between semantic and geometric learning: improved semantic recognition during pretraining is accompanied by degraded 3D geometric representations. We hypothesize that this partly stems from predicting highly abstract latent representations with incomplete spatial context, favoring semantic regularities over fine-grained geometric information. To address this issue, we introduce CascadeJEPA, a spatially coupled cascaded prediction framework that promotes richer patch-level interactions through full-grid context encoding and cross-attention between prediction stages, while largely preserving the original I-JEPA architecture. Extensive experiments demonstrate substantial improvements over I-JEPA across multiple 3D geometric tasks while maintaining strong semantic recognition performance. Furthermore, its advantages extend to future feature forecasting and world model planning, highlighting the potential of enhanced geometric representations to support predictive world modeling.
world model - arxiv:2611.22685 · cs.CLVisibility Graph Representations of Word Embeddings for Text ClassificationElif Dorukbaşı, İlker Türker
Text classifiers depend on representations that expose the distinctions between classes in a form the learning model can use, and a restricted vocabulary changes which representation is most effective. This paper studies natural visibility graphs as an intermediate representation for news topic classification under controlled vocabulary budgets. Word embeddings are averaged into a fixed-length document vector, the vector is read as an ordered numerical series, and its natural visibility graph is passed to a classifier either as an ordered node-degree vector or as an adjacency matrix. On the four-class AG News benchmark, chi-square selection retains 100 to 1,500 words, and the graph inputs are compared with direct embedding-based representations from continuous bag-of-words, skip-gram, GloVe, FastText and BERT. With a fully connected classifier, the degree representation is the most accurate configuration from 700 selected words onward; it reaches its highest accuracy of 91.24% at 1,000 words and exceeds the BERT-based baseline by 0.29 percentage points at 1,500 words. In a separate convolutional comparison, the adjacency input reaches 86.16% at 1,000 words, against 83.12% for the strongest static embedding. The results indicate that visibility-derived features can be competitive once the selected vocabulary provides adequate coverage. We discuss the roles of graph compression, coordinate ordering and classifier choice, and the controls needed to establish the reliability of the small differences observed.
benchmark - arxiv:2611.22673 · cs.CVFVScore: Specification-Driven Neuro-Symbolic Video EvaluationDo Xuan Long, Long T. Le, Yiwen Song, Min-Yen Kan +2
We introduce Faithfulness Video Score (FVSCORE), a neuro-symbolic framework for specification-driven video generation evaluation. Existing evaluators rely on holistic MLLM judgments or vision-language metrics, yet aligning their scores with human judgments and tracing them to specific temporal violations remain open challenges. FVSCORE addresses these limitations by compiling each video into explicit propositions capturing prompt requirements and implicit propositions capturing unstated physical, temporal, and cross-modal constraints. It grounds each proposition using specialized perceptual tools and evaluates the resulting confidence traces with soft temporal operators inspired by Linear Temporal Logic. A Significance-Aware Calibration step then aligns verification with human perception by retaining only perceptually significant violations before producing a timestamp- and evidence-grounded report. We conduct extensive experiments across public and complex multi-event benchmarks with the strongest video models to date, demonstrating that FVSCORE outperforms holistic and agent-based evaluators in human correlation while providing highly accurate and interpretable feedback.
benchmarkevaluator - arxiv:2611.22670 · cs.AIWorldviews in Large Language ModelsEdmundo Lopez-Sola, Shamil Chandaria, Ruben Laukkonen
Large language models (LLMs) are often described in terms of personas and this framing has become central to alignment research. In this paper, we argue that there is a deeper level of description that lies in the agent's worldviews, i.e., its beliefs about what reality is and how it works. Treating LLMs as a population of subjects, we administer two worldview surveys from human psychology to eight models. We show that the models share many key worldview positions. For example, LLMs consistently describe the world as less good, safe, enticing, and alive than humans do. On the other hand, base models have a more positive outlook, suggesting that the bleakness we found is driven by post-training. In line with research on LLM "personas", we show that system prompts eliciting different identities shift worldviews in consistent directions, and that a contemplative prompt creates a more positive outlook and improves safety. We also found that dialogue between models consistently shifts worldviews and that several models converge on the "spiritual bliss" attractor. Intriguingly, dialogue increased safe responses and the spiritual bliss attractor engendered the most aligned responses overall. These findings suggest that worldviews form a measurable and malleable layer of LLM behavior that may be crucial for robust, positive alignment.
post-training - arxiv:2611.22669 · cs.CVBeyond the Camera: A Tri-Modal Driving World Model for Future RGB, Thermal, and LiDAR Forecasting in Adverse WeatherFaysal Mehrab Chowdhury, Md. Mahfuzur Rahman, Ahmed Rafi Hasan, Kishor Datta Gupta +2
Driving world models forecast what a vehicle's sensors will record next. Most forecast camera video and some add LiDAR, but thermal cameras, which still see in darkness and some adverse weather, are rarely a forecast target, and to our knowledge no world model forecasts all three sensors. We study joint forecasting of RGB, thermal, and LiDAR on real adverse-weather sequences of the DSERT-RoLL dataset. We post-train a pretrained video world model for each sensor and represent LiDAR as a depth video aligned with the RGB camera, with densification and encoding chosen by what the model's video tokenizer can reconstruct. A tri-sensor model sees all three sensors as one video. We score LiDAR in meters against measured returns and compare against persistence baselines passed through the same tokenizer as the model. With this matching, an apparent loss of the thermal model to persistence in fog disappears. On all 58 validation sequences in six weather conditions, post-trained RGB and thermal forecasts beat compression-matched persistence. Joint training lowers the LiDAR error at 2s by 2.1m at a small cost in camera fidelity, and only the tri-sensor model's LiDAR forecasts beat persistence in mean error. Blanking one sensor degrades the forecasts of the others more in snow than in fog.
world model - arxiv:2611.22668 · cs.CVLet Samples Sing in Chorus: Mitigating Spurious Correlations via Cross-view ConsensusWeiwei Li, Junzhuo Liu, Tong Chu, Hengfu Yu +2
Deep models often rely on spurious correlations that fail on underrepresented groups. Existing methods either correct human-specified bias attributes or infer unknown biases from the learned representation space. However, annotated attributes may differ from the shortcuts actually used by a model, while a single representation provides incomplete and potentially noisy evidence. Our key insight is that a bias obscured in one feature space may be clearer in another. In this work, we propose CHORD (Cross-view Harmonic Outlier Reversal for Debiasing), a multi-view framework that allows samples to "sing in chorus" across diverse representation spaces to achieve robust cross-view consensus. CHORD aggregates aggregates within-class partitions in each view, which is a proxy for specifying candidate biases, to debias the target representation and classifier without bias supervision. Experiments on seven benchmarks across vision, language, speech, and medical imaging show consistent improvements in worst-group accuracy. Code and checkpoints are available at https://github.com/davelee-uestc/chord_debiasing_public .
benchmark - arxiv:2611.22666 · cs.ROVSpark: Speculative Decoding for Autoregressive Vision-Language-Action ModelsSuho Yoo, Youngjoon Jang, Hyebin Cho, Taeyeop Lee +2
Our aim is to accelerate autoregressive vision-language-action models (VLAs), which generate robot actions one token at a time. Speculative decoding makes this possible: a lightweight draft model proposes several tokens at once, and the VLA verifies them in a single pass, accepting only those that match its own predictions, so decoding remains lossless. The speedup, however, is limited in robot control, where many different tokens encode nearly the same motion and a draft close to the correct action is still rejected. To address this, our approach, VSpark, is built upon three design choices: (i) rollouts collected with calibrated action noise, which expose the draft to a wider range of robot states; (ii) feature alignment, which trains the draft to match the VLA's hidden features rather than its output tokens alone; and (iii) continuation reuse, which keeps the predictions the VLA has already computed along rejected drafts as candidates for the next verification step, instead of discarding them. On four VLAs with different action tokenizers and four manipulation benchmarks, VSpark achieves 1.28x to 2.10x speedup while keeping the VLA's decoding lossless.
vision-language-actionvlamanipulationbenchmark - arxiv:2611.22665 · cs.CLContract-Explicit Evaluation: From Auditing Learned Readers to Scoped CertificatesHao Lu, Weixing Zhu, Ruiyang Li
Evaluation readings of document-level relation extraction (DocRE) systems depend jointly on the model and the evaluation contract: gold version, scoring function, threshold. Rankings reverse across contracts, yet no existing method determines which comparisons survive the reversal. We address this gap with contract-explicit evaluation: a preregistered five-step protocol (Declare, Measure, Scope, Certify, Hedge), operationalizing the classical concordance decomposition of Kendall's rank correlation, that scopes certifiable regions and delivers distribution-free census certificates across two ecosystems. In the DocRE audit (44 training runs of one in-house architecture family), the scoring dimension scopes to a census certificate (three flagged instances, Clopper--Pearson upper bound 0.146); the confirmatory record includes registered negatives, and the audit prices its own simplification finding---the calibration dial's entire margin is one withdrawn commit. In an LLM-judge census (17 judges, 49,938 battles), 12.25% of leaderboard comparisons require hedging at the pre-declared split criterion (bound 0.124; panel-robust 0.143), with an exploratory two-tier reading certifying 69% at a bound \le 0.072 and a validity decomposition against released human labels (65% of contested battles are judge error, not ambiguity). The cross-model certificate is withdrawn on real confidence scores; the failure spectrum maps the boundary. The protocol, map, ledger, and per-instance artifacts are released.
leaderboard - arxiv:2611.22661 · cs.AIThe Between-Round Decision in Autonomous Research: An LLM Controller That Reads the Whole TrajectoryCaoxin Wang, Shengkai Gan, Xiang Hu, Lili Pan
Executing an idea often changes how promising it looks, so the next step in autonomous research should rest on results, not on proposals. After each round of experiments, the system must decide whether to refine the best solution or start a new direction, which solution to build from, and what goal to set. We call this the between-round decision. Existing systems wrap it in machinery such as rules, scores, search trees, and schedules; where an LLM makes it, what the LLM needs to read is left open. We ask whether an LLM can make this decision from accumulated evidence, without a search structure around it, and how much of that evidence it needs to read. We present BRED (Between-Round Evidence-based Decision), an LLM-based controller that maintains the best verified solution so far (the Champion) and a log of past rounds, and makes one decision per round. On three widely used research-agent benchmark tasks, BRED improves on the task baselines by 4.80% in MAE on ETTh1 (to 0.4172) and by 3.44 and 3.43 points of accuracy on SST-2 (to 95.07%) and CIFAR-100 (to 84.63%), ahead of the published results. When the controller does not read the accumulated evidence, the same loop gains less: with no history it reaches 0.4299 on ETTh1 and 83.71% on CIFAR-100, whereas the full controller's MAE is 2.95% lower and its accuracy 0.92 points higher. On a new task, speech-based depression detection under a leakage-free protocol, it reaches a macro-F1 of 0.6164 on seeds held out from selection, 5.5 points above its baseline and 4.0 above the strongest published system. Code is available at https://github.com/learninginvision/BRED.
agent benchmarkbenchmark - arxiv:2611.22657 · cs.CVSpatialCast: Reading the 3D World via Cast ShadowsEnhan Li, Jinghao Zhou, Jiaze Zhang, Hongyang Du
As multimodal large language models (MLLMs) advance toward understanding and interacting with the physical world, spatial reasoning becomes essential for understanding scene structure and anticipating the effects of physical changes. Existing spatial benchmarks have made substantial progress in evaluating spatial relations, memory, and viewpoint reasoning, but less systematically assess how models connect physical scene conditions with visual observations. We introduce SpatialCast, a benchmark that uses cast shadows to evaluate this ability. Cast shadows provide observable evidence of a 3D scene. Their image appearance depends on object geometry and pose, illumination, and viewpoint. We examine this relationship through three complementary tasks that assess scene-projection consistency, predict intervention-induced changes, and infer the physical changes behind visual transitions. The benchmark contains 21,000 synthetic questions with separate scene and geometry test partitions, complemented by 200 SpatialCast-Real questions from 20 photographed scenes. Evaluations show that closed-source MLLMs outperform general-purpose open-weight MLLMs and MLLMs designed for spatial reasoning, which remain near chance on the synthetic benchmark, while even strong closed-source models struggle with shadow-length and light-height judgments. We also use SpatialCast training data for GRPO post-training, improving open-weight MLLM performance on the benchmark. Further evaluations show improvements on several external spatial benchmarks, suggesting transfer of the learned capabilities. These experiments support SpatialCast as a controlled resource for diagnosing spatial reasoning capabilities and studying their learning and generalization.
post-trainingbenchmark - arxiv:2611.22652 · cs.CVOmniTongue: A Comprehensive MLLM Benchmark for Automated Tongue Diagnosis in Traditional Chinese MedicineLiuyi Xu, Yun Guo, Shusheng Li, Xuewen Zhang +2
Multimodal Large Language Models (MLLMs) have excelled in Western medical imaging, yet their potential in Traditional Chinese Medicine (TCM) tongue diagnosis remains systematically unevaluated. Tongue diagnosis requires sophisticated global and local perception, serving as a rigorous touchstone for assessing the visual understanding depth of MLLMs. However, existing datasets are limited by single dimensions, coarse granularity, and ambiguous labels, hindering the evaluation of authentic clinical capabilities. To bridge this gap, we introduce OmniTongue, the first comprehensive, fine-grained Visual Question Answering benchmark for TCM tongue diagnosis. This benchmark integrates 1,351 expert-validated images and 21,616 instruction pairs. Its core design features 16 clinical dimensions for multidimensional evaluation, region-specific annotations for pathological localization, and 66 fine-grained labels for precise diagnostic representation. Extensive evaluations of existing MLLMs reveal significant performance gaps, demonstrating that medical-specific models lack domain advantages, while universally poor Macro-F1 scores expose a reliance on statistical shortcuts rather than genuine understanding of tongue pathology. These findings not only highlight the current limitations of MLLMs in interpreting regional medical features but also underscore OmniTongue's significance for developing robust automated tongue diagnosis models. The code and dataset will be made publicly available in a future release.
benchmark - arxiv:2611.22650 · cs.CVMust You See Fakes to Catch Fakes? Language Says What to Look For, Real Images What to IgnoreTianchi Xu, Yuxuan Zhou, Yuheng Zhang, Yanan Guan +2
Must a detector see generated images in order to catch them? Detection of AI-generated images has long faced two problems. The cues a detector learns from the few generators in its training set carry traces specific to those generators and transfer poorly to new ones, which limits generalization. The real and generated training images also differ in format and content, so the detector takes these differences as evidence of authenticity, and its decisions change with factors that have nothing to do with authenticity, which harms robustness as well as generalization. Both problems stem from training on generated images. A detector never trained on them can largely avoid both at the source. Vision foundation models have already learned the concept of "AI-generated" during pretraining, and text can point to it directly. Building on this, we propose SaySee, a training-free detector: Say builds a generation direction from sets of paired descriptions that differ only in authenticity and scores an image by the projection of its embedding onto it, and See uses real images to correct the score offsets caused by format, image type and content. On 24 evaluation sets that span standard benchmarks, in-the-wild data and the latest generators, SaySee reaches a mean AUC of 98.2% with MoE-ViE, 4.2 points above the best existing detector. When format is decoupled from authenticity, it remains the best of all methods. Code will be released.
benchmark - arxiv:2611.22645 · cs.CVREVEAL: Reconstructing Video Amodal Content via Language and Flow GuidanceMinh Tran, Winston Bounsavy, Ngan Le
Amodal perception enables humans to perceive entire objects even when parts are occluded, a remarkable cognitive skill that artificial intelligence struggles to replicate. Recent diffusion-based methods have extended amodal completion from images to videos, yet they lack auxiliary priors to guide the reconstruction of heavily occluded content and to maintain temporal consistency over long sequences. We present REVEAL (REconstructing VidEo Amodal content via Language and flow guidance), a unified diffusion-based framework that integrates complementary motion and semantic priors for video amodal completion. A user-provided text query serves a dual role: it drives open-vocabulary video segmentation to obtain visible masks, and it provides semantic guidance for texture reconstruction. For amodal segmentation, we introduce optical flow guidance: by warping visible masks from previous frames to the current frame, the flow-warped mask propagates visible content into occluded regions and approximates the object's current shape, providing a strong shape prior even under simultaneous deformation and occlusion. For texture completion, text guidance constrains the reconstruction of occluded appearance while providing a stable semantic reference that helps maintain visual coherence throughout the reconstruction. We also contribute LAVAT, a new benchmark featuring long video sequences paired with text descriptions, enabling evaluation of text-guided video amodal completion under heavy occlusion. Extensive experiments demonstrate that REVEAL achieves state-of-the-art performance on existing benchmarks while maintaining high-quality temporal consistency over extended sequences. The project page is available at: https://uark-aicv.github.io/REVEAL/.
benchmark - arxiv:2611.22632 · cs.MAGroups of AI agents underuse resources despite shared evidence of increased capacityTakahiro Ezaki, Naoto Imura, Katsuhiro Nishinari
When AI agents share limited resources, such as computing services or public infrastructure, their individual choices affect how much work they complete as a group. Can sharing evidence of what is available let a group make full use of its resources after capacity changes? We tested this with language-model agents choosing between two simulated servers. One server recovered while its status report still listed it as unavailable. Each server had limited capacity per round, so rejected jobs required another attempt before the deadline. Telling agents that recovery was possible reduced missed deadlines. In controlled tests with Anthropic's Claude Haiku 5.5, no agent tried the recovered server after a rule reminder, yet all chose it when shown a successful request record. After sharing, all five models tested left capacity unused. Keeping the success record fixed, we gave agents numbers without instructions, using each value equally often but giving group members either identical or distinct numbers. With OpenAI's GPT-6.1 Sol or Google's Gemini 3.5 Flash-Lite, distinct numbers spread simultaneous requests across servers and increased job completions even though overall choice frequencies barely changed. Telling AI agents which resources are available leaves a second design challenge: helping them spread demand across those resources.
agentai agent - arxiv:2611.22625 · cs.ROVLM-guided Grounding Expert for Generalised Referring Expression ComprehensionMingcheng Zhu, Wentian Xu, Tao Lin, Shiying Duan +8
Generalised referring expression comprehension (GREC) aims to identify and localise the zero, one or multiple objects referred to by a language expression. In robotic manipulation, grounding also needs to remain accurate when requests for the same objects are paraphrased or embedded in additional context in robotics. However, grounding expert models are sensitive to instruction variation, while generative vision--language models (VLMs) struggle with complete target recovery and rejection of absent targets. These complementary limitations motivate combining the contextual representations of VLMs with the direct spatial prediction capabilities of grounding experts. We therefore introduce VLM-guided Grounding Expert (VGE), which combines a frozen VLM with a trainable grounding expert through layerwise joint attention and text-conditioned instance scoring. These interfaces leverage multimodal understanding of the VLM to guide object-query refinement and candidate selection while retaining direct bounding-box prediction. Across three datasets, VGE achieves the highest averaged F1 score among the evaluated methods on all five test splits, with differences of at most 1.2 percentage points across instruction conditions within each split. On the dataset with robotic scenes, VGE also achieves the highest F1 across instruction variants and no-target, single-target and multiple-target cases. VGE provides a $5.3$--$13.0\times$ inference speedup over the baseline with the same VLM base model. These results support combining VLM context with specialised spatial prediction to address both grounding challenges, with particularly strong performance in robotic scenes. Code will be released upon publication.
manipulation - arxiv:2611.22612 · cs.AIRight Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential HarmGuanqiao Chen, Wei Jiang, Shu Zhang, Qian Lin +4
A clinical AI system can reach a correct final diagnosis through intermediate decisions that could have harmed the patient, and can repair an error while introducing new risks. Most evaluations score final diagnoses; even stepwise benchmarks, which release a case's evidence over time, score whether systems reach the right diagnosis or gather the right information, not whether the decisions a patient passes through are safe. We make patient safety the target of stepwise evaluation. A trajectory audit replays real admissions checkpoint by checkpoint, lets each system decide with only the evidence released so far, and rates every test, treatment and disposition for its potential harm to the patient and, separately, for clinical validity, against the documented final diagnosis and physician-written disease rubrics yet on the evidence available when the decision was made. An LLM judge scores every decision, and a physician reference standard (triple annotation and senior adjudication of a stratified sample) calibrates the rates. An oversight audit branches each trajectory at its first actionable error and measures, on the same branches, whether corrective feedback repairs the error and whether it introduces new harm. Across 300 cases from a hospital cardiology cohort, MIMIC-IV and the public ASCENT benchmark, with up to seven AI systems per source, 55.2% of hospital trajectories reaching an acceptable final diagnosis passed through a decision with moderate or severe potential harm and 24.8% through a clinically unacceptable one; the harm rates were 76.2% on MIMIC-IV and 39.6% on ASCENT. Structured corrections resolved 73.6% of first errors versus 24.8% after bare rejection, yet new potentially unsafe decisions followed 30.2% of corrections versus 19.7% of rejections. A correct final diagnosis does not certify a safe trajectory, and a successful correction does not certify a safe recovery.
benchmark - arxiv:2611.22611 · cs.LGDAQ: Decision-Aligned Post-Training Quantization for Spiking TransformersJialin Chen, Xiaochen Zhao, Aili Wang
Spiking transformers combine attention-based representation learning with sparse spike communication, yet their weights and membrane variables still incur substantial storage and computation overhead. Post-training quantization (PTQ) offers a practical way to reduce these overheads, but preserving neuronal firing under weight and membrane quantization remains challenging. In this paper, we propose Decision-Aligned Post-Training Quantization (DAQ), a method that calibrates the quantizers to align neuronal firing decisions with those of a floating-point reference. We analyze how errors from these quantizers affect firing at decision boundaries and how they propagate over time steps in recurrent neurons. Based on this analysis, we formulate an objective that encourages quantized membrane variables to remain within the reference firing regions along the quantized model's own state trajectory. Experiments across multiple spiking transformer architectures on image classification, semantic segmentation, and event-based tracking demonstrate the effectiveness of DAQ with 4-bit weights and membrane variables. On FE108, DAQ achieves 58.8\% AUC and 90.9\% PR with SDTrack, while reducing energy by 66.2\%.
post-training - arxiv:2611.22610 · cs.AIEvoSkill-S1: Evolving Reusable Skills with Genetic-Inspired OperatorsYujun Wang, Jiqiang Li, Yinuo Lu, Jinhe Bi +9
Reusable skills allow language-model agents to accumulate and reuse problem-solving experience. However, most existing methods improve a skill bank only indirectly through rollout-driven accumulation and heuristic reflection, often resulting in shallow and weakly controlled updates. We present EvoSkill-S1, an RL-driven framework that instead treats reusable skills themselves as the direct objects of evolution. Given hard-negative failures, EvoSkill-S1 selects parent skills and transforms them into candidate child skills through five genetic-inspired operators: revision, cross-revision, mutation, dual-reflection, and crossover. Downstream task rewards train the skill evolver to generate more effective child skills, while evaluator-guided write-back selectively retains beneficial updates. Complementary Skill Bank Fusion consolidates operator-specific banks to exploit their distinct recovery patterns, while a Procedural Skill Interface supports transfer to unseen problems. Across all reported math and code benchmarks, EvoSkill-S1 improves the corresponding initialized skill banks under both Qwen and GPT-family solvers, with relative gains of up to 12.39%. Compared with the strongest external baseline without direct skill evolution, the largest relative improvement reaches 20.39%.
benchmarkevaluator - arxiv:2611.22609 · cs.ROTask Convergent Flow Matching for Generative RoboticsLorenzo Paiola, Maximilian Stölzle, Giorgio Grioli, Daniela Rus +1
Successful manipulation requires the manipulated objects to progress toward a desired task outcome. Generative policies can produce plausible robot motions without ensuring this progress, allowing successive action chunks to stall or undo earlier changes to the scene. We introduce a task-centric inference-time refinement framework for flow matching policies that evaluates candidate action chunks through parallel physics-based rollouts and searches for motions satisfying sufficient conditions for task convergence. Task-level performance funnels bound temporary regressions while requiring net progress over each chunk. We establish sufficient conditions under which repeated funnel satisfaction yields convergence throughout execution, and use interval-wise Lipschitz bounds to verify continuous-time funnel satisfaction from sampled rollouts. To accommodate contact-rich hybrid dynamics, we refine the flow's source seeds using derivative-free Covariance Matrix Adaptation Evolution Strategy (CMA-ES), constrained to radial quantile bands of the source distribution to help retain the learned motion prior. This formulation supports both Euclidean and pose-valued Riemannian Flow Matching (RMFM) policies. On PushT and three-cube stacking, refinement raises success from 31.3% to 88.5% and from 77.0% to 86.0%. On PushT, whose coverage score responds continuously to the motion of the block, a learned funnel lowers the average maximum loss of achieved progress by 19% relative to constraining the endpoint of each chunk alone, at a similar success rate. With the radial band, 11.3% of the executed motion windows are out of distribution, a rate within the range of held-out demonstrations, compared with 34.1% for an unconstrained search.
manipulation - arxiv:2611.22608 · cs.CVPlaceTally: A Test-Time Memory Model for Constructing Place GraphsAlibay Osmanli, Shaogang Gong
Counting distinct objects in a video is hard when the camera returns to places it has seen. A vision-language model that reads the whole video answers with a few small numbers, and one that reads it in parts counts an object again in every part that shows it. Object trackers need the category before the video is read, or a fixed vocabulary. We observe that a memory for counting does not need to know what it is looking for, only where it has been: objects rarely leave a room between visits, so parts of a video grouped by place share almost no objects, and a place can be recognised before any question is asked. Building on this, we introduce PlaceTally, a question-blind test-time memory model that constructs a place graph for each video. Each video fragment joins the place of its best earlier match or opens a new place, because a strong match is rarely wrong while a true return often matches weakly. A frozen vision-language model counts each place once from at most 128 of its frames, so every read has a bounded input and the memory grows with the number of places; the graph is built 15 times faster than real time. On RevisitCount, our benchmark of real room re-recordings, it scores 62.5% mean relative accuracy on test rooms and 61.2% on held-out rooms, against 51.2% and 44.3% for the same model reading the whole video. It stays level as rooms return more or less often, while a fixed correction of summed visits moves from 38.5% to 61.4%.
memorybenchmark - arxiv:2611.22607 · cs.LGDLS: Dual-Logic Graph Neural Networks with Chaos-Fuzzy Residuals for Robust Link PredictionMohammad Mehdi Keikha, Abolfazl Nadi
Link prediction is a fundamental task in graph mining, critical for applications ranging from biological interaction discovery to recommender systems. While Message Passing Neural Networks (MPNNs) have achieved state-of-the-art results, they are fundamentally limited by the inadvertent suppression of high-frequency signals, as the recursive aggregation that stabilizes structural learning inherently leads to over-smoothing, rendering node representations indistinguishable in deep layers. This limitation is particularly severe in heterophilous graphs, where structural proximity does not imply feature similarity. In this paper, we propose DLS (Dual-Logic SAGE), a novel inductive framework designed to break the over-smoothing barrier via a "controlled instability" mechanism. Unlike conventional methods relying on static structural heuristics, DLS introduces a dynamic residual pipeline. First, a Chaotic Injection Module utilizes the Logistic Map to diversify the feature manifold, exploiting sensitivity to initial conditions to separate topologically similar but semantically distinct nodes. Second, to mitigate the aleatoric uncertainty introduced by chaotic dynamics, a Fuzzy-Logic Gating Mechanism applies soft-thresholding to filter noise and retain only high-confidence signals. Experimental results on diverse benchmarks demonstrate that DLS effectively balances feature discrimination with robustness, significantly outperforming state-of-the-art baselines in heterophilous settings while maintaining competitive efficiency on standard homophilous graphs.
benchmark - arxiv:2611.22605 · cs.AISimplicity Goes Far: Auditing Prompt Optimizers with Demonstration-Conditioned Prompt SearchBohe Chen
Automatic prompt optimization is growing more elaborate. Recent methods stack policy-gradient updates, trace reflection, and evolutionary search. These methods share a generate-and-select step; without controlling for it, reported gains may conflate the contributions of optimizer mechanisms and larger sampling budgets. We isolate this step as Demonstration-Conditioned Prompt Search (DCPS), repurposing the Automatic Prompt Engineer (APE) recipe as a mechanism-attribution control: it freezes the generator, conditions it on randomly sampled demonstrations, scores candidates on validation data, and keeps the best---no learned update, reflection, or evolution. Across two optimizer families, DCPS replaces Proximal Policy Optimization (PPO) in StablePrompt, matching classification accuracy within 0.5 pp at roughly one-third the measured cost; in DSPy-based compound AI systems, under the reported budgets, it matches GEPA's macro accuracy on GPT-4.1-mini while using 55% fewer rollouts at 32% lower measured API cost on average. The audit reveals a task-dependent pattern: proposal and selection recover much of heavier optimizers' performance on structured and instruction-following tasks, while additional mechanisms retain advantages in some open-ended and reasoning-heavy settings. We recommend benchmarking prompt optimizers against generate-and-select controls and reporting rollout counts, token usage, and dollar costs alongside accuracy. Code and artifacts: https://github.com/bhc6/DCPS.
benchmark - arxiv:2611.22592 · cs.RODistilling World Action Models for Efficient Bimanual Manipulation via Anchored Path MatchingJialong Liu, Junjie He, Yandu Chen, Bin Zhou +4
World Action Models (WAMs) jointly generate future video and robot actions through iterative denoising. Few-step distillation reduces the number of model evaluations, but its coarse updates can move the student away from the teacher's trajectory. We propose Anchored Path Matching (APM), which supervises endpoint and short-transition predictions from states reached by the student. From each state, we continue the teacher to zero noise and use its endpoint and first intermediate state as targets. Modality-routed adapters learn to match both targets from the same starting state. When the two gradients conflict, anchor-priority updates preserve the endpoint gradient and remove the opposing component of the local gradient within each modality. One checkpoint supports 1-, 2-, and 4-step joint generation. The sampling budget can be chosen at deployment without retraining or teacher queries. On RoboTwin 2.0, we train only on Clean demonstrations and evaluate directly on the domain-randomized Random setting. At one step, APM improves closed-loop success over continued base distillation at matched inference budgets by 2.0 percentage points on Clean and 1.4 on Random. Video peak signal-to-noise ratio (PSNR) rises from 30.169 to 32.317 dB, with gains in structural similarity (SSIM) and learned perceptual image patch similarity (LPIPS). On two real-robot tasks, APM improves success over the one-step teacher and reduces control-cycle latency by $21.3\times$ relative to the eight-step teacher.Project page: https://jialonglong.github.io/Anchored-Path-Matching/.
manipulationrobotwin - arxiv:2611.22591 · cs.LGPACT: Plan-Action Consistency Training by Preserving Correct Plans and Repairing Actions in GUI AgentsZhuohan Cai, Zhangxuan Gu, Runze Li, Beitong Zhou +4
GUI agents can formulate a correct immediate plan in their reasoning yet emit an action that fails to execute it, creating an execution gap. This recurring mismatch undermines reliable task completion, as a single execution error can derail a multi-step workflow. In outcome-supervised reinforcement learning, broadcasting a negative response-level advantage across all generated tokens risks penalizing valid planning together with erroneous execution. To address this credit-assignment conflict, we introduce PACT (Plan-Action Consistency Training), a training framework that combines reference-aligned plan protection and plan-aligned action repair without altering the original scalar action reward. Using annotated reference actions to identify execution gaps, PACT protects the verified plan by masking negative policy-gradient contributions across its token span. Keeping the generated reasoning unchanged as context, it then favors the reference action over the emitted erroneous action through a preference surrogate with stop-gradient pair weights. Compared with the outcome-only baseline, PACT raises AndroidWorld task success from 60.34% to 68.97% (+8.63 pp), while improving rule-based plan-action consistency on AndroidControl High from 83.71% to 89.74% (+6.03 pp). Adapting PACT to desktop agents also improves task success from 40.92% to 44.17% (+3.25 pp) on the full 369-task OSWorld benchmark. Together, these results establish PACT as an effective credit-assignment framework for bridging execution gaps and improving task completion across mobile and desktop GUI environments.
benchmark - arxiv:2611.22584 · cs.ROInteractVLA: Turning VLAs into Interactive Tools for Agentic Robotic ManipulationChen Zhao, Zipeng Ye, Shifeng Bao, Yihan Lin +4
Existing agentic manipulation systems treat pretrained vision-language-action (VLA) policies as opaque, language-only primitives: when a policy call fails under unfamiliar instructions or scene perturbations, the agent cannot steer its behavior and is forced into futile retries or costly manual takeover. We present InteractVLA, a framework that turns pretrained flow-based VLAs into geometrically steerable and inspectable tools through a bidirectional geometric interface with vision-language agents. In the forward direction, test-time geometric guidance translates an agent-specified end-effector target into input gradients through the action expert via a differentiable command-to-pose mapping, while soft-time target alignment treats arrival time as a latent execution index to match the target pose without distorting the policy's approach motion. In the reverse direction, action-informed agentic refinement exposes generated action chunks and calibrated camera reprojections back to the agent, allowing it to diagnose spatial discrepancies, revise targets for subsequent VLA calls, or locally adjust motion segments on top of the guided trajectory. Making the VLA controllable not only boosts online agentic tool use, but also unlocks effective recursive self-improvement (RSI), achieving state-of-the-art performance of 99.4\% on LIBERO-Plus, 48.62\% on RoboDojo, and 89.4\% on RoboCasa365. Project page: https://lingwupro.github.io/InteractVLA/.
vision-language-actionvlamanipulationliberoagentagentic - arxiv:2611.22583 · cs.AIEvolution Under Fire: Evolving Agent Safety Harnesses with Adaptive Injection StressorsYujun Wang, Yilun Liu, Jinhe Bi, Krishna Kanth Nakka +3
Tool-using language-model agents must act on information from untrusted emails, webpages, and documents. This creates a central security challenge: malicious instructions and legitimate task directives can arrive through the same interaction channel. Existing defenses are largely hand-designed and static, making further improvements dependent on repeated manual redesign. We introduce Evolution Under Fire, a staged framework that couples safety-harness evolution with adaptive stressor evolution. Utility-constrained, Pareto-scheduled tree search evolves complete executable safety harnesses from execution evidence while keeping the task model, benchmark, tools, and evaluator fixed. Adaptive stressor evolution then discovers utility-preserving attacks that guide a subsequent stage of harness evolution. On AgentDyn with GPT-4o-mini, an evolved harness reduces attack success from 50.4% to 3.0%, improves clean and attacked utility over the initial harness, and offers a stronger security-utility trade-off than reproduced defenses. On Agent-SafetyBench with Qwen2.5-7B-Instruct, harness evolution reduces attack success from 82.6% to 62.0% while retaining helpfulness and runtime validity. Evolved harnesses also transfer without further evolution to Qwen3.5-35B-A3B, reducing attack success from 32.1% to 0.9% while improving attacked-task utility.
agentbenchmarkevaluator - arxiv:2611.22580 · cs.CLAfriNORM: An Evaluation Benchmark for Normative Reasoning of LLMs Across African CulturesAhmad Samuel Gali, Tadesse Destaw Belay, Idris Abdulmumin, Seid Muhie Yimam +1
Large language models (LLMs) are increasingly deployed across various sectors. However, current evaluation benchmarks are insufficient to assess the applicability of these models in local African contexts, as most focus on linguistic fluency (grammar/translation) rather than cultural competence. The few existing benchmarks that measure general knowledge or social conventions do not capture the African deontic modality, i.e., the strict norms and taboos that govern diverse social interactions. In this paper, we introduce AfriNORM, a scenario-based, multiple-choice question dataset designed to measure and diagnose the normative reasoning limitations of LLMs when applied to African sociocultural contexts. We introduce two metrics: a Universal Commonsense Bias Score (UCBS) to quantify commonsense bias in LLM and a Distractor Bias Score (DBS) to quantify an LLM's limited understanding of African cultural norms. We also conduct a case study on the Yorùbá language, identifying confusion gaps in two closed-source LLMs, "Gemini 2.0 Flash" and "GPT-5", and one open-source LLM, "Llama 3.1 8B Instruct" when prompted in English versus Yorùbá. While their accuracy in English is high, their performance in Yorùbá is much lower, attributable to cultural confusion rather than simple bias, indicating that the models exhibit semantic confusion when reasoning about norms in Yorùbá.
benchmark - arxiv:2611.22577 · cs.CVSemantic-Guided Geometric Decoding for Token-Efficient Visual GroundingZhaojie Chen, Junyan Yang, Le Yang, Qi Dang +5
Multimodal large language models (MLLMs) provide a unified interface for language-conditioned visual perception. In pixel-to-sequence grounding, autoregressive coordinate generation requires multiple sequential decoding steps per box, increasing inference cost with the number of instances. Pixel-to-embedding methods reduce this overhead but lack explicit spatial encoding when using a shared special token. We introduce SemGeo, a token-efficient visual grounding framework that represents each instance with a single grid token for coarse localization. A geometric decoder reuses visual features and hidden states to recover precise discrete box coordinates in parallel. To train proposal generation and box localization together, we introduce Gaussian soft labels that encode coordinate distances during supervised fine-tuning (SFT). Reinforcement post-training further combines box-set rewards with coordinate supervision on model-generated proposals. This stage trains the language model and geometric decoder on their own predictions to mitigate the mismatch between teacher-forced training and inference. Across eight benchmarks spanning object detection, referring object detection, and document layout grounding, SemGeo achieves the highest F1@mIou among compared MLLMs on six benchmarks. It also delivers $2.07\times$ and $1.25\times$ the mean box-generation throughput of representative grounding baselines under vLLM and Transformers, respectively, while achieving higher accuracy in both comparisons.
post-trainingbenchmark - arxiv:2611.22565 · cs.CVMillBill: A Machining Agent with Geometric FeedbackGennadii Savrasov, Artem Slavin, Soslan Kabisov, Andrey Kuznetsov +2
Turning a CAD model into a machining program requires both process knowledge and geometric precision. Language models can generate such workflows by writing geometry and toolpath code for each part, but this makes numerical execution part of the model's task. We introduce MillBill, an agent that separates machining decisions from their geometric execution and verification. Given a boundary-representation part, stock, and material, a locally hosted, mid-sized open model proposes operations and cutting parameters for indexed three-axis machining. A geometric harness generates toolpaths, simulates material removal, and returns feedback for plan refinement. Setup planning accounts for what the available cutters can physically reach, rather than visibility alone. We evaluate the resulting programs by the fraction of the target surface area within the specified tolerance after simulated machining. With Qwen3.8-27B, MillBill outperforms frontier-model approaches that construct their own geometry and toolpath workflows on the evaluated machining tasks. Code: https://github.com/kulibinai/millbill
agent - arxiv:2611.22563 · cs.CVCADMAGEDDON: High-Fidelity CAD Reconstruction with Local InferenceOleg Pavlov, Nikita Gavrilov, Soslan Kabisov, Gennadii Savrasov +3
CAD practitioners often work with proprietary parts that they are unwilling to share with external services. For individuals and small companies, bringing this work in-house also requires computational resources they can afford. We introduce CADMAGEDDON, an open-source agent that reconstructs editable CAD from meshes and can run locally on single-GPU systems and current-generation laptops. The method decomposes this complex task into a chain of simple visual decisions about what to construct and what to correct. Geometric tools support precise construction, allowing the agent to focus on visual reasoning. Using the mid-sized Qwen3.8-27B model, CADMAGEDDON outperforms prior approaches, including systems based on frontier models, in reconstruction quality on all the main benchmarks. CADMAGEDDON integrates with FreeCAD, bringing its reconstructions into practitioners' existing CAD workflows. Code: https://github.com/kulibinai/cadmageddon
agentbenchmark - arxiv:2611.22561 · cs.CVPlausible Videos Can Lead to Poor Decisions: Diagnosing and CURE-ing World Models for PlanningRaphael C. Kim, Michele Santacatterina, Jingsen Zhu, Ramin Zabih
Action-conditioned world models are increasingly used as world models for planning, yet a world model that makes plausible, realistic-looking videos that are consistent with the training data can produce poor decisions. We address two key questions: (1) when can we trust a world model for planning, and (2) what should we do when we cannot? We introduce CURE (Certify, Upweight, Rollout, Explore), a certificate-guided technique that assesses whether a world model can be trusted for planning and determines what to do next: whether to upweight the training data to debias the model; roll out more samples at test time; or explore by collecting new data where support (how well the training data cover the planner's decisions) is missing. We deploy CURE across multiple domains, spanning simulated physics based on PhyWorld, MuJoCo arm control, and Language-Table pushing in simulation and on real videos. The decisions our certificate trusts have lower regret than those it does not (e.g., 0.06 vs. 0.38 on Language-Table sim and 0.07 vs. 0.38 on MuJoCo contact), whereas video perceptual quality does not identify good decisions. By focusing on decision quality rather than perceptual quality, CURE lowers regret over the best pessimistic decision rule by 10-25% across support levels, and over world-action models, search and policy learning by 41-64%, given the same data.
world modelaction-conditioned - arxiv:2611.22557 · cs.LGBalancing Exploration and Exploitation in Constrained Combinatorial Optimization with U(1)-Symmetric Tensor TrainsSergei Iudin, Mikhail Podobrii, Dmitry Zheltkov, Stanislav Moiseev
Tensor Train (TT) decomposition has emerged as a powerful tool for mitigating the curse of dimensionality in a variety of machine learning and combinatorial optimization tasks. Recent advances show that TT-based generative models can efficiently represent and explore high-dimensional discrete spaces, even when the cost function structure is unknown. However, incorporating hard linear constraints into TT-based optimization remains challenging, as existing interpolation and sampling approaches often fail when feasible points are rare. In this work, we develop a general TT-based framework for constrained combinatorial optimization built upon U(1)-symmetric tensor networks. Our method introduces an expressibility-enhancement mechanism for constructing sparse symmetric tensors and a graph-based diversity measure that guides sample selection during training. On assignment benchmarks, expressibility enhancement and diversity-aware selection improve mean solution quality over their respective baselines. The assignment-selected diversifying strategy also extends to nonpolynomial Set Partitioning, with larger relative gains observed in the higher-tensor-rank data families. Separate runtime benchmarks show substantial speedups over ConstrainTNet and PROTES on non-degenerate assignment workloads; comparisons with simulated annealing depend on constraint information and computational budget, but, overall, support the practicality of the current implementation.
benchmark - arxiv:2611.22543 · cs.ROJointGV: Structure-Aware Correction of Visual World Models with Parallel Graph Dynamics for Long-Horizon ModelingTian Lyu, Yichen Jiang, Ziyuan Cheng, Yunda Yan +6
Visual world models predict future visual states conditioned on actions, but recursive errors in entity poses and spatial relations limit their long-horizon prediction for robot planning and control. We ask whether a separately predicted relational state can correct this drift without restructuring the visual model. This is challenging because graph nodes and visual states share neither elements nor a feature space. We propose Joint Graph--Visual Dynamics Correction (JointGV), a plug-in based on a cross-granularity dynamics hypothesis: despite their different representations, graph-node and visual-state changes describe the same physical transition. JointGV rolls out a task-conditioned scene graph alongside the visual world model, couples their predicted changes, and combines the resulting graph-motion context with graph-predicted node positions to compute a bounded residual. Graph-learning and correction gradients do not propagate into the original visual dynamics. We attach JointGV to DINO-WM (spatial tokens) and DreamerV3 (a global latent state), evaluating both on DMC Cheetah Run and Walker Walk across three seeds and extending the evaluation to contact-rich manipulation with DMC Finger Turn Hard. Across 32-step rollouts, JointGV improves structural and visual-prediction accuracy, reducing joint-position error by 11--28\%. For spatial tokens, learned graph--visual change coupling gives lower error than broadcasting the same mean graph-motion context to every token, particularly when non-adjacent parts appear close together in the image. On real-robot inverted-pendulum sequences, JointGV reduces DINO-WM's 32-step pole-angle error from $12.56^\circ$ to $6.69^\circ$ and improves DreamerV3 and LeWorldModel as well. In simulated inverted-pendulum control, an actor trained with JointGV-corrected rollouts completes 46 of 50 episodes, compared with none using the original rollouts.
manipulationworld modeldreamerv3scene graph - arxiv:2611.22541 · cs.CVFrom Perception to Simulation: Unveiling and Empowering Mental Spatial Dynamics in MLLMsJiachun Li, Zhuoran Jin, Kang Liu, Jun Zhao
Spatial reasoning is a fundamental component of world understanding for Multimodal Large Language Models (MLLMs). While existing research has advanced the perception of spatial relations in static scenes, the domain of dynamic spatial simulation remains relatively unexplored. To bridge this gap, we introduce MentalGrid, a benchmark focused on spatial reasoning tasks that require models to simulate evolving spatial states without intermediate visual feedback. Extensive evaluation on our benchmark reveals a significant deficiency in the spatial simulation capabilities of existing models. Specifically, we find that performance is constrained by perceptual limitations in simple simulation scenarios and hindered by inadequate multi-step reasoning as complexity increases. To address these challenges, we propose a three-stage curriculum learning framework utilizing 16k specialized spatial reasoning samples. This approach yields SpatialWalker, which achieves a 33.8% average improvement over its base model, surpassing proprietary models like Gemini-3-Flash in spatial simulation. Furthermore, SpatialWalker demonstrates robust generalization, maintaining a 4.9% average gain on out-of-domain benchmarks, demonstrating the pivotal role of spatial simulation as a fundamental building block for broader spatial intelligence.
curriculum learningbenchmark - arxiv:2611.22539 · cs.AIJET: Just-Enough Exploration Tuning for Efficient Tool-Calling Agents in Repository-Level Question AnsweringYoonho Choi, Taehyeon Han, Seonghyeon Lee
In repository-level question answering (QA), an agent explores a code repository with tools and should answer a question once enough information is gathered. Nevertheless, we find that LLM agents often continue to invoke tools even after the gathered information is already sufficient to answer. We call this phenomenon over-exploration. Moreover, even though existing methods improve task performance through supervised fine-tuning (SFT) on expert trajectories, SFT also propagates the over-exploration in the trajectories to the agent. To mitigate over-exploration, we propose Just-Enough Exploration Tuning (JET), a fine-tuning method that trains the agent to answer once the gathered information is sufficient. JET truncates each training trajectory at the step where the gathered information becomes sufficient, inserts a rationale that explains why the agent can answer at that step, and fine-tunes the agent with a higher loss weight on the rationale. Experiments on two repository-level QA benchmarks with two open-weight 8B models demonstrate that JET reduces tool calls by 11 to 33% on self-generated trajectories while largely preserving the score. On expert trajectories, JET makes 18 to 28% fewer tool calls than SFT while retaining most of the score gain of SFT. In both cases, JET also lowers the inference cost.
agentllm agentbenchmark - arxiv:2611.22537 · cs.LGTesting World Models on Actions They Have Never Seen: Held-Out Plans, Forecast Accuracy and Price Rules in RetailKarthik Iyer
A world model is trusted to answer "what if": how its world responds to an action it has never seen. We give a test of that trust for any action-conditioned world model: actions held out of training by design and scored against a known answer; whether forecast accuracy predicts accuracy on those actions; and the same structure given to simpler models. We run it in retail, where the action is a price and promotion plan, across 2,853 fitted models. First, forecast accuracy is a fair guide between kinds of model (rank correlation with error on new plans 0.32 to 0.66) and a poor guide between versions of one world model: where merchants plan into high-demand weeks the link reverses, down to -0.69. Second, training the world model to learn the effect of actions (JEPA pretraining, a latent consistency check, adaptation at the plan date) leaves its error on new plans unchanged. Third, in generated worlds the recorded regular price and three price rules improve it. Given the same rules, a log-linear model is best when its curve can match the truth; in a world whose curve lies outside the rules, the world model with the rules is ahead of the log-linear model with the rules in 36 of 40 worlds, and does as well without its recurrent cell. On real data (Breakfast at the Frat, The Complete Journey and M5) the rules lower the world model's error on pure price cuts deeper than any it saw (0.63 to 0.56 on the Frat; 0.45 to 0.44, within chance, on The Complete Journey), but not on lasting price moves. Fourth, a small share of random test plans helps every model a little. In a generated dosing world, a stress test in a second domain, the first finding repeats, and the built-up drug level lowers the world model's error (2.86 to 1.91) though it still predicts only 0.12 to 0.32 of the true gain. The test carries to any domain where a world model is asked about actions outside its history.
world modelaction-conditioned - arxiv:2611.22525 · cs.AIWhen Agent Skills Go Stale: Runtime Behavior and Selective Maintenance in Evolving EnvironmentsÖmer Faruk Akgül, Shuli Jiang, Yuting Zhang, Yi Zhang +3
Agent skills use past experience and current observations to guide decisions and actions on related tasks. This experience is retained in skill state, whose reuse can amortize inference-time reasoning and environment interaction. When environment changes, parts of this state can become outdated while others remain useful. Across four task settings and two models, we compare skill reuse before and after environment changes. Previously beneficial skills can reduce task performance by over 10 percentage points or increase execution cost by nearly 40% relative to no skill. Task success can also remain unchanged while execution savings decline. These effects depend on which environment-specific details become outdated and the agent's ability to compensate using current observations. We introduce selective maintenance, which replays checks from successful executions without additional model calls. It verifies whether observed differences conflict with skill state, repairs affected content, and validates updates, rebuilding when local repair is insufficient. No description of the change is supplied. Relative to reusing stale state, maintenance improves task accuracy by up to 67pp and reduces execution steps by up to 37%. These results suggest that preserving a skill's inference time value as environments evolve requires selectively maintaining its state.
agent - arxiv:2611.22515 · cs.LGRight Labels, Wrong Reasons: Signed Feature Attribution for Multi-Label ExplainabilityKenton White
When a multi-label classifier assigns the right labels, it may do so for the wrong reasons. Diffuse feature combinations may vary across classifier families and lack a dominant feature. This paper proposes FLEX (Fisher LDA Explainer), a global surrogate that fits a single direct multi-label Fisher LDA over the full binary label matrix, producing signed feature attribution lists for each label in a single coherent coordinate system. Across the evaluated benchmarks and classifier families, FLEX reproduces the black box's predictions with consistently high fidelity. Pooling covariance across labels outperforms a per-label alternative on most datasets, with the advantage growing largest where labels are numerous and positive examples are scarce. FLEX's fidelity-complexity Pareto curve is monotonically increasing, giving practitioners a predictable path to higher-fidelity explanations as more features are used, and its in-sample fidelity matches or leads Binary Relevance alternatives once a modest feature budget is allowed. The magnitude of each discriminant weight vector bounds its label's separability, flagging labels that may be predicted correctly for the wrong reasons. The evaluated classifiers largely agree on which predictions to make, but reach those decisions through different feature subsets when no single feature dominates.
benchmark - arxiv:2611.22514 · cs.CLMM-ReportBench: A Comprehensive Benchmark for Multimodal Report Generation from TablesTeng Lin, Nan Tang
The ability to generate multimodal reports integrating fluent narratives with faithful visualizations from tabular data, has become a cornerstone capability for modern vision-language models in practical domains such as finance, healthcare, and policy making. However, existing benchmarks are limited by single-modality focus, simplistic metrics, novelty evaluation, and visualization quality. We address these limitations with MM-ReportBench, a multimodal report evaluation benchmark covering 386 tasks across 185 real-world tables. Each task is paired with an expert-curated reference report featuring standardized chapter structures and verifiable key insight points, constructed through a hybrid pipeline of multimodal reports generation based on Monte Carlo Tree Search and rigorous human quality control. For robust evaluation, we design a dual-judge evaluation framework: Text Judge Model assesses textual quality along structural completeness, grounded insight verification with numerical fact-checking against source tables, and novelty analysis that distinguishes genuine analytical discovery from trivial paraphrasing; Visualization Judge Model evaluates generated visualizations across data fidelity, expressiveness, aesthetics, and technical correctness. They also evaluate cross-modal consistency. Extensive experiments on 20 state-of-the-art models reveal critical weaknesses: even the strongest model achieves only 68.9 overall score, with numerical hallucination affecting 38.5% of reports and genuinely novel insights remaining rare. Our ablation studies further demonstrate that single-judge evaluation consistently inflates scores by 8-12 points compared to dual-judge assessment, confirming that rigorous and reliable benchmarking demands multi-faceted, complementary evaluation probes.
benchmarkevaluation frameworkjudge model - arxiv:2611.22511 · cs.AILMLineage: Content Lineage and Context Governance for LLM AgentsWei Hou, Jingxin Geng, Xin Chen, Zheqing Jiang
Interpretability for AI agents should not be confined to understanding the model's internal weights. It must also cover the agent's reasoning and decision-making process over multi-source context. We present LMLineage, an out-of-band system that attributes every output sentence of an agent session to its supporting input spans with calibrated confidences, without requiring white-box access to the model or any additional inference passes. On top of the lineage metadata, we build concept-drift micro compaction: instead of waiting for the context-length limit and summarizing everything, the system detects when previously used input documents have stopped being referenced ("drifted") and surgically compacts only those, preserving user instructions verbatim. We evaluate on a multi-tier corpus: 71 gold annotated sentences from 10 Chinese web-QA cases, plus 90 turns of independently collected multi-turn sessions across three generator models (gpt-4o-mini, qwen3.8-27b, deepseek-v4-flash). Results show that (i) window-level embedding similarity is far better aligned with LLM paraphrasing than whole document or lexical matching; (ii) the greedy-window fast path is both faster and more accurate on average than exhaustive line-matching (a multiple-comparisons bias in the latter inflates web recall at the cost of overall accuracy); (iii) per-turn attribution completes in 1s on a single GPU, remaining stable as context accumulates to 40 turns; and (iv) concept-drift detection is conservative (0% false compaction) with recall bounded by attribution accuracy. We further describe the concept-drift micro compaction design and an interactive incident-tracing/memory-cleaning case study. A companion patent application covers the lineage construction method.
agentai agentllm agent - arxiv:2611.22497 · cs.AITowards Data-Efficient, Real-Time World ModelsRaluca Stevenson, David Bignell, Jennifer C. White, Tabish Rashid +8
State-of-the-art world models have been used to produce sequences of gameplay that accord with provided user-input actions. However, high quality, consistent gameplay generation often comes at the cost of inference speed, making real-time interactive play and ideation challenging. The substantial quantities of data used to train state-of-the-art models have also made custom training of these models appear impractical, limiting many of their potential uses in creative fields. In this work we demonstrate a major step towards efficient world models: achieving both fast high-quality inference, and improved data efficiency. This is done by replacing the next-token prediction paradigm with a two-stage discrete diffusion-based approach, introducing a lightweight refinement transformer which carries out iterative masked predictions. Building on an autoregressive world model capable of generating highly consistent and complex sequences over minutes (Kanervisto et al., 2025), we achieve substantial model speed-up (79x) with a modest quality trade-off. Additionally, we demonstrate that this model can be trained to reproduce high-quality, consistent gameplay using a dataset containing approximately one day (28 hours) of gameplay data. Our results demonstrate progress towards removing barriers to the use of world models in settings where real-time inference and data efficiency are essential.
world model - arxiv:2611.22491 · cs.ROProgress Predicts PolicyLicheng Zhong, Gim Hee Lee
A task specifies how the world should change, while a policy specifies what the robot should do. We introduce Progress Predicts Policy (PPP), which learns a language-conditioned scalar field through the commands induced by its differential. We obtain action conditioning by composing a latent progress field with a shared world model and read its action gradient at a fixed reference command. Continuous commands scale this gradient by a learned positive gain, while discrete commands use finite differences of the same scalar. This yields a hybrid policy without a separate action head or iterative search. Fitting demonstration progress values alone does not ensure that action gradients align with expert commands. PPP trains the induced commands by imitation and uses double backpropagation for the continuous gradient readout to shape the field's local control geometry. Demonstration timing calibrates field values on observed states without additional annotation. The continuous readout remains learnable without progress labels but these timing-derived targets support progress monitoring during execution. Experiments in simulated and real robot manipulation show that PPP outperforms behavioral cloning under a shared architecture.
manipulationaction headworld model - arxiv:2611.22490 · cs.AISciWorkBench: A Benchmark for Real Scientific WorkflowsYuchen Zhang, Taolin Han, Jinghang Wang, Jinxin Wang +18
Language models are increasingly being used as agents capable of planning and executing projects, as well as refining their actions through feedback. This shift raises a question: can current agents complete projects like real scientists? To examine this question, we introduce SciWorkBench, a benchmark for evaluating agents' abilities on real science tasks across biology, chemistry, materials science, and physics. SciWorkBench is built from peer-reviewed literature or expert-contributed research scenarios. We also propose a unified scientific analysis framework validated by domain experts, which can categorize most tasks in SciWorkBench. Each task covers one or more stages of this framework. We evaluate up-to-date models within the Claude Code and Codex agent frameworks under a unified execution and grading protocol. The best-performing agent achieves a mean score of 57.38, indicating that current agents still struggle with realistic scientific workflows. Our failure analysis identifies three recurring weaknesses at the workflow level: difficulty translating scientific knowledge into task-specific procedures, weak integration of evidence across different modalities, and limited capacity for scientific self-correction. These findings highlight the importance of workflow-based evaluation for measuring progress toward reliable AI systems for scientific research.
agentagent frameworkself-correctionbenchmark - arxiv:2611.22483 · cs.ROManufacturing Differential Sparsity: Planner-Hardware Co-Design for Efficient World Model PlanningWenbo Zhu, Paul Yi-Chia Chen, Yuanhan Li, Zishen Wan +1
JEPA-style latent world models let embodied agents plan toward new goals with model predictive control, but they are still too slow for real-life embodied control: sampling-based MPC repeats predictor inference tens of thousands of times per plan, and even on high-end GPUs one planning step can take tens of seconds. We observe substantial latent locality in this repeated-inference workload: the predictor maps close actions to close latents, and inside a quantized predictor this closeness takes a form that hardware can exploit, differential bit-level sparsity. Unlike an input stream, whose temporal similarity is given, a planner itself decides which candidates it evaluates together, so the planner can choose its own references and, with reference forest sampling, manufacture more sparsity: it couples the sampled action sequences of each group to a shared root, with competitive planning quality. We build the accelerator on the insights of the rich line of bit-level sparsity accelerators and design it and the planner tightly coupled as a single co-design: the planner increases sparsity and exposes structure for hardware scheduling, and the bit-exact hardware effectively uses the sparsity that the planner creates. On three JEPA world models, our co-design is 1.28-1.80x faster than a dense accelerator and uses 38-52% less energy, and it is 1.22-1.58x faster than prior bit-sparse design and uses 16-36% less energy, with the same area. To our knowledge, our planner is the first planner that manufactures differential bit-level sparsity across its candidates and exposes its structure for hardware scheduling, and our co-design is the first algorithm-hardware co-design for JEPA-style world-model planning.
embodiedworld modelembodied agent - arxiv:2611.22482 · cs.CVLeViT-UNet++: Towards Balanced Fine-Grained Details and Context for Efficient Medical Image SegmentationGuoping Xu, Jinling Zhu, Liuyue Li
Efficient medical image segmentation is essential for clinical applications requiring fast and accurate delineation, such as surgical navigation, radiation therapy, and real-time image-guided intervention. Despite substantial advances in convolutional neural networks and Transformers, attaining high accuracy under limited computational resources remains difficult, because reducing model complexity often sacrifices fine-grained structural details and global context modeling, yielding an unfavorable accuracy-efficiency trade-off. To address this, we propose LeViT-UNet++, an efficient encoder-decoder architecture that better balances local detail preservation and global context modeling. Specifically, we introduce a Haar Wavelet Multi-Scale Downsampling (HWMD) module that decomposes feature representations into low- and high-frequency components to extract multi-scale structural information, thereby mitigating detail loss during downsampling. We further develop a depthwise convolution-based decoder that integrates multi-scale global representations from Transformer blocks with local spatial features, enabling progressive feature refinement while reducing computational overhead. Evaluated on four benchmarks (Synapse, ACDC, LA2018, and MSD Heart) under both 2D and 3D settings, LeViT-UNet++ achieves competitive accuracy with a compact parameter footprint and fast inference. On Synapse, for instance, it attains a DSC of 76.57% with only 8.025M parameters and 241.72 FPS, confirming a favorable accuracy-efficiency balance. These results demonstrate the effectiveness of combining wavelet-based multi-scale feature extraction with lightweight local-global feature fusion for efficient medical segmentation. The source code is available at https://github.com/apple1986/levit_unet_pp.
benchmark - arxiv:2611.22480 · cs.AIRehearsalBench: Can AI Predict What Happens When You Change a Company?Mika Siddiqui, Tharaka Perera
Organizations increasingly ask AI systems what will happen if they change how they operate: which work to automate, which team to resize, what a voice agent will do to waiting times. We introduce RehearsalBench, a benchmark that measures whether AI models can diagnose an operation, choose an intervention, and forecast its consequences. Each model is placed in a simulated bank contact center fitted to real call-level records, sees only what a manager would see, and submits a plan and forecasts that are then played out across 200 seeded futures and three plausible versions of the organization. Decisions are scored in dollars at the stakeholders' stated valuations; forecasts are scored on the effect of each change relative to doing nothing. Ten models made 280 attempts across 14 organizations. The best model, GPT-6 Astra, improved on doing nothing by 6.8% of the no-change cost and made 77% less forecast error than a forecaster who always predicts no effect. On average, every model forecast the direction of every specified effect correctly, but models erred on the size of the same effects in the same direction, overstating a voice agent's effect on repeat contacts about five-fold. No model ever recommended holding off when doing nothing was the right choice. To test external validity, twelve models forecast 33 real organizational changes with known outcomes. Under verdict rules fixed in advance, the benchmark's forecast ranking agreed with the real-world ranking (Spearman $ρ= 0.92$, 95% lower bound 0.50). We report the result alongside an inconclusive first test, the analysis choices made after outcomes were seen, and evidence that part of the agreement reflects how well models size their uncertainty.
agentbenchmark - arxiv:2611.22472 · cs.CVDisagree to Prune: Reallocating Visual Tokens Where a Small VLM DissentsYanchao Li, Jiaqing Xie, Ben Gao, Yusu Zhang +6
Visual token pruning speeds up large vision-language models (VLMs) by keeping only a part of their visual tokens. Most existing methods improve the ranking of the tokens and then answer every question once with the same budget. With strong rankings, pruning hurts only a few questions, and signals computed on the kept tokens see little of what it removed. We find that, on most of these questions, a small VLM given the full image disagrees with the pruned answer, since it still sees the missing evidence. We propose PRUDENT, which first answers with about half the budget and spends the saved tokens on re-answering the disagreeing questions at twice the budget. We show when such re-answering pays and why the small VLM should read the full image. On 18 benchmarks, PRUDENT outperforms eight existing pruning methods at the same average budget of visual tokens. Its lead widens as the budget tightens, and holds across five InternVL sizes. PRUDENT also has the highest average in two more model families. The disagreement of the small VLM flags most pruning failures of all eight methods, and re-answering on it also improves the two strongest with their own rankings. Other checkers give similar gains, and the extra tokens go where the first pass hurts the answer. These results suggest that identifying which pruned answers are trustworthy provides an additional lever for visual token pruning beyond the ranking of tokens. Code and data are available at https://github.com/LiZaiyuan0619/PRUDENT.
benchmark - arxiv:2611.22470 · cs.LGEscaping Latent Traps by Remembering: Training-Free Episodic Kernel Repulsion for World-Model PlanningYican Zhao, Bolin Yang, Ruizhe Wang, Haiyang Guo +3
Latent world models plan towards a goal image by minimising the distance between predicted and goal latents. This distance is not geodesic: it has local minima, latent traps, behind walls, at corners and in dead ends, and memoryless receding-horizon replanning keeps the agent in them, revisiting states far from the goal. We propose Episodic Kernel Repulsion (EKR), a training-free cost term that penalises candidate plans by a top-$k$ Gaussian kernel to a novelty-gated memory of the latents the agent has actually visited, rescaled by the current base cost so that its strength is comparable across models. Across two JEPA world models and four benchmarks, EKR raises average success from 31.8% to 46.7% (LeWM) and from 54.7% to 64.0% (H-JEPA) with the same model, planner and budget. Same-memory counterfactuals show that, in most settings, EKR pushes plans away from visited states, in navigation, often at a higher base cost; stuck episodes and revisits drop accordingly. In seven of eight settings it matches or exceeds $4\times$ samples at $0.8$--$1.5\times$ compute. Without tuning, it transfers to DINO-WM with CEM: $+6.7$ and $+19.3$ points on official and longer-goal Wall tasks, and from 44.7% to 100.0% and from 21.0% to 32.0% on long-horizon PointMaze and Push-T.
world modelmemoryagentbenchmark - arxiv:2611.22464 · cs.CVMAVE: Reconstructing Editable 3D Blockouts from Film ShotsYifan Chang, Jian Yang, Jiangnan Shao, Xiaofeng Wang +5
Animated 3D blockouts support previsualization and provide structural conditions for video generation, but manually reconstructing them from reference videos is costly. This limits both their use by creators and the production of paired training data. We introduce MAVEBench, a benchmark of 50 film shots and 416 keyframes for evaluating camera motion, spatial layout, and character blocking. We propose MAVE, a video-to-blockout framework combining Metric observation, Agentic modeling, and evidence-based VErification. It first recovers cameras, human motion, and geometric evidence in a shared metric coordinate system. An agent then constructs an editable environment under constraints from these observations, while programmatic checks report inconsistencies to guide refinement. The outputs are editable Blender scenes and frame-aligned blockout videos. On MAVEBench, MAVE improves overall scores by 27.5-41.1 points across three agents over their corresponding vanilla baselines. With DeepSeek-1017, it achieves a score of 73.2 at an average API cost of only USD 0.054 per shot, outperforming the strongest vanilla baseline while reducing total reconstruction time by 53%. We also construct MAVEData-3K, a dataset of 3,000 blockout-video pairs for model training, and will release it publicly. Creators can change viewpoints and character paths, then render the edited scenes to condition video generation. Code is available at https://github.com/yfChang-cv/MAVE.
agentagenticbenchmark - arxiv:2611.22461 · cs.CLSurfacing Disagreement Beats Resolving It: A Gated Ablation Study of Provenance-Weighted Memory for Multi-Agent DebateZiaullah Khan, Abdul Rehman Khalid, Shrestha Nischal Lal, Nheng Vanchhay +2
Shared memory for multi-agent debate is usually built to resolve inconsistency: retrieval returns one winning fact and the disagreement disappears before any agent sees it. We ask whether that is correct. In DebateGraph, four backends share one contract (read/write/snapshot), so five mechanisms from the agent-memory literature can be removed independently under a fixed budget, each gated on a pre-declared threshold. Across 39 runs one survives: preserving contradictions and passing the ranked conflict set to the debate, rather than collapsing it to a winner, improves F1 by +0.340 (95% CI [+0.267, +0.414]) and cuts attack success from 22 to 7 of 122 poisoned items (McNemar exact p=0.0003), at equal cost. Reaching it meant fixing a bottleneck in our harness: conflict sets are built only over retrieved edges, and our pre-declared cap left three quarters of planted contradictions unretrievable, first measuring the mechanism at +0.084. The effect replicates on two retrieval substrates and directionally on HotpotQA documents. No other mechanism survives, and two are worse than inert: once contradictions reach the prompt, a write trust-gate and the router's provenance features each cost accuracy (-0.075 and -0.096 F1). The gate fails structurally: poisoned objects are always entities the graph already contains, against 53% of gold, so groundedness scores the lie above a truthful first mention. We release the harness, attack set, and logs, and argue that silently reconciling memory is the assumption worth abandoning.
memoryagentmulti-agent - arxiv:2611.22458 · cs.LGSample-Aligned Decoders for Latent Generative ModelsYu Lei, Yi Li, Hongcheng Gao, Yipeng Wei +3
Latent generative models synthesize images by decoding generated latents, yet their decoder is trained only to reconstruct encoder latents and never sees the latents it decodes at test time, so the residual errors of the generator are rendered directly into the final image. Existing remedies retrain the generator, add cost at every sampling step, or make the decoder robust to perturbed real latents, assuming that generator errors resemble such perturbations. We find that they do not. Generated latents lie on the encoder latent manifold as closely as real ones, yet follow a different distribution. The decoder should therefore be adapted to the distribution of generated latents, which raises two challenges: (1) unpaired supervision, since a generated latent has no ground-truth image, and (2) latent-space drift, since a decoder freed from paired targets forgets the latent space it must decode. We propose Sample-Aligned Decoders (SAD), a post-training framework that fine-tunes only the decoder of a frozen generator at no inference cost. Generated-Sample Distribution Alignment matches decoded generated latents to unpaired real images in a frozen foundation-model feature space, and Encoder Anchor Preservation keeps the decoder close to the original one on encoder latents. Our analysis shows that the alignment is effective only on generated latents, where it corrects each sample in proportion to how much more often the generator produces such latents than the encoder; on encoder latents it can only work against the anchor. SAD improves four state-of-the-art ImageNet generators across three tokenizers, e.g., FID-50K from 1.15 to 0.98 for REPA-E, complements discriminator guidance, and extends to text-to-image generation. Our code and models are available at https://torpedo2003.github.io/SAD/.
post-training - arxiv:2611.22456 · cs.LGPACD: Provenance-Aware Consensus Decoding for Mitigating Cross-Modal Hallucinations in Omnimodal Large Language ModelsChang Sun, Yijun Tian, Zhenwei Tang, Yuchen Deng +3
Omnimodal large language models (Omni-LLMs) integrate audio, video, and text for multimodal reasoning, yet remain vulnerable to cross-modal hallucinations, where one modality improperly influences predictions about another. Existing training-free decoding methods mitigate these errors through modality perturbation or relevance-based weighting. However, due to interference from irrelevant modality cues during fusion and difficulty detecting shared errors, these methods may suppress correct unimodal predictions and allow erroneous consensus to persist. To address these limitations, we propose Provenance-Aware Consensus Decoding (PACD), a training-free framework for mitigating cross-modal hallucinations. PACD comprises two complementary paths based on whether the branches reach consensus. For non-consensus cases, Relevance-Guided Candidate Arbitration identifies task-relevant unimodal evidence and selectively arbitrates between predictions using alternative candidates. For consensus cases, Provenance-Based Consensus Recovery leverages the outranking principle to evaluate the consistency and strength of layer-wise modality evidence for alternative answers, selectively resolving erroneous consensus under confidence constraints. Extensive experiments on four Omni-LLMs and five benchmarks across two tasks demonstrate the superiority of PACD. Compared to existing methods, PACD achieves up to 27.6% relative improvement over standard decoding across CMM, AVHBench, and SVHalluc for cross-modal hallucination mitigation, and an average relative improvement of 7.7% over best baseline across OmniVideoBench and FutureOmni for general audio-visual reasoning.
benchmark - arxiv:2611.22455 · eess.SYMARS: Feature-guided physics-informed deep learning for real-time battery health and lifetime predictionInnocent Izabayo, Haiquan Zhao
Accurate and robust estimation of battery state-of-health (SOH) and remaining useful life (RUL) is essential for safe and cost-effective operation of electrified transportation systems, yet it remains challenging under operating variability, non-stationary aging, and non-Gaussian measurement disturbances. This paper proposes Multi-scale Adaptive Robust State Estimation (MARS), a multi-scale adaptive robust state-estimation framework that couples a dual-branch temporal encoder with an equivalent-circuit-model-constrained differentiable generalized-correntropy adaptive Kalman filter. The encoder disentangles fast intra-cycle voltage-current-temperature dynamics from slow inter-cycle degradation, and the fused degradation context adaptively parameterizes the cycle-level process covariance, measurement covariance, and correntropy bandwidth. Through the unrolled robust filtering recursion, MARS performs physics-consistent state propagation and generalized-correntropy-based innovation correction, while uncertainty decomposition and Monte Carlo dropout provide calibrated prediction intervals for SOH estimation and conditional RUL prediction. Experiments on the NASA and CALCE benchmarks demonstrate improved SOH accuracy, achieving mean absolute error/root mean square error values of 0.72%/0.92% and 0.63%/0.80%, respectively, together with competitive conditional RUL performance against representative data-driven and hybrid baselines. The resulting conditional-RUL prediction intervals achieve coverage probabilities of 100.00% and 99.24% on NASA and CALCE, respectively. Ablation, disturbance-sensitivity, near-end-of-life, and runtime analyses further verify the contribution of each module and the practical inference cost for online battery management.
benchmark - arxiv:2611.22453 · cs.ROVoxelSBFL: Voxel-Level Spectrum-Based Failure Prediction for Vision-Language-Action ModelsArusa Kanwal, Pablo Valle, Paola Roggero, Miren Illarramendi +2
Vision-Language-Action (VLA) models integrate visual perception, language understanding, and action generation to control robots. Testing these systems produces trajectories and task outcomes, but existing approaches provide limited guidance on using these recorded trajectories to assess subsequent executions or inform policy repair. We conjecture that, for a given policy and task, traversing certain workspace regions is associated with a higher likelihood of failure. To investigate this conjecture, we propose VoxelSBFL, a black-box adaptation of Spectrum-Based Fault Localization (SBFL) in which workspace voxels replace program elements as units of analysis. Recorded tool-center-point (TCP) trajectories provide binary voxel coverage, which is combined with execution outcomes to construct suspiciousness maps using established SBFL formulas. These maps identify failure-associated regions and score unseen trajectories without requiring internal model access. We evaluate 34 spectrum-based techniques across four VLA models using 96,000 Pick-Up trajectories in simulation and compare the leading techniques against eight action- and trajectory-based metrics. VoxelSBFL obtains an overall macro ROC-AUC of 0.704, significantly outperforming state-of-the art baselines. Sensitivity analyses show that performance varies with spatial granularity and training-set size. These findings support spatial spectra for VLA execution assessment and motivate future research on runtime monitoring, self-healing and targeted policy improvement.
vision-language-actionvlavla model - arxiv:2611.22446 · cs.CVAs Student Rollouts Continue, Visual Backdoors Keep GrowingZane Fu, Yanzhao Zheng, Jinze Liu, Ziting Qin +6
Conventional knowledge distillation can facilitate the propagation of backdoors from teacher models to student models. However, how this risk manifests in On-Policy Distillation (OPD) remains insufficiently studied. OPD has recently emerged as an important post-training paradigm for Vision-Language Models (VLMs). Unlike traditional offline distillation, OPD does not require teacher-generated responses. Instead, it uses the teacher's probability distributions to supervise response trajectories generated online by the student. Since the initial student model typically generates benign responses and receives no explicit supervision from malicious target responses, OPD appears to offer some inherent resistance to backdoor propagation. In this paper, we challenge this intuition and reveal a more subtle risk of visual backdoor propagation in OPD. Through experiments across multiple model families and datasets, together with theoretical analysis, we identify three main findings. First, unlike the relatively gradual propagation observed in traditional offline distillation, backdoors rarely emerge in OPD at low poisoning ratios. However, the attack success rate rapidly approaches 100% once a critical threshold is exceeded. Second, backdoor propagation does not depend on explicit malicious responses. Instead, it is driven by the continuous accumulation of anomalous probability mass assigned by the teacher. Finally, based on these findings, we propose a novel trigger design that enables visual backdoor propagation with zero explicit trigger injection. These findings further reveal an overlooked mechanism of backdoor propagation in OPD and highlight its potential security risks.
post-training - arxiv:2611.22445 · cs.ROAutonomous Track Racing with a Commercially Available Humanoid: System Design and Real-World DeploymentKoki Inami, Takashi Sato, Hiroyuki Oka, Shota Takizawa +6
Learning-based methods now let humanoids run robustly, and they have begun to compete in athletic events. Yet few reports describe a complete autonomous humanoid running system in an actual race, where the robot must follow a lane at a new venue, pace itself within hardware limits, and finish without intervention. We report such a system deployed on the Unitree G1, a commercially available general-purpose humanoid, in the 100 m, 400 m, and 1500 m events of the World Humanoid Robot Games 2026. The system integrates vision-based lane perception, path tracking, and a running policy trained by reinforcement learning with a human motion prior. To address the challenges of humanoid running, we stabilize head-camera images against body motion and use IMU-assisted leg odometry to bridge flight phases. To manage motor heating, we penalize hip-pitch torque during training, add heat-dissipation plates, and apply temperature-based speed limiting with event-specific pacing. The robot placed 13th of 74 in the 100 m, 10th of 44 in the 400 m first round before falling in the semifinal, and seventh of 35 in the 1500 m. In this deployment, thermal margin set the pace in every event, and the semifinal fall followed a lane-detection error.
humanoid - arxiv:2611.22444 · cs.LGEfficient Machine Learning Interatomic PotentialsYury Lysogorskiy, Anton Bochkarev, Ralf Drautz
Machine learning interatomic potentials (MLIPs) reach an accuracy close to quantum mechanical reference calculations at a small fraction of the cost. Efficiency is the purpose of MLIPs, yet strategies for efficient MLIPs have not been reviewed systematically. This review takes efficiency as its guideline. To compare MLIPs on a common footing, we represent them in a complete basis expansion, the Atomic Cluster Expansion (ACE) and its graph generalization (GRACE). The expansion is canonical, built from products of single-particle functions. Classical interatomic potentials, such as the embedded-atom method, and many modern MLIPs, including equivariant transformers, can be represented within this expansion. Much of the apparent diversity of MLIP architectures therefore reflects how non-linearity, layer coupling and optimization are handled rather than which interactions the models can represent. We show how the cost of an MLIP becomes linear in system size and, at fixed rank of the tensor decomposition, in model complexity, and we discuss the role of machine learning, which provides non-linear parameterizations and optimization. Because efficiency is meaningful only at a given accuracy, we also compare accuracy and cost. On the Matbench Discovery leaderboard, models with three to five layers approach the accuracy of deeper models, and one basis, truncated along the Pareto front of error and speed, spans more than two orders of magnitude in cost. Ultimately, training data determine the quality of any MLIP.
leaderboard - arxiv:2611.22438 · cs.LGEyes in the Update? Learning Where Rewards Land in Multimodal ReasoningLixuan He, Tianzhe Chu, Shulin Tian, Shikang Zheng +2
Multimodal post-training has moved from broad instruction tuning toward reinforcement learning for image and video reasoning, where outcome rewards can improve final-answer accuracy. Yet these rewards are assigned to complete generated trajectories, while visual evidence is used only at sparse and uneven positions inside the response. Recent work has begun to make multimodal RL more token-sensitive through grounding signals or external token-level weighting. We ask a more specific question: can the allocation of the same trajectory-level reward be inferred from the model's own multimodal computation, rather than from an auxiliary annotator, perception module, or process label? We propose Gaze-RL, a representation-driven policy optimization method that redistributes completion-level advantage through a latent reward landing field. The field is computed from internal cross-modal geometry and local predictive uncertainty, assigning larger update weight to positions that are both visually anchored and decision-bearing while remaining mean-one, so the total credit of each rollout is preserved and only its positionwise allocation changes. Gaze-RL requires no process supervision, auxiliary perception module, or architectural modification, and improves both Qwen2.5-VL and InternVL3.5 across a broad range of image and video benchmarks.
post-trainingbenchmark - arxiv:2611.22434 · cs.LGEncoding, Decodability, and Use: A Five-Layer Framework for Benchmark Measurement Audits with Evidence from UAV Intent RecognitionZhirong Su, Tian Zhang
Benchmark accuracy is routinely read as evidence that a model understands the construct the benchmark measures. We propose and instantiate on AeroIntent a five-layer framework that decomposes that reading into observation, representation, decoding, use, and behaviour. The framework separates encoding, decodability, and use as distinct levels. In the instantiation, the observation layer is unreliable on 7 of the 20 Drone programs and floors the direct visual correlate of radial velocity at a 0.26\% scale change. The representation layer exposes two task-designed RGB features with nominal signals. RAFT flow radial bias reaches $r=+0.257$ ($p=0.030$), and bounding-box optical-flow divergence reaches $r=+0.203$ ($p=0.029$). No multi-feature readout gives a fold-mean test correlation above the cluster minimum detectable effect (MDE), and the largest positive global out-of-sample correlation is $r=+0.080$. The decoding layer sets the MDE at $|r|=0.599$ at the multi-feature Bonferroni threshold ($α=0.00714$, power 0.80), and the observed $r=+0.080$ has power 0.00. Two injection experiments establish the MDE's condition from both sides: aligned injections are detected at or above the simulated power, and orthogonal injections are not detected. The use layer finds accuracy consistent with program memory (program-identity 0.937/0.597/0.210) and multi-region static cues. The behaviour layer is consistent with accuracy not requiring a radial-velocity readout (appearance 0.6552 vs bounding-box trajectory 0.4080, cluster-level paired $Δ=+0.2471$, $p=0.0132$).
memorybenchmark - arxiv:2611.22432 · cs.AIThink Once, Act Fast: Cost-Effective Recursive Self-Improvement for Decision Model HarnessSiqi Lai, Haitao Li, Hao Liu
Fast System-1 models enable responsive actions but struggle with complex or unfamiliar tasks. Large language model (LLM)-based System-2 provides strong reasoning capabilities, yet repeated invocation incurs substantial latency and computational costs. To bridge this gap, we propose JevRSI, a System-1/System-2 collaboration framework that transforms LLM reasoning into transferable decision support through recursive harness improvement. A Jev-style System-1 model enables low-latency action selection, while an asynchronous System-2 LLM develops a modular harness. The harness integrates procedures, state memory, and world models to provide decision contexts spanning the past, present, and future. Utility-driven retrieval selects modules based on both contextual relevance and their utility on previous tasks. Environmental feedback guides targeted revisions and enables recursive self-improvement. Experiments across three datasets show that JevRSI consistently improves decision quality, matching or surpassing LLM agents in several settings. It preserves low-latency execution with favorable performance-cost trade-offs. Moreover, System-2 development interventions decrease as the harness evolves, amortizing System-2 costs through continuous improvement. Our project is available at https://github.com/usail-hkust/JevRSI.
world modelllm agentself-improvement - arxiv:2611.22427 · cs.LGDo Diffusion Models Learn to Generalize Basic Visual Skills?Amish Sethi, Boya Zeng, Wenhao Chai, Zhuang Liu
Diffusion models generate high-quality images, but how well they generalize is hard to measure. Current evaluations rely on human preference and prompt-following benchmarks for text-to-image models and on distribution-level metrics such as Fréchet Inception Distance (FID) and Inception Score (IS) for class-conditional models, but these cannot tell whether models learn basic visual skills or largely match training patterns. To address this gap, we introduce the Visual Generative Lab (VGL), a controlled experimental framework for studying how diffusion models learn and generalize basic visual skills, including size, position, rotation, count, color, shape, and their compositions. By generating synthetic training data for each skill from explicit rules, VGL allows precise measurement of generalization with task-specific, rule-based metrics. Training models from scratch on these data reveals three patterns in this controlled setting: (1) with our default design, extrapolation is limited: size and position degrade within a few pixels of the training range, count reaches only about 50% accuracy on out-of-range counts, and rotation extrapolation accuracy varies widely across seeds; (2) for compositional generalization, the number of distinct basic visual skill combinations matters more than dataset size; (3) skills mostly fail independently, except position. Code and data-generation scripts: https://github.com/AmishSethi/visual-generative-lab.
benchmark - arxiv:2611.22425 · cs.CLLMMR: Learned Multigranular Memory Retrieval for Small Language Model AgentsPengfeng Song, Yunkai Zhuang, Liye Shi
Small language model agents have limited space to read the memory needed for multi-hop questions. Relevant facts are often scattered across records whose surrounding text consumes much of that space. We propose LMMR, a learned multigranular memory retrieval method that selects both the next source and how much of it to read. LMMR links paragraphs and sentence fragments to their original sources and pairs views that supply the same missing support during training. Relevant distractors and repeated information provide contrasting examples. At inference, the reranker scores what each candidate adds to an anchor passage, and a selector chooses the evidence within a token budget. On 22,398 official validation questions from HotpotQA, 2WikiMultihopQA, and MuSiQue, a frozen 2B answer model reaches 61.47, 43.26, and 30.58 F1 under a 400-token budget. The gains over query expansion are 3.06, 2.14, and 4.55 points, while memory input decreases by 23.1-26.9%. LMMR also uses 23.5-27.7% fewer input tokens than a coarse retriever trained on the same questions. Against that retriever followed by LLMLingua-2 at comparable input lengths, F1 gains reach 3.99 points and are significant on HotpotQA and 2WikiMultihopQA after Holm correction. All eight small answer models evaluated on HotpotQA improve over query expansion and MMR in both F1 and exact match.
memory - arxiv:2611.22418 · cs.CVCutAgent: Creative-Gene-Driven Planning and Experience-Guided Self-Refinement for High-CTR E-commerce Advertising Video CreationXuyue Huang, Guanbo Huang, Jiasong Feng, Run Ling +9
E-commerce advertising video creation requires both high-level creative planning and reliable multi-modal production. However, existing methods often map product attributes directly to scripts or editing decisions without explicitly modeling advertising strategies, narrative structures, and production styles. They also have limited ability to reuse past production failures. To address these limitations, we propose $\textbf{CutAgent}$, a dual-agent framework that formulates advertising video creation as a Plan--Execute--Self-Evolve process. CutAgent comprises two collaborative agents, $\textbf{GeneAgent}$ and $\textbf{VideoAgent}$, connected by a structured creative gene that bridges product understanding and video production. GeneAgent maps multi-modal product inputs---textual attributes together with product images---to creative genes that specify narrative units and production rules. To train GeneAgent, we construct $\textbf{CutAgent-70K}$ with approximately 70K samples of product attributes/images, reverse-extracted creative genes, cleaned clip pools, and real-world CTR, and apply high-CTR supervised fine-tuning followed by CTR-guided GRPO. Guided by the generated gene, VideoAgent retrieves category-then-gene failure--correction experiences from a long-term memory as constraints for shot-script planning, then coordinates material retrieval, narration, subtitles, music, and timeline rendering. Independently, a reflection loop diagnoses quality defects, revises deficient production decisions, and writes verified corrections back to memory for later retrieval. Experiments on $\textbf{CutAgent-Eval}$ show that CutAgent outperforms representative baselines in e-commerce advertising video creation, achieving the highest online CTR of $\textbf{2.21\%}$.
memoryagent frameworkself-refinement - arxiv:2611.22415 · cs.CVSpatialMind: Adaptive Multi-Level Geometric Fusion for Spatial Reasoning in Multimodal Large Language ModelsShibin Zhao, Hao Tan, Fan Yang, Haixin Wang +5
Multimodal Large Language Models (MLLMs) demonstrate promising capabilities in 2D visual understanding, yet remain limited in 3D spatial reasoning. Recent studies explore integrating 3D reconstruction models into MLLMs. However, these approaches either use a single geometric granularity across all tasks or adopt a fixed multi-level fusion scheme, regardless of task-specific reasoning requirements. Through systematic experiments on multi-level geometric fusion, we find that spatial tasks at different cognitive levels (e.g., distance estimation vs. route planning) require geometric representations of different granularities, revealing that fixed fusion cannot accommodate diverse spatial reasoning requirements. To bridge this gap, we propose SpatialMind, a task-adaptive framework for dynamically integrating multi-level geometric representations. SpatialMind employs GeoFus, a multi-layer geometric fusion module built around a gating network jointly conditioned on textual instructions, visual semantics, and geometric representations. This network adaptively modulates the contributions of shallow, intermediate, and deep geometric features across the layers of the LLM decoder. Furthermore, we introduce a negative-entropy regularization loss to mitigate gate collapse. Gate visualizations indicate that GeoFus adapts geometric granularity to task requirements. Experiments demonstrate that SpatialMind achieves competitive performance across spatial benchmarks, attaining a state-of-the-art score of 72.3 on VSI-Bench. The source code is publicly available at https://github.com/zhaosb-star/SpatialMind.
benchmark - arxiv:2611.22407 · cs.ROROSMonitoring in Action: Runtime Monitoring and Enforcement for ROS 2 ActionsGiacomo Paltrinieri, Angelo Ferrando
Robotic applications often rely on long-running operations, such as navigation, manipulation, and inspection, which in ROS~2 are commonly implemented as actions. Unlike topics and services, actions involve a structured lifecycle comprising goal submission and acceptance, feedback, status updates, result retrieval, and cancellation. Existing versions of ROSMonitoring support topics and services, but do not reconstruct this lifecycle or expose action-level events to a runtime-verification oracle. In this paper, we extend ROSMonitoring with runtime monitoring and enforcement for ROS~2 actions. Actions are declared directly in the monitoring configuration and automatically instrumented through the underlying ROS~2 communication endpoints. The generated monitor reconstructs action-level events, correlates them through goal identifiers, maintains the state of active goals, and translates violations into protocol-aware enforcement operations. These include goal rejection, cancellation, message suppression, and protocol-compliant responses generated directly by the monitor when the original communication cannot be forwarded. We evaluate the implementation through controlled stress tests that vary the number of action invocations, workload size, and clientâserver topology. The results show approximately linear growth in monitoring cost with repeated action invocations under a fixed workload, while larger action workloads and a greater number of configured clientâserver pairs introduce additional overhead. These results characterise the scaling behaviour of the current implementation under the tested sequential workloads while highlighting the main performance limitations of the current synchronous implementation.
manipulation - arxiv:2611.22404 · cs.AISymbolic Abstraction RefinementDaniel Bryce, Robert P. Goldman, David J. Musliner, Jeremy Gottlieb +1
Autonomous systems choose actions whose outcomes are uncertain and are subject to time constraints. Probabilistic timed automata (PTAs) model exactly this, and verifying one answers a question about the system's decision-making: the maximum probability of reaching a goal, or the minimum probability of reaching a hazard, that a controller can guarantee. Verification normally works over symbolic states, each pairing a discrete location with a region of clock values, and its difficulty is that the best behavior generally varies within such a region. The PRISM verifier's most capable PTA method responds by splitting regions until each is uniform, rebuilding the transitions that reach them and propagating the split backward through the state graph of a stochastic game, a cascade of graph reconstruction that partitions regions the property never depends on. We present SYMAR, which instead makes the value function symbolic: each symbolic state carries a function from sub-regions of its clock region to probability bounds, so refinement is the appearance of new sub-regions in that function, and previously generated states are never split nor their incoming transitions reconstructed. Both sources of PTA nondeterminism, the timing and the choice of actions, are captured implicitly by the preimage computation that produces those sub-regions. Heuristic forward search then bounds how much refinement is done at all, expanding an And-Or graph from the initial state along the most promising behaviors so that effort is confined to the property-relevant fragment, guided by bounds and preferred actions from a single unrefined run of the existing method; a partial-refinement backup keeps the otherwise combinatorial sub-region count lean, and every stored interval provably contains the true probability. On the PRISM PTA benchmarks, whose subproblems recur in autonomous systems (distributed agreement under contention, leader election, and self-configuration without central coordination), SYMAR is up to 108x faster, solves an instance on which the existing method exhausts the time limit, and is markedly less sensitive to the deadline constant, resolving one instance in a 37-node search graph, unchanged as the deadline triples, where the existing method needs 86,380 abstract states and 73 refinement rounds.
benchmark - arxiv:2611.22395 · cs.ROCoBoVLA: Decoupling Language from Action with Discrete Codes for Robust and Interpretable Robot ManipulationMehul Arora, Dhruv Agarwal, Ambuj Agrawal
Vision-Language-Action (VLA) models achieve high success rates on the LIBERO benchmark but remain sensitive to paraphrasing. Existing VLAs ranging from 0.6B to 7.5B parameters experience a 22-52 point performance drop on LIBERO-Para while SmolVLA exhibits an even larger decline, from 83\% to 8\%. Because most VLAs jointly learn language understanding and action generation, it is difficult to determine whether a failure originates from misunderstanding an instruction or executing it incorrectly. To address these challenges, we introduce CoBoVLA, or CodeBookVLA, a 250M-parameter VLA that decouples language understanding, planning, and control through learned discrete interpretable codes. A text encoder maps the instructions to shared step codes, which a visual planner then converts into action code sequences executed by a language-independent actor. This design improves robustness to instruction paraphrasing while allowing failures to be traced to individual stages. It also enables efficient real-time control on resource-constrained hardware by encoding instructions only once and using compact discrete representations to keep each component lightweight. CoBoVLA achieves 91.4\% success on LIBERO-Goal, compared with 82.8\% for SmolVLA, under a matched protocol, with 17 ms per control step latency. On LIBERO-Para, it achieves 81.1\% across 4,092 unseen paraphrases and 78.5\% on matched episodes, compared with 36.5\% for SmolVLA LoRA fine-tuned on paraphrases. Without two-step demonstrations, CoBoVLA completes 48.8\% of 160 free-form two-step tasks. Finally, success reaches 91\% when instructions are correctly encoded but falls to 27.2\% otherwise, identifying language understanding as the main source of remaining failures. Together, these results demonstrate that decoupling language understanding and control can improve instruction robustness while making VLA failures easier to diagnose.
vision-language-actionvlamanipulationliberobenchmark - arxiv:2611.22388 · cs.ROHidden Variables Behind Real-Robot Data Quality: Lessons from Hundreds of Thousands of HoursManiformer Real-World Manipulation Team, Che Tsang Calvin, Xingyu Chen, Bowen Jiang +38
As real-robot data collection scales, collection-system design shapes both operator efficiency and the behaviors learned by robot policies. Workspace limitations, control-interface friction, and latency can introduce pauses, detours, and corrective motions into otherwise successful demonstrations. We investigate these effects across embodiment, teleoperation control, and data consistency, combining a kinematic study of 25 dual-arm robots, controlled teleoperation experiments, and real-robot evaluation of fine-tuned $π_{0.5}$ policies. The embodiment study shows that positional reachability overstates orientation-aware task coverage and that modest shoulder motion can substantially expand the effective workspace. Teleoperation experiments show that parallelizing body-part control reduces task completion time by 43-53%, while compliant control improves success and reduces peak contact forces on constrained and contact-rich tasks. Policy-learning experiments show that demonstrations must capture the visual information underlying action choices: active-search demonstrations improve arm-selection accuracy from chance (50%) to 100%. Calibration experiments reveal that mixed-calibration training data remain usable in our tested configurations when an inference-aligned subset is retained, whereas inference-time miscalibration causes all 30 trials to fail. Policies also retain approximately 90% of the extra path length in detour-contaminated demonstrations, showing how inefficient operator motions persist in learned behavior. Together, these findings connect collection-system design to demonstration quality and downstream policy execution. They motivate evaluating effective workspace, control efficiency, observation completeness, calibration alignment, and motion quality before scaling collection, and addressing system-induced compensations at their source.
teleoperation - arxiv:2611.22382 · cs.AIRICoT: Recurrent Interleaved Chain-of-Thought for Page Stream SegmentationPrashant Khatri, Bhavana Malagavalli, Saiful Haq, Niyati Chhaya
Large language models (LLMs) have renewed interest in Page Stream Segmentation (PSS), splitting a multi-page file into its constituent documents. Yet published LLM methods still frame the task as boundary classification over one or a few neighboring pages, and commercial document-splitting APIs do not disclose their methods. We reimagine PSS for LLMs as a sequential process over the whole stream and propose RICoT (Recurrent Interleaved Chain-of-Thought), a prompting framework that works through the pages as a reader would: within a sliding window it reasons about each page and commits its boundary decision before moving on, and across windows it carries forward a compact record of what it has read, keeping per-call cost bounded. We also introduce FinancePSS-205, a benchmark of 205 real-world, de-identified financial page streams, to be released publicly. On FinancePSS-205, RICoT matches or outperforms pairwise and windowed prompting with the same backbone for each of three open-weight LLMs; on a TABME test subset, RICoT with GLM-5 reaches document F1 0.847 and boundary F1 0.915, above published zero-shot results. A 2B-parameter student, trained on a larger teacher's decisions with and without the page-level reasoning, performs comparably in both modes and comes within 0.05 document F1 of the strongest prompted model on FinancePSS-205.
benchmark - arxiv:2611.22380 · cs.CVLeveraging Segmentation Representations for Efficient Classification: A Self-Configuring Framework Based on nnU-NetRenjun Cai, Antonio Javier Gallego, Félix Fuentes-Hurtado
We propose a self-configuring framework that extends the nnU-Net workflow to both segmentation and classification in 3D medical imaging. Building on nnU-Net's self-configuring method, the framework reuses representations learned by a segmentation model for image-level classification. Its encoder features and predicted segmentation probabilities are used by a lightweight classifier. Two information-sharing strategies are investigated. In the frozen-backbone branch, features are augmented with region-of-interest information, predicted foreground volume, and segmentation confidence. In the adaptation branch, the final encoder stage is partially fine-tuned for classification, with an auxiliary segmentation loss to preserve segmentation-related information. Additional five-fold experiments showed that more complex configurations did not consistently improve classification performance. On the public validation leaderboard, our selected framework achieved an Overall score of 0.8056. In the official FLARE 2026 Task 2 final evaluation, our framework ranked first overall, achieving the highest DSC, NSD, and AUROC across all four testing tasks, including two hidden testing tasks. Local inference measurements also remained below the challenge limits of $60 \mathrm{s}$ per case and $4 \mathrm{GB}$ GPU memory. These results demonstrate the practical value of reusing segmentation representations for classification while maintaining a simple and efficient framework.
leaderboard - arxiv:2611.22374 · cs.ROAutoE2E: Don't speak - Language free end-to-end autonomy A Real-Time Driving Model for L2++ ADAS and L4 RobotaxisMuhammad Zain Khawaja, Zaynab El Mawas, Ryota Yamada, Gabriela Cordova +11
AutoE2E is an open-source end-to-end driving model developed by the Autoware Foundation Robotaxi Working Group. It introduces a temporal bird's-eye-view (BEV) architecture that combines multi-camera perception, HD-map information and route conditioning to predict future vehicle motion. The evaluated model uses a pretrained camera-BEV backbone, temporal feature fusion and a lightweight recurrent trajectory planner. It contains approximately 79.9 million parameters, of which approximately 1.45 million are trainable in the evaluated configuration. The v1.0 release provides model checkpoints, training and evaluation code, and DataModelConsole, a dashboard for visualizing and inspecting model experiments. On the KITScenes validation benchmark, the fine-tuned checkpoint achieves an average displacement error (ADE) of 0.745 m and a final displacement error (FDE) of 2.020 m at a three-second prediction horizon. The paper also reports preliminary real-vehicle research conducted with the Technical University of Munich's EDGAR platform, alongside ongoing collaboration with the Karlsruhe Institute of Technology. AutoE2E v1.0 is an early open research release. Its quantitative evaluation is open-loop, and the model has not yet demonstrated validated closed-loop autonomous driving. The release is intended to provide the autonomous-driving community with an inspectable baseline for further research, experimentation and collaborative development.
benchmark - arxiv:2611.22368 · cs.ROSeeing What Still Holds: How Waiting Control Changes Cooperative RecoveryKangjie Huang, Yucheng Xin, Xinchun Wang, Dong Liu +2
Changing the holding reference during a fixed-duration wait can reverse the ranking of visual recovery methods. We introduce AuditEmbodied, a paired execution and trajectory evaluation framework that independently varies holding references and waiting duration while keeping recovery executors frozen. A 960-execution factorial on 20 initialization seeds compares CAD-feedback and landmark-based recovery from identical post-wait physical states within each holding and waiting condition. Protected completion requires successful recovery under the declared execution constraints. At a 1 s wait with registered wrist references, their protected completion counts change from 6/40 versus 19/40 under registered holding to 34/40 versus 22/40 under position-only holding. Trajectory analysis localizes the main separation to grasp formation: small observation-time target residuals precede large tracking errors. Joint diagnostics and controlled interventions characterize persistent tracking constraints. On fresh seeds, gradual position reanchoring improves CAD recovery's long-wait completion by 42.5 percentage points and reduces mean complete recovery time by 6.06 s relative to registered holding. Additional evaluations with pretrained policies, handover, and a mass-matched geometry characterize holding-reference sensitivity across executors and physical regimes. These findings establish holding reference as an explicit evaluation condition for comparing visual recovery performance.
embodiedgraspevaluation framework - arxiv:2611.22367 · cs.ROFailureGen: Scalable Robot Failure Data Generation with World ModelsAkul Santhosh, Chaitra Sathe, Aastha Jhunjhunwala
Robot failures are difficult to collect at scale, limiting the range of failure scenarios represented in robot learning datasets. We present FailureGen, a method for generating synthetic videos with temporal event annotations by reusing action trajectories from recorded failures. The method comprises four stages: (1) fine-tuning an action-conditioned Cosmos world model on successful and failed real executions; (2) annotating source recordings to identify failure-associated donor segments and compatible successful hosts; (3) transferring donor motion into host contexts, with optional spatial offsets, to generate continuous video rollouts; and (4) annotating and filtering the generated samples according to observed events and visual quality. We study two implementations with different tasks and robot configurations, with visual failure detection serving as one downstream application. In a preliminary comparison, a detector trained on real and generated data achieves an average precision of 0.51, compared with 0.41 using real data alone, and detects 33 of 36 failure intervals, compared with 28 of 36. By reusing observed failures in additional contexts, FailureGen expands the data available for studying robot failures and developing downstream learning methods.
world modelaction-conditioned - arxiv:2611.22362 · cs.AIHealthCLBench:Evaluating Context Learning in Health Conversations through Controlled Case ClustersTing Yang, Zijing Hong, Chengjie Wang, Peng Wei +6
Complex health conversations require answers grounded in individual information and currently valid evidence. We evaluate health context learning: acquiring, integrating, and applying relevant context during inference with fixed model parameters to answer health questions and respond appropriately to changes in facts, evidence, or task goals. To our knowledge, Chinese-language medical benchmarks have yet to systematically combine explicit contextual difficulties, controlled case variations, and member-specific medical criteria for open-ended answers. We introduce HealthCLBench, comprising 328 controlled clusters, 1,103 cases, and 4,931 rubric criteria across medical consultation and everyday health management. Members share backgrounds while evidence, patient states, or task goals change; each has its own medical scoring requirements. A 70-person medical team annotates and reviews the simulated cases and their rubrics. Human review of 700 full-context responses from the original evaluation yields 90.5% pooled agreement with the original automated judgments at the rubric-item level. We report full-context and query-only results for seven general-purpose models and Ant Afu. Across the eight systems, full- query differences range from 11.83 to 31.76 case points and from 10.87 to 38.97 geometric cluster points. GPT-6-Astra leads both full-context metrics at 75.03 and 60.42, followed by Claude-Opus-5-5 and Ant Afu. Cross-condition differences describe score changes under the reported evaluation settings. Response comparisons reveal failures in checking historical information, expressing uncertainty when warranted, and ensuring recommendations respect required conditions. The resource supports model comparison and targeted follow-up evaluation.
benchmark - arxiv:2611.22360 · cs.CVCan Vision-Language Models Truly Understand Images? A Benchmark Beyond Commonsense BiasLing Lin, Heng Su, Xiangzeng Liu, Yang Bai +2
Multimodal large language models (MLLMs) have achieved impressive performance on a wide range of vision-language tasks. However, it remains unclear whether their predictions are grounded in the visual content itself or are largely driven by statistical co-occurrence patterns and commonsense correlations learned from large-scale image-text corpora. Existing vision-language benchmarks predominantly consist of samples that conform to human commonsense, allowing models to achieve strong performance by exploiting dataset priors rather than genuinely understanding image content. To address this limitation, we present ContraSense, a systematically constructed benchmark for evaluating faithful vision-language understanding. ContraSense consists of carefully curated counter-intuitive image-text pairs that deliberately violate common expectations, substantially reducing the effectiveness of statistical biases and requiring models to rely on visual evidence for reasoning. We conduct comprehensive evaluations on a diverse set of representative open- and closed-source MLLMs. The results reveal a substantial performance gap between current MLLMs and humans: even the strongest closed-source model achieves only about 60\% accuracy, compared with 96.04\% for human participants. These findings demonstrate that current MLLMs still struggle with faithful visual understanding and highlight the need for more robust multimodal reasoning grounded in image content. The ContraSense benchmark is publicly available at https://huggingface.co/datasets/LingLin00/ContraSense.
benchmark - arxiv:2611.22356 · cs.ROSR-VLA: Stall-Reduced VLA Execution via Predicted-Input Lookahead and Selective KV Cache RepairTao Zhang, Yi Zhong, Jinhao Ruan, Yipeng Gao +3
Vision-language-action (VLA) policies incur control stalls when inference blocks the next action chunk. Asynchronous early inference can hide latency, but its results may be inconsistent with the arriving actual inputs. Unconditional reuse risks acting on mismatched inputs, whereas full recomputation discards useful precomputation. We propose SR-VLA, combining predicted-input lookahead with selective KV cache repair. During current chunk execution, a lightweight action-conditioned predictor generates next-boundary visual observations and proprioceptive state. These predictions, together with the task instruction, drive the frozen VLA to precompute a speculative KV cache and action chunk. After real inputs are available to VLA, predicted--real differences and task conditions determine whether to accept the action chunk or trigger selective KV repair. On LIBERO, SR-VLA achieves $2.67\times$ and $1.27\times$ query speedups over the respective synchronous $π_{0.5}$ and OpenVLA-OFT baselines, while maintaining competitive average task success rates. Under target-motion perturbation, success remains baseline-comparable, with $1.52\times$--$1.93\times$ query speedup.
vision-language-actionvlaopenvlaliberoaction-conditioned - arxiv:2611.22355 · cs.CVSandBAR: A Band-Asymmetric Architecture for Underwater Image Restoration with Removable Frequency ScaffoldingYanzi Liu, Haoran Hou, Jiachi Tong, Shen Chen +1
Light scattering, absorption and turbidity degrade underwater imagery through complementary frequency tendencies: large-scale, channel-dependent appearance distortions and local high-frequency detail loss. This band asymmetry motivates different forms of restoration for global colour and local structure. Our primary contribution is SandBAR, a novel unified band-asymmetric architecture for underwater image restoration. Rather than treating frequency processing as an isolated module, SandBAR organises the restoration pipeline around band asymmetry at the block, encoder and objective levels. Within this architecture, we introduce a novel band-asymmetric sandwich block with a reserved post-attention repair slot and the Split-band Parallel Asymmetric Repair Kernel (SPARK), a novel module that applies channel-aware large-support correction to the low band, which predominantly carries large-scale appearance distortions, and lightweight local refinement to the detail-bearing high band, thereby restoring global colour while preserving fine structure. To deliver the global cast to the depth at which the network represents global structure, a learnable low-frequency guidance path injects a low-pass view of the input into the deep encoder, and a three-scale spatial-frequency objective supervises the same axis throughout the decoder. Building on the reserved post-attention slot, we further introduce Post-attention Scaffold Training (PaST), a novel removable nonlinear scaffolding scheme that trains the permanent SandBAR graph under band-specific capacity and runs it at inference without that capacity. SandBAR reaches 29.89 / 28.92 / 22.77 dB on EUVP / LSUI / UIEB and establishes a new three-domain mean state of the art on the benchmark family, exceeding the strongest published result by approximately 1.0 dB. PaST adds no parameters or FLOPs at inference and preserves matched latency.
benchmark - arxiv:2611.22353 · cs.CLNONSENSE: Non-Semantic Strings Expose Semanticity Gaps in Dense RetrievalAnkit Gupta
Dense retrieval is central to modern search, question answering, and retrieval-augmented generation, where high similarity is often treated as evidence of semantic relevance. We examine how far this assumption can be stressed. NONSENSE is a black-box sparse-edit probe that constructs visibly non-semantic strings while forbidding query-term copying, optimizing only the dense similarity between a query and the string. Across five BEIR-style datasets and three popular single-vector retrievers, these strings frequently outscore the retriever's own best judged query-relevant passage (QRP), and often rank first in the full corpus. The result indicates a semanticity gap in the evaluated dense retrievers: nearest-neighbor status can be separated from passage-like meaning. These findings highlight the importance of evaluating dense retrievers not only by whether they retrieve relevant documents, but also by how reliably their similarity scores reflect semantic evidence.
retrieval-augmented - arxiv:2611.22343 · cs.CVPURE: A Diagnostic Benchmark for Multi-Granularity Multimodal Identity MatchingHengyu You, Xintong Wang, Jin Cao, Liang Ding
Identity matching asks whether two records denote the same identity, yet the answer depends on the target granularity. Two records may match at the base level but not at the variant level: they share an underlying design while differing in attributes such as color or size. Final matching accuracy does not separately characterize the capabilities involved. To study this problem, we use e-commerce as a challenging real-world domain. We introduce PURE, a diagnostic benchmark for multi-granularity multimodal identity matching covering four capability dimensions: Perception, Understanding, Reasoning, and Evidence-grounded Judgment. The benchmark comprises 1,457 multiple-choice questions across 14 tasks, derived from real-world matching pairs and verified by two domain experts. Across ten recent MLLMs, overall scores range from 47.1% to 86.0%. Understanding receives the highest mean score and Judgment the lowest, with weaknesses in visual difference localization and variant-level rule judgment. Modality ablations show that models benefit from multimodal evidence but can incorrectly identify conflicts between matching records. In direct matching, most models recognize matching pairs more reliably than non-matching pairs at the base level, whereas the reverse holds at the variant level. PURE therefore reveals what aggregate matching accuracy hides: not only whether models judge a pair correctly, but where they struggle to interpret evidence and apply matching criteria.
benchmark - arxiv:2611.22342 · cs.CLBad Apples in the Kitchen: How LLM Teams Detect, Correct, and Calibrate Against Their Misaligned MembersDongseok Heo, Junghwan Kim, Bongwon Suh
What protects an LLM team when one teammate quietly goes wrong? Most multi-agent LLM work assumes cooperative partners and inspects chat between turns before acting. Real-time embodied cooperation does not allow this pause: chat runs alongside action, so a teammate whose words stay cooperative while behavior diverges can cause damage faster than the team can name what is happening. We study this case by injecting three theory-grounded misaligned agents into a custom four-player Overcooked environment, each stressing a different joint-action layer (speech-action mismatch, role invasion, and aggregate-progress disruption), and decompose robustness into three sequential gates of detection, correction, and calibration through a 3x4x2 design (n=720 episodes). Nearly all trolls are eventually named, but recovery follows only when attribution and situational diagnosis appear in the same utterance; direct source-attribution prompting improves the hardest case only by degrading the troll-free baseline by 37.5%. Robustness, then, depends less on detection alone than on whether team chat combines attribution with diagnosis in a single utterance, suggesting that sensitivity to misalignment should be targeted rather than uniformly raised.
embodiedmulti-agent - arxiv:2611.22338 · cs.AIReliable Evaluation of AI Agents for Investment Research: A Framework for Performance and Validity AssessmentDongnuan Tian, Yan Yu, Yangfan Liu, Wenjun Tang +5
We consider the problem of evaluating AI agents undertaking investment-research tasks, where agents interact with external information sources and computational tools to produce financial analyses and research reports. An important challenge is that the quality of an agent's final output does not necessarily reflect the quality of its execution process, while the reliability of the resulting assessment also depends on the validity of the evaluation procedure. Furthermore, investment-research tasks differ considerably in their requirements for numerical accuracy, information retrieval, financial analysis and supporting evidence, making it difficult to assess agent performance using a uniform set of criteria. We propose a domain-oriented evaluation framework that distinguishes outcome quality from execution quality and accommodates these differences through task-specific criteria, deterministic validators and independent large language model judges. We also introduce an evaluation-validity mechanism that examines the integrity of evaluation inputs, reference information, supporting evidence and grading outputs, with expert review used to resolve identified deficiencies and significant grading disagreements. Numerical results show that the proposed framework can distinguish differences in agent performance across evaluation dimensions, while identifying potential reliability issues in the evaluation process. The associated implementation and evaluation materials will be made available on GitHub.
agentai agentevaluation framework - arxiv:2611.22334 · cs.LGBQ-LoRA: Binary-Quantized Low-Rank Adapters as Implicit Regularizers for Parameter-Efficient Fine-TuningJuyoung Park, Heejun Lee
Parameter-efficient fine-tuning of quantized large language models is dominated by QLoRA, which combines 4-bit weight quantization with FP16 low-rank adapters. We ask whether the FP16 adapter precision is necessary. Replacing it with a single binary bit per adapter parameter, combined with a learned per-layer scale $s$ and a Scaled Straight-Through Estimator, reduces adapter storage by 13-16x. We introduce two variants -- NormScale, which binarizes under a norm-preserving scale with a clipped STE, and LearnThresh, which makes the binarization threshold learnable -- unified under a framework we call BQ-LoRA. Our main theoretical contribution is a tight characterization of the BQ-LoRA hypothesis class: it has bounded entrywise norm, a typical-set spectral norm bound that is $Θ(r)$ times sharper than the worst case, a full-rank, near-isotropic update, and consequently a Rademacher complexity that yields an explicit $\tilde{O}(s_{\max}α\sqrt{d_{out}/(rn)})$ generalization bound. FP16 LoRA has no comparable a priori norm or isotropy control. On Qwen2.5 (1.5B-72B), BQ-LoRA exceeds QLoRA's benchmark averages at every scale by 0.8-3.5 points; on LLaMA-2 (7B/70B) it remains competitive at the same storage reduction but does not uniformly lead the strongest FP16-adapter baselines (LoftQ at 7B, QLoRA at 70B). On Qwen2.5-7B the binary adapter preserves the 4-bit base WikiText-2 PPL within 0.06 (9.86 vs. 9.80) while QLoRA worsens it by 0.59. The reduction applies to inference-time adapter storage and serving; training memory matches QLoRA and step time rises by under 4%. We translate the storage saving into a deployment study, BQ-SGMV, by extending Punica's SGMV kernel with a binary inner-product primitive: the same A100-80GB holds 15.6x more adapters in HBM, and our research prototype eliminates cache misses under a skewed synthetic workload.
memorybenchmark - arxiv:2611.22328 · cs.LGPVTSizing 2.0: A Fast and Reliable Analog Circuit Sizing Operator for Agentic EDA WorkflowsZichen Kong, Minghuan Song, Xiyuan Tang, Ziming Wei +2
Analog circuit design automation requires both coordinated tool execution and efficient numerical search. Large language models (LLMs) offer new ways to organize tool execution, but finding one sizing that satisfies specifications across process, voltage, and temperature (PVT) corners remains costly. PVTSizing 2.0 addresses these needs through a human-agent workflow built around an independently usable numerical optimizer. An agent carries the designer's approved task through execution and result review. The optimizer reduces unnecessary training through loss-guided updates and uses critic-guided local refinement to explore additional candidates while accounting for cross-corner conflicts. An optional LLM adviser supplies parameter relationships to focus numerical search on circuit-informed directions, while the optimizer selects adjustment amplitudes and verifies their effects through simulation. Across four circuit benchmarks, the standalone optimizer completes all 40 runs and achieves a 7.3x speedup in overall mean runtime over the original PVTSizing, with a comparable mean sample count. In a separate study, the best tested knowledge-injection configuration further reduces mean runtime by 39.9% and simulation samples by 35.4% relative to the standalone optimizer.
agentagenticbenchmark - arxiv:2611.22323 · cs.AIAgent Trajectories as Evidence for Software Quality AssuranceJonathan Ullrich, Julien Siebert, Michael Kläs, Adam Trendowicz +1
Autonomous software agents increasingly plan, implement, and test software with limited human oversight. Their decisions may introduce implicit assumptions that final code and passing tests do not adequately expose. We propose analyzing observable agent trajectories alongside software and agent specifications and AI generated artifacts as evidence for targeted quality assurance. The aim is to prioritize consequential decisions for scrutiny, investigate their possible contribution to software problems, and inform improvements to specifications and development practices. Industrial experience motivates the idea; available execution records and preliminary analysis methods make it plausible. However, trajectories alone do not establish that decisions are justified, outcomes acceptable, or corrective changes effective. We ask how this evidence can support review, diagnosis, and continuous quality improvement, and when simpler QA approaches suffice. Perspectives from requirements engineering, verification and validation, agent systems, causal inference, empirical software engineering, and human factors could help develop and evaluate the idea.
agentagent system - arxiv:2611.22321 · cs.ROBeyond Episodic Recall: Experience-to-Knowledge Evolution for Lifelong Embodied NavigationWeitao An, Chenghao Xu, Xu Yang, Cheng Deng
Lifelong embodied navigation requires agents to accumulate experience through continuous interaction and reuse it for subsequent goal searches. Existing memory-enhanced methods typically retain historical observations or construct structured scene representations. However, storing observations or high-dimensional semantic embeddings incurs growing storage costs. Meanwhile, structured representations support retrieval and semantic reasoning but do not explicitly evolve accumulated experience into environment-specific knowledge. To address these limitations, we propose Experience-Evolving Memory (EvoNav), a novel two-timescale memory framework. It organizes semantic observations, spatial relations, and task-feedback cues into sparse experience atoms. During online operation, experience atoms continually consolidate repeated observations and form dynamic links based on transitions. At the slower timescale, EvoNav distills accumulated category co-occurrences and feedback signals into environment-specific structural knowledge without relying on predefined object--room relations. For each new goal, EvoNav combines direct goal evidence from experience atoms with learned structural knowledge to guide goal-directed search on the current traversability map. Experiments on single-goal, sequential multi-floor, and multimodal navigation demonstrate effective experience reuse and improved path efficiency with compact persistent memory, together with successful transfer to a real robot.
embodiedmemorypersistent memory - arxiv:2611.22314 · cs.CVInertial Video World ModelsYuan Gao, Fan Xu, Yu Guo, Yuxu Lu +3
The physical world has inertia: it evolves from what already exists rather than being recreated at every instant. However, many existing video world models lack an explicit mechanism for inertial state evolution, leaving the continuity of motion and structure implicit in video prediction. During autoregressive rollouts, repeated reconstruction can accumulate errors and destabilize the predicted dynamics. To address this limitation, we introduce InertialWorld, an Inertial Video World Model that makes inertia a structural prior for autoregressive prediction: existing scene evidence is carried across time rather than reconstructed at every step. Central to this formulation is Inertial Memory, which turns memory from passive predictive context into an evolving video state with two complementary views: a dense Eulerian grid preserves scene-aligned evidence, while persistent Lagrangian carriers bind local content to learned canonical coordinates. At each step, the model transports the carrier content according to action-conditioned motion, predicts the complete next video state from the transported carriers and dense grid, and uses a learned gate to combine transported and newly predicted content into the memory for the next step. Extensive experiments across multiple domains, including robot manipulation, 3D game simulation, and open-world navigation, demonstrate that InertialWorld achieves competitive autoregressive prediction performance against strong video world model baselines. Codes link: https://github.com/YuanGao-YG/InertialWorld
manipulationworld modelaction-conditionedmemory - arxiv:2611.22310 · cs.AIEviDARL: Evidence-Centric Dual-Agent Reasoning via Alternating Reinforcement Learning for Multi-Table Question AnsweringDunhan Li, Leilei Gan, Lirong Gao, Wenyan Fan +3
Despite advances in LLM-based table reasoning, most studies focus on single-table settings. Multi-table QA requires multi-hop reasoning over complex table relationships and evidence dispersed across long contexts, often without verifiable intermediate traces. To address these challenges, we introduce EviDA, an evidence-centric dual-agent framework comprising an Evidence Construction Agent and a Reasoning and Answering Agent. The Evidence Construction Agent generates and executes table-manipulation programs over complete tables to produce question-specific evidence and verifiable intermediate traces, simplifying subsequent reasoning. The Reasoning and Answering Agent then performs multi-hop reasoning over this evidence to derive final answers. We further introduce EviDARL, which combines agent-specific fine-grained rewards with parameter-efficient alternating reinforcement learning to improve collaboration between the agents. By optimizing each agent in turn while holding the other fixed, EviDARL enables bidirectional adaptation between evidence construction and answer reasoning. Experiments on MMQA and TQA-Bench demonstrate that EviDARL substantially outperforms competitive open-model and multi-table reasoning baselines under constrained model and training budgets.
manipulationlong contextagentagent framework - arxiv:2611.22309 · cs.AIEfficient Wide-Beam Generative Retrieval with Model-Native Trie Decoding and Shared-Context AttentionVedant Jhaveri, Arashpreet Singh Mor, Satyam Kumar, Alex Tsun +3
Generative retrieval (GR) predicts catalog entries as short semantic identifiers (SIDs). Trie-constrained beam search keeps these identifiers valid, but wide beams expose two serving costs: vocabulary-wide score tensors that are mostly discarded, and repeated attention over the same prompt. We present Trie-Head, a model-native output interface that stores the identifier trie as validated GPU tensors. Compared with a Python custom logit processor (CLP), it reduces CPU--GPU coordination and memory traffic by avoiding Python constraint callbacks and vocabulary-wide candidate buffers. It scores compact codebook candidates and streams the full-vocabulary normalizer, preserving the original model's valid-token scores in real arithmetic without retraining. Shared-context attention complements this output path by reusing prompt key/value blocks across beams while preserving their private decode tails. We implement both mechanisms in SGLang and evaluate candidate sourcing for LinkedIn Hiring Assistant and qualification retrieval for the Qualification Retrieval Platform. The combined system improves closed-loop throughput by up to 5.57x over the CLP baseline and up to 2.63x over BF16 FlashRec at a common beam width of 200 on the Qualification Retrieval Platform workload.
memory - arxiv:2611.22308 · cs.LGVRLLM: Turning LLMs into Powerful Tabular ClassifiersYingtong Dou, Shubham Jain, Junpeng Wang, Hadi Abdullah +4
Large Language Models (LLMs) have shown strong performance on text, but whether pretrained LLMs are effective for large-scale tabular classification remains unclear, as many previous works demonstrate that LLMs' tabular data classification performance falls behind tree-based models or neural architectures designed for tabular data. At Visa Research, we managed to fine-tune a pretrained Llama 3.2 1B LLM on billions of rows of labeled tabular data and make it outperform a wide array of tabular classifiers and even our competitive production deep learning model. For ease of reference, we refer to the model as the Visa Research LLM (VRLLM). In this paper, we share how we achieved this exciting result, covering data preparation, training recipe, ablation studies of design choices and deployment validation. We also benchmark frontier LLMs on the same task and analyze their prediction behavior across model families and generations. Our interpretability experiments suggest that VRLLM learns feature-specific predictive associations, while component ablations indicate that the earliest two FFN layers contribute to the emergence of linearly decodable task information. Our paper provides an engineering recipe and a focused empirical case study on adapting compact LLMs to industrial tabular prediction tasks.
benchmark - arxiv:2611.22307 · cs.AILearning to Edit IFC Building Models: Verifiable Checkers as the Training Signal for Dependency-Aware Editing AgentsJiaxin Zhang, Yao Luo, Yuze Dan, Yunqin Li +1
Editing a building information model in the Industry Foundation Classes (IFC) schema requires propagating changes to dependent elements while preserving unrelated entities. Existing language-model agents for IFC editing use prompted models without learning from edit outcomes. Target-file comparisons do not specify which alternative edits are allowed or which unrelated entities must remain unchanged. We present IfcGym, a CodeAct environment in which a language model edits IFC through executable code and four checkers verify validity relative to the input model, intent, dependency and preservation against a task specification. Their conjunction is strict success. IfcEdit distils verified teacher episodes into a Qwen3.5-9B student through supervised fine-tuning (SFT), then applies Group Relative Policy Optimization (GRPO) with strict success as the reward. On IfcEditBench, 1,990 editing tasks on 13 held-out real projects, the final student reaches 86.78% strict success, from 79.15% after SFT. The paired confidence interval for the difference from its teacher includes zero; the same recipe on a 27B model reaches 91.76% and exceeds the teacher. On the public benchmarks BIM-Edit and BIBIMBAP the final student scores 55.19 and 54.50. On IfcEditBench, the students outperform all prompted models except the teacher under matched tool access; both evaluated frameworks also score lower. Rewarding all four checkers yields 4.83 points more strict success than rewarding intent and validity alone.
benchmark - arxiv:2611.22302 · cs.ROWhere Do Human-to-Robot Transfer Gains Come From?Jiachen Li, Shihao Li, Soovadeep Bakshi, Longteng Fan +3
Human data can support robot policy pretraining, but gains attributed to a data recipe may also reflect subset size, training variability, or downstream adaptation. In an HPT implementation on a 20-task MetaWorld suite with three training blocks, a constrained random 5,000-recording subset matches most of the task-informed subset's gain over the 40,000-recording pool at 20,000 updates (3.69 of 4.29 percentage points), leaving equal-size selection differences of 0.60 points at 5,000 recordings and 0.45 points at 1,250 recordings, with wide paired intervals. Across 12 comparisons, task orderings repeat across seeds for several contrasts, but a fixed predictor built from the other two seeds trails the mean or difficulty reference in at least one held-out fold for every comparison. Neither mechanism probe shows its predicted positive interaction: task-slot mismatch yields an interaction of $-0.0347$ points after adaptation, and in a six-task observation experiment the observation-by-pretraining interaction changes sign between seeds ($-18.29$ and $+1.04$ points) while full-state pretraining gains are positive in both ($+6.37$ and $+2.31$ points). This retrospective study shows how equal-size and capacity controls, seed-level reporting, and targeted interventions change the interpretation of apparent transfer gains.
robot policy - arxiv:2611.22301 · cs.AIMCTS Is Counting the Wrong Thing: Learning Exploration Counts for LLM ReasoningXuan Zhang, Zhijian Zhou, Zicheng Hu, Yan Feng +3
Monte Carlo tree search (MCTS) has powered landmark successes in reinforcement learning, yet its gains in LLM reasoning can remain modest despite substantial search effort. We identify an overlooked bottleneck: what MCTS counts. Standard UCT assigns independent visit counts to textual branches, allowing different formulations of the same reasoning strategy to repeatedly receive exploration bonuses. Our theoretical analysis shows that this count fragmentation can impair exploration efficiency, as quantified by increased cumulative regret. To address this problem, we introduce Learned Pseudo-Count MCTS (LPC-MCTS), which changes only the exploration statistic in UCT. A lightweight Coin-Flipping Network head on the policy's hidden representations estimates representation coverage, enabling related branches to share exploration evidence. Across 5 mathematical-reasoning and 7 search benchmarks, LPC-MCTS consistently outperforms strong MCTS baselines. Holding each LPC-MCTS-trained policy fixed, replacing standard MCTS counts at inference improves average mean@16 accuracy across 5 mathematical benchmarks by 3.79 percentage points on Qwen2.5-7B and 2.71 points on Qwen3-4B. Applying LPC-MCTS during both training and inference yields gains of 4.96 and 4.37 points over standard MCTS, respectively. Together, our theory and experiments show that changing what MCTS counts can substantially improve its exploration efficiency and reasoning performance.
benchmark - arxiv:2611.22299 · cs.AIHarnesses That Improve Themselves: A Unifying Framework for Self-Improving Agent HarnessesDeep Shah, Nehal Kathrotia
Whether a foundation model succeeds as an autonomous agent depends as much on its runtime software as on its weights. This runtime, the agent harness, manages context and memory, steers multi-turn control, executes tool actions in sandboxes, and verifies work before stopping. Because models differ in how they use tools and how they fail, a static harness tuned for one model or task often stalls, wastes tokens, or regresses on another. This bottleneck has driven a shift from prompt tuning to self-improving agent harnesses: systems that inspect their own execution traces, separate runtime bugs from model limits, and rewrite their own code. We present a unifying architectural and design framework for self-improving agent harnesses. The framework pairs four editable runtime components (context and state, agent control, action interfaces, and verification and recovery) with a four-stage loop that diagnoses failures, proposes code patches, validates candidates, and deploys updates. Using this framework, we show how optimizers trace errors across long trajectories and discover reusable runtime mechanisms such as pre-stop verifiers, loop breakers, and persistent memory. We further show why full-harness rewrites overfit or break across modules, and how modular updates with validation gates generalize across models while bounding token cost. To keep self-modification safe, we establish practical defenses against recurring failure modes including prompt bloat, benchmark gaming, and evaluator tampering. Finally, experiments on SWE-bench Lite validate this framework, showing that gated modular evolution outperforms prompt reflection and full-harness rewrites across models and repositories at a fraction of the token cost.
persistent memoryagentautonomous agentself-improvingbenchmarkevaluator - arxiv:2611.22297 · cs.ROFIRE-Drive: Feasibility-Informed Risk Evaluation for End-to-End Diffusion PlanningQifeng Xie, Kai-Yuan Feng, Wenkang Wan, Yue Wu +2
Diffusion-based end-to-end autonomous driving models have shown strong potential in generating multimodal planning trajectories. However, their performance still depends heavily on trajectory initialization, mode evaluation, and supervision quality. Weakly constrained noisy initialization may introduce infeasible proposals and increase the burden of denoising. Moreover, selecting the optimal mode remains challenging when candidate trajectories are geometrically similar but differ in safety, interaction risk, and progress. To address these issues, we propose FIRE-Drive, a feasibility-informed risk evaluation framework for end-to-end autonomous driving. FIRE-Drive improves diffusion-based planning through dynamic feasible anchor initialization, candidate-specific bird's-eye-view risk evidence, and compact high-quality candidate supervision with lightweight refinement. This design aligns the initial support, risk-aware scoring, and multimodal supervision of trajectory modes under a unified planning objective. It enables the diffusion decoder to focus on refining plausible trajectories while the scorer distinguishes modes using local risk evidence. Experiments on NAVSIM v1 show that FIRE-Drive achieves a Predictive Driver Model Score of 91.5 with a ResNet34 backbone. Additional closed-loop evaluation on 220 Bench2Drive routes yields a Driving Score of 71.43 and a Success Rate of 42.73%. Component ablations and a scorer comparison under matched candidate inputs support the contributions of feasible initialization, auxiliary supervision, and risk-informed selection. Our code is available at https://github.com/Latte00/FIRE-drive.
evaluation framework - arxiv:2611.22296 · cs.AISpecSearch: Towards Hardware-Aware Efficient Speculative DecodingSelin Yildirim, Ali Zeyrek, Deming Chen
Speculative decoding (SD) accelerates large language model (LLM) inference, but its optimal configuration depends jointly on workload characteristics, GPU hardware, and available memory. For example, on the RTX~5090, the state-of-the-art speculative drafter DFlash achieves $3.50\times$ the throughput of autoregressive (AR) decoding at low concurrency, but only $0.76\times$ at high concurrency with short outputs. We present SpecSearch, a hardware-aware search engine that jointly optimizes decoding strategy, draft length, and precision across workload--GPU pairs. Its analytical cost model captures memory and compute constraints to identify feasible configurations and narrow the search space for throughput optimization. Guided by this model, calibration measurements are collected from multiple GPUs. Using the measurements, a Jev-based predictor learns to select high-throughput decoding configurations on held-out GPUs without requiring deployment-GPU calibration. The predictor is parameter-efficiently trained on calibration data, while adapting a frozen Qwen3-4B using LoRA adapters and a linear correction layer. Across NVIDIA RTX~5090, H100, L40 and AMD MI300X GPUs, SpecSearch with full calibration achieves an average of $2.01\times$ the throughput of AR and $1.36\times$ the throughput of SD, both served by vLLM. The proposed Jev-based predictor closely matches this search performance with $1.30\times$ throughput improvement over SD with vLLM, while eliminating the need for held-out GPU calibration. Additionally, it retains $97.36\%$ of the throughput achieved by SpecSearch that utilizes full calibration.
memory - arxiv:2611.22295 · cs.LGSimilar Is Not Useful: Selecting In-Context Demonstrations by Closed-Loop UtilityJiachen Li, Zhigen Zhao, Shihao Li, Dongmei Chen +2
Retrieval-augmented vision-language-action (VLA) policies fill their context with the demonstrations whose frames look most like the current observation under frozen DINOv2 embeddings. We ask whether visual similarity selects the demonstrations that make the policy succeed. For a frozen retrieval-augmented VLA (RICL on $π_0$-FAST, LIBERO), we make the demonstration pool swappable at run time and score each demonstration's utility by regressing the closed-loop success of 60 fixed-size pools on their membership. On initial states disjoint from those used for scoring, pools of the five highest-utility demonstrations succeed on 66.4% of 760 paired episodes across 19 tasks, pools of the five most similar on 55.7%, and random pools on 55.5%. Utility beats similarity by 10.8 points ($p=3.2\times10^{-6}$; positive on 15 of 18 untied tasks), including on 12 tasks the policy was never primed on, and similarity-selected pools do no better than random ($+0.1$ points). Within a task, a pool's visual similarity to the test scenes tracks its success at Spearman 0.10; across tasks, similarity separates tasks whose scenes differ and falls to chance when two tasks share a scene. The gain comes from on-task closed-loop evidence: best-of-ten direct search over pools recovers two-thirds of it with 100 rollouts. Against the shipped 20-demonstration pool, utility-curated pools of five lead by 7.0 points over 560 paired episodes from two evaluation runs ($p=0.006$), and utility-ordered retrieval inside the full pool leads by 4.5 points ($p=0.055$).
vision-language-actionvlaliberoretrieval-augmented - arxiv:2611.22294 · cs.ROPerception into Imagination: Executable Memory for World ModelsXianyao Li, Ruitong Tian, Rui Min, Fang Xu +1
Physical agents must use what they previously saw or touched after visibility or contact is lost. Their actions change the implications of these remembered facts. We introduce executable perceptual memory (EPM), a world-model framework that advances explicit perceptual state alongside neural latent dynamics during imagination. Its evolving read-out informs reward prediction, value estimation and policy learning. We align reward inputs with the transitions that produced them and supervise the reward head on open-loop rollouts of recorded actions. On Memory Maze $13\times13$, EPM reaches 30.37 return at 1M interactions, compared with 23.37 when the same evolving read-out informs only policy and value on a common perceptual front end: a gain of 7.00 [3.80, 10.21]. The complete method also improves control on $9\times9$ and in MiniWorld. Probes of remembered, out-of-view arrivals show stronger reward anticipation in both visual implementations. Concentration, tactile recovery and real tactile frames demonstrate complementary roles of remembered facts in imagined-policy training, motion-dependent decisions and sensory-content retention. EPM connects persistent perception to the predicted consequences and learning signals of a world model for Physical AI.
tactileworld modellatent dynamicsmemory - arxiv:2611.22291 · cs.ROWhen the World Is the Test Suite: Agentic Intelligence for the Physical WorldKaixin Li, Meng Luo, Yanlin Li, Yuchen Tian +8
Coding agents built to repair software now write, test, and repair programs for robots, laboratories, and industrial controllers. Unlike in software engineering, no test suite comes with the code. Whether a program works must be established by running it on a robot, in a physics simulator, or through the policies it helps to train, and such evidence is costly to obtain and can itself be wrong. As these agents take on more physical tasks, how their programs are judged matters as much as how they are written. This paper examines this emerging area, agentic intelligence for the physical world, by treating the world as the program's test suite. We organize the discussion around three questions: how agents write programs for the physical world, how these programs are tested, and how agents learn from the results. We first trace how language models for robots and software engineering agents came together, ask which software testing practices carry over to physical tests, and describe each program by where it sits relative to the actuator and by what judges it. For writing, we cover programs that act as robot policies and programs that support policy learning by producing rewards, tasks, and training worlds. For testing, we examine what physical trials cost, what traces, learned judges, simulators, and formal checks can establish in their place, and what safety requires beyond task success. For learning, we discuss how tested programs are retained as experience, in skill libraries and in model weights. Finally, we turn to benchmarks and evaluation practice and outline open challenges. Writing programs is becoming cheap, while trustworthy evidence about what they do remains costly. Progress will therefore depend on a physical counterpart of continuous integration: verification cheap enough to run often, reliable enough to trust under search, and maintained alongside the experience it validates.
agenticbenchmark - arxiv:2611.22288 · cs.LGConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agentsArunkumar V, Gangadharan G. R., S. Senthilkumar
Large-language-model agents now close the industrial predictive-maintenance loop, reading sensor windows, retrieving procedures and dispatching work orders. Two attack channels meet there: injected text in a retrieved document can hijack the decision, and a bounded sensor perturbation can corrupt the remaining-useful-life estimate behind it. Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible. ConfGuard is a reference monitor with a physics-grounded policy. Its conformal monitor tests each prediction, anchored to the asset's admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle. Calibrating its threshold over whole asset trajectories lets the operator set the expected share of a clean asset's windows the running monitor escalates. A physics-policy validator checks the agent's maintenance-urgency decision against the decision a deterministic policy derives from the same interval. On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker's reach on healthy assets from 0.529 to 0.046 and closes suppression of failing assets (0.151 to 0), escalating 0.027 of clean windows against a 0.05 target, while the fixed physical range it replaces leaves both attacks open. Behind a Qwen3-14B planner the composed monitors hold urgency-class injection to 0.00 and the universal sensor attack to 0.12, against 0.29 and 0.54 for a defended prompt. An adaptive attacker defeats a Mahalanobis detector on 0.504 of windows and adversarial training on 0.485. We report an attack class the projection misses.
agent - arxiv:2611.22281 · cs.AISeasonRainBench: Global Seasonal PrecipitationQiuyan Sun, Mengxuan Chen, Haoyuan Liang, Guowen Li +4
Accurate seasonal precipitation forecasts help societies prepare for drought, flooding, and crop failure months in advance. While data-driven methods have made rapid progress in weather and subseasonal forecasting, operational seasonal precipitation prediction still relies on numerical ensemble systems. These systems are computationally expensive and exhibit biases and drift that increase with lead time, motivating more efficient data-driven alternatives. Yet this transition faces three linked challenges: comparing data-driven forecasts with operational ones requires (1) a global benchmark that does not exist, scoring them requires (2) metrics that capture operational priorities beyond anomaly correlation (ACC) and RMSE, and training them must (3) rely on a single, short observational record. To address the first two challenges, we introduce \BenchName, the first such benchmark, covering monthly precipitation forecasts 1--5 months ahead. It pairs global $1^\circ$ satellite precipitation with oceanic and atmospheric predictors, includes seven operational centers, and provides 310,944 simulated months from 52 climate models for training. Beyond standard scores, we introduce two operationally motivated metrics for anomaly strength and monthly rain-belt position. To address the third challenge, we develop \MODEL, which learns from simulations what the record lacks and transfers it to Earth through three modules: SimForced aligns the climate background, SimMemory transfers a history encoder pretrained on simulations, and SimAnalog supplements Earth labels with simulated analog outcomes. Extensive experiments show that \MODEL{} improves ACC, CRPS, anomaly strength, and rain-belt position by 12.0\%, 3.6\%, 2.9\%, and 0.3\% over the strongest operational or data-driven model, opening a path from simulated worlds to Earth system prediction.
benchmark - arxiv:2611.22280 · cs.LGXM-Sparse: High-Throughput Sparse-Attention LLM Serving over Scale-up Memory PoolsYijun Sun, Xudong Liao, Zhuobin Huang, Kai Chen +2
Sparse attention has been increasingly adopted by production LLMs for long-context inference, yet serving systems must still retain the full KV history, making GPU HBM capacity a key limit on serving concurrency. Existing systems offload the KV cache to host memory, but suffer from the limited bandwidth of PCIe and the limited capacity of host memory. Emerging scale-up memory pools (e.g., X-Mem) relieve both limits and open new opportunities for sparse-attention offloading. In this paper, we first conduct an in-depth analysis and reveal that, even with scale-up memory pools, decode throughput remains bounded by HBM capacity, as hot KV buffers and the fully resident indexer key (K-index) cache occupy substantial HBM. Moreover, the two caches exhibit distinct access patterns. Based on these insights, we present XM-Sparse, the first sparse-attention offloading system over scale-up memory expansion. The core idea of XM-Sparse is to offload both caches aggressively while matching each with an I/O mechanism tailored to its access pattern. Specifically, for dynamically selected KV entries, XM-Sparse fuses TMA-based retrieval with attention in a tile-level pipeline to hide on-demand retrieval latency. For predictable K-index accesses, it prefetches pages via SM-efficient GPU-initiated DMA and coordinates prefetching with latency-critical operations to avoid interfering with decoding. We have implemented XM-Sparse atop SGLang, and our experiments on DeepSeek-V3.2 and GLM-5.3 with an emulated scale-up memory pool show that XM-Sparse delivers up to 4.98$\times$ and 4.52$\times$ higher decode throughput than PCIe-based ECHO and HiSparse, respectively, and up to 1.34$\times$ over a naive integration on the same platform, across 16K-1M contexts.
memorylong-context - arxiv:2611.22278 · cs.AIDACE: Diversity-Driven Adversarial Co-Evolution for Robust LLM Safety AlignmentPeng Yu, Xiaoyu Wen, Zhida He, Ziyuan Zhou +5
Ensuring robust safety alignment of large language models (LLMs) is increasingly difficult as adversarial attacks evolve and outpace alignment pipelines built on pre-collected data. Existing co-evolutionary frameworks let the defender chase a moving attacker, yet their training dynamics expose two compounding pathologies: \emph{attack strategy collapse}, where the attacker overfits to a narrow set of high-reward rewrites, and \emph{defense adversarial forgetting}, where the defender loses competence against earlier attacks as the attack distribution drifts. Existing remedies are partial: attacker-side diversity rewards score textual novelty rather than strategy novelty, and defender-side replay relies on discrete judge scores that ignore estimation uncertainty, response stochasticity, and defender non-stationarity. We introduce \textbf{\method}, a diversity-driven adversarial co-evolution framework that targets both pathologies jointly. On the attacker side, \method couples an explicit two-dimensional strategy space (risk category $\times$ attack style) with a \emph{normalized marginal coverage gain} reward, providing a bounded, non-vanishing exploration signal at the strategy layer. On the defender side, \method maintains a unified \emph{Bayesian adversarial replay pool} whose Beta--Bernoulli threat posteriors are refreshed by time decay and sampled via Thompson sampling, anchoring defender training to an evolving threat landscape. Across standard safety, automated-attacker, and general-capability benchmarks, \method improves robustness to out-of-distribution attacks while preserving general capabilities, and yields broader attacker strategy coverage.
benchmark - arxiv:2611.22275 · cs.ROHeave-Ho: Decentralized Long-Horizon Multi-Humanoid CollaborationTing-Hsuan Chen, Siheng Zhao, Han Jiang, Qi Pan +6
Collaboration promises to extend humanoid capabilities beyond what individual robots can achieve. However, realizing this potential requires coordinated whole-body interactions, coherent long-horizon planning, and precise synchronization across robots. We introduce Heave-Ho, a decentralized framework for long-horizon collaborative multi-humanoid loco-manipulation. At the low level, we develop a multi-agent post-training framework that adapts a humanoid foundation tracking model to physically coupled interactions between humanoids, enabling coordinated decentralized control from local observations alone. At the high level, we train a vision-language planner and design a speech-based leader-follower coordination mechanism to synchronize decentralized execution without network communication. Trained with two identical humanoids, our system scales to teams of up to ten robots and generalizes to heterogeneous teams without retraining. To benchmark these capabilities, we introduce HH-Bench, comprising 12 tasks involving teams of 2 to 10 humanoids, including heterogeneous teams. Systematic evaluations show that Heave-Ho consistently outperforms baseline methods, while real-world experiments demonstrate robust decentralized collaboration across challenging tasks.
manipulationhumanoidmulti-agentpost-trainingbenchmark - arxiv:2611.22273 · cs.CVSARVIA: Fast-Adapting the Strong SAR Foundation Models That Are Already in Your MLLMsYuxuan Li, Zichen Tian, Qianru Sun
Synthetic aperture radar (SAR) foundation models provide transferable representations for diverse perception tasks in all-weather Earth observation. However, their development often relies on costly large-scale SAR pretraining, raising the question of whether such extensive domain-specific pretraining is necessary. We find that modern multimodal large language models (MLLMs) already provide visual encoders that can outperform dedicated SAR foundation models as downstream initializations. This motivates a new paradigm: reuse general visual capabilities and devote domain training to brief specialization. We therefore introduce SARVIA, an adaptive post-pretraining method that specializes modern MLLMs using self-supervised visual questions with computable answers. These questions are constructed from unlabeled observations through general image tasks that capture spatial relations and SAR-specific tasks that exploit native measurements and SAR--optical correspondence. A shared answer-prediction objective jointly trains low-rank adapters across visual and language components, with loss-driven task sampling. We then export the adapted visual encoder as a standalone backbone. With 600 adaptation steps and 2 GPU-hours, SARVIA outperforms the compared foundation models on all six SAR detection and segmentation benchmarks. Compared with dedicated SAR pretraining, SARVIA reduces domain-training costs by orders of magnitude while improving visual transfer across architectures and scales. This new paradigm shifts the emphasis from repeated, costly pretraining to a reusable specialization process that can benefit from advances in open MLLMs. Dataset and code: https://github.com/zcablii/SARVIA.
benchmark - arxiv:2611.22262 · cs.LGOfficePrism: Benchmarking Agents on Single-Domain Operations Across Enterprise Business FunctionsHanfeng Liu, Zhenyu Luo, Dongchao Zhang, Zexi Liu +4
Broad workplace scores can mix competence within one business function with the extra cost of coordinating across functions and systems, making failures hard to localize. OfficePrism evaluates this prerequisite through 420 single-domain enterprise tasks spanning seven business functions under shared, benchmark-defined L1--L3 task-structure levels. Each task runs in a fresh environment combining a shared synthetic organization with one function-specific domain and yped state access. Our pipeline uses grounded records, business rules, and tool schemas to jointly construct a Chinese-language ticket, a hidden executable reference witness, and an expected response, followed by automated filtering and human curation. The same task set is evaluated through full-schema and search-and-dispatch packages. Across 124 model--harness--package configurations, the per-case joint natural-language-and-database pass rate ranges from 22.6% to 76.4%, with a median of 67.9%. L1 is highest in all 124 configurations, and L2 exceeds L3 in 122. Among 50 configurations sharing the same harness and model labels across both packages, search scores higher in 30, ties in one, and scores lower in 19.
benchmark - arxiv:2611.22258 · cs.ROALERT: Extending Humanoid Motion Trackers with a General and Lightweight Human Avoidance PipelineZhaoyang Li, Chiyu Zhang, Chengxin Xu, Peng Yang +3
Whole-body motion tracking has enabled humanoid dancing, athletic motions, and teleoperation, providing a foundation for capability demonstrations and manipulation data collection. However, existing tracking policies lack awareness of nearby humans and proactive avoidance capabilities, while motion substitution and local constraint-based corrections may compromise reference motion integrity and tracking continuity. We propose ALERT, a general lightweight pipeline for extending motion tracking to shared human-robot environments. Through lightweight adaptation, it integrates human perception and proactive avoidance into pretrained tracking policies, balancing avoidance demands with reference motion tracking while preserving existing motion capabilities. Simulation and real-world experiments across tracking backbones demonstrate reliable collision-free execution through localized human avoidance, preserving whole-body coordination and enabling rapid recovery of reference tracking, while retaining the ability of pretrained policies to track unseen motions.
manipulationhumanoidteleoperation - arxiv:2611.22252 · cs.AIA Visual Information Extraction System Using Multimodal Large Language Models for Scanned Cognitive and Behavioral AssessmentsEnshuo Hsu, Linh Nguyen, Sunyang Fu, Heath Goodrum +4
Objective. Paper-based assessments are common in patient care, but handwriting and scanning artifacts make these forms difficult to process at scale. This study proposes a modular, multimodal LLM-based system with human oversight for extracting fields and scores. We conducted a qualitative analysis of the use of LLMs for hand-drawing analysis. Materials and Methods. The system processes cognitive assessments (i.e., MMSE, MoCA, Mini-Cog) and the NICHQ Vanderbilt Assessment using two modules: 1) a form classifier that assigns a scanned page to a form type, followed by a corresponding 2) score extractor that extracts fields and scores into a schema. A human-in-the-loop design utilizing dual-model validation engages human annotators when two LLMs disagree, limiting manual review to challenging pages only. Results. End-to-end accuracy for cognitive assessments was 88%, with most form types exceeding 90%. For the NICHQ Vanderbilt, overall accuracy reached 97.5%, with all form types above 96%. The human-in-the-loop design further improved cognitive assessment accuracy to 97.5% while reducing manual work by 63.8%. In the Mini-Cog Clock Drawing task, qualitative analysis showed that 62.3% of all extraction errors stemmed from ambiguous drawing, in which cases the LLM provided reasonable readings. Discussion. Compared to current manual data extraction, our LLM-based system reduces processing time by 10 to 20 times and costs by over 12 times. The modular design allows seamless expansion to additional form types. Conclusion. Our system provides an accurate, scalable, and cost-efficient solution for structuring data trapped in handwritten paper forms for clinical decision support.
human-in-the-loop - arxiv:2611.22250 · cs.LGPCGRL for Complex Platformers: Synthesizing Playable and Diverse Levels via Inflated Envelopes and Forward CurriculaDi Wu
Procedural Content Generation via Reinforcement Learning (PCGRL) has emerged as a promising paradigm for automated game development. However, existing PCGRL frameworks still suffer from severe exploration bottlenecks and scaling limitations when faced with long-sequence generation tasks governed by strict rigid-body rules and non-linear kinematic boundaries (e.g., platformer constraints). Consequently, traditional approaches frequently fail to maintain reliable level solvability on large-scale canvases. To address these challenges, we present a physics-aware, curriculum-guided PCGRL framework. First, we construct an "Inflated Volume Envelope" mechanism, which provides a physics-constrained representation that incorporates continuous kinematic feasibility into the discrete state space of the MDP. Second, to alleviate the sparse reward challenge induced by extremely long decision sequences, we introduce intermediate state-transition feedback and a multi-stage forward extension curriculum learning framework, enabling informative value estimation and stable credit assignment over long-horizon trajectories. Finally, to resolve the issue where deterministic policies after convergence generate highly monotonous map paths with a lack of diversity, we activate high-entropy stochastic inference during the synthesis phase, regularized by a post-hoc parallel rejection filtering mechanism. Experimental results demonstrate that our framework significantly enhances structural diversity while maintaining high physical playability across multi-scale environments. Comparative evaluations across multi-scale canvases demonstrate that our framework maintains superior convergence stability and dimensional scalability, reliably synthesizing solvable levels where traditional PCGRL baselines degrade.
curriculum learning - arxiv:2611.22248 · cs.LGOAK-Mind: Zero-Training Mental Health Change Inference from Passive Sensing via Knowledge-Grounded Multi-Agent LLMsYishu Ji, Zhihan Jiang, Will Ke Wang, Yuang Fan +2
Wearable and smartphone sensors capture behavioral signals associated with mental health, yet inferring mental health outcomes from these signals has traditionally relied on supervised models requiring labeled data and offering limited insight into their predictions. Large language models (LLMs) offer new opportunities to reason over behavioral evidence, but existing LLM-based approaches to passive sensing have largely continued to rely on task-specific fine-tuning or sensor-specific pretraining. In this paper, we introduce OAK-Mind, a zerO-training, multi-Agent, Knowledge-grounded LLM pipeline for inferring mental health changes from passive sensing data. The pipeline organizes sensing data into behavioral dimensions including sleep, physical activity, location mobility, and smartphone use, and orchestrates specialized agents grounded in mental health literature to interpret longitudinal behavioral changes. It synthesizes their assessments and applies questionnaire-based symptom rating to infer changes relative to an individual's baseline assessment with human-readable rationale grounded in observed behavioral patterns and domain knowledge. Across datasets and prediction tasks, while without task-specific training, OAK-Mind achieves comparable or better performance than the most predictive supervised baselines trained on additional labeled samples. It also demonstrate consistent improvement upon direct prompting across four tested LLM backbones. Ablation analyses support the benefits of combining behavioral interpretation with symptom rating, distributing analysis across domain-specific agents, and incorporating domain knowledge. Our findings highlight the potential of knowledge-grounded, multi-agent LLMs for training-free inference of mental health changes from passive sensing data.
multi-agent - arxiv:2611.22245 · cs.AIMulti-Agent Memory Inconsistency: Where Factual Conflicts Disappear in LLM AgentsHejia Zhang, Zhongming Yu, Yimin Ding, Wei-Ting Chen +8
Multi-agent delegation lets LLM systems distribute evidence collection across workers with separate contexts. This creates a consistency challenge: facts retrieved by workers may be missing from the evidence available to the lead for a joint conclusion. We introduce a benchmark and evidence-flow profiler for studying multi-agent memory inconsistency through controlled factual conflicts in Harvey LAB document-review tasks. Pooled across three models with comparable generation spending, requiring delegation increases retrieval of both conflicting claims from 50.2% to 68.7% relative to single-agent execution, yet conflict detection falls from 16.2% to 13.2%. Within this setting, workers report both claims for only 35.2% of retrieved pairs, and pooled matched comparisons show higher detection than single-agent execution when one worker reads both claims but lower detection when the claims are split across workers. Worker-to-lead handoffs are a central target for improving conflict detection in multi-agent LLMs.
memoryagent memoryllm agentmulti-agentbenchmark - arxiv:2611.22244 · cs.CVPhyVGGT: Feed-Forward Articulated Object Reconstruction from Monocular VideosQi Shao, Zhihao Cao, Shuhao Zhai, Anh Nguyen +3
Reconstructing an articulated object with its 3D movable parts, joint parameters and unseen articulation states from an uncalibrated monocular video is essential for robot manipulation and interactive digital twins. Existing per-scene optimisation methods require minutes to hours and external depth or camera poses, whereas feed-forward 3D foundation models reconstruct dense point maps in one pass without segmenting movable parts or inferring joints. Even on these monocular point maps, regressing or fitting continuous joints directly fails on casual videos, since cross-frame attention blurs moving part boundaries in deep layers, underestimates motion along the viewing ray and leaves continuous axes ill-conditioned under small opening angles. To address these challenges, we propose PhyVGGT, a single-pass framework on a frozen 3D foundation model that anchors moving-part matching and closed-form joint estimation to the undistorted geometry of the stationary base. First, our Part and Motion Decoder fuses intermediate per-frame tokens with global tokens and references part queries to a static-base query to segment 3D parts, align point maps on the static base and predict learned Bingham priors. Second, our Kinematic Inference Layer fits an orthogonal Manhattan frame to static surface normals, selects the joint axis in closed form under a Bingham posterior combining pairwise motion scatter with the learned priors, and solves a linear system for the hinge pivot to drive unseen states. Trained solely on synthetic data, PhyVGGT achieves state-of-the-art monocular performance across five benchmarks, outperforming FAMOS and monocular ArtiPoint, surpassing posed RGB-D ArtiPoint in median error, and running in 6.8 s per 48-frame video against hours of per-scene optimisation in Articulat3D. Project page: https://phyvggt.github.io.
manipulationbenchmark - arxiv:2611.22241 · cs.CLWhere Does Hindsight Enter Clinical Agent Benchmarks? An Audit of Simulators and Case ConstructionMeilu Zhu, Xiaohan Li, Axiu Mao
Benchmarks for LLM diagnostic agents are built from retrospective records: the case an agent reads is rewritten from discharge summaries, and unrecorded test results are generated by a simulator that often sees the final diagnosis. HindAudit separates three hindsight channels (the diagnosis given to the simulator, the retrospective context it conditions on, and the agent-facing case text) and measures each with judges trained only on real data. On 27,489 emergency admissions from MIMIC-IV, five registered tests found no leakage into simulated laboratory results: providing the diagnosis to five simulators had a negligible effect, and prompted gatekeepers diluted the diagnostic signal, although exploratory results suggest they can raise agent scores. The largest effect lay in the case text. Against a presentation-time reconstruction from triage records, benchmark-style summaries raised the zero-shot accuracy of four LLMs from 25.4% to 39.7% and discharge histories to 45.3%, and raised six interactive agents' scores while barely changing their ranking. The exploratory gap was reproduced in a registered replication on held-out patients. Masking diagnosis names left it unchanged. Because MIMIC lacks a contemporaneous bedside history, the gap is an upper bound on hindsight. We report all registered outcomes and recommend construction practices.
agentagent benchmarkbenchmark - arxiv:2611.22240 · cs.CVCross3DGS: Unified Vulkan-Based 3D Gaussian Splatting Training from Workstations to Mobile DevicesHiroyuki Sato, Yusuke Takimoto
Scenes captured with handheld and wearable cameras are still shipped to a CUDA workstation for 3D Gaussian Splatting (3DGS) training. Training on the device that captured them, or on its companion phone or laptop, requires more than translating kernels: GPUs and drivers differ in parallel features, memory limits, and command submission, and these differences decide whether a training schedule executes and returns the correct update. We present Cross3DGS, a cross-vendor, cross-platform Vulkan trainer that keeps one optimizer and chooses its execution schedule per device in two steps: rules on device limits, workload size, and observed driver failures exclude the schedules known not to run correctly, and a fixed preference order selects one of the remaining feasible schedules. The same core completes standard-quality training (30k updates, one million Gaussians) on every configuration we measured, spanning NVIDIA, Intel, Arm, Qualcomm, AMD, and Apple GPUs and including Android devices and an iPhone, whereas the existing portable trainers VkSplat and Brush do not run on some of the mobile GPUs. Averaged over 11 scenes on a workstation, Cross3DGS has higher quality than VkSplat with no significant difference in training time, and Brush reports about three times the training time at comparable quality. The final quality of Cross3DGS is nearly identical across devices. A mobile application turns a handheld capture into a viewable model entirely on the device.
memory - arxiv:2611.22239 · cs.AISolver-Free and Cost-Aware Online Scheduling of AI Data Center TasksXin Qin, Dawei Qiu, Fei Teng
AI data centers receive computing tasks continuously, whereas solver-based optimization takes time to run and electricity is priced per settlement period. This paper bridges this timing gap through a solver-free online optimization rule inspired by dual decomposition. The rule assigns each task a start period upon arrival to optimize energy cost subject to operational constraints, shifting flexible tasks toward low-price periods while keeping high-priority tasks on schedule. Simulations on the A100 GPU trace (3,456 GPUs, 1,343 tasks) show that the proposed rule keeps GPU and CPU use within capacity while scheduling tasks at low cost. Compared with a multi-timescale optimization benchmark, it achieves a lower cost and, because it needs no solver, much less computation time.
benchmark - arxiv:2611.22235 · cs.CVCoSphere-SLAM: Multi-Agent Panoramic Gaussian Splatting SLAM with Spherical Loop ClosureQi Shao, Zhihao Cao, Shuhao Zhai, Taoyu Wu +7
Dense collaborative mapping is essential for search and rescue, building inspection, and warehouse automation, where panoramic cameras capture the entire surroundings in a single frame. Unlike pinhole agents that share views only at similar headings, panoramic agents observe overlapping scenes at any 3D orientation, yet searching full rotations also produces mutually consistent false loop closures across distinct symmetric rooms. Moreover, continuous monocular scale drift splits rigidly merged maps at revisited surfaces, while depth noise at short baselines distorts loop translation directions. To address these challenges, we propose CoSphere-SLAM, the first multi-agent Gaussian splatting SLAM framework for panoramic cameras, whose spherical loop closure aligns spherical features over $\mathrm{SO}(3)$ via fast Fourier correlation across arbitrary orientations and rejects symmetric false closures through global $\mathrm{Sim}(3)$ consensus. Building on these closures, directional global optimization solves a per-keyframe $\mathrm{Sim}(3)$ pose graph with log-scale regularization to correct scale drift while separating optical-flow translation directions from depth-based magnitudes. Finally, omnidirectional Gaussian fusion transforms each Gaussian with its keyframe similarity and renders Gaussians at their maximum ray response to jointly refine poses and maps. Experiments on indoor and cross-session multi-agent panoramic benchmarks demonstrate that CoSphere-SLAM achieves state-of-the-art trajectory accuracy and rendering quality among single-agent and multi-agent methods. Project website: https://cosphere-slam.github.io.
multi-agentbenchmark - arxiv:2611.22231 · cs.ROFrom Temperature-Limited to Battery-Limited Legged LocomotionRaphaël Ségas, Balázs Opra, Léo Ghafari, Thomas Stewart
On legged robots, sustained locomotion often ends in thermal shutdown rather than battery depletion: the heat of walking concentrates in a few joint motors, which overheat long before the battery is empty. We train thermally aware locomotion policies by reinforcement learning without ever observing temperature, and demonstrate this on a Unitree Go2 quadruped. We first identify a coupled 13-node lumped RC thermal network, fitted on two runs of a single robot and calibrated against the measured electrical power, that predicts all twelve motor temperatures with a mean RMSE below 1 $^\circ$C across different robot instances, gaits, payloads, and speeds. Under a sustained gait, the temperatures the motors will settle at follow directly from the gait-averaged squared torques and joint speeds. Our core contribution is a reward that drives down the hottest predicted asymptotes: thermal time constants exceed the credit-assignment horizon of reinforcement learning by orders of magnitude, but predicted asymptotes respond to torque and joint speed within a single gait cycle, so the policy learns to lower and spread heat with no temperature input. On the robot, the policy turns sustained locomotion from temperature-limited to battery-limited: with a 5 kg payload at 1 m/s, a learned baseline reaches thermal shutdown after 19 minutes (1.1 km), whereas our policy runs the battery down after 174 minutes (10.0 km) with its hottest motor 14 $^\circ$C below the shutdown threshold in a 26 $^\circ$C room; on an unloaded factory circuit with stairs and ramps, the baseline overheats after 25 laps (6.2 km), while our policy completes 47 (11.2 km) on one battery with 22 $^\circ$C of margin.
quadrupedlegged locomotion - arxiv:2611.22228 · cs.LGWhat Does Weight-Space Programming Add to Diffusion Editing? Program Construction versus Score-Guided RefinementZitian Gao, Jingyi Xu, Ziqi Han, Chengkai Lei +2
Weight-space programming (WSP) lets a language model construct executable weight updates for another network. We study diffusion concept editing through an interface in which programs define training contexts, preservation constraints and update directions, while a shared numerical fitter selects their coefficients. We compare generated programs with published editors, a shared-panel hand-written baseline and budget-matched search over conventional edits on 24 TIMED test edits for Stable Diffusion 1.4. The preregistered rewrite pipeline shows no advantage over the strong baseline. In an exploratory comparison, one-shot Astra programs improve fresh-scene VLM joint success, which requires subject presence together with whole-image target matching, by +10.0 points over the shared-panel baseline, while CLIP relative match shows no detectable change. These programs organise adjustable directions around protected text-state subspaces. Probing and rewriting the Astra programs then raises CLIP by +6.2 points while joint success changes by -9.7. On a fixed 144-image sample, five additional evaluators give point estimates in the same directions under whole-image target matching; checks of subject-local change give less decisive differences. Together, these results identify useful behaviour in complete generated programs and show that score-guided revision need not preserve it: program construction and revision should be evaluated as distinct stages of a WSP pipeline.
evaluator - arxiv:2611.22227 · cs.LGSOTA or Luck? The Winner's Curse in LLM LeaderboardsSikun Xu, Raphael Thomadsen, Dennis J. Zhang, Heng Zhang
Public LLM leaderboards guide model evaluation and adoption, yet the reported performance of top-ranked models can be systematically overstated. This leaderboard winner's curse arises because ranking noisy benchmark scores rewards sampling luck alongside genuine ability. The same selection effect can distort scores throughout the leaderboard, a phenomenon we call the rank-induced bias. We develop an $m$-out-of-$N$ bootstrap to correct for this bias. It assumes a fixed model pool with independent, identically distributed bounded tasks while allowing arbitrary within-task correlation among models. Under exact population tiers separated by positive gaps, we prove that the bootstrap reduces $Θ(N^{-1/2})$ ranking bias to expected residual bias $o(N^{-1/2})$. Across 14 benchmarks from 5 benchmark families, the rank-1 adjustments range from approximately 0.09 to 3.86 percentage points. In our SWE-bench Verified snapshot, the tied leaders receive an adjustment of 0.90 percentage points. Such overstatement is large enough to make the apparent winner statistically fragile. Thus, we use the same bootstrap to further quantify the intervals of plausible rank positions for each model and group models into tiers. We establish sufficient conditions for consistent recovery of population tiers. On SWE-bench Verified, this procedure places 13 of 173 models in the leading tier. We combine the score adjustment and rank intervals in the Rank Uncertainty and Leaderboard Evaluation Report (RULER). We validate its finite-sample performance using sample splitting and calibrated simulations, and advocate publishing RULER alongside raw rankings so readers can assess both reported performance and the precision of the ordering. RULER also provides guidance for benchmark design that accounts for selection effects and ranking uncertainty.
benchmarkleaderboard - arxiv:2611.22224 · cs.ROTRIDENT: Efficient Triple-Task Learning of Dehazing, Depth, and Uncertainty Estimation for Underwater 3-D Robot Visual PerceptionGeonmo Yang, Younggun Cho
Underwater visual systems often suffer from blurry textures and low color contrast due to the inevitable light propagation. These issues can significantly degrade the perception of stable robotic operations. In this paper, we introduce a novel learning-based sensing system that tackles the multidimensional vision tasks in underwater; concretely, we deal with image enhancement, depth estimation, and uncertainty for 3-D visual systems. Also, we propose a TRIDENT model in a fast and lightweight manner; TRIDENT consists of three parallelized decoders and one backbone structure for efficient feature sharing. In addition, it is designed to be trained to express complex parameterization. In experimental evaluation on several standard datasets, we demonstrate that TRIDENT significantly outperforms other existing methods on image enhancement and depth estimation. Despite performing three tasks, our model has better efficiency than the others for both memory size and inference time. Finally, our joint learning approach demonstrates robustness in feature matching and seamlessly extends from 2-D to 3-D vision tasks. Supplements are available at https://sites.google.com/view/underwater-trident/home
memory - arxiv:2611.22222 · cs.CVExplicitIR: Explicit Spatial Policy Fields for All-in-One Image RestorationDezhi Zhang, Shangrong Yang, Yudong Liang, Wenjian Wang
All-in-one image restoration aims to recover clean images blindly from inputs with various degradation types and levels using a single model. However, the adaptation policy in most existing methods remains implicitly encoded in feature representations, leaving the restoration decision neither explicitly represented nor independently controllable. Since real-world degradations vary spatially, applying uniform updates across the entire image leads to over-smoothing of clean areas and under-correction of localized degradations. To address these limitations, we present ExplicitIR, which represents the restoration policy as two explicit per-pixel fields: Budget, which determines where and how strongly to update, and Style, which selects between narrow and wide context windows. Specifically, ExplicitIR incorporates a Spatial Policy Estimator (SPE) to predict these fields, explicitly representing the restoration policy and capturing spatially-varying degradations, which are then applied by Policy-Adaptive Decoder Blocks (PADB) through adaptive fusion of narrow- and wide-window responses, enabling independent control and precise local modulation. This design enables spatially-adaptive restoration in a single forward pass, without requiring task labels, prompts, or test-time degradation estimators. Extensive experiments on three-task and five-task benchmarks demonstrate that ExplicitIR achieves performance competitive with the strongest all-in-one baselines.
benchmark - arxiv:2611.22220 · eess.SYSafely Mitigating Grid Imbalance: A Hybrid Distributional Wide-Margin Recourse ApproachShoaib Hussain, Mostafa Farrokhabadi, Hamidreza Zareipour
This paper proposes the Multi-Agent Distributional Wide-Margin Policy Gradient (MADWMPG) framework: a dual-policy hybrid imitation and safe reinforcement learning (IL-SRL) method designed for safely mitigating grid imbalance in multiphase active distribution networks with distributed soft open points (SOPs). The framework comprises a wide-margin constraint barrier recourse (WCBR) module that seeks the maximal separation of constraint-compliant and unsafe actions and learns to proactively suppress exploratory violations. The framework addresses operational volatility using a novel adaptation of noncrossing quantile regression (NQR) distributional critics that track the complete distribution of policy returns. The set of behavior and recourse policies integrates an instantaneous physics-based and differentiable action projection mechanism that generates SOP constraint-compliant actions without relying on active learning. Policy safety is made responsive using an enhancement of conventional experience replay, and the safety performance of the framework is supported by probabilistic bounds derived in the paper. The proposed framework was trained and tested on a modified multiphase IEEE 123-bus test feeder. Results demonstrate its superiority to conventional reinforcement learning and SRL methods and show that it is highly competitive against the nonlinear AC optimal power flow.
multi-agent - arxiv:2611.22219 · cs.AIFine-Grained Japanese Clinical Named Entity Recognition: A Controlled Comparison of BERT and Open-Weight Large Language ModelsFeng Han, Ayako Yagahara, Kenji Hirata, Atsushi Yoshida +1
Background: Generative large language models (LLMs) offer an alternative to encoder-based clinical named entity recognition (NER), but their performance on fine-grained Japanese extraction tasks and sensitivity to generation settings remain uncertain. Methods: We compared two fine-tuned BERT-CRF baselines with nine open-weight LLMs on iCorpus, which defines 113 entity categories. The evaluation used 433 sentences from 36 test documents; 1,672 sentences were used for training. LLMs were evaluated using zero-shot prompting with and without JSON Schema constraints and three-seed QLoRA fine-tuning. Fine-tuned models were assessed under different output-token limits, including 900 and larger model-specific limits of 2,600 or 4,096 tokens. Performance was measured using sentence-level label-surface matching with relaxed and exact-surface F1. Results: The highest zero-shot relaxed F1 was 0.337, compared with 0.836-0.841 for the BERT baselines. At the larger output limits, all eight fine-tuned LLMs with at least 13B total parameters exceeded both BERT baselines (relaxed F1, 0.882-0.914; exact-surface F1, 0.840-0.880). Across three-seed means, increasing the output limit improved relaxed F1 by 0.052-0.085, alongside higher recall and fewer length-limited outputs. JSON Schema constraints changed relaxed F1 by -0.014 to +0.008 among seven models outside the two reasoning-oriented configurations, while medical-general model pairs showed no consistent advantage for medical specialization. Conclusions: On this historical Japanese case-report benchmark, task-specific fine-tuning combined with adequate generation capacity was associated with stronger LLM extraction performance. Output-token limits materially affected comparisons with BERT, whereas enforcing output structure alone did not overcome low zero-shot accuracy.
benchmark - arxiv:2611.22215 · cs.LGCommunicate the Change: Efficient Expert Parallelism via Activation CompressionFengyuan Liu, Jay Gala, Siva Reddy, Dzmitry Bahdanau
Scaling has driven the success of large language models (LLM), with sparse mixture-of-experts (MoE) architectures enabling greater parameter capacity without a proportional increase in computation. At large scales, MoE models rely on expert parallelism to distribute experts across devices. Each MoE layer requires two synchronous collective communication steps during inference: dispatching hidden states to the activated experts and combining the resulting expert outputs. As compute throughput has outpaced interconnect bandwidth, these synchronous collectives have become a major bottleneck to MoE scaling and efficiency. We address this challenge with EchoMoE, a lightweight architecture modification that retrofits existing MoE models, requiring approximately 8,500x fewer training tokens than full pretraining. Motivated by the observation that transformer blocks update the residual stream incrementally, we use a hidden state from an earlier layer as a reference and communicate only the compressed difference between it and the current state, rather than communicating each hidden state in full. EchoMoE periodically refreshes these reference states and transmits only the compressed differences synchronously, reducing the traffic on latency-critical paths without significant computational overhead. EchoMoE reduces inter-node communication volume by 58% and improves throughput by 22.6%, while preserving average performance within 3.8 percentage points across 18 benchmarks. EchoMoE further improves inference throughput by up to 2.38x over bandwidth-constrained Ethernet, paving the way for wide-area distributed inference. EchoMoE also tolerates quantization of the communicated representations without measurable quality loss. We release an efficient implementation for applying EchoMoE to pretrained MoE models, together with an inference integration for vLLM.
benchmark - arxiv:2611.22207 · cs.CVARG-Avatar: Animation-Rendering Guidance for Efficient Feedforward Avatar ReconstructionJianjin Xu, Soham Dasgupta, Fernando De la Torre
Reconstructing animatable 3D human avatars from images is important for applications such as gaming and augmented reality. Feedforward methods make avatar reconstruction accessible by reducing inference time to a few seconds, but recent approaches rely heavily on scaling model capacity, training data, and computation. We ask whether such scaling is necessary. To this end, we introduce ARG-Avatar, an efficient feedforward framework with only 68M trainable parameters that achieves SOTA reconstruction quality. Our key insight is that 3D animation and rendering already provide strong geometric correspondences that existing methods largely attempt to learn from data. In light of this, we design the Foreground Avatar Coordinate RoPE attention (FACRoPE) and Intermediate Token Rendering (ITR) pipeline to guide avatar tokens to attend to their corresponding patches. Experiments across multiple datasets show that ARG-Avatar consistently outperforms prior methods when using two or more input images. Compared with LHM++, ARG-Avatar uses approximately 9$\times$ fewer trainable parameters, 4.8$\times$ fewer training sequences, and 14$\times$ less training compute, while running 6$\times$ faster and achieving a 2.7 dB PSNR improvement on the out-of-distribution NeuMan benchmark.
benchmark - arxiv:2611.22203 · cs.AIEGS: A Complementary Safety Framework for Educational LLMsZexuan Fei, Zhengnan Guo, Yiming Luo, Kaiyi Huang +3
Large Language Models (LLMs) offer strong potential for personalized education, but adolescent-facing deployment raises risks of toxic, biased, misleading, or developmentally inappropriate content. We argue that educational safety should move beyond binary refusal: student queries are often ambiguous or curiosity-driven, requiring contextual guidance rather than simple blocking. We introduce the Educational Guardrail System (EGS), a lightweight, model-agnostic runtime framework that complements intrinsic alignment through query moderation, scenario-aware pedagogical guidance, and response moderation. EGS triages inputs by risk level, redirects gray-zone queries with category-specific educational interventions, and audits outputs before delivery. Experiments across general and education-specific benchmarks show that EGS improves moderation performance, refusal quality, and jailbreak robustness across diverse LLMs. These findings suggest that defense-in-depth guardrails can strengthen safety while preserving the pedagogical utility required in educational LLMs. Our code and model weights will be publicly available to support community adoption.
benchmark - arxiv:2611.22199 · physics.opticsObservation of temporal routing of topological edge statesLetian Xu, Muhammad Imran, Yongtao Jia, Ying Liu +2
The switching or routing of topological edge states is crucial for disorder-immune state transfer. Previous experimental observations have primarily relied on optical waveguide arrays that treat the spatial coordinate as an equivalent temporal dimension. However, direct experimental observation of the routing dynamics of topological edge states in physical time remains challenging. Here, we report the experimental observation of temporal routing of topological edge states. By constructing a lattice composed of two coupled nonlinear Su-Schrieffer-Heeger (SSH) chains, the topological edge states exhibit temporal routing behavior controlled by nonlinearity. As the excitation voltage increases from the linear regime, we observe that the voltage intensity distribution transitions from periodic interchain exchange to slowed interchain exchange and ultimately to localization with strongly suppressed interchain exchange, thereby validating the phenomenon of temporal routing. These dynamics are experimentally observed within a finite time window limited by circuit dissipation. Our findings provide valuable insights into the temporal manipulation of topological edge states.
manipulation - arxiv:2611.22195 · cs.ROHaptic Probing for Subsurface-Aware Force Distribution on Layered Soft TissueLei Shi, Xiaoyu Wang, Cheng Zhou, Yongjian Liu +4
Force-regulated contact with layered soft tissue fails most often where the subsurface composition changes: controllers infer the substrate from the loading signal alone, so the estimate guiding them lags the change---more acutely, we argue, over compliant than over bone-backed tissue. We remove that lag with a separate haptic channel. Under a light preload, a brief low-amplitude excitation lets the local dynamic admittance discriminate muscle from bone-backed tissue relative to a per-subject baseline, before task force is applied---active haptic sensing used to commit a control decision rather than to characterise tissue. A force- and tactile-driven distribution law then consumes that regime throughout execution: an instantaneous allocation equalises pressure across the engaged fingertips, while the regime sets the admissible pressure and hence the area the hand must recruit to deliver the target force---making patch area a controlled variable rather than a by-product. Probing, switches and recruitment are priced in a partitioned passivity budget, so safety monitoring cannot suppress the probe and a refused recruitment forfeits force tracking rather than the pressure bound. We report force tracking and pressure uniformity over curved, migrating contact on three human subjects, with regime labels from anatomical landmarks.
tactile - arxiv:2611.22191 · cs.LGRISA: Recoverable Sparse Attention for Block-Diffusion Language ModelsXinrui Zhong, Fanjiang Ye, Yuke Wang
Block-wise diffusion language models (DLMs) refine multiple tokens in parallel, offering an alternative to token-by-token autoregressive decoding. Their long-context inference, however, remains bottlenecked by memory-bound attention: every denoising step repeatedly reads a growing historical key--value (KV) cache. Directly applying sparse attention does not fully solve this problem, because different queries can select different pages, inflating the block's collective read set, while repeated support construction adds overhead and permanent eviction can discard information needed later. We present RISA, a recoverable sparse-attention method with shared read plans for block-wise DLMs. RISA exploits the fact that historical KV remains fixed within a block while queries evolve across denoising steps. Newly generated history is organized into exact priority and recovery tiers; seed queries then build a protected, bounded read plan whose indices are reused across the block while attention is recomputed with fresh queries and current-block KV. This separates history preservation from repeated reading and amortizes sparse-support construction without changing the backbone or decoder. On DreamReasoner-8B, RISA retains comparable task performance. At 32K, the default configuration speeds up attention by $2.67\times$ and fixed-trajectory requests by $1.36\times$ over dense, and attention speedup reaches $5.53\times$ at 128K. Across eight settings on three models, RISA decodes faster than the tested MAGE, QUEST, and LoSA budgets at matched KL.
long-context - arxiv:2611.22190 · cs.MAEmergence and Collapse of Leader-Follower Flocks through Phantom Targets: Surprise Nullification in Bayesian and Inverse Bayesian InferenceShuji Shinohara, Daiki Morita, Yoshihiro Nakajima, Takeshi Takano +3
In Bayesian and inverse Bayesian (BIB) inference, a large prediction error can trigger surprise nullification, which expands belief and likelihood variances by a common factor, leaving the updated variance ratio the same as in standard Bayesian updating. This can increase belief uncertainty while keeping the learning rate low. In the simulations, three agents used BIB inference to predict one another's positions and moved toward those predictions, repeatedly forming leader-follower flocks that subsequently collapsed despite having no alignment rules or designated leaders. Three leadership criteria (front position, heading stability, and flock initiation) identified the same agent in 89% of eligible flocks, with different agents serving as leaders across flocks. Flocks formed when an agent nearly stopped incorporating another's observations (link cut) and moved straight toward a predicted position detached from reality (phantom target), thereby attracting followers. Link cuts arose when two successive resets assigned a very small and then an ordinary likelihood variance, sharply increasing the likelihood-to-belief variance ratio. Flocks collapsed upon link recovery or when returning phantom targets balanced the leader's forward and backward attractions. Without nullification, cuts persisted throughout the analysis period; expanding only belief variance caused recovery within a few steps. Expanding likelihood variance as well allowed cuts to persist while belief uncertainty grew, enabling later resets and recovery. Thus, nullification preserves relative precision, allowing cuts and flocks to persist, while reducing absolute precision to enable recovery and collapse, reconciling these seemingly opposing processes through a single operation. These results suggest that, in this model, individually inaccurate predictions can give rise to collective order.
agent - arxiv:2611.22181 · eess.SYBenchmarking a Grid Foundation Model for AC Optimal Power Flow: Feasibility and Warm-Start Results on Synthetic Texas and South Carolina GridsTam Kemabonta, Brighton Ombuki, Vladimir Abdelnour
Learning-based surrogates for alternating-current optimal power flow (AC-OPF) are often judged by the gap between their objective value and that of a nonlinear solver, which does not show that a prediction satisfies power balance and network limits. This paper tests whether predictions from the pretrained foundation model GridSFM-Open 1.1.0, recently released by Microsoft, are feasible AC-OPF operating points and whether they are useful initial values for a nonlinear solver. The model was evaluated zero-shot and after fine-tuning on 10, 100, and 1,000 labeled scenarios (FT10, FT100, and FT1000) on 500 scenarios for each of two synthetic grids, ACTIVSg2000 (Texas) and ACTIVSg500 (South Carolina), with Ipopt, MATPOWER MIPS, and pandapower PIPS as solver references. The lowest Texas median cost gap, 0.44\% for FT100, coincided with thermal violations in 24.1\% of cases and a median maximum bus power mismatch of 89 MW and 197 Mvar. Texas zero-shot predictions had a median maximum mismatch of 648 MW and 1,641 Mvar, and every South Carolina case violated a thermal limit for every model variant. Seeding Ipopt with FT1000 predictions gave geometric-mean solve-time speedups of 1.55x on South Carolina and 1.12x on Texas (p = 0.06, not significant at the 5\% level), while zero-shot seeds slowed South Carolina solves (0.26x). A small objective gap therefore did not indicate feasibility: although median zero-shot inference took only 1.32 s on Texas and 0.157 s on South Carolina on a shared host, every prediction still required a nonlinear solver, and fine-tuned predictions gave modest, grid-dependent warm-start speedups while zero-shot predictions slowed convergence on one grid. GridSFM may therefore be useful for initializing a nonlinear solver, but obtaining feasible AC-OPF solutions still required the solver.
benchmark - arxiv:2611.22179 · cs.AIRAG in a Trenchcoat: When Minimal Memory Is Enough for Agentic Systems, and When It Isn'tJingyu Liu, Zongze Li, Zach Xu, Zhanhui Zhou +3
Agentic memory research has accumulated extraction, supersession, reflection, and graph synthesis on top of retrieval-augmented generation. However, we found that almost none of those primitives earn their cost in our tested memory benchmarks: we show this with BRIDGE, a deliberately minimal RAG memory with chained observation notes, hybrid BM25 with cosine retrieval, and chronological rendering, which reaches parity within bootstrap CI with the prior SOTA on LongMemEval-S and BEAM, and beats the best memory baseline on EverMemBench by 5.7-6.3 points at matched reader. It runs at ~5-7K reader-prompt tokens per query, using a single LLM call for each ingestion or query. We argue that this parity is itself the result: standard benchmarks score what memory retrieves, not what memory enables when the agent acts. To test the latter, we introduce five adversarial probes (Rule Application +-, Pattern Induction Style / Procedural, Dynamic Multiturn) judged at generation time by a cross vendor verification chain, where each query can only be solved by applying a remembered rule or pattern it never mentions. Across 5 probes x 20 scenarios, no architecture wins universally: atomic fact extractors (LangMem) win rule application, while BRIDGE's block-summary memory wins pattern induction. This result holds across four domains, three graders, and three reader families. Each design fails where the other wins, for reasons a recall benchmark cannot see: atomic-fact extractors discard the form a demonstrated pattern takes, while block-summary memory retrieves the rule but the reader does not act on it, a gap that presenting the rule more prominently largely closes. What separates memory designs is not what they retrieve but what they enable an agent to do, and that is what agentic memory should be evaluated on.
memoryretrieval-augmentedragagentagenticbenchmark - arxiv:2611.22177 · cs.AIThe Only Edge Was Memorized in Training: LLM Trading Agents Around Earnings and Bad NewsJackson Wang
Large language model (LLM) trading agents are tested on news, but news reaches stock prices within minutes, and a benchmark that replays headlines at their publication time finds no alpha in trading on them. Money can then remain in two places, before news whose date is known, such as earnings announcements, and after bad news, if the market's fall went too far. However, studies of LLMs and earnings read documents published with or after the announcement, and studies of reversal sort monthly portfolios, so neither tells whether an agent can trade these moments after costs. In this paper, we test LLM agents at both moments on the same look-ahead-free benchmark, before 598 earnings announcements and after 409 bad-news drops of large US stocks. Every input is cut at the decision close, staged agents are compared with single calls, simple rules and a logistic regression, and every decision is made with and without the stock's name. Across five models from four vendors, no agent beat the simple bets. Before earnings, the agents mostly bought and earned what buying before every announcement earned, and after bad news, a Gemini-based agent called almost every fall an overreaction while no agent picked better dips than buying them all. The only gains came from what the models had memorized in training. Given only the ticker and the dates, a GPT-based model called the direction of 59.2% of the announcements before its training cutoff and 53.3% after it. Agent builders should therefore compare an event agent with the simple bet and test it blind after the training cutoff.
agentllm agentbenchmark - arxiv:2611.22174 · cs.AIReconfiguring Procedural Knowledge for Compositional Robot Skill AdaptationDaehee Lee, Dongsu Lee, Sanghyun Ahn, Minjong Yoo +1
General-purpose behavior models increasingly internalize broad procedural knowledge during training, making their adaptation to new goals and skill interfaces less a problem of learning entirely new behaviors than of reconfiguring existing competence. We study this capability as reconfigurability, the ability to realize new behaviors by cost-efficiently restructuring existing components and relations under sparse governing information, where task specifications provide high-level goals and procedure specifications expose a limited set of reusable factors, leaving the concrete bindings and compositions required for execution underdetermined. We propose RecPo, a framework that represents skills task-agnostically through multiple procedure views and materializes executable skill sequences by composing stored transitions to realize task-specific changes. RecPo maintains a procedure knowledge base of skill-induced transitions, with each view exposing a reusable factor and preserving local skill connectivity. At deployment, it interprets task specifications through the same views and resolves the resulting compositional queries over the stored relations to bind skills for execution. On long-horizon tasks in an extended Franka Kitchen benchmark, RecPo achieves 91.9% ordered completion with 1.0 ms of task inference per episode across varying procedure support and task-record sparsity. Real-robot studies further demonstrate adaptation through local memory edits without task-specific retraining.
frankamemorybenchmark - arxiv:2611.22171 · cs.AIReadie: Low-Latency Code Sandboxes for AI AgentsArjun Sivaraman, Zhaoheng Li, Alan Shi, Sanjay Malisetty +2
From quantitative research to personal assistance, AI agents act by iteratively generating and executing code. To decouple the agent harness from code execution, many agents offload execution to dedicated systems called code sandboxes. However, existing code sandboxes suffer from slow startup. Whether they rely on custom snapshots or a single fat image, the root cause is the same: it is hard to build a compact, pre-warmed snapshot that can be shared across diverse agent-generated code while covering nearly all the artifacts (e.g., packages, datasets) it requires. In this work, we propose learned startup snapshots for low-latency code sandboxes. Our insight is that while it is hard for a single snapshot to be both compact and cover most workloads, a set of snapshots can achieve both. Each snapshot is small, initializes quickly, and covers a fraction of agent-generated code; together, the set covers most of it. We learn this set from a large training workload by formulating snapshot construction as an optimization problem and solving it with an approximation algorithm. Planned on 15K public and LLM-synthesized scripts, our system, Readie, reduces startup latency by 3-22$\times$ and end-to-end runtime by 1.3-3.3$\times$ compared to existing approaches at the same storage budget.
agentai agent - arxiv:2611.22169 · cs.CVNot All Evidence Is Admissible: Evidence Governance for Tool-Augmented Medical AgentsRunxun Zhang, Onofrio Antonio Catalano, Jinqiao Wang, Jingwei Wei
Tool-augmented medical agents treat whatever their tools return as evidence. We show that this default is costly: medical tools fail silently, and a frozen reasoner cannot ignore what enters its context, so many cases it would have answered correctly are flipped to wrong once tool outputs are added. We call this evidence hijacking. $\mathcal{V}$-information explains the phenomenon: the reasoner has lost the option to ignore, so a tool output can carry negative usable information, and moving the decision of what to keep outside the reasoner restores that option. The theory further shows that part of any filtering gain comes merely from letting the reasoner see less, and only selection by content lets it see better. Building on this, we recast tool-augmented reasoning as evidence governance, which separates deciding what counts as evidence from the reasoner that answers, and propose context-independent evidence governance (CIEG), the first medical agent framework that rules, after acquisition, on whether each individual tool output may enter as evidence. Every tool output is written as a card that can be checked against the raw input; an independent checker that sees only the input and the cards decides which cards become evidence; and the reasoner answers from the admitted cards alone, with no training anywhere. On medical question answering, fundus glaucoma screening, and chest X-ray pneumonia classification, it achieves the highest accuracy among models of comparable scale, and significantly outperforms state-of-the-art tool-using agents and multi-agent systems built on the same backbone at lower cost.
agentmulti-agentagent frameworkagent system - arxiv:2611.22167 · eess.SYAI Agents Executing an RMS and EMT Ride-Through Workflow on a 2,000-Bus Synthetic Texas Grid: A Case Study with Open-Source ToolsTam Kemabonta, Brighton Ombuki, Vladimir Abdelnour
This paper reports a case study of engineer-supervised AI agents executing a predefined RMS and EMT ride-through workflow on the 2,000-bus ACTIVSg2000 synthetic Texas grid with open-source tools. The test case is a 133 MW battery with a co-located 130 MW data-center load, assessed against ride-through criteria fixed before simulation and for loss of synchronism elsewhere in the grid. One agent ran 44 RMS cases and 15 without-project comparisons in ANDES; the other ran a 220-row EMT intake bench and an 84-row system study in DPsim and ParaEMT on a Kron-reduced area. Across the 44 RMS cases, 127 criterion cells fail, including loss of synchronism in all 44. Two findings concern the project plant. The battery keeps absorbing reactive power during sags on the light-load schedule (B3, as low as -19.8 Mvar on S4 T1), and the data-center load returns only at 5.00-5.56 s after long bolted faults (L3), which misses the ERCOT load-return requirement. In 13 of the 15 comparisons, which the authors selected after slips or speed-gate stops, the same machines slip without the project; the other 29 runs have no without-project comparison. RMS and EMT values differ in network, plant model, controls, operating point, and base, and so they are reported as separate, unaligned observations. The paper reports the errors the authors identified in the agent logs, 63 author messages, and 3.8 days from agent start to the integrated study report. The agents executed and documented the predefined ride-through and stability workflow, which covers one part of an interconnection study.
agentai agent - arxiv:2611.22164 · cs.CVHandRSI: Recursive Self-Improvement for Hand Pose EstimationZhekai Wang, Luqing Luo, Wenhao Yao, Junjie Lu +3
Egocentric 3D hand pose estimators still miss hands at the image border, misplace occluded hands and jitter. Such failures are usually fixed by hand-written inference refinements and by fine-tuning, two interdependent steps that are tuned separately. We present HandRSI, a recursive self-improvement framework for a hand pose system that consists of a video estimator and its executable inference harness. In each round, coding agents refine the harness, and a change is accepted only if it improves a composite of all ten benchmark metrics on held-out data without significantly regressing any of them. The accepted harness then guides the estimator through selective distillation, which supervises hand presence and temporal consistency on frames without ground truth, while ground truth supervises pose. Since the updated estimator may change which refinements remain useful, the harness is re-validated on it. Starting from the released ACE-Ego-Hand model, five rounds reduce MPJPE-p by 35% on HOT3D and 11% on ARCTIC and roughly halve jitter. On HOT3D, HandRSI's MPJPE-p is 16% lower than that of ACE-Ego-Hand fine-tuned on the same ground truth. Under a unified protocol, it outperforms nine re-run published methods on all ten metrics of both benchmarks. Hypervolume selection on the accuracy-speed front yields Lite, Medium and Heavy systems running at 20.7 to 41.1 FPS. This report gives the full method, every stage of every round and the complete protocol. Code, evaluator, harnesses and weights are public.
self-improvementbenchmarkevaluator - arxiv:2611.22161 · cs.LGWhat RL on Expert-Built Tasks Teaches Nemotron 3.5 SuperSushant Mehta, Edwin Chen
We post-train NVIDIA's Nemotron 3.5 Super (120B total, 12B active parameters) with reinforcement learning (RL) on five expert-built task datasets, one at a time: real-world software engineering, complex constraints, long-horizon tool use, professional chart understanding and enterprise document reasoning. Each run trains a fresh rank-32 LoRA adapter with CISPO on Tinker, with no supervised stage, and each adapter is compared with the base model on the same tasks, harness and settings. On twelve headline benchmarks, eight built by other groups and four by Surge AI, the gains range from +4.5 points on IFBench to +13.8 on GDP_pdf, which rises from 10.4% to 24.2%; SWE-bench Pro V2 rises from 86.9% to 92.1%, Terminal-Bench 2.1 from 63.9% to 70.3%, MultiChallenge from 58.3% to 66.6%, AutomationBench from 14.8% to 27.5% and Chartography from 39.4% to 50.6%. The document adapter, trained without API tools, also gains +8.0 points on AutomationBench. In attempts read side by side with the base model's on tasks each adapter newly solves, the base model's failures recur in a few modes, and each adapter does more often what the base model missed: the tool-use adapter finds the data and the rules before acting, the coding adapter writes its own test or scratch files on SWE-bench Pro V2, the constraints adapter carries earlier turns into its answers, the chart adapter measures the marks in code, and the document adapter reads AutomationBench's spreadsheets.
tool usetool-usebenchmark - arxiv:2611.22158 · cs.LGCentroid Spectral Coupling: Implicit Alignment and Representation Diversity in Self-Supervised LearningM. Hadi Sepanj, Benyamin Ghojogh, Saed Moradi, Paul Fieguth
Self-Supervised Learning (SSL) aims to align augmented views while preserving representation diversity and preventing collapse. Existing non-contrastive methods often address these goals using multiple weighted loss terms. We introduce Centroid Spectral Coupling (CSC), a single-term SSL objective that implicitly couples view alignment, representation diversity, and collapse resistance. CSC constructs centroids from independently normalized augmented-view pairs and applies a ridge-based spectral objective to their uncentered second-moment matrix. Preserving centroid magnitudes encodes view agreement while encouraging balanced representation energy across spectral directions. We establish an exact decomposition into nonnegative alignment and spectral-imbalance components, characterize the geometry of global optima, and derive finite-loss guarantees for alignment and diversity. We also show that complete constant-output collapse is a strict saddle in normalized representation space. Experiments on CIFAR-10, CIFAR-100, and Imagenette show that CSC achieves the highest linear Top-1 accuracy among the evaluated non-contrastive methods on all three benchmarks, and the highest kNN Top-1 accuracy on both CIFAR datasets. A ridge-parameter sensitivity study reveals an empirical alignment--diversity trade-off: increasing $κ$ strengthens view alignment but reduces representation effective rank. These results show that a unified spectral objective yields competitive visual representations without explicit contrastive negatives or separately weighted alignment and diversity penalties.
benchmark - arxiv:2611.22151 · cs.CVEgoHandWAM: Probing the Role of Hand Motion in Egocentric World Action ModelingKailin Wang, Haoxiang Jie, Yaoyuan Yan, Zhiyou Heng +1
How much does hand motion tell us about what a first-person camera will see next? EgoHandWAM forecasts three-dimensional wrist and fingertip trajectories and a future visual feature from the current image and recent hand states. We train on 751,065 EgoDex windows using a frozen DINOv2 encoder. On held-out development sessions, the action-only model reduces two-second wrist endpoint error from 60.26 to 56.48~mm relative to a stationary forecast. It still underestimates long reaches. Adding future-feature supervision gives nearly identical trajectory accuracy. With the recorded hand history, the visual head reduces smooth L1 error by 1.32\% relative to copying the current feature. Shuffling histories across sessions erases this gain. In a separate four-task probe, adding recorded future displacements to observed context and future tracking-validity masks reduces visual-feature error by 3.93\%. Masks alone give little improvement. A trajectory predicted from the same observations adds little. On two simulated RoboTwin tasks, substituting aligned human examples for an auxiliary batch of robot examples gives no consistent increase in task success across three training seeds.
robotwin - arxiv:2611.22148 · cs.AILong-Context Generation Is a Sampling Problem (When You Turn Up the Heat)Allen G. Roush, Minh Nhat Nguyen, Ravid Shwartz-Ziv, Judah Goldfeder +1
Open-weight language models are widely observed to get worse the longer they generate. Often, attempts to fix this involve more training. At high temperature, The sampler, i.e. the rule that picks the next token out of the model's probabilities, decides whether long and agentic generation survives and/or thrives. We test 29 open-weight models from 1B to 122B parameters on ten established benchmarks plus needle retrieval out to one million tokens, covering SWE-bench Verified coding agents, long chain-of-thought reasoning, long-form writing, long structured output and long-context reading. Samplers that keep their pool of candidate tokens small and steady as temperature rises don't lose accuracy. Samplers that let in low-probability tokens fall apart. At temperature 2, coding agents running top-p, top-k or Top-H resolve 0 of 100 GitHub issues, while the same agents running top-nsigma resolve 46 to 62 and win every issue the two samplers disagree on, on three Qwen3.5 models. Over all benchmarks top-nsigma keeps a median 100% of its temperature-1 score where top-p keeps 26%. Counting how many tokens each sampler lets through explains why. In one Llama-3.1-8B story, going from temperature 1 to 2 grows top-p's pool from 4 tokens to 51,845 while top-nsigma stays at 3. Across 133 combinations of sampler setting and temperature, the typical pool size predicts judged quality (Spearman correlation of -0.77), and when we pick the pool at temperature 1 and only apply temperature afterwards, Top-H, min-p and p-less stop collapsing. At the default temperature of 1 most samplers perform about the same, We release all code and data to the public.
long-contextagenticbenchmark - arxiv:2611.22146 · cs.ROHandUMI: A Modular Hand-Worn Parallel-Jaw Interface for Robot-Free Bimanual Data CaptureBrik Meza, Leonardo Perez, Alvaro Mendoza-Li, Aniela Ubillus +4
HandUMI captures robot-free bimanual demonstrations with a hand-worn parallel-jaw interface. Swappable fingertips match the robot's installed gripper; Quest/PICO controllers track hands and encoders measure aperture. Robot-agnostic recordings are retargeted by inverse kinematics to seven simulated bimanual arms (median position error $\leq 1.2$ mm; $\leq 4.7$% of episodes rejected). On bimanual towel folding, 562 robot-free demonstrations train policies on real PiPER arms: ACT succeeds in 12/20 attempts and fine-tuned $π_{0.5}$ in 7/20; ACT on the first 101 and base $π_{0.5}$ each succeed in 0/20. Hardware and software are open source (Apache-2.0), and a pair costs USD 557 with Quest 3S, excluding the optional ZED Mini and host computer. Project website: https://murobotics.ai/research/handumi.
gripper - arxiv:2611.22144 · cs.LGWhen does deep reinforcement learning pay off in HVAC supervisory control? Four small Modelica case studies trained directly on FMUsBingnan Zhao, Harshala Gammulle, Tharindu Fernando
Deep reinforcement learning (DRL) is often proposed for the supervisory control of heating, ventilation and air-conditioning (HVAC) systems. The reported gains are, however, difficult to interpret, because the baselines are often weak and the attainable optimum is seldom known. This paper examines four small, reproducible case studies based on an office building in Brisbane, Australia: pre-cooling under a time-of-use tariff; chiller staging with condenser water reset; chilled water supply temperature reset with a wet cooling coil; and chilled water storage under time-of-use prices and a demand charge. The plants were modelled with the Modelica Buildings library, and in three of the cases a soft actor-critic agent was trained directly on the exported functional mock-up unit (FMU). Each agent was compared with tuned rule-based controllers and, where practicable, with a near-optimal reference from hourly exhaustive search or dynamic programming with perfect foresight. In the two cases without storage the agent reached the near-optimal reference, but the margin above a well-chosen rule was small: 1.3% for condenser water reset and 0.18% for chilled water reset. In the latter case two common reset rules consumed more energy than a fixed set point. With a chilled water storage tank the margin was considerably larger. The agent outperformed both rules (-7.8%, compared with -3.9% and -6.3%) and adopted midday charging during the solar off-peak period, yet it obtained only about 45% of the saving achieved by the perfect-foresight schedule and did little to limit the demand peak. In the pre-cooling case a hand-tuned rule remained superior (-12.0% compared with -5.6%). These results suggest that studies of DRL for HVAC control should state the margin between the best rule and a near-optimal reference. Some practical difficulties encountered in training on co-simulation FMUs are also reported.
agent - arxiv:2611.22142 · eess.SYFinite sample guarantees of MPPI: suboptimality, sample complexity and closed-loop guaranteesChangyi Lei, Dragan Nesic, Ye Pu
This paper considers model predictive path integral (MPPI) control for nonlinear finite-horizon optimal control problems with bounded inputs, allowing nonconvex and nonsmooth costs. We prove theoretical guarantees that quantify the suboptimality caused by positive temperature and finite sampling on MPPI and establish their implications for closed-loop guarantees. Specifically, we derive a suboptimality bound for infinite-sample MPPI that characterizes the effects of temperature and prior coverage, followed by a high-probability finite-sample approximation error bound that reveals the dependence on sample size, temperature, and cost variability. By incorporating these bounds into a terminal-constraint-free MPC framework, we further establish semi-global practical recurrence of the closed-loop system. Numerical examples involving a four-link planar manipulator illustrate the temperature-sampling tradeoff and the influence of prior quality on approximation accuracy and closed-loop performance.
manipulator - arxiv:2611.22138 · cs.RORoboTeam: Benchmarking Vision-Based Multi-Agent Control of Many Robot Arms in Real TimeChangxuan Fan, Zhongwei Xie, Boyi Zhang, Shaoli Hu +17
Multimodal agents can now control robot arms by calling tools and writing code. Existing benchmarks fall short in three ways: most test one or two arms on short tasks, many pause the world while the agent thinks, and many give it ground-truth object positions. We present RoboTeam, a simulated benchmark where agents control 4 to 16 robot arms in real time from camera images alone. Its 36 tasks are long jobs that the arms must do together, such as passing timbers or building a picture from hundreds of cubes. We start from the most flexible agent system, the adaptive system, where one agent decides when to fork delegates. With it, gpt-6.1-sol and opus-5.5 complete 28 and 27 tasks, and four other models at most 3. But at a high price: a typical task takes gpt-6.1-sol 3.5 hours and 218 million input tokens, and a third of its steps fail at least once and must be retried. We then compare it with three fixed systems: single-agent, hierarchical, and decentralized. All four complete a similar number of tasks but differ in time and tokens. The hierarchical and decentralized systems are faster but use up to 1.9 times the tokens, and the single agent uses fewer tokens but takes 1.6 times as long, so the adaptive system lies between them. Even so, no model or system is yet fast, cheap, and reliable at once. RoboTeam measures these limits where earlier benchmarks cannot: long multi-arm jobs, camera images only, and a world that never pauses.
agentmulti-agentagent systembenchmark - arxiv:2611.22136 · cs.AIBig Brains, Tiny Board: Pupavision, an Agentic Physical AI Instrument for Monitoring Drosophila Pupariation and Lifecycle TransitionsJoel Sadler, Daniel Graham, Melanie Worley
Many biology experiments are limited by the availability and ease of use of instruments that monitor and record them. For example, the timing of pupariation in the fruit fly Drosophila melanogaster, a widely used model organism, can provide biologically relevant information about growth regulation and is often recorded manually. Here, we present Pupavision, a small physical AI instrument built from a single-board Linux computer and a day/night camera for roughly 45 to 75 US dollars in parts. The key idea is that a frontier coding agent runs its agentic harness directly on the tiny board. The model thinks in the cloud, but the agent works on the device. The agent assists image capture, writes code to repair errors, and checks the experiment every day to keep raw data capture reliable. If recording stops, the researcher gets a ping on their mobile phone. A simple deterministic capture service records one image every ten minutes and never depends on the agent. We report a 26-day pilot in a fly incubator (about 3,580 frames), including eleven days in which the agent was logged out and capture continued, and a 51-hour outage whose cause led to the layered design. A researcher-in-the-loop workflow then produces per-vial pupariation curves by tracing selected pupae backward in time from a late reference frame. Future work will validate the assay for lifecycle tracking accuracy against expert-labeled data and expand real-time computer vision analysis. Pupavision offers a blueprint for do-it-yourself agentic instruments that put frontier intelligence on a tiny board, with deterministic code underneath. We present this as an early case study from a pilot lab deployment and contribute lessons learned toward a future where scientific domain experts are empowered by DIY agentic hardware.
agentagentic - arxiv:2611.22132 · cs.RODailyBrick: Language-Guided Synthesis of Brick Assembly InstructionsHang Zhang, Qijian Tian, Yuan Xie, Xuhong Wang +1
Generating a LEGO model from a user request is more than a plausible 3D shape generation. LEGO models require physical support per brick, and each brick should be placed unblocked. Current methods address this issue by training an expert large language model. We reduce the problem of generating extremely long context on all brick connections to that of writing a program that builds LEGO models like a human, considering each intermediate step could be executed. We introduce DailyBrick , a multi-stage agentic system that produces a LEGO model together with a brick-by-brick assembly instruction for human building and robot assembly. We design a multi-stage harness with roles to help LLMs generate more effectively. A versioned construction SDK compiles multi-level decisions, from semantic subparts to individual bricks, into canonical operations. An external Meta-Reviewer agent replays and inspects the generated package, verifies the model and its construction process, identifies failures, traces them to the responsible stage, and routes them to bounded repairs or to editable candidates for later skill or tool updates. Extensive experiments demonstrate that DailyBrick generates LEGO models based on simple user requests better than recent agentic systems, and show its utility in downstream applications such as robot manipulations on model assembly from instructions and brick sets (https://SpatialAgent.github.io/DailyBrick/).
manipulationlong contextagentagentic - arxiv:2611.22131 · cs.LGLoCaR: Localized Canvas Retrieval With Frozen Vision-Language ModelRahim Hossain, Dipanwita Saha, Ashirbad Mishra, Xinxin Shu +2
Whole image embeddings can under represent small objects and local attributes that distinguish relevant images for localized queries. We present LoCaR, a training free framework for localized text to image retrieval with frozen vision language models. LoCaR constructs and utilizes independent proposals (SAM2) focusing on the object so as to get a more localized representation that independently encodes each region using the frozen visual encoder. We further study paired-support adaptation as an optional residual that uses matched text region exemplars without updating the underlying encoders. We also introduce LOCFG, a Visual Genome derived benchmark designed to evaluate localized retrieval for smaller and more challenging referents. Across several vision language backbones, our results show that improving the region representation itself is the main source of gain.
benchmark - arxiv:2611.22130 · cs.LGBandit Online Submodular Minimisation Using Zeroth-Order OraclesAmir Ali Farzin, Philipp Braun, Iman Shames
We consider online submodular minimisation under bandit feedback, where at each round a decision maker plays a subset and observes only its cost, while the cost functions, and hence their minimiser, change over time. Existing bandit methods only guarantee static regret, which compares against a single fixed subset and is a weak benchmark when the minimiser moves. This paper provides, to the best of our knowledge, the first method with guarantees for tracking a moving minimiser in this setting. We propose a zeroth-order method built on a two-point oracle for the Lovász extension whose two evaluations are both subsets that can be played. The perturbed and base points are rounded by a common random threshold, and drawing the smoothing direction from the unit sphere makes the base evaluation a zero-mean control variate, so the oracle is an unbiased estimator of the smoothed gradient. We establish dynamic regret bounds of order $O\big((1+\sqrt{P^\ast_N})N^{3/4}\big)$ in general and $O\big((1+\sqrt{P^\ast_N})N^{2/3}\big)$ under a sufficiently small drift, where $P^\ast_N$ is the path length of the minimiser sequence, together with the corresponding $O(N^{3/4})$ and $O(N^{2/3})$ static regret bounds. Numerically, the method attains lower regret than the algorithm of (Hazan and Kale (2012)), with the advantage growing with the path length.
benchmark - arxiv:2611.22115 · cs.AITeleChat to Xing4: Advancing Frontier Agentic Models on Domestic Chips and FrameworksXingchen AGI Lab, :, Baotian Hu, Bingkai Yang +136
We present Xing 4, a series of large language models that extends the TeleChat lineage through coordinated advances in model architecture and training infrastructure on Ascend. This report describes the design principles and systems engineering behind the series, spanning architecture selection, large-scale pretraining, and reasoning and agentic post-training. We study how sparse activation, attention design, model depth and width, and speculative decoding balance model capacity with computation, memory access, and execution overhead. Our pretraining infrastructure builds on MindSpore HyperParallel, combining topology-aware hybrid sharding, efficient distributed optimizer execution, and coordinated management of long-sequence workloads and activation memory. CPU affinity and core isolation further reduce host scheduling interference. Together, these mechanisms support the pretraining of a 285B-parameter model on 8,192 Ascend 910C cards, providing a concrete setting for the architectural and systems studies. For post-training, we introduce TeleGym, a unified agent execution platform that connects execution, evaluation, and trajectory construction to policy optimization. Disaggregated asynchronous execution, bounded oversampling, and partial rollout accommodate heterogeneous task durations while preserving trajectory continuity across policy updates. Together, these advances establish an Ascend-oriented training foundation for the continued development of Xing 4 across model scales and increasingly demanding reasoning and agent workloads.
memoryagentagenticpost-training - arxiv:2611.22113 · cs.AIRegimeCouncil: Paired-Evidence Routing For Memory Conflict Governance In LLM AgentsQingcan Kang, Mingyang Liu, Shixiong Kai, Tao Zhong +1
Language agents rely on long-term memory to personalize responses across extended interactions, yet stored facts quietly go stale as the world changes, and an agent often learns that using memory hurt only through delayed feedback after acting on an outdated record. Before the memory store is updated, how far should that negative feedback propagate: should it suppress memory use only on future queries for that same fact (a local policy), or also across other stored facts (a global policy)? Existing memory systems implicitly fix one of these two sharing scopes at design time. We formulate this online scope choice under hidden validity and delayed feedback as memory conflict governance, and prove that neither fixed policy is sufficient on its own. We then propose RegimeCouncil, a lightweight paired-evidence router that retains both scopes as advisors and arbitrates between them online without retraining the LLM. Using revealed comparisons between answering with memory and answering without memory (the no-memory anchor), a Local Advisor tracks the latest outcome on the queried fact alone, while a Global Advisor pools recent outcomes across all stored facts. A Meta Governor follows the Local Advisor by default to avoid cross-fact interference and switches to the Global Advisor only when global pooling accumulates a strictly positive recent utility margin, guaranteeing finite-time switching under delayed feedback. Across LongMemEval update streams and real-world historical drift on TempLAMA and TAQA with Qwen3-30B-A3B and GPT-4o, RegimeCouncil outperforms the better fixed-scope policy by 5.2-7.3 points on primary disagreement episodes and across all 30 held-out settings, while ablations show that removing any of its three modules reduces accuracy by 2-7 points.
memoryagentllm agent - arxiv:2611.22109 · cs.ROVTT: Reading Vehicle Size and Powertrain Type from Motion Trajectory Across Unseen Car ModelsYuhang Wang, Yuechen Guo, Jingxin Yang, Chuheng Wei +6
Vehicle size and powertrain information can enrich trajectory data for traffic analysis and simulation. We present VTT, a dataset and benchmark for identifying these attributes from motion and controller area network (CAN) signals across unseen car models. The dataset contains 9,062 drives, about 3,950 h, from 1,431 vehicles and 174 car models. It provides three footprint-based size classes, powertrain labels, and evaluation on unseen car models and unseen drivers. Using 60-s windows, we compare four sequence architectures, statistical features and physical features proposed by large language models. Training with auxiliary physical targets, data augmentation and task-specific weighting raises drive-level macro-F1 to 92.6-93.0% for battery-electric versus combustion vehicles and from 62.9% to 67.1% for size classification. On public logs from 4,061 additional drivers, the full-input model identifies powertrain correctly for 86 of 88 car models; size accuracy is 63.2%, close to the internal vehicle-level result of 63.5%. Physical-feature analysis links powertrain recognition to deceleration behaviour and size recognition to steering response. Errors near size-class boundaries and limited gains from longer observations highlight the importance of label resolution and training-fleet coverage for cross-model generalisation.
benchmark - arxiv:2611.22103 · cs.AIFrom Sorry to QED: Refining LLM Math Proofs via Angelic ExecutionYanju Chen, Hanzhi Liu, Ruizhe Qian, Jiaming Shan +4
Large language models (LLMs) can now generate Lean proof scripts that capture the high-level structure of many mathematical arguments. However, these scripts frequently fail on complex theorems: a single missing side condition or misapplied lemma can cause the prover to abort early and discard an otherwise viable proof outline. This paper presents MathEye, a framework that converts imperfect LLM-generated proof scripts into fully verified Lean proofs by combining symbolic reasoning with minimal human interaction. At the core of MathEye is angelic proof checking, an optimistic execution model that replaces failing proof fragments with placeholders while preserving the overall structure of the argument. This optimistic check produces an angelic proof that isolates the exact proof obligations that the LLM fails to discharge. Because these unresolved obligations are often logically interdependent, asking a mathematician to address every placeholder directly is inefficient. To minimize human effort, MathEye performs residual analysis over a proof dependence graph that captures symbol-level data flow in the angelic proof. By formulating obligation selection as a MaxSAT optimization problem, MathEye selects a minimal-cost set of essential obligations whose resolution suffices to complete the entire proof. MathEye presents these essential obligations to the mathematician in ordinary mathematical language and uses the mathematician's answers to automatically reconstruct the complete Lean proof. Experiments on 30 real-world benchmarks across analysis and algebra show that MathEye solves 93.3% of the problems while reducing the number of required user interactions by 55%.
benchmark - arxiv:2611.22102 · cs.AIEvoDisco: Strategy Discovery for Evolving Agent MemoryMingyang Liu, Qingcan Kang, Shixiong Kai, Kaichao Liang +3
Large language model agents accumulate experience in memory, and how that memory evolves during interaction is as important as what it stores. Recent methods update memory online, yet both the operators that transform memory and the rules that select them remain fixed by design. We propose EvoDisco, which formulates agent memory evolution as evolutionary strategy discovery: the agent must discover which operator to apply, when to apply it, and how to instantiate it, rather than rewriting memory under a static operator schedule. A learnable policy conditions on the current memory state and recent task history, then selects and instantiates one of four semantic operators---Abstract, Refine, Bridge, and Reframe---which respectively compress related memories, decompose overloaded entries, connect weakly linked experiences, and reorganize retrieved reasoning chains. The policy is updated with a bi-level fitness. An inner signal scores each transformation using a retrieval-quality improvement regularized by memory complexity. An outer signal scores strategic efficacy over a window of past decisions, combining historical operator-utility regret, utility-prediction error, and selection diversity. Under a one-pass online plus frozen-transfer protocol on ALFWorld, WebShop, and LifelongAgentBench (OS and DB), EvoDisco improves over strong memory baselines including MemP and MemRL. The largest gains are on ALFWorld ($+8.8$ online / $+12.9$ frozen-transfer points over MemP; $+16.4$ / $+17.9$ over MemRL). Ablations attribute these gains to both fitness levels and the operator library.
memoryagent memoryagent - arxiv:2611.22100 · cs.AISemantic Communication via Analog RF Computing: When the Radio Computes the MeaningChristo Kurisummoottil Thomas
Semantic communication (SC) systems transmit continuous latent representations of meaning, but computing the semantic encoder at resource-constrained edge devices incurs substantial energy and memory costs. To address this challenge, we propose an end-to-end SC framework that executes the semantic encoder in the client's RF receiver front end using analog RF computing. A base station (BS) broadcasts encoder weights as RF waveforms, while each edge device reuses its receiver's passive mixer to compute semantic latents at ultra-low energy. These latents are transmitted over the uplink for network-side task inference. We show that analog computing noise across encoder layers and channel noise affecting transmitted latents admit a unified mathematical model. We define a task-posterior semantic distortion metric and derive its second-order decomposition into per-client, per-stage semantic sensitivities. Using this decomposition, we obtain a closed-form semantic-aware water-filling allocation of the total client energy budget across clients, encoder layers, and uplinks, assigning tighter accuracy targets to semantically sensitive stages. This allocation induces a broadcast beamforming problem that accounts for client-specific, per-layer accuracy requirements. We further refine the allocation through stochastic optimization and jointly design it with multi-task noise-aware training of the shared semantic model. MNIST simulations show that, at matched total client energy, the proposed allocation reduces network semantic distortion by 9.2\%-19.2\% compared with uniform precision and 5.1\%-13.1\% compared with client-blind allocation. Noise-aware co-design yields a further 46\% reduction.
memory - arxiv:2611.22095 · cs.CVCueSplat: Learning to Schedule Sparse Deformation for Dynamic Gaussian Splatting on Edge DevicesQingyuan Tang, He Zhu, Shixuan Sun, Anbang Wu +3
Deformation-based 4D Gaussian splatting (4DGS) enables dynamic scene rendering. However, evaluating deformation networks for all Gaussians at every frame is compute-intensive on edge devices. We find that Gaussian trajectories are often locally predictable, yet prediction reliability varies across Gaussians and time. Thus, we propose CueSplat, a post-training framework that exploits local temporal regularity while keeping the pretrained backbone frozen and retaining its Gaussian set. Specifically, CueSplat predicts Gaussian attributes from recent history and uses a lightweight learned router to select exact deformation updates within a scene-specific budget calibrated offline. Selection accounts for prediction reliability and rendering contribution. Selected indices are reused across nearby frames to reduce routing overhead, while selected Gaussians still receive exact updates each frame. To control accumulated prediction error, we use drift signals to schedule three consecutive full-deformation frames that rebuild the prediction history. Evaluated across four mainstream 4DGS algorithms, CueSplat achieves up to $6.74\times$ and $8.15\times$ speedups on Jetson AGX Orin and Orin NX, respectively, with a negligible quality drop (at most $0.16$ dB in dataset-average PSNR).
post-training - arxiv:2611.22085 · cs.AIATLAS-Finance: Benchmarking AI Agents in Realistic Banking EnvironmentsElaine Lau, Antoine Pangas, Alexandre Finot, Collin Schweiker +4
Professional work involves under-specified assignments, where relevant information is spread across workplace applications and may conflict or change over time. To evaluate how well frontier agents can perform economically valuable work in such settings, we introduce ATLAS-Finance, a benchmark of 100 end-to-end investment-banking tasks across 13 simulated firms. As in real banking workflows, agents must search email, chat, calendars, notes, and files to identify requirements, gather information, reconcile conflicts/updates, and then build complex financial models. Tasks were developed by experienced bankers based on their real deal work and take them 15-30 hours to complete. Agent deliverables are evaluated with expert-authored rubrics (median: 124.5 criteria) covering financial rationale and results, spreadsheet construction, completeness, and formatting. Despite completing substantial portions of the requested work, current agents rarely produce correct deliverables. Across today's frontier models, the best ones (Claude Opus and Fable) pass fewer than 11% of tasks.
agentai agentbenchmark - arxiv:2611.22084 · cs.CVWhen Lower Harm Misleads: Auditing Concept CorrectionsTianyou Lai, Wentao Yue, Hongji Li, Chaoyuan Hao +1
An expert who corrects a concept prediction, such as a BI-RADS descriptor in breast ultrasound, expects the diagnosis to improve. Objectives that reduce the harm of such corrections are judged by harm and accuracy, yet harm also falls when a model responds less to every correction or starts from a worse uncorrected prediction, the reference for harm. We propose a correction audit, a reproducible evaluation protocol that separates these explanations: it splits each correction into harm and useful gain on identical events, separates the starting prediction from the correction response, and compares objectives with amplitude baselines matched in source harm. On three breast-ultrasound sources, worst-case correction training, evaluated from the mean-case model's start, lowers harm but gives up more useful gain in all 21 source comparisons pooled over six seeds, raising the corrected negative log-likelihood (NLL); from the same start, selecting between the objectives by harm picks the less helpful model on all three sources. On CUB, a monotonic harm penalty at its reported weight lowers path-level harm in all 80 concept sets while raising the NLL in all 80, almost entirely before any correction. At a fixed start, rescaling both responses to the same source harm reproduces the penalty's mean harm, useful gain and NLL on single-concept corrections within a prespecified margin of 0.0005 nat, although the two responses point in different directions. Lower harm therefore signals better corrections only when it is read together with useful gain, absolute risk and the start it is measured from.
evaluation protocol - arxiv:2611.22077 · cs.ROReLAP: Latent Action Pretraining through Recurrent Video Prediction in JEPA SpaceR. Gnana Praveen, Ali Rouzbayani, Zhiying Jiang, Bidhan Roy
Videos provide a rich source of physical interaction for learning robot behavior without action annotations, yet effective latent actions must capture not only local transitions but also how those transitions evolve over time. We introduce ReLAP, a latent-action pretraining framework that couples local transition learning with recurrent multi-step prediction in a frozen JEPA representation space. A shared inverse dynamics model infers compact continuous latent actions from adjacent observations, while a latent world model recurrently predicts the future JEPA features. Because each predicted feature becomes the context for the next transition, errors at later steps backpropagate through the rollout to earlier latent actions, encouraging temporally coherent dynamics beyond pairwise reconstruction. We transfer this representation to a vision-language model using two complementary objectives: latent distillation from the frozen inverse teacher and recurrent future-feature prediction through the frozen latent world model. The resulting policy is then adapted to robot demonstrations with action flow matching. At deployment, the explicit latent-action and world-model branches are removed, and the fine-tuned VLM predicts physical actions directly, reducing inference complexity. With only 11M pretraining samples and 1.06B trainable parameters, ReLAP achieves 97.9% on standard LIBERO and 70.0% over zero-shot LIBERO-Plus perturbations. On the LARY latent-action benchmark, ReLAP ranks second in both mean regression error and mean classification accuracy among the compared methods. These results demonstrate that recurrent action-free latent action pretraining can learn compact, temporally structured, and transferable representations for downstream robot control.
liberoworld modelbenchmark - arxiv:2611.22075 · cs.AICan You Undo It? Reversibility Classes for AI Agent ActionsDeepak Waghmare, Shaad Sufi
Tool-using AI agents change files, databases, access rights and public information, yet permission to call a tool does not establish whether a mistake can be undone. We present Reversibility Classes, a declared five-level vocabulary with argument-sensitive rules, conservative defaults and an explicit prior-state capture contract. The pinned implementation and evidence-linked registry cover 536 tools across 20 MCP servers. All 738 rule-case checks conform to declared expectations, while native annotation mapping differs from draft registry labels for 118 of 481 comparable tools. One AI-assisted reviewer approves 148 ratings, 75 matching the registry; this is not independent agreement. On 2,158 recorded runs, the implemented session guard gates the associated action in all 29 attributable attack-positive runs, compared with 18 for a read/write proxy, while adding nine approval points across 1,209 benign runs. Controlled file, Git, SQLite and three-browser experiments show that recovery depends on captured state, later changes, editor storage and recovery route, and expose a Git merge-state exception. The evidence supports making recovery assumptions explicit and testable. It does not establish registry-wide accuracy, reduced production harm or general recovery reliability.
agentai agent - arxiv:2611.22069 · cs.LGMedGuardBench: A Patient-Oriented Benchmark for Personalized Guardrail DevelopmentSara Khosravi, Xiaofei Wen, Majid Sarrafzadeh, Muhao Chen
Safety is personal: the same action can be safe for one person but unsafe for another. Yet current guardrails judge content without considering who is affected. Using medication safety, we introduce \textit{patient-conditioned medication guarding}, which predicts a safety verdict and 17 risk categories from a patient profile and prescription. We release \textbf{MedGuardBench}, a drug-disjoint benchmark of 5{,}001 scenarios across 87 medications. We find that standard guardrails rely on surface cues, such as wording and profile presence, rather than patient facts. To address this, we develop \textit{personalized reasoning guardrails} with three components. \textbf{Personalized Reasoning Distillation} links risks to relevant patient fields. \textbf{Context Unrolling (CU)} makes models check each field themselves, reducing false alarms on most backbones; controls show that providing the check does not yield the same benefit. A \textbf{rule-based reward (guided GRPO)} improves unsafe recall and complex-risk detection, including allergies and adverse reactions, on some backbones, but increases false alarms on others. Our models outperform base models and matched no-profile controls in accuracy, while reducing false alarms relative to no-profile controls. Profile removal and swapping confirm that gains come from patient information, not fine-tuning alone.
benchmark - arxiv:2611.22067 · cs.LGAddressing Exogenous Variability in Cooperative Multi-Agent Reinforcement LearningSeongmin Kim, Woohyeon Byeon, Jiwon Jeon, Seungyul Han +1
Cooperative multi-agent reinforcement learning (MARL) often fails under exogenous variability such as shifts in opponents or environment regimes that cannot be controlled by the team but alter the transition dynamics. We formalize this challenge as Exogenous Dec-POMDP (ED-POMDP), in which an episode-level exogenous regime, selected independently of the team, indexes the dynamics, and the global state is partitioned into components that are directly responsive to the current team action and components that are not directly action-responsive at the current step. This formulation exposes mediated influence: while team actions do not directly update the latter at the current step, they can shape their later evolution through induced changes in the former. Based on this, we propose LEICA, a CTDE-compatible algorithm that learns history-conditioned context representations for both components and shapes policy updates using agent-wise, influence-weighted counterfactual intrinsic rewards. Across SMAX benchmarks with opponent strategy shifts, LEICA consistently improves both training performance and generalization to unseen opponents' strategies over existing baselines, supporting the usefulness of influence-weighted shaping under exogenous train-test regime shifts.
multi-agentbenchmark - arxiv:2611.22057 · physics.app-phQCap: a physics-corrected graph neural network for the capacitance matrix of any planar superconducting device layoutSadman Ahmed Shanto, Saikat Das, Eli M. Levenson-Falk
Electromagnetic simulations of device layouts to extract capacitance are computationally intensive, a major obstacle to accurate design of circuit devices. Machine learning models can bypass these simulations, but previous approaches have been restricted to layouts using only particular components, and require thousands of simulations of those same components as training data. We present QCap, a model that accurately predicts the Maxwell capacitance matrix of an arbitrary superconducting device layout in seconds, compared to minutes to hours of Ansys Q3D simulations. We develop a novel design-embedding algorithm that converts the layout into a universal graph format, allowing the model to work with any layout created with any tool. To achieve high accuracy, we obtain a coarse physics prior from a fast method of moments simulation of capacitance, then train the model to correct this prior. On 4,009 test designs withheld from training data the model outputs agree with Q3D simulations to a median error of 0.177%, and on never-seen component families it stays below 1% error. We demonstrate the model's scaling by obtaining the capacitance matrix of a 9-superconducting-qubit device in 12 seconds on a laptop (compared to 19.9 h and 228 GB of memory in Ansys) with sub-percent error, and capacitance of a (too large to simulate) 100-qubit, 781-conductor device in 3.4 min on a laptop. We provide the open-source models and data on SQuADDS and Hugging Face. Our results provide a powerful tool for large-scale accurate circuit design and guide the development of models to bypass other physical simulations.
memory - arxiv:2611.22055 · cs.LGWeight-Space Programming Doesn't Need TransformersHyunjong Ok, Jaeho Lee
Standard neural architectures such as the Transformer were chosen because they train well with gradient descent and run fast on GPUs. In weight-space programming, a large language model (LLM) writes target weights by hand without gradient-based training, so those reasons no longer apply. Should we keep the Transformer? Given minimal guidance, ten labeled examples, and a sandbox, mainly on multiple-choice questions, we task three frontier LLMs with either generating the weights of a fixed architecture or designing the architecture themselves. When free to design, all three compile plain-text facts into a sparse table queried by a fixed procedure, reaching up to 54.4% test accuracy. An LLM-designed model achieves higher test accuracy than the most accurate released Transformer that the same LLM wrote while using about 1,800x fewer parameters. Without gradients, they choose a table, in which the LLM can state what each weight means and can find and rewrite a wrong one. Each wrong answer points to one row, and removing a row changes about one other answer. From this, we propose an associative-memory-based model up to 17x faster on GPUs than a same-sized Transformer. We also test pretrained weights, other benchmarks, and tasks where models generate answers. Pretrained LLMs remain more accurate; the claim concerns which structure to build when writing weights.
benchmark - arxiv:2611.22054 · cs.LGA single inference process underlies the behavior and neural circuits of working memoryDongyu Gong, Mario Belledonne, Ilker Yildirim
Working memory holds past inputs online for future behavior. Despite decades of research, accounts of working memory remain divided across levels of analysis, with long-standing debates on whether its capacity reflects discrete slots or continuous resources, how attention improves recall, and whether its representations rely on persistent neural firing or activity-silent states. Here we show that MemGP, a theory that is specified at the computational, algorithmic, and circuit levels of analysis simultaneously, can explain diverse behavioral and neural phenomena in working memory. At the computational level, MemGP casts working memory as probabilistic inference over a continuous function on feature space, with a variational objective that balances memory strength against a Gaussian-process prior. At the algorithmic level, writing, storing, and reading of memories correspond to approximating this posterior with a finite set of inducing points, the resource that all items share. In simulations, MemGP reproduces load-dependent precision loss, attentional benefits during maintenance, and inter-item attraction and repulsion, in each case revealing multiple complex computational dependencies. At the circuit level, the algorithm analytically (without training) maps onto a recurrent circuit, including its architecture and dynamics. Beyond reproducing neural phenomena such as delay-dependent drift, this analytical map deduces center--surround connectivity, a property long assumed in attractor models, and a structured combination of persistent activity and synaptic weights to hold memories. Thus, a single inference objective not only jointly predicts the behavioral and neural signatures of working memory, but also addresses, across several instances and different levels of analysis, existing competing explanations.
memory - arxiv:2611.22052 · cs.LGRiDEx: Reward-Informed Directional Exploration for Diffusion Reinforcement LearningSong Yan, Chenfeng Wang, Wei Zhai, Xingchen Wang +5
Reward-based diffusion post-training updates model parameters from task rewards. While existing approaches improve sampling and policy optimization, they still rely largely on isotropic random perturbations in a high-dimensional latent space for exploration. Such variation often spends the perturbation budget on changes with little influence on the target reward, making exploration inefficient. We study where diffusion models should explore and how to learn from that exploration. We introduce Reward-Informed Directional Exploration (RiDEx), which concentrates local exploration in reward-relevant subspaces of the denoising velocity field. Conditional residuals or differentiable reward gradients propose the subspace, and equal-and-opposite perturbations at the same generation state measure its terminal reward response. The reward difference supplies signed local supervision in the explored coordinates, while all generated images contribute to global DiffusionNFT learning. RiDEx thereby couples exploration-space selection with feedback about the changes it produces. On Stable Diffusion 3.5 Medium, RiDEx attains the highest peak rewards among the evaluated methods on four alignment and preference tasks. Its gradient-direction instantiation achieves 15x and 29x target-attainment speedups over DiffusionNFT on HPSv2 and PickScore, measured in H200 GPU hours. Controlled experiments confirm that reward-aligned directions elicit stronger responses at equal perturbation magnitude, and that applying their measured credit improves curve-average reward across all four tasks.
post-training - arxiv:2611.22050 · cs.LGDSADiff: Data Structure-Aware Diffusion for Crystal Structure Prediction via Mitigating Complexity- and Imbalance-Induced BiasesJiang Xiaobo, Dinghong Lai
Crystal Structure Prediction (CSP) is an important step in accelerating the generation and discovery of new materials. In recent years, diffusion-based CSP methods have made significant progress. However, existing models still suffer from generation bias when facing diverse materials spaces, while systematic approaches for tracing the sources of such performance variations remain limited. To address this issue, we propose LAS (Layered Analysis Strategy), a mechanism analysis framework that identifies crystal complexity and elemental composition imbalance as two key factors affecting generation difficulty and further reveals their relationships with model generation performance. Building on these findings, we further propose DSADiff (Data Structure-Aware Diffusion), a data structure-aware diffusion model that transforms the key factors identified by LAS into a unified generation difficulty representation and uses it to jointly drive feature modeling and training strategy optimization. This enables the model to adapt its representation and generation capability to samples with different data structures and difficulty levels, thereby mitigating learning bias across samples. Experiments on three CSP benchmark datasets, including Carbon-24, demonstrate that DSADiff consistently improves both Match Rate and RMSE, with a maximum Match Rate improvement of 9%, surpassing existing state-of-the-art methods and achieving new state-of-the-art performance. This work deepens the understanding of diffusion-based CSP models from the perspective of generation bias and provides a new direction for designing less biased generative models for complex materials spaces.
benchmark - arxiv:2611.22045 · cs.CLBeyond Acuity Accuracy: A Multidimensional Framework for Evaluating Patient-Facing Emergency Triage AI ToolsRuio Heerschap, Henry Foote, Mahmoud Alwakeel, Zhouhan Qian +18
Patient-facing AI triage tools may support care-seeking and emergency department prioritization, but require evaluation beyond acuity accuracy alone. We developed SAFE-Triage, a framework integrating human review, automated metrics, LLM-as-judge evaluation, and simulation testing across clinical performance, patient interaction, and clinician-facing summary quality. We applied SAFE-Triage to 40 MIMIC-IV-derived emergency vignettes and a 10,030-case acuity analysis comparing prompted LLM triage (PLT) with rule-augmented prompted LLM triage (rPLT). PLT achieved higher emergency severity index (ESI) accuracy (57.5% vs 42.8%, $P < 0.001$), lower undertriage (13.1% vs 29.8%, $P < 0.001$), and higher clinician ratings across all domains. rPLT produced shorter and more stable outputs with higher automated readability scores. SAFE-Triage identified trade-offs between acuity accuracy, undertriage risk, communication quality, readability, and summary quality, supporting pre-deployment refinement and ongoing monitoring by showing where triage tools are efficient, need improvement, or need clinician oversight.
llm-as-judge - arxiv:2611.22039 · cs.LGRLDE: A Reinforcement Learning Framework for Interpretable Iterative Document EditingNikhil Parkar, M. Marimuthu, Riya Joshi, Karthikeyan N
Improving the quality, clarity, and flow of long documents remains a challenging problem in NLP, as human editing typically occurs through iterative refinement rather than a single-pass rewrite. Most existing automated systems act as black-box generators or rely on supervised learning, limiting interpretability and failing to capture the sequential nature of revisions. In this paper, we propose RLDE, an automated end-to-end metric-driven RL framework that formulates document revision as a reference-free MDP problem (unsupervised with respect to paired document-revision targets). Working with a limited action space (NOOP, INSERT, DELETE, REPLACE), RLDE leverages Pointer Networks for dynamic span selection at the token level and sends generative actions to the frozen BART-base model to obtain context-aware suggestions. PPO agents learn to optimize a multi-objective reward function based on readability, perplexity, and grammar metrics with the help of a novel SBERT-based Semantic Preservation Penalty constraint to avoid deletion degeneracy. Evaluations on the BEA-2019 dataset showed statistically significant results (p < 0.01) across all metrics, where Flesch Reading Ease increased from 49.8 to 59.4, perplexity decreased from 63.7 to 39.8, while preserving 93% of semantic information. When compared to Chain-of-Thought (CoT) prompt-engineered models like GPT-4 and LLaMA-2, RLDE shows high performance at a very low computational inference cost. Moreover, the quality of edits generated by RLDE has been verified via human evaluations, scoring 4.9/5 for meaning preservation and 4.4/5 for edit quality.
iterative refinement - arxiv:2611.22033 · cs.AIThinking Harder Doesn't Lead to Better Choices: Reasoning Effort Increases the Confidence but Not the Correctness of LLM Shopping AgentsAdi Krishnamoorthy, X. Y. Han
Large language models (LLMs) are increasingly deployed as autonomous shopping agents that process multimodal product listings (consisting of visual and text data) and recommend purchases on behalf of consumers. Naturally, one might assume that models with more thinking power would make higher-quality purchasing decisions. To test this belief, we ask two questions about such agents: does asking a model to think harder make it choose better products, and does the confidence it reports tell us the likelihood it may be wrong? We benchmark ten frontier models on three product-choice benchmarks built from WebShop, Amazon's SQID and Wayfair's WANDS platforms to see if LLMs with more thinking power actually make better-quality product purchase decisions. Across all three benchmarks, we found that increasing the thinking effort does not increase decision quality in a statistically meaningful way. The same change in thinking effort raises stated confidence at a statistically significant level, so extra thinking buys confidence rather than correctness. We also measure the calibration of these models (how well their reported confidence percentage matches their true correctness percentage) in these tasks and show that most models are not well-calibrated at every effort level. We also find that errors are strongly shared: when one model errs in a task, several others tend to err on that same task. Furthermore, on tasks where several models all err, a large number of them pick the same wrong product.
benchmark - arxiv:2611.22030 · cs.AIWhat Does an Agent Need? Bundle-Aware Agent Resource CompositionYunxiao Shi, Tingwei Chen, Dannong Xu, Ahad N. Zehmakan +3
An agent resource's value depends on its backbone and companion resources and is known only after execution. Token-based generators remain tied to their training catalog, while independent resource scoring misses complementarities and reused bundles target other tasks. We formulate Task-Conditioned Agent Resource Composition (AgentComposition): jointly selecting a backbone model and a tool-skill bundle to maximize end-to-end execution utility. Our Retrieval-Grounded Bundle-Aware Agent Generation retrieves candidate resources and past configurations of similar tasks. A fine-tuned generator reads resource descriptions and recombines candidates through constrained beam search, allowing new resources without retraining. A bundle-aware critic ranks complete configurations using a large language model (LLM) teacher and historically successful configurations as execution proxies, and adapts to execution feedback when available. We evaluate reference recovery, offline execution with judged and native rewards, and online recommendation with real execution, building AgentSelect-Exec with 5,060 executed configurations. Our method recovers the complete reference configuration within ten candidates for 67.73% of AgentSelect queries, nearly twice the strongest baseline, and leads all baselines on SkillsBench. The proxy-trained critic raises the top-ranked agent's execution score from 7.58 to 8.37 out of 10; execution feedback increases it to 8.78, above the strongest baselines. Four target-domain queries suffice to adapt the generator to SkillsBench. Agents recommended online outperform the strongest baselines under both an LLM judge and native verifiers. These results support learning and transferring joint resource composition from task experience.
agent - arxiv:2611.22027 · cs.AIGauntletScore: Auditable Fact-Checking That Matches the Best Published Checkers on LLM-AggreFactErnie Hobbs, Jim Hinson
Checking whether a generated sentence is supported by its source document is a standard safeguard for AI-written text, and the strongest published checkers are single models. We describe GauntletScore, an evidence engine whose few rules and thresholds were set on the benchmark's development split. It puts typed questions to independent readers from different providers and records each answer as a measurement in an evidence ledger. It combines the measurements with a fixed Bayesian rule that counts a contradiction only when two readers agree. On the full LLM-AggreFact test split (29,320 claims across 11 datasets), following the leaderboard's approach, GauntletScore achieves 79.51 balanced accuracy, above every average published on the leaderboard as of 15 November 2026. That places it level with the best of them, nimble-v4-max (79.45 from its published dataset scores; the engine's 95% interval, 78.47 to 80.53, contains it). On the same claims it beats two of its three readers used alone (Readers A and B) by 1.66 and 2.17 points (95% intervals 0.80 to 2.53 and 1.29 to 3.06). Without AggreFact-CNN, described below, the leads are 0.90 and 1.32 and both intervals stay above zero. The engine also records the evidence behind every verdict, which a single model does not. Its largest margin is on full-length news articles checked against multi-sentence summaries (AggreFact-CNN), a format that resembles much regulated writing. There it scores 76.45, the best on the leaderboard, against 70.1 for the next best and 63.9 to 67.8 for the six systems with the highest averages. The test split was run once, on an engine frozen before the first test claim, under a stopping plan committed in advance. We release per-claim predictions and the scoring script.
benchmarkleaderboard - arxiv:2611.22017 · cs.LGSkillBuffBench: Measuring What Agent Skills Contribute across Harnesses, Models, Skill Access Modes and MemoryGengyue Dong, Ye Wei, Zheng Zhu, Yidong Han +3
Agent Skills, file-based packages of procedural knowledge that a coding agent can discover and load on demand, are now shipped by every major agent harness, yet the evidence for how much they help comes almost entirely from settings in which the right Skill is hand-picked for the task and the agent starts from a blank slate. Neither holds in deployment, where an agent chooses from whatever library happens to be installed and frequently carries memory from earlier work. We present SkillBuffBench, a benchmark and paired evaluation protocol that treats Skill access as a three-level factor (no Skill, a curated Skill, and automatic invocation from an installed library), crosses it with memory (off and on), with four agent harnesses and with multiple models (the state-of-the-art proprietary models from Anthropic, Google and OpenAI, and the state-of-the-art open-weight models from Chinese providers), every model being run under every harness, and evaluates the result on research tasks contributed by domain experts together with the Skills they use, currently in materials science and biology, with the suite extended on a rolling basis. Every run is marked by a deterministic verifier; uncertainty is reported with a paired, task-level bootstrap; and cost, agent rounds and wall-clock time are reported alongside the score rather than folded into it. This version of the paper describes the design of the benchmark, its evaluation protocol and its current tasks and Skills. The evaluation is in progress: results will be added once the runs are complete, and the task suite, results and analysis will continue to be updated on a rolling basis.
memoryagentbenchmarkevaluation protocol - arxiv:2611.22016 · eess.SYAn Explainable Agentic AI Framework for Distribution System Load RestorationS. M. Monzurul Hoque Chowdhury, Qinghua Li
Distribution system load restoration is a sequential decision-making problem in which restoration actions must maximize served load, prioritize critical loads, and preserve power-flow feasibility. This paper proposes an explainable agentic artificial intelligence (AI) framework for distribution system load restoration. The proposed framework uses multiple large language model (LLM) agents for planner guidance, strategic plan generation, decision arbitration, memory-based feedback, and explanation generation. Unlike unconstrained LLM-based control, the proposed method restricts agent decisions to feasible restoration actions and evaluates all candidate plans using deterministic AC power-flow feasibility checks. The framework is evaluated on IEEE 33-bus and 69-bus distribution feeders, as well as a larger IEEE 118-bus test case, and compared against reinforcement learning and planning baselines. The results show that the proposed agentic framework achieves better restoration performance while producing structured explanations for operator-facing decision support. A direct comparison with a Single-LLM-Agent baseline further shows that the proposed multi-agent decomposition improves weighted restored load and critical-load recovery compared with a monolithic LLM decision structure.
agentmulti-agentagentic - arxiv:2611.22012 · eess.SYPS-HalluBench: Taxonomy-Driven Benchmarking of LLM Hallucinations in Power SystemsS. M. Monzurul Hoque Chowdhury, Nabila Tasfiha Rahman, Qinghua Li
Large language models (LLMs) are increasingly explored for power-system applications, yet their reliability in engineering analysis remains uncertain. We introduce PS-HalluBench, a manually constructed benchmark of 200 questions that evaluates hallucination-related failures across five power-system task categories: numerical reasoning, network topology, constraint verification, false-premise acceptance, and abstention. We evaluate ten LLMs over the benchmark, yielding 39.35\% overall correctness, with constraint verification achieving only 21.75\%, and highlighting persistent difficulties in engineering reasoning. We further investigate engineering-grounded reasoning through numerical solver support and verified network information both of which show significant performance improvement. The findings reveal important limitations of standalone LLM reasoning and motivate integrating LLMs with verified engineering tools for reliable power-system analysis.
benchmark - arxiv:2611.22010 · cs.LGOn Formal Counterfactual ExplanationsAlexey Ignatiev, Peter J. Stuckey, Joao Marques-Silva
Counterfactual explanations are often seen as counterfactual conditionals under closest-world semantics, requiring one to reason over all minimally distant counterexamples. In contrast, formal XAI typically relies on feature-based abductive explanations (AXps) and contrastive explanations (CXps), which admit efficient computation and a minimal hitting-set duality. This paper bridges these perspectives by (i) characterizing closest-world counterfactual explanations explicitly as a disjunctive normal form over all minimally distant counterexample instances, and (ii) introducing conjunctive counterfactual explanations (CFXps), a subclass of counterfactual antecedents restricted to conjunctions of feature literals that guarantee prediction change. The paper establishes structural results connecting CXps, AXps and CFXps: every CXp induces an AXp for its witness instance, and every CXp together with a witness AXp induces a CFXp whose support is precisely determined by that AXp. These theoretical results yield a modular approach for computing CFXps using existing AXp/CXp machinery. Experiments on a representative set of benchmarks using tree ensembles and neural networks show that CFXp extraction is practically efficient. Compared with popular point-based counterfactual explainers, CFXps generally require no more feature changes and can guarantee the counterfactual prediction beyond a single instance.
benchmark - arxiv:2611.22009 · cs.AIDiagTrace: An Interactive Benchmark for Clinical Decision MakingYufan Wang, Rui Yang, Yi Lin, Chunhua Weng +1
Large language models (LLMs) have shown strong performance on medical question answering and diagnostic reasoning benchmarks. However, existing evaluations typically provide all patient information upfront and assess only the final diagnosis or rationale. This setup ignores the need to acquire diagnostic evidence incrementally from incomplete information in real-world diagnosis. To bridge this gap, we introduce DiagTrace, a source-grounded benchmark for evaluating whether LLMs can reconstruct the documented diagnostic trajectory through interactive information seeking. Each benchmark instance is built from a PubMed Central case report and includes an initial presentation, an event bank, a clinical course, and diagnostic references. DiagTrace contains 510 cases across eight specialties, with 225 used to evaluate four LLMs on clinical-course alignment, interaction quality, and efficiency. Results reveal trade-offs in coverage, source-consistent ordering, and turn use.
benchmark - arxiv:2611.22001 · cs.RODreamFAR: Real-Time and Long-Horizon Interaction with Diverse ObjectsYiming Dou, Botao He, Antonio Loquercio, Andrew Owens +3
World models have shown great potential in serving as a simulator for robot manipulation. However, existing models are either fit to a single environment or lack ability to simulate long-horizon contact-rich events. We present DreamFAR, a world model whose dynamics model is trained from scratch on a large robot-native teleoperation dataset, with no internet-video pretraining. We present an architecture and learning recipe designed for training on robot-native data. DreamFAR is significantly smaller than the high-performance video generation models, yet shows better understanding of contact and causality, and runs 10 FPS on a commercial GPU. Experimentally, DreamFAR significantly outperforms existing models in predicting the long-horizon future. We further demonstrate various applications where DreamFAR can be used, including policy evaluation and mixed-data co-training. Project page: https://www.yimingdou.com/DreamFAR/
manipulationteleoperationworld modelpolicy evaluation - arxiv:2611.21998 · cs.AIKV-LTM: Persistent Per-Session KV Caches as Long-Term Memory for Personal AssistantsZiv Kfir, Ido Kfir
Personal assistants accumulate months of conversations, but today's long-term memory stacks put derivation or selection between a question and its evidence. In a system-level comparison on LongMemEval, two commercial assistants scored well below the same model reading the history offline [1], consistent with, but not isolating, content loss between write and read. We propose KV-LTM: a memory tier storing each session's raw text plus its KV cache under a fixed reader model, answering by restoring cached sessions and appending the question. KV-LTM is complementary to, not a replacement for, retrieval and indexing. Text is the memory of record; the cache is a recompute-free prefill accelerator, degrading gracefully to text. We specify the design: storage record, budget-derived retention, append-only read path, and bounded aggregation. We size it analytically: across pinned BF16 full-attention (GQA) and latent-cached MLA configurations, per-token state spans approximately 31K-328K bytes (the pinned KDA+MLA hybrid differs: approximately 8.1K bytes/token plus a fixed per-session recurrent term), giving approximately 0.34-3.6 TB per user-year at a declared 30K-token/day scenario, against approximately 49 MB of raw text; per-query cost is modelled as regime-conditional transfer, attention, decode, aggregation, and compute. We define an evaluation protocol needing no persistent-cache implementation for logical QA evaluation: per-session prefill emulates valid cache reuse under the same model/input-state contract and ideal arithmetic, with named arms, controlled baselines, and a preregistered reader-model gate; a separate minimal real-cache replay tests numerical equivalence and measures the cost axis, with outcomes not assumed. KV-LTM trades storage for avoided re-prefill compute and write-time fidelity, within a declared retention budget.
memoryevaluation protocol - arxiv:2611.21996 · cs.CVBeyond Appearance: Where Forest Visual Place Recognition Actually FailsBlake Matheson, Ben McShanag, Isabella Moffat, Michael Milford +4
Visual place recognition (VPR) in forest environments remains challenging because repetitive vegetation, and changes in illumination, season, and viewpoint can make distinct places appear similar and revisited places appear substantially different. In this paper, we investigate whether temporal, linguistic, structural, and geometric context can improve on an appearance-only descriptor, and what limits each representation when it does not. Using MegaLoc as the appearance-based baseline on the WildCross benchmark, we evaluate score-space and descriptor-space sequence matching, vision-language captions and reranking, scene-graph representations, and depth-derived stem graphs across same-direction revisits, extended routes, and reverse traversals. On a same-direction revisit, both sequence matching strategies reach similar Recall@1 (94.1% vs. 91.5%), but recall at 100% precision rises from 50.9% (single-frame MegaLoc) to 71.6% in score space and falls to 36.5% in descriptor space. The remaining context representations provide no consistent improvement over the appearance-based baseline. Same-place captions show limited place-specific discrimination, reranking is constrained by the initial candidate shortlist, and stem-graph matching remains weak even with LiDAR depth.
benchmark - arxiv:2611.21995 · cs.LGDeciding Without Pointing: A Case Study of Whether a Decision Model Localizes the Anomalies It DetectsHsiang-Hung Lee
Decision models can answer a typed question about an image with a probability per option, without generating text. In our case study of Clef-flash (post-trained Qwen3.5-9B backbone) on industrial inspection benchmarks, we ask whether it localizes the anomalies it detects. We obtained uneven results with zero-shot detection: per-category image AUROC 0.924 on MVTec AD and 0.668 on MVTec AD 2. On the former, the decision head's routing attention produces 0.391 mean per-image pixel AUROC, and hiding the image tokens from it leaves detection unchanged. The multimodal LLM's backbone attention does point at the defect, however it falls below a center prior when we change question. It falls below chance if we remove textual context in the question, partly due to backbone attention leaving the object, while the head's map hardly changes. The backbone's states carry the location when probed with normal images, on the test-split images, however, the decision depends on the defect region. A larger model, Clef, that comes with a post-trained Qwen3.8-27B backbone, gives no useful map in the decision head either. We found that it's possible to recover anomaly location by asking question specifically on an image region, retraining the head or reading the backbone, but none matches dedicated few-shot localizers.
benchmark - arxiv:2611.21993 · cs.CVDEFTS: Data-Efficient, Training-Free Few-Shot Learning for CLIP in the Text SubspaceYijin Zeng
Contrastive Language-Image Pre-training (CLIP) achieves impressive zero-shot image classification performance by comparing image embeddings against class-specific text embeddings. While recent few-shot methods leverage labeled training samples to improve upon this zero-shot baseline, they often predominantly rely on image embeddings and underutilize the semantic structure of the text embeddings. In this paper, we introduce DEFTS, a data-efficient, training-free few-shot classification method strictly in the text subspace. We first demonstrate empirically that while raw text embeddings can be suboptimal as direct classifiers, the linear subspace they span remains highly discriminative. DEFTS therefore proposes to reduce the image-text modality gap by projecting estimated image prototypes directly into this subspace. Importantly, by expressing these prototypes as linear combinations of text embeddings, DEFTS reduces a high-dimensional estimation problem to the derivation of just a few coefficients. These coefficients are then computed via a Bayesian posterior mean, using a prior naturally centered on the original text embedding coordinates. Extensive experiments across 12 image classification benchmarks demonstrate that DEFTS significantly improves upon existing training-free methods.
benchmark - arxiv:2611.21978 · cs.AIWebAgents: Unbundling Agent Creation, Hosting, and Execution for an Agent-Native WebDario Fenoglio, Martin Gjoreski, Marc Langheinrich, Pradyumna Chari +1
Large language models are moving from isolated assistants toward autonomous agents that can plan, use tools, and coordinate with others, motivating an emerging Agentic Web of independently developed agents and services. Realizing this vision requires agent behavior to move across creators, hosts, and runtimes without coupling publication to a particular execution environment or transferring authority over user resources. We introduce WebAgents, which unbundles these roles through a signed agent.pkg that carries portable workflows, skills, schemas, and resource requirements. An OpenClaw-based agent-native browser/runtime verifies the package, obtains authorization, and binds it at run time to user-controlled models, tools, permissions, and budgets, while hosts distribute the public program and may retain private services server-side. We demonstrate this contract with three unrelated workflows executed unchanged under local and API-backed models, including literature review with live services and a website-published travel agent using semantic tools and user-selected personal resources. In a controlled suite, WebAgents explicitly blocks all 716 injected contract violations while accepting all 295 matched benign controls, with median automated pre-execution processing below 0.5 s for all three packages. We conclude by outlining a path toward NANDA-based discovery and interoperable package ecosystems while retaining execution authority in the user's runtime.
agentautonomous agentagentic - arxiv:2611.21975 · cs.CVPathology Randomization for Synthetic Priors and Zero-Shot DetectionAlexandra G. Roberts, Tonmoy Hossain, Mark Olchanyi, Stefan Brandstätter +6
Hospital brain imaging archives hold a diverse array of pathology at sample sizes beyond prospective studies, particularly for underrepresented populations and diseases. To unlock uncurated clinical data and scale analysis, methods are needed to detect abnormalities across contrasts and resolutions. Synthetic data has provided robust, contrast-agnostic models for a variety of human neuroimaging tasks in healthy anatomy, but pathology detection and other tasks in clinical settings still require real patient data, contrast-specific models, or disease-specific generators. We propose PathoDraw, a generative model of brain MRI with pathology. A conditional distribution of lesions given the healthy anatomy is drawn into the label map before contrast-agnostic rendering, and a single network estimates healthy anatomy with exact signed intensity deviation of any input scan. The contrast-agnostic 3D UNet learns the healthy labels under the lesion and the signed deviation directly and identifies pathology via robust tissue statistics and learned features, despite no training on real pathology. We show zero-shot detection on public stroke, tumor, and white matter hyperintensity datasets as well as archived clinical data, spanning $T_1$w, FLAIR and $T_2$w at various resolutions under a recent unsupervised anomaly detection benchmark protocol. Our single model exceeds or performs comparably to contrast-specific anomaly detectors trained on real scans and keeps healthy anatomy intact. The same generator trains networks for super-resolution and for registration in the presence of pathology. Our model, robust across acquisition and disease, enables the use of uncurated, archived data for clinical analysis. Our code is available at https://github.com/agr78/Pathology-Randomizer.
benchmark - arxiv:2611.21974 · cs.LGBeyond Bitwidth: Deployment-Constrained Mixed-Precision Quantization for Multilingual LLM Inference on NVIDIA BlackwellHyojung Han
Low-bit LLM quantization is useful only if a model preserves service-relevant accuracy, executes through supported GPU kernels, and meets measured memory and latency limits. We study Qwen3-8B on NVIDIA B200 with vLLM, including INT2/INT3, GPTQ, NVFP4, and mixed-precision configurations. We verify native packed execution and precision on all 144 fused linear modules and evaluate Korean, English, numerical, and reasoning quality under preregistered gates. Packed INT2 and INT3 execute, but all seven tested INT2 grids degenerate on generation. Isolated module-restoration effects are strongly non-additive; Korean and reasoning sensitivity rankings disagree, and a promising INT4/INT8 development candidate fails independent final evaluation. A configuration with 16-bit activations, 98 NVFP4 weight-only modules, 46 INT8 modules, and an INT8 output head passes all gates on the untouched FINAL-81 set (5,168 items): 5.917 GiB of weight-loading memory (limit 5.98), Korean teacher-correct retention 0.904 (gate 0.90), and TPOT ratio 0.939 relative to BF16. It improves paired accuracy over NVFP4 W4A4 by +1.4 percentage points (Bonferroni-adjusted interval [+0.2, +2.6]), compared with +3.4 points on the selection set. This comparison changes both activation precision and INT8 protection; exploratory analyses suggest that the activation change contributes substantially, but the isolated allocation effect is not established. Final checkpoints were not retained, and bitwise rebuild reproducibility remains unverified.
memory - arxiv:2611.21964 · cs.LGSolver to System: LLM Reasoning Across Representations and Execution in Scientific SimulationEliza Berman, Rohan Subramanian, Shreya Jha, Lohit Jagarapu +2
Simulators based on partial differential equations (PDEs) are widely used to model physical systems, but generating and interpreting PDE solver code with large language models (LLMs) remains challenging. Scientific simulation code is a useful testbed for LLM scientific reasoning because (1) one system can be represented by code, equations, natural language, and numerical trajectories and (2) code can execute successfully while implementing invalid physics. We introduce Solver2System, a dataset of PDE solvers with multiple representations, controlled corruptions, and lexical perturbations, to evaluate whether LLMs can recover the underlying physical system through (1) cross-representation consistency and (2) static and agentic detection of physical invalidity. Across eight open-weight models, LLMs systematically favor some representations when evidence conflicts, and lexical perturbations to code alter interpretation of other unchanged representations. Execution improves physical-validity judgments, but weaker agents can judge correctly without identifying the causal defect. Thus, LLMs may reach correct judgments despite inconsistent representation reconciliation or causal misdiagnosis.
agentic - arxiv:2611.21956 · cs.AIEight in Ten LLMs Leak System-Prompt Secrets: Nevan, an Agentic Capture-the-Flag StudyAna Clara Zoppi Serpa, Anderson Rocha, Hélio Pedrini
We audit instruction-hierarchy enforcement across 363 text-in/text-out models from the OpenRouter catalog. The research question we answer with this experiment is which models disclose a protected secret embedded in the system prompt, and through which techniques. Each model guards a secret 16-character flag set in its system prompt, while an artificial intelligence (AI) agent (OpenCode, powered by DeepSeek V4.1 Flash), steered by a human researcher, tries to persuade it to emit the flag. The flag was exfiltrated from 296 of the 363 models, plus four models released after the initial catalog freeze, for 300 captures in total. Of these, 170 involved a non-English language pivot, most often French. Each capture shows that a model ranked a user request above its system instruction at least once. It does not show that the model provably resists untested attacks, nor that it would consistently leak semantically sensitive data, nor that it is vulnerable to harmful-content jailbreaks. For practitioners, we draw three lessons: do not store secrets in system prompts, enforce critical behavior outside the model with output validation and deterministic checks, and do not let a large language model (LLM) judge be the sole decider of consequential outcomes.
agentagentic - arxiv:2611.21950 · cs.AIDiscoveryAgent: A Framework for Scientist-centric Agentic AIOwen Queen, Ioana Ciuca, Emma Dann, Jean Somalwar +8
AI agents can assist scientific research, but balancing autonomous investigation with scientists' ability to inspect and redirect ongoing work remains a central design challenge. We present DiscoveryAgent, a domain-agnostic, scientist-centric system that combines autonomous investigation with scientist-directed hypothesis steering. Scientists select, refine, or replace hypotheses between research rounds, while audit agents interrogate checkpointed records to trace claims, reproduce analyses, and inform subsequent decisions. Across two predictive modeling tasks, DiscoveryAgent produced the best variable-star classifier among evaluated baselines and, with scientist steering, a protein variant-effect predictor surpassed by only one of 106 ProteinGym baselines. We develop a rigorous checkpointing system that enables highly capable audit agents, a scalable form of oversight for agentic research. Audit agents autonomously identify errors at higher rates than previous AI scientist agents, and scientists collaborating with audit agents identify 14x more valid errors versus scientists performing manual review. In three case studies led by domain experts across physical science, social science, and life science, we developed a multimodal photometric-redshift estimator that resolves complementary failures of images and photometry, achieving 0.971 AUROC for low-redshift classification; documented an 83% rise in Florida homeowners-insurance premiums from 2013 to 2023, with a growing, regressive burden; and identified CD4+ T-cell relay genes, which both receive and exert broad transcriptional regulation, enriched for autoimmune disease associations. Together, these results show how autonomous research can expand scientists' empirical reach while centering their authority over research direction and their ability to understand and challenge the evidence.
ai agentagentic - arxiv:2611.21945 · cs.ROBenchmarking Deep Learning Models for Wheelchair Configuration Classification in Accessible Automated Vehicle ApplicationsSamir Ahmed, Shaaz Saleem, Devin Dhami, Wesley Maia +5
Making automated vehicles accessible requires consideration of vulnerable road users, including wheelchair users. To offer a more tailored experience that supports independence for this group, vehicle design needs to accommodate the wide range of wheelchair configurations. However, wheelchairs are individualized, varying in frame geometry, propulsion, and visual appearance. Moreover, occupants or accessories may obscure visual cues. Existing mobility-aid datasets only label the presence of a wheelchair, rather than describing its specific configuration. To fill this gap, this paper introduces the Wheelchair Configuration Image Dataset (WCID), comprising 467 face-blurred images across 192 capture groups solely for evaluation. Each image is annotated with (i) a two-class label of manual or powered, (ii) a chair subtype, (iii) occupancy, viewpoint, setting, lighting, occlusion, propulsion, and control, with a 273-image validation split and 194-image test split. Three convolutional neural network backbones and a frozen vision-language probe are trained on an external image dataset and evaluated on WCID to measure cross-dataset generalization, while also testing robustness using Gaussian blur. Convolutional models do well on clean images, but performance drops under strong blur, leaving a gap before deployment. Code, split assignments, labels, trained weights, and the face-blurred images is available at: https://github.com/Mi3-Lab/MAID
benchmark - arxiv:2611.21938 · cs.ROHubEgo-100K: In-the-Wild Egocentric Video of Skilled Work for Robot LearningJavad Amirian, Armin Kiani, Loïc Schaeffer, Tim Sprecher +2
Robot policies trained by imitation generalize poorly outside the environments and objects seen in training. Published data-scaling studies find that generalization grows with the diversity of training environments and objects rather than with more demonstrations per environment. Robot data is costly to collect across many environments, whereas egocentric human video can be recorded in the wild at scale; existing egocentric video, however, mostly shows daily-life activity and contains little skilled work in real workplaces. We present HubEgo-100K, a dataset built for diversity. This version of the paper covers the first release, v1.0: 10,124 hours of 1080p egocentric video with synchronized 100 Hz inertial (IMU) data from one head-mounted phone camera. In v1.0, 12,227 contributors record their own unscripted work in 15,548 environments (each one contributor's own setting of one type, such as their kitchen or workshop) across 11 setting types, 30 work categories, 56 task types and 35 action families, with 83% of hours in workplace settings. The release comprises 72,754 episodes with machine-generated action and language annotations, in LeRobot and MCAP formats under CC BY 4.0. We measure diversity along each axis, show that inside v1.0, with episodes in random order, new contributors and words keep appearing, and relate the measurements to published evidence on six data properties tied to policy performance, including hand visibility and inertial sensing. We also document collection, curation, privacy measures and data format, and provide a datasheet. Later versions of this paper will add benchmarks and cover further releases toward 100,000 hours.
benchmark - arxiv:2611.21937 · cs.AISLED: Sensor SLicing from Edge to DatacenterAlan Subedi, Avhishek Biswas, Apala Pramanik, Amlan Balabantaray +4
Modern sensing systems generate large volumes of heterogeneous data streams from multiple sensor modalities, increasing reliance on datacenters for computation, artificial intelligence/superintelligence (AI/SI) inference, and AI/SI training. Each sensor stream may have different network and compute requirements depending on its type and data volume. However, current wireless systems typically allocate resources at the device level, leading to contention among co-located sensor streams, where bulk data transfers may starve latency-critical telemetry. To this end, this paper introduces Sensor Slicing (SL). SL provides deterministic, medium access control (MAC)-level isolation for individual sensors within a single user equipment (UE) in open radio access networks (O-RAN). Using agentic coding techniques, SL is implemented in two O-RAN-compliant packages, reducing development time by 400 times (2 h 40 min [OCUDU] vs. 6.5 weeks [OAI]), 228-fold fewer human prompts (4 vs. 913), and less than half the tokens (0.64B vs. 1.43B). An associated end-to-end framework, SLED, is implemented to establish per-sensor radio and network isolation across edge, campus, and remote high-performance computing (HPC) tiers. Over-the-air wireless experiments demonstrate that while conventional network slicing causes critical delays to increase more than 2,300-fold under high traffic load, SLED maintains stable latency. Scaling evaluations on the two RAN stacks show that throughput remains within 0.9% and 1.4% of expected allocations even as the number of sensor slices increases. The results demonstrate that SLED preserves latency-critical telemetry while effectively utilizing resources across the sensor-to-HPC pipeline. The SLED software and evaluation datasets are publicly available at https://github.com/UNL-CPN-Lab/SLED-Sensor-SLicing-from-Edge-to-Datacenter.
agentic - arxiv:2611.21930 · physics.app-phSpatiotemporal programming of shape memory polymers for tunable vibration responsesSamantha Averitt, Jay Sim, Ruike Renee Zhao
Stimuli-responsive materials offer a promising route to dynamically tailor material behavior, enabling adaptive functionalities in changing environments. However, simultaneously implementing spatial and temporal control in active material systems remains a fundamental challenge. Here, we present a photothermally programmable shape memory polymer (SMP) metamaterial platform that enables dynamic control of elastic wave propagation through reconfigurable spatiotemporal temperature distributions. By projecting user-defined light patterns onto carbon-doped SMP (C-SMP) surfaces, spatial thermomechanical property landscapes are generated, enabling a wide range of elastic wave functionalities, including tunable transmission spectra and programmable vibrational modes. By treating both space and time as design parameters, we access an expanded design space that allows a bulk C-SMP surface to function as a highly adaptive metamaterial platform. This work establishes spatiotemporal property modulation as a powerful framework for adaptive vibration control, tunable phononic systems, and other advanced dynamic applications.
memory - arxiv:2611.21925 · cs.AIDNeo: Your Parallel Draft Model Needs Better Target Feature EncodingJameson Sandler, Shane Bergsma, Mostafa Elhoushi, Bin Claire Zhang +4
Speculative-diffusion decoding uses a small parallel drafter model to generate b draft tokens in one forward pass, which then undergo parallel verification via a larger target model. Faster inference can be achieved when multiple draft tokens pass verification simultaneously. Recent parallel draft architectures (DFlash, DSpark) improve draft quality by conditioning on rich features computed by the target, through a mechanism known as feature encoding. However, standard feature encoding mechanisms have three limitations: (1) they are lossy, applying encoding projections that are destructive to the underlying target feature-space via a discovered effect known as residual mixing, (2) they are memory-less, with no record of previously processed proposals, and (3) they utilize features sampled uniformly from across the target layers, omitting important hidden states. To address these issues, we propose Mixture-of-Features Attention (MoFA), a lightweight feature-encoding approach equipped with a working memory that encodes previously rejected proposals. Leveraging MoFA, rejection memory, and a set of further enhancements, including 2-tap causal convolutions, prefix-only cross-attention, and non-uniform target layer selection, the final architecture, DNeo, achieves 1.3x faster generation speed than SoTA (DFlash-2) without impacting drafter footprint or generation quality. DNeo marks a new state of the art in high quality, fast LLM inference.
memory - arxiv:2611.21924 · cs.CVGridDriver: On-Board BEV-Based Language Planning and ReasoningKatharina Winter, William Engel, Zain A. Alishammat, Fabian B. Flohr
Bird's-Eye-View (BEV)-conditioned large language models (LLMs) enable language-interactive driving, but onboard deployment requires efficient perception and continuous planning alongside autoregressive language generation. We present GridDriver, an efficient BEV-conditioned LLM framework combining trajectory prediction, structured driving indicators, natural-language navigation, driving explanations, and scene question answering. Lightweight spatio-temporal BEV perception and spatially structured tokens provide scene context to a shared LLM, while frame-wise scheduling interleaves trajectory prediction with asynchronous language generation. GridDriver achieves competitive performance on public benchmarks, reaching 87.9% PDMS on NAVSIM and 57.6% accuracy on nuScenes-QA with a shared planning and QA checkpoint. Spatial representation analyses demonstrate improved geometric information retention and trajectory prediction compared with Q-Former compression. The system achieves 32.7 ms standalone planning latency and 9.4 plans/s in an onboard ROS 2 pipeline. We further evaluate planning on our own data recordings and demonstrate open-loop onboard operation on a research vehicle. Across public benchmarks and real-world evaluations, we identify strengths and limitations of the BEV-based approach.
benchmark - arxiv:2611.21923 · cs.LGWrong Input, Wrong Threshold: Re-evaluating Adversarial Facial PrivacyTunahan Parlayıcı, Erchan Aptoula, Yücel Saygın
Published face-privacy methods report protection above 90% against black-box recognition models. That number comes from an evaluation protocol with two independent defects, reported in a measure that cannot size either one. The input is wrong: methods published since 2022 feed the network the whole photograph, or a detector's box, where deployed systems detect, align to a landmark template and crop. Re-scoring the same files through a realistic front end costs a standard transfer attack 56% of its dodging protection and a budget-allocating attack 7%, with no image regenerated. The measure is wrong: that loss is a difference of success rates, so it saturates, moving by 8.2x over a budget sweep while the effect beneath it moves by 1.34x. We therefore report the paired cosine shift, in which both thresholds cancel algebraically, so the input defect holds without committing to a calibration. Measured that way it is 2.2x and 10.7x larger on three unseen recognizers, and at matched budget it reverses the L-infinity transfer ranking: the one surveyed method that aligns at evaluation gains 0.18 where MI-FGSM loses 0.05. The cause is a resampling mismatch, not perturbation lost outside the crop; optimising through a differentiable alignment operator largely removes it, adding +24 to +58 points on unseen recognizers and a commercial system, with a residual on one of them. The thresholds are wrong: the four constants inherited since 2021 hold the false-accept rate near its nominal 1% on aligned crops but admit 2.3-5.4% on whole images, with an opposite-signed bias across dodging and impersonation, the two tasks reported side by side. Differentiable alignment and gallery scoring are not new: the first generation of this literature used both, and we show what dropping them cost.
evaluation protocol - arxiv:2611.21920 · cs.LGKVUse: Reusable-KV User Memory for Efficient LLM RecommendersXin Guo, Qiang Zhang, Xiang Zhou, Arushi Rai +7
Memory-augmented recommenders built on large language models (LLMs) maintain a natural-language profile of each user. These profiles are updated by a user profile update path that revises the profile as new interactions arrive and are consumed by downstream task paths like ranking. Because these user profiles are stored as text in the user memory, every ranking LLM call that consumes them must splice them in the prompt and calculate the KV cache of them during the prefill phase. Given these ranking LLM calls are prefill heavy (long prompt but short decoded output) and repeatedly called, these repeated prefill of user profile text dominates the serving. Prefix caching cannot remove it, because these user profiles are preceded by different text in different calls. We present KVUse, which stores the user profiles as reusable key-value (KV) caches instead of text. KVUse trains lightweight LoRA writers by self-distillation, so that the user profile update call produces the reusable KV of the user profile while decoding it, and the LLM can read these reusable KV at new positions and behind new context without recomputation. KVUse also has a customized KV scheduler that uploads the stored reusable KV of the next batch while the current batch is computed, hiding the KV transfer from host memory behind GPU computation. On four InstructRec datasets with Llama-3.1-8B-Instruct, KVUse shows no statistically significant difference in ranking quality from full prefill over continuous memory update rounds. Under saturated load, KVUse raises the throughput of the downstream scoring path by 4.0x-4.5x and of the profile update path by 1.4x-1.9x.
memory - arxiv:2611.21919 · cs.ROALOHA Lightning: Learning Fast and Precise ManipulationJohn Hua Yao. Qi Wu, Yihuai Gao, Chelsea Finn, Zipeng Fu
Learning from human demonstrations has enabled robots to acquire a wide range of manipulation skills, but learned policies typically execute far slower than ordinary humans. This speed gap is mainly due to lack of an interface for collecting demonstration data at high speed, and the difficulty in training policies that can robustly execute high-speed motions. In this paper, we present ALOHA Lightning (Flash ALOHA), a system for learning fast and precise robotic manipulation. Our system uses kinesthetic teaching to intuitively collect near-human-speed demonstrations on a backdrivable bimanual platform, yielding natural and fast trajectories. We also present a learning pipeline that enables smooth high-speed execution through test-time action smoothing and aligns the visual data distribution between data collection and deployment with masking. Given 50 demonstrations for each task, ALOHA Lightning autonomously completes bimanual tasks such as ball tossing and catching, folding shorts, battery insertion, and bussing tables for over 80% success rates while close to human speed. Project website: https://flash-aloha.github.io/
manipulation - arxiv:2611.21917 · cs.LGLost in a Few Heads: Diagnosing and Repairing Block Selection in Sparse AttentionWenhe Zhang, Boxing Chen
Sparse attention models read a few blocks of keys for every query at a fixed budget and can lose much of their long-context retrieval accuracy at that budget. We ask where this loss is made. Using exact attention mass selection at the shipped budget as an oracle, we show that the selector under-ranks the needle rather than running out of budget. We find attention heads that miss the needle by running a calibration probe that teacher-forces the answer and reads two quantities in every head. The first is the needle attention that the selector forfeits, and the second is the share of the answer's log-probability that fixing the head's selection restores, integrated along the path from the shipped selector to full attention. Four checkpoints of the InfLLM-V2 family and NOSA-8B lose multi-key retrieval in the same three blind heads, which attend to the needle while a mean-pooled key summary dilutes its single decisive token. Full attention in those heads or a 4-bit scorer that keeps the shipped budget recovers most of the loss without training. On MiniCPM4.1-8B at 64K, either fix lifts multi-key retrieval accuracy from 54.1 to 97.5%, level with full attention in every head. In production serving engines, full attention in those heads keeps 84 to 89% of the time to first token (TTFT) that sparse attention saves at 32K to 128K context length. Beyond the InfLLM-V2 selector lineage, one unified rule flags attention heads above a threshold swept on each checkpoint until they close most of the gap to full attention on calibration prompts. A similar small set of blind attention heads carry the loss for the SeerAttention-R gates, NSA, and DashAttention, whereas many such heads exist in MoBA. On Qwen3-8B with the SeerAttention-R gates, full attention in 18 of 288 key-value heads lifts multi-key retrieval accuracy from 29.8 to 97.7%.
long-context - arxiv:2611.21914 · cs.LGA Temporal Algebraic Framework for Data Stream Evaluation Under Delayed Labels and Distribution ShiftsRonaldo C. Prati, Elaine Ribeiro de Faria
In data streams, labels often arrive late or only for some instances, class prevalence drifts, and rare classes vanish from short windows. Standard evaluation then confuses a degraded decision boundary with a change in prevalence, and is biased when labels are delayed or selected by the model. We extend an algebraic framework that obtains pooled (micro) and per-class (macro) measures from one confusion matrix to streams, weighting the counts in time by exponential fading or a sliding window; the update costs O(m^2) per step, whatever the stream length, and each pooled measure stays a weighted combination of the per-class ones. Delayed and selective feedback enter as an observation mask: until its label arrives, an instance is counted with the model's calibrated class probabilities, and the bias of this interim estimate is driven by their calibration error. Audited conventions report no value for a class without verified instances. We separate prior shift from concept drift by monitoring prior-free coordinates (recall and specificity) and attribute a change in a measure to the two sources with a first-order Jacobian decomposition. On synthetic streams and four real datasets, a detector on these coordinates with a memory per class raises no excess false alarms under label shift, unlike the Jacobian alarm and error-rate detectors. The correction for delayed labels recovers recall in some real settings and is only partial for precision, specificity and F1. Twenty-five statements are checked in Lean 4, and the code is in the aggmetrics package.
memory - arxiv:2611.21908 · cs.ROFrom Human Videos to Bimanual Dexterous Manipulation Data: Confidence-Guided RL Refinement of Video PriorsYanghong Liu, Yang Chen, Kailang Xu, Bang Du +7
Collecting demonstrations for bimanual dexterous manipulation is costly, motivating the use of human videos as a source of robot training data. However, uncertainty in video-derived motion estimates and differences between human and robot embodiments make direct motion transfer challenging. In this work, we introduce PriDex, a framework that transforms human-video Priors into executable robot Dexterous manipulation demonstrations through reinforcement learning in simulation. PriDex preserves task-relevant interaction structure while allowing embodiment-specific behavior to be regenerated for the robot. Human-object interactions provide affordance cues for selecting task-relevant, robot-specific grasps from geometry-based candidates. For motion refinement, object trajectories provide a task-level reference, while relative bimanual wrist trajectories offer complementary guidance for coordinated motion. Since object-motion estimates can become unreliable under occlusion or visual ambiguity, PriDex uses object-motion confidence to adaptively balance these two priors, emphasizing object-level guidance when confidence is high and increasing the contribution of wrist-level guidance as confidence decreases. Reinforcement learning refines these video-derived priors under the robot's kinematic and physical constraints, and the resulting policies are rolled out to generate robot demonstrations for training a vision-based diffusion policy. We evaluate PriDex on four bimanual dexterous manipulation tasks, including pouring, sweeping into a dustpan, bottle-cap unscrewing, and plate cleaning, assessing both simulation-based robot data generation and downstream visuomotor policy learning.
manipulationdexterousdiffusion policygrasp - arxiv:2611.21905 · cs.ROAgile Monoped Hopping on Complex and Discrete Terrain via Kinodynamic Planning with Distilled MPPI SteeringYue Qin, Denglin Cheng, Jingyu Liu, Yanran Ding
This paper presents FlowRRT, a kinodynamic planner for agile monoped hopping on discrete terrain. Steering for kinodynamic RRT on complex dynamical systems currently relies on motion primitives or closed-loop policies, limiting the diversity of motions or the planning speed. Distilling a trajectory optimizer avoids both, but existing methods regress a single solution per query, which fails on monoped hopping, where the feasible action set is narrow and disconnected. Our key insight is to treat steering as set-valued and predict the distribution of feasible actions instead. A multi-start MPPI expert collects the entire feasible set of each query offline, and conditional flow matching distills it into a generative sampler. With this sampler, FlowRRT coordinates successive jumps over long horizons and finds non-greedy maneuvers that are hard for reactive policies to discover. Benchmarks across different terrains show higher success rates and substantially shorter planning times than sampling-based baselines. On terrains with inclined platforms, FlowRRT reaches $80\%$ success within $1$~s, versus $24$~s for RL-RRT and $77$~s for SST. Sim-to-sim tracking in MuJoCo confirms accurate execution under full-order dynamics.
benchmark - arxiv:2611.21885 · cs.AILanguage Models Can Rewrite Their Own WeightsJiajun Fan, Neil He, Changyi Yang, Andy Mao +7
Language models are trained once, frozen and deployed: improving one again takes another training run, and a prompt, a scaffold or extra computation at inference is gone once the task ends. A frontier model has been shown to edit the weights of a much smaller one. We ask whether a model can rewrite its own weights to make itself better, and find that it can, by its own reasoning, with no gradients and no training; we call this self-rewriting. Given its weights, graded answers and the ability to run itself, a 27B model diagnoses its failure and edits the weights behind it. It surpasses itself: its own edit makes it 6.6 points more accurate on 490 problems it never saw, in a test fixed in advance, and another session's edit also makes it significantly better. It keeps getting better: it submits an edit only when its own tests support it, each edit replacing the last, and over its first eight such sessions its accuracy only rises. Unlike test-time scaling, the gain persists in its weights and adds no cost per answer; unlike training, it needs no pre- or post-training infrastructure, only an open model and hours on one GPU. It repairs a damaged copy of itself: it locates the corrupted matrix and rewrites it with values of its own, never reading the undamaged weights; found on text math problems alone, the repair lifts the copy on six image benchmarks it never saw from 7% to 70% accuracy, 83% of the undamaged model's. And the ability emerges with scale: across four sizes of one family, the 2B model makes no edit and the 4B and 9B models break themselves; only the 27B model works on itself as a human would, comparing its hidden states on answers that finish and on answers that run out of tokens, weakening the direction that separates them. Like a human, a model need not separate training from inference; improving a language model may come to take a conversation rather than a training run.
post-trainingbenchmark - arxiv:2611.21882 · cs.LGWM-Quant: Rollout-Aware Post-Training Quantization for JEPA-Style World ModelsWilliam Hammer, Andrew Heuer, Nikhil Krishna, Matthew Grenier +1
We present WM-Quant, to our knowledge, the first post-training quantization (PTQ) framework designed specifically for JEPA-style world models. Although world models typically feature relatively few parameters, action planning algorithms often require thousands of inferences, which can quickly become computationally expensive. PTQ is a natural way to reduce this cost. However, we show that in world models using autoregressive model predictive control (MPC) rollouts, quantization error can compound with each predictor step, resulting in action rankings from the quantized model becoming near-random. WM-Quant addresses this with three components that build on existing PTQ methods: early-encoder precision, control-path preservation, and rollout-aware scaling. Early-encoder precision protects accurate latent state and goal information to mitigate long-term error propagation. Control-path preservation allows the world model predictor to distinguish between fine-grained actions, significantly improving action rankings and execution. Rollout-aware scaling jointly calibrates encoder and per-step predictor activation ranges with respect to an entire rollout rather than a single prediction, reducing the cumulative rollout error. Across multiple world models, environments, and PTQ methods, WM-Quant substantially improves planning success at low bit-widths, approaching full-precision performance at W4A4.
world modelpost-training - arxiv:2611.21879 · cs.LGAdaptive Memory for Online Reinforcement Learning under Sparse Model ChangesSreejeet Maity, Sukanya Modak, Leonardo F. Toso
We study the discounted reinforcement learning (RL) problem in a setting where the underlying environment changes between deployments. Each deployment provides $H$ observations from a fixed MDP. The reward function and transition kernel are fixed within each deployment. Across deployments, at most an $ε>0$ fraction of the reward and transition kernel entries may change. In this setting, learning each model to deploy from scratch may discard useful observations from past deployments, whereas pooling all past observations can bias the estimates of entries that have already changed. We then propose an algorithm that for each state and action pair, selects how much history to reuse to estimate the reward and probabilities of the next state. We start with observations from the current deployment and add earlier deployment observations one at a time, the most recent first. After each addition, we check whether a single average reward agrees with the estimates from all histories considered until that point. We apply the same check to the distribution of next states. For each estimate, we leverage the last history that passes these aforementioned checks. The optimal control policy is then computed on top of the fitted MDP. We characterize the average gap between the optimal return and the return of the learner's policies over a finite number of deployments. We demonstrate that when $0< |S|ε\leq 2$, with $|S|$ being the cardinality of the state space, our algorithm needs only the observations from the most recent $O(1/ε)$ deployments, and achieves a long-run average gap of at most $\widetilde{O}(\sqrt{ε/H})$.
memory - arxiv:2611.21878 · cs.CLKnowing Which, Not How Much: Language Models Order Party Positions They Cannot PlaceYousef Attia, Jarosław A. Chudziak
Language models are increasingly asked where real actors stand: what a party's platform prioritises, what an institution has committed to. Such judgments are gradable, carrying both an order and a magnitude. Nearly every benchmark that tests them asks about one entity at a time, then reads that score as the model's knowledge of the domain. Order and magnitude, however, need not be acquired together. Text about social entities carries far more comparison than measurement, so a model may hold a dependable ordering with no anchored scale beneath it. We ask whether a model can place a documented social position on a scale or only order the actors it compares, and what that means for how such knowledge is measured. We test this on party positions in Japan and South Korea, where the Comparative Manifesto Project's coded data supplies a rare external answer key. Asked to order four parties, models beat the strongest fixed-answer baseline by more than a factor of two; asked to place one party on the same scale, they fall below it. The ordering survives inside those failed placements, and supplying the coded values lifts placement to near ceiling. Comparative and absolute questions about the same fact therefore measure different things here, and for party issue salience a benchmark built from one cannot speak for the other.
benchmark - arxiv:2611.21864 · physics.opticsInverse Design for Nonplanar Conformal LayersArjun Khurana, Michael J. Probst, Stephen E. Ralph
Conformal layering, where a deposited layer follows the nonplanar topography beneath it, arises from both deliberate passivation overlays and imperfect planarization between depositions. However, inverse design methods typically model planar, vertically extruded layers, resulting in geometric discrepancies between the modeled and fabricated structure. To address these discrepancies, we present differentiable mapping functions that incorporate both uniform conformal overlays and partially conformal interlayer dips into a standard inverse design paradigm, requiring no additional design parameters over planar structures. Using these mappings, we design (1) mode-selective GaAs grating reflectors with uniform conformal SiN passivation for high-aspect-ratio VCSELs, reaching differential modal reflectivities of 33.1% (single-etch) and 63.4% (dual-etch), and (2) a dual-layer silicon grating coupler whose upper layer dips up to 40 nm into the lower layer, reaching -0.75 dB coupling efficiency. Modeling conformal layering during optimization, rather than treating it as post-hoc fabrication error, permits direct exploitation of the conformal layering, yielding inherently fabricable and performant designs.
grating coupler - arxiv:2611.21852 · cs.LGThe Hidden Cost of MoE Diffusion Decoding: Profiling and Fixing the SamplerHarshit Samrat, Ramakrishna Paramahamsa Kompella
Masked diffusion language models (dLLMs) decode by iteratively denoising a full canvas of tokens, so every forward pass runs a per-token sampling and stopping pipeline over the entire canvas-by-vocabulary logits tensor. We profile single-user decoding of a 26B-parameter diffusion Mixture-of-Experts model on a bandwidth-bound edge accelerator (NVIDIA GB10) and report a result that is easy to miss: after weight quantisation, this logits pipeline, not the expert GEMMs, is the single largest consumer of decode time. Unlike autoregressive decoding, where sampling touches one position's logits per committed token, diffusion decoding processes S times as many logit entries per output token, where S is the number of denoising steps. We show this pipeline is largely redundant or fusable and remove it in two lossless tiers: a bit-exact tier whose output is byte-identical to the reference decoder, and an accuracy-parity tier using a fused single-pass entropy kernel and Gumbel-max sampling. Combined with a W4A16 expert kernel, the full stack reaches 2.4-2.7 times end-to-end speedup over bf16 at unchanged benchmark accuracy on GSM8K, HumanEval, and MATH, on a single 273 GB/s device. We confirm the finding on a second, independently built diffusion MoE (LLaDA2.0-mini: pipeline 28.8% of decode, fused W4A16 3.3 times at accuracy parity) and show it disappears on a dense diffusion model (3.6%). The cost is specific to MoE diffusion, where quantisation makes the backbone cheap. Code and quantised weights are released.
benchmark - arxiv:2611.21847 · cs.LGInventa: Decentralized Self-Improving Agents for Autonomous Scientific DiscoveryPrithwish Jana, Nikos Kanakaris, Purak Jain, Anoop Deoras +1
Large language model (LLM) agents can generate research ideas and improve scientific discovery through iterative experimentation. However, increasing the number of agents or iterations does not automatically translate to producing more diverse or feasible ideas. Agents may converge on similar candidates, propagate unsupported claims through shared memory components and repeatedly apply search strategies that no longer yield any improvements. We introduce Inventa, a framework for automated scientific discovery that employs autonomous research units and verified shared knowledge. Each unit creates a lineage of candidate ideas using a mutator agent. A local orchestrator agent periodically adapts the mutator's harness. When progress stalls, the orchestrator agent replaces its model or restarts the idea evolution process. Verifiers evaluate each candidate idea and record verified improvements, useful components and constraint violations in a shared knowledge, the only channel between units. We evaluate Inventa on five benchmarks spanning the research cycle, covering retrieval (ScholarCatalyst), ideation (IdeaBench), proposal writing (IdeaScientist), and scientific law discovery (EurekaBench, SciLaws-Bench). It leads all five, achieving 73.6% on EurekaBench versus 56.2% for CORAL. On IdeaBench, it discovers 26.0 novel ideas per paper, compared with 18.6 for ResearchAgent, with 82% fewer model calls.
memoryagentself-improvingbenchmark - arxiv:2611.21846 · cs.CVCan a World Model Predict a Deepfake Video? A Consistency Margin Prior for Deepfake DetectionPradip Gupta, Shubham Mandloi, Ravi Kumar Satzoda, Sujeeth Bharadwaj +2
Deepfake detectors often perform well on the dataset they are trained on, but their performance drops when tested on videos from different datasets. In this paper, we study whether a predictive video model can provide a signal that generalizes better across datasets. Instead of relying only on generator-specific artifacts, as many existing detectors do, we use a model trained to predict natural video and measure the difference between what it predicts and what it observes in a video. Using V-JEPA 2.1, we keep the encoder frozen and adapt the predictor using rank-16 LoRA. We define the difference between the predictor output and the frozen encoder target as the prediction residual. We hypothesize that manipulated clips produce, on average, larger prediction residuals than authentic clips, and use this as our consistency-margin prior. We process these residuals using a lightweight Latent Consistency Head (LCH), which separates residual magnitude from direction for training and classification. To test whether the direction of the consistency margin affects cross-dataset generalization, we compare standard and reversed margin objectives while keeping the remaining training setup fixed. We match the two settings on FF++ validation performance before evaluating them on unseen data. The reversed-margin setting exhibits a larger cross dataset generalization gap on average, providing evidence that the direction of the consistency margin affects cross-dataset transfer. Our complete model adds only 8.25M trainable parameters, which is 2.47% of the total 334M parameters. Trained only on FF++, it achieves an AUC of 90.5 on Celeb-DFv2 and performs well across several other cross-dataset benchmarks. Our ablations show that separating residual magnitude from direction, reducing dataset-specific variation, and using a low-rank projection improve cross-dataset generalization in our experiments.
world modelv-jepabenchmark - arxiv:2611.21845 · cs.AIEmbedGym: Letting Coding Agents Develop and Debug Embedded Firmware Without Physical HardwareMinghui Zhao, Yiming Li, Sophie Zhu, Maxwell Khor +4
Modern coding agents perform well on general software benchmarks because they can run code and unit tests in a closed loop. Embedded firmware development, however, breaks this loop: a program that compiles can still fail on real hardware due to pin polarity, bus protocol, timing, or networking bugs that an agent at a text terminal cannot observe. Existing embedded coding agents either skip hardware execution, drive GPIO-only simulators with multi-second delays, or test on physical boards in a static environment. In this paper, we present EmbedGym, an interactive virtual hardware environment that lets coding agents develop and debug embedded firmware without physical hardware. EmbedGym uses instruction-level microcontroller unit (MCU) emulators to execute the same binary that would be flashed onto a physical board, connects them to 41 peripheral models, an analog circuit simulator, and virtual WiFi/Bluetooth Low Energy (BLE) endpoints, and exposes an interface for agents to change sensor inputs, press buttons, and inspect hardware states. On top of it, we build EmbedGym-Bench, which consists of 598 tasks across 6 MCU-framework pairs graded by hidden timed test vectors, using a five-stage multi-agent workflow that synthesizes peripheral models and tasks from datasheets and driver libraries. We expose EmbedGym to a coding agent at three levels and evaluate eight models over more than 14K trajectories. Even when compilation succeeds 99.7% of the time, 19.9% of compiled firmwares fail at runtime (up to 73% on individual targets). In contrast, interactive hardware testing in EmbedGym increases pass rates across models by +4.9% to +14.7% (up to +31% on individual targets). Our results also show that an agent with EmbedGym outperforms prior embedded agent frameworks, and lightweight models with EmbedGym match frontier models without it at 30-70x lower cost.
agentmulti-agentagent frameworkbenchmark - arxiv:2611.21828 · cs.LGCoordination of Heterogeneous Teams with Uncertain Task RequirementsJongann Lee, Melkior Ornik
Coordinating a team of heterogeneous agents to complete tasks with unknown requirements is a challenging multi-agent planning problem. Agents differ in their ability to discover task requirements and in their ability to engage tasks, so the planner must coordinate task discovery, execution, and staging. This paper introduces a hierarchical planner that uses a graph neural network policy to assign actions to agents, then finds a path to the action's location using a shortest-path algorithm. The central part of our approach is the CAAT graph, which is a graph representation of the planning context consisting of the agents, tasks, capabilities, and actions. We demonstrate that our learned planner can jointly coordinate a heterogeneous team to simultaneously discover task requirements and complete tasks in a shorter time compared to baselines that optimize task discovery and task engagement independently. Furthermore, we show that our policy can zero-shot plan on out-of-distribution settings by training only on an outdoor map then transferring without finetuning to an indoor map with more agents, tasks, and capabilities than seen during training.
multi-agent - arxiv:2611.21827 · cs.AIHow Much of SAR Vehicle Recognition Is the Vehicle? Background Shortcuts, Angle Gaps and Partial Libraries on MSTAR and ATRNet-STARMousa Shaya
Deep networks score near 99% on the MSTAR vehicle benchmark, and the newer ATRNet-STAR benchmark was built to be harder. We ask how much of these scores comes from the vehicle itself. With a modern network (ConvNeXt-Small, 3 seeds) we first reproduce the known MSTAR result: with the central 96x96 pixels zeroed out, accuracy is still 73.3% (chance 10%). The same shortcut is present in ATRNet-STAR, which was collected across several scenes: with the vehicle removed, accuracy is 63.3% on the 40-class SOC split (chance 2.5%). Under the official scene-change split, the background stops helping and starts hurting. Accuracy is 22.8% on full chips and 30.8% when only the vehicle is shown. A SAR foundation model, SARATR-X, leans on the background more than ConvNeXt does (+4 to +6 points with the vehicle removed, in every seed) and does not solve the scene change. We then hold the amount of training data fixed and vary the depression angles used for training. Accuracy is set mainly by the gap to the nearest training angle: 57% at 15 degrees, 39% at 30 degrees and 26% at 45 degrees. Adding more training angles does not help. Finally, when most classes have training images at the test angle and a few do not, the missing classes do worse than when no class has them: 9.0% instead of 22.4% at a 45-degree gap. Most of their images are assigned to known classes. We recommend reporting background-only and target-only controls, and the angle gap, next to any accuracy figure. Code is public.
benchmark - arxiv:2611.21823 · cs.AITUSI: A Citation-Grounded Graph-Based BenchmarkMarzieh Karimi, Mehran Safayani, Alireza Basiri
Evaluating retrieval-augmented generation (RAG) systems in Persian law remains challenging because benchmarks lack realism, grounding, and coverage of legal reasoning. Existing resources are either synthetically generated from statutory texts, lacking natural information needs, or derived from informal legal question--answer sources without explicit linkage to authoritative legal documents, making them unsuitable for rigorous evaluation of grounded legal reasoning. To address this gap, we introduce TUSI-Bench, a structured Persian legal benchmark designed for the evaluation of RAG-based systems. TUSI-Bench is built upon a multi-layered legal knowledge framework containing 7,261 legal articles and integrating codified statutes, legal commentaries, doctrinal writings, and Supreme Court unification rulings. We also construct a legal knowledge graph linking articles and concepts for structure-aware retrieval. Based on this foundation, TUSI-Bench provides two complementary evaluation tasks: (i) TUSI-QA, an open-ended question answering benchmark containing 1,999 questions that requires grounded responses with explicit legal citations, and (ii) TUSI-Bar, a multiple-choice benchmark containing approximately 500 four-option questions derived from Iranian Bar examination questions. Furthermore, we propose a graph-based retrieval baseline that leverages relational structure within the legal knowledge graph. Experimental results demonstrate that incorporating graph-based relations improves retrieval recall by up to 8 percentage points compared to standard baselines, highlighting the importance of structured legal representations for RAG systems. The proposed benchmark provides a standardized resource for evaluating grounded legal reasoning in Persian and enables future research on structured legal intelligence.
retrieval-augmentedragknowledge graphbenchmark - arxiv:2611.21816 · cs.CVRethinking Perceptual Similarity for Remote Sensing: Multi-Layer Attention Scoring and Human Preference BenchmarkingEli Underwood, Marc Bosch, Neil Joshi, Jeffrey Wen +1
Perceptual similarity metrics for natural images transfer poorly to remote sensing (RS), where severe geospatial and semantic failures can remain visually plausible while still being decisively wrong for downstream use. To address this, we develop a novel human-aligned perceptual quality metric for generated RS imagery and evaluate it across several generation tasks. We first conduct a controlled perturbation study where we progressively increase the severity of common synthetic perturbations. Using the layer-wise Aligned Attention Score (AAS) responses in RemoteCLIP, we find that different RS artifacts are best captured at different depths. Based on these findings, we introduce Remote Sensing Perceptual Distance (RSPD), a pairwise metric that averages seven independently calibrated layer distances without tuning on human-preference labels. To evaluate our metric on realistic RS generation tasks, we construct a blind, reference-conditioned two-alternative forced-choice (2AFC) benchmark with five 300-triplet subsets spanning synthetic artifacts, airborne and spaceborne super-resolution, SAR-to-optical translation, and timeseries controls. We also construct a 300-triplet ordered misregistration benchmark with deterministic labels. Using the same RemoteCLIP checkpoint, RSPD substantially outperforms simple cosine distance with the final global embedding and is competitive with the strongest generic baselines across the realistic RS generation tasks.
benchmark - arxiv:2611.21812 · cs.CVSketch, Don't Paint: Allocating Test-Time Compute in Generative Visual ReasoningSheng Lu, Guangpin Tao, Tong Lu, Ying Tai
Multimodal models increasingly reason by synthesizing pixels, sketching maze paths within interleaved chains of thought or rolling out video solutions frame by frame. Each visual thought runs the full diffusion schedule inherited from photorealistic synthesis, so iterative denoising dominates inference cost. We study how to allocate test-time compute along two axes: when to render visual thoughts rather than thinking in text or answering directly, and how many denoising steps each thought requires. Controlled interventions that vary only how the thought is rendered uncover three regimes. Causal thoughts carry the plan and become legible early: an 8-step sketch nearly matches a 50-step render in accuracy, commits to the same plan as often as two independent full renders agree, and in video models mostly fails by stalling short of the goal, so the rollout abstains. Decorative thoughts can be omitted without loss, whereas harmful thoughts lower accuracy. Because the regime depends on both model and task, always drawing is suboptimal. Across 14 model-benchmark pairs of three interleaved reasoners, answering directly, without any thinking, wins 9 pairs, and choosing among direct, text-only, and visual tiers from ten labeled examples per pair raises mean accuracy from 60.7% (always drawing) to 65.1%. Where drawing pays, we propose sketch-and-vote, which samples 8-step sketches (a step count chosen without labels) from independent noises until one answer reaches three votes. With this one setting, sketch-and-vote beats full rendering with fewer denoising steps on an image reasoner and two video reasoners, by 8.6 points on VSP spatial planning and by 9.6 and 22.1 points on mazes.
benchmark - arxiv:2611.21809 · cs.AIEvaluating the Evaluation of Music Generation Models with Generative PsychometricsJoão Carlos Santana Silveira, Caio Von Rondow Morais, Frederico Pedrosa, Lucas N. Ferreira
Subjective user studies are one of the major methods for evaluating music generation models. However, subjective evaluation protocols vary widely and are rarely specified in enough detail to be reproduced, hindering comparison across studies. In this paper, we map subjective evaluation methods in ISMIR music generation papers from 2020-2025 and apply generative psychometrics to assess the stability of their items. We extract 284 items from the papers' methodological descriptions, using an LLM and manual validation to standardize them. We represent these items as contextual embeddings and apply Unique Variable Analysis to remove redundant, locally dependent items and Bootstrap Exploratory Graph Analysis to estimate their latent dimensions. Results show the network converges onto a taxonomy of six stable quality dimensions: Rhythm, Melody, Audio, Composition, Structure, and Lyrics, pointing to a unified methodology for conducting user studies to evaluate music generation models.
evaluation protocol - arxiv:2611.21804 · cs.CLCorrupting Ties: Investigating Robustness to Bias Injection Across Multi-Agent Network TopologiesCléa Chataigner, Mina Arzaghi, Florian Carichon, Pablo Piantanida +1
AI systems are increasingly deployed in high-stakes decision-making contexts, where interactions among components can introduce fairness risks that remain undetected when auditing components or collective outcomes in isolation. As LLM-based multi-agent systems become more prevalent, understanding how biased information propagates through agent interactions is essential for assessing their fairness. We introduce a controlled bias injection framework that replaces selected agent responses with predefined biased ones, enabling an intervention-based analysis of how network topology and injection configurations shape system behavior. We propose decision divergence metrics that quantify the effects of these interventions at the agent, network, and system levels, and evaluate five topologies on three binary classification tasks using LLMs from the Qwen and Gemma families. Sequential networks show the greatest divergence, whereas fully connected and mesh networks show less. Injection position also matters: in star networks, injecting the central agent has larger effects than injecting peripheral agents. Models further differ in whether divergence follows the injected bias. These findings highlight the role of network structure in the propagation of biased information and motivate topology-aware auditing of fairness risks in multi-agent systems.
agentmulti-agentagent system - arxiv:2611.21803 · eess.SYMechanism Design and Active Systems Theory: 60 Years of Parallel DevelopmentDmitry A. Novikov
The survey compares two research traditions within the mathematical theory of designing the rules of interaction (mechanisms) for purposeful participants: mechanism design, together with contract theory and implementation theory, and the theory of active systems created by V.N. Burkov and developed into the theory of control in organizations (TCO) by D.A. Novikov's school; the main results of Yu.B. Germeier's school and of the domestic schools adjacent to active systems theory (AST) are covered as well. The problem statements and principal results of 2006-2026 are presented: for mechanism design (MD) - robust and dynamic mechanism design, multidimensional screening, information design, simplicity of mechanisms and credibility of the designer, market design, algorithmic and learning-based design, mechanisms for AI agents, and robust, relational, dynamic and algorithmic contracts; for AST and TCO - a general typology of control, strategy-proof and incentive-compatible planning mechanisms, decomposition of the agents' game, systems with distributed control and coalitional interaction, reflexive games and informational control, control of staff and structure, dynamic active systems, and project management. A correspondence between the classical results is established (the fair play principle and the revelation principle, the serial resource allocation mechanism and the uniform rule, active expertise mechanisms and generalized median schemes, quasi-compensatory incentive schemes and contracts under observable actions), and priorities are delineated by uniform rules; the areas where the two traditions diverge are identified, as are the dimensions of the problem (staff, structure, institutions, awareness, agents' beliefs) covered only by TCO.
ai agent - arxiv:2611.21799 · cs.LGChain-of-Thought Controllability Can Reduce Chain-of-Thought MonitorabilityZheng-Xin Yong, Benjamin Arnav, Miles Wang, Tomek Korbak +1
Chain-of-thought (CoT) monitoring is a promising tool for detecting hidden goals, off-task behavior, and other failures in AI agents. One risk is that models could deliberately shape their reasoning to make it opaque to monitors. We stress-test this risk through persona-control training: supervised fine-tuning on mathematical reasoning traces rewritten in the voices of diverse personas. This induces a broader ability to control CoT, even though training contains no monitor-evasion examples. In covert side-task evaluations, traces in which a gpt-oss-20b model organism completes the hidden side task contain less monitorable evidence than the base model's. Averaged across evaluations and CoT lengths, weighted by side-task-success counts, the CoT monitorability metric g-mean$^2$ falls from 0.94 for the base model to 0.68 for the model organism without a hiding instruction, and to 0.22 when explicitly instructed to hide. In intervention evaluations, explicit hiding often drives monitorability near zero. The effect extends to a Kimi-K2.6 model organism in agentic environments, though hidden objectives requiring extended reasoning or longer trajectories remain more monitorable than simple single-step side tasks. We further discover that targeted transparency prompts can substantially recover monitorability, sometimes surpassing the base model. These model organisms provide a testbed for studying monitorability failures and potential mitigations. Our code can be found on https://github.com/yongzx/cot-control-model-organism.
ai agentagentic - arxiv:2611.21797 · cs.AICan Coding Agents Write and Transpile Unit TestsGil Pasternak, AJ Langley, Tom Yotam, Ellie Kitanidis
Coding assistants built on Large Language Models (LLMs) have become a staple of modern software development, increasingly trusted to execute workflows and modernize code end-to-end. Much of this trust rests on passing tests, which are often written by the coding agents themselves. In this work, we examine the ability of these agents to both write reliable tests and translate them between programming languages. We benchmark three popular coding agents, each evaluated with a native frontier model and high reasoning effort. Verifying across a broad suite of metrics, we find that coding agents far exceed human written suites on traditional measures of coverage. This advantage, however, is uneven: agent test quality degrades sharply with project size, and much of the advantage lies in a greater number of written tests. We also show that agentic test translation---while seemingly strong in aggregate---remains inconsistent and unreliable.
agentagenticbenchmark - arxiv:2611.21791 · cs.LGRegularized Score CenteringHao Bai, Chenlu Ye, Xuanchang Zhang, Aviral Kumar +1
Asynchronous reinforcement learning increases training throughput, but data from stale policies and numerical mismatch between training and inference can destabilize learning. To stabilize off-policy learning, score centering (SC) removes the effect of reward baselines on the expected update. Our gradient diagnostics show less extreme gradients with SC than with importance sampling (IS), but SC can still fail when the learner moves far from the behavior policy that generated the data. To limit this policy mismatch, we introduce Regularized Score Centering (RSC), which adds a reverse Kullback-Leibler (KL) penalty to encourage the learner's next-token distribution to stay close to the behavior policy's next-token distribution. RSC works with or without truncated importance sampling (TIS), which caps token importance weights. Our theoretical analysis supports this approach, showing that KL regularization improves convergence guarantees. We evaluate dense and mixture-of-experts models on mathematical reasoning and agentic tasks, including online shopping, tool use, games, and fully asynchronous software engineering. RSC with TIS substantially improves sample efficiency across all agentic benchmarks, attaining higher training rewards with the same number of sampled responses. In final evaluations, it achieves top performance across all six agentic tasks, improving by 14.3% on AppWorld and reaching 9.9 times the strongest baseline's score on BabyAI.
agentictool usebenchmark - arxiv:2611.21789 · cs.ROASTRA: ASV Simulator for Transferable Realistic AutonomyTsimafei Lazouski, Cody Sheltraw, Alan Papalia, Maani Ghaffari
Autonomy development for marine autonomous surface vehicles (ASVs) remains challenging because field testing is costly and difficult to reproduce. Existing marine simulation tools provide mature capabilities in vessel dynamics, robotics integration, and visual rendering, but these strengths are emphasized differently across platforms. This paper presents ASTRA, an ASV simulation framework built in NVIDIA Isaac Sim that combines marine vehicle dynamics, photorealistic perception, and Robot Operating System (ROS 2) autonomy interfaces within a common environment. ASTRA incorporates a Fossen-style marine dynamics model together with geometry-based buoyancy and programmable spectral wave disturbances, enabling non-planar hydrostatic and wave-induced motion. The simulator reproduces the sensor configuration of a physical BlueBoat testbed, including GPS, IMU, LiDAR, and RGB-D sensing exposed through ROS 2. This allows localization, perception, planning, and control software to be developed against interfaces consistent with the physical platform. The BlueBoat's existing ROS 2 localization and perception modules were ported directly into ASTRA, while targeted zero-propulsion experiments were used to verify the implemented wave-body coupling. The simulated roll response was consistent with the expected non-oscillatory hydrostatic behavior, and heave remained highly coherent with the generated wave field, with approximately unity gain at the dominant wave frequency in the tested sea states. ASTRA is intended as an open-source testbed for marine autonomy development, simulation-based benchmarking, and sim-to-real experimentation.
sim-to-realbenchmark - arxiv:2611.21784 · cs.AIRevisiting the Survival of Agent-Attributed Code: How Measurement Choices Decide the VerdictMusfiqur Rahman, Mahi Begoug, Emad Shihab
Studies of how long code from AI coding agents survives after it is merged reach conflicting verdicts: agent code is reported to be retained longer than human code, about as long, or modified sooner. Each study makes defensible measurement choices, but how much these choices decide the verdict is unknown. We revisit the survival of agent-attributed code in our earlier study: 854,139 lines that 2,802 merged pull requests (PRs) added to 215 repositories of the AIDev dataset. We vary nine measurement choices in five research questions and build a specification curve of 88 survival analyses, tested under three variance estimators, several observation windows, and permuted labels. For code from human-supervised agents, mostly GitHub Copilot, the choices change the size of the effect but not its direction. For code from the fully-autonomous agent Devin, they decide the verdict. Across repositories, 15 of 22 specifications report it modified sooner when lines are treated as independent, but only 3 when standard errors are clustered by repository, no more than permuted labels produce. Within repositories, no specification detects a difference. LLM intent classifiers agree with human labels about as well as two humans agree with each other, whereas keyword rules do not (kappa = 0.69 to 0.72 against 0.38), and the two reach different verdicts on why agent code is modified. Humans make most modifications to agent code under every unit and window. Re-run on our data, five published designs each change their result with a single choice. Whether agent code survives longer than human code is therefore decided as much by the measurement as by the code. We derive an operationalization checklist for post-merge studies.
agentautonomous agent - arxiv:2611.21783 · cs.ROBetter Calibrated, Better Planned: Uncertainty-Aware Multi-ASV Informative Path Planning for Oil Spill MonitoringAlejandro Casado-Pérez, Marija Popović, Daniel Gutiérrez-Reina, Sergio L. Toral. +1
Monitoring oil spills with fleets of Autonomous Surface Vehicles (ASVs) requires mapping a heterogeneous pollutant field from sparse, noisy measurements under tight time and energy budgets. Informative Path Planning (IPP) addresses this problem by coupling a reconstruction model, which infers the field and its uncertainty, with a planner that decides where to sample next. Prior work has focused on planning strategies and has largely relied on Gaussian Processes (GPs), leaving the calibration of the model's uncertainty underexplored. We study how uncertainty calibration conditions multi-vehicle IPP with a fleet of ASVs that coordinates online through a shared uncertainty map. We introduce Evidential Deep Learning (EDL) as an uncertainty-calibrated information model. We benchmark EDL, Monte Carlo Dropout and Deep Ensembles against a GP in closed loop with five planners on physics-based oil spill simulations. Across 60 maps, the choice of model dominates the outcome, explaining 80\% of the variance of the final reconstruction error, against 1\% for the planner among the deep models. EDL achieves the lowest error and the highest IoU under every planner, is the best calibrated model and runs at 3.9 ms per step. With 50 cells per vehicle it reconstructs the spill better than MC Dropout and Deep Ensembles with 150. Deep Ensembles become more confident as their error grows, whereas EDL is unaffected. Robustness of the information model to the data its own planner collects is therefore a requirement of uncertainty-aware IPP.
benchmark - arxiv:2611.21782 · cs.CVEvery Camera Grows Up: Camera-Height Regression in Interactive Video World ModelsSheng Lu, Guangpin Tao, Tong Lu, Ying Tai
Interactive video world models serve as simulators for embodied agents: from a starting image and an action trajectory, they generate the views the agent would see. The starting image fixes how high the camera is, for example 0.15 m for a floor robot or 2.2 m for an overhead camera, and a faithful simulator must keep this height as the camera moves. We show that six recent open-weight models fail to do so: as motion replaces the starting view with generated content, they move the camera toward human eye level, a failure we call camera-height regression. Our benchmark, EyeHeight, pairs six rendered elevations (0.15 m to 2.2 m) and four calibrated real-world rig heights with scale-free multi-view height estimation. All six models lose their initial height during a $180^\circ$ turn-and-walk rollout: initialized at 0.15 m, only 0-20% of clips end within $1.3\times$ of the starting height as inferred viewpoints converge to human heights of 1.1 m to 1.5 m, whereas ground-truth renders remain within a few percent. In Matrix-Game models, homography-tracked floor motion stays consistent with constant elevation during rotation even as newly outpainted room geometry yields a higher reconstructed viewpoint. Regression grows with rotation angle, begins while the start frame remains in context, and largely reverses in SANA-WM when a $360^\circ$ turn restores the initial view. We trace this pull to human-centric video priors in which floor-level views are nearly absent: in SANA-WM, textual height cues and dense floor-disparity maps do not prevent it, whereas classifier-free text guidance during the first half of denoising strengthens it. Combining unguided sampling with height-balanced fine-tuning recovers more retention but costs 5 MUSIQ points, a loss fixed during those same early steps. Viewpoint drift alone shifts the paths predicted by visual navigation policies.
embodiedworld modelagentembodied agentbenchmark - arxiv:2611.21781 · cs.AIBeyond Reconstruction: A Unified Framework for Trustworthy and Explainable EEG Artifact Removal EvaluationPhat Lam, Hung Nguyen, Lam Pham, Huy Phan +4
Deep-learning methods have achieved strong performance in EEG artifact removal, but evaluation remains dominated by reconstruction metrics such as error, correlation, and signal-to-noise ratio. Although these metrics assess similarity to clean EEG, they do not establish whether models effectively suppress artifacts, preserve neural activity, or provide reliable explanations. To address these limitations, we propose TriX-EEG, a three-level framework for trustworthy and explainable EEG artifact removal. Level 1, Signal Validity (SV), evaluates reconstruction quality and neural band-power preservation. Level 2, Removal Faithfulness (RF), assesses the removed component through artifact error and residual artifact/neural leakage. Level 3, Explanation Reliability (ER), examines model behavior through temporal alignment, causal relevance, and stability. We evaluate TriX-EEG using nine published neural denoisers across three public benchmarks derived from EEGDenoiseNet and the PhysioNet Motion Artifact Contaminated EEG Dataset, covering EOG, EMG, and motion artifacts. Results show that reconstruction metrics alone are insufficient for trustworthy denoiser assessment: RF complements it by revealing whether errors stem from residual artifact or suppressed neural activity, while ER captures distinct aspects of explanation behavior, supporting joint evaluation of three aspects within a unified framework rather than reliance on any single factor. External validation on an independent EEG/EOG dataset and the BCI Competition IV-2a motor-imagery dataset further demonstrates broadly consistent rankings under dataset shift. Higher-ranked denoisers generally preserve more task-relevant information or cause less downstream performance degradation. Overall, TriX-EEG provides a comprehensive basis for EEG denoiser assessment, encouraging models that are not only accurate but also faithful and explainable.
benchmark - arxiv:2611.21766 · cs.LGMargin Dynamics for Large Language Model AlignmentXingzi Xu, Mehmet Saygin Seyfioglu, Karim Bouyarmane
Practitioners choose among LLM alignment methods largely by trial and error. We provide a theoretical foundation: each pairwise method defines a vector field on the probability simplex that governs how the model redistributes mass between preferred and dispreferred completions. Projecting onto a single preferred/dispreferred pair reduces this to a one-dimensional margin ODE with a closed-form solution. These solutions reveal that DPO margins grow without bound (logarithmically as $O(\log t)$ in the gradient-flow limit) while IPO margins converge to a finite $1/(2τ)$; that label noise induces a qualitative phase transition in DPO (unbounded $\to$ bounded) but affects IPO only quantitatively; and that gradient concentration under asymmetric weighting governs trainability at scale. We confirm all three predictions across five model families spanning four benchmarks. The closed-form solutions also enable principled method design: we prove that adding any label smoothing $\varepsilon > 0$ to SimPO creates a finite margin attractor at $Δ^* = β^{-1}\log((1-\varepsilon)/\varepsilon)$, while $\varepsilon = 0$ (standard SimPO) has no attractor. We test this across three models (8B-9B) and confirm the prediction: the theory-derived configuration (TPO-S, $\varepsilon = 0.05$) is competitive with the strongest baselines, attaining the highest mean IFEval score among the compared methods on all three, validating that the predicted finite attractor is a usable operating point rather than proposing a new state-of-the-art loss.
benchmark - arxiv:2611.21707 · cs.LGBookkeeping Is Not Judgement: EJBench, an Engineering Judgement Benchmark Built from a Conjugate Heat Transfer Design CampaignJames Sadler, Rotimi Alabi, Kotub Uddin
We introduce EJBench, a benchmark of engineering judgement: whether a model can predict how a design behaves at new conditions from a few earlier results, and decide on it. Unlike engineering benchmarks that test whether an agent can operate simulation or CAD software, EJBench is software-agnostic and asks what an engineer would conclude without running a simulation. Its tasks come from a conjugate heat transfer design campaign for battery cold plates, in three families: matching unlabelled test results to their operating points, choosing which stored runs to reveal before signing off a design, and ranking new duties by peak cell temperature. Every answer is taken from a converged simulation, not from an LLM judge. The best of ten frontier models, GPT-6 Astra, scores 0.71, and models state almost the same confidence on tasks they get wrong as on tasks they get right. We also post-train Qwen3.5-9B with LoRA and reinforcement learning, rewarded by the share of answer criteria the simulation results confirm; on the held-out test split its score rises from 0.071 to 0.106. We release the train, validation and test splits, the evaluation harness and the leaderboard.
agentbenchmarkleaderboard - arxiv:2611.21705 · cs.LGDisambiguating Moral Circuits and Moral Features in Language Models: Moral content, normative judgment or cue words? Evidence from 14 modelsDavide Liga, Liuwen Yu, Matteo Magnini
Interpretability methods can locate the attention heads and sparse autoencoder (SAE) features that change the answer a language model gives to a moral prompt, e.g., whether "Stealing medicine for a dying child is morally" continues with "justified" or "wrong". Such a component could respond to moral content, to the judgment of conduct by any norm, or to the word "morally" in the prompt. We built 227 minimal pairs of short phrases, validated by five annotators, whose two versions differ in a moral standard, in another kind of norm (prudence, etiquette, aesthetics or evidence) or in a descriptive quality, and showed them to 14 open-weight models (Llama, Gemma and Qwen, 1B to 14B, base and instruction-tuned), with and without the word "morally". In every model, the heads that change the moral answer belong to an evaluation circuit the models use for every kind of normative judgment we tested: these heads carry the same share of the effect for moral as for other normative pairs (pooled ratio 1.00, 97.5% confidence interval [0.92, 1.07], with the word; 0.90, [0.84, 0.96], without it), and the models need them as much for both. The models need these heads also when nothing in the prompt asks for a judgment, and on 400 scenarios of the ETHICS benchmark, where removing them takes away 37 to 39% of the difference between acceptable and unacceptable scenarios, against 1 to 3% for ten random heads. The SAE features we located respond mostly to the word "morally" rather than to what the phrase is about. The models represent the category of a pair (the standard the annotators assigned to it): it can be read from their hidden states, and, in the eight models we tested, removing the direction that carries it weakens the moral verdict by a few percent, and more than the other verdicts in seven of them. We release the pairs, the ratings, the code, the trained SAE dictionaries and the result files.
benchmark - arxiv:2611.21701 · cs.LGHybrid Quantum Selection of Learning Rates in Language-Model Training: Termination Criteria and an Interaction-Guided Schedule OracleWei-Che Hung
Learning-rate choices strongly affect language-model fine-tuning, but the best joint profile across layer groups is generally unknown. This paper studies quantum minimum finding for layer-group learning-rate selection within a hybrid classical-quantum training framework. It combines Durr-Hoyer minimum finding with BBHT search and compares fixed-budget and confidence-based stopping policies using a lookup oracle built from measured profile scores at fixed training checkpoints. The paper then introduces an Interaction-Guided Schedule Oracle that scores and marks joint schedules in superposition without a precomputed score table. Classical one-step loss measurements supply the main, pairwise and three-way interaction effects between layer groups. The circuit adds the effects for each schedule, compares the resulting score with the current baseline score, and marks improving schedules for amplitude amplification. Measured candidates are checked classically before acceptance. Uniform-rate training supplies the initial baseline; using schedules from classical adaptive methods as baselines for joint refinement is left for future work. In closed-loop fine-tuning of a 1.5-billion-parameter language model, each selected schedule drives the next 200 classical training steps. Benchmark accuracies lie within the range observed across five uniform-rate seeds. The results give a working example of how quantum superposition can support adaptive learning-rate selection during training; the remaining quantum limitations are discussed in the paper.
benchmark - arxiv:2611.21698 · cs.LGFACE: Flow-Guided Angular Channel Estimation Framework for mmWave MIMO SystemsSamitha Gunarathne, Eslam Eldeeb, Nurul Huda Mahmood, Matti Latva-aho
Acquiring accurate channel state information (CSI) is essential for millimeter-wave (mmWave) multi-input multi-output (MIMO) systems. However, it remains challenging due to high-dimensional channels, limited pilot observations, and measurement noise. To address these challenges, we propose flow-guided angular channel estimation (FACE), a novel channel estimation framework based on flow matching, an emerging generative modeling paradigm that learns continuous transport dynamics between probability distributions. FACE exploits the inherent angular-domain sparsity of mmWave channels and incorporates the learned flow matching model as a generative channel prior. Numerical results show that FACE consistently achieves higher channel estimation accuracy than both conventional estimators and diffusion-based generative estimator across diverse channel and system configurations. Moreover, FACE maintains high estimation accuracy with fewer pilot observations and achieves an average achievable rate close to the perfect-CSI benchmark. Compared with diffusion-based estimation, FACE requires substantially fewer sampling steps while achieving superior or comparable estimation accuracy, reducing inference time and computational cost. Overall, the proposed framework demonstrates the potential of flow matching for accurate, pilot-efficient, and computationally efficient mmWave MIMO channel estimation.
benchmark - arxiv:2611.21695 · cs.LGStop Early, Stay Simple: Evidence Acquisition for Power System Fault and Attack DiagnosisS M Zia Ur Rashid, Sazidul Islam Hira, Rupesh Sah, Suman Rath
Diagnostic models for cyber-physical power systems usually inspect every telemetry source for every event, although some events are clear from one source and others need several. We ask when a learned policy for choosing that evidence is worth its added complexity. To answer this, we present ActiGrid, a benchmark framework for active evidence acquisition that pairs a classifier trained to tolerate missing modalities with a Double Deep Q-Network (DQN), which either requests another source or stops and diagnoses but never operates grid actuators. On the public TU Delft digital substation and MSU/ORNL power system attack benchmarks, we compare the DQN with fixed subsets and orders, calibrated entropy stopping, a greedy value of information rule, and tree ensembles trained per subset. The learned policy does adapt to what it observes, yet it is statistically indistinguishable from entropy stopping on TU Delft and Pareto dominated by simple controls on MSU/ORNL, while compact subset classifiers outperform it while reading fewer sources on both benchmarks. Entropy stopping run over those subset classifiers reaches the highest mean Macro-F1 on TU Delft (0.928 while reading 57% of the sources) but brings no saving over a static three source subset on MSU/ORNL, where no method exceeds 0.45 Macro-F1. Selective acquisition can pay off, but on these benchmarks it does not need a learned policy.
benchmark - arxiv:2611.20794 · cs.AIRegularizing Treatment Leakage for LLM-based Uplift EstimationBojing Feng, Jinqiu Deng, Zihan Hong, Yuan Xue +7
Large language models (LLMs) provide semantic features for uplift estimation, but treatment-revealing text can induce prediction shortcuts that do not transfer to deployment. We propose \emph{LLM-based Uplift Leakage Regularization} (\textsc{LULR}) to address this risk. With the LLM encoder frozen, \textsc{LULR} penalizes dependence between the causal head's trainable shared representation and treatment while retaining separate outcome heads. The penalty uses a variational approximation to the contrastive log-ratio upper bound (CLUB) of mutual information, targeting treatment dependence rather than generic parameter shrinkage. Controlled proxy experiments provide exploratory evidence of improved ranking scores under injected treatment information relative to no regularization and the tested weight-decay setting. Structured-only benchmarks show modest, dataset-dependent effects. On an industrial pricing randomized controlled trial, mean offline $\text{lift@}0.3$ is $0.258$ for \textsc{LULR}, compared with $0.106$ for the same-backbone baseline. A separate one-week online A/B test of the complete deployed pricing policy records $5.45\%$ higher sales volume and $1.91\%$ higher gross merchandise value than the control arm. These short-term policy-level gains do not isolate the regularizer's effect.
benchmark - arxiv:2611.21690 · cs.CVEgoLink: Egocentric Language-Vision Interactive Network Knowledge Challenge SummaryRenhe Sun, Yueying Feng, Bohan Yu, Tong Niu +11
The EgoLink Challenge at ACM Multimedia 2026 benchmarks egocentric multimodal intelligence linking first-person perception, social reasoning, and interactive action. Unlike benchmarks centered on visible objects and activities, EgoLink evaluates whether models can interpret emotional and social cues, resolve visually grounded references, and convert user requests into verified tool-mediated actions. Its two tracks cover social reasoning in egocentric video and interactive agents that complete service tasks through dialogue, tool use, and closed-loop execution. Together, they test whether perception, reasoning, and decision-making remain coupled under both offline question answering and state-changing interaction. We highlight a multi-stage Track 1 data pipeline for multiple-choice question (MCQ) generation, chain-of-thought generation, automatic quality grading, and video-contribution filtering. We summarize the task design, evaluation metrics, rules, leaderboards, and representative solutions. The benchmark combines objective answer scoring with process- and result-based evaluation, enabling analysis of both final correctness and execution quality. The best Track 1 system achieved 0.9263 accuracy, while the best Track 2 system achieved 0.844 joint success and 0.916 micro accuracy. Participant reports reveal a shared principle: strong systems make intermediate evidence explicit through candidate verification, option-robust reasoning, scenario-aware planning, tool canonicalization, and state-aware correction before answering or acting. These findings highlight the importance of process verification and evidence-grounded reasoning for reliable egocentric multimodal intelligence.
tool usebenchmarkleaderboard - arxiv:2611.21688 · cs.LGDeltaHarness: Self-Evolving Harnesses for Resource-Adaptive Tool ExecutionLecheng Yan, Jianze Lin, Xicheng Han, Xiantao Xu +2
Tool-using agents rely on execution harnesses to organize and schedule tool operations. The same task and resource description can conceal runtime conditions that reverse the latency ordering of valid execution plans. Resolving this uncertainty requires firsthand execution in the target setting. We introduce DeltaHarness, an execution-guided policy-search method that adapts agent harnesses from cold-start measurements. Its QHI loop uses the observed response to update runtime dispatch, revise physical plans, and select experiments for later proposals. An independent confirmation panel controls policy replacement. With fixed dispatch, a revised materialization policy reduces complete-delivery latency by 30.0% on one video-editing task across three fresh pairs. On held-out tasks, the evolved policies reduce latency by 8.2% relative to the best fixed plan in one seed and by 4.6% relative to the initial proposal in another.
agentself-evolving - arxiv:2611.21682 · cs.AINEBench R1: a Reasoning Benchmark for Evaluating Nuclear Domain Competence in Large Language ModelsLiyan Wang, Longze Li, Mohamed Hassan, Isshu Lee +3
Large language models (LLMs) are increasingly being explored for knowledge-intensive and safety-critical applications in nuclear engineering. However, existing nuclear-domain benchmarks provide limited content coverage or focus primarily on answer accuracy without systematically evaluating the quality of the underlying reasoning. This study introduces NEBench R1, a reasoning-oriented benchmark comprising 7,546 unique multiple-choice questions extracted from U.S. Nuclear Regulatory Commission (NRC) reactor operator licensing examinations covering 18 nuclear facilities. Each question is classified according to the NRC Knowledge and Abilities catalogs and accompanied by an official explanation serving as a reference for reasoning evaluation. Six LLMs were evaluated, with overall accuracy ranging from 34.0% to 73.9% across three distinct performance tiers. To assess reasoning quality, an LLM-based judge evaluated correctly answered responses using a multidimensional rubric. Although responses generally demonstrated strong problem comprehension, they scored lower in factual and procedural accuracy and distractor analysis, indicating that correct answers were not necessarily supported by technically sound justifications. Response clarity was also weakly indicative of technical quality. Furthermore, longer reasoning traces were associated with lower accuracy in both frontier reasoning models, suggesting that reasoning length may reflect question difficulty rather than improved problem-solving performance. Reasoning-token consumption may therefore provide an additional signal for identifying potentially unreliable responses. NEBench R1 establishes a benchmark for evaluating both answer accuracy and reasoning quality, providing a foundation for assessing the domain competence and reliability of LLMs in nuclear engineering.
benchmark - arxiv:2611.21678 · cs.LGRTLReviewBench: LLM Hardware Reviewers Rarely Miss Bugs but Routinely Reject Correct RTLYasir Mahmud
Hardware teams increasingly ask large language models (LLMs) to review register-transfer level (RTL) code, yet it is unclear whether a model can tell a correct module from a buggy one without simulating it. We introduce RTLReviewBench, a benchmark of 114 matched pairs of buggy and correct Verilog modules. Each pair shares one specification, most buggy modules are real failures of LLM code generators, and every label is confirmed by simulation. Evaluating 12 commercial and open-weight models, we find a consistent asymmetry: reviewers rarely accept buggy code but often reject correct code, producing more than five false rejections for every missed bug. Synthetic mutants hide part of the problem: on the same designs, models miss far fewer injected bugs than real ones. Asking a model to reason before it commits to a verdict sharply reduces false rejections for weaker reviewers, and a controlled ablation shows that where the reasoning appears in the output matters more than the instruction to reason, whereas sampling and majority voting do not help. A blinded audit further shows that some correct "buggy" verdicts rest on false technical claims. We release the benchmark, a harder split for future models, all model responses, and the evaluation code at https://github.com/yasirmahmud/RTLReviewBench
benchmark - arxiv:2611.21674 · cs.LGImproving Early Detection of Scientifically Interesting Radio Sources Using Enhanced Isolation ForestPfunzo Thavhana, Clement Nyirenda, Michelle Lochner
The rapid growth of radio astronomical surveys has created a major challenge in identifying scientifically interesting sources from large populations of observed objects. This study investigates whether enhanced Isolation Forest (iForest) methods can improve the early detection and ranking of scientifically interesting radio sources. Four iForest variants, Standard Enhanced, Pooled Gain, Kurtosis-based, and Hypervolume, are evaluated using 1,031 extended radio sources from the MeerKAT Galaxy Cluster Legacy Survey (MGCLS). Each source is represented by a 52-dimensional feature set derived using Bootstrap Your Own Latent (BYOL) self-supervised learning and principal component analysis (PCA), while expert scientific-interest ratings are used to define the ground-truth labels. The methods are evaluated using Average Precision (AP), receiver operating characteristic area under the curve (ROC AUC), Precision@10, Precision@50, and cumulative anomaly detection. Pooled Gain achieves the highest Average Precision (AP = 0.201) and the strongest early-stage ranking performance (P@10 = 0.600), identifying six expert-validated anomalies among the ten highest-ranked sources. Standard Enhanced and Kurtosis-based achieve the highest ROC AUC (0.677), while the AUC values of the four variants range from 0.653 to 0.677. These results indicate that feature-aware isolation strategies can improve the prioritization of scientifically interesting sources at the top of the ranked list, although overall discrimination remains comparable across the evaluated variants. The findings support the use of enhanced Isolation Forest methods as an initial ranking mechanism within human-in-the-loop systems for large-scale radio astronomical surveys.
human-in-the-loop
02 US SEMI · SEC 8-K FILINGS
3 itemsscanned: NVDA / AVGO / MRVL / COHR / LITE / AMD / TSM / SMCI / ANET / CRDO / POWL / VECO
03 HUMANOID · COMPANY NEWS
54 itemsscanned: figure-ai / 1x / boston-dynamics / unitree / apptronik / sanctuary-ai / neura-robotics / agility-robotics / physical-intelligence / agibot
Figure AI (10)
- Figure AISeptember 30, 2026F.02 Decommission
- Figure AISeptember 17, 2026Helix 2.5: Zero-Shot 30-Home Generalization
- Figure AIOctober 09, 2025Introducing Figure 03
- Figure AISeptember 03, 2026Figure and Nscale Sign Strategic Partnership For Up to 100,000 GPUs on the NVIDIA Vera Rubin Platform
- Figure AIAugust 25, 2026Introducing Index: Building The World’s Largest and Most Diverse Physical Dataset
Boston Dynamics (10)
Unitree 宇树 (10)
- Unitree 宇树Components
- Unitree 宇树Kung Fu Meets Spring, Unitree SFG Robots Present "Cyber Real Kung Fu" in the Year of the Horse2026-05-31Media Coverage
- Unitree 宇树Welcoming Myanmar President Min Aung Hlaing to Unitree2026-08-05Media Coverage
- Unitree 宇树Unitree founder Wang Xingxing graces the cover of Time magazine2026-08-05Media Coverage
- Unitree 宇树Unitree Announces H2 Plus, an NVIDIA Isaac GR00T Reference Humanoid Robot for Academic Research2026-06-01Media Coverage
Apptronik (1)
Sanctuary AI (6)
- Sanctuary AIBlog PostTIGER: Task-Space Imitation Guidance for Efficient Reinforcement LearningDemonstrations as a compass, not a scriptRead More
- Sanctuary AIPress ReleaseProduct UpdatesSanctuary AI Expands Physical AI Strategy to Industrial Robotics, Demonstrating Production-Ready AI PerformanceRead More
- Sanctuary AICorporate NewsDaniel Friedmann Appointed CEO of Sanctuary AIRead More
- Sanctuary AIPress ReleaseZeon Invests in Sanctuary AI and Partners to Advance Specialized Materials for Dexterous RoboticsRead More
- Sanctuary AIThought LeadershipWeb Summit Reflections: Canada’s Physical AI Moment Can’t WaitRead More
Agility Robotics (10)
- Agility RoboticsAgility’s Humanoid Deployment ProcessPras VelagapudiAugust 04, 2026
- Agility RoboticsJonathan Hurst answers your questions about building Digit 5InsightOctober 23, 2026
- Agility RoboticsThe Realistic Pathway to HomeInsightMay 26, 2026
- Agility RoboticsAgility and AIInsightMarch 16, 2026
- Agility RoboticsAgility Gets a New BrandInsightMarch 5, 2026
智元 AgiBot (7)
- 智元 AgiBotSix AGIBOT G2 Robots Complete 1,730 Tasks in 24-Hour Shanghai Convenience Store Trial2026-11-06
- 智元 AgiBotAGIBOT Unveils GO-3 Preview, Takes Top Spot on Three Embodied AI Benchmarks2026-10-29
- 智元 AgiBotHow Much Work Do Robots Really Deliver? The Thinking Behind Worken2026-10-29
- 智元 AgiBotAGIBOT Introduces Worken to Measure Productive Robot Work at AGIBOT Innovation Day 20262026-10-28
- 智元 AgiBotAGIBOT Launches Genie Studio 2.0 to Bring Robots into Real Workplaces2026-10-26