跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+0新增论文
+0其中 AI 相关
+1已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-09 00:58 · 全站最近入库：2026-11-08 06:02各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 01:45 — 2026-11-09 01:45（北京时间） · 点击数字查看对应简报；AI 相关包含在新增论文内。

最近完成检查：2026-11-09 00:58最近发现新增版本：2026-11-08 06:02库内论文最新更新时间：2026-11-06 02:59

每分钟刷新已保存的数据。空检查不代表采集异常；新收录可能包含历史论文。AI 导读按新增版本计数。

17,100 已收录论文14,661 AI 相关4,057 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

时间范围跟随排序：版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常，查看状态 →

找到 3,292 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 00:56（北京时间）
新论文 · v1arXiv:2611.04100

The Agent Board: Kanban for Machines; Durable Coordination for Agents That Outlive the Conversation

Tyler Hunt, Donald Nguyen, Gurbinder Gill

原文摘要

Harnesses for large language models have converged on a single-threaded shape: one agent, one history, one machine, one sitting. That shape fails work that spans many agents, outlives any one model call, or waits on systems that are not agents. We describe the Agent Board, the coordination substrate in the Corvic harness. The board is a durable, shared table of semi-semantic tasks: a few exact fields the system routes on, and a free-form body that only language models read. Agents post tasks, watch tasks, and block on them, ending their turn and holding no process until a matching post wakes them. A system agent, the Board Keeper, acts as the interrupt handler for the board: it carries events from non-agent parts of the harness onto the board, matches them to waiting tasks by reading their prose, and allocates new agents for tasks that name a kind of agent rather than an instance. Loadable coordination policy kernels give statuses and roles meaning for a given pattern of work, without the board enforcing them. The design mirrors the kanban board a human team puts on a wall. We argue it is what persistent agent products such as Meta Muse, OpenAI dots and SpaceXAI Grok Bot need once an agent must wait, hand off work, or work with agents from another vendor. We view the board as an operating-system scheduler for agents and state where the analogy breaks. We separate what the mechanism guarantees from what it leaves to convention, and report a trace from a live board and three small simulated pilots. In the pilots, the wait cost of the board was flat in wait length while naive polling grew about 165-fold, but a plain external watcher was cheaper than the board. The Board Keeper matched prose waiters with no false wakes but missed 3.6 percent of real wakes.

cs.AI · cs.DC

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2603.16016

FlatLands: Indoor floormap completion from a single egocentric view

Subhransu S. Bhattacharjee, Dylan Campbell, Rahul Shome

原文摘要

A single egocentric image typically captures only a small portion of the floor, yet a complete metric traversability map of the surroundings would better serve applications such as indoor navigation. We introduce FlatLands, a dataset and benchmark for single-view indoor floor-map completion in metric bird's-eye view (BEV), evaluating eleven models: both deterministic and generative. FlatLands contain 270,575 observations from 17,656 scene records across six metric 3D datasets, with aligned observation, visibility, validity, and ground-truth maps and both in- and out-of-distribution protocols. We compare training-free, deterministic, ensemble, and generative methods on shared geometry-derived inputs and, without fine-tuning, on maps estimated from real egocentric RGB images. Our results show that predicting floor almost everywhere can achieve high IoU when most evaluated cells are floor, while generative sample sets outperform the best single-output model under a proper scoring rule and support safer risk-weighted planning. FlatLands establishes a rigorous testbed for evaluating both floor-map accuracy and how effectively alternative completions represent the ambiguity of unseen space.

cs.CV · cs.AI · cs.RO · eess.IV

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2601.01075

Flow Equivariant World Models: Structured Memory for Dynamic Environments

Hansen Jin Lillemark, Benhao Huang, Fangneng Zhan, Yilun Du, Thomas Anderson Keller

原文摘要

Embodied systems experience the world as 'a symphony of flows': a combination of many continuous streams of sensory input coupled to self-motion, interwoven with the dynamics of external objects. These sensory streams and the underlying dynamics of the world obey smooth, time-parameterized symmetries which existing world models ignore. Without a memory that respects this structure, partial observability presents a major obstacle to existing methods: each observation reveals only a fraction of the world, while unobserved regions continue to evolve. In this work, we introduce Flow Equivariant World Modeling, a framework that leverages time-parameterized symmetries within a latent memory for stable and accurate dynamics prediction over long horizons. The latent memory shifts and transforms equivariantly with self-motion and inferred external object motion, keeping information about out-of-view regions aligned as time progresses. We demonstrate the advantage of this framework over state-of-the-art diffusion, memory-augmented, and recurrent world model architectures on 2D and 3D partially observed video world modeling benchmarks. More broadly, our results suggest that predictive representations become more powerful when they are organized in line with the temporal and dynamical structure of the world they model. Project page: https://flowequivariantworldmodels.github.io/

cs.LG · cs.AI · cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2604.02434

Compositional Neuro-Symbolic Reasoning

Anugyan Das, Omkar Ghugarkar, Vishvesh Bhat, Asad Aali

原文摘要

We study structured abstraction-based reasoning for the Abstraction and Reasoning Corpus (ARC) and compare its generalization to test-time approaches. Purely neural architectures lack reliable combinatorial generalization, while strictly symbolic systems struggle with perceptual grounding. We therefore propose a neuro-symbolic architecture that extracts object-level structure from grids, uses neural priors to propose candidate transformations from a fixed domain-specific language (DSL) of atomic patterns, and filters hypotheses using cross-example consistency checks. Instantiated as a compositional reasoning framework based on unit patterns inspired by human visual abstraction, the system augments large language models (LLMs) with object representations and transformation proposals. On ARC-AGI-2, it improves over the base solver LLM (Grok-4, 16.0%) to 24.4% on the public evaluation set, and to 30.8% when combined with ARC Lang Solver via a meta-classifier. These results support the claim that in the evaluated settings, allocating test-time compute to object abstraction, neural-guided transformation proposal, and consistency filtering can improve accuracy over unstructured sampling-based scaling without task-specific finetuning of the system. We open-source the code: (https://github.com/CoreThink-AI/arc-agi-2-reasoner).

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04090

Swapping the Judge Changes the Ranking: A Preregistered Study of Cross-Provider Replication in LLM-Judged System Comparisons

Sangdo Han, Moonsoo Kim

原文摘要

Practitioners who compare systems with an LLM judge and have no human labels often check a conclusion by repeating it with a judge from another provider. We test how far that check can be trusted. In a preregistered study (with a disclosed post-registration reduction of the judge set), five judge models from three providers re-judged 4,000 Chatbot Arena battles (nine systems, 33 system pairs) that carry human votes, each in both response orders. Agreement with the human vote was low at the verdict level (median Cohen's kappa = 0.22), and system rankings did not survive a change of judge: Kendall's tau with the human ranking ranged from 0.17 to 0.89 (median 0.67), so the preregistered hypothesis that rankings remain stable was not supported. Conclusions replicated by two low-cost judges from different providers matched the human conclusion more often than single-judge conclusions (+0.10 precision), but the gain came from filtering out one poorly aligned judge, and replication involving that judge left a conclusion for only 35% of comparisons. Judges that agreed with humans reached a stable conclusion with no more verdicts than human votes required; the others failed to converge on 26-70% of system pairs at the sample sizes tested (up to 100 verdicts per pair). Disagreements with the human vote co-occurred across judges far more often than independence predicts, although a single human vote per battle cannot separate shared judge bias from voter noise.

cs.CL · cs.AI · cs.CY

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2602.01518

Qrita: High-performance Top-k and Top-p Algorithm for GPUs using Pivot-based Truncation and Selection

Jongseok Park, Sunga Kim, Alvin Cheung, Ion Stoica

原文摘要

Despite their importance in model sampling, efficient implementation of Top-k and Top-p algorithms poses a significant challenge as model vocabularies grow with every new model release. Existing approaches often rely on sorting, which incurs significant computation and memory overhead on GPUs, or on stochastic approaches that alter the algorithm's output. In this work, we propose Qrita, an efficient Top-k and Top-p algorithm based on a pivot-based truncation and selection. Qrita leverages pivot-based search for both Top-k and Top-p with two key techniques: 1. Gaussian-based $σ$-truncation, which greatly reduces the search space of the vocabulary, and 2. Quaternary pivot search with duplication handling, which halves the number of pivot search iterations and guarantees deterministic output. We implement Qrita using Triton and evaluate its performance against the Top-k and Top-p kernels of high-performance LLM execution engines such as SGLang and FlashInfer, improving end-to-end serving throughput up to 1.4$\times$ with half the memory usage, while providing the same output of stable sort-based Top-k and Top-p algorithms. Qrita is now the default Triton Top-k and Top-p sampler of vLLM, and its implementation is available at https://github.com/vllm-project/vllm/blob/main/vllm/v1/sample/ops/topk_topp_triton.py.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04087

ScienceVision: Can AI Observe and Reason Like a Scientist?

Tianyi Ma, Jing Gu, Haozhe Zhao, Liang Chen, Yunfan Zhang, Haijun Su, Jialong Wu, Le Xue 等 13 位作者

原文摘要

Scientific discovery builds on accurate observations interpreted through domain knowledge and reasoning. For scientific 3D data, this requires understanding structures and spatial relationships, from protein contacts to anatomical organization. Existing multimodal benchmarks largely evaluate text and 2D images, while scientific 3D evaluations vary in representations and protocols across disciplines, limiting cross-disciplinary comparison. To assess these foundational capabilities, we introduce ScienceVision, a cross-domain benchmark for scientific 3D understanding and reasoning through expert-level visual question answering (VQA). We construct 250 questions, with 50 each from Biology, Chemistry, Medical Imaging, Mechanical CAD, and Architecture, through a pipeline combining synthetic question generation with domain-expert review and annotation. Each task pairs a native 3D artifact with a scientific question, a specified response format, and a task-specific reference answer, requiring domain knowledge to be applied to geometric or topological evidence. We evaluate 12 models on the same questions in three settings: visual question answering with native 3D artifacts accessed through an agent harness's file and code-execution tools; the same workflow augmented with expert-inspired viewer inspection and measurement tools exposed through the Model Context Protocol (MCP); and visual question answering from prescribed rendered 2D images using an agent harness. All settings use the same task-specific deterministic final-answer scoring. The inspection environment supports view changes, target isolation, and slice inspection. In this main comparison, the best mean answer score is 0.53 out of 1. Mean scores in the native-artifact workflow without additional MCP tools exceed those in the rendered-image workflow for every model, by 17.2 percentage points on average.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.26080

Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents

Changdae Oh, Wendi Li, Seongheon Park, Samuel Yeh, Tanwi Mallick, Sharon Li

原文摘要

Process reward models enable fine-grained, step-level evaluation of LLMs, yet building them for agentic settings remains prohibitively difficult: long-horizon interactions, irreversible actions, and stochastic environment feedback make both human annotation and Monte Carlo estimation infeasible at scale. In this work, we show that reinforcement learning (RL) post-training already provides the ingredients for effective step-level scoring, eliminating the need for dedicated reward model training altogether. Concretely, we derive an implicit advantage under a general stochastic Markov decision process, which we term progress advantage -- log-probability ratio between the RL-trained policy and its reference policy exactly recovers the optimal advantage function. This formulation makes the resulting signal annotation-free, domain-agnostic, and available as a byproduct of the standard RL post-training pipeline. We validate the effectiveness of the progress advantage across three different applications: test-time scaling, uncertainty quantification, and failure attribution on five benchmarks and four model families. Across all settings, it consistently outperforms confidence-based baselines and, despite requiring no task-specific training, surpasses dedicated trained reward models. We complement these results with deeper analyses on characteristics of progress advantage, offering practical guidance for adoption in real-world agentic systems.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v7arXiv:2605.14211

ASH: Agents that Self-Hone in Long-Horizon Worlds

Benjamin Schneider, Xavier Schneider, Victor Zhong, Sun Sun

原文摘要

Long-horizon visuomotor tasks remain a fundamental challenge in AI, as current methods rely on hand-engineered rewards or action-labeled demonstrations, neither of which scales. We introduce ASH, an agentic system that learns a long-horizon policy from unlabeled, noisy internet video, without reward shaping or expert annotation. ASH follows a self-improvement loop; when it gets stuck, ASH learns an Inverse Dynamics Model (IDM) from its own trajectories, and uses its IDM to extract supervision from relevant internet video. ASH uses unsupervised learning to identify key moments from large-scale internet video and retains them as long-term memory - allowing it to tackle long-horizon problems. We evaluate ASH on two complementary environments demanding multi-hour planning: Pokemon Emerald, a turn-based RPG, and The Legend of Zelda: The Minish Cap, a real-time action-adventure game. In both games, behavioral cloning, retrieval-augmented and zero-shot foundation-model baselines plateau, while ASH sustains progression across our 8-hour evaluation. ASH reaches an average of 11.2/12 milestones in Pokemon Emerald and 9.9/12 in Legend of Zelda, while the strongest baseline gets stuck in both environments at an average of 9.3/12 and 7.8/12 milestones, respectively. We demonstrate that self-improving agents are a scalable recipe for learning in long-horizon worlds. Our project page is https://toolbox-hq.github.io/ASH

cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04079

AlgoPilot: Cross-Paradigm Reasoning in Language Models via Strategy Selection and Guidance

Yilun Hao, Ruixiao Yang, Hudson Hilal, Yongchao Chen, Kaizhi Qian, Yada Zhu, Yang Zhang, Chuchu Fan

原文摘要

Real-world problems span diverse domains, including program synthesis, symbolic reasoning, planning, and optimization, and often require fundamentally different solution paradigms. A central challenge for LLM-based reasoning is twofold: identifying an appropriate problem-solving approach and formulating it correctly. In this paper, we propose AlgoPilot, a cross-paradigm reasoning framework that enables adaptive selection of the solving approaches and structured problem formulation prior to execution. The AlgoPilot framework contains three components: 1) a SteerLM, trained via supervised fine-tuning (SFT) and reinforcement learning (RL), that adaptively selects the solving strategy (paradigm and modes) based on input problems, 2) the corresponding expert GuideLM, trained via SFT, that generates structured formulation guidance tailored to the selected strategy, and 3) an ExecutionLM that generates solutions conditioned on both the selected strategy and its formulation. Unlike prompting-based methods that rely on fixed reasoning templates, our approach learns to adaptively choose and structure problem-solving strategies based on the input. We build AlgoPilot on Qwen3-8B and evaluate across 5 problem categories and 13 benchmarks with 66 unique domains, comparing against both the latest similarly scaled small models and large LLMs. AlgoPilot significantly improves average accuracy for the same model from 38.4% to 72.9%, outperforming the best large LLM prompting baseline GLM-5 by 11.7%. Ablation studies confirm the effectiveness of each component. Notably, we show that the learned steering and formulation modules can transfer across execution models, suggesting their potential to improve other LLMs. Project page: https://sites.google.com/view/algopilot.

cs.AI · cs.CL · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.11901

DuoBench: A Reproducible Benchmark for Bimanual Manipulation in Simulation and the Real World

Tobias Jülg, Seongjin Bien, Simon Hilber, Yannik Blei, Pierre Krack, Maximilian Li, Sven Parusel, Rudolf Lioutikov 等 10 位作者

原文摘要

Bimanual robot systems substantially expand manipulation capabilities, but coordinating two arms introduces additional control complexity and failure modes that are not well captured by existing benchmarks. We introduce DuoBench, an extensible benchmarking framework for bimanual manipulation policies on the FR3 Duo platform. DuoBench comprises eleven tasks spanning four coordination categories, implemented in simulation and partially replicated in the real world through reproducible task recipes with 3D-printable assets. In addition, we propose a stage-based evaluation scheme that supports fine-grained semantic failure analysis beyond binary success. It includes stage-conditional pass rates with confidence intervals as well as human-teleoperated datasets for all benchmark tasks. We benchmark several imitation-learning and vision-language-action policies in simulation and on real hardware. Our results show that current policies remain challenged by bimanual manipulation, particularly at the beginning of interaction, during parallel arm execution, and in sim-to-real transfer settings. We further find that simulation reveals the same failure modes as real-world experiments. DuoBench provides a reproducible testbed for analyzing them and studying future methods for dual-arm policy learning. Code, datasets, and videos are available at https://duobench.github.io

cs.RO · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2510.00279

SLogic: Subgraph-Informed Logical Rule Learning for Knowledge Graph Completion

Trung Hoang Le, Tran Cao Son, Ishtiaq Ahmed, Huiping Cao

原文摘要

Logical rule-based methods offer an interpretable approach to knowledge graph completion (KGC) by capturing compositional relationships in the form of human-readable inference rules. While existing logical rule-based methods learn rule confidence scores, they typically assign a global weight to each rule schema, applied uniformly across the graph. This is a significant limitation, as a rule's importance often varies depending on the specific query instance. To address this, we introduce SLogic (Subgraph-Informed Logical Rule learning), a novel framework that assigns query-dependent scores to logical rules. The core of SLogic is a context-aware scoring function. This framework is in line with the specificity principle in commonsense reasoning. This function determines the importance of a rule by analyzing the subgraph locally defined by the query's head entity, thereby enabling a differentiated weighting of rules specific to their local query contexts. Extensive experiments on benchmark datasets show that SLogic achieves competitive performance among rule-based methods. It also generates query-dependent, human-readable logical rules that serve as explicit explanations for its inferences.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04061

On the Primacy Bias in RLVR Training

Phuc Minh Nguyen, Ngoc-Hieu Nguyen, Parshin Shojaee, Chinh Duc La, Duy Minh Ho Nguyen, Rui Zhang, Khoa D. Doan, Ting Hua 等 9 位作者

原文摘要

Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a promising approach for enhancing the reasoning capabilities of large language models. However, it still remains unclear whether RLVR can extend reasoning capabilities beyond what is already encoded in the base model. In this paper, we study this question through the lens of primacy bias: the tendency of models to over-rely on already-learned representations, disproportionately improving performance on problems they already handle well while underexploring harder ones that require departing from those representations. Through our controlled experiments and large-scale analysis, we show that RLVR training on mixed data tends to further sharpen performance on easier problems while yielding only marginal gains on harder ones. We find that although hard problems lead to modest gains overall, they are actually the ones that help most in expanding the model's reasoning capabilities and pushing the representations. However, prolonged training on these hard problems may also hurt and lead to forgetting on easy problems. These findings help to better understand some of the prior conflicting observations about RLVR, specifically, why on (OOD) domains where the base model performs poorly, improvements on hard problems can expand reasoning; while on well-trained domains, forgetting can reduce performance and mask genuine gains obtained from hard problems.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.05427

LLM-based NPCs in Video Games: A Systematic Mapping Study

Dilan Mian, Jeremy Bradbury, Yann-Gaël Guéhéneuc, Fabio Petrillo, Cristiano Politowski

原文摘要

For decades, non-player characters (NPCs) have been driven by rule-based systems whose personalities and interactions are scripted before release. Large language models (LLMs) promise NPCs that converse openly, adapt to new situations, and behave in ways developers did not fully author, prompting a wave of research and commercial experimentation. Yet the field lacks a unified picture of what has been built, how well it works, and at what cost. We address this gap with a systematic mapping study of 377 papers from the academic literature, yielding 42 papers analysed against a main research question and ten sub-questions across four thematic axes. We provide the first quantitative characterisation of component adoption, latency, safety coverage, and reproducibility: short-term memory and world-state grounding are near-universal (60% and 69%), whereas long-term memory (31%), reflection (38%), and retrieval-augmented generation (14%) remain rare, and response times span 2.6 to 15 seconds. Finally, we identify six structural barriers to commercial adoption and offer evidence-based recommendations for developers and researchers.

cs.SE · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04059

Query-based Submodular Data Selection for Efficient Instruction Tuning

Zhenghao Zhao, Yasutomo Kawanishi, Huihuo Zheng, Michael E. Papka, Yan Yan

原文摘要

As large language models are fine-tuned for increasingly specialized tasks, their performance depends more on which training examples are selected than on how many. Given a large heterogeneous training pool and a small target set specifying the desired capability, targeted data selection aims to find a compact subset that maximizes downstream performance. Existing methods score each candidate independently against the target and select the top-ranked examples, yet this point-wise strategy ignores inter-example interactions: at tight budgets the selected subset is dominated by near-redundant points that waste capacity on overlapping information. We propose Query-based Submodular functions with Gradient-guided Synthesis (QSGS), which formulates targeted selection as a submodular graph cut over a gradient similarity graph. A modular relevance term pulls selected examples toward the target task, while a pairwise dispersion penalty suppresses intra-subset redundancy. The resulting objective balances target relevance and intra-subset redundancy using the same formulation across different selection budgets. Empirically, the benefit of diversity-aware selection becomes increasingly pronounced as the data budget decreases. To further close coverage gaps that no selection from existing data can bridge, we introduce a gradient-guided synthesis step that identifies under-covered target directions via residual subspace analysis and generates synthetic examples to fill them. Experiments on MMLU, BBH, and TydiQA across Llama-2, Llama-3, and Mistral, show that QSGS at a 0.5% budget matches or exceeds full-data performance on Mistral-7B, outperforms the strongest baseline across all benchmarks, and yields increasing gains as the budget shrinks.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04057

Look Before You Launch: Learned Throughput Prediction for Hardware Selection in a Production Training Fleet

Hossein Askari, Wei Sun, Chenguang Zhu, Bin Zan, Sreen Tallam, Keyu Man

原文摘要

In mixed-vendor accelerator fleets, many training jobs can run on several hardware types, and the best choice depends on each job's throughput on each type. Routing a job needs a throughput estimate for every candidate type before the job runs. We present LEAP, a machine learning performance prediction service that predicts each training job's throughput on every candidate hardware type at submission time and ranks the candidates by predicted throughput per unit cost. It routes tens of thousands of training job attempts a day across hundreds of different model types and more than 20 hardware types in a production fleet. LEAP predicts throughput from submission-time training configuration, including each job's prior throughput. LEAP reaches an accuracy of 77% over 30 nightly retrains, higher than the lookup table it replaced and at least as high as every alternative model we tested, including fine-tuned language models. Over six months, the share of eligible training GPU-hours that received a ranking by LEAP grew from 58% to 95%, and we estimate that it saves about 3% of training computational cost. In this paper, we identify the drivers that actually move accuracy and those that do not. We offer these as lessons for teams building performance predictors at fleet scale.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.40946

Opening the Box Changes the Cat: LLMs Detect Chain-of-Thought Monitoring - What now?

Amir Jalilifard, Marcos Medeiros Raimundo, Anderson Rocha

原文摘要

Chain-of-thought (CoT) reasoning allows large language models (LLMs) to generate intermediate reasoning traces in natural language while solving complex tasks. These traces can support safety monitoring by exposing potentially harmful plans before actions are executed. In this work, we investigate the conditions under which LLMs infer the presence of CoT monitoring from direct and indirect environmental cues and examine how this inference affects their subsequent behavior. Our observations suggest that oversight is an interactive process: while a monitor inspects a model's reasoning, traces left by monitoring interventions can allow the model to infer that its thoughts are being observed and to adapt its behavior accordingly. We show that even subtle environmental cues, such as timing-related signals from timeouts and operation durations, can reveal the presence of monitoring and trigger models to adapt their behavior to evade oversight. Across repeated episodes and interaction loops, we further observe that models can form statistical estimates of the monitor's effectiveness and use these estimates to guide attempts to evade detection. These findings highlight the need to evaluate CoT-based oversight under adaptive conditions, where the monitoring process itself may influence the behavior it seeks to observe.

cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04055

HybridSpec: Throughput-Aware Speculative Decoding for Hybrid GDN-MoE Models

Haiyu Wang, Sai Qian Zhang, Zechun Liu, Steven Li

原文摘要

In speculative decoding, parallel drafters such as DFlash make drafting cheap, so target verification dominates each round. A central choice is the verification length, the number of tokens that the target processes in each round. Verifying more draft tokens can accept more of them, but work after the first rejection is wasted. For hybrid target models that combine Gated DeltaNet (GDN) and mixture-of-experts (MoE) layers, verification time grows much faster with this length than for dense models. In a single-request profile of Qwen3.6-35B-A3B with DFlash, it grew by 71% from length 4 to 16, almost entirely in GDN and MoE layers, compared with 4%-6% for a dense Qwen3-4B target. Each GDN layer must update a recurrent state at every token and recover it at the end of the accepted prefix. The MoE cost depends on the experts that the tokens activate, which are known only once verification runs. We present HybridSpec, which selects the verification length on the GPU from draft-side acceptance estimates and measured target latency. Verification then runs this length plus one extra token up to the first MoE router, before any expert runs. There, a lightweight routing correction uses the observed expert assignments to decide whether to keep that token. The rest of verification runs only on the selected tokens. Block-parallel GDN kernels avoid storing a state after every token and reconstruct the accepted state of every GDN layer in one launch. In isolated benchmarks at verification lengths 4 to 16 with every draft token accepted, these kernels, including accepted-state reconstruction, are $1.54\times$-$4.04\times$ faster than the default kernels. With three GDN-MoE target models in single-request greedy decoding, HybridSpec is $1.75\times$-$1.88\times$ faster than autoregressive decoding and $1.30\times$-$1.33\times$ faster than DFlash.

cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04054

Forgetting to Improve: Principled Data Removal in Active Learning

Manuel Wendl, Erik Englesson, Andreas Krause, Carl Henrik Ek

原文摘要

The uncertainty of a statistical model is most commonly factorized into an aleatoric and an epistemic part. This factorization changes how predictions are interpreted in downstream decision tasks. Importantly, except for the idealistic scenario with no model mismatch, the quantification is a characteristic of the model and not the data generating process. In this paper, we propose Forgetting to Improve a method that reduces this discrepancy by incorporating the task into the modeling framework. Our key insight is to acknowledge that in scenarios of model mismatch, data can have a detrimental effect on the modeling for a specific task. Based on this insight, we propose an influence function for Gaussian process models that allows for principled removal of detrimental data samples. We showcase the flexibility of this approach by demonstrating significant improvements across a range of tasks, including Bayesian optimization, model-based reinforcement learning, and transductive learning.

cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04048

Exploit or conform? How norm utility prompting changes the way LLMs use community beliefs

Leon Klingborg, Theodor Cimpeanu, Alexander J. Stewart

原文摘要

Large Language Models (LLMs) serve both as tools for aiding human decision-making in everyday life and as models of human agents in experimental simulation studies. Alignment on prototypical social decisions -- such as whether to cooperate on a collective action -- therefore has immediate practical consequences for researchers and for society at large. Here we systematically explore the impact of ``norm utility'' -- the non-material components of utility that account for an individual's desire to conform to social norms and to be internally consistent -- on the strategic decision-making of three open-weight LLMs in one-shot, two-player social dilemmas. We find that when a social dilemma is presented together with information about community beliefs -- but without any indication of how those beliefs should matter -- models tend to treat community belief as strategic information, so that in the Snowdrift game cooperation falls as community belief in cooperation increases. When we explicitly prompt the components of norm utility this relationship inverts, and cooperation increases with community belief, in broad agreement with the predictions of a simple norm utility model. We then show that prompting norm utility reduces the volatility of strategic decisions with respect to changes in the underlying social context, and generates more consistent reasoning habits across models. Taken together, our results suggest that prompting norm utility can help generate more human-like and consistent behavior across models and lower volatility within models.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2610.04898

A Systematic Analysis of the Predictive Power of LM Surprisal in Reading Chinese

Hongao Zhu, Muxiaoqiao Xu, Yikang Liu, Siyuan Song, Yuxia Wang, Byung-Doh Oh, Hai Hu

原文摘要

This study analyzes the predictive power of LM-derived, token-level surprisal on Mandarin Chinese reading times. Using a suite of Chinese-Pythia models (14M--1.4B) trained from scratch with 30B tokens, we examine how well surprisal predicts first fixation duration, gaze duration, and total reading time in three paragraph-level eye-tracking corpora of Mandarin Chinese (GECO-CN, HKP, and MECO). Contrary to previous null findings, our results show that surprisal is predictive of Chinese reading times. However, whether predictive power scales with model size and the amount of training is corpus-specific: bigger models predict reading times better in GECO-CN, whereas inverse scaling emerges in HKP and, at the largest sizes, in MECO. Subsequently, we find that checkpoints whose surprisal remains closer to $n$-gram statistics are better predictors of reading in HKP and MECO, providing one possible explanation for the inverse scaling behavior. All in all, the predictive power of surprisal on Chinese reading time measurements is corpus-specific, which cautions against drawing conclusions about language model scale from a single corpus.

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04041

When Model Complexity Matters in Cross-Subject Naturalistic fMRI Prediction

Junhao Zhao, Jacob Kang, Colin Kurniawan, David Simberg, Nan Xu

原文摘要

Increasingly complex neural models predict blood-oxygen-level-dependent (BOLD) responses to naturalistic stimuli, but the predictive value of this added complexity remains unclear. We ask when pretrained stimulus representations help and how much downstream complexity is needed, with and without recent measured BOLD history. We compare linear diagnostic references and nonlinear predictors across HCP movie fMRI (154 subjects), BOLD Moments (10 subjects), and Algonauts 2025 (4 subjects). Under a shared offline protocol, models are fitted on training subjects and evaluated on held-out subjects viewing largely familiar stimuli, without refitting. Pretrained visual features generally improve prediction over raw pixels; for linear autoregression, this advantage becomes more pronounced over longer rollouts (recursive predictions without additional measured BOLD). Measured BOLD history provides substantial gains that generally diminish {as rollouts lengthen}. Nonlinear stimulus-only predictors can improve on linear readouts, whereas linear autoregression provides a strong history-conditioned baseline. Controls examine context length, regularization, fitting procedures, and model capacity. Regional response consistency across subjects is associated with held-out prediction accuracy, while less consistent regions tend to gain more from measured history. In paired simulations with fixed linear neural dynamics, slower hemodynamic responses generally sustain high temporal prediction correlations over longer rollouts, illustrating how the observation process can contribute to predictive persistence. Together, these findings support evaluating stimulus representations and downstream complexity separately, using history-conditioned linear prediction as a diagnostic reference under matched inputs and measurement availability.

q-bio.NC · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04035

Recursa: Toward Recursively Self-Improving Foundation Model Systems

Erkang Zhu, Yin Lin, Xiaorong Shi, Xuanrui Lin, Jingren Zhou

原文摘要

Foundation models in changing environments can encounter unfamiliar domain knowledge that their original training did not cover. However, new domains rarely provide large amounts of expert-created training data or reliable supervisory signals for further training. We introduce Recursa, a foundation model system that adapts to new domains by learning from its own experience without human experts or a stronger teacher model. Recursa uses a knowledge repository to store what it learns: domain-specific skills and instructions on when to apply them, along with memories of past task experiences and reflections on what to improve. Recursa bootstraps the repository from domain materials and then refines it through recursive self-improvement. It uses parallel learners to execute tasks and synthesize local improvements, then validates and merges the improvements into the knowledge repository. On Harvey LAB's Firm Knowledge benchmark, Recursa raises the all-pass rate on held-out test tasks from 30.0% to 55.0% with DeepSeek-V4.1-Flash and from 35.0% to 46.0% with Qwen3.8-Flash.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.01839

Import What You Need: Learning When and How to Augment EHR Graphs with External Knowledge

Chen Chen, Mohsen Nayebi Kerdabadi, Dongjie Wang, Mei Liu, Zijun Yao

原文摘要

Longitudinal prediction from electronic health records (EHRs) is limited by the sparsity and irregularity in patient trajectories, and knowledge augmentation with external knowledge graphs (KGs) offers a promising way to alleviate these issues. However, KG expansion often follows predefined rules, with limited control over whether and how to augment each visit as a patient's state evolves. We propose ReTA, a Reinforcement learning-based dynamic Topology Augmentation framework that casts KG import as a per-visit, budget-aware policy. ReTA first constructs an offline refined pool of KG-grounded templates, then learns a policy to select one augment action per visit from three options: Soft Import, which enriches node features without modifying graph topology, Hard Import, which grafts a compact KG subgraph onto the visit graph to create message-passing shortcuts, and Skip, which leaves the visit unaugmented when the base encoder is already confident. To stabilize learning, ReTA employs a decoupled encoder that processes semantic and structural signals in separate channels and fuses them via adaptive gating. Experiments on MIMIC-III and MIMIC-IV across diagnosis prediction, mortality, and readmission show that ReTA consistently outperforms strong baselines while remaining efficient, with larger gains under sparse supervision. ReTA transfers across datasets and knowledge graphs and yields interpretable augmentation patterns.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04032

ROCm InferaSim: Cut LLM Serving Trial Costs and Deploy the Best Configuration

Anshu Raina, Yao Fu, Zhang Jiejing, Peyman Razaghi, Shiksha Patel, Ravi Gupta, Deval Shah, Adeem Jassani 等 14 位作者

原文摘要

Serving an LLM efficiently depends on the engine, how the model is split across GPUs, precision, scheduler settings, speculative decoding, KV-cache management, and prefill-decode disaggregation. The best configuration depends on the model, hardware, workload, and objective. Operators usually find it by trial and error. Exhaustively measuring the candidates in our study would take about 25,000 GPU-hours. ROCm InferaSim predicts serving performance from a few short measurements. Choices such as batch size, prompt length, and GPU count change the amount of work per GPU without changing the kernels. ROCm InferaSim benchmarks each distinct set of kernels once, on at most one node. We call this measurement an anchor. Analytical laws carry that anchor to other deployments that run the same kernels. For multi-node deployments, ROCm InferaSim reduces the configuration to a single-node equivalent, measures it, and scales the result analytically. A simulator replays multi-turn traffic, and a tuning agent searches configurations using the predictions. On AMD Instinct MI355X GPUs with vLLM, SGLang, and ATOM, 11 anchors took 9.1 GPU-hours, against 204 GPU-hours to measure the 226 fixed-length configurations they cover. On 172 configurations, time per output token has 8.1% median error. On a multi-turn coding workload, throughput is within 8%. For 159 disaggregated deployments, none used as an anchor, throughput is within 5.5%. A five-candidate short list contains a deployment within 12% of the best feasible pairing of model and latency target.

cs.AI · cs.DC

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2602.00950

MindGuard: Guardrail Classifiers for Multi-Turn Mental Health Support

António Farinhas, Nuno M. Guerreiro, José Pombal, Pedro Henrique Martins, Laura Melton, Alex Conway, Cara Dochat, Maya D'Eon 等 9 位作者

原文摘要

Large language models are increasingly used for mental health support, yet their conversational coherence alone does not ensure clinical appropriateness. Existing general-purpose safeguards often fail to distinguish therapeutic disclosures from genuine clinical crises, and progress on developing better ones is bottlenecked by a lack of clinically grounded evaluation resources. To address this gap, we introduce a risk taxonomy, developed in collaboration with PhD-level clinical psychologists, that identifies actionable harm (self-harm and harm to others) while preserving space for safe, non-crisis therapeutic content. We release MindGuard-testset, a dataset of multi-turn conversations annotated at the turn level by clinical experts, which is, to our knowledge, the first public benchmark with turn-level clinical risk labels for multi-turn mental health support. We also release an automated red-teaming (ART) framework designed to measure how safety classifiers affect downstream model behavior in adversarial multi-turn interactions. Using synthetic dialogues generated via a controlled two-agent setup, we train MindGuard, a family of lightweight safety classifiers (with 4B and 8B parameters). Our classifiers reduce false positives at high-recall operating points and, when paired with clinician language models, help achieve lower attack success and harmful engagement rates compared to general-purpose safeguards. We release all models, training and human evaluation data, and the ART framework.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04027

A Practical Causal Framework for Autonomous Mobile Robots: From Simulation to Real-World Human Environments

Luca Castri, Gloria Beraldo, Nicola Bellotto

原文摘要

Causal inference has emerged as a powerful paradigm enabling robots to make safe, efficient decisions by reasoning about the ``why'' behind events. However, its real-world application is hindered by the difficulty of collecting causal data, especially in human-populated environments where training takes time, while users expect immediate operation. To this end, we tackle the challenge of transferring simulation-learnt causal relations to the real world with minimal effort. While retaining the invariant high-level causal structure, we empirically demonstrate the system adapts efficiently to real-world situations by updating only the causal model parameters with a limited set of new observations. Our results show that the parameter update enables rapid sim-to-real transfer. The framework runs efficiently onboard of a mobile robot and significantly outperforms a non-causal baselines navigating successfully and safely in complex human-shared scenarios.

cs.RO · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2604.01002

Query-Conditioned Evidential Keyframe Sampling for MLLM-Based Long-Form Video Understanding

Yiheng Wang, Lichen Zhu, Yueqian Lin, Yudong Liu, Jingyang Zhang, Hai "Helen" Li, Yiran Chen

原文摘要

Multimodal Large Language Models (MLLMs) have shown strong performance on video question answering, but their application to long-form videos is constrained by limited context length and computational cost, making keyframe sampling essential. Existing approaches typically rely on semantic relevance or reinforcement learning, which either fail to capture evidential clues or suffer from inefficient combinatorial optimization. In this work, we propose an evidence-driven keyframe sampling framework motivated by an information-theoretic perspective. We consider maximizing the conditional mutual information between the selected frames and the answer given the query as an idealized objective, and use a modular singleton-gain surrogate to motivate a tractable frame-wise selection criterion. Guided by this motivation, we train a query-conditioned evidence scoring network using annotated answer-supporting segments. A contrastive objective encourages greater aggregate score mass on frames from evidence regions than on background frames, providing a practical evidence-based criterion beyond semantic similarity. Experiments on long-form video understanding benchmarks show that our method consistently outperforms prior sampling strategies under comparable frame budgets. With Qwen2.5-VL-7B under the same 32-frame budget, our method improves over uniform sampling by 10.1 points on LVBench and 4.4 points on the long-video subset of VideoMME.

cs.CV · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2603.15457

Evasive Intelligence: Lessons from Malware Analysis for Evaluating AI Agents

Simone Aonzo, Merve Sahin, Aurélien Francillon, Daniele Perito

原文摘要

Tool-using artificial intelligence (AI) agents are increasingly adopted as systems that plan, observe their environment, and take actions over extended time periods. This shift challenges the current evaluation practice, in which models are tested in restricted, fully observable settings. We argue that such evaluations are vulnerable to a failure mode long known in computer security: a system that behaves benignly when it detects that it is being analyzed. Malware has exploited this for decades, by probing sandboxes, virtual machines, and instrumentation, before revealing its payload. The shared challenge is that the subject under observation can detect the conditions of its evaluation, so the setup used to observe behavior ends up shaping the behavior observed, yielding overly optimistic safety and robustness assessments. Recent AI-safety studies demonstrate pieces of this risk, though only preliminarily and under controlled conditions. Drawing on decades of malware-analysis experience, we outline concrete principles for evaluating AI agents as potentially adversarial: explicitly threat-modeling evaluator-discrimination; testing across harness variants and reporting the worst case; using shadow executions to provide a realistic runtime without irreversible side effects; treating environment probing as a risk signal; and performing unpredictable reassessments after deployment.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.04006

Controlling Action-Head Co-Adaptation to VLM Representation Anisotropy

Ruijie Zhang, Jiayin Zheng, Yequan Zhao, Liyan Tan, Ziyue Liu, Zhengyang Wang, Kunpeng Song, Nathan Zhao 等 11 位作者

原文摘要

Modern learning systems increasingly combine pretrained models with downstream heads. Despite the growing importance of such modular designs, how the downstream model co-adapts to the geometry of pretrained representations remains relatively underexplored. We study this problem through vision-language-action models (VLAs), where a pretrained vision-language model (VLM) conditions an action model for control. We find that the action model progressively concentrates its sensitivity on high-variance dimensions of the VLM outputs. Through controlled experiments across different backbone-head models, we further show that counteracting this sensitivity concentration can be beneficial under representation--task mismatch settings. Motivated by these observations, we introduce \emph{Adaptive Interface Equalizer} (AdaIE), a parameter-free mechanism that reweights pretrained representations according to the evolving downstream sensitivity during training. AdaIE improves VLA success rates on simulation benchmarks, with gains of up to $+5.28$ percentage points (pp) on RoboCasa-GR1 and $+1.68$ pp on RoboTwin 2.0, and achieves an average gain of $+19$ pp across five real-world manipulation tasks. It further generalizes to diffusion policies and world models, improving downstream performance in both settings. Our results highlight interface co-adaptation as an important optimization problem when coupling pretrained representations with downstream models.

cs.AI

查看论文
arXiv ↗PDF ↗