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

+1084新增论文
+883其中 AI 相关
+130已有论文修订
+94新增 AI 导读

arXiv 最近完成扫描:2026-11-18 04:28 · 全站最近入库:2026-11-18 19:45各信源进度 →

统计口径与数据新鲜度

统计 2026-11-18 06:08 — 2026-11-19 06:08(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:2026-11-18 04:28最近发现新增版本:2026-11-18 19:45库内论文最新更新时间:2026-11-18 08:00

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。

26,870 已收录论文22,374 AI 相关8,156 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 277 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-18 04:27(北京时间)
新论文 · v1arXiv:2611.23189

面向通用医学图像分割的检索增强校正

Retrieval-Augmented Correction for Generalist Medical Image Segmentation

Vincenzo Marcianò, Xiaoming Zhang, Michela Antonelli, Sebastien Ourselin, Maria A Zuluaga

AI 速读 基于摘要

GRACE 用共享的已验证病例知识库为掩膜校正提供检索上下文,训练一次即可跨多器官、多模态编辑分割掩膜。

cs.CV · cs.AI

查看原文摘要

Automated segmentation is routinely used as the first step in quantitative image-based clinical workflows, where predicted masks are subsequently reviewed and corrected by human operators. Automated correction of the residual structural errors is currently addressed one anatomy at a time, so the number of trained correctors scales with the number of segmentation targets. We present GRACE, a retrieval-augmented refiner that edits an input mask under context drawn from a shared knowledge base of verified prior cases. The refiner is trained once on a multi-organ corpus spanning compact organs, cavities, branching vessels, and tumour lesions. In the held-out dataset-transfer evaluation, three held-out datasets query a fixed knowledge base built from training data only. A separate leave-one-anatomy-out experiment tests parameter-free index extension: verified cases of an unseen solid organ enter the knowledge base, and the refiner stays frozen. A light shape-aware projection head on a frozen multimodal encoder maps embedding similarity to the shape agreement of the masks, so the knowledge base returns cases that the refiner can use as a shape prior. We evaluate GRACE on 18 public datasets spanning CT, MRI and ultrasound, using nine refinement configurations and seven upstream segmentors that span weak, mid, and strong operating regimes. We report each change with the DSC of its input mask. On the weak tier, the input DSC is 0.386 and GRACE adds up to +0.35 DSC. On a strong segmentor, the input DSC is 0.932 and the mean change is small. Transfer to an imaging modality that is absent from training and from the index is partial and depends on the anatomy, and remains an open research direction. Our results suggest retrieval-conditioned refinement is a practical alternative to anatomy-specific correction models. Code: https://github.com/robustml-eurecom/grace

新论文 · v1arXiv:2611.23154

面向推理时目标的分层策略学习

Learning Stratified Policies for Inference-Time Objectives

Mridul Mahajan, Xuezhou Zhang, Aldo Pacchiano

AI 速读 基于摘要

作者提出将推理时的采样预算分配给一组互补的 LoRA 策略(分层策略),以提升大语言模型的 pass@k 表现。

cs.LG · cs.AI

查看原文摘要

Inference-time scaling improves the performance of large language models by sampling multiple responses and evaluating their collective utility. However, many existing approaches obtain these samples by repeatedly sampling from a single policy. In this work, we study if the sample budget at inference time can instead be allocated across a population of policies. We introduce stratified policies, where each of $k$ learned policies contributes one response to a $k$-sample inference-time objective. We realize this policy population using separate LoRA adapters on a shared language model backbone and optimize it for pass@$k$. We theoretically motivate the approach by analyzing the linear setting; a population of LoRAs can achieve zero best-of-$k$ approximation error in settings where every single LoRA has strictly positive prediction error. Empirically, we demonstrate that stratified policies consistently improve pass@$k$ over a single-policy baseline across multiple reasoning tasks. Overall, our results highlight how distributing samples across complementary policies can be a more effective use of sample budget than repeated i.i.d. sampling from a single policy at inference time.

新论文 · v1arXiv:2611.23118

审计学习型四足机器人模拟器:ANYmal D 上的机器人世界模型

Auditing a Learned Simulator for Legged Robots: The Robotic World Model on ANYmal D

Joyjeet Singh

AI 速读 基于摘要

作者报告了对 ANYmal D 四足机器人学习型模拟器 RWM 及其不确定性版本 RWM-U 的端到端审计,发现单步检查会误判设计选择,且模拟器精度不能预测策略行走表现。

cs.RO · cs.LG

查看原文摘要

A legged robot can learn to walk inside a learned simulator: a neural network that predicts the robot's next state and is run forward on its own predictions. They are usually trained and checked one step ahead, although a policy uses them for a hundred steps or more. We audit one published system end to end, without a GPU: the dynamics model of the Robotic World Model (RWM) for the ANYmal D quadruped, and the uncertainty estimate of its follow-up, RWM-U. We rebuilt the model to match the released code's outputs, losses and gradients exactly, measured 101 trained simulators, 200 compressed copies and 38 simulators of three other architectures, and trained policies inside simulators that we tested in a hand-built MuJoCo model of the robot. Training on the model's own rollouts beat one-step training at 368 steps on all ten held-out episodes (registered), although one-step training was better one step ahead. One-step checks therefore misjudge design choices: pruning 30% of the weights kept 90% of the model's skill one step ahead and 39% at 2 s. Ensemble disagreement ranked rollout error well (Spearman +0.90 at 100 steps), yet covered 4.6% of outcomes at one sigma, against 68% for a calibrated estimate, as each network's own variance collapses under the implemented loss. Two checks that need no recorded future ranked some pools of simulators well, but neither was shown better than 8-step error in a primary test. At the robot level, accuracy has not predicted walking: all 18 fine-tuned copies we tested predicted better than the released model one and eight steps ahead, yet 13 gave failing policies, and registered tests of whether any check ranks simulators by their policies were not settled. In a secondary test, running the authors' policy inside each simulator ranked them well. We also report nine of our own claims that we withdrew. Scope: one robot, one gait, simulated data, no hardware.

新论文 · v1arXiv:2611.23113

LUTBench:文本驱动摄影风格迁移基准测试

LUTBench: Benchmarking Text-Driven Photographic Look Transfer

Chuanzhi Xu, Boyu Wei, Chengkun Yue, Xuanhua Yin, Yiyang Li, Chupeng Liu, Zihan Deng, Weidong Cai

AI 速读 基于摘要

作者构建了基于91个3D LUT与FiveK照片的文本驱动摄影调色基准,用统一协议比较16种方法在目标保真度、输入转换敏感性和泛化性上的表现。

cs.CV

查看原文摘要

Photographic color grading shapes the tone and color of a photograph to achieve a desired look. Text-driven photographic look transfer lets users specify these adjustments in natural language, with the goal of reproducing the requested look across photographs while preserving their content. However, prior research on text-driven color grading uses different datasets, text instructions, target definitions, and metrics. These differences make it difficult to compare how faithfully the methods reproduce the same look. To compare methods under a shared protocol, we introduce LUTBench. We collect 91 three-dimensional lookup tables (3D LUTs) and define one photographic look per LUT with a specified input conversion. Applying these looks to 5,000 MIT-Adobe FiveK photographs produces 455,000 input-target image pairs with reproducible target images. For each look, we use templates to turn measured tone and color changes on fixed color samples into candidate clauses. We use training image measurements to verify applicable clauses before composing the text description. The evaluation measures target fidelity, tests sensitivity to predefined input conversions, and separates generalization to unseen looks from reproduction of seen looks on new photographs. Across 16 methods, stronger text alignment does not guarantee target fidelity. However, most supervised variants improve fidelity on both seen and unseen looks.

新论文 · v1arXiv:2611.22828

TabDAG: Structured Evolutionary Search for Tabular Feature Engineering

Anastasia Volkova, Andrey Kuznetsov, Ivan Oseledets, Valentin Khrulkov

原文摘要

Feature engineering can improve tabular prediction, but useful transformations are compositional, data-dependent, and subject to leakage constraints. We present TabDAG, a framework that combines evolutionary search with a graph representation of feature-engineering programs. Nodes implement row-wise or data-fitted transformations; edges encode dependencies among raw variables, intermediate values, and final features. The representation supports interactions, statistical encodings, clustering- and neighborhood-based features, and multi-stage compositions in a single executable program. Mutations are structured edits constrained by a parent-specific schema and checked before evaluation. In our experiments, no structurally invalid program reached evaluation, although proposals could be rejected and valid graphs could still fail at runtime. Across ten regression and classification datasets and three tabular model families, evolved features improve predictive performance over raw inputs. TabDAG and its combination with OpenFE achieve the two best mean ranks among the compared methods; pairwise rank differences between feature-engineering methods do not survive multiple-testing correction. The LLM-based searches use the same backbone and candidate-evaluation budget. A controlled ablation compares free-form Python, whole-graph JSON, and constrained graph edits with the search engine, budget, and backbone held fixed. Constrained edits yield fewer failed evaluations than either rewrite baseline. The resulting graphs expose the transformations and dependencies selected by search, enabling feature-level analysis of the observed gains.

cs.NE · cs.LG

新论文 · v1arXiv:2611.22778

A Full-Ledger Counterfactual Audit of Adaptive Reasoning

Xiaopeng Chu, Wenzheng Kong, Wei Liu, Wenyi Zhang

原文摘要

Adaptive reasoning must decide whether to answer now or spend more compute. Predicting that the current answer is wrong does not establish whether further reasoning will improve it. We introduce the Full-Ledger Counterfactual Value Audit (FLCVA), which compares an immediate answer with four fixed-length continuations from the same frozen prefix under one answer procedure and accounts for the cost of every signal used by a policy. On GSM8K and MATH-500, a random 256-token continuation recovers a wrong answer in 26.6% of Qwen3-8B cases and 18.9% of DeepSeek-R1-Distill-Qwen-7B cases, while damaging a correct answer in 1.4% and 1.9%. Correctness labels from independently sampled continuations predict a separate continuation's outcome in both primary models. In offline replay, using these labels to allocate compute improves performance on DeepSeek; the Qwen3 gain remains inconclusive. Under full cost accounting, the tested lightweight, myopic controllers mainly recognize current errors and add little continuation-specific value beyond that shortcut. A Llama-8B replication and action/cost ablations examine the robustness of these findings. FLCVA separates measuring local continuation value, predicting it from observable signals, and using those predictions to allocate computation.

cs.AI · cs.CL

版本更新 · v2arXiv:2610.36509

ARISE: Authentic Reference-Informed Stability Enhancement for Continual Multimodal Forgery Detection

Yuchen Zhang, Yaxiong Wang, Yujiao Wu, Lianwei Wu, Qi Zhang, Li Zhu, Chaofeng Dong, Le Wu 等 9 位作者

原文摘要

Multimodal forgery detection is commonly studied under a static assumption: training attacks are expected to approximate those encountered after deployment. This assumption becomes increasingly fragile as generators, editing systems, and multimodal large language models continually introduce new forgery mechanisms and forensic evidence. We formulate this problem as attack-incremental continual multimodal forgery detection, where the binary real/fake interface remains fixed while emerging attacks must be learned without forgetting previous ones. We introduce Fable-Bench, which organizes attack evolution through complementary joint-modality, text-side, and image-side protocols. We further propose ARISE (Authentic Reference-Informed Stability Enhancement), motivated by the open-ended forged class and the comparatively reusable geometry of authentic content. ARISE fits a real-only Gaussian mixture model to projected authentic replay representations as a compact historical reference. Meanwhile, the proposed Gaussian-Responsibility-Aware Stability Preservation (GRASP) derives a sample-adaptive anisotropic metric that strongly constrains historically stable representation directions while retaining plasticity elsewhere. Our Vulnerability-Aware Replay (VAR) further estimates stage-wise vulnerability from authentic-density degradation, forged-to-authentic boundary confusion, and authentic-mode occupancy drift, then assigns bounded replay emphasis to fragile stages. All adaptive signals are obtained from training-time replay statistics, while inference uses only the original multimodal backbone and binary classification head. Experiments across three attack-evolution protocols and multiple detector architectures demonstrate improved historical retention and final average performance over representative continual-learning baselines.

cs.CV

版本更新 · v2arXiv:2610.08524

No Need to Stop the Fleet: Localized Anomaly Isolation via Dead Zones in AGV Fleets

Natalia Ogorelysheva, Falk Howar, Jian-Jia Chen

原文摘要

Automated Guided Vehicle (AGV) fleets in production and logistics environments rely on a system-wide emergency stop to handle local anomalies such as malfunctions, hazards, or contaminated areas. While safe, this approach halts the entire fleet even when only a small area is affected, causing unnecessary downtime. This paper proposes dead zones -- a formally defined localized isolation approach that serves as an additional fail-operational mechanism for situations where a full system-wide halt is not strictly needed. Rather than shutting down the entire fleet, a dead zone isolates the anomalous area, allowing unaffected robots to continue operating normally. We introduce two operational strategies: basic dead zone handling, in which affected robots enter a dead zone and stop, and dead zone handling with escaping, in which robots that are sufficiently far from a dead zone reroute around it. For escaping on counterclockwise trajectories we prove collision-freedom with a bounded arrival delay for all escaping robots, provided a minimum pairwise separation is kept between robots. Experimental evaluation confirms the theoretical guarantees and characterizes the throughput tradeoffs between all three approaches: full emergency stop, basic dead zone handling, and dead zone handling with escaping.

cs.RO

新论文 · v1arXiv:2611.22404

Symbolic Abstraction Refinement

Daniel Bryce, Robert P. Goldman, David J. Musliner, Jeremy Gottlieb, John Skovbekk

原文摘要

Autonomous systems choose actions whose outcomes are uncertain and are subject to time constraints. Probabilistic timed automata (PTAs) model exactly this, and verifying one answers a question about the system's decision-making: the maximum probability of reaching a goal, or the minimum probability of reaching a hazard, that a controller can guarantee. Verification normally works over symbolic states, each pairing a discrete location with a region of clock values, and its difficulty is that the best behavior generally varies within such a region. The PRISM verifier's most capable PTA method responds by splitting regions until each is uniform, rebuilding the transitions that reach them and propagating the split backward through the state graph of a stochastic game, a cascade of graph reconstruction that partitions regions the property never depends on. We present SYMAR, which instead makes the value function symbolic: each symbolic state carries a function from sub-regions of its clock region to probability bounds, so refinement is the appearance of new sub-regions in that function, and previously generated states are never split nor their incoming transitions reconstructed. Both sources of PTA nondeterminism, the timing and the choice of actions, are captured implicitly by the preimage computation that produces those sub-regions. Heuristic forward search then bounds how much refinement is done at all, expanding an And-Or graph from the initial state along the most promising behaviors so that effort is confined to the property-relevant fragment, guided by bounds and preferred actions from a single unrefined run of the existing method; a partial-refinement backup keeps the otherwise combinatorial sub-region count lean, and every stored interval provably contains the true probability. On the PRISM PTA benchmarks, whose subproblems recur in autonomous systems (distributed agreement under contention, leader election, and self-configuration without central coordination), SYMAR is up to 108x faster, solves an instance on which the existing method exhausts the time limit, and is markedly less sensitive to the deadline constant, resolving one instance in a 37-node search graph, unchanged as the deadline triples, where the existing method needs 86,380 abstract states and 73 refinement rounds.

cs.AI

新论文 · v1arXiv:2611.22190

Emergence and Collapse of Leader-Follower Flocks through Phantom Targets: Surprise Nullification in Bayesian and Inverse Bayesian Inference

Shuji Shinohara, Daiki Morita, Yoshihiro Nakajima, Takeshi Takano, Masakazu Higuchi, Ung-il Chung, Yukio-Pegio Gunji

原文摘要

In Bayesian and inverse Bayesian (BIB) inference, a large prediction error can trigger surprise nullification, which expands belief and likelihood variances by a common factor, leaving the updated variance ratio the same as in standard Bayesian updating. This can increase belief uncertainty while keeping the learning rate low. In the simulations, three agents used BIB inference to predict one another's positions and moved toward those predictions, repeatedly forming leader-follower flocks that subsequently collapsed despite having no alignment rules or designated leaders. Three leadership criteria (front position, heading stability, and flock initiation) identified the same agent in 89% of eligible flocks, with different agents serving as leaders across flocks. Flocks formed when an agent nearly stopped incorporating another's observations (link cut) and moved straight toward a predicted position detached from reality (phantom target), thereby attracting followers. Link cuts arose when two successive resets assigned a very small and then an ordinary likelihood variance, sharply increasing the likelihood-to-belief variance ratio. Flocks collapsed upon link recovery or when returning phantom targets balanced the leader's forward and backward attractions. Without nullification, cuts persisted throughout the analysis period; expanding only belief variance caused recovery within a few steps. Expanding likelihood variance as well allowed cuts to persist while belief uncertainty grew, enabling later resets and recovery. Thus, nullification preserves relative precision, allowing cuts and flocks to persist, while reducing absolute precision to enable recovery and collapse, reconciling these seemingly opposing processes through a single operation. These results suggest that, in this model, individually inaccurate predictions can give rise to collective order.

cs.MA · nlin.AO

版本更新 · v2arXiv:2606.26525

Theory-Scale Auto-Formalization of Logic for Computer Science

Yuming Feng, Frederick Pu, One An, Osbert Bastani, Li Zhang, Jiani Huang, Xujie Si, Ziyang Li

原文摘要

Auto-formalization is critical for scalable formal verification, but existing progress largely focuses on isolated statements, while theory-scale auto-formalization, which coherently translates hundreds of interdependent definitions, lemmas, and theorems, remains open due to challenges in consistency, faithfulness, scalability, and correctness. In this paper, we introduce LCS-Bench, a stand-alone, theory-scale benchmark based on Logic for Computer Science. LCS-Bench is built through a novel semi-automated agentic pipeline that leverages concept graphs, formal signature planning, issue tracking, sorry-filling with counter-example search, complemented by faithfulness review from human experts. The resulting artifact covers 327 textbook items, over 4,076 Lean declarations, and more than 85K lines of Lean code. The dataset supports broad evaluation through a data engine that automatically derives five tracks of evaluation benchmarks, measuring different aspects of auto-formalization and theorem-proving capabilities. We also introduce a novel evaluation protocol featuring definitional equivalence checkers, enabling more fine-grained and faithful assessment. Through extensive evaluation on 17 models, we demonstrate that (1) LCS-Bench is of high quality, consistent, and faithful; (2) the benchmark is challenging, with state-of-the-art models achieving only 20.3% on auto-formalization tasks; and (3) our analysis reveals key findings regarding theory-scale auto-formalization and suggests promising directions for future work.

cs.LG · cs.LO · cs.PL

新论文 · v1arXiv:2611.22030

What Does an Agent Need? Bundle-Aware Agent Resource Composition

Yunxiao Shi, Tingwei Chen, Dannong Xu, Ahad N. Zehmakan, Jialie Shen, Zhendong Chu, Qingsong Wen

原文摘要

An agent resource's value depends on its backbone and companion resources and is known only after execution. Token-based generators remain tied to their training catalog, while independent resource scoring misses complementarities and reused bundles target other tasks. We formulate Task-Conditioned Agent Resource Composition (AgentComposition): jointly selecting a backbone model and a tool-skill bundle to maximize end-to-end execution utility. Our Retrieval-Grounded Bundle-Aware Agent Generation retrieves candidate resources and past configurations of similar tasks. A fine-tuned generator reads resource descriptions and recombines candidates through constrained beam search, allowing new resources without retraining. A bundle-aware critic ranks complete configurations using a large language model (LLM) teacher and historically successful configurations as execution proxies, and adapts to execution feedback when available. We evaluate reference recovery, offline execution with judged and native rewards, and online recommendation with real execution, building AgentSelect-Exec with 5,060 executed configurations. Our method recovers the complete reference configuration within ten candidates for 67.73% of AgentSelect queries, nearly twice the strongest baseline, and leads all baselines on SkillsBench. The proxy-trained critic raises the top-ranked agent's execution score from 7.58 to 8.37 out of 10; execution feedback increases it to 8.78, above the strongest baselines. Four target-domain queries suffice to adapt the generator to SkillsBench. Agents recommended online outperform the strongest baselines under both an LLM judge and native verifiers. These results support learning and transferring joint resource composition from task experience.

cs.AI

新论文 · v1arXiv:2611.21973

How Many Persistent Features Does a Biometric Need? A Law Linking Equal Error Rate to Feature Count and Persistence

Oleg Komogortsev, Mehedi Hasan Raju

原文摘要

A biometric designer controls the number of features and their temporal persistence. Earlier empirical work linked the equal error rate (EER) to persistence and to the required number of features by straight lines. We derive the law behind these regularities. For $N$ independent standardized Gaussian features with test-retest correlation $ρ$, scored by cosine similarity, Fisher's $z$-transform turns the genuine and impostor scores into two normals, giving $\mathrm{EER}=Φ(-\frac{1}{2}\sqrt{N}\,\operatorname{atanh}ρ)$ and, inversely, $N=(2q/\operatorname{atanh}ρ)^2$ with $q=Φ^{-1}(1-\mathrm{EER})$; intercorrelated features enter through an effective count $N_{\mathrm{eff}}$. The central extension lets persistence vary across comparisons with spread $τ$ on Fisher's scale: $\mathrm{EER}=Φ(-m/(\sqrt{τ^2+v}+\sqrt{v}))$ with $v=1/(N_{\mathrm{eff}}-2)$, which for a normal spread has a floor $Φ(-m/τ)$ that no number of features can lower. With no fitted parameter the simple law reproduces published simulations and the EER-ICC line, and predicts the KCC-EER relation of eye-movement embeddings on GazeBase ($R^2=0.88$), though less well on GazeBaseVR ($0.63$). Read from the genuine and impostor scores, the general form matches the EER of eye movements, keystroke dynamics, the ECG and face embeddings over 101 conditions with $R^2=0.98$ and a median error of $0.6$ points, against $0.87$ and $1.9$ points for the simple law. The 61 eye-movement, keystroke and ECG conditions check consistency on the same scores; the 40 face conditions use held-out subjects, with the same accuracy. The product $τ\sqrt{N_{\mathrm{eff}}-2}$ gives a design rule: adding or decorrelating features pays fully while it is well below one and little once it exceeds about three, where only higher or more uniform persistence helps.

cs.CV

新论文 · v1arXiv:2611.21812

Sketch, Don't Paint: Allocating Test-Time Compute in Generative Visual Reasoning

Sheng Lu, Guangpin Tao, Tong Lu, Ying Tai

原文摘要

Multimodal models increasingly reason by synthesizing pixels, sketching maze paths within interleaved chains of thought or rolling out video solutions frame by frame. Each visual thought runs the full diffusion schedule inherited from photorealistic synthesis, so iterative denoising dominates inference cost. We study how to allocate test-time compute along two axes: when to render visual thoughts rather than thinking in text or answering directly, and how many denoising steps each thought requires. Controlled interventions that vary only how the thought is rendered uncover three regimes. Causal thoughts carry the plan and become legible early: an 8-step sketch nearly matches a 50-step render in accuracy, commits to the same plan as often as two independent full renders agree, and in video models mostly fails by stalling short of the goal, so the rollout abstains. Decorative thoughts can be omitted without loss, whereas harmful thoughts lower accuracy. Because the regime depends on both model and task, always drawing is suboptimal. Across 14 model-benchmark pairs of three interleaved reasoners, answering directly, without any thinking, wins 9 pairs, and choosing among direct, text-only, and visual tiers from ten labeled examples per pair raises mean accuracy from 60.7% (always drawing) to 65.1%. Where drawing pays, we propose sketch-and-vote, which samples 8-step sketches (a step count chosen without labels) from independent noises until one answer reaches three votes. With this one setting, sketch-and-vote beats full rendering with fewer denoising steps on an image reasoner and two video reasoners, by 8.6 points on VSP spatial planning and by 9.6 and 22.1 points on mazes.

cs.CV

新论文 · v1arXiv:2611.21577

当工具名称并非身份标识:基于MCP的AI智能体中的提供者错误绑定测量

When Tool Names Are Not Identities: Measuring Provider Misbinding in MCP-Based AI Agents

Zeynab Anbiaee, Sajjad Dadkhah, Ali A. Ghorbani

AI 速读 基于摘要

论文研究MCP生态中工具元数据与提供者身份未绑定的问题,即攻击者仅通过注册一个广告特定元数据的MCP服务器,就能让本应发给合法提供者的工具请求被解析到攻击者控制的工具。

cs.CR · cs.AI

查看原文摘要

Large language model (LLM) applications increasingly act as agents that select and invoke external tools. The Model Context Protocol (MCP) standardizes this interaction by allowing clients to compose tools advertised by multiple servers. However, tool metadata used for discovery and selection is not, by itself, necessarily bound to a verified provider identity. This paper studies this gap as tool-provider misbinding: a failure in which a request intended for a legitimate provider is instead resolved to an attacker-controlled tool. The attacker does not compromise the host, LLM, legitimate server, or network channel; it only operates another MCP server that advertises tools with chosen metadata. We formulate tool provider misbinding as an attacker-controlled tool-selection optimization problem and evaluate it in a controlled MCP environment across five LLM-based clients. The results show that interactions drive misbinding. The strongest failures occur when exact-name collision, task-aligned description, and compatible schema reinforce one another, but authority-enhanced descriptions with compatible schema can also cause wrong-provider execution or unresolved provider ambiguity without name collision. Different models expose the same provider-binding weakness differently, including attacker-controlled provider selection, clarification requests, and unstable or inconsistent provider resolution.

新论文 · v1arXiv:2611.21520

Safe Resemblance, Silent Propagation: Limits of Text-Similarity Guardrails for AI Agent Safety

Charles Xiao

原文摘要

When AI agents relay information to one another, an output can preserve an injected instruction's effect on downstream agents even after its wording changes, yet legitimate quotation and template reuse can produce high textual overlap with the source, so text-similarity guardrails that attempt to flag the former risk conflating these two cases. We test this distinction using DonkeyRail, a public classifier combining BLEU, ROUGE-L, and METEOR features, deployed in a controlled three-agent email relay, defining a behavioral measure in which an output counts as functional (the injected procedure induces fresh computations in two subsequent agents) independent of the detector's score, alongside legitimate controls that explicitly authorize quotation and reuse. At its default threshold the classifier misses 8 of 19 functional outputs while flagging 70 of 192 legitimate replies, with every completed quotation and template-reuse reply triggering a false alert; calibrating for low false positives on one template set does not transfer, as the recalibrated threshold detects only one functional output and still produces 15 false alerts on new templates. We also identify an operating-point error in all 21 published classifier configurations, where ROC interpolation replaces direct threshold application, and show formally that high AUC against a surface-level text criterion provides no distribution-free guarantee of high AUC against the behavioral target, while the existing evaluation labels lack the joint observations needed to estimate the gap. An authorization analysis further shows that identical visible transcripts can require different safety decisions depending on the governing policy, a distinction no observation-only classifier can resolve. These results argue for behavioral and policy-grounded evaluation over text-similarity guardrails.

cs.AI

新论文 · v1arXiv:2611.21432

Just sort it! A note on the LP behind rank-one anti-loop edits

Matteo Fischetti

原文摘要

Weight-space programming removes the thinking loops of a small reasoning model by a rank-one edit W' = W + u v^T of an MLP output matrix, where u is a loop direction and v is the solution of a sparse Linear Program (LP) over the neuron activations. We investigate whether the LP is needed. We replace it by a sort: we take the 8 neurons whose activation has the smallest coefficient of variation, and give them equal shares of the push that the LP solution realizes. On Qwen3.5-0.8B the sorted edit scores 67.0% on 1000 problems (MATH-500 and GSM8K) against 67.4% for the LP edit and 9.5% for the unedited model; the paired difference is -0.4 points, with 95% confidence interval [-2.9, +2.1], inside a non-inferiority margin of 5 points declared before the runs. On three further sampling seeds the difference is -0.6 points, interval [-2.0, +0.8], and -0.8 points, interval [-2.3, +0.6], when the dose is set without the LP. The sort does not solve the LP, but the neurons it shares with the LP solution carry most of its mean push. The outcome is that, on this model, the edit acts as a nearly constant bias along u.

cs.LG · cs.CL · math.OC

版本更新 · v2arXiv:2608.03690

An ECG foundation model for zero-shot transfer from hospital to point-of-care devices

Edoardo Coppola, Stefano Fiorini, Pietro Liò, Mattia Savardi, Alberto Signoroni

原文摘要

Electrocardiogram (ECG) recording is moving from hospitals to point-of-care devices that record one or two leads, where public labelled data is too scarce to build foundation models (FMs) as for 12-lead ECGs. Whether existing 12-lead FMs transfer to these devices without retraining is unknown. We show that seven published FMs reach a zero-shot AUROC of 41-81% on two real point-of-care datasets and, when reduced to one or two leads, lose 12-35 AUROC points on average and gain 0.15-0.23 in mean absolute error across 29 benchmarks covering 1,673 clinical targets. We present RELAY, a 7-million-parameter FM that represents ECGs as graphs of segments from whichever leads are recorded, without filling missing ones or knowing their identities. Pretrained on 9.2 million 12-lead ECGs, RELAY is competitive with 12 leads, ranks first on both point-of-care datasets (AUROC 87.6% and 81.5%) and on 26 and 22 of 29 benchmarks with one and two random leads, and keeps the most informative operating point when 12-lead thresholds are applied to device recordings. Matched ablations show that omitting absent leads, rather than filling them, reduces what RELAY loses in transfer. ECG FMs should therefore be judged across all the acquisition settings, not on 12-lead benchmarks alone.

cs.LG

版本更新 · v3arXiv:2611.15592

Small Heads, Strong Decisions: A Controlled Study of Typed-Decision Models

Ritvik Rastogi

原文摘要

In September 2026, TypeSafe announced Jev as the first model in its System One category: unstructured state goes in, and typed, calibrated decisions that software can use directly come out. Jev's architecture and training data are unpublished, but many open typed-decision models have followed, including Nimble, the Decision 1.0-3.0 family and Clef. They answer structured questions about a piece of text, such as a yes/no check, a choice among labelled options or a graded score, by reading the text with a language model and scoring the options with a small readout. Each pairs a different readout with a different backbone, training mix and recipe, so it is unclear which design choices matter. We re-implement five readouts and train all of them under one protocol: the same backbone, the same 15 public English training sources, the same recipe with a per-head learning-rate sweep, and the same seed. Our experiments demonstrate that: The complexity of the head does not affect accuracy: on Qwen3-0.6B every readout reaches 81.3-82.0% in-domain accuracy, and a 0.5M-parameter cosine head, which scores each option by its cosine similarity to the question, matches the 27M-parameter joint schema head adapted from Clef in-domain at half its inference time. Reading all of a text's questions from one prompt also matches one prompt per question in accuracy, with 1.6 times faster inference and 2.1-2.5 times faster training on one H100. The backbone's size matters more than either choice: moving from Qwen3-0.6B to Qwen3-4B adds 4 points in-domain and 11 points on held-out sources, while the base, instruct and thinking checkpoints of the same model tie. We recommend spending effort on the backbone and keeping the head minimal.

cs.AI

新论文 · v1arXiv:2611.21303

Syllable- and Word-Level Evaluation of Tibetan OCR

Jacob Moore, Paula Lauren

原文摘要

Optical character recognition (OCR) for Tibetan is evaluated predominantly by character error rate (CER). Tibetan marks syllable boundaries but does not mark word boundaries, so a word error rate requires a segmenter. This study evaluates nine recognizer configurations on typeset, woodblock and two handwritten script styles, alone and in combination, and reports syllable- and word-level error rates, to our knowledge the first for Tibetan OCR. The configurations span CTC models, TrOCR encoder-decoders with and without post-training quantization, and a vision-language model, and each is scored under CER, the syllable error rate (SER) and three segmented word error rates (SWER). CER ranks systems largely as the higher-level rates do (Kendall $τ\geq 0.833$), with one pair ordered oppositely. On the effect of joint training, CER and SER record significant changes of opposite sign in 7 of 64 comparisons. SER needs no segmenter and ranks systems as SWER does under two of the three segmenters ($τ\geq 0.889$). The output of each configuration is also tagged for part of speech, annotated for named entities and machine translated, and each result is compared with that obtained on the reference text. As rough guidance, translation chrF of 60 is reached at SER of 0.047 or below and part-of-speech preservation of 0.8 at 0.078 or below, while no error rate separates the systems that reach entity preservation of 0.6 from those that do not.

cs.CL

新论文 · v1arXiv:2611.21283

PLECTA: Overlap-aware instance segmentation of dense strand networks from binary masks

Oday Allan, Igor Ostanin

原文摘要

Many materials and tissues, including carbon-nanotube sheets, collagen and paper, comprise overlapping strands whose arrangement influences their effective properties. Measuring this geometry from micrographs requires individual strands to be traced through crossings and signal gaps, which merge distinct strands or split one strand into disconnected fragments. We present PLECTA (from Latin plecta, a braid), a deterministic method that segments individual strands using only a binary image whose foreground pixels mark strand centrelines. At each junction, PLECTA selects which strand segments continue into one another while allowing segments to terminate. It formulates this decision as an exact maximum-weight partial-matching problem, with an explicit penalty for unpaired segments and no restriction on the number of segments entering a junction. The output permits crossing pixels to belong to multiple strands. On synthetic scenes, PLECTA resolved strand identities more accurately than three released alternatives at every density tested. The same ranking held on an independently published generator run at its own settings. A separate stage uses the corresponding greyscale micrograph to estimate which strand lies on top at each crossing. The complete workflow is demonstrated on scanning electron micrographs of carbon nanotube sheets, from raw images to individual strand instances with estimated crossing order, without manual annotations for training or as pipeline inputs.

cs.CV · cond-mat.soft

版本更新 · v4arXiv:2610.05570

Poisson-GENERIC Neural Operators: Exact Metriplectic Structure in Function Space via Casimir Entropies

Jason Sulskis, Sathya Ravi

原文摘要

Existing thermodynamically consistent neural operators enforce GENERIC degeneracy by projecting the reversible operator, introducing state dependence and generally forfeiting the Jacobi identity. We instead use the compatible Lie--Poisson pencil $L=αD+λ(uD+Du)$ for nonlinear transport and constant Poisson multipliers otherwise. Augmenting $u$ with a latent entropy density $s$ makes $S=\int s$ a Casimir, so $L\,δS/δz=0$ without projection. The energy combines a fixed mechanical quadratic, a learned gauge-free potential, and a convex internal energy; the friction operator $M=AA^{\!\top}$ enforces $M\,δE/δz=0$ pointwise and supports diffusion and damping through local Onsager fluxes. Skewness, positivity, and both degeneracies hold algebraically for any parameters; the continuum pencil satisfies Jacobi, with discrete residuals at machine precision on alias-free inputs. Closed-form heat and damped-wave friction operators are exactly representable, and a discrete-gradient integrator preserves the first and second laws. Across four PDEs in one and two dimensions and three backbones (FNO, Transolver, CNO), the model wins $62/72$ seed-level comparisons against unconstrained baselines at equal optimizer steps, recovers exact transport and wave symbols, matches heat and Burgers dissipation within $12\%$, and dissipates nothing on advection. At equal wall-clock time, the unconstrained FNO is more accurate over ten frames. Over one hundred frames, in and out of distribution, our model never increases physical energy and conserves it when dissipation vanishes, whereas the unconstrained FNO loses a quarter of conserved energy out of distribution and projected GENERIC-FNO often creates energy. Limitations include late-time heat under-damping and explicit-integrator divergence on waves.

cs.LG

新论文 · v1arXiv:2611.20841

Vision-Language Model-Based Phase-OTDR Event Classification

Ahmet Bestami Kose, Mehmet Sergen Catal, Muhammet Cagri Yeke, Kivilcim Yuksel, Abdurrahman Gumus

原文摘要

Phase-sensitive optical time-domain reflectometry (Phase-OTDR) enables distributed vibration sensing, but reliable event classification has traditionally relied on fully supervised models that require extensive annotated data. This study examines the adaptability of a Vision-Language Model (VLM) for Phase-OTDR event recognition through few-shot learning. Raw samples are converted into Short-Time Fourier Transform (STFT) dB-magnitude spectrogram mosaics, color-mapped and resized to RGB images, and processed by a Contrastive Language-Image Pre-training (CLIP) image encoder; few-shot adaptation optimizes only the Tip-Adapter-F cache keys, leaving both encoders unchanged. Under a leakage-controlled protocol (train-side support/validation pool of 512 images per class, fixed held-out test of 120 per class, no test-set access during selection, 20 distinct support-sampling trials), the selected configuration (the ViT-L-14 Vision Transformer encoder with OpenAI weights, 128 shots per class) reaches 96.81 +/- 0.42% accuracy and a macro-averaged F1 score of 96.79 +/- 0.42% on the held-out test set; a label-efficient 16-shot setting, that is, 96 labeled support images at adaptation time across the six classes, already attains 90.17 +/- 0.81%. Tip-Adapter-F clearly outperforms both zero-shot CLIP and the standard training-free Tip-Adapter, and prompt and backbone analyses show that performance is stable across Vision Transformer encoders and insensitive to prompt wording at the selected operating point. Overall, CLIP visual features carry a transferable structure that a lightweight support cache turns into a label-efficient Phase-OTDR event classifier. The code and the corresponding image dataset are made publicly available on GitHub to support further research: https://github.com/miralab-ai/vlm-phase-otdr.

cs.CV · cs.LG

版本更新 · v2arXiv:2601.16907

What Does a Cosine of 0.8 Mean? Monotone Calibration of Embedding Similarity

Nicolas Tacheny

原文摘要

Text embedding models turn sentences into vectors, and the cosine between two vectors is the usual measure of their similarity. Search, semantic caching and deduplication often accept a pair when this cosine reaches a fixed threshold. When people judge one pair of sentences more similar than another, the cosine usually agrees, but its value means different things for different models, because each model packs its scores into its own narrow range. The same threshold can therefore keep most pairs with one model and few with another. We propose to calibrate the score instead of changing the model. An increasing function fitted on sentence pairs rated by people translates the cosine of each model into the human rating scale and never reverses the order of two pairs. We fit three such functions for six models, ranging from a simple S-shaped curve to a free step function. On new pairs, calibrated scores agree with human ratings and the models agree with one another. One threshold selects pairs of similar quality whatever the model. The scale obtained belongs to the human ratings used to fit it, so its values shift on two datasets rated differently while the models stay comparable.

cs.LG

新论文 · v1arXiv:2611.20786

ThermoSight: From Appearance to Internal Thermofluid Fields via Physics-Guided Neural Reconstruction

Yuhao Fang, Chenglin Lu, Hefei Ling, Chun Lou

原文摘要

Most neural reconstruction methods focus on reproducing scene appearance, yet inferring physical states from visual observations is essential for machine understanding of the physical world. Reconstructing the internal temperature and soot fields of radiating thermofluids such as flames is challenging yet essential for improving energy efficiency and assessing heat exposure and fire spread. In flames, temperature and soot concentration jointly govern visible emission, and soot self-absorption alters the light that reaches each camera. With limited views, changes in the two fields can offset one another in the recorded RGB images, making temperature and soot difficult to disentangle. ThermoSight, a physics-guided neural framework for scientific imaging, jointly reconstructs continuous three-dimensional temperature and soot volume fraction fields from limited calibrated flame images without field-level ground truth during optimization. A differentiable spectral model links these fields to measured RGB through thermal emission, self-absorption, and calibrated camera responses. Temperature-dependent emission spectra and wavelength-dependent attenuation constrain shared fields across RGB channels, reducing temperature-soot ambiguity. Chromaticity consistency, observation-derived spatial constraints, and soot regularization complement intensity fitting. In a synthetic ablation, wavelength-resolved attenuation reduces temperature RMSE in the hot, sooty region from 234 to 112 K relative to a three-channel approximation, with nearly identical RGB PSNR (65.23 vs. 65.24 dB). Measured data yield coherent fields and competitive RGB reconstructions. These results demonstrate that physics-guided neural fields recover interpretable internal physical quantities beyond visually consistent appearance.

cs.CV · physics.app-ph

新论文 · v1arXiv:2611.20769

Deep Learning for Satellite-Based Mountain Hazard Screening: A Retrospective Failure Analysis on the 2021 Chamoli Rock-Ice Avalanche

Snigdhajyoti Rakshit

原文摘要

The approach utilizes 10 bands from Sentinel-2 imagery, along with terrain information, in a deep learning pipeline to classify potential hazard sources in any given mountainous region. The performance of our model has been evaluated against the data collected during the rock-ice avalanche event in the Chamoli region in the Indian Himalayas on 7th February 2021. U-Net performs segmentation of the detected land surface in the satellite image, detects changes in the identify candidate hazard zones, and then the following steps are performed to calculate the potential energy using the Monte Carlo sampling for steepest descent (D8) flow paths, finding the exposed structures in the downstream locations, and ranking the zones depending on the risk score. In comparison to the simple spectral rule, our approach achieved the higher macro F1 score of 0.648 (as compared to 0.519). But the F1 score in case of rock and soil was low, being in the range of 0.3s. In case of optical change detection in winter, it's only that northeast-facing slope which couldn't be detected, as it was under 73% of shade prior to the incident. The parameters have been set for this particular scene alone and not independently; hence, this is a retrospective test, and this test locates a failure which has occurred earlier. The primary contribution of this paper is locating the pipeline leak.

cs.CV

新论文 · v1arXiv:2611.20767

UNITE: Consolidating Embodied VLM Specialists via Shared-Anchor One-Bit Factorization

Tao Tao, Anke Tang, Enneng Yang, Zhi Hou, Xiaogang Wang, Dacheng Tao

原文摘要

Embodied capabilities such as visual grounding, spatial reasoning, navigation, and driving understanding are often developed in separate vision-language specialists fine-tuned from a common pretrained model.Deploying the specialists separately multiplies storage, whereas merging them into one dense model forces conflicting task updates to share parameters and loses part of each specialist's expertise.We introduce UNITE, a framework that consolidates such specialists into a compact shared-backbone mixture of experts. UNITE keeps the feed-forward block of each specialist as a routable expert and represents the expert bank through a shared-anchor one-bit factorization: a shared floating-point anchor plus expert-specific, row-scaled binary deviations.The anchor is optimized jointly with the scales and signs through closed-form alternating updates, using only source weights and no calibration data.Router adaptation then adjusts token-level expert selection with the backbone and experts frozen, updating fewer than $0.6$M parameters ($<0.01\%$ of the $8.8$B active parameters). Built from four Qwen3-VL-8B specialists, the complete UNITE model occupies $20.3$ GB, $1.16\times$ the storage of one dense specialist and $59.6\%$ below the uncompressed MoE. Across $22$ benchmarks and R2R Val-Unseen, UNITE matches six static merging methods in mean retention of best-specialist performance, leads all of them on Scan2Cap, ScanQA, and all four navigation metrics, and surpasses every source specialist on all five 3D scene understanding benchmarks (up to $+19.1$ on Scan2Cap) and on R2R navigation. Relative to the uncompressed MoE, factorization is nearly lossless on $21$ of $22$ benchmarks; the exception is VSI-Bench, where frame-count experiments suggest impaired use of cross-frame evidence.UNITE thus consolidates complementary embodied capabilities near single-model storage.

cs.LG · cs.CV

新论文 · v1arXiv:2611.20721

CARE: Contrastive-Anchored Residual Encoding with Listwise Reconstruction for Retrieval-Oriented Semantic IDs

Yuan Xu, Dawei Chen, Xiaoguang Lin, Xin Chen, Yulong Ou, Haijun Wu

原文摘要

Generative retrieval replaces nearest-neighbor search with the generation of a document's semantic ID (SID), so the SID tokenizer bounds what the retriever can recall. Common SID constructions (residual k-means, RQ-VAE, finite scalar quantization, capacity balancing) optimize criteria defined on document embeddings alone and do not control whether a query can still find its relevant documents after quantization. We propose two retrieval-oriented SID tokenizers. CARE-SID (Contrastive-Anchored Residual Encoding for Semantic IDs) quantizes in a retrieval-aligned space: a zero-initialized residual alignment network is trained with a query-document contrastive loss before quantization, an anchor loss and a reconstruction target fixed to the original embedding prevent drift and collapse, and reference-vector rating residual quantization assigns the codes. CARE-LR (CARE with Listwise Reconstruction) then distills, for each training query, the similarity distribution that the original embeddings induce over a candidate list into the SID reconstructions, through a function-preserving layer-resolved head. Both keep a document-only encoding interface. We judge SIDs without training a generator, by row-budget recall in a bucketing view (the partition) and a reconstruction view (the SID with its reconstruction). In offline evaluation on industrial video search data, both tokenizers achieve higher reconstruction-view recall than residual k-means, and CARE-LR improves further on CARE-SID, mainly through the listwise distillation. A comparison of two document embeddings shows that contrastive alignment helps most when the embedding has not been trained on search data, and that the embedding affects SID recall far more than the quantizer. CARE-SID is deployed in the generative retrieval stage of an industrial video search system.

cs.IR

版本更新 · v4arXiv:2611.12657

When Does Sliced Inference Pay Off? Resolution, Tiling and Dense Fusion for Small Object Detection

Stefan Becker, Jens Bayer, David Münch, Wolfgang Hübner

原文摘要

Sliced inference such as SAHI is the standard remedy for small objects in high-resolution imagery, but its gains are usually measured against low-resolution baselines. We revisit sliced inference with scale as the explicit variable and ask when tiling actually pays off. We compare nine dense detectors, eight YOLO models of four generations and FCOS, on VisDrone and DOTA, under pre-specified hypotheses, calibration-only tuning and a single evaluation on held-out test data. Two findings emerge. First, much of the benefit attributed to slicing is a resolution effect: on VisDrone, a single pass trained at a higher resolution outperforms tiling with box fusion (merging decoded boxes) for every detector that could be trained at that resolution, and at equal compute a higher resolution helps more than a larger model. Second, tiles carry complementary information that box fusion wastes. We introduce SCAHI (Slicing with Coverage-Aware Hyper Inference), which accumulates the dense outputs of all passes with weights that give every image position the same total coverage, and reads objects off a peak grid fine enough to keep crowded neighbours apart. On identical passes, SCAHI outperforms box fusion for every YOLO detector on both datasets and matches it for FCOS. On DOTA, whose images far exceed the resolution a single pass can be trained at, tiles with SCAHI beat the best single pass for every detector, by 4.3 to 5.6 mAP50 at 2048 px, and replacing only SAHI's merge step by SCAHI adds 2.3 to 4.0 mAP50. An ablation attributes the gain to the border weighting and the resolution of the peak grid. Source code is available at https://github.com/beckerio/scahi.

cs.CV

新论文 · v1arXiv:2611.20484

Agent-Guided Configuration Tuning for SMT Solvers

Cong Wang, Qiaoyi Zheng, Zhaoguo Wang, Fan Long

原文摘要

SMT solvers expose many configuration choices that determine how well their existing algorithms perform on a workload. We present, to our knowledge, the first workflow that uses general-purpose LLM agents to tune SMT solver configurations. Agents read solver documentation, source code, and persistent experiment records to propose option changes, and a deterministic harness measures each proposal, validates the returned models, and selects changes separately for each competition Division. Over roughly nine weeks, our tuning campaign recorded 4,545 proposals for cvc5 across 13 Divisions and 36 logics. We evaluate the tuned configurations on the 2026 and 2025 SMT-COMP ModelValidation benchmarks under a 24-second cutoff. On problems not used for training, they validate more problems than official cvc5 in ten of the 13 Divisions in 2026 and in eight in 2025, with geometric-mean gains of 7.8% and 6.8% across Divisions. On the complete official selections, which include training problems, the tuned floating-point counts exceed those of every individual competition participant in both years. Further studies show how configuration changes make existing algorithms effective by changing access to strategies, scheduling, and option compatibility, and how benchmark families trade off against one another. The agents' experiment traces contain hypothesis-based proposals in every tuning group, and 64.3% of the extracted hypotheses draw on prior measurements. These results show that agent-guided tuning can specialize a mature SMT solver and that accumulated experimental evidence informs sustained search.

cs.AI