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：2026-11-08 21:48 · 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 2026-11-07 22:09 — 2026-11-08 22:09（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：2026-11-08 21:48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 14,661 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-08 21:46（北京时间）
版本更新 · v2arXiv:2610.38210

GAAC: Lossless KV Compression for Bandwidth-Efficient LLM Inference on Personal Devices

Yipin Guo, Arun M George, Siddharth Joshi

原文摘要

Weight-only, lossy quantization substantially reduces the storage and memory requirements for serving large language models (LLMs), but the growing key-value (KV) cache remains a bottleneck for long-context inference on personal devices. We present Group-Adaptive Additive Codec (GAAC), a lossless KV compression and serving system that exploits exponent redundancy to reduce storage and accelerate attention without introducing KV approximation. GAAC encodes BF16 exponents using calibrated per-channel base-plus-offset codes over contiguous ranges. GAAC adapts code widths across small groups while preserving KV values bit-for-bit. A fused attention kernel reconstructs KV tiles on the fly, avoiding full-cache decompression. Block-wise compression and fixed-capacity slots support incremental cache updates and paged serving. Across five tested models, GAAC achieves 1.405-1.407$\times$ lossless BF16 KV compression. We also study compression performance on low-precision tensors FP8 E4M3 caches ( 1.09-1.14$\times$) and E5M2 caches (1.25-1.30$\times$). On NVIDIA DGX Spark, GAAC accelerates decode attention at a 64K-token context by 1.19-1.27$\times$ and end-to-end W4A16 decode steps by 1.04-1.15$\times$. In serving, it reduces resident KV memory by approximately 26% and increases aggregate throughput by 13-29% under comparable KV-memory budgets. On Apple M4, GAAC achieves 1.02-1.25$\times$ attention speedups, demonstrating portability beyond NVIDIA GPUs. Code is available at https://github.com/Intelligent-Microsystems-Lab/GAAC.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.33857

TrackTok: Object-Centric Video Tokenization with Semantically Persistent Tokens

Nhi Pham, Christopher Wewer, Bernt Schiele, Jonas Fischer, Jan Eric Lenssen

原文摘要

We introduce TrackTok, an efficient video tokenization framework that produces object-centric tokens, i.e., tokens that consistently encode the same semantic objects across time, even as it moves, deforms, or undergoes partial occlusion. Conventional patch-based or existing holistic tokenizers produce token identities often tied to fixed image locations. In contrast, our approach learns temporally persistent object-centric representations that bind visual features across frames into stable semantic units. This enables substantially more compact, continuous latent video representations, reducing the number of tokens required for downstream generative modeling with state-of-the-art flow models, while preserving the semantic structure needed for high-quality synthesis. Our tokenizer offers a favorable quality-efficiency trade-off. TrackTok achieves comparable class-conditional generative FVD on UCF-101 dataset at 8 times fewer latent tokens and better latency, and outperforms existing 1D tokenization baselines on frame prediction task on Kinetics-600 across low token budgets. These results suggest that object-aligned and temporally persistent tokenization is a promising direction for scalable video generation, where reducing latent token counts is critical for efficient modeling. Our code for this work is available at https://github.com/phamleyennhi/tracktok.

cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2503.13344

DeepSTEP: Simultaneous Tracking and Estimation of Pose for Animals and Humans

Shashikant Verma, Harish Katti, Soumyaratna Debnath, Yamuna Swamy, Shanmuganathan Raman

原文摘要

Maintaining a consistent identity across video frames while estimating pose is crucial for downstream tasks such as behavior analysis and action recognition, yet existing pipelines largely decouple tracking and pose estimation. Top-down methods rely on per-frame detections followed by identity association, leading to increased computational cost and error propagation, whereas bottom-up approaches often struggle to preserve identity reliably in multi-instance settings. To address these limitations, we introduce DeepSTEP (STEP), a Transformer-based framework for joint tracking and pose estimation across diverse animal species and humans. STEP unifies localization, identity maintenance, and keypoint estimation into a single model, eliminating the need for per-frame detections or post hoc matching. Our approach leverages discriminative model weight prediction to propagate target identity across frames while predicting anatomically meaningful keypoints. Discriminative model prediction typically requires predefined target states. In contrast, STEP assumes only an initial bounding box in the first frame, removing the need for explicit keypoint initialization. This is enabled by two components: Gaussian Map Soft Prediction (GMSP) and Offset Map Regression Adapter (OMRA), which allow direct keypoint prediction and simplify inference. Starting from the initial bounding box, STEP tracks the subject and estimates its pose in subsequent frames. We train and evaluate STEP on datasets spanning multiple species, demonstrating strong performance and greater efficiency than existing methods. By jointly modeling tracking and pose, STEP provides an effective and scalable solution.

cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.14171

CSI-JEPA: Towards Foundation Representations for Ubiquitous Wi-Fi Sensing with Minimal Supervision

Xuanhao Luo, Zhizhen Li, Yuchen Liu

原文摘要

Channel state information (CSI) provides a widely available sensing modality for human and environment perception, but existing CSI sensing models usually rely on task-specific supervised training and require substantial labeled data for each task, device, user, or environment. This limits their scalability in practical deployments where unlabeled CSI is abundant but labeled data is costly to collect. In this paper, we present CSI-JEPA, a self-supervised predictive representation learning framework for label-efficient and transferable Wi-Fi sensing. CSI-JEPA learns reusable temporal-spectral representations from unlabeled CSI samples by predicting latent features of masked channel regions from visible context. To better match the physical structure of CSI, it jointly tokenizes CSI along the time and subcarrier dimensions and introduces a channel variation-aware masking strategy that prioritizes regions with stronger local channel dynamics. After pretraining, the encoder is frozen and adapted to downstream sensing tasks using lightweight task-specific adapters. We evaluate CSI-JEPA on seven real-world Wi-Fi sensing tasks spanning diverse objectives and deployment settings. The results show that CSI-JEPA improves downstream sensing performance over competitive baselines, achieving up to 10.64 percentage points (pp) mean accuracy gain over the supervised Transformer and matched-budget label savings of up to 95.0%. Across unseen tasks, environments, hardware platforms, and CSI organizations, frozen CSI-JEPA improves downstream adaptation by up to 22.92 pp over non-pretrained baselines, demonstrating strong representation transferability.

cs.LG · cs.NI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07081

Wide View Synthesis from Overlapping Perspective Cameras in Autonomous Driving

Kaixuan Zhang, Zhipeng Xiong, Minxian Li, Mingwu Ren, Xiatian Zhu

原文摘要

Multi-perspective-camera systems are widely adopted since they provide broad visual coverage for autonomous driving, motivating the synthesis of a unified wide view from spatially separated observations. In this work, we formulate wide view synthesis (WVS), which aims to synthesize a geometrically consistent wide view at a specified virtual viewpoint from multiple perspective cameras. Unlike conventional image stitching, WVS requires reasoning about 3D geometry and visibility to handle depth-dependent parallax and occlusions arising from spatially separated camera centers. While vanilla 3D Gaussian Splatting (3DGS) enables high-fidelity view synthesis, its reliance on per-scene optimization limits its applicability to driving scenes. Furthermore, WVS introduces additional challenges, *e.g.*, peripheral stretching under perspective projection and photometric inconsistencies across cameras. To address these challenges, we propose **Wide-View Gaussian Splatting (WV-GS)**, a feed-forward Gaussian-splatting framework that synthesizes wide views without per-scene optimization. Specifically, we introduce Piecewise Perspective Rendering (PPR) to mitigate peripheral stretching through local perspective projections sharing a common camera center, and Spatial Affine Photometric Modeling (SAPM) to account for camera-dependent photometric variations during training. We further introduce **WideDrive**, a synthetic benchmark providing dense ground-truth wide views and adapt a LiDAR-projected sparse-reference protocol to real-world datasets. Extensive experiments on multiple datasets demonstrate the effectiveness of WV-GS for WVS. Dataset and code will be publicly released.

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07078

Optimistic Hedge in General-Sum Games: Fundamental Limits and Regret Bounds via Temporal Approximation

Yuval Dagan, Nadav Eisen, Liad Erez, Tomer Koren

原文摘要

Optimistic Hedge, a multiplicative weight update with one step of extrapolation, is one of the most natural regret-minimizing dynamics for games, and characterizing its regret has been a longstanding open question. Daskalakis, Fishelson and Golowich proved that if all $m$ players in a finite general-sum game follow Optimistic Hedge with a fixed learning rate, each player's regret is polylogarithmic in the horizon $T$. Subsequently, Ha showed that an appropriate choice of learning rate makes each player's regret bounded independently of $T$. This bound, however, is non-constructive and its dependence on the number of actions $n$ is not explicit, leaving open whether Optimistic Hedge can achieve a constant regret bound which is polynomial in the size of the game. We show that the answer hinges on the number of players. In two-player games, there is a learning rate for which Optimistic Hedge guarantees $O(n\log n)$ regret independent of the horizon. In three-player games, however, no such bound is possible: for any learning rate up to an absolute constant and for sufficiently large horizon $T$, there is a game with $n$ actions per player in which some player's regret is $2^{Ω(n)}$, and more generally, for every $m\ge3$ we construct $m$-player games in which some player's regret is $2^{Ω(mn)}$; for learning rates above $2^{-Ω(mn)}$, that player's regret even grows linearly in $T$. Our upper bound is based on a temporal approximation framework for the analysis of Optimistic Hedge. Starting from the variance-based RVU bound of Daskalakis et al., we approximate the losses over time-localized windows by well-behaved functions, whose total variation is a small fraction of their cumulative variance. In $m$-player games, the same framework gives a significantly simplified proof of the recent $O(\sqrt{m}\log n\log T)$ regret bound due to Ha.

cs.LG · cs.GT

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07073

When Verification Is Not Agreement: Cross-Agent LLM-to-Dafny Formalization

Priyanath Maji

原文摘要

A verified Dafny postcondition certifies that a formula holds, but it does not guarantee that the result was reached through genuine independent reasoning. We introduce a cross-agent formalization protocol where a Specifier model translates clinical trial criteria into a Dafny predicate, and an Implementer model writes an operational method checked against it. A same-model control establishes that even with no capability gap, independent re-extraction from identical text agrees only 59.3% of the time. Evaluating four model configurations on 96 clinical trial documents, we find that agreement is highly sensitive to how model capability is allocated across the two roles: the GPT-4.1-nano/GPT-4.1-mini configuration achieves 53.0% agreement when the stronger model is the Implementer, but only 14.3% when the roles are reversed. The strongest tested configuration, GPT-5.6-luna/GPT-5.4-mini, improves agreement to 34.4% but remains below the same-model baseline and is dominated by direct contradiction (42.7%). We also detect a critical verification artifact in which a weaker Implementer satisfies the postcondition by invoking the Specifier's predicate directly; after structural auditing, corrected agreement falls from 14.3% to 13.2%. Our results show that verification success alone is insufficient evidence of independent cross-agent agreement, and that model capability does not guarantee agreement when formalizing the same source independently.

cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2501.10172

Mean and Variance Estimation Complexity in Arbitrary Distributions via Wasserstein Minimization

Valentio Iverson, Stephen Vavasis

原文摘要

Parameter estimation is a fundamental challenge in machine learning, crucial for tasks such as neural network weight fitting and Bayesian inference. This paper focuses on the complexity of estimating translation $\boldsymbolμ \in \mathbb{R}^l$ and shrinkage $σ\in \mathbb{R}_{++}$ parameters for a distribution of the form $\frac{1}{σ^l} f_0 \left( \frac{\boldsymbol{x} - \boldsymbolμ}σ \right)$, where $f_0$ is a known density in $\mathbb{R}^l$ given $n$ samples. We highlight that while the problem is NP-hard for Maximum Likelihood Estimation (MLE), it is possible to obtain $\varepsilon$-approximations for arbitrary $\varepsilon > 0$ within $\text{poly} \left( \frac{1}{\varepsilon} \right)$ time using the Wasserstein distance.

cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07072

Automated Feedback to Peer Tutors Increases Their Talk Moves and Learner Reasoning but Not the Balance of Talk

Dorottya Demszky, Yann Hicke, Joy Yun, Mariah Olson, Max Karpawich, Anees Iqbal, Matt Wu, Jimmy Huang 等 10 位作者

原文摘要

Automated feedback on classroom discourse improves instruction in professional and online teaching contexts, and recent work suggests that delivering similar feedback to learners may strengthen these effects by sharing responsibility for participation. Whether this holds in small-group volunteer tutoring has not been tested. We present results from a randomized controlled trial across two four-week SAT mathematics bootcamps on the Schoolhouse.world platform, involving 1,030 volunteer tutors, 7,443 learners and 6,508 recorded sessions. Tutors were randomly assigned to receive no feedback, automated feedback on their own discourse, or that feedback together with participation feedback delivered to their learners under one of two motivational framings. Feedback to tutors increased eliciting of learner ideas, uptake of learner contributions, and learner reasoning by 8-15%, and the pattern replicated across both bootcamps. However, tutor talk time, which averaged 91.5% in the control group, did not change in any condition, and learner-facing feedback did not further change discourse. Interviews with 17 tutors indicate that they adopted the moves the feedback recommended but were limited by learners who did not respond. Interviews with 13 learners suggest why the learner-facing feedback had little effect: it reported a group-level rather than a learner-level statistic, some learners preferred to listen, and learner-to-learner interaction was rare in these sessions. Automated feedback can change how tutors invite learner thinking, but shifting how much learners talk likely requires changes beyond feedback.

cs.CL · cs.ET · cs.HC

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.34763

The Physical Cost of Context: Quality-Constrained Resource Accounting for Coding Agents

Zhongming Yu, Yitong Ma, Hengjia Yu, Boqin Yuan, Xiangye Lin, Hejia Zhang, Jishen Zhao

原文摘要

Coding agents resend a growing context on every model call, and context policies are usually judged by the prompt tokens they save. Executed work also depends on call count and prefix reuse: a caching runtime prefills only the tokens after the longest cached prefix. A shorter prompt can therefore cost more if it rewrites cached history. Our quality-gated protocol splits executed prefill into logical prompt minus reused prefix and measures KV residency, time, and energy. On 450 localization cases derived from SWE-rebench V2, preloading ten retrieved blocks (Eager) and adding anchored history compaction (Compact) both pass a five-point non-inferiority gate. In a separate 900-cell study, Eager cuts model calls by 35.7% and the GPU-energy proxy by 33.2% against on-demand search (Grep); Compact saves nothing further. Enabling prefix caching leaves Eager's prompt volume similar (+8.3%; its 95% CI spans zero) but cuts its uncached prefill by 67.7%. Deleting history can turn a token saving into extra work. With mini-SWE-agent on 100 SWE-bench Verified instances, keeping only the last five observations (Mask-5) cuts Eager's prompt volume by 33.3% but loses 135.9k reusable tokens per task. Uncached prefill rises by 608.5%, the energy proxy by 125.6%, and the policy misses the quality gate. Policies thus save physical work by shortening the trajectory and can multiply it by editing reusable history.

cs.SE · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07071

Shaping & Monitoring Models through their Internal Representations

Thomas Fel, Matthew Kowal, Matt Feiszli, Satchel Grant, Andrew Lee, Chris Earls, Fatih Dinc, Vasudev Shyam 等 26 位作者

原文摘要

We propose Mechanistic Shaping, a unified framework for shaping models through their internal representations. Mechanistic Shaping is applied during training so that models follow specifications (rules) about which concepts they express, measured via their outputs or internal representations (e.g., no harmful content), and also which concepts they rely on, i.e., what they use to make their decisions (e.g., not a candidate's gender when screening a résumé). Across multiple domains (vision, medical imaging, language), scales (up to 32B parameters), and training paradigms (classification, SFT, DPO, GRPO), we find that Mechanistic Shaping substantially reduces targeted behaviors while preserving task performance. Since training against a representation can also modify it, we ask whether shaped concepts remain monitorable, a question that depends on the class of monitors considered. We therefore introduce a hierarchy of such classes, which separates a concept that escapes one probe from a concept that is truly gone. We characterize three geometric mechanisms of circumvention: the concept can drift away from the monitor, be erased from the layer, or persist in higher-order statistics as a nonlinear encoding, which we explain through a product-code construction. Although other mechanisms may arise in different settings or at larger scales, these three account for all of our observations, and each can be detected by a matching monitor. In practice, we find that shaped concepts largely remain monitorable: refitted linear probes recover the concept in most settings, and the original probe remains reliable whenever gradients do not flow through it. Taken together, these results point to the potential of using interpretability as a tool for training, and call for a correspondingly rigorous science of interpretability-based training.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2510.02906

FinMultiHopQA: A KG-Grounded Multi-Hop Financial QA Benchmark for Evidence-Driven Reasoning

Abhinav Arun, Reetu Raj Harsh, Bhaskarjit Sarmah, Stefano Pasquali

原文摘要

Multi-hop financial question answering often requires composing evidence distributed across filing sections, fiscal years, and peer companies. Prior financial QA benchmarks address parts of this problem individually, but none jointly support controlled multi-hop taxonomies, document topologies, motif-level reproducibility, and provenance-linked auditability at scale. We introduce FinMultiHopQA, a benchmark of 15,600 qualitative and quantitative questions, grounded in FinReflectKG, a temporally indexed financial knowledge graph derived from SEC 10-K filings with 17.5M+ triples. The pipeline instantiates statistically salient 2-3 hop typed KG motifs across intra-document, inter-year, and cross-company regimes. Beyond closing this gap, the same construction lets us hold each question's underlying evidence constant while varying only its presentation - embedding it in noisy surrounding context or changing its order - giving a controlled instrument to study a secondary question at scale: using six such presentations, we test whether reasoning LLMs' performance depends on evidence structure. Results confirm this: despite their reasoning capacity, judged correctness remains consistently lower under less-structured presentations of the same evidence - most sharply for quantitative cross-company reasoning and 3-hop branch motifs - indicating that reasoning alone does not fully offset how evidence is structured; KG-linked presentation also uses fewer input tokens by design. We release the benchmark, provenance annotations, model outputs, evaluation records, and supporting technical documentation.

q-fin.CP · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v6arXiv:2603.13535

From Proxy to Certificate: Two-Sided Ollivier-Ricci Bounds from Balanced Forman Curvature

Giorgio Micaletto, Tebe Nigrelli

原文摘要

Balanced Forman (BF) curvature is used in graph learning as a cheap stand-in for Ollivier-Ricci (OR) curvature, yet it does not lower-bound OR curvature. We show that the BF value of an edge certifies a two-sided interval for its lazy OR curvature without solving any optimal transport problem, that BF coincides with a quadrangle-augmented Jost-Liu (JL) bound with its positive parts removed, and that when the 2-hop neighborhood is available, the interval can be sharpened at a cost of $\mathcal O(d_{\max}^{2.5})$ per edge. On 23 benchmark graphs, our quadrangle-augmented bound is tighter than the bounds of JL and Tian-Lubberts-Weber (TLW) on 13.8-100% of the edges and reduces their mean gap to the exact curvature by a median of 31.2%; intersected with the bounds of TLW, the intervals settle the curvature sign of more than half of the edges on 21 of the 23 graphs. On 15 real networks with up to 1,049,866 edges, a certify-then-solve procedure answers sign queries exactly while solving a median of 5.9 times fewer transport problems, for median speedups of 2$\times$ over exact evaluation, measured against a min-cut solver that is itself about two orders of magnitude faster than the widely used $\mathtt{GraphRicciCurvature}$. The main results are machine-checked in Lean 4.

stat.CO · cs.DM · cs.LG · math.CO · math.DG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07069

Finite-Time Convergence and Regret Guarantees for Neural Reinforcement Learning in POMDPs

Jingjuan Huang, Zhiyao Zhang, Alvaro Velasquez, Nathaniel D. Bastian, Jia Liu

原文摘要

Reinforcement learning (RL) in Partially Observable Markov Decision Processes (POMDPs) is computationally challenging due to the curse of long-range dependence on history. Fortunately, recent studies have shown that a range of POMDPs that satisfy the filter stability assumption can be well approximated by a superstate Markov Decision Process (MDP) based on a truncated history in a finite-memory window, with a value gap that decays with the window length. However, existing finite-time convergence analyses on such superstate MDP surrogates are either limited to over-simplified linear function approximation or rely on an i.i.d. sampling oracle that is violated by POMDP trajectories, both of which severely limit the applicability of these theoretical results in practice. This motivates us to bridge the gap by investigating RL in POMDPs with nonlinear neural value function approximation. Our main contributions are threefold: i) We first establish a new finite-state MDP point-wise approximation error bound, which lays the foundation for our RL algorithm design and analysis in POMDPs; ii) By applying Temporal-Difference (TD) learning to POMDP-generated superstate transitions and targeting the Bellman fixed point of the superstate MDP, we prove a finite-time critic error bound with nonlinear neural value function approximation; and iii) By combining the neural critic function approximation error bound with the policy mirror descent framework, we obtain a finite-time sublinear regret against the optimal POMDP value. To the best of our knowledge, such a finite-time sublinear regret based on neural function approximation is the first in the literature for finite-memory POMDPs. Our numerical experiments also verify our theoretical results.

cs.LG · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07066

Cooperative Multi-Agent Reinforcement Learning via Epigraph-Form Guided Exploration

Sungil Son, Hoseong Jung, Dahyun Oh, H. Jin Kim

原文摘要

Discovering cooperative behavior in multi-agent reinforcement learning (MARL) is challenging due to the combinatorial complexity of joint state-action spaces, which hinders the emergence of coordinated behaviors from trial-and-error alone. Intrinsic rewards are often used to aid discovery, but naively combining them with team objectives can distort the learning signal, compromising task performance. In this paper, we propose EFXPLORER, a constrained exploration framework that maximizes exploration objectives subject to a constraint that preserves established task performance. We solve the resulting problem using an epigraph reformulation that introduces adaptive exploration budgets. This approach separates intrinsic rewards from task objectives and regulates exploration through task feasibility. To further encourage diverse and temporally extended exploration, we incorporate a successor distance-based intrinsic reward that captures long-horizon dependencies. Empirically, our method outperforms strong baselines and induces novel cooperative strategies across SMAX, VMAS, and MPE benchmark suites.

cs.AI · cs.MA

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2603.01367

DUEL: Exact Likelihood Reshapes Masked Diffusion Scaling Laws and Beyond

Gilad Turok, Yair Schiff, Chris De Sa, Volodymyr Kuleshov

原文摘要

Masked diffusion models (MDMs) are trained to predict tokens in every order but generate in one. As such, MDM likelihood is an intractable sum over all orders. The evidence lower bound (ELBO) is used as a proxy for likelihood, but it is a loose bound and assumes positions are unmasked uniformly at random. Frontier MDMs, however, select positions with deterministic rules such as confidence. To resolve this mismatch we introduce DUEL, a framework that unifies common deterministic MDM samplers. DUEL computes the likelihood exactly, under the sampler that is actually used during generation. DUEL is a simple plug-and-play algorithm applicable to any pretrained MDM with no retraining. To compute the likelihood of a text sequence, simply replay generation, reveal the true tokens instead of sampled ones, and sum their log-probabilities. Evaluating with DUEL instead of the ELBO changes what we conclude about MDMs. The perplexity gap between MDMs and autoregressive (AR) models shrinks by up to 32% in-domain and 82% zero-shot. In scaling laws, the reported 16x MDM-AR compute gap falls to 9x under DUEL, showing over 1/3 of it is an evaluation artifact. Scoring multiple-choice answers with likelihood via DUEL instead of the ELBO improves LLaDA-8B accuracy on all six tasks, by up to 9.7 points. DUEL also selects samplers correctly: it matches a gold-standard LLM judge on every decided comparison, which generative perplexity and MAUVE fail to do. DUEL reveals that MDMs are more competitive with AR models than the ELBO suggests.

cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v3arXiv:2602.17659

When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs

Yu Fang, Yuchun Feng, Dong Jing, Jiaqi Liu, Yue Yang, Zhenyu Wei, Daniel Szafir, Mingyu Ding

原文摘要

Vision-Language-Action models (VLAs) promise to ground language instructions in robot control, yet in practice often fail to follow language faithfully. When presented with instructions that lack strong scene-specific supervision, VLAs suffer from counterfactual failures: they act based on vision shortcuts induced by dataset biases, repeatedly executing well-learned behaviors and selecting objects frequently seen during training regardless of language intent. To support systematic study, we introduce LIBERO-CF, the first counterfactual benchmark for VLAs that evaluates language following capability by assigning alternative, feasible task instructions under the same LIBERO scene layouts. Our evaluation reveals that counterfactual failures are prevalent yet underexplored across state-of-the-art VLAs. To mitigate these failures, we propose Counterfactual Action Guidance (CAG), a simple yet effective dual-branch inference scheme that explicitly strengthens language conditioning in VLAs. CAG combines a standard VLA policy with a parallel language-unconditioned branch, amplifying the difference between their action predictions induced by language instructions. This design reduces reliance on vision shortcuts, improves robustness on counterfactual tasks, and requires neither additional demonstrations nor modifications to the base VLA architecture or its pretrained weights. Extensive experiments demonstrate consistent improvements across diverse VLAs. On LIBERO-CF, our training-free variant, TF-CAG, improves $π_{0.5}$ by 9.7% in language grounding and 3.6% in task success. When using a separate Vision-Action (VA) module, VA-CAG achieves larger gains of 15.5% and 8.5%, respectively. In real-world evaluations, VA-CAG improves language grounding by 9.4% and task success by 17.2% on average.

cs.CV · cs.RO

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07065

Stochastic Dynamics of Generative AI: From Autoregressive Models to Interacting LLM Networks and Neural SDEs

Vikram Krishnamurthy, Yiming Zhang, Luke Snow

原文摘要

This paper develops a stochastic dynamical-systems perspective on three classes of generative AI models. First, we formulate large language models (LLMs) as high-dimensional nonlinear autoregressive models with attention-based dependencies. This formulation connects pretraining by conditional maximum likelihood to sequential token generation and provides a probabilistic interpretation of LLM operation. Second, we consider networks of interacting LLMs and characterize their collective behavior through mean-field ordinary differential equations. These dynamics provide a framework for understanding emergent preferential attachment and persistent structural biases, including type-dependent glass-ceiling effects. Third, we consider neural stochastic differential equations (SDEs) as continuous-time generative models for stochastic trajectories, including those observed at irregular time intervals. We discuss their training with particular emphasis on Wiener-space cubature methods based on rough-path analysis as an efficient alternative to Monte Carlo estimation. Together, the three settings address probabilistic prediction, information exchange and emergent behavior in interacting networks, and path-space generative modeling.

cs.LG · eess.SP · eess.SY

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07063

SD-LoRA: Training-Time Structural Distillation into LoRA for Few-Shot Vision-Language Adaptation

Mohammed Rahman Sherif Khan Mohammad, Ardhendu Behera, Sandip Pradhan, Swagat Kumar, Yonghuai Liu

原文摘要

Few-shot adaptation of vision-language models must balance two competing goals: learning fine-grained visual distinctions from very limited supervision and preserving the efficiency that makes frozen CLIP-like models attractive for deployment. Existing lightweight adaptation methods, including prompt tuning, LoRA, and cache-based adapters, typically supervise global image embeddings and therefore underuse token-level structure. Conversely, structure-aware methods exploit local evidence but often introduce additional inference-time computation. We propose SD-LoRA, a structural distillation framework that uses token-level graph reasoning only during training and transfers its effects into a lightweight LoRA-adapted CLIP model. During training, a heterogeneous graph teacher models patch-patch, global-local, and visual-text relations over high-resolution ViT tokens. Rather than retaining this graph at test time, SD-LoRA transfers its structural bias into the shared CLIP-LoRA backbone through implicit distillation and Prototype Predictive Alignment, which aligns graph-conditioned representations with support-derived class prototypes. At inference, the graph teacher is discarded; prediction requires only a single CLIP-LoRA forward pass and a cache-based classifier. Across 11 few-shot benchmarks, SD-LoRA improves over a strong CLIP-LoRA baseline by 2.4/1.8/3.0/2.4/2.0 percentage points in the 1/2/4/8/16-shot settings, respectively, without additional graph computation at test time. Results on base-to-novel generalization, cross-dataset transfer, out-of-distribution robustness, and calibration further show that training-time structural supervision improves both accuracy and transfer reliability. Code is available at https://github.com/MR-Sherif/SD-LoRA.git.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07061

Segment What It Can Draw: Querying a Text-to-Image Generator with Its Own Images for Skin Lesion Segmentation

Hongyu Cao, Guangfu Guo, Jinghan Zhang, Dakshak Keerthi Chandra, Ying Zhang, Kunpeng Liu

原文摘要

Skin lesion segmentation without annotated masks requires the target to be specified through other cues. Existing approaches rely on dermoscopy-specific rules, annotation-derived model priors, or user-selected regions. A general text-to-image diffusion transformer offers a text interface to the target, but its concept-word attention is unreliable on dermoscopy images: in the model's attention-output space, the word-derived skin-to-lesion direction is poorly aligned with the direction separating real skin from real lesions. We show that the direction derived from generated healthy-skin and lesion images is substantially better aligned. Based on this finding, we propose GENIQ, a training-free method that constructs initial skin and lesion prototypes from generated images and refines them using the most skin-like and lesion-like patches from unlabelled dermoscopy images. It scores test patches along the refined direction and applies a gated threshold to suppress contextual spill-over around small lesions. Without inference-time annotations or additional models, GENIQ achieves Dice scores of 0.85-0.88 on three dermoscopy test sets. It outperforms zero-shot and word-based attention methods on all three sets, and a hand-designed dermoscopy pipeline in Dice and IoU on two.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.34499

Can Statistics Tell Whether an Undeciphered Text Is Language? Calibrated Tests on the Voynich Manuscript

Xin Zhang, Victor S. Sheng

原文摘要

Statistics are often the only evidence of whether a text in an unread script records a language, but meaningless text shares many of them. A test means little until it is run on texts whose answer is known. We build such tests and apply them to the Voynich manuscript, whose text may be meaningless copying or a language hidden by a cipher. We write both accounts as programs, fit them with the same search budget and check them on held-out statistics. The copying process of Timm and Schinner predicts these statistics better than any cipher with one key, but ciphers with two keys, or that encode only the word classes of a real text, do equally well. A trace of language that needs no decipherment says more. In 28 texts in 16 natural languages the same word pairs keep sharing a line, and the manuscript shows this at half the strength of the weakest. Word codes fall to this level only with heavy miscopying or a plaintext whose word pairs recur weakly, such as abbreviated Latin. Solvers that read enciphered controls find nothing readable in the manuscript under letter ciphers, word codes or codes with one word per letter or syllable. One cipher remains, a nomenclator with two keys, which writes frequent words whole and spells the rest, and whose scribe miscopies more than half of the words. Our tests exclude it in all languages but Arabic and Hebrew. Planted messages show that a short message could still hide.

cs.CL · cs.CR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07055

Expanding the Quality-Coverage Frontier with Competitive Flow Matching

Wenxuan Peng, Richard Zhang, Haitian Zheng, Jaskirat Singh, Tsai-Shien Chen, Eli Shechtman, Bharath Hariharan, Hadar Averbuch-Elor 等 9 位作者

原文摘要

Flow matching (FM) generators often face a quality-coverage tradeoff. Inference-time techniques such as classifier-free guidance raise quality, but at the cost of coverage. We hypothesize this limitation stems from a basic property of the objective: an intermediate noisy image could correspond to many clean images, yet the model can only predict a single velocity, at best averaging over the multiple possibilities. To mitigate this, we introduce competition among predictions during training. We achieve this by augmenting the predictor with a randomly drawn low-dimensional conditioning code $z$. For each training example, we sample $K$ values of $z$ to produce $K$ candidate velocities that compete to explain the target, and apply the loss only to the best-matching candidate. This winner-take-all objective allows different modes to shape different predictions, rather than being averaged into a single compromise. Furthermore, this term naturally enables a new form of guidance, auxiliary conditioning scaling, which complements traditional guidance methods. Together, these two guidance mechanisms show that our method substantially improves the quality-coverage frontier. We validate on ImageNet-256, using two different testbeds, SiT and RAEv2. Given less than 3% of the original SiT pretraining compute for post-training, at the 67% Recall point, our method reduces $\mathrm{FD}_{r,6}$ from 19.22 to 5.91. On RAEv2, at the 70% Recall point, it reduces $\mathrm{FD}_{r,6}$ from 16.83 to 3.18.

cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.00700

Value Over Language Model: Detecting Original Contribution in Writing

Vibhhu Sharma, Thorsten Joachims, Sarah Dean

原文摘要

LLMs have been rapidly adopted across writing tasks, prompting the development of tools for detecting LLM-generated text. Yet, these tools largely measure how much of a document's surface text was written by an LLM and aren't fundamentally designed to measure how much of the information content or ideas originated from the LLM itself rather than being supplied by the user in the prompt. In this work, we design a framework that measures how much value a person adds on top of what a language model could have easily produced by itself. The method requires no training or labeled data and never scores the document's surface text, insulating it from stylistic confounders. Instead, it extracts the document's content at increasing levels of granularity, uses an LLM to reconstruct the document from each partial representation, and compares these reconstructions with those produced from the task description alone. We call this framework Value Over Language Model (VOLM), which measures a document's contribution relative to a replacement-level document that an LLM could produce from the task description alone. We evaluate VOLM with a specific instantiation of this framework across three domains: news articles, ICLR peer reviews, and argumentative essays. VOLM separates human-authored documents from matched LLM-generated documents produced from generic task descriptions, while remaining substantially invariant to content-preserving transformations, including LLM-based reconstruction and round-trip translation. We further find that increasingly constrained content extractors reduce residual differences between LLM-generated and humanized text, demonstrating the importance of disentangling informational content from stylistic variation. We hope these results encourage further work on specialized instantiations of the framework and on assessing human contributions in LLM-assisted writing more generally.

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07054

Reclassification Is All You Missed: Surfacing Variant Reinterpretation Need in Online Health Communities at Population Scale

Kamalakkannan Ravi, Mengshu Nie, Annapurna Poduri, Cong Liu

原文摘要

Genetic variants are routinely reclassified in ClinVar after a patient is tested, sometimes from uncertain to pathogenic, yet patient recontact is inconsistent, with no clear duty and responsibility diffused among labs, providers, and patients. Many affected people discuss their results in public online health communities that clinical recontact often does not reach. We built PRISM, an agentic large-language-model pipeline that screens these posts, checks whether the variant was later reclassified, and drafts a disclaimer-bearing outreach message, in a no-contact dry run. Across 3.9 million posts from 32 communities, agent-predicted reinterpretation-relevant discussion is rare (0.288%) but concentrated in specialist genetics communities (6.30% versus 0.10% in general medical-advice communities) and growing fivefold since 2014; PRISM surfaces 123 candidate classification differences, 5 of them from an uncertain submission to pathogenic or likely pathogenic. On a held-out set of 30 real-post drafts, a clinician judged 27 factually grounded and 15 fit to send: factual accuracy is necessary but not sufficient for clinically acceptable outreach. PRISM contributes a population-scale reinterpretation-need map with code and data to be released upon publication.

cs.CL · cs.AI · cs.CY · q-bio.GN

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.27041

Robustness of Internal Signals for Hallucination Detection in Vision-Language Models

Allison Chen, William Yang, Salma Abdel Magid, Jonathan Williams, Olga Russakovsky, Esin Tureci

原文摘要

When given an image and a text prompt, vision-language models (VLMs) can produce seemingly sophisticated text responses. However, they are also prone to hallucinate objects in their output, which can have detrimental consequences in safety-critical applications. To detect these hallucinations, certain methods interestingly leverage internal signals from the VLM as it generates the text response; potentially, these methods use the same signals that led to a hallucination to detect it. We conduct an in-depth analysis to understand the robustness of these detection methods. First, we show that a naive baseline of a generated word's token position in the VLM text response achieves competitive performance. Further, we reveal how some detection methods rely on token position and use this heuristic as a shortcut. Second, we explore detection methods' sensitivity to language and vision modalities. Our experiments suggest that strong detection performance is often driven by detecting language-based hallucinations, and when methods rely on token position, they tend to be less visually grounded and exhibit reduced robustness when hallucinations occur earlier in a text response. Lastly, we demonstrate that different detection methods can carry complementary signals; when ensembling methods together, adding weaker methods to an ensemble of higher performing ones still offers performance improvements. Altogether, while there may be useful internal VLM signals to detect hallucinations, our investigation reveals important limitations of the current methods that try to leverage them.

cs.CV · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07051

Closed-Loop Agent Evaluation: Layered Measurement and Self-Calibration

Ruifeng Ren, Gun Ahn

原文摘要

Once an agent is live, teams want to know whether it is doing its job, not just whether it throws errors. Yet most evaluation is one-off: a benchmark run, a look at traces, a single score. Its conclusions neither check themselves nor reach the agent. We propose closed-loop evaluation, an evaluation method for deployed agents. It models evaluation as performance estimation under biased raters and censored outcomes, and it requires every level's conclusion to trace back to atomic data and to be computed by versioned rules. The method covers four levels of analysis (span, trace, conversation and engagement), measures along a spine of Execution, Substance, Trajectory and Results, treats Conduct as a constraint throughout, and uses Provenance and Adjudication components to certify the evidence and the evaluators. Evaluator agents generated by an Agent Factory carry these components, each with a domain knowledge pack and an adapter harness. Two loops close the evaluation: an inner loop audits the evidence, rules and metrics themselves, and an outer loop uses per-component regression checks to drive agent improvement. We also give a pipeline that runs this process automatically. Using the business-analysis agent SharpInsight as a case and porting the method to public conversation data, we show how problems differ across levels and how evidence-and-rule judgment is more reliable than holistic scoring (ICC 0.86 vs. 0.81; kappa 0.81 vs. 0.63).

cs.AI · cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07050

Authorization Propagation in Agent2Agent Systems: An Experimental Security Evaluation of Multi-Hop Delegation

Gabriele Donno, Kashif Mehmood, Katina Kralevska

原文摘要

Delegated Agent2Agent (A2A) workflows create a security problem when the agent sending a request is different from the principal from which the authority originated. This paper evaluates a layered downstream-verification design in a synthetic four-party healthcare reimbursement workflow. The prototype separates current-caller context, transitive authorization evidence, per-hop Ed25519 request authentication, and independent replay state for requests and authorization artifacts. Of 15 adversarial conditions, 14 are rejected; an isolated delegation-path test exposes the absence of semantic authority attenuation. A separate authorized-intermediary experiment shows that a holder of the shared authorization-evidence HMAC key can modify the represented authority and generate valid evidence. In the fixed-depth local benchmark, the full configuration has a median security-processing latency of 5.11 ms and a 3.87 kB serialized payload. The results show the value of separating request authentication from inherited authority while identifying origin-verifiable delegation and semantic attenuation as requirements for stronger A2A authorization propagation.

cs.MA

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07049

Gaussians Can Subtract: Neural Dissolve Fields for Hybrid Scene Rendering

Yiyan Wang, Haohan Liu, Tetsuo Shibuya

原文摘要

Hybrid Mesh-Gaussian rendering combines compact textured surfaces with flexible 3D Gaussian primitives, offering an efficient alternative to purely Gaussian scene representations for structured environments. Existing scene-reconstruction hybrids commonly follow an additive formulation, where Gaussian residuals compensate for errors left by the mesh. However, not every Mesh-Gaussian conflict is caused by missing content: residual geometry errors, occlusion mismatch, texture-projection artifacts, and view-dependent appearance can make the same mesh content consistent in one view but inconsistent in another. We therefore formulate hybrid rendering as a multi-view inconsistency problem and present Gaussolve, a subtract-and-refill framework that selectively dissolves inconsistent mesh content before residual Gaussians refill the revealed regions. We introduce a lightweight Neural Dissolve Field, parameterized by scalar spherical-harmonic Gaussians, to learn where and from which viewing directions mesh content should be dissolved. A continuous transmittance fusion rule progressively exposes residual Gaussians behind the dissolved mesh, preserving their reconstruction gradients and avoiding hard visibility discontinuities. Experiments on structured and semi-structured scenes demonstrate higher visual fidelity with fewer Gaussian primitives and faster rendering. Project page: https://ooosn.github.io/gaussians-can-subtract/

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07048

Adapting the Process of Adaptation: Darwinian, Lamarckian, and Agent-Directed Evolution in AI

Andrew W. Lo, Chaoyi Zhao

原文摘要

Evolution has shaped intelligence for billions of years. Artificial intelligence (AI) now raises a striking possibility: intelligence may begin to direct its own evolution. As AI agents become capable of modifying and evaluating their own behavior, how quickly could they improve, and how can that improvement remain safe? We develop a theoretical framework comparing Darwinian, Lamarckian, and agent-directed evolution following environmental change. Darwinian evolution relies on natural selection, Lamarckian evolution adds inherited learning, and agent-directed evolution allows agents to generate and evaluate behavioral variants and choose how extensively to search. When learning is sufficiently effective, these mechanisms exhibit progressively faster adaptation. Generating more variants accelerates adaptation, while optimal search effort depends on learning ability, problem difficulty, and experimentation costs. However, an agent's ability to direct its improvement does not ensure that improvements are evaluated against the intended objective. When the evaluation target is misspecified, additional search accelerates convergence toward the wrong behavior, even as measured performance improves. In recursive self-improvement, such errors may persist across generations, posing challenges for AI safety. Independent evaluation and oversight must keep pace with autonomous AI development. The central challenge is not simply to accelerate the evolution of intelligence, but to preserve humanity's ability to guide its direction, limit its speed, and intervene when necessary.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.07042

Follow the Policy, Fail the Benchmark: Auditing and Revising Policy-Based Decision Benchmarks

Yanru Xiao, Yifei Cai

原文摘要

Policy-based decision benchmarks evaluate whether systems apply written rules to financial and operational cases. Yet a rule-following response can be marked wrong when the benchmark misstates the required decision. We audit nine public policy-based decision benchmarks through source tracing, program execution, and controlled checks. We find unsupported targets, reference errors that survive replay or propagate as solved steps, and evaluators that reject acceptable answers or credit incomplete decisions. To measure the scoring impact, we rescore unchanged model responses. On the 96 LegalBench SARA numeric cases, correcting the scorer raises acceptance by more than 80 percentage points for each of three models, without changing references or responses. In RuleArena Tax, restoring a required zero floor changes 27 of 300 references and narrows an observed model gap. A controlled federal-exemption task with explicit ownership and eligibility assumptions also shows that errors in supplied intermediate answers can affect final decisions. We present BAR (Benchmark Audit and Revision), a versioned collection of audit evidence and revisions for reevaluation.

cs.CL · cs.SE

查看论文
arXiv ↗PDF ↗