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 16:58 — 2026-11-21 16:58(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:2602.00092

通过宪法解释与控制模型行为:基于原子概念编辑

Interpreting and Controlling Model Behavior via Constitutions for Atomic Concept Edits

Neha Kalibhat, Zi Wang, Prasoon Bajpai, Drew Proud, Wenjun Zeng, Been Kim, Mani Malek

AI 速读 基于摘要

提出一种黑盒可解释性框架,通过原子概念编辑学习可验证的自然语言“宪法”,以解释和控制模型在多种任务中的行为。

cs.LG · cs.AI · cs.CL · cs.CV

查看原文摘要

We introduce a black-box interpretability framework that learns a verifiable constitution: a natural language summary of how changes to a prompt affect a model's specific behavior, such as its alignment, correctness, or adherence to constraints. Our method leverages atomic concept edits (ACEs), which are targeted operations that add, remove, or replace an interpretable concept in the input prompt. By systematically applying ACEs and observing the resulting effects on model behavior across various tasks, our framework learns a causal mapping from edits to predictable outcomes. This learned constitution provides deep, generalizable insights into the model. Empirically, we validate our approach across diverse tasks, including mathematical reasoning and text-to-image alignment, for controlling and understanding model behavior. We found that for text-to-image generation, GPT-Image tends to focus on grammatical adherence, while Imagen 4 prioritizes atmospheric coherence. In mathematical reasoning, distractor variables confuse GPT-5 but leave Gemini 2.5 models and o4-mini largely unaffected. Moreover, our results show that the learned constitutions are highly effective for controlling model behavior, achieving an average of 1.86 times boost in success rate over methods that do not use constitutions. Our open-sourced code can be found at https://github.com/google-deepmind/atomic_concept_edits

新论文 · v1arXiv:2611.21480

RESILAGENT: A Three-Pillar Architecture for Cyber-Physical Resilience in Electrical Power and Energy Systems

Nikolaos Kekatos, Georgios Koutidis, Konstantinos Antonakopoulos, Stylianos Basagiannis, Sokratis Katsikas, Georgios Kavallieratos, Alexios Lekidis, Tom Nianios 等 9 位作者

原文摘要

Electrical Power and Energy Systems face hybrid threats that combine cyber intrusions, physical sabotage and climate-driven stress, and routinely outpace today's siloed monitoring stacks. We present RESILAGENT, a concept architecture organised around three operational pillars, Sensing & Data Intelligence, Resilience Analytics & Reporting, and Incident Response & Recovery, mediated by a Cloud-based Orchestrator and exercised through a closed Detect-Simulate-Decide-Sign loop. The framework fuses agentic artificial intelligence and multimodal large language models for cross-modal detection, real-time digital twins for what-if simulation, and a permissioned blockchain anchor for tamper-evident threat intelligence and CER/NIS2-aligned evidence. We instantiate it on two scenarios, a transmission operator under coordinated cyber compromise, sabotage and heatwave, and a distribution operator under a supply-chain compromise during a storm, and report an IEEE-bus micro-benchmark that quantifies each pillar through false-data-injection attack detection, cyber-physical fusion lift, N-1 contingency screening and anchor cost.

cs.CR · cs.AI · eess.SY

新论文 · v1arXiv:2611.21476

生成式AI与高等教育评估危机:审视大学水平评估的“防AI”策略

Generative AI and the Crisis of Higher Education Assessments: Examining 'AI-Proofing' Strategies for College-Level Assessments

Bonam Mingole, Davis Yadav, Hui Min Lee, S. Shyam Sundar, Amulya Yadav

AI 速读 基于摘要

该研究收集了17个学科、37位教师设计的75道大学评估题,测试三种先进LLM的答题表现,并分析教师“防AI”策略是否有效约束模型。

cs.AI

查看原文摘要

Higher education is confronting an academic-integrity challenge unlike any in its history. Students can now outsource homework, take-home assignments, and online examinations to large language models (LLMs) that complete them in seconds. Existing research on evaluating LLM performance on academic assessments has focused largely on standardized examinations, or on questions from college-level courses drawn from a single discipline. Far less is known about how LLMs perform on college-level assessment questions spanning multiple disciplines, that instructors have purposely designed to be AI-resistant, and about which features of such questions, if any, meaningfully constrain LLM performance. We address this gap by collecting 75 questions from authentic college coursework assessments, spanning 17 disciplines, which were contributed by 37 faculty members from ten U.S. universities and one U.K. university. Of these, 61 questions were intentionally designed by faculty members to be 'LLM-proof', while the remaining 14 questions were sourced from online courses. We then (i) evaluate three state-of-the-art LLMs' ability to correctly answer these questions with and without supplementary course materials; (ii) identify five strategies, through thematic analysis, that faculty used to make their questions harder for LLMs to solve; and (iii) test whether these strategies impact LLMs' ability to answer the questions accurately. Our experimental results show that the best-performing model scored approximately 83% without supplementary course materials and 85% with them, and we observed noteworthy differences across models, answer formats, and instructor tweaking strategies. We discuss implications for assessment design, academic-integrity policy, and the broader ethics of deploying LLMs in educational settings.

新论文 · v1arXiv:2611.21473

Network Intrusion Detection Adaptability under Spurious Labels: A Noise-Aware Few-Shot Class-Incremental Learning Framework

Zijian Zhao, Davide Di Monda, Domenico Ciuonzo, Yipeng Wang, Antonio Pescapè

原文摘要

Network Intrusion Detection Systems (NIDSs) must incrementally adapt to emerging attacks, even when only a limited amount of labeled network traffic is available. Consequently, Few-Shot Class-Incremental Learning (FSCIL)-based NIDSs have gained growing attention. However, existing FSCIL-based NIDSs are highly vulnerable to label noise in newly-observed attack classes. To address this limitation, we propose NAFN, a noise-aware FSCIL-based NIDS framework that reduces the impact of mislabeled data during incremental adaptation. NAFN consists of two components: (i) Noise Detection, which identifies and removes mislabeled biflows using noise-detection algorithms, and (ii) Feature Sampling, which estimates class-conditional embeddings from clean biflows, calibrates them with base-class statistics, and resamples synthetic features to recover the original biflow count. The framework follows the Rethink Few-Shot (RFS) teacher-student paradigm with lightweight nearest-neighbor or logistic regression classifiers, and various pretraining & memory selection strategies. Experiments on CSE-CIC-IDS2018, Edge-IIoT, and IoT-NID datasets show that NAFN effectively preserves prior knowledge and maintains strong incremental performance even under 50% label noise, consistently improving new-class F1-scores over non-denoised baselines, up to +52%, while being time-efficient and robust in cross-network and high-noise incremental settings. The implementation is publicly available to foster reproducibility.

cs.NI · cs.AI · cs.LG

新论文 · v1arXiv:2611.21471

子过程约束的马尔可夫决策过程

Subprocess-Constrained Markov Decision Processes

Jiarui Gan, Debmalya Mandal

AI 速读 基于摘要

提出子过程约束MDP,要求任意可达历史后的延续期望折扣成本满足状态相关界,并证明历史依赖策略可多项式时间近似求解。

cs.AI

查看原文摘要

Constrained Markov decision processes (CMDPs) typically bound expected cumulative cost from an initial state. Such a guarantee can be insufficient when a policy must remain feasible after every history it may encounter. We introduce \emph{subprocess-constrained MDPs} (S-CMDPs), which require the expected discounted cost of each continuation, conditional on any reachable history, to satisfy a bound determined by the current state. This stronger requirement changes the structure and complexity of the problem: we show that stationary policies can be strictly suboptimal, and finding even an approximately optimal stationary policy is NP-hard. We adapt a recently introduced value-set iteration framework to show that history-dependent policies surprisingly overcome both obstacles. Our polynomial-time algorithm represents such a policy implicitly: its expected reward is within $ε$ of the optimal feasible value, it violates each subprocess cost bound by at most $δ$, and its action distribution can be computed in polynomial time for any given history.

版本更新 · v3arXiv:2602.12612

Self-EvolveRec:基于LLM方向性反馈的自演化推荐系统

Self-EvolveRec: Self-Evolving Recommender Systems with LLM-based Directional Feedback

Sein Kim, Sangwu Park, Hongseok Kang, Wonjoong Kim, Jimin Seo, Yeonjun In, Kanghoon Yoon, Hyunsik Jeon 等 9 位作者

AI 速读 基于摘要

提出Self-EvolveRec框架,通过用户模拟器提供定性批评、模型诊断工具提供定量内部验证,构建方向性反馈闭环,并让诊断工具与模型协同演化,以改进LLM驱动的推荐系统代码演化。

cs.IR · cs.AI

查看原文摘要

Traditional methods for automating recommender system design, such as Neural Architecture Search (NAS), are often constrained by a fixed search space defined by human priors, limiting innovation to pre-defined operators. While recent LLM-driven code evolution frameworks shift fixed search space target to open-ended program spaces, they primarily rely on scalar metrics (e.g., NDCG, Hit Ratio) that fail to provide qualitative insights into model failures or directional guidance for improvement. To address this, we propose Self-EvolveRec, a novel framework that establishes a directional feedback loop by integrating a User Simulator for qualitative critiques and a Model Diagnosis Tool for quantitative internal verification. Furthermore, we introduce a Diagnosis Tool - Model Co-Evolution strategy to ensure that evaluation criteria dynamically adapt as the recommendation architecture evolves. Extensive experiments demonstrate that Self-EvolveRec significantly outperforms state-of-the-art NAS and LLM-driven code evolution baselines in both recommendation performance and user satisfaction. Our code is available at https://github.com/Sein-Kim/self_evolverec.

新论文 · v1arXiv:2611.21465

用少量雅可比样本热启动神经动力学模型

Warm Starting Neural Dynamics Models From Few Jacobian Samples

Maxime Rochkoulets, Joppe De Jonghe, Mariya Ishteva

AI 速读 基于摘要

提出 TDec-Init,利用少量转移雅可比样本通过张量分解解耦出权重与非线性,从而热启动神经网络动力学模型。

cs.LG · cs.AI · eess.SY

查看原文摘要

Model-based reinforcement learning (MBRL) leverages a model of the environment, used for planning and generating samples without interacting with the real world. Motivated by evidence that informed initialization significantly improves the efficiency of MBRL, we propose TDec-Init, an algorithm to warm-start neural dynamics models using transition Jacobians. Using a decoupling algorithm based on the decomposition of a tensor of stacked Jacobian samples, we recover weights and flexible nonlinearities, and use them to initialize a feedforward neural network. In the low-data regime, we show on several benchmark environments that our approach outperforms random initialization and value training, and either matches or outperforms Sobolev training, while requiring fewer training steps between episodes.

新论文 · v1arXiv:2611.21457

视线之外,记忆之中:紧凑视觉潜变量在超高分辨率遥感中保留局部证据

Out of Sight, Into Memory: Compact Visual Latents Retain Local Evidence in Ultra-High-Resolution Remote Sensing

Meng Tang, Siyuan Du, Yuxin Liu, Shuaishuai Cao, Di Wang, Haonan Guo

AI 速读 基于摘要

作者报告,将高分辨率裁剪编码为紧凑潜变量记忆令牌,能在丢弃裁剪令牌后保留局部证据,提升超高分辨率遥感问答准确率。

cs.CV · cs.AI

查看原文摘要

Question answering over ultra-high-resolution (UHR) remote sensing scenes poses a tension for multimodal large language models: global overviews obscure small objects and local attributes, whereas detailed crops consume substantial context. Compressing inspected crops into short sequences of continuous visual latents offers a compact memory once crop tokens are discarded, yet diagnostic studies on natural images find that models often ignore latent content, leaving unclear whether compact latents can retain fine-grained targets lost in downsampled UHR remote sensing scenes. To isolate evidence retention from region selection, we encode high-resolution crops into latent memory tokens for frozen backbones and evaluate both their predictive utility and content dependence. Across two generations of Qwen-VL on LRS-GRO, a 64-token memory achieves 7.6- to 9.9-fold compression of retained crop tokens and improves accuracy by 2.5 to 4.8 percentage points over the downsampled global image alone, recovering most of the high-resolution crop gain with consistent behavior on XLRS-Bench. Substituting memory from another image or another region of the same scene reduces accuracy by 5.8 to 7.4 percentage points and yields lower accuracy than removing memory altogether, demonstrating that predictions depend on the encoded regional evidence. Restoring the source crop within the same model reduces the accuracy loss caused by zeroing or removing memory. These findings establish that compact visual latents retain useful local evidence in UHR remote sensing and identify evidence availability as a key factor governing their contribution.

新论文 · v1arXiv:2611.21456

Expertise capture in workplace AI

Andreas Ekelhart, Ahmad Mohsin, Iqbal H. Sarker, Helge Janicke

原文摘要

AI assistants are becoming part of how knowledge workers think and perform, accumulating the prompts, exemplars, memory and interaction histories that encode a worker's professional judgment. This infrastructure typically runs under employer or platform control. We name the resulting process expertise capture, the absorption of worker-applied knowledge and judgment into AI resources under organizational control. We develop a framework of contribution forms, technical residences and transformation families, and show that how far a contribution can be attributed, separated and revoked depends on where it resides and how it has been transformed, properties fixed by system design long before any dispute arises. When a worker departs, the organization retains a capability shaped by their judgment while the worker loses access to it, and the asymmetry compounds as contributions move from inspectable artifacts into derivatives, learned components and successor systems. We outline three measures, write-time provenance, extraction of separable contributions and an executable exit procedure, that make an agreed allocation technically performable, and make explicit what cannot be separated at all.

cs.CR · cs.AI

新论文 · v1arXiv:2611.21449

Replicating and Extending Weight-Space Programming: Frontier LLMs Compile Programs into Weights and Read Input Layers

Roksana Goworek, Alexander Hickey, Federico Nanni, Martin O'Reilly

原文摘要

Researchers at MIT recently reported that GPT-6 Astra, a frontier large language model (LLM), can write the weights of small neural networks directly in a single reply, and can read the raw weights of small networks to say what they were trained to do. In a three-day sprint, we replicated and extended these results. Our four findings: (1) Writing weights reproduces. Astra's networks for the game Breakout scored 514, 550 and 307 (versus MIT's 517), and scored just as well on two modified versions of Breakout. Claude Opus 5.5 can also write working networks, but often runs out of output tokens. (2) What is new is translating programs into weights. Earlier models could write a good policy as code a generation before they could write network weights, and the networks the models write behave like compiled programs: two models independently decompiled Astra's Breakout network into Python matching its actions in all tested states. On classifying handwritten digits, Astra's networks scored 50% on average, below a simple classifier (82%). (3) Reading weights is mostly reading the input layer. Astra and Opus 5.5 identify which game a network plays almost perfectly, but on three new games we designed and trained ourselves, Astra recovered where things are on the board, while mostly missing how the game works. (4) Auditing from weights alone is possible but shallow. Without executing code, Astra picked out a small language model poisoned during training in 10 of 10 runs, one poisoned by extra training afterwards in 7 of 10, and a malicious adapter in 23 of 32 cases, but never identified the trigger or malicious behaviour. Open-weight models lag behind. From this early study, we conclude that LLMs can read and write small networks, but that this currently amounts to compiling hand-designed programs and reading the layout of the input layer. It is a promising but early tool for interpretability.

cs.CL · cs.AI

新论文 · v1arXiv:2611.21441

Learning to Reason with Predictive Gains for Generative Recommendation

Zhuoyang Li, Teng Shi, Jiakai Tang, Weijie Yu, Guofu Xie, Xiao Zhang, Lei Kou, Yang Yu 等 13 位作者

原文摘要

Reasoning-enhanced generative recommendation (GR) has attracted increasing attention for leveraging the reasoning capabilities of large language models (LLMs) to improve next-item prediction. However, existing methods primarily optimize reasoning through Group Relative Policy Optimization (GRPO) with rule-based rewards determined by final recommendation performance, which faces two key limitations: (i) unmeasured predictive contributions within reasoning, where the predictive contributions of individual reasoning steps are not directly reflected in optimization signals, and (ii) weakly discriminative learning signals within groups, where different reasoning trajectories within the group often receive indistinguishable rewards, limiting effective policy optimization. In this paper, we propose GainReason, a simple yet effective training framework that optimizes reasoning through Predictive Gain (PG). Specifically, PG measures changes in target Semantic ID (SID) log-likelihood induced by individual reasoning sentences, providing contribution-aware and discriminative rewards. We further theoretically establish the alignment between expected PG and the recommendation ranking objective. Building on this signal, we develop PG-based Reasoning Optimization (PGRO) to optimize reasoning generation with sentence-level rewards, supported by a custom causal attention mechanism for efficient PG computation. Additionally, an auxiliary PG-based Reasoning Utilization (PGRU) task adaptively strengthens reasoning-conditioned prediction to complement PGRO. Extensive experiments on three public benchmarks demonstrate that GainReason consistently outperforms competitive baselines and achieves substantial performance gains from reasoning.

cs.AI

新论文 · v1arXiv:2611.21439

A Reinforcement Learning Method to Navigate SMEFT Operator Space

Jacky Kumar

原文摘要

In particle physics, discrepancies between experimental measurements and Standard Model (SM) predictions are commonly interpreted within new theoretical models. In this context, the Standard Model Effective Field Theory (SMEFT) provides a general framework encompassing a broad class of models. However, exploring the enormous space of possible SMEFT operators rapidly becomes a challenging combinatorial problem: many operators can contribute simultaneously to different observables, their effects are correlated, and distinct operator combinations can provide comparably good descriptions of the data. In this work, we formulate the exploration of the SMEFT operator space as a sequential decision-making problem using reinforcement learning (RL). Our RL agent employs a lightweight actor-critic architecture with approximately 40K trainable parameters in total. Rather than searching for a single best-fit theory, our method is designed to discover a diverse set of high-quality theory hypotheses. We test the method in a complex phenomenological environment comprising 132 observables across several sectors of particle physics, including 24 measurements that deviate from their SM predictions by more than $1.5σ$. We show that the learned RL policy efficiently constructs multi-operator theories that substantially improve the global fit. Comparisons with independent random-subset and random-sequential searches demonstrate the ability of the RL policy to navigate this large combinatorial space. Our results establish RL as a systematic framework for exploring diverse patterns of new physics (NP) interactions within SMEFT.

cs.LG · cs.AI · hep-ex · hep-ph

版本更新 · v2arXiv:2611.16591

Singularity-Bench: Can Agents Build Agents?

Yang Zhou, Krish Agarwal, Zhuoming Chen, Zerui Wang, Zhenyu Zhang, Wenshuang Li, Xinpeng Wei, Zhichen Zeng 等 19 位作者

原文摘要

Frontier AI agents increasingly succeed on realistic, long-horizon tasks and are beginning to improve the systems through which they operate. Yet existing benchmarks often restrict agents to narrow action spaces, use simplified environments, or rely on verification methods that are unreliable for evaluating arbitrary agent-generated systems. We introduce Singularity-Bench, an open-ended benchmark for autonomous AI development with six end-to-end tasks spanning pre-training, mid-training, post-training, inference, the agent harness, and cluster orchestration. Each task requires agents to implement core systems from a minimal starting point and evaluates them against mature human-engineered baselines on identical hardware. Additionally, through extensive red-teaming covering 9 of the 10 categories in the OWASP Top 10 for Agentic Applications, we expose reward-hacking vulnerabilities in the conventional Harbor framework and develop Breakwater, an extension of Harbor that blocks all attacks tested in our study. Across 11 agents built from 7 frontier models and 2 harnesses, Opus 5.5 with Claude Code matches or outperforms the human baseline on all six tasks with a Singularity-Index of 1.25, suggesting it may have the skills needed to reach the singularity. The remaining agents average only 0.60. Our vision is for Singularity-Bench to track how much of AI development agents can already automate and when they may transition from assisting to taking end-to-end responsibility for building stronger successors.

cs.AI

版本更新 · v2arXiv:2611.12529

Perceived Agency During Agent-Avatar Control Transitions in Virtual Reality

Jingyi Zhang, Ruijun Sun, Klara Brandstatter, Nels Numan, Anthony Steed

原文摘要

Prior beliefs about whether a virtual human (VH) is controlled by a human or a computer can shape how users interpret and respond to its behaviour. As emerging hybrid VH systems allow control to transfer between human operators and computer programs during an ongoing interaction, it remains unclear whether perceived agency updates with the actual controller or continues to reflect users' prior beliefs and first impressions. We investigated this question in a within-subject VR study with 24 participants across four conditions: always computer-controlled, computer-to-human, human-to-computer, and always human-controlled. We designed an interaction task in which participants acted as interviewers, asking the VHs a fixed set of questions. Before each interaction, participants were informed whether the VH was human- or computer-controlled, matching the actual controller at the beginning, but were unaware of possible control transitions. We found that perceived agency was strongly influenced by the controller presented at the beginning of the interaction, with no evidence of updating following the control transitions. Social evaluations including likeability and co-presence were generally higher in conditions where human control occurred later and covered a greater proportion of the interaction. The results suggest that prior beliefs and first impressions continued to shape perceived agency, whereas social evaluations were more sensitive to the VH's actual behaviour over the interaction.

cs.HC · cs.AI

版本更新 · v2arXiv:2610.28537

Beyond 4 bits: Data-Centric Post-training Quantization and Recovery

Rhea Sanjay Sukthanker, Riccardo Grazzi, James Hensman, Pashmina Cameron

原文摘要

Post-training quantization (PTQ) compresses the weights of a trained model with a limited compute and data budget. Down to 4 bits, state-of-the-art PTQ methods are near-lossless; below this limit, the tested scalar PTQ configurations collapse to near-random accuracy and quality becomes acutely sensitive to choices such as the quantization representation and calibration data. At such low bitwidths, recovery fine-tuning becomes crucial, and its effectiveness depends heavily on these same choices. Yet there is little systematic evidence on how the quantization method, the quality and task alignment of calibration data, and the design of the recovery procedure interact to determine the final 2-bit model. We present a systematic study of 2-bit quantization and recovery fine-tuning across model families, scales, and data distributions, and distill our findings into practical, architecture-dependent best practices for obtaining an accurate 2-bit model. This recipe, which combines a vector codebook representation, a small task-aligned calibration mixture, and a short codebook-only distillation stage, restores 63-68% average accuracy across seven question-answering tasks at 2 bits (a 71-73% BF16 ceiling), versus the near-random 29% of the evaluated scalar PTQ pipelines, at roughly half the compute of a full-model QAT baseline. Given the same calibration and recovery data, EfficientQAT, a trained scalar baseline, avoids the collapse. However, at a matched rate it trails the vector recipe on generation on every model (2.8-10.9 AvgGen; within noise on Qwen3-MoE) and on question answering on three of the four studied models (2.7-7.8 AvgQA). We also find that the best recovery data depends on the calibration data, an interaction confirmed on a held-out model. Our architecture-aware map of the 2-bit design space shows that representation, calibration data, and recovery data must be chosen jointly.

cs.AI

新论文 · v1arXiv:2611.21423

Emergent Steering Beyond Endpoint Alignment in Chemical Reaction Models

Yili Shen, Kehan Guo, Haomin Zhuang, Xiangliang Zhang

原文摘要

Machine learning models for chemical reaction prediction are usually evaluated by endpoint accuracy, but endpoint correctness alone does not reveal how a model represents chemical transformation or why it fails. We introduce UniRxnAxis, a unified forward-retrosynthesis framework that models both tasks as direction-conditioned endpoint redistribution in a shared latent space. By aligning source states, transformed states, and encoded target endpoints, UniRxnAxis enables directional auditing of learned latent updates. This shared latent geometry allows the learned displacement (from the encoder source state $S_0$ to the decoder-facing transformed state $S_1$) to be decomposed into an endpoint-parallel component and an endpoint-orthogonal residual, which we call emergent steering. Our evaluation shows that emergent steering is functionally important: in retrosynthesis, it recovers most of the model's utility, remains effective under random, shuffled, and prototype controls, and directly affects which bond edits are promoted or suppressed. The residual-over-parallel ordering persists across alignment strengths, after decoder-interface calibration, against angle-matched random axes, and in a second graph-edit architecture, although its magnitude depends on the reference axis and optimization. The decomposition also supports a proxy taxonomy of model-side failures, showing how reaction models can be audited beyond endpoint accuracy.

cs.LG · cs.AI

版本更新 · v2arXiv:2602.14869

Concept Influence: Leveraging Interpretability to Improve Performance and Efficiency in Training Data Attribution

Matthew Kowal, Goncalo Paulo, Louis Jaburi, Tom Tseng, Lev E McKinney, Stefan Heimersheim, Aaron David Tucker, Adam Gleave 等 9 位作者

原文摘要

As large language models are increasingly trained and fine-tuned, practitioners need methods to identify which training data drive specific behaviors, particularly unintended ones. Training Data Attribution (TDA) methods address this by estimating datapoint influence. Existing approaches like influence functions are both computationally expensive and attribute based on single test examples, which can bias results toward syntactic rather than semantic similarity. To address these issues of scalability and influence to abstract behavior, we leverage interpretable structures within the model during the attribution. First, we introduce Concept Influence which attribute model behavior to semantic directions (such as linear probes or sparse autoencoder features) rather than individual test examples. Second, we show that simple probe-based attribution methods are first-order approximations of Concept Influence that achieve comparable performance while being over an order-of-magnitude faster. We empirically validate Concept Influence and approximations across emergent misalignment benchmarks and real post-training datasets, and demonstrate they achieve comparable performance to classical influence functions while being substantially more scalable. More broadly, we show that incorporating interpretable structure within traditional TDA pipelines can enable more scalable, explainable, and better control of model behavior through data.

cs.AI · stat.ML

新论文 · v1arXiv:2611.21408

Do Embedding-Map Distances Guide LLM Ideation? Four Tests Against What Came Next

Javier Aguilar Martín

原文摘要

A popular ideation recipe treats an embedding map of existing work as a guide: pick points far from everything and decode them into ideas. I test four ways such distances could guide ideation against what appeared later. (1) Far random directions are not text: no later item reaches a raw cosine above about 0.10 with one (after centring or whitening, 22 of 24 cells stay out of reach), and an LLM decoder shown a target's neighbours, with score-only feedback, writes those neighbours in all six target sets. A neighbour-free search writes coherent text at median cosine 0.57 and 0.47 to interior gaps, beating the nearest existing item on 15 and 14 of 20, but only 0.07 and 0.06 to far points. (2) On k-means cells of a past-only map, the registered hole score mostly tracks cell size and position; three density-based gap scores forecast which cells grow fastest at or below chance (AUC 0.30 to 0.66) and add nothing to momentum in Steam and arXiv, with patents inconclusive; more isolated new Steam games are slightly better reviewed. (3) LLM continuations of arXiv papers stay closer to the source than papers that cite it, under three user prompts and against proposal rewrites (gpt-4o) and April-December 2023 papers (OLMo). Critique lengthens this step beyond content-free revision, replicated on 150 new papers. Revealing a real follow-up's title turns the ideas by 0.032 along a follow-up-specific direction beyond revision; reviewer, oracle and distance feedback add 0.13, 0.08 and 0.34 of that turn (upper bounds 0.46, 0.46, 0.70), and their turn is not separable from revision or their longer step. (4) For inspiration material graded by distance, no inverted U was detected, on an anticipation margin that tracks format more than content. Original tests were registered before running, follow-ups (thirteen requested in two AI-agent reviews) after the main results; all are dated only by local git commits.

cs.LG · cs.AI · cs.CL · cs.DL

新论文 · v1arXiv:2611.21407

Harness Leverage Across Model Size: A Ladder of Five Local Agent Harnesses from 0.75B to 4.7B

Hao Wang, Ting Huang

原文摘要

A harness turns a model into a system that completes tasks: it builds the prompt, exposes tools and decides when work is done. Harness comparisons hold the model fixed. Model comparisons hold the harness fixed. We vary both: five shipped harnesses, 18 artifact-producing tasks, four locally deployable open-weight models from 0.75B to 4.7B and a 35.5B anchor, with one deterministic checker scoring all 360 ladder cells. The completion-controlling harness, Mingbird, has the highest mean at every size. Of four pre-registered paired contrasts, two survive Holm correction, at 0.75B and 4.7B. Thirteen of sixteen paired intervals exclude zero under the deployment reading, with leads from 0.075 over the strongest rival to 0.819 over the weakest. The completion gate is not the carrier: disabling all six executable checks moves the mean by +0.080, +0.117, +0.003 and -0.004, every interval including zero, and only at 1.9B does disabling it create empty runs. Removing one repeat-blocking rule alone costs 0.411 at 0.75B and leaves eight of eighteen tasks with no artifact. The harness effect is a progress effect: it decides whether a run leaves anything on disk. Of 357 usable cells, 127 leave none of their declared artifacts and 85 of those exit normally, so a scorer that reads exit status credits work that was never written. Whether the effect grows or shrinks with size, four rungs cannot settle: a floor arm produces a rise, a ceiling a decline, and a cross-family rung a peak.

cs.AI · cs.SE

新论文 · v1arXiv:2611.21405

Shaping AI's Beliefs Before It Generates: Human Sense of Agency in Creative AI

Zi Wang, Mahima Pushkarna

原文摘要

AI systems for creative media generation expand human capabilities yet risk diminishing human agency. A core problem is anchoring, where the user's thinking becomes anchored to a polished, generated artifact, shifting the user's role from active authoring to reactive editing. We propose pre-generation belief controls that can empower creators to inspect, revise, and steer the AI's interpretation before any media is generated, so that the initial anchor is set by the human. We formalize four design principles and instantiate them in the Proactive Co-Creator (PaCC), a system for early-stage conceptual ideation and prototyping that proactively surfaces its assumptions and uncertainty through clarifications, belief graphs, and media attributes, inviting users to shape the AI's beliefs before generating the final outcome. Using PaCC as a design probe in a small-scale, single-arm exploratory user study, we observed that participants reported directionally positive sense of agency and process control relative to their familiar prompt-based AI tools, noting that surfaced assumptions helped them think through details they had overlooked, whereas their feelings about ownership and authorship diverged

cs.AI · cs.HC

版本更新 · v2arXiv:2601.23163

Probing the Trajectories of Reasoning Traces in Large Language Models

Marthe Ballon, Brecht F. Verbeken, Vincent Ginis, Andres Algaba

原文摘要

Large language models (LLMs) increasingly solve difficult problems by producing reasoning traces before emitting a final response. However, it remains unclear how accuracy and decision commitment evolve along a reasoning trajectory, and whether intermediate trace segments provide answer-relevant information beyond generic length or stylistic effects. Here, we propose a trajectory-probing protocol for evaluating how a reasoning model's predicted answer evolves along a partial reasoning trace. The protocol (1) generates a model's full reasoning trace, (2) truncates it at fixed token-percentiles, and (3) injects each partial trace back into the model, measuring the model's induced answer distribution. We apply the protocol to seven open-source reasoning models (Qwen3-4B/8B/14B, gpt-oss-20b/120b, DeepSeek-R1-Distill-Llama-8B and Gemma-4-31B-it) across four benchmarks (GPQA Diamond, MMLU-Pro, Omni-MATH-2 and CompMath). The protocol reveals that accuracy and decision commitment generally increase as the percentage of provided reasoning tokens grows, and length, form, and token-identity controls confirm that gains stem from instance-specific semantic content rather than context length or generic reasoning style effects. Furthermore, cross-model probing shows that even stronger models frequently repeat the source model's errors when given its partial reasoning trace, while allowing further reasoning substantially improves recovery. Together, these results provide a reproducible, model-agnostic measurement of reasoning trace dynamics that generalizes across answer formats and can reveal model-specific failure modes invisible to aggregate accuracy.

cs.LG · cs.AI

新论文 · v1arXiv:2611.21383

The Mask Is Not the Object: Volumetric Supervision for 3D Gaussian Segmentation

Jeonghwan Cho, Minsu Kim, Junyoung Hong, Seon Joo Kim

原文摘要

3D Gaussian Splatting enables object-level operations on Gaussian primitives, yet existing segmentation methods are largely trained and evaluated through rendered views. This mismatch can produce accurate rendered masks from incomplete 3D selections since alpha-blended feature supervision concentrates learning on dominant contributors while leaving occluded or low-contribution Gaussians weakly supervised. We propose VODA, a plug-and-play supervision objective that exploits local depth structure to group per-pixel contributors and directly supervises all Gaussians in selected depth clusters, bypassing contribution-weighted gradient delivery. To evaluate volumetric completeness, we introduce Removal-3D, a synthetic object-removal benchmark, and the Removal Depth Error (RDE), a metric comparing predicted and ground-truth depth changes. Across four rendering-based baselines, VODA consistently improves volumetric segmentation on Removal-3D while maintaining or improving performance on LERF and ScanNet. Further analysis shows that VODA recovers lower-contribution Gaussians, while simple gradient reweighting and spatial propagation fail to reproduce comparable gains.

cs.CV · cs.AI

新论文 · v1arXiv:2611.21382

VisualArtifactSpec 1.0: Evaluating Visual Artifacts under Task Specifications

Shuyi Ouyang, Xiaojun Tang, Zijian Zou, Xintong Zhang, Mingli Song, Guannan Zhang, Hongwei Xue

原文摘要

Visual artifact tasks test agents' ability to translate task specifications and source materials into outputs that meet practical needs. Evaluation across artifact families helps distinguish broadly applicable authoring capabilities from strengths confined to a particular medium. We introduce VisualArtifactSpec 1.0 (VAS), a benchmark of 211 creation and modification tasks spanning presentations, static graphics, animations and videos, and web interfaces. Each task specifies authoring objectives, source materials, constraints, and deliverables. Existing benchmarks provide limited support for systematically evaluating how authoring choices suit different audiences, purposes, and media across artifact families. VAS evaluates two aspects separately: Task Fulfillment verifies explicit requirements and applicable functionality, while Artifact Performance assesses Content, Organization, Visuals, and Delivery. Task-specific rating anchors instantiate shared criteria for the intended audience, purpose, and medium, with judgments grounded in source materials and artifact inspection records. We evaluate nine model-harness configurations on all 211 tasks. Overall scores range from 42.12 to 70.52, with Artifact Performance accounting for 73.4% of the gap between the highest- and lowest-scoring configurations.

cs.AI

新论文 · v1arXiv:2611.21707

Bookkeeping Is Not Judgement: EJBench, an Engineering Judgement Benchmark Built from a Conjugate Heat Transfer Design Campaign

James Sadler, Rotimi Alabi, Kotub Uddin

原文摘要

We introduce EJBench, a benchmark of engineering judgement: whether a model can predict how a design behaves at new conditions from a few earlier results, and decide on it. Unlike engineering benchmarks that test whether an agent can operate simulation or CAD software, EJBench is software-agnostic and asks what an engineer would conclude without running a simulation. Its tasks come from a conjugate heat transfer design campaign for battery cold plates, in three families: matching unlabelled test results to their operating points, choosing which stored runs to reveal before signing off a design, and ranking new duties by peak cell temperature. Every answer is taken from a converged simulation, not from an LLM judge. The best of ten frontier models, GPT-6 Astra, scores 0.71, and models state almost the same confidence on tasks they get wrong as on tasks they get right. We also post-train Qwen3.5-9B with LoRA and reinforcement learning, rewarded by the share of answer criteria the simulation results confirm; on the held-out test split its score rises from 0.071 to 0.106. We release the train, validation and test splits, the evaluation harness and the leaderboard.

cs.CE · cs.AI · cs.CL · cs.LG

新论文 · v1arXiv:2611.21375

Proceedings Eighth International Workshop on Formal Methods for Autonomous Systems

Matt Luckcuck, Maike Schwammberger, Akhila Bairy

原文摘要

This EPTCS volume contains the papers from the Eighth International Workshop on Formal Methods for Autonomous Systems (FMAS 2026), which was held on the 17th and 18th of November 2026. FMAS 2026 was co-located with the 27th International Conference on Formal Engineering Methods (ICFEM), and hosted at the Leonardo Royal Grand Harbour Hotel in Southampton.

cs.LO · cs.AI

新论文 · v1arXiv:2611.21373

WeVisForensics: Self-Evolving Agent for Omni AI-Generated Video Detection

Yipeng Du, Yuxi Li, Ying Zhang, Chong Sun, Chen Li, Jing Lyu, Ying Tai

原文摘要

Recent advances in video generation call for reliable detection grounded in explicit forensic evidence. However, current methods primarily depend on single-pass chain-of-thought reasoning that overlooks brief or localized artifacts, or on predefined multi-agent workflows that offer limited flexibility in selecting relevant inspections. Even with adaptive investigation, optimizing the investigation policy alone does not directly improve the forensic capabilities of fixed experts. We introduce WeVisForensics, an agentic framework in which a main investigator agent coordinates a self-evolving library of forensic inspector sub-agents spanning visual and audio cues. The investigator learns which artifact type to examine, over which window, and when to conclude, and these interactions provide shared feedback for both levels of agents. Our dual-timescale policy optimization strategy, DTPO, closes this loop through fast reinforcement learning of the investigator and slower refinement of the inspectors, guided by annotated artifact intervals, to report subtle artifacts they once dismissed. Furthermore, we curate ForenCue-30K, combining role-specific forensic annotations and interaction trajectories to supervise individual inspections and their coordination, with an accompanying reinforcement-learning pool. Experiments show that WeVisForensics averages 95.2% detection accuracy across in-domain, out-of-domain and agentic benchmarks, 9.0% above the strongest prior detector, while grounding its verdicts in localized evidence.

cs.CV · cs.AI

新论文 · v1arXiv:2611.21367

Premise Recovery in Software Design: What Does This Decision Rest On?

Priyabanta Sandulua, Islam Mesabaha, Sebastian J. Vollmer

原文摘要

Design decisions shape software architecture, and writing them down preserves the reasons behind it. Architecture decision records (ADRs) capture them, each with its context, choice, and consequences. A decision can depend on earlier ones: reversing one of them would force its review. We call these its premises. Yet ADRs do not always name their premises. When a decision is reversed, engineers supersede its ADR and search by hand for what it affects. That search can miss ADRs resting on an unnamed premise. We frame this as premise recovery: given an ADR with identifiers masked, find its premises among earlier records. We collect 1,060 ADRs from 36 repositories and label 1,529 candidate pairs for 60 target ADRs. Two LLM annotators agree at \k{appa} = 0.859. An independent human annotator, blind to their labels, agrees with the stated/implicit tag on 30 of 32 premises (\k{appa} = 0.875). We compare 7 open-weight models with 2 baselines that use no LLM. Qwen3.8-27B reaches F1 0.756, against 0.466 for word overlap; 2 models added post hoc score similarly. On the 11 premises whose titles barely overlap the target, it finds 0.758, while word overlap finds none. Reasoning raises its recall of implicit premises but not its F1. Word overlap with as many predictions gains 69% as much. We find only 2 to 3 stale dependents, though we searched for unlinked ones in only 30 of 66 supersessions. Open models could thus suggest unwritten premises when a decision is reversed, with precision between 0.68 and 0.78.

cs.SE · cs.AI · cs.MA

新论文 · v1arXiv:2611.21362

What You Post Is What You Symptom: Estimating Psychometric Networks from Social Media Discourse

Ştefana-Arina Tăbuşcă, Ana Sabina Uban, Liviu P. Dinu

原文摘要

Symptom networks in psychopathology are estimated almost exclusively from self-report questionnaires, restricting inference to whatever symptom set a given instrument covers. We introduce a framework for symptom-resolved network estimation directly from social media discourse and apply the framework to depression and eating disorders, with a third anorexia corpus as cross-dataset validation. To our knowledge, we are the first to estimate symptom-resolved psychometric networks from social media text, compare them to clinical co-endorsement networks, and replicate transdiagnostic linguistic markers across 3 independently recruited corpora spanning 2 disorders simultaneously. For depression, expression scores correlate with BDI-II totals at $ρ{=}0.434$, 20 of 21 items reach FDR significance, and split-half reliability is 0.935. The Network Comparison Test detects few edge differences between modalities (3-5\% of edges), while bridge symptoms diverge. For eating disorders, the item-level results are exploratory. Identified linguistic features replicate the same direction across all 3 corpora, and they cluster in three construct families.

cs.CL · cs.AI · cs.CY · cs.SI

新论文 · v1arXiv:2611.21348

Representational Change Under Self-Consuming Training: A Sparse Autoencoder Analysis

Suchit Gupte, Zhongteng Cai, Xueru Zhang, Mohammad Mahdi Khalili

原文摘要

Large language models are increasingly exposed to model-generated text, raising the possibility that future models will be trained recursively on the outputs of their predecessors. Prior work on such self-consuming training loops has largely focused on output quality and downstream performance, leaving their effects on internal representations poorly understood. We study how representations evolve across five generations of self-consumption and whether sparse autoencoders (SAEs) remain reliable under this shift. At each generation, we compare a lineage trained on synthetic data with a matched control trained on real data under the same training settings. We evaluate how well SAEs trained before self-consumption transfer to later generations, measuring reconstruction and sparsity across SAE widths, architectures, and model layers. We then fine-tune the original SAE on each generation and use sparse probing to test whether task-relevant information remains concentrated in a small number of latents or becomes more distributed across the representation. We find that zero-shot SAE performance degrades progressively under self-consumption, with growing reconstruction and sparsity gaps that vary substantially across layers. Fine-tuning largely restores reconstruction, but not feature-level usefulness: task-relevant information remains accessible in the full latent space while becoming less concentrated in individual features. These results show that self-consumption induces representational drift and that reconstruction fidelity alone is insufficient to establish that an adapted SAE remains equally useful for interpretation.

cs.LG · cs.AI

版本更新 · v2arXiv:2611.08684

Cancellation Is Not Enough: What Retries Leave Behind in an Overloaded LLM Engine

Xindan Zhang, Fei Wang

原文摘要

Clients and proxies in front of an LLM engine retry calls that time out. After a transient overload, these retries can keep the engine from recovering even when calls arrive below its capacity. Since the engine may keep computing for clients that have left, propagating cancellation looks like the natural remedy. In this paper, we test it on vLLM with overload experiments on three open models, and with scheduler traces of one that follow each client attempt to the execution it creates in the engine. We find that with retries on, cancellation alone did not restore service in any run, and only 12-38% of the calls that arrived after the overload were delivered. The traces suggest why: retries that vLLM ran typically started prefill less than half a second before they timed out. Executions that vLLM later aborted took 79-85% of the prompt tokens it computed, and vLLM's prompt-token counter missed a third to half of all computed prompt tokens. A retry budget restored service by shedding retries but left many of the overload's calls undelivered. Request coalescing delivered every call from the overload onward, some late, unless cancellation had already aborted the running execution. Overload controls should thus be judged by the executions that retries add and the calls that clients receive.

cs.AI