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

+549新增论文
+443其中 AI 相关
+62已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-21 12:19 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 17:48 — 2026-11-21 17:48(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:2026-11-21 12:19最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 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。

29,158 已收录论文24,181 AI 相关8,992 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 6,976 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-21 12:17(北京时间)
版本更新 · v2arXiv:2610.23909

When the Editing Intent Is Split: A Cross-Modal Jailbreak Attack for Large Image Editing Models

Xianmin Ye, Zhiyuan Fan, Tianju Liu, Yuhang Wang, Zhenxing Niu, Haichang Gao

原文摘要

Large image editing models increasingly infer editing intent from both textual and visual inputs, creating an underexplored safety risk when harmful intent is distributed across modalities. In this work, we formalize this split-intent vulnerability and propose the Split-Intent Jailbreak Attack (SIJA), which asymmetrically distributes an unsafe editing intent by specifying the editing referent in text while concealing the safety-critical action and specification through graphical visual semantics in an auxiliary cue. To systematically study this threat, we introduce the Split-Intent Image Editing Safety Benchmark (SI2ESBench) for evaluating image editing models under split-intent inputs. Extensive experiments on representative commercial and open-source models demonstrate that SIJA effectively compromises state-of-the-art image editing systems, achieving ASRs of 68.28% on GPT Image 2 and 91.38% on Qwen-Image-3.0-Pro. Our results expose a mismatch between multimodal editing and safety assessment: the distributed intent can be recovered to execute the requested transformation, yet its harmfulness often remains undetected by existing safeguards. To mitigate this vulnerability, we propose editing-intent reconstruction, a defense principle that explicitly evaluates the complete transformation implied by the joint multimodal context and substantially improves rejection of split-intent attacks. Our findings expose a cross-modal safety gap in modern image editing systems and highlight the need to assess the jointly inferred transformation rather than individual inputs in isolation. Project page: https://github.com/yexianmin/SIJA.

cs.CV · cs.AI

版本更新 · v4arXiv:2610.29271

ARCANA: A Benchmark for Abstraction and Analogical Reasoning

Ruqi Chen, Zilei Wang, Wei Wang

原文摘要

Analogical reasoning is widely considered a fundamental component of human intelligence. Few benchmarks are designed to explicitly test far analogical reasoning - reasoning between domains that are superficially dissimilar yet share deep relational or structural similarities. In this work, we introduce an analogical reasoning benchmark called Abstraction and Reasoning Corpus of ANAlogy (ARCANA). The tasks of ARCANA require performing far analogical mappings from real-world objects or physical phenomena to abstract grid-based transformations, different from the well-known ARC-AGI benchmark which primarily emphasized innate human priors. We validate ARCANA through a human study, demonstrating that humans solve these tasks reliably. In contrast, evaluations across a range of large language models show a distinct performance gap compared to humans, revealing systematic failure modes that remain challenging for current models. To address this gap, we propose ConceptSIR, a reasoning framework based on concept slippage, introspection, and reflection. Experiments demonstrate that ConceptSIR enables standard LLMs to solve several challenging ARCANA tasks that were unsolved by direct prompting in our evaluation. Overall, our results suggest that far analogical reasoning remains an open problem for contemporary LLMs. All the materials are available at sites.google.com/view/arcanacogtest. The complete codebase, including benchmark, and ConceptSIR, is publicly available at https://github.com/density-11/arcana-test. All ARCANA data and task instances can be accessed and explored at https://arcanacogtest.org/.

cs.AI

新论文 · v1arXiv:2611.20882

Accelerate Bench: Evaluating Agents for Application-Level Program Optimization

Bowen Liu, Yuhan Li, Nuo Chen, Weile Luo, Miao Peng, Qintong Zhang, Qifan Zhang, Jia Li 等 10 位作者

原文摘要

Application optimization requires coding agents to discover useful changes and coordinate them across a complete program while preserving the required computation. We introduce Accelerate Bench, a benchmark of 32 CPU optimization tasks adapted from scientific applications and supercomputing competitions. Each task provides a working program and specifies the workload, correctness requirements, timing boundaries, and permitted edits, leaving agents to choose where and how to optimize. Public testing supports iterative development, and a separate verifier scores the final submission. On benchmark version 1.1, we evaluate 16 coding agents, each pairing a model with a harness. Their Score@1 ranges from 25.5% to 67.4%. Most of their programs are correct, and the size of the speedup separates them: the two leading agents are at least 2x faster on 22 or more of the 32 tasks, while the other fourteen reach 2x on 8 to 17 and leave 125 of their 412 correct outcomes below 1.25x. Controlled replays show why coordination across an application matters: identical source code runs between 0.29x and 2.12x depending only on build and runtime settings, and one code change is worth 1.01x or 6.4x depending on another. Accelerate Bench pairs executable workloads with agents' programs and trajectories to evaluate both their gains and the process behind them.

cs.AI

新论文 · v1arXiv:2611.20873

Re-Reading Is Not Free: Token Traces and Cache-Aware Energy Estimation for LLM Agents in Software Engineering

Tina Vartziotis, Ketevan Peranidze, Rodopi Kosteli, Ezgi Sarikayak, George Dasoulas, Michael Keckeisen, Sotirios Kotsopoulos, Francesca Dominici

原文摘要

As large language models (LLMs) evolve from conversational assistants into autonomous agents embedded in software development workflows, estimating their environmental footprint requires moving beyond conventional token-based accounting. Existing energy estimators focus on isolated request-response interactions, overlooking the repeated processing of accumulated context, cache reuse, and work delegated to subagents. We analyze nearly one billion tokens collected across 42 sessions and 752 task episodes over ten weeks of real-world AI-assisted software development. Cached-context reads make up 94.8% of all API-reported tokens, while uncached input accounts for only 0.017%; for every generated token, approximately 260 cached tokens are read. Context grows 5.3-fold over a session compared with 1.4-fold within individual tasks. To quantify the energy implications, we extend an existing analytical inference-energy model to account for cached-context processing, yielding a cache-aware hybrid estimator. It estimates approximately 250 kWh for the workload, compared with approximately 10 kWh from the output-based estimator EcoLogits, a 25-fold difference largely driven by cached-context reads. Neither estimator has been validated on long cached contexts, but no modeling assumption we varied closes the gap. Starting each task from a fresh context would lower our estimate by about 60%. We also find that parent-session logs report only a subagent's final internal call, undercounting delegated tokens by a factor of ten; for this workload the correction changes the energy estimate by about 5%, but token-volume studies are fully exposed to it. Our findings expose a substantial gap between how current energy estimators and agentic workloads treat context, highlighting the need for workflow-level energy accounting that captures context accumulation, cache reuse, and delegated execution.

cs.SE · cs.AI

新论文 · v1arXiv:2611.20861

Scenario-Based Risk Certificates for Multi-Property Vulnerability Discovery in Parametric MDPs

Zhiming Chi, Andrea Turrini, Yong Li, Lijun Zhang

原文摘要

Safety-critical systems such as network protocols and distributed algorithms depend on environmental parameters whose exact values are uncertain. Parametric Markov decision processes (pMDPs) model such systems by making transition probabilities symbolic functions of the parameters. For a single quantitative requirement, parametric verification asks which parameter valuations satisfy the requirement. In practice, however, engineers often monitor several safety and performance requirements at once, and one valuation may make several requirements vulnerable simultaneously. In this paper we study multi-property vulnerability discovery under requirement-wise semantics: given a pMDP and M quantitative requirements, we construct candidate regions of the parameter space that may contain violations, and study budgets for random testing under a specified parameter distribution. To solve this, we adopt a two-layered approach in which the scenario-approximation layer provides PAC guarantees on identifying conservative vulnerability regions for each requirement while the testing layer treats random parameter exploration as coverage of overlapping regions and yields a stopping certificate computed from the marginal region masses alone; a composition theorem then bounds the expected non-discovery score by the scenario approximation error plus the testing risk from the two layers. For confirmed discoveries, candidate hits are checked by model checking and budgets use independently certified discovery probabilities. Experiment evaluate these certificates using empirical surrogates and independent confidence bounds.

cs.CR · cs.AI

版本更新 · v2arXiv:2610.11171

VAMR: Multi-Question Agentic Reasoning for Efficient Long-Form Video Understanding

Runquan Gui, Hanzhu Chen, Zehao Wang, Hanxin Zhu, Xin Li, Zhibo Chen

原文摘要

Long-form video understanding often involves multiple questions about different aspects of the same recording. Yet existing video agents typically process each question through an isolated tool-use trajectory. This repeatedly restarts video exploration and memory construction, missing opportunities to acquire evidence jointly and progressively build a shared understanding that supports the complete question set. We introduce \textbf{VAMR} (\textbf{V}ideo \textbf{A}gent for \textbf{M}ulti-Question \textbf{R}easoning), which coordinates all questions about a video through one shared tool-use trajectory. At each round, a persistent policy model can invoke tools for one or more unresolved questions and submit answers for questions with sufficient evidence. Question-conditioned visual perception retrieves fine-grained clues for several questions in one call, while layered multi-question memory integrates reusable context into a shared video story and preserves separate evidence for individual questions. After supervised fine-tuning initializes this interaction protocol, we propose question-horizon policy optimization (\qhpo) to optimize shared trajectories in which questions progress and finish at different rounds. Specifically, a question-level critic estimates the value of each active question, while round alignment maps each question advantage to the rounds that directly serve it before the aligned advantages are aggregated to optimize the shared actor. Across LVBench, Video-Holmes, and LongVideoBench, VAMR achieves the highest accuracy overall and the fewest reasoning rounds among iterative methods. On LVBench, it reaches 62.1\% accuracy, exceeding VideoARM by \textbf{4.3} points while reducing reasoning rounds and processed frames by \textbf{85.9\%} and \textbf{61.4\%}.

cs.CV · cs.AI

新论文 · v1arXiv:2611.20848

MolReps: Taking SELFIES, Making SMILES, Teaching LLMs

Ken Deng, Garrett Morris

原文摘要

The same molecule can be expressed in multiple, syntactically distinct representations, yet it remains unclear whether large language models (LLMs) can recognize and relate these equivalent encodings. We introduce MolReps, a pipeline that generates 9,600 question-answer pairs across ten representations for conversion, normalization, and 2D structure recognition. Post-training on MolReps improves text-only and multimodal models, MolQwen3-4B-Instruct (MQ3-4B-I) and MolQwen3-VL-4B-Instruct (MQ3-VL-4B-I), from $4.01\%$ to $75.13\%$ and from $7.11\%$ to $73.14\%$ (InChIKey match) on the molecule-overlap test set, respectively. On the molecule-disjoint test set, post-training increased validity from $20.35\%$ to $58.12\%$ for MQ3-4B-I and from $17.19\%$ to $59.34\%$ for MQ3-VL-4B-I. However, InChIKey match remained low, decreasing from $3.86\%$ to $3.09\%$ and from $6.44\%$ to $2.29\%$, respectively. Furthermore, the resulting models achieve a relative $27.84\%$ improvement in average over their base model on MolRepBench, a comprehensive benchmark for evaluating molecular property prediction. These results show the effectiveness of targeted post-training for enhancing molecular representation understanding and its potential to improve downstream molecular tasks.

physics.chem-ph · cs.AI · cs.CL

新论文 · v1arXiv:2611.20844

DetourBench: Trajectory-Level Evaluation of Interactive Agents with Exact Minimum-Action Bounds

Hanlin Tian, Sihan Zhu, Zhao Yang, Yuxiang Wang, Zhiquan Hu, Yu Mi, Hongquan Zhu, Qiufei Hu

原文摘要

Benchmarks judge language-model agents mainly by whether they reach the correct final state, not by how they got there. Endpoint scoring cannot separate an agent that retrieved exactly the evidence a decision required from one that read most of the environment, nor an agent that never located the evidence from one that stopped a single action short. Settling this needs a reference cost, and authored solutions establish only that a task is solvable, not that no shorter route exists. We present DetourBench, which constrains the environment until that minimum is computable. Each task is a finite, partially observable transition system with unit-cost accesses whose resources become globally addressable after discovery. Exhaustive search yields the full-information minimum action count $L_0$ with a replayable witness; from graph-recoverable terminal states it also yields exact residual work $R_T$, which separates an episode that stopped one access short from one that wandered, while an irreversible wrong commitment is recorded as having no finite residual distance rather than assigned one. Success, attainment, residual progress, and normalized detour are reported separately. Across 375 scored episodes --- five model endpoints, 25 integrated tasks, three repeats --- success ranges from 58.7% to 96.0% while conditional attainment ranges from 0.803 to 0.890 and orders the models differently. Among recoverable failures, residual work separates near-complete trajectories from costly incompletion and from irreversible commitment. We open-source the instances, oracle witnesses, and evaluation harness at https://github.com/DatagridsAI/DetourBench.

cs.AI · cs.CL

新论文 · v1arXiv:2611.20837

Evidence Diet: When Graph Evidence Helps LLMs Rank Competing Candidates

Yuchong Hu, Zhao Zhang, Ziyao Zhao, Haoyu Wang, Jing Xiao

原文摘要

When several candidates have graph support, does it matter which candidate receives which supporter and path counts? We study how LLM rankers use collaborative evidence on Amazon Books and MovieTV. Swapping user-level evidence degrades ranking, while matched item substitutions have little effect. On fixed clean-trained checkpoints, unrelated catalog text lowers Hit@1 from 44.77% to 38.70%, with 99.9% of rankings still valid. Across all three paired seeds, the NDCG@5 difference between count and path prompts was larger after count training than after path training. Degree matching preserves target paths while giving more rivals support. Across the full degree-matched development pools, adding candidate-level graph records to history improves NDCG@5 by 2.28 points on Books and 2.54 on MovieTV. When the target and a rival both have support, path- and count-trained rankers benefit from retaining counts. On these development queries, adding counts numerically to history scores outperforms presenting them as text. For count-trained rankers, exchanging intact count pairs among supported candidates preserves support and token length, yet lowers NDCG@5 by 4.06 points on 141 Books queries and 3.26 on 104 MovieTV queries. A frozen evaluation on 274 previously unused Books users with supported targets and rivals confirms the attribution effect: correct assignment improves pairwise accuracy by 6.07 percentage points (95% CI [4.18, 8.00]).

cs.IR · cs.AI

新论文 · v1arXiv:2611.20826

Conversing with a Conversational Agent: A Real Conversation, Not a Simulated One

Christophe Denis

原文摘要

The claim that interactions with artificial conversational agents merely simulate human conversation rests on a philosophical distinction between reproducing the observable features of an activity and actually performing it. Although this distinction is appropriate for certain physical phenomena, its application to linguistic and symbolic activities is far less straightforward. This article argues that conversations with large language model (LLM)-based agents are genuine conversational activities rather than mere simulations, without thereby attributing consciousness, subjective understanding, or human intentionality to artificial systems. Drawing on the distinction between subjective and objective simulation, the article first examines the limitations of defining conversation exclusively through the intrinsic properties of its participants. It then develops a relational and semiotic account of conversational activity, informed by linguistic pragmatics, Lacan's Conversing with a Conversational Agent theory of the mirror, Baudrillard's concept of the simulacrum, Simondon's philosophy of individuation, and Peirce's semiotics. These perspectives make it possible to distinguish the reality of conversational interaction from the appearance of artificial subjectivity and from the epistemic validity of the meanings produced. The argument culminates in the concept of Hybrid Semiotic Fabrication, which characterizes meaning-making as a distributed process involving human participants, artificial intelligence, corpora, and technical arrangements. Recognizing the reality of human-AI conversation therefore entails neither anthropomorphizing artificial agents nor granting their outputs epistemic authority. Rather, it requires a more precise account of the relational, semiotic, and normative conditions under which conversational meaning is produced and justified.

cs.AI

新论文 · v1arXiv:2611.20819

Surprise-Gated Memory for Live Speaker Diarization

Louis Mouchon, Alexis Lainault

原文摘要

We transpose the Surprise-Gated Memory recipe, a frozen encoder plus a calibrated novelty signal that gates a one-shot episodic store consolidated by periodic replay, from continual visual learning to live speaker diarization. The moment-to-moment question a streaming diarizer must answer, whether an arriving voice belongs to an already-heard speaker or a new one, is a known-versus-novel decision, exactly the decision the surprise gate is built for. On a frozen fine-tuned DiariZen backbone, our online system SGM-Diar recovers 4.8 of the 9.2 diarization-error points separating a naive online baseline from the offline oracle on AMI-SDM (52%) and 1.9 of 2.3 points on VoxConverse (82%), in the end-of-meeting revisable mode, with calibration frozen on development data and transferred cross-domain without adjustment. These are single-seed runs on three AMI meetings and five VoxConverse sessions, so the ordering of conditions rather than the absolute rates is the result. Beyond error rate, we measure the gate's behaviour directly and find a coherent memory-like profile: one-shot speaker recall at 0.857, a novelty response that habituates as speakers become familiar, re-identification that is invariant to the length of intervening silence, and a false-novelty rate of 0.001. Two ablations reproduce, now on audio, the central prediction of the visual SGM work: a sliding replay window is worse than no replay at all. The memory is strictly intra-meeting and is wiped at the end, and no gradient touches the run-time system. We report SGM-Diar as a proof-of-concept and state its limitations explicitly.

cs.SD · cs.AI

新论文 · v1arXiv:2611.20815

ForeSay: Tell the Future in Words, Branch the Details in Noise for Precipitation Nowcasting

Xudong Ling, Wenjie Luo, Chaorong Li, Tianxi Huang, Tao He, Guiduo Duan

原文摘要

Precipitation nowcasting involves two distinct scales of uncertainty: macroscopic storm evolution and fine-scale local convective variations. Current generative ensemble methods entangle both scales within independent random noise draws. This coupling causes two practical limitations: every ensemble member must execute the full sampling trajectory from scratch, so an N-member ensemble costs N times a single forecast; moreover, divergent trajectories spatially misalign heavy-rain echoes across members, causing standard pixel-wise averaging to blur critical high-intensity extremes. We present ForeSay, a generative nowcasting framework that decouples macroscopic evolution prediction from local stochastic detail generation. Given historical radar frames, a vision-language model first predicts a textual scenario of future storm dynamics to condition a Rectified Flow generator. Guided by this shared macroscopic prior, ensemble members share early trunk sampling steps to establish the storm trajectory and branch with fresh noise only in later steps to explore local convective details. Because members then differ mainly in local details, self-anchored support-consensus decoding replaces pixel averaging with discrete support voting to preserve sharp heavy-rain cores. Across six datasets, ForeSay achieves the best CSI at the highest precipitation thresholds, with relative improvements over the strongest baseline on each dataset ranging from over 10% to 84%. Our 8-member ensemble retains the forecasting skill and diversity of eight independently sampled runs while reducing computational cost from 8x to 5.2x that of a single forecast.

cs.AI

新论文 · v1arXiv:2611.20812

The Cybersecurity Dataset Observatory: Which Data for AI in Cybersecurity?

Francesco Cerasuolo, Gabriele Mangiacapre, Antonio Pescapè

原文摘要

AI is increasingly used for cybersecurity tasks such as intrusion and malware detection, in enterprise, industrial, vehicular and mobile networks. Its effectiveness depends on data: models need representative network traffic, both to be trained and to be evaluated realistically. Public datasets of cybersecurity network traffic, however, are scattered across laboratory pages, repositories and aggregators, and are continuously released, corrected and withdrawn: 92 of the 234 datasets we catalog were released in 2023 or later. At this pace, surveys that list them are outdated soon after publication, since they describe the field only as it was when they were written. We present the Cybersecurity dataset observatory, a public and continuously updated catalog of cybersecurity datasets based on network traffic. New datasets are added as they are found or submitted, every record is periodically re-checked against its original source, to which it links rather than hosting the data, and every change is recorded in a public change log. This paper documents the first release: we define a traffic-centered scope and a seven-axis taxonomy, and review 234 datasets (204 original captures and 30 derivatives) across 14 domains. The review shows a field that grows quickly but unevenly, where 91 datasets distribute only processed records, a few laboratories produce a large share of the data, and UAV, wireless and V2X networks have little public evidence, while 5G has few real captures. We then relate the catalog to common uses in AI research, outline a procedure for selecting datasets, and discuss benchmark pitfalls and open gaps.

cs.CR · cs.AI

版本更新 · v2arXiv:2608.00005

RubricReviewer: From Direct Critique to Rubric-Driven Review via Agentic Evidence Synthesis and Human-Aligned Generation

Shuyu Guo, Wenxiang Hu, Yuyue Zhao, Lun Zhou, Yougang Lyu, Xiaohui Yan, Zhaochun Ren

原文摘要

Peer review at major venues is under unprecedented submission pressure, motivating the use of large language models (LLMs) as review assistants. Existing LLM-based reviewers, however, face two structural limitations. First, they map manuscripts directly to reviews, leaving the underlying rubric implicit and entangling its derivation with the judgement. Second, the prevailing paradigms each capture only half of a good review: training-free agents gather broad evidence but produce undirected critiques, while training-based reviewers inherit human discriminative judgement together with its noise and uneven coverage. We introduce RubricReviewer, a fully rubric-driven framework that addresses both limitations. It makes rubric generation an explicit intermediate step, so that both review generation and the final assessment are conditioned on paper-adaptive rubrics. It further combines a training-free agent (Scout) that gathers external evidence with a human-aligned trained model (Aligner) that consumes this evidence, fusing the strengths of both supervision sources. Experiments on real-world submissions show that RubricReviewer produces reviews that are markedly more comprehensive and more discriminative than prior systems, and exhibits the strongest robustness against adversarial prompt-injection attacks. Ablation studies further confirm the necessity of each component.

cs.CL · cs.AI

版本更新 · v2arXiv:2610.41398

Distilling Knowledge from Search: Decoder Model Factory for Quantum LDPC Codes

Anil Bilgin, Wanda Hou, Pranav Srinivas Murali, Yi-Zhuang You

原文摘要

We introduce GraphMP, a learning-based message-passing decoder for quantum low-density parity-check codes that transfers without fine-tuning across code families, code sizes, physical error rates, and noise models. Its local message passing keeps computation scalable as the decoding graph grows. We further embed GraphMP in a beam-search algorithm that progressively refines candidate error patterns, trading additional decoding time for lower logical error rates. To turn this decoding effort into lasting improvements, we propose the decoder model factory, a paradigm for training successive models on data collected by earlier models during decoding. In our implementation, verified search trajectories provide reusable training data for distilling knowledge into subsequent models. Across generations, the model factory yields a search-enhanced decoder that is broadly competitive with Relay-BP. On the largest code block, which encodes $4612$ logical qubits, it achieves a lower logical error rate and about $20$ times the decoding throughput of the standard Relay-BP baseline. These results motivate a coupled quantum-classical architecture with two interacting cycles: rapid decoding protects quantum information, while slower learning adapts the models to changing noise conditions. The long-term vision is a scalable decoder model factory that continually produces high-performance, low-latency decoders for evolving quantum hardware.

quant-ph · cond-mat.dis-nn · cs.AI

版本更新 · v3arXiv:2503.08823

ResBench: Benchmarking LLM-Generated FPGA Designs with Resource Awareness

Ce Guo, Tong Zhao

原文摘要

Field-Programmable Gate Arrays (FPGAs) are widely used in modern hardware design, yet writing Hardware Description Language (HDL) code for FPGA implementation remains a complex and time-consuming task. Large Language Models (LLMs) have emerged as a promising tool for HDL generation, but existing benchmarks for LLM-based code generation primarily focus on functional correctness while overlooking hardware resource usage. Furthermore, current benchmarks offer limited diversity and do not fully represent the wide range of real-world FPGA applications. To address these shortcomings, we introduce ResBench, the first resource-focused benchmark explicitly designed to distinguish between resource-optimized and inefficient LLM-generated HDL code. ResBench consists of 56 problems across 12 categories, covering applications from finite state machines to financial computing. Our open-source evaluation framework automatically tests LLMs by generating Verilog code, verifying correctness, and measuring resource usage. The experiments, which primarily analyze Lookup Table (LUT) usage, reveal significant differences among LLMs, demonstrating ResBench's capability to identify models that generate more resource-optimized FPGA designs.

cs.AR · cs.AI · cs.CL · cs.ET · cs.LG

版本更新 · v2arXiv:2605.13534

Scaling Retrieval-Augmented Reasoning with Parallel Search and Explicit Merging

Jiabei Liu, Wenyu Mao, Junfei Tan, Chunxu Shen, Lingling Yi, Jiancan Wu, Xiang Wang

原文摘要

Deep search agents have proven effective in enhancing LLMs by retrieving external knowledge during multi-step reasoning. However, existing methods often generate a single query for retrieval at each reasoning step, limiting information coverage and introducing high noise. This may result in low signal-to-noise ratios (SNR) during search, degrading reasoning accuracy and leading to unnecessary reasoning steps. In this paper, we introduce MultiSearch, an RL-based framework that addresses these limitations through multi-query retrieval and explicit merging of retrieved information. At each reasoning step, MultiSearch generates queries from multiple perspectives and retrieves external information in parallel, expanding the scope of relevant information and mitigating the reliance on any single retrieval result. Then, the agent consolidates and refines retrieved information at the merging process, improving the SNR and ensuring more accurate reasoning. Additionally, we propose a reinforcement learning framework with a multi-process reward design to optimize agents for both multi-query retrieval and information consolidation. Extensive experiments on seven benchmarks demonstrate that MultiSearch outperforms baseline methods, enhancing the SNR of retrieval and improving reasoning performance in question-answering tasks.

cs.AI

版本更新 · v2arXiv:2607.21186

On the effects of emulated quantum layers on CNN performance and explainability for OCT and dementia imaging

Guillermo Rubiños Rodríguez, Martín Ottavianelli, Mateo Alonso, Gonzalo Blázquez Gil, Boris-Stephan Rauchmann, Pablo Díez-Valle, Sergio Altares-López

原文摘要

Hybrid quantum-classical convolutional neural networks have been proposed as an alternative to classical deep learning models. However, executing network components on quantum hardware imposes fundamental limitations, such as gradient estimation through repeated measurements, and increasing the number of qubits can lead to trainability problems such as barren plateaus. Moreover, the reported success of these hybrid networks has been linked to quantum architectures that can be efficiently simulated on classical computers. We investigate whether small, classically simulated quantum circuits can serve as useful components within larger models, as a substitute for purely classical layers. We present a systematic comparison between a Hybrid Quantum-inspired CNN (HQiCNN) and standard CNN architectures, where the models differ only in a single intermediate layer: an emulated quantum circuit in the HQiCNN and classical counterparts in the CNN. The models are evaluated on two medical imaging datasets while varying training set size, learning rate, batch size, circuit entanglement and observable se. Under a strict parameter budget, the HQiCNN outperforms the classical baselines. When the classical layer is given more parameters, overall performance becomes comparable, with the HQiCNN showing improvements only in the intermediate-data regime. Removing entanglement yields comparable performance and richer observable sets yield no performance improvement. Finally, we introduce two SHAP-based metrics for comparing model explanations, $|$SHAP$|$IoU and EMD$_{\mathrm{pos}}$. They show that all architectures attend to anatomically plausible regions, and that the spatial attention agreement between them increases with training set size. These results indicate that, under specific data regimes, quantum-inspired hybrid models are a competitive alternative to classical CNNs for medical image classification.

quant-ph · cs.AI · cs.LG

新论文 · v1arXiv:2611.20779

EEGProbe: Do Vision-Language Models Really Understand EEG? A Benchmark for Signal-Grounded EEG Signals Evaluation

Haozhe Zhu, Dan Yuan, Chenghao Gu, Feike Xu, Yue Wen, Shin'ichi Konomi

原文摘要

Vision-language models (VLMs) have shown strong multimodal capabilities across scientific and biomedical domains, yet their ability to understand electroencephalography (EEG) signals remains unclear. We introduce EEGProbe, a signal-grounded benchmark for evaluating whether general-purpose VLMs can extract and reason from minimally processed, channel-wise EEG without task-specific feature engineering or EEG-specific adaptation. EEGProbe contains 620 EEG-centered questions across 13 subtasks and three capability categories: EEG Signal Classification, EEG Cross-modal Alignment, and EEG Neurophysiological Reasoning. To reduce shortcut-based answering, EEGProbe preserves the temporal structure of the recordings where possible, minimizes benchmark-induced processing, and applies signal-necessity testing followed by expert verification. We evaluate 11 representative open- and closed-source VLMs under a unified zero-shot protocol. The strongest model, GPT-5.6 Sol, achieves an overall score of only 41.63\%, with cross-modal EEG alignment emerging as a major bottleneck. We further examine model behavior from a neurophysiological perspective and analyze how temporal, spectral, spatial, and physiological EEG structure relates to model successes and failures. Inference-time neurophysiological guidance can reshape model predictions by helping models recognize the neurophysiological patterns present in the input and connect them with relevant neurological knowledge, narrowing the gap between general multimodal competence and EEG understanding. EEGProbe provides a testbed for current VLMs and a basis for developing future multimodal models that can interpret EEG evidence more reliably. The EEGProbe benchmark will be publicly available at https://huggingface.co/datasets/2122ZHU/EEGProbe.

cs.AI · cs.CV · cs.HC

新论文 · v1arXiv:2611.20771

Nemotron 3.5 Bolt E4B 4GB: An Open and Efficient Small Language Model for Fast Conversational Agents

NVIDIA, :, Adi Margolin, Aditya Malte, Aishwarya Chhabra, Alex Hodson, Ameya Sunil Mahabaleshwarkar, Ambrish Dantrey 等 113 位作者

原文摘要

We introduce Nemotron 3.5 Bolt E4B 4GB, an open and efficient small language model further post-trained from Gemma-4-E4B-it and optimized for speed and intelligence per GB of memory used for fast conversational agents. The model carries forward Gemma's strong conversational capabilities and further improves multi-turn tool use, instruction following, and input grounding for fast, local voice/device assistants, interactive personas, and embedded/Edge interfaces. The model combines natural conversation with multi-turn agentic capabilities to power voice/device assistants, interactive characters and robotic/IoT interfaces running on GeForce RTX, Jetson, and RTX/DGX Spark. Experiments on 21 benchmarks show that Nemotron 3.5 Bolt E4B 4GB achieves the highest average accuracy while generating the fewest output tokens in both reasoning and non-reasoning modes, compared with Gemma-4-E4B-it and Qwen3.5-4B. The model is released with a DFlash drafter. With DFlash, Bolt E4B achieves $2.74\times$ the per-user decode speed of inference without speculative decoding at concurrency~1, and $1.21\times$ the generation throughput of Gemma-4-E4B-it with MTP at batch size 1.

cs.CL · cs.AI · cs.LG

版本更新 · v2arXiv:2609.11900

MindTopo: Can Foundation Models Reason in Topological Space?

Yunfei Ge, Anbang Liu, Qineng Wang, Johnalbert Garnica, Jianwen Lyu, Zihan Wang, Reuben Tan, Jianfeng Gao 等 12 位作者

原文摘要

Spatial reasoning depends not only on metric properties such as distance, angle, and shape, but also on topological relations that remain invariant under continuous deformation. Cognitive science identifies these relations as foundational to spatial understanding, yet foundation-model evaluations largely focus on metric or viewpoint-dependent relations. We introduce MindTopo, a benchmark of topological intuition across five properties grounded in cognitive science and formal topology, namely continuity, separation, order, enclosure, and knots. MindTopo evaluates each property at two cognitive levels. Reasoning asks a model to identify topological relations or infer how they change. Planning instantiates a foundation model as a closed-loop agent whose policy selects environment actions. MindTopo provides a Core Set of 3,900 instances across 13 procedurally generated task types, split into an Easy and a Hard level, and an Extended Set of 11,035 original instances. On the Core Set, we benchmark 13 multimodal large language models (MLLMs) and study agent configurations augmented with image and video generation, covering seven video generative models in planning settings. Every MLLM performs better on reasoning than on planning. On the Easy level, where human scores are available, the strongest model trails humans in both reasoning and planning. On Qwen3-VL-2B-Instruct, supervised fine-tuning and reinforcement learning improve reasoning more than planning. Generated observations retain local cues and reach solved endpoints, yet none of the 1,890 evaluated video rollouts passes all process-validity checks.

cs.AI · cs.CL · cs.CV

版本更新 · v3arXiv:2410.07158

quanda: An Interpretability Toolkit for Training Data Attribution Evaluation

Dilyara Bareeva, Galip Ümit Yolcu, Anna Hedström, Niklas Schmolenski, Thomas Wiegand, Wojciech Samek, Sebastian Lapuschkin

原文摘要

Training data attribution (TDA) methods estimate the influence of individual training samples on a model's predictions, offering a principled lens for interpreting neural networks. Despite rapid methodological progress, TDA evaluation remains scattered across papers and rarely reproducible, making fair comparison between methods challenging. We introduce quanda, a Python toolkit that standardizes TDA evaluation. quanda features a comprehensive collection of evaluation metrics and ready-to-use benchmarks spanning image classification, text classification, and language modeling, alongside a uniform interface for integrating diverse TDA implementations. We showcase quanda through a comparative study of existing TDA methods and find that no approach excels uniformly across metrics, with clear trade-offs between ground-truth faithfulness and downstream task performance. By consolidating the tooling around TDA, quanda equips the community to evaluate attribution methods rigorously and reproducibly. The toolkit is thoroughly tested, documented, and available as an open-source library on PyPI and at https://github.com/dilyabareeva/quanda.

cs.LG · cs.AI

版本更新 · v2arXiv:2512.05239

A Survey of Bugs in AI-Generated Code

Ruofan Gao, Amjed Tahir, Peng Liang, Teo Susnjak, Ali Soltanian Fard Jahromi, Foutse Khomh

原文摘要

AI coding systems are increasingly used to help developers write code at scale, improving software development efficiency and boosting productivity. While these models and agents have shown strong performance, concerns persist regarding the quality of the AI-generated code. Although prior work has reported a range of quality issues associated with AI-generated code, these findings remain scattered and lack a systematic synthesis. A comprehensive review is lacking that demonstrates the bug types, patterns, distribution, their correlation with specific AI coding systems, and the remediation strategies proposed to address them. To fill this gap, we conduct a systematic literature review of existing studies on AI-generated code to develop a comprehensive understanding of bugs in the generated code and provide a foundation for future improvements in AI-assisted coding. Our results show that bugs are widespread in AI-generated code, with functional and stability bugs emerging as the most frequently reported categories across the reviewed studies, a substantial proportion of which is attributed to model hallucination. Model size, instruction-following capability, and the quality and diversity of training data are key factors influencing bug occurrence. To mitigate bugs, prompt engineering techniques are widely used; however, certain domain-specific or semantically complex issues remain difficult to resolve automatically and still require human intervention.

cs.SE · cs.AI

版本更新 · v4arXiv:2602.10085

CODE-SHARP: Continuous Open-ended Discovery and Evolution of Skills as Hierarchical Reward Programs

Richard Bornemann, Pierluigi Vito Amadori, Antoine Cully

原文摘要

A core quality of general intelligence is the ability to open-endedly expand and evolve its set of mastered skills autonomously. While recent Foundation Model (FM) driven approaches have shown promising results towards this goal, they typically rely on significant human-in-the-loop engineering, limiting their transferability to novel environments. To address this, we introduce Continuous Open-ended Discovery and Evolution of Skills as Hierarchical Reward Programs (CODE-SHARP), a framework that leverages FMs to open-endedly grow and evolve an archive of Python programs encoding skills to train a generalist agent policy entirely from scratch via reinforcement learning, directly from source code. These programs, termed Skills as Hierarchical Reward Programs (SHARPs), each encode a local success condition and a set of prerequisites delegated to previously discovered SHARPs. At runtime, SHARPs dynamically route the agent through their prerequisite chain based on the current state, rewarding each completion along the way, requiring the agent to learn only the marginal behaviour each new SHARP introduces, enabling efficient learning of long-horizon skills without any pre-defined rewards. On Craftax-Classic and XLand-MiniGrid, agents trained fully autonomously by CODE-SHARP outperform previous works by 6x and 2.7x in median performance and are the only agents capable of crafting iron tools and mining diamonds. On AI2-THOR, CODE-SHARP achieves 66.7% median success versus 0% for ELLM and OMNI, using only the initial environment state instead of the full source code. Scaled to Craftax-Extended, CODE-SHARP trains a generalist agent on over 90 discovered SHARPs, enabling the agent to solve challenging long-horizon tasks zero-shot, matching agents trained on ground-truth rewards.

cs.AI

版本更新 · v3arXiv:2603.05963

Skeleton-to-Image Encoding: Enabling Skeleton Representation Learning via Vision-Pretrained Models

Siyuan Yang, Jun Liu, Hao Cheng, Chong Wang, Shijian Lu, Hedvig Kjellstrom, Weisi Lin, Alex C. Kot

原文摘要

Recent advances in large-scale pretrained vision models have demonstrated impressive capabilities across a wide range of downstream tasks, including cross-modal and multi-modal scenarios. However, their direct application to 3D human skeleton data remains challenging due to fundamental differences in data format. Moreover, the scarcity of large-scale skeleton datasets and the need to incorporate skeleton data into multi-modal action recognition without introducing additional model branches present significant research opportunities. To address these challenges, we introduce Skeleton-to-Image Encoding (S2I), a novel representation that transforms skeleton sequences into image-like data by partitioning and arranging joints based on body-part semantics and resizing to standardized image dimensions. This encoding enables, for the first time, the use of powerful vision-pretrained models for self-supervised skeleton representation learning, effectively transferring rich visual-domain knowledge to skeleton analysis. While existing skeleton methods often design models tailored to specific, homogeneous skeleton formats, they overlook the structural heterogeneity that naturally arises from diverse data sources. In contrast, our S2I representation offers a unified image-like format that naturally accommodates heterogeneous skeleton data. Extensive experiments on NTU-60, NTU-120, and PKU-MMD demonstrate the effectiveness and generalizability of our method for self-supervised skeleton representation learning, including under challenging cross-format evaluation settings.

cs.CV · cs.AI

版本更新 · v2arXiv:2501.13946

Hallucination Mitigation using Agentic AI Natural Language-Based Frameworks

Diego Gosmar, Deborah A. Dahl

原文摘要

Hallucinations remain a significant challenge for Generative AI, undermining trust in AI systems and their reliability. This study investigates how orchestrating multiple specialized AI agents can mitigate hallucinations, with a focus on systems that use Natural Language Processing (NLP) for agent interaction. We design a pipeline that feeds 310 prompts, purposefully crafted to induce hallucinations, into a front-end agent; its outputs are reviewed and refined by second- and third-level agents that detect unverified claims, add explicit disclaimers and clarify speculative content. We introduce four novel Key Performance Indicators (KPIs) - Factual Claim Density, Factual Grounding References, Fictional Disclaimer Frequency and Explicit Contextualization Score - aggregated into a Total Hallucination Score (THS) and assessed by a dedicated fourth-level evaluator agent. Inter-agent context is carried by the Open Floor Protocol (OFP, formerly OVON, Open Voice Network), whose structured JSON messages let each reviewer pass its assessment of hallucination likelihood, and the reasons behind it, to the next stage. Across the 310 prompts, THS improves by a mean of -0.131 per prompt from the front-end agent to the final reviewer (Cohen's d = 2.38), with lower scores on 302 of 310 prompts. The gain is carried mainly by explicit disclaimers and contextualization, making speculative content more transparent and clearly demarcated from factual claims. The results show that multi-agent orchestration with structured, NLP-based exchange of meta-information can substantially strengthen the epistemic framing of AI-generated responses and improve their explainability.

cs.CL · cs.AI · cs.MA

版本更新 · v2arXiv:2611.17442

Racing the Clock: Teaching VLNs Budget-Conditioned Object Search

Maitree Meher, Siming He, S. Shankar Sastry, Jitendra Malik, Claire Tomlin, Jonas Frey

原文摘要

How do action budgets shape vision-language navigation (VLN) exploration, and what expert supervision best teaches budget-aware search? We study budget-conditioned object search: finding as many distinct instances of a requested category as possible within a budget, with discovery defined geometrically by camera field of view and proximity. Distinct discoveries provide a denser evaluation signal than single-target success. We cast expert design as a choice of scene abstraction: a single budget-indexed Bellman recursion, applied to different abstractions of the full environment, yields experts that differ in what they know. An object-level Bellman expert is given every target location and reduces search to budgeted routing. A room-level Bellman expert retains observed room connectivity, doorway evidence, and cross-scene semantic priors but withholds the contents of unexplored rooms, so it must search for targets that its student also cannot see. We distill both experts into Uni-NaVid under matched training conditions and also evaluate greedy-frontier supervision on held-out HM3D scenes. Changing only the stated budget changes learned trajectories. Despite higher object-level expert return, room-level supervision achieves the highest mean collection among evaluated VLNs at every tested budget. At 2000 steps, it finds 72.7% of reachable targets, versus 68.5% under greedy-frontier supervision, 65.5% under object-level supervision, and 44.8% for the pretrained VLN. These results show that expert-return rankings need not predict supervision quality for VLN exploration, and motivate choosing an expert's abstraction with the student's information in mind rather than by expert return alone.

cs.RO · cs.AI · cs.CV · cs.LG

新论文 · v1arXiv:2611.20736

GainCtrl: Predicting Remaining Gains for Efficient LLM Agent Execution

Kangwen Zhao, Jianfeng Cai, Ruopei Sun, Mingxiao Feng, Peng Zhang, Wengang Zhou, Houqiang Li

原文摘要

Large language models (LLMs) are widely used to power agents that interact with tools and environments to solve complex tasks. Repeated attempts improve the quality of tool-using agents, but continued execution incurs additional costs even when it is unlikely to improve the best result ultimately returned. This raises the question of when to stop an attempt to reduce execution cost while preserving task-level Best-of-N quality. To address this question, we introduce GainCtrl, a gain-based execution controller for LLMs. In GainCtrl, a lightweight CPU predictor uses interaction histories to estimate expected positive gains over retained scores, which are then used to guide stopping decisions with calibrated thresholds to limit the loss in mean Best-of-N task quality. We evaluate the controller through trajectory replay across three environments and three LLM actors. Comparisons of interpolated quality-cost Pareto frontiers show that GainCtrl generally uses fewer tokens than baselines at matched target quality. Fixed policies also achieve quality close to that of full execution while substantially reducing token consumption. The predictor runs at millisecond latency on a single CPU thread, supporting repeated monitoring without access to LLM internals.

cs.AI

新论文 · v1arXiv:2611.20733

MemEmo: Evaluating Emotion in Memory Systems

Peng Liu, Zhen Tao, Jihao Zhao, Ding Chen, Yansong Zhang, Cuiping Li, Zhiyu Li, Hong Chen

原文摘要

Memory systems address the challenge of context loss in Large Language Models during prolonged interactions. However, as an essential tool for warm companionship and personalized services, the effectiveness of memory systems in processing affective information remains underexamined. To bridge this gap, we propose an emotion-enhanced memory evaluation benchmark to assess the performance of mainstream and state-of-the-art memory systems in integrating and harnessing emotional information. First, we design a benchmark called \textbf{MemEmo} to assess the affective reasoning and tracking capabilities of memory systems across four dimensions: emotional information extraction, emotional memory updating, emotional memory question-answering, and emotional conflict detection. Second, we construct the \textbf{H}uman-\textbf{L}ike \textbf{M}emory \textbf{E}motion (\textbf{HLME}) dataset to provide an empirical foundation for scrutinizing memory systems. Finally, We evaluate five representative memory systems on the HLME dataset, reporting 11 performance metrics across four dimensions. Experimental results reveal that aggregating scores for all tested memory systems is up to 45\%, far below 100\% scale, exposing a profound disparity between memory system and affective intelligence.

cs.CL · cs.AI

新论文 · v1arXiv:2611.20732

Routing Beyond Task Categories: A Cost-Aware LAB-Bench Study

Ahmed Tariq, Leo Guinan

原文摘要

We tested whether features of biology questions or of a model's first answer can improve accuracy within an inference budget. In a LAB-Bench pilot, we compared a category router with one that adds six structural features. Neither reached the primary 80% accuracy target on development. At the 50% and 60% secondary targets, both chose Haiku for all 120 held-out questions and scored 71/120. We then froze a new comparison before collecting responses on 144 additional questions. Semantic routing scored 86/144. A policy that could defer after reading Haiku's answer scored 87/144. Both stayed within the inference budget of USD 0.678 per 1,000 queries. Against a 95% Haiku and 5% GPT mixture, the accuracy gains were 0.66 percentage points for semantic routing and 1.35 points for deferral. The paired 95% intervals were [-0.28,2.25] and [-0.14,3.47] points. A simple rule that escalated Haiku nonanswers scored 92/144 at USD 0.710 per 1,000 queries, above the budget. A retrospective nonanswer rule with a calibration-selected escalation probability gave an expected 85.63/144. At the learned deferral policy's observed inference cost, post hoc interpolation gave an expected 86.14/144. In a later development check, Astra and Gemini each scored 30/40. The best tested policy in each router family matched that score. The pair covered 33/40 in hindsight. The routers missed the 32/40 criterion, so the proposed 120-question evaluation did not start.

cs.AI