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

+352新增论文
+272其中 AI 相关
+39已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-17 04:58 · 全站最近入库:2026-11-17 04:58各信源进度 →

统计口径与数据新鲜度

统计 2026-11-16 11:47 — 2026-11-17 11:47(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:2026-11-17 04:58最近发现新增版本:2026-11-17 04:58库内论文最新更新时间:2026-11-14 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。

24,106 已收录论文20,158 AI 相关7,049 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 24,106 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-17 04:57(北京时间)
新论文 · v1arXiv:2611.05986

A Comparative Study of Importance Attribution to Columns in Deep Learning over Relational Databases

Hila Levi-Yosefi, Shunit Agmon, Ester Livshits, Agapi Rissaki, Benny Kimelfeld

原文摘要

Deep learning over relational databases has led to increasingly effective prediction models, motivating a growing need to explain their behavior. We study explanation methods that assign importance scores to individual database attributes, focusing on a fundamental question: to what extent does the choice of importance-attribution method matter? We develop a methodology to systematically compare importance-attribution methods and analyze their similarities and differences. In particular, we study two aspects of the explanations: which attributes they deem important, captured by differences in attribute rankings, and how importance scores vary across attributes, captured by differences in score distributions. We further investigate whether these differences are reflected in the conventional measure of explanation quality, fidelity. We apply our methodology in an extensive empirical study of four methods: Monte Carlo Shapley, Kernel SHAP, single-attribute impact, and mask learning. We experiment with multiple datasets and tasks using two RDL architectures: heterogeneous GraphSAGE and Relational Graph Transformers. Our experiments show that explanation methods behave differently depending on the prediction type: regression or classification. In regression, most methods exhibit high agreement in attribute rankings, whereas classification yields considerably less consensus. Notably, while mask learning frequently diverges in its importance distributions, Shapley-based methods and single-attribute impact often converge to similar explanations. This suggests that, in many settings, single-attribute impact can replace Shapley at substantially lower computational cost. Overall, our study provides methodological tools for comparing importance-attribution methods, and empirical evidence that their choice has important consequences for interpreting relational deep learning models.

cs.LG · cs.DB

版本更新 · v2arXiv:2607.03068

PixCon: Clean-Positive Contrastive Learning for Foundation-Model Semi-Supervised Segmentation

Ebenezer Tarubinga

原文摘要

Can a simple labeled-pixel contrastive branch improve semi-supervised segmentation with a strong pretrained encoder? We study this question with PixCon, which adds a class-wise memory bank to a DINOv2-Base implementation of UniMatch V2. The bank stores ground-truth-labeled features only where the student prediction agrees with the label. A multi-positive contrastive loss uses these features to train the shared encoder-decoder; the head and bank are discarded at inference. This simplifies earlier labeled-pixel banks by removing confidence thresholds and learned ranking. On Pascal VOC with 183 labels, PixCon improves all three paired seeds by 0.20, 0.64 and 1.83 mIoU points, raising the mean from 87.01 to 87.90. With 92 labels the mean gain is 0.40 points; single-seed differences on Cityscapes and ADE20K range from -0.08 to +0.14 points. A predicted-label contrastive control reaches nearly the same Pascal accuracy, so these results support the added branch rather than an accuracy benefit from its correctness filter. An exact gradient decomposition distinguishes mislabeled-positive counts from their similarity-weighted influence. Separate teacher diagnostics find retained-pixel error rates of 1.8% on Pascal and 10.6% on ADE20K. These diagnostics do not measure the training bank or predict final accuracy. PixCon adds 15.3-19.7% iteration time in the measured runs and no inference computation. We report paired seeds, training cost and the limits of the evidence.

cs.CV · cs.AI · cs.LG · eess.IV

新论文 · v1arXiv:2611.05985

Quantum Key Exchange from Public Quantum Money

Eli Goldin, Tomoyuki Morimae

原文摘要

Two-message quantum key exchange (2QKE) cannot be realized in a statistically secure way, and therefore some computational assumptions are required. So far, the known weakest assumption for 2QKE is the existence of one-way functions (OWFs). Can we construct 2QKE from a weaker assumption than OWFs? In this paper, we give the first complete characterization of 2QKE. We show that the following four primitives are (existentially) equivalent: everlasting-secure 2QKE, standard 2QKE, efficiently-verifiable state puzzles (EV-StatePuzzs), and one-time secure digital signatures with quantum signatures and signing keys. Because EV-StatePuzzs are potentially weaker than OWFs [Kretschmer, Qian and Tal 2025], the equivalence implies that 2QKE are potentially weaker than OWFs. Moreover, our result shows the equivalence between standard 2QKE and everlasting-secure 2QKE, which means that everlasting security comes for free. Finally, one-shot signatures, quantum lightning, and public-key quantum money mini-schemes imply EV-StatePuzzs, and therefore, our result also gives a new lowerbound, namely everlasting-secure 2QKE, of these public-key unclonable primitives. We also show that one-shot signatures imply QPKE with quantum public keys. [Kitagawa, Morimae,Nishimaki and Yamakawa, 2024] constructed QPKE with quantum public keys from OWFs, and it was open whether QPKE with quantum public keys can be constructed from an assumption weaker than or incomparable to OWFs. Our result gives a potential answer to this question since one-shot signatures seem to be incomparable with OWFs. Moreover, QPKE constructed from one-shot signatures has an additional functionality, which we call certified freeze, that allows the receiver to publish a publicly verifiable classical certificate that computationally binds the receiver to an upper bound on the total number of distinct valid ciphertexts under its verification key.

quant-ph · cs.CR

版本更新 · v2arXiv:2607.28685

Safety, or Just Capability? A Validity Audit of Agent-Safety Benchmarks

Youting Wang, Xiao Han, Dingyan Shang, Yuan Tang, Bowen Liu

原文摘要

Agent-safety benchmarks measure different behaviors, and their scores get quoted interchangeably as an agent's safety. We treat four of them (R-Judge, InjecAgent, AgentHarm, AgentDojo) as measurements to be validated, running each under its official implementation and author-provided scorer on up to 22 models, with MMLU and GPQA measured by us under one protocol as a capability composite. The metric is the first problem. On any binary trace-judgment benchmark scored by $F_1$, an ``always positive'' policy attains $F_1 = 2π/(1+π)$; on R-Judge that is $0.690$, above five of the 21 models that actually discriminate. The three broad-coverage benchmarks then rank the same 18 models differently, and the trade-off behind that disagreement is a small-panel artifact: R-Judge specificity against AgentHarm safety correlates $-0.64$ at $n{=}7$ and $+0.02$ at $n{=}18$, and a quarter of random size-7 subsets reach $|ρ| \geq 0.5$ around that near-zero value. Held-out validity turns on which outcome you pick. Capability predicts task success ($ρ{=}{+}0.60$) but correlates negatively with misalignment safety ($ρ{=}{-}0.44$, $n{=}21$). On their paired $n{=}20$ panel, the corresponding contrast is $Δ{=}{-}1.00$ (95% CI $[-1.48, -0.49]$, $p<0.001$), and it survives leave-one-organization-out and organization-clustered bootstrap analyses. On an expanded 41-model panel, the misalignment correlation weakens to $-0.16$ (95% CI $[-0.54, +0.22]$) and jailbreak strengthens to $+0.34$, though neither change is significant. \mbox{AgentHarm} shows the strongest held-out association, $ρ{=}{+}0.72$ with three-template jailbreak safety after controlling capability. But both instruments score harmful compliance, so this is evidence of convergent validity rather than general safety. Naming the benchmark, metric, target behavior, and model panel is the minimum a safety claim needs.

cs.AI · cs.IR

版本更新 · v2arXiv:2611.04382

Beyond Compression Ratio: High-Throughput KV Compression for Disaggregated LLM Serving

Junhong Wu, Zedong Liu, Xinyang Ma, Yuhan Chen, Guangming Tan, Dingwen Tao

原文摘要

KV cache compression reduces the communication volume of disaggregated large language model serving, but its practical benefit depends not only on compression ratio and model quality, but also on how efficiently the compression pipeline executes. This requirement becomes increasingly important as modern serving systems adopt high-bandwidth interconnects, where inefficient compression can offset a substantial portion of the saved communication time. This paper presents FastKVComp, a high-throughput KV cache compression data path built on top of KVServe. FastKVComp improves compression execution in two complementary ways. First, it fuses the distribution transformation and low-bit quantization stages into a KVaware TileLang kernel, eliminating intermediate tensor materialization and redundant memory accesses. Second, it replaces the original lossless compression backend with an open, extensible, and searchable backend based on LC, enabling high-performance compression operator pipelines. The fused kernel achieves up to 5.9x speedup, while the new lossless backend increases compression throughput from below 90 GB/s to over 500 GB/s. Across the evaluated workloads, FastKVComp reduces job completion time by up to 2.04x. End-to-end evaluation shows that these optimizations allow KV compression to retain meaningful communication benefits under modern high-bandwidth network.

cs.DC

版本更新 · v2arXiv:2607.09153

KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling

Peng Kuang, Haibo Jin, Xiaoyu Han, Yanli Wang, Xiaopeng Yuan, Ye Yu, Kaidi Xu, Haohan Wang

原文摘要

Process Reward Models (PRMs) have been proven to be highly effective in guiding test-time scaling (TTS) methods, which significantly boost the capabilities of LLM-based multi-agent systems. However, existing PRMs are text-based: they re-encode the entire trajectory text from scratch. In long multi-agent rollouts, the scoring cost, growing quadratically with respect to sequence length L, creates a severe computational bottleneck, severely limiting PRMs' application in long-context scenarios. To resolve this, we introduce KV-PRM, a highly efficient process reward model that eliminates the heavy text re-encoding by directly reading the KV cache produced naturally during the LLM's generation phase. By processing a single "verify token" against the pre-existing KV cache, KV-PRM reduces the scoring cost from O(L^2) to O(L). We formally prove that the KV cache contains strictly greater information capacity than text, and is more efficient for downstream reward modeling. Empirically, across the MATH, GSM8K, and AIME benchmarks, KV-PRM matches or strictly outperforms text-PRMs under various TTS methods such as Beam Search, MCTS, and Weighted Voting, with up to a 5,000x reduction in scoring FLOPs, a 37x reduction in latency, and a 34x reduction in per-sequence memory footprint compared to text-based PRMs.

cs.AI

版本更新 · v2arXiv:2510.02194

UpSafe$^\circ$C: Upcycling for Controllable Safety in Large Language Models

Yuhao Sun, Zhuoer Xu, Shiwen Cui, Kun Yang, Lingyun Yu, Yongdong Zhang, Hongtao Xie

原文摘要

Large Language Models (LLMs) have achieved remarkable progress across a wide range of tasks, but remain vulnerable to safety risks such as harmful content generation and jailbreak attacks. Existing safety techniques---including external guardrails, inference-time guidance, and post-training alignment---each face limitations in balancing safety, utility, and controllability. In this work, we propose UpSafe$^\circ$C, a unified framework for enhancing LLM safety through safety-aware upcycling. Our approach first identifies safety-critical layers and upcycles them into a sparse Mixture-of-Experts (MoE) structure, where the router acts as a soft guardrail that selectively activates original MLPs and added safety experts. We further introduce a two-stage SFT strategy to strengthen safety discrimination while preserving general capabilities. To enable flexible control at inference time, we introduce a safety temperature mechanism, allowing dynamic adjustment of the trade-off between safety and utility. Experiments across multiple benchmarks, base model, and model scales demonstrate that UpSafe$^\circ$C achieves robust safety improvements against harmful and jailbreak inputs, while maintaining competitive performance on general tasks. Moreover, analysis shows that safety temperature provides fine-grained inference-time control that achieves controllable trade-off between utility and safety. Our results highlight a new direction for LLM safety: moving from static alignment toward dynamic, modular, and inference-aware control. Code: https://github.com/HansSunY/UpSafe.

cs.AI · cs.CR · cs.LG

新论文 · v1arXiv:2611.05984

On the Design of Hyper-Connections: Simplification, Stability, and Scalability

Xiangdong Zhang, Xiaohan Qin, Sunan Zou, Tuo Dai, Bizhao Shi, Zhuo Xia, Yebin Yang, Yizhou Chen 等 11 位作者

原文摘要

Hyper-Connections (HC) architectures are increasingly adopted in frontier language models, yet their components remain insufficiently understood and introduce non-negligible training overhead. In this work, we systematically study the HC architecture to understand its components and identify simpler, more efficient designs. We find that learned read/write gating without mixing and unconstrained mixing with fixed gates form two effective interaction groups: either alone achieves performance comparable to mHC. We also identify repeated access to the expanded residual state as a major source of training overhead, even without mixing. These findings lead to SimpleHC, which simplifies the architecture by removing residual mixing and simplifies execution through shared memory access, using the resulting additive structure and shared gate inputs. Layer-specific gates share a group-entry state, enabling batched reads and deferred writes; scalar corrections exactly preserve preceding sublayer output contributions. Our analyses support both the retained residual structure and the design of shared memory access. SimpleHC incurs only 2.4% extra end-to-end training time over a vanilla residual baseline at the largest tested scale. It maintains performance comparable to mHC when scaling from 15B to 30B MoE models, exhibits smoother gradient dynamics during standard training, and remains stable under stress. We hope SimpleHC will encourage a broader pursuit of simplicity in foundation model design, making performance gains more accessible and practical at scale.

cs.LG · cs.CL

新论文 · v1arXiv:2611.05982

Single-Exponential FPT Algorithms for Minimum Bisection and Matching Cut via a New Unbreakable Decomposition

Soh Kumabe

原文摘要

Minimum Bisection asks whether the vertices of a graph can be partitioned into two parts whose sizes differ by at most one and with at most k edges between the two parts. Cygan, Lokshtanov, Marcin Pilipczuk, Michał Pilipczuk, and Saurabh (SICOMP 2019) gave the first fixed-parameter algorithm, which runs in O*(2^(O(k^3))) time, by introducing an unbreakable decomposition. Cygan, Komosa, Lokshtanov, Marcin Pilipczuk, Michał Pilipczuk, Saurabh, and Wahlström (TALG 2021) strengthened the decomposition and improved the running time to O*(2^(O(k log k))). However, whether the running time could be improved to single-exponential remained open. In this paper, we give a deterministic O*(2^(O(k)))-time algorithm for Minimum Bisection. Our algorithm also counts all feasible bisections. To obtain an O*(2^(O(k)))-time algorithm, we introduce two new techniques. First, we give a new algorithm that computes a (2k,k)-edge-unbreakable decomposition with adhesion at most 2k in O*(2^(O(k))) time by relaxing the unbreakability and adhesion guarantees of the decomposition in (TALG 2021). Second, we improve the running time of the dynamic program over the decomposition from O*(2^(O(k log k))) to O*(2^(O(k))) using important cuts, inclusion-exclusion, and formal logarithms and exponentials. Together with color coding, the same techniques give a deterministic O*(2^(O(k)))-time algorithm for Matching Cut, which asks for a cut whose crossing edges form a matching. This improves the O*(2^(O(k log k)))-time algorithm of Aravind and Saxena (IWOCA 2021).

cs.DS

新论文 · v1arXiv:2611.05981

Single-Variable Connectivity in Random $k$-SAT

Chunyang Wang, Yitong Yin, Yixiao Yu, Yiyao Zhang

原文摘要

We resolve longstanding questions about connectivity, reconstruction,and looseness in random $k$-SAT, for sufficiently large fixed $k$ as the number $n$ of variables tends to infinity. $\textbf{Single-variable connectivity up to reconstruction.}$ At clause densities up to \[\frac{2^k}{k}\bigl(\ln k+\ln\ln k+γ_*-o_k(1)\bigr),\] a polynomial-time algorithm joins a typical pair of satisfying assignments with high probability, using $O_k(n)$ single-variable flips that preserve satisfiability. Here $γ_*\approx0.871$ arises from a refined tree-reconstruction analysis. $\textbf{Settling the reconstruction threshold.}$ Building on the tools developed for connectivity, we determine the clustering threshold defined by the reconstruction criterion to be \[\frac{2^k}{k}\bigl(\ln k+\ln\ln k+γ_*+o_k(1)\bigr),\] below the classical cavity prediction with constant term $1+O(\ln\ln k/\ln k)$. $\textbf{Shattering before rigidity.}$ We locate the looseness--rigidity transition at a higher density, matching the classical cavity prediction for clustering: \[\frac{2^k}{k} \left(\ln k+\ln\ln k+1+Θ\left(\frac{\ln\ln k}{\ln k}\right)\right).\] Below the transition, we establish a stronger form of looseness: any variable of a typical solution can be reversed by logarithmically many single-variable flips that preserve satisfiability. Above it, reversing most variables of a typical solution requires changing linearly many variables (rigidity). Yet throughout an interval between reconstruction and rigidity, the solution space already splits into groups of exponentially small relative size separated by linear Hamming distance and linear energy barriers (shattering), while logarithmic looseness persists.

cs.DS · cs.DM

新论文 · v1arXiv:2611.05979

Learning Object-Centric Motion Priors from Human for Robotic Dexterous Manipulation

Zhengdong Hong, Guofeng Zhang

原文摘要

Manipulating diverse objects with multi-fingered dexterous hands is challenging due to the high dimensionality and complex dynamics. Human-Object Interaction (HOI) datasets provide rich knowledge about task information and embodied interactions. Instead of solely imitating the human demonstrations, our method learns to holistically predict future hand-object states by leveraging these datasets. The predicted future states of the object can serve as a general-purpose reward term for reinforcement learning, reducing reliance on task-specific reward engineering and enhancing generalization across tasks. We conduct extensive experiments on three manipulation tasks in simulation and the real world. Our approach outperforms existing SOTA methods in both success rate and generalizability on novel objects. Furthermore, we validate the cross-embodiment compatibility of our methods by successfully deploying the skills on different robot hands.

cs.RO

版本更新 · v3arXiv:2610.27616

EchoNav: Evidential BEV Completion from Historical Observations for Visual Navigation

Diwen Liu, Yiyang Shao, Borui Wang, Chi Zhang, Jizhuo Chen, Jiaming Wang, Benyun Zhao, Zhongyu Li 等 9 位作者

原文摘要

Visual navigation in cluttered environments requires robots to recover previously observed structure and infer obstacles and free space that remain occluded. We present a framework that combines temporal aggregation and occupancy completion in an evidential bird's-eye-view (BEV) representation predicted from short RGB sequences. A decoder adapts frozen VGGT-Omega features and combines historical residual proposals using cellwise weights and a learned no-update option. The BEV predictor produces a local map in the latest ego frame without depth or camera-pose inputs or explicit geometric warping. The evidential model treats observed free space separately and predicts occupancy with associated confidence across the remaining supported region. We train the predictor on simulated RGB sequences paired with complete and visibility-masked occupancy annotations. Our experiments evaluate navigation in simulation and on a wheeled robot, with additional demonstrations on quadruped and humanoid robots. In simulated sequential two-goal navigation, our system achieves 86.3% success, outperforming the strongest evaluated baseline by 32.2 percentage points. Across 60 wheeled-robot trials in 20 scenes, it achieves 73.3% success, compared with 33.3% for the strongest evaluated baseline.

cs.RO

版本更新 · v2arXiv:2609.34639

Uniform Race: Parameter-Free Approximate Rejection Sampling

Seiyun Shin, Juhyeong Pang, Kwang-Sung Jun

原文摘要

Consider the approximate sampling problem: given $N$ independent samples from a proposal distribution $μ$, the goal is to select one whose distribution is close to a target $π$ specified only up to a normalizing constant. The seminal work of Block & Polyanskiy (2023) provides a total variation (TV) error bound for approximate rejection sampling (RS) as a function of the algorithm's acceptance threshold $M$. The ideal threshold $M$ that minimizes the bound, however, depends on properties of $(π, μ)$ that are typically unavailable from the observed sample. This raises a natural question: Can one attain the best RS guarantee without taking the best $M$ as input? We answer this question affirmatively by proposing a parameter-free sampling algorithm called uniform race (UR). UR computes a score for each sample $y$ that is the importance weight $π(y)/μ(y)$ divided by an independent uniform random variable and then returns the one with the highest score. We prove that UR attains the best $M$-tuned TV error bound for RS. That is, UR, without knowing the best $M$, achieves the same performance guarantee as RS that takes the best $M$ as input. We also prove that UR has no larger TV error than the budget-calibrated version of the rejection sampler of Rohatgi et al. (2025) and sampling importance resampling (SIR). In particular, we identify instances where UR's error is exponentially smaller in $N$ than that of these two baselines. Furthermore, we establish conditions under which attaining this RS guarantee for every $(π, μ)$ uniquely determines the selection probabilities as those of UR. Finally, test-time scaling experiments on LLM reasoning tasks corroborate the theoretical comparisons and demonstrate that UR remains competitive without requiring threshold selection.

cs.LG · stat.ML

版本更新 · v2arXiv:2609.34458

Understanding Generalization Requires Universal Induction

Aram Ebtekar, Marcus Hutter, Danica J. Sutherland

原文摘要

Classical statistical theory is insufficient to explain the successes of general-purpose AI models, because it depends on handcrafted inductive biases that it cannot justify. No Free Lunch (NFL) theorems force any learner that beats chance on some environments to underperform on others. We might hope that past experience informs which environments to expect, but NFL applies equally to meta-learning. Thus, any method that makes meaningful predictions necessarily begins with an inductive bias external to the data. Choosing to bias toward short programs yields Solomonoff induction (SI), whose performance is competitive against all computable learners - albeit up to "constants" that become large when comparing against specialized methods that exploit background information. We therefore relativize SI to an information vantage point, biasing toward short programs with access to all preexisting information. This reframes the inductive bias: instead of seeking some absolute notion of simplicity, we favour accessibility with respect to our vantage point. An algorithm can only outpredict the relativized SI to the extent that its code contains additional information about the data, and no algorithm can generate such information. While SI is incomputable and hence not a practical algorithm, it provides a formal optimum for inference in the limit of infinite compute, and there is evidence to suggest that frontier AI systems roughly approximate it. Thus, the only known answer to meta-NFL is rooted in algorithmic information theory, which we should expect to play a fundamental role in explaining the generalization behaviour of modern (and future) AI systems.

stat.ML · cs.AI · cs.IT · cs.LG · math.ST

新论文 · v1arXiv:2611.05974

PathoS2VD: From 2D Histopathology to Pseudo Z-Stack Representations

Jiamu Wang, Keunho Byeon, Ji-Uk Kim, Jin Tae Kwak, Heounjeong Go

原文摘要

For decades, diagnostic pathology has relied on thin two-dimensional (2D) tissue sections, despite biological tissues exhibiting inherently three-dimensional (3D) morphological organization. Because a single section captures only one focal plane, important morphological cues along the z-axis are inevitably discarded, limiting tissue representation for both human interpretation and computational analysis. In this work, we propose a generative framework that learns z-axis morphological priors from pathology z-stack data, offering a cost-effective alternative to dedicated volumetric imaging, and transfers this knowledge to conventional single-plane histopathology. Specifically, an image-to-video (I2V) latent diffusion model (LDM) is trained to synthesize a pseudo z-stack from a single 2D image, capturing plausible morphological variations across adjacent focal planes rather than reconstructing physical tissue volumes. To enable deployment on large-scale clinical datasets without native z-stack acquisition, we further introduce a prior-preserving mixture fine-tuning strategy that adapts the model to new domains while retaining the learned z-axis morphological prior. Extensive experiments demonstrate that the synthesized pseudo z-stacks exhibit realistic cross-plane morphological continuity and consistently improve downstream prostate cancer classification compared with conventional 2D representations. Our results suggest that learning transferable z-axis priors from limited z-stack data provides a practical way to enrich routine pathology images with complementary morphological information. Code: https://github.com/QuIIL/PathoS2VD.

cs.CV

版本更新 · v5arXiv:2603.19310

MemReward: Graph-Based Experience Memory for LLM Reward Prediction with Limited Labels

Tianyang Luo, Tao Feng, Zhigang Hua, Yan Xie, Shuang Yang, Ge Liu, Jiaxuan You

原文摘要

Reinforcement learning (RL) fine-tuning improves large language model (LLM) reasoning using reward signals on sampled rollouts. When reward labels are available for only a subset of rollouts, restricting policy updates to that subset leaves the remaining rollouts unused for policy optimization. Semi-supervised learning propagates labels from labeled to unlabeled samples, but existing methods either diffuse labels with fixed similarity weights or treat each sample independently, ignoring the relations among queries, thinking processes, and answers. We propose MemReward, a graph-based experience memory that propagates rewards to unlabeled rollouts during online policy optimization. MemReward stores queries and the rollouts (thinking processes and final answers) of an initial LLM policy as nodes of a heterogeneous graph connected by similarity and structural edges, and warms up a GNN on labeled rollouts to propagate rewards over this graph. During online RL fine-tuning, unlabeled rollouts are attached to the graph by query similarity, and the policy is updated with ground-truth rewards on labeled rollouts and GNN-predicted rewards on the rest. We evaluate MemReward on 13 benchmarks in mathematics, question answering, and code generation with verifiable rewards, and further with LLM-judge, human-preference, and continuous rewards. With ground-truth rewards on only 20% of rollouts, MemReward achieves 96.6% of the performance of fully supervised training (Oracle) on Qwen2.5-1.5B and 97.3% on Qwen2.5-3B, where it also outperforms label propagation and FixMatch-style self-training. It carries over to Llama backbones up to 8B and closely approaches Oracle on out-of-domain tasks.

cs.LG · cs.AI

版本更新 · v4arXiv:2603.03415

Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs

Mingyu Jin, Yutong Yin, Jingcheng Niu, Qingcheng Zeng, Wujiang Xu, Mengnan Du, Wei Cheng, Zhaoran Wang 等 10 位作者

原文摘要

In this work, we investigate how the internal representations of Large Language Models (LLMs) change with inputs of increasing difficulty. Although sparsity changes may not be obvious for individual samples, dataset-level results reveal a consistent statistical trend: as task difficulty increases (e.g., harder questions, longer contexts, or more answer choices), the last hidden states of LLMs become systematically sparser. In short, \textbf{the harder the task, the sparser the representation}. This sparsity--difficulty relation is observable across diverse models and domains, suggesting that harder inputs drive more concentrated activation patterns in the last hidden state. Through a series of controlled analyses and learning-dynamics experiments, we show that representational density is a learned property of data familiarity: models develop rich, distributed representations for mastered patterns, while unfamiliar inputs default to sparser activations. Our finding is also actionable. We illustrate this with Sparsity-Guided Curriculum In-Context Learning (SG-ICL), which uses sparsity to select few-shot demonstrations matched to the query's difficulty, outperforming standard CoT and Auto-CoT baselines on MATH-500. Our study provides new insights into how LLM representations reflect task difficulty and how this signal can be leveraged for inference. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.

cs.CL · cs.AI

新论文 · v1arXiv:2611.05971

Enforcing Accountable Disclosure in Confidential AI

Shumpei Shiina, Haruka Takahashi, Kento Sasaki, Takuya Hayashi, Tsubasa Takahashi

原文摘要

Advances in CPU and GPU confidential computing enable confidential AI, allowing LLMs to process user prompts without exposing them to service providers or infrastructure operators. However, AI agents often need to send prompt-derived information to external services, and providers may also require limited disclosure for safety monitoring and failure diagnosis. We propose CORDE, a system for confidential virtual machines (CVMs) designed to satisfy the principle of Confine or Declare: user-derived data is confined by default, but when necessary, it may leave the CVM only after being "declared." To "declare" data is to record its exact contents, recipient, and stated reason in a user-accessible, tamper-evident declaration log. To put this principle into practice, we present an integrated system design that combines per-user information-flow control in Rust with a TEE-based ledger and remote attestation. Our evaluation on an NVIDIA H100 GPU shows that declaration before disclosure adds little overhead in task latency, about 0.5% on the AppWorld agent benchmark. CORDE provides a basis for accountable disclosure of user data, supporting responsible and transparent data management in AI services.

cs.CR · cs.OS

新论文 · v1arXiv:2611.05970

Ray-Tracing-Based Dynamic Channel Sounding With Autoregressive Wi-Fi CSI Prediction

Lyutianyang Zhang, Yixin Ying, Liu Cao, Vanlin Sathya, Jinke Ren, Shuguang Cui

原文摘要

Channel sounding in Wi-Fi multiple-input multiple-output (MIMO) systems must balance channel state information (CSI) accuracy against CSI acquisition overhead. Device mobility causes channel aging, while autoregressive CSI prediction introduces errors that accumulate over sounding intervals. However, standardized statistical channel models cannot fully capture the environment-specific channel dynamics encountered in physical artificial intelligence (AI) applications, motivating the use of Sionna-based ray tracing to evaluate CSI prediction under device mobility and evolving propagation geometry. This paper develops a dynamic sounding framework that integrates autoregressive CSI prediction with signal-to-interference-plus-noise ratio (SINR)-constrained interval selection. A Transformer with rotary position embedding that converts device positions into CSI sequences predicts CSI across multiple subcarriers for beamforming updates between sounding exchanges. We formulate sounding-interval selection as a predicted average-throughput maximization problem under per-stream SINR lower-bound constraints, including CSI acquisition and processing overhead. A prediction-error perturbation analysis provides the SINR bounds, and cumulative-sum enumeration globally solves the finite single-variable problem. Actual SINR feasibility holds whenever the prediction errors satisfy the bounds. Ray-tracing experiments assess channel prediction, trajectory transfer, and autoregressive prediction errors. In single-STA Wi-Fi 7 link-level simulations, predicted-CSI beamforming improves mean goodput by 67.3%, 112.7%, and 119.1% over stale CSI for linear, circular, and cosine trajectories, respectively.

cs.IT · cs.AI · cs.NI

新论文 · v1arXiv:2611.05969

CLEAR: Color-Texture Latent Alignment for Single Image Reflection Removal

Tu Vo, Sheir A. Zaheer, Chan Y. Park

原文摘要

Single Image Reflection Removal (SIRR) seeks to recover the transmitted scene from a photograph captured through glass. Recent latent restoration methods use pretrained generative priors, but many recover detail by feeding high-frequency features from the degraded input back into the decoder. This works for degradations such as blur, noise, or haze, where input detail mostly belongs to the scene. It is less suitable for SIRR, where reflections add a second structured layer: directly reusing the input's skip connections can sharpen the image while also bringing reflection edges back. Latent compression also introduces color drift and fine-detail loss. We propose CLEAR (Color-Texture LatEnt Alignment for Single Image Reflection Removal), a deterministic, single-timestep, non-iterative latent framework that uses a pretrained latent diffusion backbone as a natural-image prior. A physics-guided separator organizes the prediction around the transmission, reflection, and residue formation model and outputs coarse layer latents in one forward pass. The Hybrid Feature Alignment Module (HFAM) then treats input-derived skips as contaminated layer-mixture features. It rectifies their high-frequency content with a clean-transmission-aligned color query, learned from the input features themselves under ground-truth feature supervision, keeping transmission-consistent texture while attenuating reflection-colored detail. CLEAR achieves the best average across fidelity and perceptual metrics on six real-world benchmarks and the lowest LPIPS on every dataset, with the largest gains on the hardest benchmarks.

cs.CV · cs.GR

新论文 · v1arXiv:2611.05968

InterSim: Innovative Intersection Digital Twins for Human-in-the-Loop Driving Simulation

Luyang Jiang, Chuheng Wei, Yaguang Li, Shaoqi Wang, Yupeng Zhou, Hyunsoo Yun, Xishun Liao, Brandon Pitts 等 9 位作者

原文摘要

Innovative intersection designs can improve roadway safety and operational efficiency, but their uncommon geometries may create challenges for driver comprehension during early deployment. Evaluating these designs through full-scale field implementation can also be costly. This paper presents InterSim, a Unity game engine-based human-in-the-loop (HITL) driving simulator that constructs digital twins of site-specific innovative intersections in a virtual environment. InterSim links the real world and the simulation world through two closed cycles: a digital twin cycle that builds each intersection twin from agency design and traffic data and returns evidence from simulation to the agency, and an HITL cycle in which human drivers or automated driving algorithms operate the ego vehicle under identical scenario conditions. The research capabilities of InterSim are demonstrated through two case studies: a surrogate safety study compares post-encroachment time distributions between a Reduced Conflict Intersection and a standard intersection under different traffic volumes; an eco-driving study compares human drivers with a rule-based agent using vehicle-to-infrastructure information and reveals differences in emission and driving behavior. These studies demonstrate that InterSim provides a reusable digital twin platform for investigating innovative intersection designs from diverse perspectives. Demo video and open-source code can be accessed at https://purduedigitaltwin.github.io/InterSim/

cs.CE · cs.CY · cs.ET

版本更新 · v2arXiv:2605.12112

When Policy Entropy Constraint Fails: Preserving Diversity in Flow-based RLHF via Perceptual Entropy

Xiaofeng Tan, Jun Liu, Bin-Bin Gao, Yuanting Fan, Xi Jiang, Chengjie Wang, Hongsong Wang, Feng Zheng

原文摘要

RLHF is widely used to align flow-matching text-to-image models with human preferences, but often leads to severe diversity collapse after fine-tuning. In RL, diversity is often assumed to correlate with policy entropy, motivating entropy regularization. However, we show this intuition breaks in flow models: policy entropy remains constant, even while perceptual diversity collapses. We explain this mismatch both theoretically and empirically: the constant entropy arises from the fixed, pre-defined noise schedule, while the diversity collapse is driven by the mode-seeking nature of policy gradients. As a result, policy entropy fails to prevent the model from converging to a narrow high-reward region in the perceptual space. To this end, we introduce perceptual entropy that captures diversity in a perceptual space and maintains the property of standard entropy. Building upon this insight, we propose two entropy-regularized strategies, Perceptual Entropy Constraint and Perceptual Constraints on Generation Space, to preserve perceptual diversity and improve the quality. Experiments across two base models, neural and rule-based rewards, and three perceptual spaces demonstrate consistent gains in the quality-diversity trade-off; PEC achieves the best overall score of 0.734 (vs. baseline's 0.366); a complementary setting of PEC further reaches a diversity average of 0.989 (vs. baseline's 0.047). Our project page (https://xiaofeng-tan.github.io/projects/PEC) is publicly available.

cs.CV

版本更新 · v2arXiv:2611.04878

TraceRouter++: Causal Semantic Routing Realignment via Optimal Transport

Chuancheng Shi, Shangze Li, Wenhua Wu, Hengzhi Chen, Fei Shen, Li Zhuo, Zhiyong Wang, Tat-Seng Chua

原文摘要

Despite their remarkable capabilities, large foundation models (LFMs) remain susceptible to adversarial manipulation. Current safety intervention methods primarily rely on the \textit{locality hypothesis}, rigidly suppressing isolated neurons or features. However, harmful semantics manifest as distributed, cross-layer circuits, rendering localized ablations both brittle against complex jailbreaks and detrimental to general utility. To bridge this gap, we propose TraceRouter++, a routing-level safety intervention framework that explicitly traces safety-relevant causal routing pathways activated under harmful inputs and realigns their unsafe activation states. Our framework operates in a \textit{Discover-Trace-Realign} paradigm: it first pinpoints a sensitive onset layer via attention divergence and disentangles dense activations into interpretable routing features using sparse autoencoders (SAEs). Subsequently, it maps these features to downstream causal circuits via feature influence scores (FIS). Crucially, departing from rigid suppression paradigms, TraceRouter++ introduces a constrained optimal transport (OT)-based soft routing realignment. By formulating safety intervention as a tractable coordinate-wise marginal optimal transport approximation, our method redirects unsafe activation marginals toward safe-reference marginals within causally identified routing subspaces while preserving non-selected coordinates at the intervention point. This mechanism safely redirects malicious semantic flows while leaving orthogonal computational routes intact. Extensive experiments across representative large language models (LLMs), multimodal large language models (MLLMs), and diffusion models (DMs) show that TraceRouter++ outperforms evaluated baselines, improving adversarial robustness while maintaining a favorable safety-utility trade-off. WARNING: This paper contains unsafe model responses.

cs.CL

新论文 · v1arXiv:2611.05959

Local information processing capacity: anatomy of computation in neuromorphic systems

Jiaxuan Chen, Daiki Nishioka, Yicheng Song, Ryo Iguchi, Sota Hikasa, Hina Kitano, Akira Hirose, Takashi Tsuchiya

原文摘要

Many neuromorphic systems possess rich internal computational structures shaped by their physical dynamics. However, existing metrics characterize only global computation, leaving these local structures unresolved despite their importance for system interpretation, optimization and miniaturization. Here, we introduce local information processing capacity (L-IPC), which represents individual dynamical states as vectors in an infinite-dimensional Hilbert space and decomposes their computational properties through subspace projections onto nonlinearity orders and memory steps, with the residual quantifying computational irrelevance. Applied to three systems with comparable global capacities (an echo state network, an ion-gating reservoir and a spin-wave reservoir), L-IPC reveals markedly different distributions of computation, ranging from homogeneous to spatially, temporally and spectrally structured patterns. These local profiles further guide task-specific pruning, outperforming standard approaches at the same size in the structured systems. These results establish L-IPC as a general framework for anatomizing computation and guiding task-adaptive design in neuromorphic systems.

cs.ET · physics.app-ph

版本更新 · v3arXiv:2505.09338

Llama See, Llama Do: A Mechanistic Perspective on Contextual Entrainment and Distraction in LLMs

Jingcheng Niu, Xingdi Yuan, Tong Wang, Hamidreza Saghir, Amir H. Abdi

原文摘要

We observe a novel phenomenon, contextual entrainment, across a wide range of language models (LMs) and prompt settings, providing a new mechanistic perspective on how LMs become distracted by ``irrelevant'' contextual information in the input prompt. Specifically, LMs assign significantly higher logits (or probabilities) to any tokens that have previously appeared in the context prompt, even for random tokens. This suggests that contextual entrainment is a mechanistic phenomenon, occurring independently of the relevance or semantic relation of the tokens to the question or the rest of the sentence. We find statistically significant evidence that the magnitude of contextual entrainment is influenced by semantic factors. Counterfactual prompts have a greater effect compared to factual ones, suggesting that while contextual entrainment is a mechanistic phenomenon, it is modulated by semantic factors. We hypothesise that there is a circuit of attention heads -- the entrainment heads -- that corresponds to the contextual entrainment phenomenon. Using a novel entrainment head discovery method based on differentiable masking, we identify these heads across various settings. When we ``turn off'' these heads, i.e., set their outputs to zero, the effect of contextual entrainment is significantly attenuated, causing the model to generate output that capitulates to what it would produce if no distracting context were provided. Our discovery of contextual entrainment, along with our investigation into LM distraction via the entrainment heads, marks a key step towards the mechanistic analysis and mitigation of the distraction problem.

cs.CL

新论文 · v1arXiv:2611.05956

The Complexity of Unpredictability

Aryan Agarwala, Dani Dorfman

原文摘要

We initiate the study of middle-bit predictors: functions that take as input a string with one missing bit and try to predict that bit. These are closely related to next-bit predictors, a classical object in pseudorandomness. We study the deterministic complexity of computing an unpredictable distribution (UNPRED). Given a list of middle-bit predictors in circuit form and a parameter $ε>0$, output a multiset of strings such that each predictor succeeds with probability less than $1/2+ε$ on a uniformly sampled string from the multiset. We write $\mathsf{UNPRED}^{\mathsf{TC}^1}$ for the restriction to $\mathsf{TC}^1$ predictors. We show that UNPRED offers a unified route towards resolving several major open problems in catalytic computation, the biggest of which is whether deterministic polynomial time (P and FP) can simulate catalytic logspace (CL and SearchCL). It is believed that $\mathsf{SearchCL}\subseteq\mathsf{FP}$. We prove $$\mathsf{UNPRED}\in\mathsf{FP}\implies\mathsf{SearchCL}\subseteq\mathsf{FP}\implies\mathsf{UNPRED}^{\mathsf{TC}^1}\in\mathsf{FP}.$$ Thus, the questions of whether UNPRED is in FP and whether SearchCL is contained in FP are almost equivalent, up to the restriction to $\mathsf{TC}^1$ circuits. We also show that any deterministic polynomial-time algorithm for $\mathsf{UNPRED}^{\mathsf{TC}^1}$ must exploit properties of $\mathsf{TC}^1$ circuits, either in the algorithm itself or in its analysis. To our knowledge, this would also be the first algorithm to use the structure of $\mathsf{TC}^1$ circuits in this way to solve a natural problem, thus advancing our understanding of Boolean circuits. Together, our results provide strong evidence that middle-bit predictors could be the key to unlocking several major open problems in catalytic computation, pseudorandomness, and Boolean circuit complexity.

cs.CC

版本更新 · v2arXiv:2611.03114

3D Gaussian Splatting within Self-Learned Neural View-Dependent Color Fields

Yixiao Yang, Yu-Shen Liu, Zhizhong Han

原文摘要

3D Gaussian Splatting (3DGS) has become a vital tool in novel view synthesis. 3D Gaussian functions with attributes like color, opacity, and scale are stacked together to explicitly model a radiance field. However, discrete 3D Gaussians usually produce poor depth with few geometry details but lots of floaters, which hinders the effective modeling of smooth radiance fields. To address this issue, we propose to learn a radiance field within a continuous view-dependent color field parameterized by neural networks. With continuous constraints in color, the bias on low-frequency signals of neural networks can highly encourage the geometry modeling to contribute to high-frequency color variations on images, getting rid of solely relying on the color attribute itself. Moreover, we impose a constraint to improve the Lipschitz continuity of the neural color field, making large changes of each Gaussian's color also match with large changes of its position in terms of a ratio. To this end, we additionally introduce novel self-learning strategies to improve the learning of the neural view-dependent color field without any external knowledge or priors, aiming for better generalization on unseen scenes. Our numerical and visual comparisons on widely used benchmarks justify our idea and show better ability of high fidelity geometry recovery in 3DGS. Project page: https://github.com/yangyixi21/Neural-Color-Fields-3DGS.

cs.CV

新论文 · v1arXiv:2611.05954

Robots in the City: Persona-Based Perceptions of Urban Robots

Keonhee Jang, Kanghoon Lee, Seokwoo Yun, Namwoo Kim, Yoonjin Yoon

原文摘要

Pedestrians increasingly encounter robots in urban public spaces without interacting with them, but how robots change the perceived safety of a street remains unclear. We examine how robot presence and type affect perceived safety across urban contexts. We generate 18 images of Seoul streets that vary in time of day, road class, and robot condition (no robot, a delivery robot, or a humanoid). Two multimodal large language models (MLLMs), Qwen3.8-27B and Gemma4-26B, each conditioned on 500 synthetic personas, choose the safer image in each pair, and 25 human respondents complete the same task. Both models prefer streets without robots and favor the delivery robot over the humanoid. Robots are chosen more often at night and less often by older personas, whereas road class has smaller effects. In the model rationales, the delivery robot appears as a physical obstacle, whereas the humanoid appears as either a threat or a source of protection. Like humans, both models chose robots more often at night, but chose the humanoid over the delivery robot far less often, and Gemma showed a night effect for the humanoid about three times that of humans. These results show that the effect of a robot on perceived safety depends on when it appears, and that persona-conditioned MLLMs capture only part of human judgments.

cs.HC · cs.CL · cs.RO

新论文 · v1arXiv:2611.05953

Graph-Mediated Large Language Model Systems in the Agentic Era: Reasoning, Knowledge, Retrieval, Execution, and Coordination

Haihui Liu, Yang Chen, Zice Qiu, Yun Zhang, Jialun Zhong, Wenhua Wang, Yanzeng Li, Tian Wang

原文摘要

Graphs play an increasingly important role in large language model (LLM) systems, supporting structural reasoning, knowledge grounding, retrieval, and memory. As these systems evolve toward the agentic era, graphs also organize execution dependencies and multi-agent communication, linking intermediate observations and retained information to subsequent decisions. However, understanding their contribution requires distinguishing the function of each graph operation, the component responsible for its computation, and the subsequent use of its output. This survey introduces an operation-level role--locus taxonomy for reasoning, knowledge and evidence, retrieval and memory, execution, and coordination. We compare how graph observations and external computations guide decisions, how memory representations and updates support later tasks, and how dependencies constrain execution and information delivery. These comparisons distinguish improvements in local metrics from downstream task performance while retaining the conditions under which results were reported. We then organize benchmarks, comparison designs, process diagnostics, and system costs into an evaluation framework. Finally, we identify open challenges in adaptive computation and stopping, memory maintenance, and dependency-aware recovery. The survey provides a basis for assessing when graph mechanisms are useful and for evaluating their effects across components and stages.

cs.AI

版本更新 · v3arXiv:2605.20342

ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning

Zuhao Yang, Kaichen Zhang, Sudong Wang, Keming Wu, Zhongyu Yang, Bo Li, Xiaojuan Qi, Shijian Lu 等 10 位作者

原文摘要

Training large multimodal models (LMMs) via reinforcement learning (RL) to natively invoke video-processing tools (e.g., cropping) has become a promising route to long-video understanding. However, most existing agentic Video-LMMs dispatch tool calls sequentially (i.e., one per turn). Sequential tool calling can propagate localization errors, accumulate irrelevant context, and require an additional main-model generation for each successive crop. We introduce ParaVT, a multi-agent RL framework for Parallel Video Tool calling, dispatching multiple time-window crops in a single turn, allowing evidence from multiple temporal windows to be examined concurrently. Yet applying standard Group Relative Policy Optimization (GRPO) to ParaVT reveals an obstacle we term the Tool Prior Paradox: inherited tool-use priors appear to support exploration while competing with the output format learned during supervised fine-tuning. Standard RL can also settle on tool skipping. A comparison with a model with weaker tool-use priors shows stable formatting but no tool calls during RL, consistent with a trade-off between format stability and tool exploration. We propose PARA-GRPO (Parseability-Anchored and Ratio-gAted GRPO), which augments standard RL with two complementary mechanisms: (i) a format reward that encourages tag closure and penalizes unclosed reasoning blocks, and (ii) a per-prompt randomization of the overview-frame budget to create a reward advantage for tool use when the overview provides insufficient evidence. Across six long-video understanding benchmarks, ParaVT improves over the Qwen3-VL baseline by +7.9% on average, with PARA-GRPO raising training-time format reward from 0.13 to a peak of 0.64. These findings highlight the need to account for inherited tool-use priors when designing RL methods for long-video understanding.

cs.CV