跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+1112新增论文
+875其中 AI 相关
+133已有论文修订
+94新增 AI 导读

arXiv 最近完成扫描:2026-11-19 20:16 · 全站最近入库:2026-11-19 19:09各信源进度 →

统计口径与数据新鲜度

统计 2026-11-18 20:33 — 2026-11-19 20:33(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:2026-11-19 20:16最近发现新增版本:2026-11-19 19:09库内论文最新更新时间:2026-11-19 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。

27,982 已收录论文23,249 AI 相关8,559 有修订112 HF 社区发现

浏览论文库

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 20,663 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-19 20:10(北京时间)
新论文 · v1arXiv:2611.21056

Evaluation-Free Recursive Self-Improvement for Coding Agents

Zhe Ren, Guowei Rong, Yimeng Chen, Nanbo Li, Piotr Piękos, Mateusz Ostaszewski, Dandan Guo, Jürgen Schmidhuber

原文摘要

Recent self-improving coding agents have achieved strong benchmark performance by iteratively modifying their own executable implementations. A common pattern uses downstream task-derived evidence to decide which candidates or lineages continue evolving, which we term evaluate-and-select. Obtaining this signal requires repeatedly executing evolved agents on downstream tasks, so broader exploration incurs increasing evaluation cost, while repeated optimization against a fixed evaluation distribution may encourage benchmark-specific adaptation. We ask: can coding agents self-improve without downstream task evaluation inside the evolution loop? We introduce Explore-then-Integrate (ELITE), which integrates implementation mechanisms rather than selecting a single exploratory child to continue the lineage. At each generation, ELITE proposes diverse focused improvements, implements them independently from the same parent, and inspects the resulting programs to identify complementary, redundant, and conflicting mechanisms before reimplementing a compatible subset as one coherent descendant. No downstream task outcomes or task-solving trajectories of evolved candidates are used to guide evolution; only deterministic checks of structural and operational validity are used inside the loop. On SWE-bench Pro, ELITE improves the initial agent from 60.0% to 85.6%, compared with 86.7% for the Darwin Godel Machine (DGM) and 88.3% for the Huxley-Godel Machine (HGM), while using zero candidate task evaluations and reducing observed wall-clock evolution time by 8.8x and 3.2x, respectively, under the evaluated configurations. In cross-benchmark transfer without further adaptation, ELITE averages 73.3% on SWE-bench Verified and 69.3% on Polyglot. Together, these results demonstrate the feasibility of coding-agent self-improvement without downstream candidate task evaluation in our setting.

cs.CL · cs.NE

新论文 · v1arXiv:2611.21053

System 1 agents: a method for local, reliable and fast agentic AI

Eleftherios Tzanis, Michail E. Klontzas

原文摘要

Agentic artificial intelligence could automate analyses of sensitive clinical data, but current agents depend on large language models (LLMs). Cloud LLMs expose patient data. Local LLMs hallucinate, and an invented value in a tool call propagates through every later step. Here we present a method to build System 1 agents: local agents that call tools without generating text. A deterministic compiler turns each tool's schema and the request into multiple-choice questions whose options come from the request and the schema, then a one-pass decision model answers them. On 1,952 requests to 157 biomedical tools, the best System 1 agent chose the right tool, or correctly declined, for 97.0% of requests, significantly ahead of all 18 local LLM-agent settings (best 95.2%). The System 1 agent filled 97.0% of 6,056 stated argument values. No System 1 call contained invented content, against 5.5-19.7% of LLM calls. The System 1 Agents framework is freely available as open-source software at https://github.com/eltzanis/SystemOneAgents.

cs.AI

新论文 · v1arXiv:2611.21044

Body of Evidence: Triangulating 3D Human Pose with Geometry-Conditioned Depth

Erik Brorsson, Kilian Freitag, Josef Bengtson, Kristofer Bengtsson, Lars Hammarstrand, Knut Åkesson, Fredrik Kahl

原文摘要

Even in multi-camera setups, occlusions can leave individual body joints with only a single usable 2D observation. Such an observation defines a viewing ray but leaves the joint's depth undetermined. We introduce Body of Evidence (BoE), a multi-view 3D human pose estimator that predicts per-camera joint depths alongside 2D joint locations. To reduce monocular depth ambiguity, we propose geometry-conditioned depth prediction, which combines image features with previously reconstructed 3D body geometry as contextual depth cues. We further introduce learnable depth-aware triangulation (LDAT), which augments differentiable algebraic triangulation with predicted depth constraints, enabling reconstruction even from a single contributing camera while preserving the benefits of geometric triangulation when multiple views are available. The reconstructed 3D poses subsequently guide depth prediction in the next iteration, allowing depth estimates and 3D geometry to progressively refine each other. Experiments on CMU Panoptic demonstrate state-of-the-art in-domain performance and superior average performance across evaluated camera subsets and unseen camera arrangements. BoE is particularly effective under sparse camera coverage, outperforming the previous state of the art by 10.6 AP25 points in the challenging three-camera setting.

cs.CV

版本更新 · v3arXiv:2607.11505

Proxy OPD: On-Policy Distillation with Transferable Relative Proxy Update for Large Language Models

Daocheng Fu, Rong Wu, Yu Yang, Jianbiao Mei, Licheng Wen, Xuemeng Yang, Yong Liu, Botian Shi 等 10 位作者

原文摘要

Reward-based reinforcement learning (RL) improves large language models but incurs substantial costs when repeated across targets. On-policy distillation (OPD) reuses experts by matching their distributions on target-generated trajectories, yet differing base policies can confound reward-induced updates with pre-existing policy differences, potentially distorting transferred update directions. We propose Proxy OPD (P-OPD), a post-training method that transfers reward-induced relative policy updates using the target's frozen base policy as a reference. A proxy undergoes reward-guided exploration, yielding token-level log-probability changes between its initial and optimized policies. These signals guide learning on target-generated trajectories, with calibration controlling signal absorption, enabling reuse of reward-induced improvements across targets. Experiments on Qwen3-family targets in mathematical reasoning, code generation, and instruction following show that P-OPD improves targets and outperforms conventional OPD in most settings. For 1.7B-to-8B transfer, average Math Mean@16 and Code Mean@8 scores exceed OPD by 8.3 and 3.4 points, respectively. Mechanism analyses show more faithful reward-induced supervision and controllable adaptation. Cross-scale and cross-generation experiments confirm signal transferability and reusability, while multi-domain distillation reveals synergistic gains and cross-domain interference. In the measured 4B-to-8B setting, target adaptation after proxy exploration takes approximately 15\% of direct target RL time; repeated signal reuse amortizes the one-time exploration cost across compatible models. These findings support relative policy updates as effective, transferable, and composable supervision for an ``explore once, transfer multiple times'' post-training paradigm.

cs.LG · cs.AI

版本更新 · v2arXiv:2605.06196

The Granularity Axis: A Micro-to-Macro Latent Direction for Social Roles in Language Models

Chonghan Qin, Xiachong Feng, Ziyun Song, Xiaocheng Feng, Jing Xiong, Lingpeng Kong

原文摘要

Large language models (LLMs) are prompted to adopt social roles from individuals to institutions, but whether their representations encode this difference in social granularity remains unclear. We discover a contrast-based Granularity Axis, the difference between mean macro- and micro-role hidden states, that aligns with the dominant geometry of role space: in Qwen3-8B, its cosine with PC1 is $0.972$, and PC1 explains $52.6\%$ of the variance. Across 75 roles at five levels, with 90,000 role-conditioned and 1,200 default responses per model, projections increase monotonically and remain stable across layers, prompts, and response-quality filters. The structure transfers to Llama-3.1-8B-Instruct and scales to Qwen3-32B and Llama-3.1-70B-Instruct, whose PC1 shares are $45.99\%$ and $31.23\%$. Independent five-rater validation ($α=0.961$) preserves the ordering; identical continuations and affect/formality controls confirm scale-related variation beyond generated wording. The axis is also behaviorally causal: positive steering moves Llama from $2.00$ to $3.17$ on a five-point macro scale for prompts admitting local responses. Steering strength and stability depend on the model's default operating regime. Together, these findings establish social granularity as a representational primitive: a continuous, ordered, causally manipulable direction organizing role-conditioned generation across model families.

cs.AI · cs.CL

新论文 · v1arXiv:2611.21040

Towards wearable-scale, transition-aware sleep staging for clinical populations

Suvadeep Maiti, Keya Sinha, Raju S. Bapi

原文摘要

Automatic sleep staging models are predominantly developed on healthy sleepers and generalize poorly to clinical populations with fragmented sleep. Across three cohorts (patients with stroke, isolated REM sleep behaviour disorder, and other sleep disorders), we observe 2-3x more awakenings per hour, nearly twice as much N1, and less REM sleep than in healthy sleepers. We propose a label-efficient teacher-student framework for wearable-scale sleep staging. SleepGPT (136M parameters), adapted using labels from 25% of training patients, supervises a causal CNN-GRU (31k parameters; 122 KB) that stages EEG, EOG, or both in real time. We introduce Gcv-TAD, a new black-box transition-aware distillation loss. It emphasizes uncommon stage transitions while discounting those that vary across patients, using a prior estimated from labelled patients. In patient-wise 10-fold cross-validation, Gcv-TAD improves per-patient macro F1 over supervised training on the same labels by +3.2 to +7.6 points on average. Gains are largest on N1, and performance approaches full-label training.

cs.HC · cs.AI

版本更新 · v2arXiv:2606.02355

SIRI: Self-Internalizing Reinforcement Learning with Intrinsic Skills for LLM Agent Training

Zhongyu He, Yuanfan Li, Fei Huang, Tianyu Chen, Siyuan Chen, Xingyang Li, Meng Hsuan Yu, Xiangrong Liu 等 12 位作者

原文摘要

Long-horizon LLM agents can benefit from reusable skills, yet existing skill-based methods often rely on external skill generators during training or persistent skill retrieval at inference, increasing engineering complexity, context length, and deployment latency. We propose Self-Internalizing Reinforcement learning with Intrinsic skills (SIRI), a three-phase framework that enables agents to discover, validate, and internalize skills without external skill generators or inference-time skill banks. SIRI first warms up the policy with GiGPO to acquire basic interaction ability and collect successful skill-free trajectories. It then performs self-skill mining, where the current policy summarizes compact skills from its own successful plain rollouts and validates them through paired skill-augmented and skill-free rollouts. Finally, SIRI distills only beneficial skill-guided action tokens into the plain policy using trajectory-level utility and action-level advantage. At inference, the agent runs with the original prompt only. On ALFWorld and WebShop with Qwen2.5-7B-Instruct, SIRI improves GiGPO from 0.908 to 0.930 on ALFWorld and from 0.728 to 0.813 on WebShop, outperforming prompt-based, RL-based, and memory-augmented baselines. Further analysis shows that our self-mining strategy can achieve performance comparable to distillation with closed-source large model.

cs.AI · cs.LG

新论文 · v1arXiv:2611.21036

Non-Myopic Safety Filtering with Learned Intervention Cost-to-Go

Johannes Autenrieb, Mark Spiller, Daniel Silvestre, Namhoon Cho, Jason J. Choi

原文摘要

Control barrier function (CBF) safety filters minimize instantaneous deviation from a nominal control command, but a sequence of pointwise-minimal interventions need not minimize cumulative intervention effort. This paper proposes a discrete-time non-myopic safety filter using a learned approximation of the optimal intervention cost-to-go over CBF-admissible policies. The approximation enters through a relaxed Bellman-residual constraint, while the original barrier condition remains hard. Feasibility is equivalent to that of the conventional CBF filter, and forward invariance remains independent of approximation accuracy under the stated assumptions. We derive a trajectory-wise performance certificate that requires neither the optimal value function nor an approximation-error bound. Under additional approximation and horizon assumptions, a suboptimality bound contains an explicit endpoint approximation-error term together with accumulated tolerance and relaxation terms. Numerical results for a collision-avoidance problem show an approximately 50% reduction in mean cumulative intervention effort relative to conventional CBF filtering over 100 initial conditions not used during training or parameter selection. The improvement is consistent across three independently trained networks, while maintaining the prescribed safety constraints and requiring less online computation than the tested predictive filter.

eess.SY · cs.LG

新论文 · v1arXiv:2611.21035

FlowRow: Learning Navigation Behaviors in Agricultural Environments with Long-Horizon Goal-Conditioned Flow Matching Policies

Utkarsh Bajpai, Cédric Pradalier, Stéphanie Aravecchia

原文摘要

Autonomous monitoring and intervention in semi-structured agricultural environments, such as orchards and vineyards, require mobile robots to navigate to locations across diverse field configurations. This entails performing navigation behaviors including in-row driving, row turning and skipping, and inter-row traversal. Recent learning-based approaches have enabled robust in-row navigation from onboard perception, but navigation across rows typically requires a priori knowledge of field geometry to specify and tune row-transition maneuvers, preventing arbitrary transitions between rows. We address this challenge by framing navigation in these environments as a goal-directed navigation task, where a long-horizon goal is expressed relative to the robot and a planner generates local paths based on the goal and the robot's perceptual observations. We present FlowRow, a flow-matching-based local planner which models the distribution of diverse agricultural navigation behaviors as $\mathrm{SE}(2)$ paths using the long-horizon goal and depth perception. To enable systematic evaluation, we introduce RowNavBench, a navigation benchmark for semi-structured agricultural environments developed in Isaac Sim, comprising 900 episodes across grape vineyards, lemon orchards, and apple orchards with varying row lengths, row and plant spacings, plant positions, and weed growth. FlowRow achieves 87% success, compared to 44% for the strongest generative behavior cloning baseline retrained on the same demonstrations. We will release our benchmark, data, and code upon acceptance.

cs.RO

版本更新 · v2arXiv:2611.18099

Eco3R: Pinned-Key Inference for Efficient and Streamable Feed-Forward 3D Reconstruction from Long Videos

Hongsen Cao, Shanxin Yuan, Mona Jaber, Ahmed Sayed

原文摘要

Feed-forward multi-view transformers such as VGGT recover the cameras and geometry of a video in one pass, but the cost of their cross-frame self-attention grows quadratically with the frame count, and their accuracy drops on long videos. Most remedies approximate attention over all frames under a fixed computational budget or select a few keys by their content. We hypothesize that the keys of a long video can be selected independently of video content, and our experiments on VGGT and $π^3$ support it: a small number of uniformly sampled frames can serve as the keys of every frame. We present Eco3R, a training-free inference framework whose keys form a Pinned Key Set of 31 such frames. A pin pass stores their tokens in a Scene Cache, and the other frames read it in independent batches that reproduce joint inference with these keys exactly, at a constant cost per frame. An Online Stream updates the pinned frames during recording. Six middle global blocks align the frames and form a reusable VGGT module: the Alignment Score reads their attention to flag failures, and Alignment Refocusing reduces them by fine-tuning only these blocks (3.6 GPU hours). Eco3R ($π^3$) matches $π^3$'s success rate on 1,000-frame ScanNet videos while running 13.8 times faster. With refocusing, Eco3R (VGGT) raises VGGT's success rate from 70% to 82% while running 8.5 times faster on 500 frames. Eco3R can reconstruct sequences of up to 4,000 frames on a single A100 GPU, while Eco3R ($π^3$) streams 1,000 frames at 18.6 fps, achieving the lowest online absolute trajectory error.

cs.CV · cs.AI

新论文 · v1arXiv:2611.21033

POKE: Probe Offline, Keep Efficiently for Vision-Language-Action Models

Jaehyuk Lee, Yeajin Lee, Suhyun Bae, Donghun Lee

原文摘要

Visual token pruning reduces the inference cost of Vision-Language-Action (VLA) models, but attention-based scores do not directly measure a token's effect on the predicted action. We define action sensitivity as the magnitude of the first-order action change when a token is replaced by its camera's mean token and estimate it for all tokens with a single backward pass. Computing it online is costly: scoring on the full (unpruned) graph needs an extra forward pass at every control step. Scoring on the pruned forward pass is cheaper but causes scoring-graph lock-in: removed tokens lose their direct paths through later layers and tend to receive lower scores, making those locations less likely to return. On replayed observations, even with a random conditioning mask, pruned-graph scoring re-selects 59% of the mask's tokens, versus 6% expected by chance; attention histories show a similar bias. POKE (Probe Offline, Keep Efficiently) computes full-graph action sensitivities offline and distills them into a student with under 0.3M parameters. At inference, the student scores the current observation without backpropagation or previous masks, and score-guided diversity selection retains the tokens. Keeping 32 of 512 visual tokens on LIBERO, POKE achieves 95.2% success on OpenVLA-OFT and 90.8% on $π_{0.5}$ at $1.85\times$ and $1.26\times$ speedups, outperforming attention-, diversity-, and distillation-based pruners, with larger margins on LIBERO-Plus.

cs.CV

版本更新 · v2arXiv:2611.11995

Infrathon: Benchmarking Language Agents for Real-World, Long-Horizon AI Infra Optimization

Zhengyu Hu, Zheyuan Xiao, Yike Wang, Hamish Ivison, Haoyu Wang, Yige Yuan, Rulin Shao, Yuetai Li 等 21 位作者

原文摘要

In this work, we study whether frontier agents can continually improve real-world AI infrastructure over long horizons. We introduce Infrathon, a benchmark of diverse, long-horizon AI infrastructure optimization tasks across different levels. We evaluate six models across 150 runs with 24-hour budgets, totaling over 10,000 GPU-hours. Our results show that model rankings vary with the evaluation horizon, while 135 of the 150 runs achieve further improvements during the second half of their budgets. These findings indicate that shorter evaluations can overlook both optimization gains and changes in relative model performance. We also find agents can discover solutions that outperform human-designed baselines: Claude Opus 5.5 increases the mean KV-cache compression ratio from $6.82\times$ with the human-designed Quant-VideoGen codec to $19.86\times$. Analysis of agent trajectories identifies task violations, unreliable testing, and premature completion as barriers to sustained improvement. Together, our Infrathon framework provides resources for constructing long-horizon tasks, evaluating frontier agents, and discovering good infrastructure solutions. Code is available at https://github.com/InfrathonX/Infrathon

cs.AI · cs.CL

版本更新 · v2arXiv:2511.15245

The Walls Have Ears: Unveiling Cross-Chain Sandwich Attacks in DeFi

Chuanlei Li, Zhicheng Sun, Jing Xin Yuu, Xuechao Wang

原文摘要

Cross-chain interoperability is a core component of modern blockchain infrastructure, enabling seamless asset transfers and composable applications across multiple blockchain ecosystems. However, the transparency of cross-chain messages can inadvertently expose sensitive transaction information, creating opportunities for adversaries to exploit value through manipulation or front-running strategies. In this work, we investigate cross-chain sandwich attacks targeting liquidity pool-based cross-chain bridge protocols. We uncover a critical vulnerability where attackers can exploit events emitted on the source chain to learn transaction details on the destination chain before they appear in the destination chain mempool. This information advantage allows attackers to strategically place front-running and back-running transactions, ensuring that their front-running transactions always precede those of existing MEV bots monitoring the mempool of the destination chain. Moreover, current sandwich-attack defenses are ineffective against this new cross-chain variant. To quantify this threat, we conduct an empirical study using cross-chain transaction data from the Symbiosis protocol (November 21, 2022 to April 4, 2026) and Debridge (December 21, 2022 to April 4, 2026), along with a tailored heuristic detection model. Our analysis indicates that this attack has indeed had a significant impact on DeFi.

cs.CE

新论文 · v1arXiv:2611.21029

Sparse Attention for Visual Generation: Toward Algorithm-System Co-Design

Majun Zhang, Yuan Guo, Decheng Wu, Jiayang Xu, Changhao Pan, Fan Zhuo, Xiaoda Yang, Chenyuhao Wen 等 9 位作者

原文摘要

Rapid advances in image and video generation have enabled high-fidelity synthesis at higher resolutions and longer durations, making the quadratic cost of dense self-attention a major bottleneck as visual token sequences grow. Sparse attention seeks to preserve generation quality at lower computational cost by retaining informative query-key interactions and omitting less consequential ones, with end-to-end gains depending on the cost of constructing sparse support and the efficiency with which the system maps the resulting workload to hardware. We survey sparse attention for visual generation through a 4+1+3 analytical framework that links algorithmic design to system implementation. The framework examines how static and dynamic methods use source information and the evidence derived from it to decide which interactions to retain and at what granularity, while accounting for the roles of reuse in reducing repeated computation and correction in compensating for omitted contributions. We relate these algorithmic choices to their implementation on hardware through the framework's three system dimensions of data organization, execution, and dataflow. We summarize common evaluation metrics and discuss open challenges and research directions, providing researchers and practitioners with a reference for the joint design of sparse attention algorithms and systems.

cs.CV

新论文 · v1arXiv:2611.21696

Layerwise Learning-Rate Scaling in Vision Networks: Effective-Rate Identification, Power-Mean Aggregation, and Budget Dependence

Chandini Vysyaraju, Raghuvir Duvvuri, Radu Timofte, Dmitry Ignatov

原文摘要

Layerwise learning-rate scaling assigns different learning rates to different parts of a network. Although established in fine-tuning pretrained models, its effects in vision networks trained from scratch are commonly assessed by training each candidate recipe. We show that, in this setting, the rate-related component of a layerwise strategy is well summarized by an effective learning rate computable directly from its multiplier profile. Across matched-control comparisons, changes in test accuracy closely track changes in training loss (r = -0.996), which supports an optimization-progress explanation of the short-budget rate effect. Increasing the untuned base learning rate tenfold shifts the rate-matched contrast upward in all six tested architectures (p = 0.016) and increases the number of architectures in which a published discriminative schedule, excluded from model fitting, outperforms the uniform control from two to five of six (p = 0.031). Within the evaluated strategy space, a parameter-weighted power mean with exponent 0.35 (95% confidence interval [0.20, 0.50]) accounts for 42.6% of within-architecture variance, compared with 36.4% for the arithmetic mean. Ranking candidates by this scalar without training improves the selected candidate's mean percentile rank from the 26th to the 10th and reduces its mean accuracy shortfall relative to the best candidate from 3.70 to 1.12 points. The rate-related effect decays as the training budget increases; relative to tuned baselines, the overall strategy effect falls within the prespecified practical-equivalence margin in two of three architectures. A separate grouping-related contrast can persist at long budgets, although its direction and magnitude depend on the architecture.

cs.LG

版本更新 · v2arXiv:2606.31502

Fully Automated Deep Learning Segmentation of RPE Loss and Ellipsoid Zone Integrity in Geographic Atrophy on Real-World OCT

Wolf-Dieter Vogl, Hlynur Skulason, Oliver Leingang, Ursula Schmidt-Erfurth, Amir Sadeghipour, Ariadne Whitby

原文摘要

Purpose: To develop and validate a fully automated deep learning method for segmenting retinal pigment epithelium (RPE) loss, ellipsoid zone (EZ) loss, and EZ thickness in geographic atrophy (GA) on optical coherence tomography (OCT). Methods: Three convolutional neural networks segment Bruch's membrane, the EZ boundaries, and RPE loss in Spectralis OCT volumes; EZ loss and thickness are derived from the EZ boundaries. The networks were developed on 298 volumes (GA, intermediate and neovascular age-related macular degeneration, healthy) and evaluated against double-graded expert annotations in an independent real-world set of 43 eyes with GA. Further analyses assessed inter-reader reliability (25 eyes), test-retest repeatability between screening and baseline visits of two phase 3 trials (68 eyes), and the effect of B-scan density (61 eyes). Results: Mean Dice similarity coefficients were 0.88 (95% confidence interval [CI], 0.84-0.90) for RPE loss and 0.87 (95% CI, 0.83-0.89) for EZ loss; the mean absolute EZ thickness error was 2.08 $μ$m. Agreement with an independent reader group was similar to inter-reader agreement. Test-retest intraclass correlation coefficients were 0.995 (RPE loss) and 0.988 (EZ loss). Lower B-scan densities had little effect on lesion area but markedly increased the error in measuring 3-month change. Conclusions: On real-world OCT, the fully automated method was accurate, with agreement comparable to human readers, and reliable, with high test-retest repeatability. Translational Relevance: Fully automated, reproducible quantification of RPE and photoreceptor loss on routine OCT could support standardized structural endpoints in GA clinical trials and objective assessment of GA in routine ophthalmic care.

cs.CV

版本更新 · v4arXiv:2610.26396

The Exhaustive Peres Transform: Extraction and Universal Coding

Aur{é}lien Garivier

原文摘要

Run to exhaustion, Peres's iteration of von Neumann's procedure yields a bijection of {0,1}^N that splits every string into fair bits and a residual. The residual is self-delimiting and determines the number of ones, so the fair bits are exactly uniform given the residual under every exchangeable law. Each residual one accounts for a power of two of the weight of the input, and the residual of every input of length N has fewer than sqrt(3) N^0.793 ones. This sparsity gives a short proof, with an explicit remainder, that the fair bits per symbol tend almost surely to the entropy for Bernoulli sources and every finite alphabet. The Bernoulli likelihood factorises along the recursion, and code B, which codes the residual at plug-in biases, has the length of the two-part code. Code G gamma-codes the gaps of the residual, with no estimate and no arithmetic coder: on every string its length is at most the logarithm of the type class plus O(N^0.793 log N). What is more, a concatenation of J pieces costs its piecewise empirical entropy plus a term that is sublinear when J = o(N), without the cuts being sent. After a Burrows-Wheeler transform of the reversed string, both codes are universal for stationary ergodic sources: code B through a context tree of growing depth, with regret O(2^k log N) against Markov sources of order k, and code G with no tree, competing on every individual sequence with the empirical conditional entropy of every order. For laws invariant under permutations inside the pieces of a partition, the fair bits not mixed across the pieces are exactly uniform given everything else; after the BWT, the fair bits of every stationary source fall short of their length by o(N) bits of entropy, with an exponential tail for Markov sources. Experiments measure the finite-length price of these guarantees.

cs.IT

版本更新 · v3arXiv:2607.19358

LISA: Linear-Indexed Sparse Attention for Efficient Long-Context Reasoning

Yu Zhao, Zekun Zhang, Fan Jiang, Bo Zeng, Xintong Wang, Chonghan Liu, Shimin Shan, Yichao Du 等 9 位作者

原文摘要

Long chain-of-thought (CoT) reasoning models such as DeepSeek-R1 keep lengthening their inference chains under test-time scaling, yet standard self-attention incurs quadratic cost in context length, making long-CoT deployment costly in production. We therefore introduce LISA (Linear-Indexed Sparse Attention), a plug-and-play replacement for self-attention that requires no full pretraining. LISA combines a linear-attention branch for long-range memory with a Lightning Indexer that dynamically selects the most relevant tokens for a sparse self-attention branch. The two branches are fused through a learned gate, so the model retains broad context while focusing computation on informative tokens. We train LISA in two stages. In Stage 1, a linear-attention branch is integrated with sliding-window attention and distilled from a frozen teacher's full-attention distribution. In Stage 2, the Lightning Indexer replaces the static window and is trained with a per-head KL-divergence objective to mimic teacher attention patterns. The resulting model reduces eager-mode self-attention computation from quadratic in sequence length to linear in the number of generated tokens and the selected context budget. Experiments on DeepSeek-distilled Qwen models show that LISA cuts measured eager-mode mean sample latency by 23.9% at 16K context without degrading long-context reasoning quality. Across nine mathematical and code reasoning benchmarks, the gated LISA-1.5B model improves the equal-weight benchmark average by 4.2 percentage points relative to the full-attention baseline.

cs.AI

新论文 · v1arXiv:2611.21020

SWE-OPD: On-Policy Distillation for Agentic Coding

Yuhang He, Yanyu Chen, Chuanzhe Guo, Jie Zhu, Zichen Tian, Hongyu Ge, Zixuan Ren, Huang He 等 14 位作者

原文摘要

In this work, we study on-policy distillation (OPD) for transferring agentic coding capabilities from a large teacher model to a smaller student model. Prior work has focused primarily on reasoning and short-horizon agentic tasks. We present SWE-OPD, a minimal OPD study for software engineering, where the student interacts with executable repositories and the teacher supervises the prefixes visited by the student. Across three teacher--student pairs, two training sources, and three SWE benchmarks, we find that on-policy supervision transfers coding capability: supervision from Qwen3.6-27B increases the SWE-bench Pro resolve rate of Qwen3.5-9B from 35.9% to 39.4%. Refreshing trajectories is not strictly required either, as an offline variant with cached teacher scores still improves over the student baselines while cutting end-to-end GPU-hours by 59.4%--72.6% and halving peak GPU allocation, at benchmark-dependent accuracy. Furthermore, we extend the offline variant to corss-tokenizer scenarios: distilling GLM-4.5-Air from DeepSeek-V4-Flash-0731 raises its SWE-bench Pro and Multilingual resolve rate from 35.8% to 39.7% and from 41.8% to 46.2%, around 4% and 4.6% improvement, respectively. We also find that a decreasing loss, however, does not necessarily indicate improved agent performance: under top-$K$ reverse-KL loss, one student exhibits severe repetition despite a decreasing loss. We identify a support-mass gradient term as a potential mechanism underlying this failure.

cs.AI · cs.LG

新论文 · v1arXiv:2611.21019

MemDoctor: Stage-wise Error Attribution and Repair for MLLM Agent Memory

Ruiyang Jiang, Xinyang Liao, Weikai Huang, Yuqi Niu, Junjie Zhou, Yubin Zheng

原文摘要

Multimodal large language models (MLLMs) now assist users across sessions of images and text, carried by an external memory of what was said and shown. When such an assistant answers wrongly, the failure usually began earlier, in how a record was written, updated or retrieved, and the answer alone does not say where. We formalize multimodal memory as four stages, Write, Update, Retrieve and Answer, and define nine error types across them. A stage-resolved analysis on Mem-Gallery traces most failures to a source stage, and most of those lie before the answer. Injecting known errors against fact-preserving control edits, we then follow each tainted record downstream. Early errors survive to the answer as apparent evidence, and hurt most when no other record holds the fact. A diagnosis that trusts the quoted evidence therefore blames the stage where the error surfaces, not the one where it began. We propose MemDoctor, in which the MLLM that built the memory diagnoses its own failures. It traces each needed fact backward through the four stages, accepts a diagnosis only when the cited evidence is found in memory, and repairs that stage before answering again. On a 4B backbone, MemDoctor improves accuracy on both Mem-Gallery and MemLens, outperforming all evaluated 8B baselines, and the same loop also improves MuRAG and A-Mem.

cs.CL

版本更新 · v4arXiv:2602.05302

PieArena: Ranking and Profiling Language Agents in Realistic Negotiation Scenarios

Chris Zhu, Sasha Cui, Will Sanok Dufallo, Runzhi Jin, Zhen Xu, Linjun Zhang, Daylian M. Cain

原文摘要

We present an in-depth evaluation of LLMs' ability to negotiate, a central business task that requires strategic reasoning and theory of mind to create and claim economic value. To do so, we introduce PieArena, a large-scale negotiation benchmark grounded in multi-agent interactions over realistic scenarios adapted from MBA negotiation courses at an elite business school. We evaluate language agents across three pairing regimes: mirror-play, cross-play, and human-LM play. We develop a ranking model for continuous negotiation payoffs that yields order-invariant, uncertainty-quantified leaderboards while correcting for systematic experimental asymmetries. We further study the effects of shared-intentionality agentic scaffolding and find asymmetric gains, with large improvements for mid- and lower-tier LMs and diminishing returns for frontier LMs. As calibration anchors, we collect human-human and human-LM negotiation data from trained business school students, finding that a representative frontier language agent matches or exceeds this human baseline in our evaluation settings. Beyond deal outcomes, PieArena provides a multi-dimensional behavioral profile that reveals cross-model heterogeneity in instruction compliance, computation accuracy, as well as judge-assessed deception and reputation, illustrating the value of evaluation beyond outcome-only leaderboards.

cs.AI

新论文 · v1arXiv:2611.21015

A physics-constrained emulator for time-discretised landslide runout modelling

Ye Chen, Lorenzo Nava, Maximillian Van Wyk de Vries, Fawu Wang

原文摘要

Numerical simulations of landslides are commonly used to forecast impact areas. Full physics-based models provide accurate predictions, but their computational expense can preclude their use for large parameter sweeps. To achieve efficient modelling for such cases, we developed a physics-constrained neural network to emulate full landslide runout in a time-discretised manner. This emulator is trained on over 200,000 samples, each consisting of a pair of landslide thicknesses and velocities on a given DEM with a fixed time interval. To constrain the model's physical consistency, we define physics-based loss functions based on mass conservation and the system's kinetic energy loss. The model demonstrates high fidelity across the test dataset, achieving root mean square errors of approximately 0.06 m for thickness and 0.11 m/s for velocity in single-step predictions. In autoregressive rollout, these errors evolved modestly to about 0.21 m and 0.09 m/s after 110 iterations, while greatly improving computational efficiency by more than 50 times.

cs.LG · physics.geo-ph

新论文 · v1arXiv:2611.21009

Learning and Verifying Stabilizing Controllers with Structured Neural Control Lyapunov Functions

Jose Luis Ledesma Martin, Patricia Pauli

原文摘要

This work investigates the joint learning of neural controllers and control Lyapunov functions (CLFs) for nonlinear discrete-time systems. We exploit the structural properties of Input Convex Neural Networks and Polyak-Łojasiewicz Networks to guarantee positive definiteness of CLF candidates by construction, reducing the conditions requiring formal verification. We develop a mixed-integer programming framework to formally verify the remaining stability conditions and introduce an iterative region of attraction (ROA) expansion procedure to enlarge the certified ROA without retraining. Experiments on an inverted pendulum and a path-tracking system demonstrate the advantages of structured neural CLFs, particularly PLNets, and illustrate how their structural bounds influence the geometry of the certified regions. Our code is available at: https://github.com/JLedesMart/Structured-Neural-Control-Lyapunov-Functions.

eess.SY · cs.LG

版本更新 · v2arXiv:2603.27460

Project Imaging-X: A Survey of 1000+ Open-Access Medical Imaging Datasets for Foundation Model Development

Zhongying Deng, Cheng Tang, Ziyan Huang, Jiashi Lin, Ying Chen, Junzhi Ning, Chenglong Ma, Jiyao Liu 等 126 位作者

原文摘要

Foundation models have demonstrated remarkable success across diverse domains and tasks, primarily due to the thrive of large-scale, diverse, and high-quality datasets. However, in the field of medical imaging, the curation and assembling of such medical datasets are highly challenging due to the reliance on clinical expertise and strict ethical and privacy constraints, resulting in a scarcity of large-scale unified medical datasets and hindering the development of powerful medical foundation models. In this work, we present the largest survey to date of medical image datasets, covering over 1,000 open-access datasets with a systematic catalog of their modalities, tasks, anatomies, annotations, limitations, and potential for integration. Our analysis exposes a landscape that is modest in scale, fragmented across narrowly scoped tasks, and unevenly distributed across organs and modalities, which in turn limits the utility of existing medical image datasets for developing versatile and robust medical foundation models. To turn fragmentation into scale, we propose a metadata-driven fusion paradigm (MDFP) that integrates public datasets with shared modalities or tasks, thereby transforming multiple small data silos into larger, more coherent resources. Building on MDFP, we release an interactive discovery portal that enables end-to-end, automated medical image dataset integration, and compile all surveyed datasets into a unified, structured table that clearly summarizes their key characteristics and provides reference links, offering the community an accessible and comprehensive repository. By charting the current terrain and offering a principled path to dataset consolidation, our survey provides a practical roadmap for scaling medical imaging corpora, supporting faster data discovery, more principled dataset creation, and more capable medical foundation models.

cs.CV · cs.AI

新论文 · v1arXiv:2611.21006

The Exact Optimal Delivery Scheme for the Cyclically $L$-Separated Placement in Multi-Access Coded Caching

Kai Wan, Huimei Wei, Minquan Cheng, Giuseppe Caire

原文摘要

This paper investigates the $(K, L, M, N)$ multi-access coded caching (MACC) system introduced by Hachem et al. The system consists of a server storing $N$ equal-sized files, $K$ cache nodes, each with a storage capacity of $M$ files, and $K$ users, each accessing $L$ consecutive cache nodes in a cyclic wrap-around topology. The server transmits coded information to the users via an error-free capacitated broadcast channel, such that each user obtains its own requested file as a combination of the server transmission and the accessible caches' content. Our objective is to minimize the worst-case transmission load over all possible user demands. First, for the cyclically $L$-separated placement, we derive a new converse bound using the acyclic index coding bound. By carefully counting all valid cache supports, the proposed converse improves upon an existing bound for the same placement. We then develop a linear coded caching scheme based on a novel determinant-based interference alignment strategy, which achieves the new converse bound exactly. Exploiting the determinant structure of a random generating matrix (each of whose columns corresponds to one user), this scheme simultaneously aligns interference across all users into low-dimensional subspaces, while preserving the decodability of their desired packets. Furthermore, by averaging acyclic index coding bounds over cyclic user permutations, we derive a general converse bound applicable to arbitrary uncoded cache placements. By combining the proposed scheme with memory sharing, we demonstrate that the achievable transmission load is within a factor of two of the global bound under uncoded placement. Finally, theoretical and numerical comparisons demonstrate that the proposed scheme generally achieves a strictly lower load than existing schemes.

cs.IT

新论文 · v1arXiv:2611.21003

K2-Horizon: A Frontier Fleet of Open Source Language Models

K2 Team, Zhengzhong Liu, Haonan Li, Xuezhe Ma, Alexander Moreno, Subham Sekhar Sahoo, Rupesh Kumar Srivastava, Mikhail Yurochkin 等 95 位作者

原文摘要

We present K2-Horizon, a fleet of six fully open language models, from a 0.9B dense model to a 375B-A23B MoE, built for general-purpose reasoning and agentic work across math and science, software engineering, deep research, office automation, and beyond. The models lead or are competitive with strong open models of their size on agentic and coding benchmarks. On the modeling side, to strengthen reasoning, we introduce Mixture-of-Value Attention (MoVA), applying MoE to attention value projections, and softplus attention gating, a more flexible, theoretically motivated gate. Our 36B-A4B model uses both and nearly matches a 32B dense model trained on the same data, with one-seventh the active parameters. We also introduce Uno, lossless diffusion adapters that speed up decoding by up to 2.2x without a separate draft model. Our release follows the LLM360 definition of open source for transparency and reproducibility: the full development history, including intermediate checkpoints from every training stage, pre-training data, training logs, and all training, evaluation, and infrastructure code, notably our production-scale xLLM framework for dense and MoE training and the RL360 reinforcement-learning stack. Beyond the data itself, we release and document in detail every pipeline that built it, the part of model development that open releases most rarely disclose. These pipelines produced roughly 10T synthetic pre-training tokens, showing that synthetic data can be a major ingredient of pre-training at scale. Four of the models share one 22T-token pre-training corpus and recipe, revealing consistent scaling behavior and enabling controlled studies of scale and sparsity. All artifacts are available at https://ifm.ai/k2/

cs.AI

新论文 · v1arXiv:2611.20999

How Can Size, Ceiling, and Time Bounds Affect the Complexity of Nonuniform Automata Families?

Tomoyuki Yamakami

原文摘要

In the past literature, families of two-way finite automata and pushdown automata having limited resource-bounded complexity have been studied in direct connection to (mainstream) space-bounded complexity classes equipped with Karp-Lipton style advice of limited size when all inputs given to the automata have bounded length. We acknowledge three major factors--size, ceiling, and time--of such families, which have a significant impact on the computational/descriptional complexity of finite and pushdown automata families, where the "size" refers to the (stack-)state complexity, the "ceiling" refers to the maximum input length, and the "time" refers to the execution time of underlying automata. In this line of study, we further explore these effects caused by different sizes and ceilings. We exploit automata-based reducibility and translatability to space-bounded advised Turing machines and auxiliary automata.

cs.CC · cs.FL

版本更新 · v3arXiv:2609.35040

GAZEleak: Passcode Inference Against Eye-tracking XR Devices Through External Observation

Hwanjo Heo, Junhee Lee, Jinwoo Kim

原文摘要

Mixed-reality headsets such as Apple Vision Pro replace the touch screen with gaze-as-pointer interaction: the wearer looks at a target and confirms with an air pinch. Because the display is inside the headset and the eye tracker is walled off from third-party software, such input is widely assumed to be unobservable to bystanders, a built-in defense against the shoulder-surfing that plagues phones and laptops. We present GAZEleak, a side-channel attack that recovers gaze-driven input from external video of head motion alone, under a strictly local, physical-observer threat model: the adversary only films the wearer from across the room and installs no software on the device. The attack exploits the centrally coupled eye-head motor program: gaze shifts recruit small, target-dependent head reorientations that project into sub-degree pose changes recoverable from commodity video. GAZEleak implements a measurement-based, sparse-optical-flow inference pipeline for users' 6-digit device passcodes. On a preliminary front-view dataset from three author-subjects who were aware of the attack hypothesis, GAZEleak places the true code within the top ten guesses for 56% (10 of 18) of test codes under a cross-person protocol with no labeled victim data, and for every code of the most exposed subject. The performance is subject-dependent: no passcode from the least exposed subject reaches the top ten, although its median guessed passcode rank is 12,786 rather than 500,000 expected from an uninformative ordering. These results provide preliminary evidence that gaze-coupled head motion can expose passcode information under controlled conditions, while motivating broader evaluation across users, behaviors, and capture settings.

cs.CR

新论文 · v1arXiv:2611.20995

CONGA: Compiling Gaussian Fiber Geometry for Structured Bayesian Inference

Víctor Gallego

原文摘要

In structured Bayesian models, a large latent block is often Gaussian conditional on a smaller set of parameters, yet generic inference methods relearn this geometry during warmup. We study how this structure can be used to construct transports for posterior inference. The conditional mean and precision determine exact affine maps, including symmetric Brenier and triangular Knothe-Rosenblatt maps, that separate a standard normal block from the remaining non-Gaussian core. KL and relative Fisher objectives then reduce to objectives on this core. We introduce CONGA, a compiler that identifies candidate Gaussian blocks and selects a temporal, spectral, or sparse implementation according to their structure and computational cost. Marginalization requires an identity established from the model; numerical evidence alone permits only a target-preserving change of variables, and a failed check leaves the model unchanged. On the Stock-Watson inflation model, CONGA improves minimum ESS per second by 12.5x for NUTS. It also restores convergence where the raw target fails: for WALNUTS on the same model, and for NUTS on a 6760-dimensional electricity load model reduced to 34 core coordinates. Experiments on graph-indexed fields show how coloring and fill affect the choice of map, including cases where retaining the original target is preferable. Code at https://github.com/vicgalle/conga

stat.CO · cs.LG · stat.ML

新论文 · v1arXiv:2611.20994

Development of an HPLC Analytical Method Based on Large Language Models:A Case Study of the Eczema External Wash Formula

Kexin Guo, Xueting Sun, Xinyan Zhao, Wei Wang, Xingchu Gong

原文摘要

Traditional Chinese medicine (TCM) formulas feature complex chemical compositions, and their HPLC quality control method development is time-consuming and experience-dependent. Large language models (LLMs) provide a promising strategy for intelligently assisting analytical method construction. Taking the Eczema External Wash Formula as the research object, this work constructed a human-machine collaborative workflow involving intelligent recommendation, experimental validation and iterative optimization for HPLC method development. Among five prompt strategies tested, expert-role setting with structured reasoning performed best. After inputting formula information, the LLM proposed five chromatographic schemes. Through experimental screening and optimization, the final HPLC method used a SunShell C18 column, mobile phase of 0.1% aqueous formic acid-acetonitrile for gradient elution, with seven markers separated within 55 min. Twenty components were characterized by LC-MS. The validated method exhibited satisfactory linearity, precision, repeatability, stability and recovery. This indicates LLMs can reduce the workload of chromatographic condition exploration and boost the development efficiency of TCM formula analytical methods, supporting intelligent TCM quality control.

cs.HC