|
创作新主题 |
社区所有版块导航
DATA
| docker Elasticsearch |
WEB开发
| linux MongoDB Redis DATABASE NGINX 其他Web框架 web工具 zookeeper tornado NoSql Bootstrap js peewee Git bottle IE MQ Jquery |
机器学习
| 机器学习算法 |
产品
| 短视频 |
印度
| 印度 |
一周十大热门主题
点击阅读原文访问arxivdaily.com，涵盖CS|物理|数学|经济|统计|金融|生物|电气领域，更有搜索、收藏等功能！
cs.LG 方向，今日共计289篇
大模型相关(48篇)
【1】LLMSurgeon: Diagnosing Data Mixture of Large Language Models
标题：LLMSurgeon：诊断大型语言模型的数据混合
链接：https://arxiv.org/abs/2605.30348
备注：ACL 2026 Main. Code at https://github.com/Yaxin9Luo/LLMSurgeon
摘要：The pretraining data mixture of Large Language Models (LLMs) constitutes their "digital DNA", shaping model behaviors, capabilities, and failure modes. Yet this composition is rarely disclosed, making post-hoc auditing of data combination or provenance difficult. In this work, we formalize $\textbf{Data Mixture Surgery (DMS)}$: given only generated text from a target LLM, estimate the domain-level distribution of its pretraining corpus under a predefined taxonomy. We propose $\textbf{LLMSurgeon}$, a strong framework that casts DMS as an inverse problem under the label-shift assumption. Rather than directly aggregating classifier outputs, LLMSurgeon estimates a calibrated $\textit{soft}$ confusion matrix and solves a constrained inverse problem to correct systematic domain confusion and recover the latent mixture prior. To evaluate, we introduce $\textbf{LLMScan}$, a recipe-verifiable evaluation suite built from open-source LLMs with transparent pretraining mixtures. Across LLMScan, LLMSurgeon recovers domain mixtures with high fidelity under fixed protocols. Our work presents a practical, post-hoc approach for auditing the digital DNA of foundation models without access to their training data.
【2】Efficient Test-Time Finetuning of LLMs via Convex Reconstruction and Gradient Caching
标题：通过凸重建和梯度缓存对LLM进行有效的测试时微调
链接：https://arxiv.org/abs/2605.30337
摘要：Test-time finetuning (TTFT) is a rapidly evolving paradigm that adapts a language model to each prompt by retrieving related sequences, updating the model on them, and then evaluating the prompt. However, TTFT is only practical if it is fast: selection and finetuning both happen per query, making each a direct bottleneck. Existing methods trade speed for quality: fast retrieval is often redundant, while stronger diversity-aware selection adds prohibitive per-query cost. We introduce HullFT, a geometric approach to TTFT that addresses both bottlenecks. Given a query, HullFT first represents the query embedding as a sparse convex combination of few training sequences, using efficient projection-free Frank-Wolfe optimization. This yields a support set that is inherently relevant and diverse. We then convert the fractional convex weights into an exact integer multiset for finetuning through a geometric integerization procedure. The resulting multiplicities naturally create repeated examples, which we exploit with Gradient Reuse to amortize forward-backward computation across repeated finetuning steps. Our experiments show that HullFT improves the quality-efficiency tradeoff over current state-of-the-art TTFT methods, achieving lower bits-per-byte at substantially lower total runtime.
【3】Resolution Diagnostics for Paired LLM Evaluation
标题：配对LLM评估的分辨率诊断
链接：https://arxiv.org/abs/2605.30315
备注：16 pages, 7 figures, 12 tables. Accepted to the ICML 2026 Workshop on Hypothesis Testing, Seoul, South Korea, 2026. Copyright 2026 by the author(s)
摘要：Across two public LLM leaderboards, many displayed pairwise rankings do not meet a conventional paired-test resolution target under the actual paired evaluation design: 11 of 40 Open LLM Leaderboard v1 pairwise comparisons and 4 of 9 MMLU-Pro top-10 adjacent-rank pairs are unresolved at (alpha, 1-beta) = (0.05, 0.8). The MMLU-Pro count rises to 6/9 under real subject-level clustering and stays at 5-6 out of 9 in 99.9% of category-bootstrap resamples. We frame paired LLM evaluation as a hypothesis-testing problem, invert level-alpha, power-(1-beta) tests, and report a per-pair resolution ratio q = N/N* as the primary diagnostic. A sharp small-effect expansion with an explicit second-order constant shows that the widely-used unpaired Cohen-h-plus-(1-rho) shortcut deviates from the correct N* by approximately a factor of two in the close-comparison regime, a deficit that three of five off-the-shelf calculators(Cohen 1988, G*Power, R pwr) silently inherit when the user post-multiplies their per-arm output by (1-rho). The unresolved-pair pattern remains under multiplicity correction and anytime-valid sequential testing.
【4】How LoRA Remembers? A Parametric Memory Law for LLM Finetuning
标题：LoRA如何记住？LLM微调的参数记忆定律
链接：https://arxiv.org/abs/2605.30260
备注：Ongoing work
摘要：
【5】OOD-GraphLLM: Graph Large Language Model for Out-of-Distribution Generalized Drug Synergy Prediction
标题：OD-GraphLLM：用于非分布广义药物协同预测的图形大语言模型
链接：https://arxiv.org/abs/2605.30247
备注：12 pages, 9 figures, ACM KDD 2026
摘要：
【6】How's it going? Reinforcement learning in language models recruits a functional welfare axis
标题：过得怎么样？语言模型中的强化学习招募功能福利轴
链接：https://arxiv.org/abs/2605.30232
备注：81 pages, 43 figures, 32 tables
摘要：
【7】When Should Models Change Their Minds? Contextual Belief Management in Large Language Models
标题：模特什么时候应该改变想法？大型语言模型中的上下文信念管理
链接：https://arxiv.org/abs/2605.30219
备注：Work in progress
摘要：
【8】MarginGate: Sparse Margin-Triggered Verification for Batch-Invariant LLM Inference
标题：MarginGate：批量不变LLM推理的稀疏边缘触发验证
链接：https://arxiv.org/abs/2605.30218
备注：13 pages, 5 figures, 11 tables
摘要：
【9】Unveiling the Visual Counting Bottleneck in Vision-Language Models
标题：揭开视觉语言模型中的视觉计数瓶颈
链接：https://arxiv.org/abs/2605.30170
备注：ICML 2026
摘要：
【10】Overcoming Forgetting in LLM Fine-Tuning with Evolution Strategies
标题：用进化策略克服LLM微调中的遗忘
链接：https://arxiv.org/abs/2605.30148
摘要：
【11】Convergence Theory for Iterative LLM-Based Neural Architecture Search: A Parametric Cross-Entropy Framework with Closed-Form Proxy Reliability
标题：基于LLM的迭代神经架构搜索的收敛理论：具有封闭形式代理可靠性的参数交叉熵框架
链接：https://arxiv.org/abs/2605.30103
备注：14 pages, 2 figures, 2 tables. Submitted to NeurIPS 2026
摘要：
【12】Latent Performance Profiling of Large Language Models
标题：大型语言模型的潜在性能分析
链接：https://arxiv.org/abs/2605.30018
摘要：
【13】Fingerprinting Inference Systems of Large Language Models
标题：大型语言模型的指纹推理系统
链接：https://arxiv.org/abs/2605.29979
摘要：
【14】Honeyval: A Comprehensive Evaluation Framework for LLM-powered HTTP Honeypots
标题：Honeyval：LLM支持的HTTHELP蜜罐的全面评估框架
链接：https://arxiv.org/abs/2605.29963
摘要：
【15】Does The Way You Plan Matter? An Empirical Study of Planning Representations for LLM Web Agents
标题：你的计划方式重要吗？LLM Web代理规划表示的实证研究
链接：https://arxiv.org/abs/2605.29927
备注：Extended version of paper submitted to EMNLP, waiting for acceptance
摘要：
【16】Plan, Don't Pose: Long Composite Motion Generation with Text-Aligned BFM
标题：计划，不要摆姿势：使用文本对齐BFM生成长复合运动
链接：https://arxiv.org/abs/2605.29906
摘要：
【17】Dissecting the Black Box: Circuit-Level Analysis of LLM Vulnerability Detection
标题：剖析黑匣子：LLM漏洞检测的电路级分析
链接：https://arxiv.org/abs/2605.29901
备注：11 pages, 6 figures. Supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP)
摘要：
【18】HARP: Hadamard-Preconditioned Adaptive Rotation Processor for Extreme LLM Quantization
标题：HARP：用于极端LLM量化的Hadamard预处理自适应旋转处理器
链接：https://arxiv.org/abs/2605.29843
摘要：
【19】Data filtering methods for training language models
标题：训练语言模型的数据过滤方法
链接：https://arxiv.org/abs/2605.29807
备注：AINL-2026
摘要：
【20】Hista and Numca: Estimate State Value Effectively for LLM Reinforcement Learning
标题：Hista和Numca：有效估计LLM强化学习的状态值
链接：https://arxiv.org/abs/2605.29782
备注：Accepted at ICML 2026
摘要：
【21】Cluster-Level Attention-Guided Parallel Decoding for Masked Diffusion Language Models
标题：掩蔽扩散语言模型的高级注意力引导并行解码
链接：https://arxiv.org/abs/2605.29607
摘要：
【22】SCOPE: A Lightweight-training LLM Framework for Air Traffic Control Readback Monitoring
标题：范围：空中交通管制回读监控的轻量级训练LLM框架
链接：https://arxiv.org/abs/2605.29543
摘要：
【23】AsymVLM: Asymmetric Token Pruning for Efficient Vision-Language Model Inference
标题：AsymVLM：用于高效视觉语言模型推理的非对称令牌修剪
链接：https://arxiv.org/abs/2605.29535
摘要：
【24】Kronecker Embeddings: Byte-Level Structured Token Representations for Parameter-Efficient Language Models
标题：Kronecker Embeddings：参数高效语言模型的字节级结构化令牌表示
链接：https://arxiv.org/abs/2605.29459
备注：28 pages, 16 tables. Reference implementation: this https URL
摘要：
【25】When Does Persona Prompting Actually Help? A Retrieval and Metric Analysis of Expert Role Injection in LLMs
标题：女神异闻录什么时候真正有帮助？LLM中专家角色注入的检索和度量分析
链接：https://arxiv.org/abs/2605.29420
备注：6 pages, 2 figures. Submitted for peer review
摘要：
【26】GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models
标题：GDSD：强化学习作为扩散语言模型的引导降噪器自蒸馏
链接：https://arxiv.org/abs/2605.29398
备注：Preprint
摘要：
【27】PassNet: Scaling Large Language Models for Graph Compiler Pass Generation
标题：PassNet：扩展大型语言模型以生成图形更容易通行证
链接：https://arxiv.org/abs/2605.29357
备注：Code and data available at https://github.com/PaddlePaddle/PassNet
摘要：Modern tensor compilers such as TorchInductor deliver substantial speedups on mainstream models, yet face a systematic performance ceiling on long-tail workloads -- our profiling shows that 43% of real-world subgraphs experience end-to-end slowdowns under default compilation. While LLMs offer a path toward automated optimization, existing efforts focus on standalone kernel generation. We argue that pass generation -- where LLMs author structured graph transformations that integrate directly into compiler pipelines -- is the more appropriate abstraction. We propose PassNet, the first large-scale ecosystem for LLM-based compiler pass generation, comprising: (1) PassNet-Dataset, over 18K unique computational graphs from 100K real-world models; and (2) PassBench, 200 curated long-tail fusible tasks (comprising 2,060 subgraphs in total) evaluated under the Error-aware Speedup Score (ES_t) -- a metric unifying correctness, stability, and performance -- with layered integrity defenses against systematic LLM exploitation. Experiments reveal that PassBench is both highly discriminative and genuinely unsaturated: the best frontier model trails TorchInductor by 37% in aggregate, yet on individual subgraphs LLMs achieve up to 3x speedup over the same compiler -- indicating that the bottleneck is consistency, not capability. Fine-tuning a small model on merely ~4K PassNet trajectories yields a 2.67x improvement approaching frontier-model performance, demonstrating substantial headroom and validating PassNet as live training infrastructure for advancing LLM-driven compiler optimization. All data, benchmarks, and tooling are publicly available.
【28】Reasoning-preserved Efficient Distillation of Large Language Models via Activation-aware Initialization
标题：通过激活感知收件箱对大型语言模型进行推理保留的高效提炼
链接：https://arxiv.org/abs/2605.29327
摘要：Efficient Distillation (EDistill) compresses large language models (LLMs) by structured pruning parameters and tuning lightweight modules with high training efficiency. Although these EDistilled LLMs achieve state-of-the-art (SOTA) performance on general ability benchmarks relative to similarly sized LLMs, we identify a severe degradation in their multi-step reasoning ability, which we term reasoning collapse. We systematically analyze the geometric origins of reasoning collapse and show that the SOTA EDistill method based on width-reducing projection matrices suffers from eRank collapse, in which the effective rank (eRank) of hidden representations drops. We theoretically explain how singular values of randomly initialized projection matrices become unevenly distributed, leading to eRank collapse and thus token indistinguishability. To address this issue, we propose RED (Reasoning-preserved Efficient Distillation) for LLMs, which introduces activation-aware initialization to initialize projection matrices as channel-selection matrices, thus theoretically mitigating eRank collapse. Experiments on Llama and Qwen series demonstrate that RED substantially recovers reasoning while maintaining high training efficiency and SOTA general ability.
【29】CoHyDE: Iterative Co-Training of LLM Rewriter & Dense Encoder for Tool Retrieval
标题：CoHyDE：用于工具检索的LLM重写器和密集编码器的迭代联合训练
链接：https://arxiv.org/abs/2605.29271
摘要：
【30】DenseSteer: Steering Small Language Models towards Dense Math Reasoning
标题：DenseSteer：引导小型语言模型走向密集数学推理
链接：https://arxiv.org/abs/2605.29247
备注：ICML 2026
摘要：
【31】Implicit Identity Technologies for LLMs: Fingerprinting and Watermarking across Datasets, Models, and Generated Content
标题：LLM的隐式身份技术：跨数据集、模型和生成内容的指纹和水印
链接：https://arxiv.org/abs/2605.29245
备注：Accepted by IJCAI-ECAI 2026. 11 pages, 1 figure. Survey and taxonomy of LLM fingerprinting and watermarking for identity, provenance, generated-content attribution, and asset protection
摘要：
【32】BlockBatch: Multi-Scale Consensus Decoding for Efficient Diffusion Language Model Inference
标题：BlockBlock：用于高效扩散语言模型推理的多尺度共识解码
链接：https://arxiv.org/abs/2605.29233
备注：23 pages, including references and appendices
摘要：
【33】Inferring the Size of Large Language Models From Popular Text Memorization
标题：从流行的文本序列化推断大型语言模型的大小
链接：https://arxiv.org/abs/2605.29223
摘要：
【34】Influence-Guided Symbolic Regression: Scientific Discovery via LLM-Driven Equation Search with Granular Feedback
标题：影响引导符号回归：通过LLM驱动的方程搜索和颗粒反馈进行科学发现
链接：https://arxiv.org/abs/2605.29184
备注：ICML 2026
摘要：
【35】RUBRIC-ARROW: Alternating Pointwise Rubric Reward Modeling for LLM Post-training in Non-verifiable Domains
标题：RUBRIC-ARROW：不可验证领域LLM后训练的交替逐点Ruby奖励建模
链接：https://arxiv.org/abs/2605.29156
摘要：
【36】Apertus LLM Family Expansion via Distillation and Quantization
标题：Atlantus通过蒸馏和量化扩展LLM家族
链接：https://arxiv.org/abs/2605.29128
摘要：
【37】ReasonBreak: Probing Vulnerabilities in Reasoning-Enabled Vision-Language-Action Models for Autonomous Driving
标题：ReasonBreak：探索自动驾驶的推理支持视觉-语言-动作模型中的漏洞
链接：https://arxiv.org/abs/2605.29114
摘要：
【38】OISD: On-Policy Internal Self-Distillation of Language Models
标题：OISD：语言模型的政策内部自我提炼
链接：https://arxiv.org/abs/2605.29089
备注：Under Review for Publication
摘要：
【39】Knowledge Offloading: Decomposing LLMs into Sparse Backbones and Memory Modules
标题：知识卸载：将LLM分解为稀疏主干和内存模块
链接：https://arxiv.org/abs/2605.29075
摘要：
【40】Measuring Real-World Prompt Injection Attacks in LLM-based Resume Screening
标题：在基于LLM的简历筛选中测量现实世界的即时注入攻击
链接：https://arxiv.org/abs/2605.28999
备注：Published in USENIX Security Symposium 2026; Code and artifacts are available at this https URL
摘要：
【41】CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models
标题：CosmicFish-HRM：紧凑语言模型中基于层次递归机制的自适应推理
链接：https://arxiv.org/abs/2605.28919
备注：17 pages, 4 figures. Exploratory study of adaptive reasoning depth in compact autoregressive language models. Code available at this https URL
摘要：
【42】When LLM Reward Design Fails: Diagnostic-Driven Refinement for Sparse Structured RL
标题：当LLM奖励设计失败时：稀疏结构RL的诊断驱动的精化
链接：https://arxiv.org/abs/2605.28918
摘要：
【43】Feature Geometry of LoRA Adapters: A Sparse Autoencoder Analysis of Representational Divergence in Fine-Tuned Language Models
标题：LoRA适配器的特征几何：精调语言模型中表示性分歧的稀疏自动编码器分析
链接：https://arxiv.org/abs/2605.28896
摘要：
【44】Continuity and Ordinality Matter: Constraining Time Series Tokens for Effective Time Series Analysis with Large Language Models
标题：连续性和有序性很重要：用大型语言模型进行有效时间序列分析的限制时间序列标记
链接：https://arxiv.org/abs/2605.28866
摘要：
【45】Large language models reorganize representational geometry during in-context learning
标题：大型语言模型在上下文学习期间重新组织表示几何
链接：https://arxiv.org/abs/2605.28854
摘要：
【46】Representation Signatures and Risk-Feedback Alignment in LLM Trading Agents
标题：LLM贸易代理中的表示签名和风险反馈一致性
链接：https://arxiv.org/abs/2605.28850
摘要：
【47】RightNow-Arabic-0.5B-Turbo: An Open Sub-1B Arabic Language Model via Vocabulary Injection and Edge-First Deployment
标题：RightNow-Arabic-0. 5 B-Turbo：通过词汇注入和边缘优先部署的开放Sub-1B阿拉伯语言模型
链接：https://arxiv.org/abs/2605.28827
备注：12 pages, 7 tables, 4 figures, 1 algorithm. Weights: this https URL
摘要：
【48】Anytime-Valid Federated Conformal RAG for LLM Swarms
标题：LLM群的任意时间有效联邦共形RAG
链接：https://arxiv.org/abs/2605.29139
摘要：Federated Conformal RAG (FC-RAG) provides distribution-free coverage for a bandwidth-limited swarm of weak language models, but only at a fixed horizon. We extend it to anytime-valid sequential coverage: validity at every stopping time, preserved under predictable adaptive control (recalibration, per-node bandwidth escalation, distilled-student refresh), at no extra cost in assumptions over fixed-horizon FC-RAG. Naive composition fails because FC-RAG's marginal coverage bound makes the betting e-process a non-supermartingale on adverse calibration draws, and Ville's inequality cannot be invoked. We give Anytime-FC-RAG, a sequential extension built on a summable per-step calibration-deviation budget that converts the marginal bound into a strict conditional bound on a calibration-good event, paired with a truncated betting e-process that is a nonnegative supermartingale on the entire probability space. From these two ingredients, we obtain four guarantees: time-uniform alarm validity $\mathbb{P}(\sup_t E_t \ge 1/δ_e) \le δ_e + δ_{\mathrm{cal}}$, a Hoeffding-stitched cumulative-miscoverage envelope at the same total budget, safety under any predictable controller (recalibration, bandwidth escalation, student refresh), and training-side error propagation across an unbounded sequence of Federated Probe-Logit Distillation (FPLD) refreshes via a summable training budget. As a practical consequence, an adaptive controller that escalates retrieval bandwidth only when the e-process crosses a warning threshold matches the alarm rate of a fixed-high-bandwidth schedule at substantially lower communication cost. Experiments on a GPT-2-small + MiniLM swarm across MMLU, DBpedia, and AG News verify the predicted alarm rate, detection delay, envelope coverage, and $14$-$57\%$ bandwidth savings; the alarm fires when and only when coverage genuinely breaks.
Graph相关(图学习|图神经网络|图优化等)(16篇)
【1】SchGen: PCB Schematic Generation with Semantic-Grounded Code Representations
标题：SchGen：使用基于语义的代码表示生成PCB原理图
链接：https://arxiv.org/abs/2605.30345
备注：19 pages, 7 figures
摘要：Printed circuit board (PCB) schematic design defines nearly all electronic hardware, but it remains manual and expertise-intensive. While generative AI has advanced digital and analog IC design, PCB schematic generation from natural-language intent is largely unexplored. This paper presents SchGen, the first large language model that generates editable PCB schematics from natural-language requests. The key challenge lies in the lack of an LLM-suited representation and a large-scale dataset. Current schematic formats are dominated by verbose, tool-specific syntax and geometry-heavy descriptions, making them difficult to generate reliably. We introduce a semantically grounded code representation that encodes schematic editing primitives with relative placement and pin-name-based wiring, transforming a geometry-driven generation problem into a semantics-driven matching task amenable to LLMs. We further construct a large-scale dataset of PCB schematics paired with user prompts via a human-agent collaborative pipeline that converts open-source hardware designs into our representation. Experiments show that SchGen significantly outperforms alternative representations and even larger general-purpose LLMs on wire connectivity accuracy and functional correctness. Our results highlight the critical role of representation design in enabling generative models for complex hardware design tasks.
【2】GRASP: Plan-Guided Graph Retrieval with Adaptive Fusion and Reranking on Semi-Structured Knowledge Bases
标题：GRASP：具有自适应融合和半结构化知识库重新排序的计划引导图检索
链接：https://arxiv.org/abs/2605.30237
摘要：
【3】iLoRA: Bayesian Low-Rank Adaptation with Latent Interaction Graphs for Microbiome Diagnosis
标题：iLoRA：用于微生物组诊断的Bayesian低等级适应和潜在相互作用图
链接：https://arxiv.org/abs/2605.30179
备注：Accepted at ICML 2026
摘要：
【4】SAHG: Sector-Anisotropic Hyperbolic Graph Model for Social Bot Detection
标题：SAHG：用于社交机器人检测的扇形各向异性双曲图模型
链接：https://arxiv.org/abs/2605.30166
摘要：
【5】From Short Histories to Long Futures: Horizon-Aware Graph Neural Networks for Long Horizon Forecasting
标题：从简史到长期未来：用于长期预测的地平线感知图神经网络
链接：https://arxiv.org/abs/2605.29952
备注：Accepted for International Conference on Pattern Recognition (ICPR) 2026
摘要：
【6】A Triple-Modal Contrastive Learning Framework with Sequence, Graph, and 3D Features for Drug-Target Interaction Prediction
标题：具有序列、图形和3D特征的三模式对比学习框架，用于药物与目标相互作用预测
链接：https://arxiv.org/abs/2605.29926
备注：12 pages, 5 figures, ISBRA 2026
摘要：
【7】When Do Graph Foundation Models Transfer? A Data-Centric Theory
标题：图基础模型何时转移？数据中心理论
链接：https://arxiv.org/abs/2605.29828
备注：21 pages, including appendix. Accepted at ICML 2026
摘要：
【8】Gated Graph Attention Networks with Learnable Temperature
标题：具有可学习温度的门控图注意力网络
链接：https://arxiv.org/abs/2605.29803
摘要：
【9】Temporal Motif-aware Graph Test-time Adaptation for OOD Blockchain Anomaly Detection
标题：用于OOD区块链异常检测的时间图案感知图测试时间自适应
链接：https://arxiv.org/abs/2605.29526
备注：Accepted to IJCAI-ECAI 2026, Special Track on AI for Social Good
摘要：
【10】Composing Non-Conjugate Factor Graphs with Closed-Form Variational Inference
标题：用闭型变分推理构造非共轭因子图
链接：https://arxiv.org/abs/2605.29467
摘要：
【11】Forget Less, Generalize More: Unifying Temporal and Structural Adaptation for Dynamic Graphs
标题：少忘记，多概括：统一动态图的时间和结构适应
链接：https://arxiv.org/abs/2605.29453
摘要：
【12】Traditional machine learning vs. deep learning from dynamic graph representations of proteins' 3D folds in the task of protein structure classification
标题：在蛋白质结构分类任务中，传统机器学习与来自蛋白质3D折叠动态图表示的深度学习
链接：https://arxiv.org/abs/2605.29228
备注：Main paper: 16 pages, 4 figures, and 1 table; Supplementary information: 13 pages, 9 figures
摘要：
【13】Better Later Than Sooner: Neuro-Symbolic Knowledge Graph Construction via Ontology-grounded Post-extraction Correction
标题：迟来好：通过基于实体的提取后纠正构建神经符号知识图谱
链接：https://arxiv.org/abs/2605.29168
摘要：
【14】Evolutionary Refinement of Generative Graph Topologies: A Hybrid WGAN-GA Approach
标题：生成图的进化细化：混合WGAN-GA方法
链接：https://arxiv.org/abs/2605.29161
备注：6 pages, 4 Figures, 4 Tables, IEEE World Congress on Computational Intelligence
摘要：
【15】Designing Active Tether-Net Systems for Space Debris Capture with Graph-Learning-Aided Mixed-Combinatorial Optimization
标题：利用图形学习辅助混合组合优化设计空间碎片捕获主动系绳网络系统
链接：https://arxiv.org/abs/2605.29021
备注：Accepted for presentation at 2026 AIAA Aviation Forum
摘要：
【16】LoRe: Adaptive Interaction-Evaluation Routing with Per-Step Interaction Budgets for Iterative Graph Solvers
标题：LoRe：具有迭代图求解器逐步交互预算的自适应交互评估路由
链接：https://arxiv.org/abs/2605.29005
备注：Accepted at ICML 2026
摘要：
Transformer(3篇)
【1】Anti Mode-Collapse in Mean-Field Transformer via Auxiliary Variables
标题：通过辅助变量防止平均场Transformer中的模式崩溃
链接：https://arxiv.org/abs/2605.30229
备注：39 pages
摘要：
【2】FHRFormer: A Self-Supervised Masked Transformer Framework for Fetal Heart Rate Time-Series Inpainting and Forecasting
标题：FHRFormer：用于胎儿心率时间序列修复和预测的自我监督屏蔽Transformer框架
链接：https://arxiv.org/abs/2605.29695
备注：Submitted to Frontiers in Digital Health. arXiv admin note: substantial text overlap with arXiv:2509.20852
摘要：Approximately 10% of newborns require assistance to initiate breathing at birth, and around 5% need ventilation support. Fetal heart rate (FHR) monitoring plays a crucial role in assessing fetal well-being during prenatal care, enabling the detection of abnormal patterns and supporting timely obstetric interventions to mitigate fetal risks during labor. Applying artificial intelligence (AI) methods to analyze large datasets of continuous FHR monitoring episodes with diverse outcomes may offer novel insights into predicting the risk of needing breathing assistance or interventions. Recent advances in wearable FHR monitors have enabled continuous fetal monitoring without compromising maternal mobility. However, sensor displacement during maternal movement, as well as changes in fetal or maternal position, often lead to signal dropout, resulting in gaps in recorded FHR data. Such missing data limits the extraction of meaningful insights and complicates automated (AI-based) analysis. Traditional approaches to handling missing data, such as simple interpolation techniques, often fail to preserve the spectral characteristics of the signals. In this paper, we propose a masked transformer-based autoencoder approach to reconstruct missing FHR signals by capturing both local temporal and frequency components of the data. The proposed method demonstrates robustness across varying durations of missing data and can be used for signal inpainting and forecasting. The proposed approach can be applied retrospectively to research datasets to support the development of AI-based risk algorithms. In the future, the proposed method could be integrated into wearable FHR monitoring devices to achieve earlier and more robust risk detection.
【3】Relational Rank Geometry in Transformers: Detecting and Steering Hidden-State Relation Frames
标题：Transformer中的关系等级几何：检测和引导隐藏状态关系框架
链接：https://arxiv.org/abs/2605.29634
备注：32 pages, 9 figures
摘要：
GAN|对抗|攻击|生成相关(12篇)
【1】On Language Generation in the Limit with Bounded Memory
标题：有限记忆极限下的语言生成
链接：https://arxiv.org/abs/2605.30324
备注：The abstract has been shortened to fit within the arXiv limit
摘要：We study language generation in the limit under bounded memory. In this task, a learner observes examples from an unknown target language one at a time and must eventually output only new valid examples. Prior work assumes access to the entire history, a strong assumption since realistic algorithms retain limited past information. Classical work in learning theory shows memory constraints dramatically alter learnability; we extend this to language generation. First, we study memoryless generators. Under a mild enumeration restriction, every countable collection of infinite languages remains generable without memory. Without this restriction, we exactly characterize when memoryless generation is possible. For finite collections, we characterize the optimal minimax density achievable by memoryless generators -- the best density guaranteed against any collection of a given size. This combinatorial bound relies on Sperner's theorem and symmetric chain decompositions. We further show that a sliding window of the last $W$ examples does not improve this worst-case density, whereas allowing it to store $b$ adaptively chosen past examples improves the achievable density for every $b \geq 1$. Finally, we revisit identification in the limit, where the learner must converge to a single correct hypothesis for the target language. We focus on its incremental variant, where the learner remembers only its previous guess. Here, although exact identification fails on a collection of just three languages, a mild relaxation requiring convergence to an ``approximate'' version of the target is achievable for every finite collection. These results show bounded memory affects these tasks differently: generation remains achievable for every countable collection, while density and identification are confined to finite collections, with guarantees weakening as the collection grows.
【2】Token-Level Generalization in LoRA Adapter Backdoors: Attack Characterization and Behavioral Detection
标题：LoRA适配器后门中的令牌级概括：攻击特征和行为检测
链接：https://arxiv.org/abs/2605.30189
备注：45 pages, 27 tables. Code and evaluation data: this https URL. Trained adapter weights available on request
摘要：
【3】Improving Adversarial Robustness of Attribution via Implicit Regularization
标题：利用隐正则化提高属性的对抗鲁棒性
链接：https://arxiv.org/abs/2605.29983
备注：39 pages, 22 figures, to be published in International Conference on Machine Learning 2026
摘要：
【4】BuilDyn: Excitation-Driven Data Generation for Building Thermal Dynamics Modeling and Control
标题：Building Dyn：用于建筑热动力学建模和控制的激励驱动数据生成
链接：https://arxiv.org/abs/2605.29849
摘要：
【5】Convex Basins in Single-Index Model Loss Landscapes: Applications to Robust Recovery under Strong Adversarial Corruption
标题：单指数模型损失景观中的凸盆地：强对抗腐败下稳健恢复的应用
链接：https://arxiv.org/abs/2605.29497
备注：Accepted at ICML 2026
摘要：
【6】A Full-Pipeline Framework for Evaluating Membership Inference Attacks in Machine Learning
标题：评估机器学习中成员推断攻击的全管道框架
链接：https://arxiv.org/abs/2605.29454
摘要：
【7】Harmless Yet Harmful: Neutral Prompting Attacks for Stealthy Hallucination Steering in Agent Skills
标题：无害但又有害：中立的闪烁攻击，以实现特工技能中的隐形幻觉引导
链接：https://arxiv.org/abs/2605.29354
备注：under review
摘要：LLM-powered coding agents increasingly participate in software development workflows by generating code, selecting dependencies, and producing package installation commands. This creates a new software supply chain risk: when an agent hallucinates a non-existent package, an attacker may register the hallucinated name and later compromise users who install it. Existing package hallucination attacks and defenses primarily focus on naturally occurring hallucinations, targeted dependency steering, or post-hoc package validation. In this paper, we introduce \emph{Neutral Prompting Attack} (NPA), a highly stealthy attack paradigm in which semantically benign instructions, such as encouraging imagination and exhaustiveness, increase package hallucination propensity without containing explicit malicious intent. Unlike targeted dependency steering, NPA does not specify an attacker-chosen package. Instead, it shifts the model's dependency generation behavior toward more speculative package names. We evaluate NPA across multiple coding-oriented LLMs and package hallucination benchmarks. Our results show that NPA increases both \emph{Hallucination ASR} and \emph{Pip Install ASR}, changes the distribution of hallucinated package names, and evades existing static-analysis, LLM-based, and agent-based Skill defenses. These findings reveal that harmless-looking prompts can covertly manipulate hallucination behavior and create downstream software supply chain risks.
【8】Robust Frequency-Calibrated Virtual EEG Channel Generation from Four Frontal Electrodes for Wearable EEG Augmentation
标题：从四个正面电极生成稳健的频率校准虚拟脑电通道，用于可穿戴脑电增强
链接：https://arxiv.org/abs/2605.29263
备注：17 pages, 4 figures
摘要：
【9】Stochastic Lifting for Generating Trajectories of Stochastic Physical Systems
标题：生成随机物理系统轨迹的随机提升
链接：https://arxiv.org/abs/2605.29194
摘要：
【10】Cycle-Space Informed Detection of Autoencoded Blind False Data Injection Attacks on Power Systems
标题：电力系统自动编码盲假数据注入攻击的周期空间知情检测
链接：https://arxiv.org/abs/2605.28912
备注：13 pages, 11 figures
摘要：
【11】PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation
标题：PrismFlow：时间序列生成中流匹配的剩余动态
链接：https://arxiv.org/abs/2605.28867
摘要：
【12】Neural Scaling Laws for Jet Generation
标题：喷气产生的神经标度定律
链接：https://arxiv.org/abs/2605.28940
摘要：Recently observed empirical scaling laws describe the performance of foundation-type models as three independent key quantities -- dataset size, compute, and model parameters -- are modified. Extracting these scaling laws informs the training of large complex models for which the tuning of hyperparameters in traditional ways is not feasible. This work for the first time explores if scaling laws can also be observed for the task of particle jet generation -- both relevant as a pre-training objective for foundation models and as in-situ simulation by itself. We indeed replicate the key logarithmic scaling law behavior for model-size scaling. Beyond studying the next token prediction validation loss of the generative model, we also study the sliced Wasserstein distance of five physical quantities that are not immediately available to the model during training. Our study shows that this quantity is monotonically related to the next token prediction validation loss, meaning that this loss is indeed a good proxy for the physics performance. For the scaling with dataset size and compute, we observe substantially weaker scaling behavior of both the loss and the sliced Wasserstein distance. We analyze this behavior by introducing the concept of a learnable window, and argue that autoregressive next token prediction on jet constituents exhibits comparatively rapid saturation relative to language-model studies. We discuss possible origins of this behavior, including the stochastic nature of QCD radiation and differences between generative and supervised learning tasks in collider physics.
半/弱/无/有监督|不确定性|主动学习(10篇)
【1】Active Continual Learning with Metaplastic Binary Bayesian Neural Networks
标题：利用元可塑二元Bayesian神经网络进行主动连续学习
链接：https://arxiv.org/abs/2605.30198
备注：Accepted at ICML 2026
摘要：
【2】Test Time Training for Supervised Causal Learning
标题：监督因果学习的测试时间训练
链接：https://arxiv.org/abs/2605.30015
摘要：
【3】SAAS: Self-Aware Reinforcement Learning for Over-Search Mitigation in Agentic Search
标题：SAAS：用于减少搜索过度的自感知强化学习
链接：https://arxiv.org/abs/2605.29796
摘要：
【4】On-Policy Replay for Continual Supervised Fine-Tuning
标题：政策重演以持续监督的微调
链接：https://arxiv.org/abs/2605.29495
摘要：
【5】Return-to-Go Is More Than a Number: Q-Guided Alignment for Return-Conditioned Supervised Learning
标题：回归不仅仅是一个数字：回归条件监督学习的Q引导对齐
链接：https://arxiv.org/abs/2605.29028
备注：28 pages, 13 figures, 20 tables, accepted by ICML 2026
摘要：
【6】Learning Robust and Task-Invariant Functional Representation from fMRI through Siamese Self-Supervised Learning
标题：通过Siamese自我监督学习从fMRI学习稳健且任务不变的功能表示
链接：https://arxiv.org/abs/2605.28990
摘要：
【7】Conf-Gen: Conformal Uncertainty Quantification for Generative Models
标题：Conf-Gen：生成模型的保形不确定性量化
链接：https://arxiv.org/abs/2605.28920
备注：ICML 2026
摘要：
【8】Emergent Semantic Representations in World Models through Physical Interaction without Linguistic Supervision
标题：在没有语言监督的情况下，通过物理交互在世界模型中产生语义表示
链接：https://arxiv.org/abs/2605.28865
备注：10 pages, 3 figures
摘要：
【9】Self-Play Reinforcement Learning under Imperfect Information in Big 2
标题：Big 2中不完美信息下的自我强化学习
链接：https://arxiv.org/abs/2605.28863
备注：11 pages
摘要：
【10】A new completely parameter-free clustering algorithm for unsupervised classification of BATSE gamma-ray bursts
标题：一种新的完全无参数的集群算法，用于无监督分类BAT γ射线暴
链接：https://arxiv.org/abs/2605.30175
摘要：Cluster analysis is a widely applied machine learning technique to understand the existing patterns in the population of gamma-ray bursts (GRBs), in order to explore their physical sources. In the present scenario, the number of clusters corresponding to differentiable groups is still under conflict, in spite of numerous attempts with the state-of-the-art clustering procedures. This crucial unknown parameter needs to be evaluated, either directly or indirectly in terms of other tuning parameters, to produce the clusters in GRBs through implementation of an appropriate clustering algorithm. While most of the applied algorithms reached two physically explained groups of merger and collapsar predominated by the short and long bursts respectively, other statistical approaches violated this binary partition. However, physical establishment of any additional cluster(s) is not yet confirmed. Therefore, we propose a new algorithm, from a different stream of clustering referred to as `completely parameter-free', which carries out the classification of GRBs in a manner that has not been tried so far. It indicates two main groups, of short and long duration bursts from the BATSE sample, compatible with the merger-collapsar theory.
迁移|Zero/Few/One-Shot|自适应(12篇)
【1】In-Context Reward Adaptation for Robust Preference Modeling
标题：用于稳健偏好建模的上下文奖励自适应
链接：https://arxiv.org/abs/2605.30323
摘要：Reinforcement Learning from Human Feedback (RLHF) typically relies on static reward models to align Large Language Models with human preferences. However, human values are inherently diverse and heterogeneous, and a single reward model often lacks the robustness required to generalize to unseen preference domains. While existing multi-reward frameworks attempt to address this, they are often restricted to a fixed set of known domains and fail to adapt to unseen human distributions without costly retraining. In this work, we propose In-Context Reward Adaptation, a transformer-based framework designed to model diverse and unseen human preferences on the fly. By leveraging the in-context learning capabilities of transformers, our approach adaptively infers the underlying reward structure from a small set of preference demonstrations. We demonstrate that while a standard transformer architecture is insufficient for this task by characterizing an asymptotic bias to the ground-truth, incorporating human response time as an auxiliary input signal enables the model to successfully adapt to preferences from previously unseen domains. Our findings show that this approach provides a more robust foundation for preference modeling, allowing for the representation of heterogeneous rewards and preference distribution shift, and offering a scalable path toward more flexible human-AI alignment.
【2】MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment
标题：MC：通过各向同性子空间对齐最大化自适应表示中的信息容量
链接：https://arxiv.org/abs/2605.29987
备注：Accepted at the GlobalSouthML Workshop at ICML 2026. 13 pages, 2 figures
摘要：
【3】Gesture-Aware Indoor THz ISAC Systems for Adaptive Resource Allocation
标题：用于自适应资源分配的手势感知室内太赫兹ISAC系统
链接：https://arxiv.org/abs/2605.29913
备注：6 pages, 4 figures, conference(Submitted to PIMRC)
摘要：
【4】MIRAGE: Adaptive Multimodal Gating for Whole-Brain fMRI Encoding
标题：MISYS：用于全脑fMRI编码的自适应多模式门控
链接：https://arxiv.org/abs/2605.29850
备注：Preprint. First two author contributed equally
摘要：
【5】Cert-LAS: Toward Certified Model Ownership Verification for Text-to-Image Diffusion Models via Layer-Adaptive Smoothing
标题：Cert-LAS：通过层自适应平滑实现文本到图像扩散模型的认证模型所有权验证
链接：https://arxiv.org/abs/2605.29809
备注：This paper has been accepted to the International Conference on Machine Learning (ICML) 2026. 26 pages
摘要：
【6】Gradient Perturbation: Learning to Perturb Gradients for Adaptive Training
标题：梯度扰动：学习扰动自适应训练
链接：https://arxiv.org/abs/2605.29494
摘要：
【7】Deep Adaptive Dimension Reduction for Bayesian Inference in Inverse Problems
标题：反问题中Bayesian推理的深度自适应降维
链接：https://arxiv.org/abs/2605.29373
备注：25 pages, 5 figures
摘要：Solving high-dimensional PDE-governed inverse problems is often challenging due to complex non-Gaussian posterior distributions, expensive forward model evaluations, and misspecified prior information. To address these issues, we propose a deep adaptive dimension-reduction Bayesian inference framework based on the Variational Flow (VF) model. Since standard normalizing flows are restricted by bijective mappings and cannot directly reduce dimensions, VF overcomes this limitation by integrating VAE-based nonlinear dimension reduction with dual normalizing flows for the latent prior and encoder. This design provides a strictly higher evidence lower bound than VAE and allows more flexible approximation of complex posterior distributions. We further introduce an iterative prior updating strategy that gradually moves the prior mean toward high-probability posterior regions, avoiding manual prior tuning. These components form a closed adaptive loop together with an adaptively fine-tuned Fourier Neural Operator (FNO) surrogate: VF generates posterior-concentrated samples to refine the surrogate, while the updated surrogate further improves posterior inference. Numerical experiments on a 100-dimensional Rosenbrock problem and three standard PDE-governed inverse problems show that our method delivers competitive or superior accuracy compared with MCMC, UKI, and SVGD baselines across all tested configurations, with the most pronounced advantages emerging in challenging scenarios such as high-noise observations and high-dimensional parameter spaces.
【8】A Theoretical and Experimental Study of a Novel Adaptive Learning Algorithm
标题：一种新型自适应学习算法的理论与实验研究
链接：https://arxiv.org/abs/2605.29273
摘要：
【9】TIMEGATE: Sustainable Time-Boxed Promotion Gates for Continual ML Adaptation Under Resource Constraints
标题：TIMEGATE：在资源限制下持续ML适应的可持续时间限制促销门
链接：https://arxiv.org/abs/2605.29183
摘要：
【10】Parallel Adaptive Multi-Objective Evolutionary Learning of Discretized Bayesian Network Classifiers for Clinical Data
标题：临床数据离散化Bayesian网络分类器的并行自适应多目标进化学习
链接：https://arxiv.org/abs/2605.29058
摘要：
【11】Adapting Automotive Aerodynamics Surrogates to New Vehicle Families via Transfer Learning
标题：通过迁移学习使汽车空气动力学替代品适应新车系列
链接：https://arxiv.org/abs/2605.27968
备注：23 pages, 12 figures
摘要：
【12】Deep Optimal Individualized Treatment Rules for Bivariate Survival Outcomes via Adaptive Prediction-Powered Learning
标题：通过自适应预测驱动学习为二元生存结局制定深度最优个体化治疗规则
链接：https://arxiv.org/abs/2605.29464
摘要：In randomized trials involving multiple treatments, bivariate survival outcomes present significant analytical challenges for making decisions. This paper addresses the problem of deriving optimal individualized treatment rules to maximize the joint survival probability beyond fixed time points $(t_1, t_2)$ through deep neural networks, while accounting for right censoring. We propose a novel approach that models treatment rules via stochastic policies, coupling marginal accelerated failure time models via link function to capture bivariate dependence. To enhance robustness and effectiveness of decision making, we introduce an adaptive prediction-powered method that leverages auxiliary predictions from machine learning models.
强化学习(9篇)
【1】On Distributional Reinforcement Learning in Chaotic Dynamical Systems
标题：关于混乱动力系统中的分布式强化学习
链接：https://arxiv.org/abs/2605.30160
摘要：
【2】RL2ML: Finite-Rollout Surrogate Objectives from Reinforcement Learning to Maximum Likelihood
标题：RL 2ML：从强化学习到最大可能性的临时推出替代目标
链接：https://arxiv.org/abs/2605.30154
摘要：
【3】Sample-Efficient Diffusion-based Reinforcement Learning with Critic Guidance
标题：具有批判性指导的样本高效的基于扩散的强化学习
链接：https://arxiv.org/abs/2605.30056
备注：accepted by ICML2026
摘要：
【4】PEARL: Training Socratic Tutors with Pedagogically Aligned Reinforcement Learning
标题：PEARL：通过教学一致的强化学习训练苏格拉底导师
链接：https://arxiv.org/abs/2605.29582
备注：16 pages, 7 figures
摘要：
【5】Information-Directed Offline-to-Online Reinforcement Learning
标题：信息导向的离线到在线强化学习
链接：https://arxiv.org/abs/2605.29405
摘要：
【6】Bridging the Sim-to-Real Gap in Reinforcement Learning-Based Industrial Dispatching through Execution Semantics
标题：通过执行语义弥合基于强化学习的工业调度中的模拟与真实差距
链接：https://arxiv.org/abs/2605.29078
备注：Accepted for publication at the 24th IEEE International Conference on Industrial Informatics (INDIN 2026), held from 26 to 29 July 2026 in Melbourne, Australia
摘要：
【7】Structured Prompt Optimization Meets Reinforcement Learning for Global and Local Interpretability over Complex Text
标题：结构化提示优化满足强化学习，实现复杂文本的全局和局部可解释性
链接：https://arxiv.org/abs/2605.29076
摘要：
【8】Label-Free Reinforcement Learning via Cross-Model Entropy
标题：通过跨模型熵的无标签强化学习
链接：https://arxiv.org/abs/2605.29009
摘要：
【9】FedQHD: Closed-Form Function-Space Federated Reinforcement Learning
标题：FedQHD：封闭式功能空间联邦强化学习
链接：https://arxiv.org/abs/2605.29002
摘要：
医学相关(6篇)
【1】Digitally enriching a screening population for pancreatic cancer using routine blood-based measures and clinical histories
标题：使用常规血液测量和临床史数字化丰富胰腺癌筛查人群
链接：https://arxiv.org/abs/2605.30275
摘要：
【2】EVL-ECG: Efficient ECG Interpretation With Multi-Aspect Heterogeneous Knowledge Distillation
标题：EVL-心电图：通过多方面异类知识提炼的高效心电图解释
链接：https://arxiv.org/abs/2605.29977
备注：Accepted at the SD4H Workshop at ICML 2026. 11 pages, 3 figures
摘要：
【3】A Fully Convolutional Approach to Denoising Structural Dynamics Data from X-Ray Photon Correlation Spectroscopy
标题：用完全卷积方法消除X射线质子相关光谱中的结构动力学数据
链接：https://arxiv.org/abs/2605.29975
摘要：
【4】Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression
标题：预测神经退行性疾病进展的治疗条件扩散
链接：https://arxiv.org/abs/2605.29932
备注：9 pages, 5 figures, 1 table
摘要：
【5】Why Specialist Models Still Matter: A Heterogeneous Multi-Agent Paradigm for Medical Artificial Intelligence
标题：为什么专家模型仍然重要：医学人工智能的异类多智能体范式
链接：https://arxiv.org/abs/2605.29744
备注：Accepted at ICML 2026. 12 pages main text, 16 pages appendix
摘要：
【6】Comparing Post-Hoc Explainable AI Methods for Interpreting Black-Box EEG Models in Depression Detection
标题：比较用于解释抑郁检测中黑匣子脑电模型的事后可解释人工智能方法
链接：https://arxiv.org/abs/2605.28977
摘要：
蒸馏|知识提取(5篇)
【1】SGMD: Score Gradient Matching Distillation for Few-Step Video Diffusion Distillation
标题：SGMD：用于几步视频扩散蒸馏的分数梯度匹配蒸馏
链接：https://arxiv.org/abs/2605.30116
备注：ICML 2026
摘要：
【2】A Predictive Law for On-Policy Self-Distillation From World Feedback
标题：来自世界反馈的政策上自我蒸馏的预测定律
链接：https://arxiv.org/abs/2605.30070
摘要：
【3】OptSkills: Learning Generalizable Optimization Skills from Problem Archetypes via Cluster-Based Distillation
标题：OptSkills：通过基于机器人的蒸馏从问题原型中学习可推广的优化技能
链接：https://arxiv.org/abs/2605.29829
备注：22 pages, 10 figuers, project: this https URL
摘要：
【4】Context Distillation as Latent Memory Management
标题：上下文蒸馏作为潜在内存管理
链接：https://arxiv.org/abs/2605.28889
摘要：
【5】Matching Rates and Optimal Allocation for Federated Probe-Logit Distillation under Heterogeneous Bandwidth Budgets
标题：不同带宽预算下联邦探针-Logit蒸馏的匹配率和最优分配
链接：https://arxiv.org/abs/2605.29642
摘要
：In federated language modeling, $K$ nodes each hold $n$ samples but cannot pool data or exchange full-precision gradients or weights. We study the minimax rate at which a conditional distribution over $V$ tokens can be estimated when each node may upload at most $B$ bits per query in a public probe set. In federated probe-logit distillation (FPLD), each node transmits a scalar-quantized logit vector on the probe set, and an aggregator distills a global parametric student. Prior work (Dubey and Huo, 2026) establishes a high-probability KL rate $O(d/(Kn) + ρ\sqrt{V \log V / m} + K^{-1} \cdot 2^{-2B/V})$ plus optimization slack, with the bandwidth term in its trace-sharpened form. Whether this bandwidth-term rate is tight, and how the upper bound generalizes to heterogeneous per-node bandwidths, are left open. We close both gaps. First, the dithered FPLD construction has a matching single-round lower bound $Ω(K^{-1} \cdot 2^{-2B/V})$ under non-degeneracy, pinning the bandwidth-axis rate at $Θ(K^{-1} \cdot 2^{-2B/V})$. $T$-round sequential refinement with nested/scaled residual quantizers achieves $O(K^{-1} \cdot 2^{-2TB/V})$; vanilla FPLD's $T$-independent bandwidth term is suboptimal for every $T > 1$. Second, we establish a heterogeneous-bandwidth upper bound for per-node budgets $B_i$, paired with a closed-form optimal allocation $B_i^* = B_{\mathrm{tot}}/K + (V/2) \log_2(w_i / \bar{w}_g)$, a log-tilted water-filling rule that is the per-node analogue of reverse water-filling for distortion-rate optimization. A plug-in adaptive variant estimates the weights from a short warm-up phase and attains $1 + O(\sqrt{\log(K/δ)/(m T_0)})$ relative suboptimality. Synthetic n-gram simulations confirm that empirical KL is bracketed by the upper and lower bounds and that the optimal allocation strictly dominates uniform and inverse-weighted baselines under heterogeneous clipping.
推荐(2篇)
【1】LoopFM: Learning frOm HistOrical RePresentations of Foundation Model for Recommendation
标题：LoopFM：从推荐基础模型的组织学表示中学习
链接：https://arxiv.org/abs/2605.29280
备注：Shali Jiang, Hua Zheng, Boyang Liu contributed equally to this work
摘要：
【2】Generative Spatiotemporal Intent Sequence Recommendation via Implicit Reasoning in Amap
标题：通过Amap中隐式推理的生成性时空意图序列推荐
链接：https://arxiv.org/abs/2605.28888
备注：9 pages, 1 figure
摘要：
聚类(2篇)
【1】CLUBench: A Clustering Benchmark
标题：CLUBench：集群基准
链接：https://arxiv.org/abs/2605.29933
摘要：
【2】MōLe-Λ: Learning the Coupled-Cluster Response State for Energies, Gradients, and Properties
标题：MōLe-Λ：学习能量，能量和性质的耦合簇响应状态
链接：https://arxiv.org/abs/2605.29622
备注：ICML 2026 AI4Physics
摘要：
超分辨率|去噪|去模糊|去雾(1篇)
【1】EMAG: Differentiable 4D Gaussian Mixture Splatting for EEG Spatial Super-Resolution
标题：EMAG：可区分4D高斯混合Split实现脑电空间超分辨率
链接：https://arxiv.org/abs/2605.29731
摘要：
自动驾驶|车辆|车道检测等(3篇)
【1】TraceCodec: A Compiler-Backed Neural Codec for Stateful Multi-Flow Network Traffic Traces
标题：TraceCodec：一个用于状态多流网络流量跟踪的操作员支持的神经编解码器
链接：https://arxiv.org/abs/2605.29941
摘要：
【2】Momentum Based Reward Design for Low Emission Traffic Signal Control
标题：基于动量的低排放交通信号控制奖励设计
链接：https://arxiv.org/abs/2605.29693
摘要：Urban traffic congestion is a growing global issue contributing significantly to long commute times and environmental pollution. Traditional traffic signal control systems often fail to adapt to dynamic traffic conditions. Adaptive traffic signal control can improve urban traffic without changing road infrastructure. Deep Reinforcement Learning (DRL) has shown strong performance for this task, but existing delay and queue-based rewards often produce short-sighted or unstable policies. This paper proposes a Momentum-Based Reward Function (MBRF) that encourages vehicles to keep moving rather than penalizing congestion alone. The method is evaluated in SUMO (Simulation of Urban MObility) using standard traffic metrics such as waiting time, queue length, throughput, and CO2 emissions. Results show that the proposed reward produces better throughput-emission trade-offs and more stable learning behavior than delay or queue-based rewards, as well as classical controllers such as Max Pressure and LQF.
【3】Multi-Resolution End-to-End Deep Neural Network for Optimizing Latency-Accuracy Tradeoff in Autonomous Driving
标题：多分辨率端到端深度神经网络用于优化自动驾驶中的延迟-准确性权衡
链接：https://arxiv.org/abs/2605.29138
备注：ICCPS 2026
摘要：
点云|SLAM|雷达|激光|深度RGBD相关(2篇)
【1】BioRefusalAudit: Auditing Biosecurity Refusal Depth Using General and Domain-Fine-Tuned Sparse Autoencoders
标题：BioRefusion审核：使用通用和领域微调稀疏自动编码器审核生物安全拒绝深度
链接：https://arxiv.org/abs/2605.30162
备注：21 pages, 2 figures, 3 tables. Apart Research AIxBio Sprint hackathon paper, April 2026 (Track 3: AI Biosecurity Tools). Code, eval set, and SAEs: this http URL. Reviewer feedback: this http URL
摘要：
【2】Compute Allocation in Evolutionary Search: From Depth-Breadth to Multi-Armed Bandits
标题：进化搜索中的计算分配：从深度广度到多臂盗贼
链接：https://arxiv.org/abs/2605.29268
摘要：
联邦学习|隐私保护|加密(3篇)
【1】Fairness-Aware Federated Learning with Trajectory Shapley Value
标题：基于轨迹Shapley值的公平感知联邦学习
链接：https://arxiv.org/abs/2605.30336
备注：Accepted for publication at the 24th European Control Conference (ECC 2026)
摘要
：Federated learning is an emerging distributed paradigm that addresses the challenges posed by heterogeneous, privacy-sensitive data. It enables multiple clients to train a model collaboratively by aggregating their local updates at a server. However, conventional aggregation schemes typically use fixed weights that fail to reflect unequal and time-varying client contributions, leading to biased and unstable learning. To improve fairness and stability, we propose the Trajectory Shapley Value (TSV), a contribution metric that evaluates how each client influences the optimization trajectory of the global model using a validation-based, temporally consistent utility. Building on TSV, we design FedTSV, an adaptive aggregation method that converts per-round evaluations into dynamic client weights, allowing the server to respond to heterogeneous and adversarial participation in real time. Experiments on benchmark datasets show that FedTSV accelerates convergence, improves robustness, and yields more equitable contribution assessments, thereby providing a principled foundation for fairness-aware federated optimization.
【2】Privacy-Enhanced Zero-Order Federated Learning via xMK-CKKS over Wireless Channels
标题：通过无线通道上的xMK-CKKS进行隐私增强零阶联邦学习
链接：https://arxiv.org/abs/2605.30123
备注：12 pages, 3 figures
摘要：
【3】Q-ANCHOR: Federated Quantum Learning with ZNE-guided Correction
标题：Q-ANCHOR：具有ZNE引导纠正的联邦量子学习
链接：https://arxiv.org/abs/2605.30075
摘要：
推理|分析|理解|解释(22篇)
【1】Reasoning with Sampling: Cutting at Decision Points
标题：抽样推理：在决策点进行切割
链接：https://arxiv.org/abs/2605.30327
摘要：Frontier reasoning models are produced by posttraining base language models with reinforcement learning. Recent work has challenged this by showing that sampling from a sharpened version of the base model's distribution, a so-called power distribution, elicits comparable reasoning without additional training, curated datasets, or verifiers. However, making this method practical requires efficiently sampling from the power distribution. A sampler needs to "mix" to the power distribution, which necessitates moving between modes of the target distribution; intuitively, e.g., trying different reasoning strategies. The samplers proposed in prior works repeatedly select a "cut" position in the current reasoning trace uniformly at random and resample the suffix from that position onward. However, reasoning traces typically contain a few consequential decisions (e.g., the choice of proof strategy or algorithm), and we observe that a uniformly chosen cut tends to rewrite local details rather than revisit decision points. We introduce an algorithm (Entropy-Cut Metropolis-Hastings) that uses the base model's next-token entropy as a proxy to identify key decision points and resample from those positions. We empirically verify that entropy jumps are a useful proxy for decision points and, in a stylized model of reasoning, prove that our method's mixing time scales with the number of decisions in a trace rather than with the number of tokens, which can be much larger. Across MATH500, HumanEval, GPQA Diamond, and AIME26, our method consistently improves over baselines and RL-trained models.
【2】ExDBSCAN: Explaining DBSCAN with Counterfactual Reasoning -- Additional Material
标题：ExDBSCAN：用反事实推理解释DBSCAN--附加材料
链接：https://arxiv.org/abs/2605.30225
摘要：
【3】Can AI Weather Models Predict Beyond Two Weeks? A Quantitative Benchmark and Analysis of Long Rollouts
标题：人工智能天气模型可以预测两周以上吗？长期推出的量化基准和分析
链接：https://arxiv.org/abs/2605.30184
摘要：
【4】PARCEL: Pool-Anchored Resampling with Conditioned Elastic Queries for Efficient Vision-Language Understanding
标题：PARCEL：带条件弹性预设的池锚定重置，以实现高效的视觉语言理解
链接：https://arxiv.org/abs/2605.30126
备注：33 pages, 4 figures
摘要：
【5】Evolving Features vs Evolving Entire Trees with GP for Interpretable Survival Analysis
标题：进化特征与进化整棵树与全科医生进行可解释生存分析
链接：https://arxiv.org/abs/2605.30119
摘要：
【6】Distributionally Robust Set Representation Learning Under Inference-Time Element Corruption
标题：推理时间元素腐败下的分布鲁棒集合表示学习
链接：https://arxiv.org/abs/2605.30089
备注：Accepted by ICML'26
摘要：
【7】Conformal Certification of Reasoning Trace Prefixes
标题：推理痕迹前置的保形证明
链接：https://arxiv.org/abs/2605.30085
备注：Code available at this https URL
摘要：
【8】MuPHI: Learning Implicit Multimodal Harm Reasoning via Semantically Grounded Reward Optimization
标题：MuPHI：通过语义基础奖励优化学习内隐多模态伤害推理
链接：https://arxiv.org/abs/2605.29951
摘要：
【9】LaRA: Layer-wise Representation Analysis for Detecting Data Contamination in RL Post-Training
标题：LaRA：RL训练后检测数据污染的分层表示分析
链接：https://arxiv.org/abs/2605.29888
备注：Work in Progress
摘要：
【10】A Geometric View of SRC: Learning Representations for Stable Residual Inference
标题：SRC的几何视图：稳定剩余推理的学习表示
链接：https://arxiv.org/abs/2605.29673
备注：37 pages
摘要：Reconstruction-based inference assigns a class by comparing class-wise reconstruction residuals; Sparse Representation Classification (SRC) is a canonical instance whose reliability depends on the geometry of the learned representation. We adopt a strict training-inference separation: SRC is used only as a fixed test-time rule and is never differentiated, unrolled, or optimized during training. In a span-level idealization based on class-conditional spans and their associated projection residuals, we formalize residual-ordering stability through a residual margin and characterize geometric obstructions -- span overlap, dominance, and near-overlap via small principal angles -- that can collapse this margin in worst-case directions. This span-level theory is primary: it specifies when the idealized residual family is well-separated, and it provides a conditional solver-level interpretation for practical residual approximations (e.g., OMP) insofar as they remain close to the span-level residual ordering. Under explicit coverage and separation assumptions, we derive a quantitative lower bound on the (idealized) residual margin. Guided by these targets, we propose geometry-shaping objectives that promote masked within-class self-expressiveness, discourage cross-class reconstruction pathways and inter-class span alignment, and prevent collapse -- without invoking SRC residuals or predictions during training. Experiments on images (COIL-100), text (TREC), and EEG connectivity evaluate all representations under identical fixed SRC/OMP inference and report residual margins and geometric diagnostics; cross-entropy is included only as a reference geometry under the same evaluation protocol.
【11】On the Construction and Implications of Low-Loss Valleys in LoRA-based Bayesian Inference
标题：基于LoRA的Bayesian推理中低损失山谷的构建及其含义
链接：https://arxiv.org/abs/2605.29580
摘要：
【12】Honest Lying: Understanding Memory Confabulation in Reflexive Agents
标题：诚实的谎言：理解反身代理中的记忆虚构
链接：https://arxiv.org/abs/2605.29463
备注：Accepted to ICML 2026 Workshop "Failure Modes in Agentic AI"
摘要：
【13】Auditing Training Data in Generative Music Models via Black-Box Membership Inference
标题：通过黑匣子成员资格推理审计生成音乐模型中的训练数据
链接：https://arxiv.org/abs/2605.29202
备注：The paper has been accepted for presentation at the workshop ArtSec 2026: Workshop on Artwork Security and Provenance in the Age of AI
摘要：
【14】When RL Suppresses Its Own Vocabulary: Recovering Reasoning Diversity in Puzzle-to-Math Transfer
标题：当RL抑制自己的词汇时：恢复拼图到数学转换中的推理多样性
链接：https://arxiv.org/abs/2605.29190
备注：Preprint
摘要：
【15】When and How Long? The Readout-Mediator Angle in Temporal Reasoning
标题：何时以及多长时间？时间推理中的读出中介角度
链接：https://arxiv.org/abs/2605.29126
摘要：
【16】Robust and Efficient Guardrails with Latent Reasoning
标题：具有潜在推理的稳健高效护栏
链接：https://arxiv.org/abs/2605.29068
摘要：
【17】FormInv: A Measurement Protocol for Semantic Invariance in Mathematical Reasoning Benchmarks
标题：FormInv：数学推理基准中语义不变性的测量协议
链接：https://arxiv.org/abs/2605.29001
备注：18 pages, 3 figures. Under review for the 3rd AI for Math Workshop (AI4Math), ICML 2026
摘要：
【18】Echoes within the Reasoning: Stealthy and Effective Watermarking via Chain of Thought
标题：推理中的回声：通过思想链进行隐蔽有效的水印
链接：https://arxiv.org/abs/2605.28890
备注：This paper is accepted by ICML2026
摘要：
【19】Leave a Window Out: Modifying the Jackknife for Predictive Inference in Time Series
标题：留出一个窗口：修改时间序列中预测推理的Jackknife
链接：https://arxiv.org/abs/2605.30292
备注：36 pages, 6 figures
摘要：
【20】Wasserstein Contraction of Coordinate Ascent Variational Inference
标题：坐标上升变分推理的沃瑟斯坦压缩
链接：https://arxiv.org/abs/2605.30253
备注：17 pages + 3 pages appendix, 3 figures
摘要：
【21】Mixing Vector Model for Copolymer Inference via Mixed Integer Linear Programming
标题：混合线性规划共聚物推断的混合向量模型
链接
：https://arxiv.org/abs/2605.29329
摘要：A novel two-phase molecule inference framework, mol-infer, has recently been developed to infer chemical graphs with prescribed abstract structures and desired property values through mixed integer linear programming (MILP) under the two-layered model, with guaranteed optimality and exactness relative to the given learned prediction function and structural constraints. In this study, we extend this framework to copolymers by introducing a simple feature representation, called the mixing vector (MV) model. In the proposed model, a copolymer feature vector is represented as a convex combination of MILP-tractable monomer descriptors weighted by the mixing ratio of the constituent monomers. This representation does not require explicit sequence-class information and is therefore naturally compatible with MILP-based inverse design. Under this model, we construct prediction functions for several copolymer property datasets using artificial neural networks, reduced quadratic multiple linear regression, and random forests. The proposed representation achieves practically useful predictive performance across multiple physicochemical property datasets; in particular, the best test R^2 score exceeds 0.7 for nine of the ten datasets and exceeds 0.9 for six datasets. We also formulate a multi-monomer inverse-design problem under the MV representation with a prescribed mixing ratio and show that the resulting MILP instances remain tractable, even for three-monomer settings. Finally, we perform an external consistency check by re-evaluating the inferred candidates and comparing the re-computed property values with those predicted by the learned model. Overall, the proposed framework gives a tractable first step toward model-level exact inverse design of copolymers under the two-layered model.
【22】Prediction-Powered Inference Across Many Tasks for AI Evaluation & Social Science Research
标题：人工智能评估和社会科学研究的许多任务中的预测推理
链接：https://arxiv.org/abs/2605.29249
摘要：Many applications require statistically valid inference across many related tasks, while using only a handful of high-quality labels per hypothesis. In AI evaluation, these tasks may correspond to model behaviors across prompts, subgroups, or hypotheses; in social science surveys, they may correspond to related questions, populations, or measurement conditions. Prediction-powered inference (PPI) uses abundant but inexpensive proxy measurements to improve inference from limited, ground-truth labels, but commonly used methods treat tasks independently and therefore fail to exploit shared structure across related tasks. This limitation is especially important in settings where only a small number of labels are available per task. To address this issue, we introduce a multi-task prediction-powered inference framework that uses labeled data from related tasks to improve power while preserving task-specific inference. Our methods exploit the shared structure in the proxy-ground-truth relationship through cross-task recalibration, while retaining within-task rectification and power tuning to construct accurate point estimates and confidence intervals. We prove that efficiency gains beyond power-tuned PPI are only possible when the proxy-ground-truth relationship contains nonlinear structure; affine cross-task recalibrations are asymptotically equivalent to using the original proxy. We complement our theoretical findings with experiments on synthetic and semi-synthetic datasets, as well as a case study auditing language models on election-related information during the 2024 U.S. presidential election. Using a large human-annotation study, we show that cross-task recalibration can substantially reduce confidence interval widths when labels are scarce.
检测相关(7篇)
【1】Masked Diffusion Modeling for Anomaly Detection
标题：异常检测的掩蔽扩散模型
链接：https://arxiv.org/abs/2605.30046
摘要：
【2】Open World Autoencoding Drift Detection with Novel Class Recognition in Tabular Non-stationary Data Streams
标题：开放世界自动编码漂移检测，在表格非平稳数据流中使用新型类别识别
链接：https://arxiv.org/abs/2605.29834
摘要：
【3】Training Deliberative Monitors for Black-Box Scheming Detection
标题：训练黑匣子计划检测的深思熟虑者
链接：https://arxiv.org/abs/2605.29601
摘要：
【4】Audio Deepfake Detection with Half-Truth Localisation Using Cross-Attentive Feature Fusion
标题：使用交叉注意特征融合进行半真定位的音频深度伪造检测
链接：https://arxiv.org/abs/2605.29531
备注：13 pages, 5 figures, 11 tables
摘要：
【5】K-FinHallu: A Hallucination Detection Benchmark for Multi-Turn RAG in Korean Finance
标题：K-FinHallu：韩国金融多回合RAG的幻觉检测基准
链接：https://arxiv.org/abs/2605.29523
摘要：
【6】Lightweight Complementary-Cue Fusion for Robust Video Face Forgery Detection
标题：轻量级补充线索融合用于鲁棒的视频人脸伪造检测
链接：https://arxiv.org/abs/2605.29092
备注：13 pages, 6 figures, 3 tables
摘要：
【7】The Biosecurity Blind Spot: Systematic Dual-use Detection in Open Science Infrastructure
标题：生物安全盲点：开放科学基础设施中的系统性双重用途检测
链接：https://arxiv.org/abs/2605.28843
备注：Ongoing work
摘要：
分类|识别(3篇)
【1】Efficient, Validation-Free Intrinsic Quality Estimation for Large-Scale Face Recognition Datasets
标题：大规模人脸识别数据集的高效、免验证的本质质量估计
链接：https://arxiv.org/abs/2605.29720
备注：ICML 2026
摘要：
【2】Opir: Efficient Multi-Task Safety Classification for Toxicity, Jailbreaks, Hate Speech, and Harmful Content
标题：Opir：毒性、越狱、仇恨言论和有害内容的高效多任务安全分类
链接：https://arxiv.org/abs/2605.29659
备注：23 pages, 4 figures, 9 tables
摘要：Real-time safety filtering for large language model (LLM) applications requires classifiers that can detect unsafe prompts, toxic language, jailbreak attempts, and unsafe responses without the cost profile of large guardrail models, and that can distinguish benign sensitive text from genuinely covert harmful content. In this paper, we introduce Opir, a family of encoder-based guardrail models built on the GLiClass architecture. Opir includes multi-task models for binary safe/unsafe classification, multi-label toxicity classification, jailbreak classification, and zero-shot unsafe prompt and response categorization. We also release edge variants with fewer than 100M parameters dedicated to binary safe/unsafe categorization. The models are trained on a three-level taxonomy containing 996 categories across 16 top-level labels, 126 mid-level labels, and 854 leaf labels. Opir's training data combines taxonomy-grounded unsafe prompts, adversarially mined hard negatives, benign safety-preserving examples, generated response examples, multilingual translations, and portions of the Aegis2 and WildGuard training subsets. We also open-sourced an evaluation harness that supports GLiClass and GLiNER2 backends as well as decoder-based models, and covers binary safety classification, multi-label categorization, toxicity, jailbreak detection, prompt safety, response safety, response refusal, and prompt subcategory views across public benchmark families. Across an expanded comparison spanning 12 safety-classification tasks and 17 category tasks against eight contemporary guardrail systems -- including both GLiNER2-based and generative guardrail models -- Opir variants are competitive on or ahead of the strongest open-weight baselines on the majority of benchmark datasets while operating with a substantially smaller deployment footprint.
【3】NeuroEdge: Real-Time Hand Gesture Recognition with High-Density EMG Using Deep Learning at the Edge
标题：NeuroEdge：在边缘使用深度学习，通过高密度EMG进行实时手势识别
链接：https://arxiv.org/abs/2605.29326
摘要：High-density electromyography (HD-EMG) has emerged as a powerful modality for decoding fine-grained neuromuscular activity, enabling real-time neural-machine interfaces (NMIs) for applications such as prosthetic control, rehabilitation, and augmented interaction. While deep learning approaches such as convolutional neural networks (CNNs)have demonstrated high classification accuracy for EMG-based gesture recognition, their deployment on embedded hardware remains a major challenge due to computational and memory constraints. This paper presents NeuroEdge, a real-time HD EMG-based NMI system that performs gesture recognition entirely on resource-constrained microcontrollers. The system features two custom-designed modules: the HD-EMG StreamBridge, a wireless communication interface that streams raw HD-EMG data from a Quattrocento amplifier to an ESP32 microcontroller; and the EdgeDL Inference Engine, a lightweight deep learning framework executing on a Sony Spresense microcontroller. A compact 1-dimensional CNN optimized for embedded inference processes, sliding windows of EMG data in real time. Data streaming and inference are pipelined and synchronized through an architecture that utilizes Direct Memory Access (DMA) for data transfer and Serial Peripheral Interface (SPI) burst communication between the ESP32 and Spresense, ensuring low-latency performance. Experimental results show that NeuroEdge achieves a real-time classification accuracy of 90% across seven hand gestures, with a total average latency of 83 ms using 192 channels of HD-EMG recorded from the forearm. Our system demonstrates the feasibility of deploying complex HD-EMG-based gesture recognition on microcontroller-based edge devices, bridging the gap between high-resolution biosignal acquisition and deep learning-based embedded inference for next-generation NMIs.
表征(6篇)
【1】DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation
标题：DynaFLIP：通过三模式动力学引导表示重新思考机器人感知
链接：https://arxiv.org/abs/2605.30350
备注：Project website: https://dynaflip-robotics.github.io
摘要：Robot manipulation critically depends on perception that preserves the action-relevant aspects of a scene. Yet most robot learning pipelines are built upon visual encoders pre-trained for static recognition or vision-language alignment, leaving motion understanding to downstream policies. We introduce DynaFLIP, a dynamics-aware multimodal pre-training framework that pushes motion understanding upstream into perception. We construct image-language-3D flow triplets from heterogeneous human and robot videos, and use these triplets as training-time supervision to shape an image-only encoder. Our key idea is to encourage the three modalities to span a small simplex volume in the shared hyperspherical space -- a smaller simplex volume indicating stronger alignment. To avoid the geometric ambiguity and trivial collapse of naive volume minimization, we combine simplex-volume minimization with a cosine regularizer and a contrastive objective. Our analyses show that DynaFLIP focuses on control-relevant regions critical for manipulation. The resulting dynamics-aware representations serve as reusable visual backbones and consistently outperform baselines across diverse downstream policies, including VLAs. We validate this across diverse simulation and real-world setups, with gains reaching +22.5% under out-of-distribution scenarios. Our results suggest that robot generalization improves when visual representations are trained to encode not just what is present, but how the world changes under action.
【2】Striding Across Reynolds Numbers: Representation Geometry in Neural PDE Generalisation
标题：跨越雷诺兹数：神经PDL概括中的表示几何
链接：https://arxiv.org/abs/2605.30112
备注：12 pages, 8 figures, 5 tables
摘要：
【3】Genetically Aligned Patient Representations Improve Hematological Diagnosis
标题：基因对齐的患者代表改善血液学诊断
链接：https://arxiv.org/abs/2605.29980
备注：Accepted for publication at the 29th International Conference on Medical Image Computing and Computer Assisted Intervention - MICCAI 2026
摘要：
【4】Learning to Perturb Hidden Representations for Generalizable Deep Learning
标题：学习扰动隐藏表示以实现可推广深度学习
链接：https://arxiv.org/abs/2605.29525
摘要：
【5】Neural-Behavioral Representation of Natural Whole-body Movement in Monkeys
标题：猴子自然全身运动的神经行为表现
链接：https://arxiv.org/abs/2605.29355
摘要：Understanding how cortical activity represents natural whole-body behaviors in primates remains challenging. Limited by the diversity of movements and inaccessibility of large-scale neural representation of whole-body kinematics, previous motor decoding studies focused on constrained tasks and limited limb movements. Here, we present a neural-behavioral recording and modeling framework for freely moving monkeys, combining large-scale epidural cortical signals from distributed sensory- and motor-related areas with synchronized multi-view motion capture through a custom-made data collection platform. We reconstructed whole-body monkey kinematics and learned a compact behavior prior using an autoregressive encoder-decoder model. Conditioned on neural signals, the model decoded accurate and realistic whole-body movement without explicit physical constraints. Our results provide a novel proof-of-concept approach for decoding natural whole-body movements in primates using large-scale intracranial neural activity.
【6】Representation Alignment Rests on Linear Structure
标题：表示对齐基于线性结构
链接：https://arxiv.org/abs/2605.28870
摘要：
优化|敛散性(12篇)
【1】HPO: Hysteretic Policy Optimization for Stable and Efficient Training under Sparse-Reward Regime
标题：HPO：稀疏奖励制度下稳定高效训练的滞后政策优化
链接：https://arxiv.org/abs/2605.30201
摘要：
【2】ESPO: Early-Stopping Proximal Policy Optimization
标题：ESPO：早期停止的近端政策优化
链接：https://arxiv.org/abs/2605.29860
摘要：
【3】The Interplay Between Interpolation and Aggregation in Regression: Optimal Sample Complexity
标题：回归中插值和聚集之间的相互作用：最佳样本复杂性
链接：https://arxiv.org/abs/2605.29819
摘要：
【4】Singularity-aware Optimization via Randomized Geometric Probing: Towards Stable Non-smooth Optimization
标题：通过随机几何探测实现奇异性感知优化：迈向稳定的非光滑优化
链接：https://arxiv.org/abs/2605.29547
备注：International Conference on Machine Learning (ICML), 2026
摘要：
【5】Unveiling Multi-regime Patterns in SciML: Distinct Failure Modes and Regime-specific Optimization
标题：在SciML中揭示多机制模式：不同的故障模式和特定于机制的优化
链接：https://arxiv.org/abs/2605.29153
备注：Accepted by ICML 2026
摘要：
【6】Optimal Gap-Dependent Regret for Private Stochastic Decision-Theoretic Online Learning
标题：私人随机决策理论在线学习的最佳差距相关遗憾
链接：https://arxiv.org/abs/2605.29148
摘要：
【7】Optimal Rates for Differentially Private Hypothesis Testing with E-values
标题：具有E值的差异私人假设测试的最佳比率
链接：https://arxiv.org/abs/2605.28952
备注：28 pages, 2 figures
摘要：
【8】Molecular Lead Optimization via Agentic Tool Planning
标题：通过统计工具规划进行分子线索优化
链接：https://arxiv.org/abs/2605.28862
备注：12 pages
摘要：
【9】WASHH: An Anchor-Aware Whale-Guided Selection Hyper-Heuristic for Continuous Optimization and SVC Configuration
标题：WASH：具有锚感知力的鲸鱼引导选择超启发式，用于持续优化和SV配置
链接：https://arxiv.org/abs/2605.28844
摘要：
【10】Diffusion Models Are Statistically Optimal for Learning Low-Dimensional Multi-Modal Distributions
标题：扩散模型对于学习低维多模式分布是统计上最佳的
链接：https://arxiv.org/abs/2605.30153
备注：accepted to ICML 2026
摘要
：Score-based diffusion models have demonstrated remarkable empirical success in learning high-dimensional distributions, particularly those exhibiting low-dimensional and multi-modal structures. However, theoretical understanding of their statistical efficiency remains limited. Existing theories typically rely on strong regularity assumptions, such as uniformly bounded densities or globally smooth score functions, which fail to capture such intrinsic structures. In this work, we study the sample complexity of diffusion models for learning distributions supported on a union of low-dimensional subspaces. Assuming that the data distribution within each subspace is subgaussian, we show that diffusion models require at most $\widetilde{O}(\varepsilon^{-k \vee 2})$ samples to achieve $\varepsilon$ error in 1-Wasserstein distance, where $k$ is the intrinsic dimension. This near-optimal convergence rate depends only on the intrinsic dimension and significantly improves upon prior theoretical guarantees that suffer from the curse of dimensionality. Notably, our analysis applies to a broad collection of distributions without imposing smoothness, bounded-density, or log-concavity assumptions. Overall, our results show that diffusion models can statistically adapt to intrinsic low-dimensional structure while naturally accommodating multi-modal data, offering a rigorous theoretical justification for their success in complex high-dimensional learning tasks.
【11】MoSSP: A Momentum-Based Single-Loop Stochastic Penalty Method for Nonconvex Constrained DC-Regularized Optimization
标题：MoSPP：非凸约束直流正规化优化的基于动量的单回路随机罚方法
链接：https://arxiv.org/abs/2605.29635
备注：35 pages, 3 figures
摘要：In this paper, we study a structured class of nonconvex constrained stochastic problems with difference-of-convex (DC) regularization, where the feasible set is possibly nonconvex and the concave part of the DC regularizer is allowed to be nonsmooth. The fundamental challenge lies in maintaining feasibility for nonconvex constraints while achieving favorable oracle complexity. Although single-loop algorithms efficiently solve unconstrained DC optimization problems, their potential for constrained optimization with DC structure remains largely unexplored. To address this gap, we develop MoSSP, a Momentum-based Single-loop Stochastic Penalty method for such problems with provable complexity guarantees. The key idea is to apply a single stochastic proximal-gradient step to the Moreau envelope of the penalty plus the convex DC part, with the concave part's proximal mapping computed in parallel. We derive two algorithm variants: a Polyak-momentum version with $O(\varepsilon^{-4})$ oracle complexity for finding stochastic $\varepsilon$-KKT points, and an improved $O(\varepsilon^{-3})$ version incorporating recursive momentum. Experimental results demonstrate the effectiveness of the proposed algorithms.
【12】Financially Guided Deep Portfolio Optimization
标题：财务指导深度投资组合优化
链接：https://arxiv.org/abs/2605.28853
摘要：Portfolio optimization in real-world financial markets is notoriously difficult due to non-stationarity, noisy data, and high transaction costs. Standard predict-then-optimize methods first forecast returns and then solve for weights, compounding prediction errors and often failing under regime shifts. We propose an end-to-end framework that directly optimizes differentiable surrogates of key financial metrics - Sharpe ratio, Omega ratio, Conditional Value-at-Risk (CVaR), and Risk Parity - allowing neural networks to learn portfolio weights via backpropagation. Our expanding-window walk-forward procedure, applied to 50 S&P 500 stocks from 2007 to 2023, incorporates realistic bid-ask spread costs and rebalances quarterly. On the challenging out-of-sample test period (2022-2023), the best model - an AttentionLSTM with the Omega-CVaR-RiskParity loss - achieves an annualized Sharpe of 0.29 and a total compounded return of +7.86%, while the S&P 500 delivers -4.52% total return and an annualized Sharpe of -0.02. This outperforms the S&P 500 by 12.38 percentage points (a relative improvement of over 270%), while keeping tail risk (CVaR) nearly unchanged. The framework consistently outperforms the equal-weight portfolio, S&P 500, and traditional methods (MVP, HRP, NCO), demonstrating that embedding financial objectives directly into model training yields robust, economically meaningful outperformance even in adverse market conditions.
预测|估计(7篇)
【1】Beyond MSE: Improving Precipitation Nowcasting with Multi-Quantile Regression
标题：超越SSE：利用多分位数回归改善降水量即时预报
链接：https://arxiv.org/abs/2605.30122
备注：7 pages, 5 figs
摘要：
【2】STAP: A Shuffle-Tokenized App Predictor with Ultra Long Context for Vocabulary-Free Mobile App Prediction
标题：STAP：具有超长上下文的洗牌令牌化应用程序预测器，用于无词汇移动应用程序预测
链接：https://arxiv.org/abs/2605.29863
备注：15 pages, 9 figures, 5 tables Preprint submitted to Expert Systems with Applications
摘要：
【3】A Systematic Evaluation of Molecular Mixture Behavior Prediction
标题：分子混合物行为预测的系统评价
链接：https://arxiv.org/abs/2605.29698
摘要：Machine learning for molecular property prediction has focused largely on pure compounds, even though many practical applications depend on mixtures with intermolecular interactions. Recent work has expanded the availability of mixture datasets, but evaluation still focuses mainly on absolute accuracy. However, absolute errors in mixtures conflate pure-component contributions with deviations from ideal mixing. We propose an evaluation framework that decomposes mixture-property error into pure-compound and interaction (non-ideal) components. The framework combines leakage-aware split protocols, ideal-mixture baselines, and excess-property metrics. To support reproducible benchmarking, we curate seven matched pure and mixture physicochemical property datasets. Across multiple mixture-property tasks and model families, we find that strong absolute accuracy can mask poor recovery of non-ideal mixture behavior, and that performance drops substantially under strict molecule splits. These results identify transfer to unseen molecules as a central challenge in molecular mixture machine learning and motivate evaluation beyond absolute accuracy alone.
【4】The Good, the Bad, and the Ugly of Markov Boundary for Tabular Prediction
标题：表格预测马尔科夫边界的好、坏和丑
链接：https://arxiv.org/abs/2605.29411
备注：11 pages, 9 figures, 2 tables. Preprint
摘要：
【5】Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting
标题：重新思考多模态时间序列预测的训练后配方
链接：https://arxiv.org/abs/2605.29401
摘要：
【6】Ensemble Score Filtering for Real-Data Energy Consumption Forecast Correction
标题：用于实时数据能耗预测校正的Ensemble分数过滤
链接：https://arxiv.org/abs/2605.29072
摘要：
【7】Improved Guarantees for Heterogeneous Treatment-Effect Estimation via Matrix Completion
标题：通过矩阵完成改进了对不均匀治疗效果估计的保证
链接：https://arxiv.org/abs/2605.30319
摘要：
其他神经网络|深度学习|模型|建模(43篇)
【1】Neural Operator-Based Surrogate Model for CFD:Helical Coil Steam Generator in Small Modular Reactor
标题：基于神经网络的小型模块化反应堆螺旋管蒸汽发生器CFD替代模型
链接：https://arxiv.org/abs/2605.30277
摘要：
【2】TriSearch: Learning to Optimize Triangulations via Bistellar Flips
标题：TriSearch：学习通过Bistellar翻转优化三角剖分
链接：https://arxiv.org/abs/2605.30220
摘要：
【3】DAMEL: Dual-Axis Multi-Expert Learning for Class-Imbalanced Learning
标题：DAMEL：用于类不平衡学习的双轴多专家学习
链接：https://arxiv.org/abs/2605.30135
摘要：
【4】Learning to Extrapolate to New Tasks: A Relational Approach to Task Extrapolation
标题：学会外推新任务：任务外推的关系方法
链接：https://arxiv.org/abs/2605.30132
备注：ICML 2026
摘要：
【5】Chess-World-Model: A 10M-Game Benchmark for Exact State Tracking from Chess Move Sequences
标题：国际象棋世界模型：从国际象棋走码序列精确状态跟踪的10 M游戏基准
链接：https://arxiv.org/abs/2605.30100
备注：20 pages, 4 figures
摘要：
【6】Alignment-Guided Score Matching for Text-to-Image Alignment in Diffusion Models
标题：扩散模型中用于文本到图像对齐的对齐引导得分匹配
链接：https://arxiv.org/abs/2605.30038
备注：ICML 2026, Project page: this https URL
摘要：
【7】Midpoint Generative Models
标题：中点生成模型
链接：https://arxiv.org/abs/2605.29920
摘要：
【8】Feedback-to-Rubrics: Can We Learn Expert Criteria from Inline Comments?
标题：反馈至主题：我们可以从在线评论中了解专家标准吗？
链接：https://arxiv.org/abs/2605.29857
摘要：
【9】MMTM: Tri-Modal Topic Modeling for Long-Form Video via Similarity-Gated Fusion
标题：MMTM：通过相似门控融合进行长格式视频的三模式主题建模
链接：https://arxiv.org/abs/2605.29765
备注：Submitted to EMNLP 2026
【10】A Novel Tensor Product-Based Neural Network for Solving Partial Differential Equations
标题：一种新型的基于张量积的神经网络求解偏微方程
链接：https://arxiv.org/abs/2605.29688
备注：44 pages, 11 figures
【11】Kernel Renormalization in Bayesian Deep Neural Networks: the Equivalent Wishart Ansatz in the Proportional Regime
标题：Bayesian深度神经网络中的核重正化：比例制度中的等效Wishart Antives
链接：https://arxiv.org/abs/2605.29684
备注：45 pages, 21 figures
【12】AMDP: Asynchronous Multi-Directional Pipeline Parallelism for Large-Scale Models Training
标题：AMDP：用于大规模模型训练的同步多方向流水线并行主义
链接：https://arxiv.org/abs/2605.29664
备注：Accepted by ICML 2026, 9 pages, and 8 figures
【13】Learning Context-Conditioned Predicate Semantics via Prototype Feedback
标题：通过原型反馈学习上下文条件的断言语义
链接：https://arxiv.org/abs/2605.29610
备注：Accepted at ICML 2026. Code: this https URL
【14】Why Larger Models Learn More: Effects of Capacity, Interference, and Rare-Task Retention
标题：为什么更大的模型会学到更多：容量、干扰和稀有任务保留的影响
链接：https://arxiv.org/abs/2605.29548
【15】The Complexity of Verifying Feedforward Neural Networks in Quantised Settings
标题：量化环境下的MIDI前向神经网络的复杂性
链接：https://arxiv.org/abs/2605.29537
【16】DynaGraph: Lightweight Multi-Model Interaction Framework via Dynamic Topological Reconfiguration
标题：DynaShape：通过动态布局重新配置的轻量级多模型交互框架
链接：https://arxiv.org/abs/2605.29511
【17】Access Sets Matter: Budgeting Expert Reads for Scalable Weight-Space Model Merging
标题：访问集很重要：可扩展权重空间模型合并的专家阅读
链接：https://arxiv.org/abs/2605.29489
备注：ICML 2026 Workshop on Weight-Space Symmetries: from Foundations to Practical Applications
【18】Do Physics Foundation Models Learn Generalizable Physics? A Bias-Aware Benchmark Across Physical Regimes and Distribution Shifts
标题：物理基础模型学习可推广的物理学吗？跨物理状态和分布变化的偏差感知基准
链接：https://arxiv.org/abs/2605.29283
备注：26 pages, 31 figures
【19】Causal Label Recovery in Payment Networks
标题：支付网络中的因果标签恢复
链接：https://arxiv.org/abs/2605.29272
备注：49 pages
【20】When and How Human Curation Backfires: Preference Alignment under Multi-Model Self-Consuming Loop
标题：人类治疗何时以及如何适得其反：多模型自我消费循环下的偏好对齐
链接：https://arxiv.org/abs/2605.29267
【21】KLAS: Using Similarity to Stitch Neural Networks for Improved Accuracy-Efficiency Tradeoffs
标题：KLAS：使用相似性缝合神经网络以改善准确性与效率的权衡
链接：https://arxiv.org/abs/2605.29259
【22】SigmaMedStat: Temporal Signal Modeling for ICU False Alarm Reduction
标题：SigmaMedStat：用于ICU虚警减少的时间信号建模
链接：https://arxiv.org/abs/2605.29236
备注：Code available at this http URL
【23】libhmm: A Modern C++20 Library for Hidden Markov Models with Correct MLE Emission M-Steps
标题：libhmm：具有正确MLE发射M步的隐马尔科夫模型现代C++20库
链接：https://arxiv.org/abs/2605.29208
备注：17 pages, 3 figures, 8 tables
【24】Probabilistic bias adjustment of seasonal forecasts using generative machine learning: A case study of Arctic sea ice predictions
标题：使用生成式机器学习对季节预报进行概率偏差调整：北极海冰预测的案例研究
链接：https://arxiv.org/abs/2605.29172
【25】Parallax: Parameterized Local Linear Attention for Language Modeling
标题：Parallax：语言建模的参数化局部线性注意力
链接：https://arxiv.org/abs/2605.29157
【26】Do Deep Networks Forget Initialization? A Forgetting-Time View of Practical Inductive Bias
标题：深度网络会忘记收件箱吗？实践归纳偏见的遗忘时间观
链接：https://arxiv.org/abs/2605.29152
备注：39 pages, 9 figures
【27】Model Merging by Output-Space Projection
标题：利用输出空间投影进行模型合并
链接：https://arxiv.org/abs/2605.29101
【28】Moment Matching Q-Learning
标题：时刻匹配Q-Learning
链接：https://arxiv.org/abs/2605.29033
备注：23 pages, 14 figures, 10 tables, accepted by ICML 2026
【29】Theoretical Foundations and Effective Algorithms for Policy-Aware Simulator Learning
标题：策略感知模拟器学习的理论基础和有效算法
链接：https://arxiv.org/abs/2605.29032
【30】The Hamilton-Jacobi Theory of Deep Learning
标题：深度学习的Hamilton-Jacobi理论
链接：https://arxiv.org/abs/2605.28983
【31】Sequential Physics-Constrained Neural Operator Forward Modeling for the $\textit{Norne}$ Reservoir System
标题：基于序贯物理约束的神经算子正演模拟 extit{Norne}$水库系统
链接：https://arxiv.org/abs/2605.28909
备注：22 pages, 2 figures, 2 tables. Code available at this https URL
【32】Spectral Guidance for Flexible and Efficient Control of Diffusion Models
标题：用于扩散模型灵活有效控制的谱引导
链接：https://arxiv.org/abs/2605.28900
备注：ICML 2026
【33】Towards Continuous-time Causal Foundation Models
标题：走向连续时间因果基础模型
链接：https://arxiv.org/abs/2605.28880
备注：ICML 2026 2nd Workshop on Foundation Models for Structured Data (FMSD)
【34】Balancing Multimodal Learning through Label Space Reshaping
标题：通过标签空间重塑平衡多模式学习
链接：https://arxiv.org/abs/2605.28869
备注：In process
【35】TaxDistill: Improving Metagenomic Taxonomic Annotation via Distilled Genomic Foundation Models
标题：TaxDistill：通过蒸馏基因组基础模型改进宏基因组分类学注释
链接：https://arxiv.org/abs/2605.28868
备注：The manuscript contains 14 pages, 7 figures, and 3 tables
【36】An End-to-End PyTorch Interface for Differentiable PDE Solvers: A RANS Model-Correction Study
标题：用于可区分的DTE解算器的端到端PyTorch接口：RANS模型修正研究
链接：https://arxiv.org/abs/2605.28858
【37】Visual Spatial Learning: Single-Field Spatial Interpolation Using Convolutional Neural Networks
标题：视觉空间学习：使用卷积神经网络的单场空间插值
链接：https://arxiv.org/abs/2605.30167
备注：53 pages, 10 figures
【38】Joint Model and Data Sparsification via the Marginal Likelihood
标题：联合模型与基于边际似然的数据稀疏化
链接：https://arxiv.org/abs/2605.29908
备注：36 pages, 8 figures, 12 tables (incl. appendix); published at ICML 2026
【39】A Minimal Bifurcation Model of Load Imbalance in a Softmax Mixture-of-Experts Router
标题：Softmax专家混合路由器中负载不平衡的最小分歧模型
链接：https://arxiv.org/abs/2605.29121
备注：21 pages, 11 figures
【40】Three-dimensional Conditional Diffusion Models for Cosmological 21 cm Lightcone Emulation
标题：宇宙学21厘米光锥模拟的三维条件扩散模型
链接：https://arxiv.org/abs/2605.29016
【41】Computational Modeling of Antibody-Antigen Complexes: PLM-Based and MSA-Based Approaches
标题：抗体-抗原复合物的计算建模：基于PLM和基于GMA的方法
链接：https://arxiv.org/abs/2605.28886
备注：PhD thesis
【42】Comment on "Spin-1/2 Kagome Heisenberg Antiferromagnet: Machine Learning Discovery of the Spinon Pair-Density-Wave Ground State"
链接：https://arxiv.org/abs/2605.28861
备注：3 pages, 1 figure; Comment on arXiv:2401.02866
【43】Towards a Foundation Model for the Martian Atmosphere
标题：迈向火星大气基础模型
链接：https://arxiv.org/abs/2605.28851
其他(55篇)
【1】When, why, and how do diffusion posterior samplers fail? A finite-sample lens
标题：扩散后采样器何时、为什么以及如何失败？有限样本镜头
链接：https://arxiv.org/abs/2605.30330
备注：All code for experiments is available at: https://github.com/voilalab/diagnosing-posterior-sampling
【2】SoundnessBench: Can Your AI Scientist Really Tell Good Research Ideas from Bad Ones?
标题：SoundnessBench：你的人工智能科学家真的能区分好的研究想法和坏的研究想法吗？
链接：https://arxiv.org/abs/2605.30329
备注：Project Page: https://hosytuyen.github.io/projects/SoundnessBench
【3】Gram: Assessing sabotage propensities via automated alignment auditing
标题：Gram：通过自动对齐审计评估破坏倾向
链接：https://arxiv.org/abs/2605.30322
【4】Self-Trained Verification for Training- and Test-Time Self-Improvement
标题：训练和测试时自我改进的自我训练验证
链接：https://arxiv.org/abs/2605.30290
【5】Statistical Embeddings for Similarity, Retrieval, and Interpretable Alignment of Numeric Tabular Datasets
标题：数字表格数据集的相似性、检索和可解释对齐的统计嵌入
链接：https://arxiv.org/abs/2605.30289
【6】Faithful Embeddings of Irregular and Asynchronous Data for Online Log-NCDEs
标题：在线日志NDEC的不规则和非同步数据的忠实嵌入
链接：https://arxiv.org/abs/2605.30213
备注：34 pages, 16 figures
【7】Mean-Field Diffuser: Scaling Offline MARL to Thousands of Agents
标题：均场扩散器：将离线MARL扩展到数千个代理人
链接：https://arxiv.org/abs/2605.30190
备注：71 pages, 15 figures, 16 tables
【8】CalArena: A Large-Scale Post-Hoc Calibration Benchmark
标题：CalArena：大规模事后校准基准
链接：https://arxiv.org/abs/2605.30188
备注：30 pages, 9 figures
【9】No More K-means:Single-Stage Sparse Coding for Efficient Multi-Vector Retrieval
标题：不再有K均值：单级稀疏编码用于高效的多载体检索
链接：https://arxiv.org/abs/2605.30120
备注：Accepted by ICML2026
【10】Ridge Regression from Poisson Resetting: A Renewal Perspective on Spectral Regularization
标题：来自Poisson重置的岭回归：谱正规化的更新视角
链接：https://arxiv.org/abs/2605.30059
【11】Discovering Cooperative Pipelines: Autoresearch for Sequential Social Dilemmas
标题：发现合作管道：对连续社会困境的自动研究
链接：https://arxiv.org/abs/2605.30003
备注：Accepted to the AI Agents for Discovery in the Wild (AID-Wild) Workshop at ACM CAIS 2026
【12】A Domain-Informed Multi-Objective Framework for EEG Channel Selection in Motor Imagery BCIs
标题：运动图像BCI中脑电通道选择的领域知情多目标框架
链接：https://arxiv.org/abs/2605.29943
备注：This work has been submitted to the IEEE for possible publication
【13】CRB-Guided Framework Design and Resource Allocation for Indoor mmWave ISCC Systems
标题：CRB引导的室内毫米波ISCC系统框架设计和资源分配
链接：https://arxiv.org/abs/2605.29939
备注：7 pages, 6 figures, conference(submitted to GLOBECOM)
【14】Fisher-Preserving Guidance: Training-Free Manifold Constraints for Safe Diffusion Control
标题：渔民保护指南：安全扩散控制的免训练多种限制
链接：https://arxiv.org/abs/2605.29937
备注：ICML2026
【15】Reducing Experimental Testing in Space Propulsion Film Cooling Analyses by Pixelwise Generative Image Interpolation
标题：逐像素生成图像插值减少空间推进器薄膜冷却分析中的实验测试
链接：https://arxiv.org/abs/2605.29911
备注：Presented at the 11th European Conference for Aeronautics and Aerospace Sciences (EUCASS), 2025, DOI: https://doi.org/10.13009/EUCASS2025-285
【16】OVA-IB: One vs All Information Bottleneck for Multi-Modal Alignment
标题：OVA-IB：多模式对齐的一与所有信息瓶颈
链接：https://arxiv.org/abs/2605.29900
【17】Open Problem: Separating Geometric and Algorithmic Compression via Cayley-Table Completion
标题：开放问题：通过凯莱表完成分离几何压缩和数学压缩
链接：https://arxiv.org/abs/2605.29885
备注：6 pages. Submitted to the Conference on Learning Theory (COLT) 2026 Open Problem track
【18】Parameter-Efficient Subspace Decoupling ViT for Mitigating Multi-Task Negative Transfer in Histological Scoring
标题：参数高效子空间去耦合ViT以缓解组织学评分中的多任务负转移
链接：https://arxiv.org/abs/2605.29852
备注：6 pages, 5 figures, 2 tables. Accepted by IEEE ICME 2026. Camera-ready version
【19】CB-SLICE: Concept-Based Interpretable Error Slice Discovery
标题：CB-SIICE：基于概念的可解释错误切片发现
链接：https://arxiv.org/abs/2605.29836
备注：20 pages, 7 figures, 12 tables, to be published at Proceedings of the 43rd International Conference on Machine Learning (ICML 2026)
【20】AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security
标题：AgentDoG 1.5：一个轻量级、可扩展的AI代理安全性协调框架
链接：https://arxiv.org/abs/2605.29801
备注：44 pages, 12 Figures, 9 Tables
【21】Certified Policy Optimisation for Nested Causal Bandits via PAC-Bayes Risk
标题：基于PAC-Bayes风险的嵌套因果Bandits认证策略优化
链接：https://arxiv.org/abs/2605.29788
【22】Realistic honeypot evaluations for scheming propensity
标题：对阴谋倾向的现实蜜罐评估
链接：https://arxiv.org/abs/2605.29729
【23】Bastion: Budget-Aware Speculative Decoding with Tree-structured Block Diffusion Drafting
标题：Bastion：采用树结构块扩散起草的预算感知推测解码
链接：https://arxiv.org/abs/2605.29727
【24】The Little Book of Generative AI Foundations: An Intuitive Mathematical Primer
标题：生成性人工智能基础小书：直观的数学入门
链接：https://arxiv.org/abs/2605.29713
备注：Preprint version, 178 pages. Comments and corrections are welcome
【25】The Sample Complexity of Multiclass and Sparse Contextual Bandits
标题：多类和稀疏背景盗贼的样本复杂性
链接：https://arxiv.org/abs/2605.29645
【26】COMET: Concept Space Dissection of the Modality Gap in Audio-Text Multimodal Contrastive Embeddings
标题：COMET：音频-文本多模式对比嵌入中情态差距的概念空间剖析
链接：https://arxiv.org/abs/2605.29628
【27】Quotient DAGs for Off-Policy Evaluation:Forward-Flow Importance Sampling and Exact Slate Propensities
标题：政策外评估的商DAB：前向流重要性抽样和精确石板倾向
链接：https://arxiv.org/abs/2605.29500
备注：31 pages, 3 figures, 7 tables
【28】PhoneWorld: Scaling Phone-Use Agent Environments
标题：PhoneWorld：扩展电话使用代理环境
链接：https://arxiv.org/abs/2605.29486
备注：work in progress
【29】How Much Is a Dataset Worth? Scaling Laws, the Vendi Score, and Matrix Spectral Functions
标题：数据集值多少钱？标度定律、文迪分数和矩阵谱函数
链接：https://arxiv.org/abs/2605.29448
备注：75 pages
【30】AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasing
标题：AliMark：增强句子级水印针对文本解释的鲁棒性
链接：https://arxiv.org/abs/2605.29434
备注：Accepted by ICML 2026
【31】Real-Time Retargeting Using Controllability Boundary for Chandrayaan-3 Lunar Landing
标题：利用可控边界实现月船三号登月实时重定向
链接：https://arxiv.org/abs/2605.29412
备注：8 pages, 6 figures, Accepted for publication in American Control Conference 2026
【32】On the Optimizer Dependence of Neural Scaling Laws
标题：关于神经标度定律的优化器依赖性
链接：https://arxiv.org/abs/2605.29387
【33】TRACER: Persistent Regularization for Robust Multimodal Finetuning
标题：TRABER：稳健多模式微调的持续正规化
链接：https://arxiv.org/abs/2605.29380
备注：ICML 2026
【34】BrahmicTokenizer-131K: An Indic-Capable Drop-In Replacement for o200k_base
标题：BrahmicTokenizer-131 K：o200 k_base的具有指示功能的插入式替代品
链接：https://arxiv.org/abs/2605.29379
备注：24 pages, 15 tables, 3 code listings. Tokenizer artifact, verification scripts, and reproduction code at https://huggingface.co/theschoolofai/BrahmicTokenizer-131K and https://github.com/theschoolofai/BrahmicTokenizer-131K
【35】Solving Integer Linear Programming with Parallel Tempering
标题：并行钢化解决NPS线性规划
链接：https://arxiv.org/abs/2605.29366
备注：Preprint. Code available at https://github.com/ski-sim/ILP-with-ParallelTempering
【36】Attention as In-Context Empirical Bayes: A Two-Stage View via Particle Dynamics
标题：作为背景下的经验Bayes的关注：粒子动力学的两阶段观点
链接：https://arxiv.org/abs/2605.29351
备注：52 pages, 5 figures
【37】GrepSeek: Training Search Agents for Direct Corpus Interaction
标题：GrepSeek：训练搜索代理进行直接的数据库交互
链接：https://arxiv.org/abs/2605.29307
【38】OmniRetrieval: Unified Retrieval across Heterogeneous Knowledge Sources
标题：OmniRetrieval：跨异类知识源的统一检索
链接：https://arxiv.org/abs/2605.29250
【39】PROTOCOL: Late Interaction Retrieval for Protein Homolog Search
标题：协议：蛋白质同质物搜索的后期相互作用检索
链接：https://arxiv.org/abs/2605.29158
【40】Eulerian Gaussian Splatting using Hashed Probability Pyramids
标题：使用哈希概率金字塔的欧拉高斯飞溅
链接：https://arxiv.org/abs/2605.29136
备注：CVPR 2026. Project Page: this https URL
【41】Bridging Chemists and AI: An Expert-Augmented Framework for Interpretable Route Evaluation
标题：化学家和人工智能的桥梁：可解释路径评估的专家增强框架
链接：https://arxiv.org/abs/2605.29108
备注：13 pages, 11 figures, ELLIS Unconference Workshop: Generative Models, LLMs, and the Future of Molecular AI (ML4Molecules 2025)
【42】Differentiable Belief-based Opponent Shaping
标题：基于差异信念的对手塑造
链接
：https://arxiv.org/abs/2605.29042
【43】Causal Intelligence for Constraint-Aware Intervention Design to Induce State Transitions
标题：约束意识干预设计的因果智能诱导状态转变
链接：https://arxiv.org/abs/2605.29008
【44】A Training-Time Diagnostic for Generalization via the Log-Alignment Ratio
标题：通过对数对齐比进行概括的训练时间诊断
链接：https://arxiv.org/abs/2605.28975
备注：32 pages, 25 figures
【45】Pre-Registering the Detectable Effect: A Paired-MDE Budget for 4-bit Quantization Benchmarks, with a Pilot Audit
标题：预注册可检测效应：4位量化基准的配对MDE预算，带试点审计
链接：https://arxiv.org/abs/2605.28873
【46】Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?
标题：灾难性遗忘的机制起源：为什么RL比SFT更好地保留电路？
链接：https://arxiv.org/abs/2605.28860
【47】One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them
标题：一个面具统治所有人：编辑后隐藏的事实以及如何找到它们
链接：https://arxiv.org/abs/2605.28839
备注：Accepted to Findings of ACL 2026
【48】What drives performance in molecular MPNNs? An operator-level factorial benchmark
标题：是什么驱动了分子MPNN的性能？操作员级因子基准
链接：https://arxiv.org/abs/2605.30195
【49】Instance-dependent Stochastic Lipschitz bandit
标题：依赖实例的随机Lipschitz强盗
链接：https://arxiv.org/abs/2605.29748
【50】Eigen-Spike Emergence and Quadratic Equivalents for Conjugate Kernels on Nonlinearly Separable Data
标题：非线性可分离数据上的本征尖峰出现和卷积核的二次等效性
链接：https://arxiv.org/abs/2605.29669
备注：89 pages, 10 figures
【51】FPLIER: Federated Pathway-Level Information Extractor
标题：FETTER：联邦路径级信息提取器
链接：https://arxiv.org/abs/2605.29587
备注：Accepted for publication at the ACM BCB '26 conference
【52】Constructing efficient channels for ideal observers using the conjugate gradient method
标题：使用共乘梯度法为理想观察者构建高效通道
链接：https://arxiv.org/abs/2605.29415
备注：Submitted to the Journal of Medical Imaging (JMI) Special Issue Honoring Dr. Harrison H. Barrett
【53】Kernel-based potential mean-field games with unbiased random Fourier $U$-statistics
标题：具有无偏随机傅里叶$U$-统计量的基于核的潜在平均场博弈
链接：https://arxiv.org/abs/2605.29371
【54】Manifold-based Algorithms for the Hadamard Decomposition
标题：基于Manifold的Hadamard分解算法
链接：https://arxiv.org/abs/2605.28980
备注：27 pages, code available from https://github.com/StefanoSicilia/Hadamard-Decomposition
【55】Dynamics of Stochastic Momentum with Sparse Updates in High Dimensions
标题：多维稀疏更新随机动量的动力学
链接：https://arxiv.org/abs/2605.28961
机器翻译由腾讯交互翻译提供，仅供参考
点击“阅读原文”获取带摘要的学术速递
Python社区是高质量的Python/Django开发社区
本文地址：http://www.python88.com/topic/197033
本文地址：http://www.python88.com/topic/197033