机器学习学术速递[5.25] na

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203篇

大模型相关(33篇)

【1】LLMs as Noisy Channels: A Shannon Perspective on Model Capacity and Scaling Laws
标题：作为噪音渠道的LLM：模型容量和缩放定律的香农视角
链接：https://arxiv.org/abs/2605.23901

【2】Strong Teacher Not Needed? On Distillation in LLM Pretraining
标题：不需要强老师吗？论LLM预训练中的蒸馏
链接：https://arxiv.org/abs/2605.23857

【3】It's the humans, not the data: Geopolitical bias in LLMs originates in post-training, amplified by the language of the prompt
标题：是人类，而不是数据：LLM中的地缘政治偏见源于训练后，并被提示的语言放大
链接：https://arxiv.org/abs/2605.23825

【4】Hierarchical Concept Geometry in Language Models Emerges from Word Co-occurrence
标题：语言模型中的分层概念几何由词共现产生
链接：https://arxiv.org/abs/2605.23821

【5】Advanced AI Service Provisioning in O-RAN through LLM Engine Integration
标题：通过LLM引擎集成在O-RAN中进行高级人工智能服务配置
链接：https://arxiv.org/abs/2605.23809

【6】Debiased Negative Mining Improves Out-of-distribution Detection with Pre-trained Vision-Language Models
标题：去偏负挖掘通过预训练的视觉语言模型改进分布外检测
链接：https://arxiv.org/abs/2605.23797

【7】LLM-driven design of physics-constrained constitutive models: two agents are better than one
标题：LLM驱动的物理约束本构模型设计：两种药剂比一种药剂好
链接：https://arxiv.org/abs/2605.23754

【8】Operator Learning for Reconstructing Flow Fields from Sparse Measurements: a Language Model Approach
标题：从稀疏测量重建流场的操作员学习：语言模型方法
链接：https://arxiv.org/abs/2605.23712

【9】CVSearch: Empowering Multimodal LLMs with Cognitive Visual Search for High-Resolution Image Perception
标题：CVSearch：通过认知视觉搜索增强多模式LLM，以实现高分辨率图像感知
链接：https://arxiv.org/abs/2605.23655

【10】Structure-Guided Entity Resolution: Fine-Tuning LLMs for Robust Name Matching in Complex Linguistic Contexts
标题：结构引导实体解析：微调LLM，以在复杂语言上下文中实现稳健的名称匹配
链接：https://arxiv.org/abs/2605.23597

【11】Push Your Agent: Measuring and Enforcing Quantitative Goal Persistence in Long-Horizon LLM Agents
标题：推动你的代理：衡量和执行长期LLM代理的量化目标持续性
链接：https://arxiv.org/abs/2605.23574

【12】When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems
标题：当尽管执行正确，规划仍失败时：基于LLM的多智能体系统的认知校准
链接：https://arxiv.org/abs/2605.23414

【13】Instance-Optimal Estimation with Multiple LLM Judges on a Budget
标题：对预算进行多个LLM法官的实例最优估计
链接：https://arxiv.org/abs/2605.23362

【14】GENSTRAT: Toward a Science of Strategic Reasoning in Large Language Models
标题：GENSTRAT：迈向大型语言模型中的战略推理科学
链接：https://arxiv.org/abs/2605.23238

【15】Empirical Bayes Conformal Prediction for Vision and Language Models
标题：视觉和语言模型的经验性Bayes保形预测
链接：https://arxiv.org/abs/2605.23189

【16】Positional Failures in Long-Context LLMs: A Blind Spot in Reasoning Benchmarks
标题：长上下文LLM中的位置故障：推理基准中的盲点
链接：https://arxiv.org/abs/2605.23170

【17】PoisonForge: Task-Level Targeted Poisoning Benchmark for Instruction-Tuned LLMs
标题：PoisonForge：用于指导调整的LLM的任务级定向中毒基准
链接：https://arxiv.org/abs/2605.23168

【18】What Does the Server See? Understanding Privacy Leakage from Large Language Models in Split Inference
标题：服务器看到什么？理解分裂推理中大型语言模型的隐私泄露
链接：https://arxiv.org/abs/2605.23158

【19】The Attribution Contract: Feature Attribution for Generative Language Models
标题：归因契约：生成性语言模型的特征归因
链接：https://arxiv.org/abs/2605.23080

【20】GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs
标题：GEMQ：MoE LLM的全球专家级混合精度量化
链接：https://arxiv.org/abs/2605.23078

【21】ModeSwitch-LLM: A Lightweight Phase-Aware Controller for Cross-Mode LLM Inference on a Single GPU
标题：ModeSwitch-LLM：一种轻量级的相感知控制器，用于在单个图形处理器上进行跨模式LLM推理
链接：https://arxiv.org/abs/2605.23057

【22】Do Language Models Know What Not to Say? Causal Evidence for Statistical Preemption in LLMs
标题：语言模型知道什么不该说吗？LLM中统计抢占的原因证据
链接：https://arxiv.org/abs/2605.23039

【23】PACE: Two-Timescale Self-Evolution for Small Language Model Agents
标题：PACE：小型语言模型代理的两时间尺度自我进化
链接：https://arxiv.org/abs/2605.23019

【24】Learnability-Informed Fine-Tuning of Diffusion Language Models
标题：扩散语言模型的可学习性微调
链接：https://arxiv.org/abs/2605.22939

【25】Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Models
标题：代码转换器追踪视觉语言模型中的视觉基础和幻觉
链接：https://arxiv.org/abs/2605.22902

【26】From Residuals to Reasons: LLM-Guided Mechanism Inference from Tabular Data
标题：从残余到原因：LLM引导的表格数据机制推理
链接：https://arxiv.org/abs/2605.22897

【27】Agentic-VLA: Efficient Online Adaptation for Vision-Language-Action Models
标题：统计学VLA：视觉-语言-动作模型的高效在线适应
链接：https://arxiv.org/abs/2605.22896

【28】When Do LLMs Reason? A Dynamical Systems View via Entropy Phase Transitions
标题：LLM什么时候会推理？从熵相变看动力系统
链接：https://arxiv.org/abs/2605.22873

【29】The Readout Shortcut: Positional Number Copying Dominates Arithmetic CoT Readout in Small Language Models
标题：读出时间表：位置数时间表在小型语言模型中主导算术CoT读出
链接：https://arxiv.org/abs/2605.22870

【30】Reading Calibrated Uncertainty from Language Model Trajectories
标题：从语言模型轨迹读取校准的不确定性
链接：https://arxiv.org/abs/2605.22864

【31】PrefBench: Evaluating Zero-Shot LLM Agents in Hidden-Preference Personalized Pricing Negotiations
标题：PreBench：在隐藏偏好个性化定价谈判中评估Zero-ShotLLM代理
链接：https://arxiv.org/abs/2605.22855

【32】LLM Sparsity Prior for Robust Feature Selection
标题：LLM稀疏性优先级，可实现稳健特征选择
链接：https://arxiv.org/abs/2605.23102

【33】MadEvolve: Evolutionary Optimization of Trading Systems with Large Language Models
标题：MadEvolve：具有大型语言模型的交易系统的进化优化
链接：https://arxiv.org/abs/2605.23007

Graph相关(图学习|图神经网络|图优化等)(11篇)

【1】SeedER: Seed-and-Expand Retrieval from Knowledge Graphs
标题：SeedER：从知识图中播种并扩展检索
链接：https://arxiv.org/abs/2605.23753

【2】Contrast to Detect: Dynamic Graph Contrastive Regularization for Unsupervised Anomaly Detection in Multivariate Time Series
标题：检测对比：多元时间序列中无监督异常检测的动态图对比正规化
链接：https://arxiv.org/abs/2605.23744

【3】Graph-based Complexity Forecasts in UK En Route Airspace Using Relevant Aircraft Interactions
标题：使用相关飞机相互作用进行英国途中空域的基于图的复杂性预测
链接：https://arxiv.org/abs/2605.23696

【4】Relevant Walk Search for Explaining Graph Neural Networks
标题：解释图神经网络的相关游动搜索
链接：https://arxiv.org/abs/2605.23673

【5】S$^3$GNN: Efficient Global Mixing and Local Message Passing for Long-Range Graph Learning
标题：S$#3$GNN：用于长期图学习的高效全局混合和本地消息传递
链接：https://arxiv.org/abs/2605.23467

【6】Weisfeiler-Leman Is Incomplete on Simple Spectrum Graphs, so Canonicalize Them
标题：简单谱图的Weisfeiler-Leman是不完全的，应规范化
链接：https://arxiv.org/abs/2605.23446

【7】Reinforcement Learning for Microcanonical Graph Ensemble with Assortativity Constraints
标题：具有组合性约束的微规范图集合的强化学习
链接：https://arxiv.org/abs/2605.23285

【8】Self-supervised Adversarial Purification for Graph Neural Networks
标题：图神经网络的自监督对抗净化
链接：https://arxiv.org/abs/2605.23239

【9】Scalable Heterogeneous Graph Foundation Models for Data-Driven Optimal Power Flow in Smart Grids
标题：智能电网数据驱动最优潮流的可扩展异类图基础模型
链接：https://arxiv.org/abs/2605.23194

【10】SciAtlas: A Large-Scale Knowledge Graph for Automated Scientific Research
标题：SciAtlas：自动化科学研究的大规模知识图谱
链接：https://arxiv.org/abs/2605.22878

【11】Cross-attention-based bipartite graph neural network for coupled nodal and elemental field prediction in large-deformation sheet material forming
标题：基于交叉注意力的二部图神经网络用于大变形板材成形中的节点和元素场耦合预测
链接：https://arxiv.org/abs/2605.22845

Transformer(4篇)

【1】Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers
标题：好的代币搜寻：视觉几何Transformer的代币选择搭便车指南
链接：https://arxiv.org/abs/2605.23892

【2】Training-Free Looped Transformers
标题：免训练环形Transformer
链接：https://arxiv.org/abs/2605.23872

【3】Certification from Examples is Hard for Circuits and Transformers under Minimal Overparametrization
标题：在最小过度参数化下，电路和Transformer很难获得示例认证
链接：https://arxiv.org/abs/2605.22964

【4】Tensor Cache: Eviction-conditioned Associative Memory for Transformers
标题：张量缓存：Transformer的驱逐条件联想记忆
链接：https://arxiv.org/abs/2605.22884

GAN|对抗|攻击|生成相关(7篇)

【1】Adversarial Vulnerability Under Temporal Concept Drift: A Longitudinal Study of Android Malware Detection
标题：时间概念漂移下的对抗漏洞：Android恶意软件检测的纵向研究
链接：https://arxiv.org/abs/2605.23623

【2】Sample-wise Targeted Adversarial Attacks on Test-time Adaptation
标题：对测试时适应的样本有针对性的对抗攻击
链接：https://arxiv.org/abs/2605.23411

【3】Prudent-Banker: No Extra Fees for Baseline Safety in Adversarial Bandits With and Without Delays
标题：审慎银行家：有和没有延迟的敌对盗贼的基线安全无需额外费用
链接：https://arxiv.org/abs/2605.23351

【4】WMAttack: Automated Attack Search for Adversarial Evaluation of World-Model Agents
标题：WMAttack：世界模型代理对抗评估的自动攻击搜索
链接：https://arxiv.org/abs/2605.23220

【5】FastKernels: Benchmarking GPU Kernel Generation in Production
标题：FastKernels：在生产中对图形处理器内核生成进行基准测试
链接：https://arxiv.org/abs/2605.23215

【6】Dithering Defense: Adversarial Robustness of Vision Foundation Models via Multi-Level Floyd-Steinberg Dithering
标题：抖动防御：通过多层浮动-斯坦伯格抖动的视觉对抗鲁棒性基金会模型
链接：https://arxiv.org/abs/2605.23065

【7】Steered Generation via Gradient-Based Optimization on Sparse Query Features
标题：通过对稀疏查询功能的基于对象的优化来引导生成
链接：https://arxiv.org/abs/2605.23040

半/弱/无/有监督|不确定性|主动学习(7篇)

【1】Commutator-Induced Uncertainty in VAEs
标题：VAE中交换机引起的不确定性
链接：https://arxiv.org/abs/2605.23449

【2】Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling
标题：通过半监督伪标签实现标签高效的数据集修剪
链接：https://arxiv.org/abs/2605.23198

【3】Self-Improving In-Context Learning
标题：自我改进的情境学习
链接：https://arxiv.org/abs/2605.23180

【4】Worse than Random: The Importance of a Baseline for Unsupervised Feature Selection
标题：比随机更糟糕：无监督特征选择基线的重要性
链接：https://arxiv.org/abs/2605.22973

【5】Dirichlet-Based Monte Carlo Dropout for Uncertainty Estimation in Neural Networks
标题：神经网络不确定性估计的基于Dirichlet的Monte Carlo Dropout
链接：https://arxiv.org/abs/2605.23635

【6】Selective Ambulance Dispatch Under Contextual Travel-Time Uncertainty
标题：基于出行时间不确定性的救护车选择性调度
链接：https://arxiv.org/abs/2605.23378

【7】Uncertainty-aware classification and triage of structural heart disease using electrocardiography and echocardiography metrics
标题：使用心电图和超声心动图指标对结构性心脏病进行不确定性意识分类和分诊
链接：https://arxiv.org/abs/2605.22968

迁移|Zero/Few/One-Shot|自适应(3篇)

【1】Adaptive Mass-Segmented KV Compression for Long-Context Reasoning
标题：用于长上下文推理的自适应质量分段KV压缩
链接：https://arxiv.org/abs/2605.23200

【2】Robust OT-Guided Generative Residual Domain Adaptation for Bike-Sharing Demand Prediction under Temporal Domain Shift
标题：用于时态变换下自行车共享需求预测的稳健OT引导生成剩余域自适应
链接：https://arxiv.org/abs/2605.23115

【3】FusionSense: Tri-Stage Near-Sensor Learning for Runtime-Adaptive Multimodal Edge Intelligence
标题：FusionSense：用于运行时自适应多模式边缘智能的三阶段近传感器学习
链接：https://arxiv.org/abs/2605.22868

强化学习(7篇)

【1】Less Effort, Shorter Proofs: Reinforcement Learning for Security Protocol Analysis in Tamarin
标题：更少的努力，更短的证明：Tamarin中用于安全协议分析的强化学习
链接：https://arxiv.org/abs/2605.23643

【2】Understanding Goal Generalisation in Sequential Reinforcement Learning
标题：理解顺序强化学习中的目标概括
链接：https://arxiv.org/abs/2605.23565

【3】Reflex: Reinforcement Learning with Reflection Symmetry Exploitation in State-Based Continuous Control
标题：反射：基于状态的连续控制中利用反射对称性的强化学习
链接：https://arxiv.org/abs/2605.23415

【4】Curriculum reinforcement learning with measurable task representation learning
标题：具有可测量任务代表学习的课程强化学习
链接：https://arxiv.org/abs/2605.23372

【5】Pure Exploration for a Good Policy in Reinforcement Learning with Bandit Feedback
标题：纯粹探索Bandit反馈强化学习中的良好政策
链接：https://arxiv.org/abs/2605.23182

【6】Infra-Bayesian Reinforcement Learning Agents Outperform Classical RL For Worst-Case Robustness
标题：在最坏情况下的鲁棒性方面，Q-Bayesian强化学习代理优于经典RL
链接：https://arxiv.org/abs/2605.23146

【7】Classical State Preparation for Variational Quantum Algorithms via Reinforcement Learning
标题：通过强化学习为变分量子算法准备经典状态
链接：https://arxiv.org/abs/2605.23138

元学习(1篇)

【1】Cost-Effective Model Evaluation with Meta-Learning
标题：使用元学习进行经济有效的模型评估
链接：https://arxiv.org/abs/2605.23595

符号|符号学习(1篇)

【1】When Good Equations Get Bad Scores: Improving Symbolic Regression Through Better Parameter Optimization
标题：当好的方程得到坏的分数：通过更好的参数优化改进符号回归
链接：https://arxiv.org/abs/2605.23272

医学相关(1篇)

【1】FederatedRSF : Federated Random Survival Forests for Partially Overlapping Medical Data
标题：FederatedRSF：部分重叠医疗数据的联邦随机生存森林
链接：https://arxiv.org/abs/2605.22954

推荐(2篇)

【1】Expand More, Shrink Less: Shaping Effective-Rank Dynamics for Dense Scaling in Recommendation
标题：扩大更多，缩小更少：塑造推荐中密集扩展的竞争排名动态
链接：https://arxiv.org/abs/2605.23191

【2】Building a privacy-preserving Federated Recommender system for mobile devices
标题：为移动设备构建保护隐私的联合推荐系统
链接：https://arxiv.org/abs/2605.22924

超分辨率|去噪|去模糊|去雾(1篇)

【1】Diffusion-based Denoising Beats Vanilla Score Matching in Parameter Estimation: A Theoretical Explanation
标题：参数估计中基于扩散的去噪优于Vanilla Score匹配的理论解释
链接：https://arxiv.org/abs/2605.22950

自动驾驶|车辆|车道检测等(3篇)

【1】Detecting Drunk Driving Using Off-the-Shelf Smartwatches
标题：使用现成的智能手表检测酒后驾驶
链接：https://arxiv.org/abs/2605.23663

【2】CBANet: A Compact Attention-Based CNN-BiLSTM Network for Aggressive Driving Event Detection
标题：CBANet：一个基于注意力的紧凑型CNN-BiLSTM网络，用于攻击性驾驶事件检测
链接：https://arxiv.org/abs/2605.23471

【3】SpinFlow: A Physics-Informed Spin Field Framework for Traffic Phase Inference and Transition Detection
标题：SpinFlow：一个用于交通相推断和转变检测的物理信息旋转场框架
链接：https://arxiv.org/abs/2605.23306

点云|SLAM|雷达|激光|深度RGBD相关(1篇)

【1】The Implicit Bias of Depth: From Neural Collapse to Softmax Codes
标题：深度的隐性偏差：从神经崩溃到Softmax代码
链接：https://arxiv.org/abs/2605.23087

联邦学习|隐私保护|加密(1篇)

【1】FIRMA: FIbonacci Ring Model Aggregation for Privacy-preserving Federated Learning
标题：FIRIS：用于隐私保护联邦学习的Fibonacci环模型聚合
链接：https://arxiv.org/abs/2605.22898

推理|分析|理解|解释(8篇)

【1】How Hard is it to Rig a Benchmark? A Social Choice Analysis of Leaderboard Robustness
标题：设定基准有多难？排行榜稳健性的社会选择分析
链接：https://arxiv.org/abs/2605.23628

【2】Unextractable Protocol Models: Collaborative Training and Inference without Weight Materialization
标题：不可提取的协议模型：没有权重具体化的协作训练和推理
链接：https://arxiv.org/abs/2605.23464

【3】Convex Compositional Reasoning Models
标题：凸成分推理模型
链接：https://arxiv.org/abs/2605.23395

【4】Understanding and Improving Noisy Embedding Techniques in Instruction Finetuning
标题：理解和改进教学微调中的噪音嵌入技术
链接：https://arxiv.org/abs/2605.23171

【5】Archimedean Copula Inference via Taylor-Mode AD
标题：通过泰勒模式AD的阿基米德Copula推理
链接：https://arxiv.org/abs/2605.23134

【6】Robots That Know What to Ask: Recovering Misaligned Rewards through Targeted Explanations
标题：知道该问什么的机器人：通过有针对性的补偿来恢复失调的回报
链接：https://arxiv.org/abs/2605.22986

【7】KAPLAN: Kolmogorov-Arnold Prognostic Learnable Activation Networks for Survival Analysis
标题：KAPLAN：用于生存分析的Kolmogorov-Arnold预测可学习激活网络
链接：https://arxiv.org/abs/2605.23082

【8】Real-Time Earthquake Magnitude Classification from Initial P-Waves: Models, Dataset, and Comparative Analysis for South Asia
标题：根据初始P波进行实时地震震级分类：南亚的模型、数据集和比较分析
链接：https://arxiv.org/abs/2605.22836

检测相关(4篇)

【1】VACE: Learning Geometrically Structured Representations for Time Series Anomaly Detection
标题：VACE：学习用于时间序列异常检测的几何结构化表示
链接：https://arxiv.org/abs/2605.23504

【2】Convex Low-resource Accent-Robust Language Detection in Speech Recognition
标题：语音识别中的凸低资源口音鲁棒语言检测
链接：https://arxiv.org/abs/2605.23235

【3】CALAD: Channel-Aware contrastive Learning for multivariate time series Anomaly Detection
标题：CARAD：多元时间序列异常检测的队列感知对比学习
链接：https://arxiv.org/abs/2605.23139

【4】MELT: A Behavioral Trace Dataset for High-Risk Memecoin Launch Detection
标题：MELT：用于高风险Memecoin启动检测的行为跟踪数据集
链接：https://arxiv.org/abs/2602.13480

分类|识别(5篇)

【1】Optimal Dimension-Free Sampling for Regularized Classification
标题：用于正规化分类的最佳无干扰抽样
链接：https://arxiv.org/abs/2605.23726

【2】Class-Dependent Hybrid Data Augmentation for Multiclass Migraine Classification under Severe Class Imbalance
标题：严重类别失衡下多类别偏头痛分类的类别相关混合数据增强
链接：https://arxiv.org/abs/2605.23453

【3】Smoothed Elicitation Complexity for Approximate $Γ$-calibration of Discrete Classification Tasks
标题：离散分类任务的大约$¥ $-校准的平滑启发复杂性
链接：https://arxiv.org/abs/2605.23017

【4】Classification of IED-free EEG Responses for Assisted Epilepsy Diagnosis
标题：辅助癫痫诊断的无IED脑电反应分类
链接：https://arxiv.org/abs/2605.22858

【5】JointHRRP-Net: A Statistically Constrained Decoupling Network for Joint Target and Jamming Recognition in Composite Jamming
标题：JointHRRP-Net：一种用于复合干扰中联合目标和干扰识别的统计约束脱钩网络
链接：https://arxiv.org/abs/2605.22857

表征(6篇)

【1】Uncovering the Latent Potential of Deep Intermediate Representations
标题：揭示深度中间代表的潜在潜力
链接：https://arxiv.org/abs/2605.23033

【2】RADAR: Relative Angular Divergence Across Representations
标题：雷达：代表之间的相对角度分歧
链接：https://arxiv.org/abs/2605.23028

【3】Learned Relay Representations for Forward-Thinking Discrete Diffusion Models
标题：前瞻性离散扩散模型的学习中继表示
链接：https://arxiv.org/abs/2605.22967

【4】Human-Centered Learning Mechanics: A Dynamical Framework for Entropy-Regulated Representation Learning
标题：以人为本的学习机制：一个基于熵的再现学习的动态框架
链接：https://arxiv.org/abs/2605.22940

【5】Approximate Machine Unlearning through Manifold Representation Forgetting Guided by Self Mode Connectivity
标题：自模式连通性引导下流形表示遗忘的近似机器学习
链接：https://arxiv.org/abs/2605.22871

【6】PilotWiMAE: Pilot-Native Representation Learning for Wireless Channels
标题：PilotWiMAE：无线通道的Pilot-Native表示学习
链接：https://arxiv.org/abs/2605.22856

优化|敛散性(14篇)

【1】Complete-muE: Optimal Hyperparameter Transfer and Scaling for MoE Models
标题：Complete-muE：MoE模型的最佳超参数传输和缩放
链接：https://arxiv.org/abs/2605.23893

【2】Approaching I/O-optimality for Approximate Attention
标题：接近I/O最佳化以获得大约的注意力
链接：https://arxiv.org/abs/2605.23751

【3】Optimization of randomized neural networks for transfer operator approximation
标题：随机神经网络的转移运算符逼近优化
链接：https://arxiv.org/abs/2605.23689

【4】B-GRTO: Bootstrapped Group Relative Tool Optimization for Referring Segmentation
标题：B-GRTO：用于引用分段的引导组相对工具优化
链接：https://arxiv.org/abs/2605.23500

【5】Automated Random Embedding for Practical Bayesian Optimization with Unknown Effective Dimension
标题：有效维未知的实用Bayesian优化的自动随机嵌入
链接：https://arxiv.org/abs/2605.23473

【6】Hinge Regression Trees and HRT-Boost: Newton-Optimized Oblique Learning for Compact Tabular Models
标题：铰链回归树和HRT-增强：紧凑表格模型的牛顿优化斜向学习
链接：https://arxiv.org/abs/2605.23422

【7】An Open-Source Training Dataset for Foundation Models for Black-box Optimization
标题：黑匣子优化基础模型的开源训练数据集
链接：https://arxiv.org/abs/2605.23417

【8】Coupling-Robust Accuracy in Multiphysics Physics Informed Neural Networks via Kronecker-Preconditioned Optimization
标题：通过克罗内克预条件优化实现多物理场物理信息神经网络中的耦合鲁棒准确性
链接：https://arxiv.org/abs/2605.23391

【9】Score-Based One-step MeanFlow Policy Optimization
标题：基于得分的一步MeanFlow策略优化
链接：https://arxiv.org/abs/2605.23365

【10】Convex Optimization for Alignment and Preference Learning on a Single GPU
标题：在单个图形处理器上进行对齐和偏好学习的凸优化
链接：https://arxiv.org/abs/2605.23244

【11】Lipschitz Optimization for Formal Verification of Homographies
标题：同形表形式验证的Lipschitz优化
链接：https://arxiv.org/abs/2605.23203

【12】Anytime Training with Schedule-Free Spectral Optimization
标题：使用无计划频谱优化随时训练
链接：https://arxiv.org/abs/2605.23061

【13】ImProver 2: Iteratively Self-Improving LMs for Neurosymbolic Proof Optimization
标题：ImProver 2：用于神经符号证明优化的迭代自我改进LM
链接：https://arxiv.org/abs/2605.22885

【14】WeCon: An Efficient Weight-Conditioned Neural Solver for Multi-Objective Combinatorial Optimization Problems
标题：WeCon：多目标组合优化问题的高效权重条件神经求解器
链接：https://arxiv.org/abs/2605.22876

预测|估计(4篇)

【1】Parametric Prior Mapping Framework for Non-stationary Probabilistic Time Series Forecasting
标题：非平稳概率时间序列预测的参数先验映射框架
链接：https://arxiv.org/abs/2605.23402

【2】Assessing Predictive Models for Fairness Based on Movement Patterns
标题：基于运动模式评估公平性预测模型
链接：https://arxiv.org/abs/2605.23234

【3】PaP-NF: Probabilistic Long-Term Time Series Forecasting via Prefix-as-Prompt Reprogramming and Normalizing Flows
标题：PaP-NF：通过以前置为提示重编程和标准化流程进行概率长期时间序列预测
链接：https://arxiv.org/abs/2605.23219

【4】RAG4Outcome: A Retrieval-Augmented Multimodal Framework for Prognostic Prediction in Chronic Osteomyelitis
标题：RAG 4 Outcome：用于慢性骨髓炎预后预测的检索增强多模式框架
链接：https://arxiv.org/abs/2605.22833

其他神经网络|深度学习|模型|建模(31篇)

【1】Leveraging Foundation Models for Causal Generative Modeling
标题：利用基础模型进行因果生成建模
链接：https://arxiv.org/abs/2605.23861

【2】Learning Dynamic Stability Landscapes in Synchronization Networks
标题：同步网络中的动态稳定性景观学习
链接：https://arxiv.org/abs/2605.23708

【3】Learning Through Noise: Why Subliminal Learning Works and When It Fails
标题：通过噪音学习：潜意识学习为何有效以及何时失败
链接：https://arxiv.org/abs/2605.23645

【4】DiLaDiff: Distilled Latent-Augmented Diffusion for Language Modeling
标题：DiLaDiff：用于语言建模的蒸馏潜伏增强扩散
链接：https://arxiv.org/abs/2605.23605

【5】Preisach Attention: A Hysteretic Model of Sequential Memory
标题：Preisach注意力：序列记忆的滞后模型
链接：https://arxiv.org/abs/2605.23603

【6】How Many Training Samples Are Needed for the Inverse Kinematics Solutions by Artificial Neural Networks
标题：人工神经网络逆运动学解需要多少训练样本
链接：https://arxiv.org/abs/2605.23583

【7】Is Dimensionality a Barrier for Retrieval Models?
标题：主观性是检索模型的障碍吗？
链接：https://arxiv.org/abs/2605.23556

【8】Goal-Conditioned Agents that Learn Everything All at Once
标题：一次学习一切的目标条件特工
链接：https://arxiv.org/abs/2605.23551

【9】Precise: SDE-Consistent Stochastic Sampling for RL Post-Training of Flow-Matching Models
标题：精确：流匹配模型RL后训练的SDP一致随机抽样
链接：https://arxiv.org/abs/2605.23522

【10】Learning partially observed systems with neural Hamiltonian ordinary differential equations
标题：用神经Hamilton常微方程学习部分观察系统
链接：https://arxiv.org/abs/2605.23510

【11】Non-normal spectral signatures of instability in neural network training dynamics
标题：神经网络训练动态中不稳定性的非正态谱特征
链接：https://arxiv.org/abs/2605.23476

【12】Learning Individual Dynamics from Sparse Cross-Sectional Snapshots
标题：从稀疏横截快照中学习个体动态
链接：https://arxiv.org/abs/2605.23470

【13】Sparse In-Network Learning via Shortest-Path Backpropagation and Finite-Rate Gating
标题：通过最短路径反向传播和伪速率门控的稀疏网内学习
链接：https://arxiv.org/abs/2605.23424

【14】What Linear Probes Miss: Multi-View Probing for Weight-Space Learning
标题：线性探测器错过了什么：重量空间学习的多视图探测
链接：https://arxiv.org/abs/2605.23410

【15】Hybrid Quantum-Classical Corrective Diffusion Modeling for Meteorological Downscaling
标题：气象缩减的混合量子经典修正扩散模型
链接：https://arxiv.org/abs/2605.23403

【16】Diffusion Domain Expansion: Learning to Coordinate Pre-trained Diffusion Models
标题：扩散域扩展：学习协调预先训练的扩散模型
链接：https://arxiv.org/abs/2605.23275

【17】Learning-Augmented Online Scheduling with Parsimonious Preemption
标题：具有节俭抢占的学习增强在线调度
链接：https://arxiv.org/abs/2605.23255

【18】Enhancing Deep Neural Network Reliability with Refinement and Calibration
标题：通过细化和校准增强深度神经网络的可靠性
链接：https://arxiv.org/abs/2605.23249

【19】Accelerating Divisible Load Processing Through Machine Learning: A Practical Framework for Large-Scale Workloads
标题：通过机器学习加速可分割负载处理：大规模工作负载的实用框架
链接：https://arxiv.org/abs/2605.23247

【20】Encrypted Neural Networks without Overflows
标题：没有溢出的加密神经网络
链接：https://arxiv.org/abs/2605.23096

【21】Model Collapse as Cultural Evolution
标题：文化进化中的模式崩溃
链接：https://arxiv.org/abs/2605.23054

【22】Open Multimodal Datasets and Open-Source Software for Data-Driven Modeling of Multiphase Transport and Thermal Systems
标题：开放多峰数据集和开源软件，用于多相传输和热力系统的数据驱动建模
链接：https://arxiv.org/abs/2605.23037

【23】World Machine: Towards Generative World Modeling for Time-Series
标题：世界机器：走向时间序列的生成式世界建模
链接：https://arxiv.org/abs/2605.23025

【24】MARGIN: Runtime Confidence Calibration for Multi-Agent Foundation Model Coordination
标题：MARGIN：多智能体基础模型协调的工作组置信度校准
链接：https://arxiv.org/abs/2605.22949

【25】Latent Cache Flow: Model-to-Model Communication Without Text
标题：潜在缓存流：没有文本的模型到模型通信
链接：https://arxiv.org/abs/2605.22863

【26】Expressive Power of Deep Homomorphism Networks over Relational Databases
标题：关系数据库上深度同胚网络的表现力
链接：https://arxiv.org/abs/2605.22852

【27】The Misattribution Gap: When Memory Poisoning Looks Like Model Failure in Agentic AI Systems
标题：错误归因差距：当记忆中毒看起来像是抽象人工智能系统中的模型故障时
链接：https://arxiv.org/abs/2605.22842

【28】The physics of AI weather models
标题：人工智能天气模型的物理学
链接：https://arxiv.org/abs/2605.23778

【29】Learning Kernel-Based MDPs from Episodic Preferential Feedback
标题：从情景偏好反馈学习基于核心的MDPs
链接：https://arxiv.org/abs/2605.23650

【30】Operationalizing Individual Fairness via Gradient Descent and Bradley-Terry Models
标题：通过梯度下降和布拉德利-特里模型实现个人公平
链接：https://arxiv.org/abs/2605.23145

【31】VAMP-Diff: VampPrior Latent Diffusion for Photoplethysmography Modeling
标题：Vamp-Diff：VampPrior潜在扩散用于光电容积脉搏波建模
链接：https://arxiv.org/abs/2605.22851

其他(48篇)

【1】CHRONOS: Temporally-Aware Multi-Agent Coordination for Evolving Data Marketplaces
标题：CHRONOS：针对不断发展的数据市场的时间感知多代理协调
链接：https://arxiv.org/abs/2605.23887

【2】Entrywise Error Bounds for Spectral Ranking with Semi-Random Adversaries
标题：具有半随机对手的谱排序的入口误差界
链接：https://arxiv.org/abs/2605.23854

【3】Valid and Expressive Copulas for Irregular Multivariate Time Series
标题：不规则多元时间序列的有效且表达性Copula
链接：https://arxiv.org/abs/2605.23632

【4】HARNESS-LM: A Three-Phase Training Recipe for Harnessing SLMs in Sponsored Search Retrieval
标题：HARNESS-LM：在赞助搜索检索中利用CRM的三阶段训练方案
链接：https://arxiv.org/abs/2605.23572

【5】MARS: Magnitude-Aware Rank Statistics
标题：MARS：幅度感知排名统计
链接：https://arxiv.org/abs/2605.23563

【6】When One Point Is Not Enough: Addressing Ambiguous Instances in Dimensionality Reduction by Splitting
标题：当一点还不够时：通过拆分来解决简化维度中的模糊问题
链接：https://arxiv.org/abs/2605.23540

【7】CoSPlay: Cooperative Self-Play at Test-Time with Self-Generated Code and Unit Test
标题：CoSPlay：使用自生成代码和单元测试在测试时进行合作自玩
链接：https://arxiv.org/abs/2605.23491

【8】Onsager-Machlup Posterior Transport for Deep Gaussian Processes
标题：深高斯过程的Onsager-Machup后验输运
链接：https://arxiv.org/abs/2605.23434

【9】Every Component is a Lookup: Token Attribution and Composition from a Single Decomposition
标题：每个组件都是一个预设：来自单一分解的代币属性和合成
链接：https://arxiv.org/abs/2605.23393

【10】Decoupling Spatio-Temporal Adapter for Fine-Grained Badminton Action Localization
标题：用于细粒度羽毛球动作本地化的时空适配器
链接：https://arxiv.org/abs/2605.23355

【11】Contrastive Distribution Matching for Amortized Sequential Monte Carlo in Discrete Diffusion
标题：离散扩散中摊销顺序蒙特卡罗的对比分布匹配
链接：https://arxiv.org/abs/2605.23346

【12】Multi-Gate Residuals
标题：多门残余
链接：https://arxiv.org/abs/2605.23259

【13】A Simple Plug-in for Improving Eviction-Based KV Cache Compression
标题：一个简单的插件，用于改进基于驱逐的KV缓存压缩
链接：https://arxiv.org/abs/2605.23258

【14】RelPrism: A Multi-Faceted Pre-training Framework with Self-Generated Tasks for Relational Databases
标题：RelPrism：一个面向关系数据库的多方面自生成任务预训练框架
链接：https://arxiv.org/abs/2605.23241

【15】Entropy Equivalence Testing
标题：熵等效性测试
链接：https://arxiv.org/abs/2605.23225

【16】Any-Dimensional Invariant Universality
标题：任意维度不变普遍性
链接：https://arxiv.org/abs/2605.23156

【17】When Determinants Are Not Enough: Private Rare Switching
标题：当决定因素还不够时：私人罕见转换
链接：https://arxiv.org/abs/2605.23131

【18】Exploiting Longitudinal Context in Clinician-Verified Interactive Lesion Tracking
标题：在临床医生验证的互动病变追踪中利用纵向背景
链接：https://arxiv.org/abs/2605.23118

【19】Dreaming Smoothly and Sample Efficiently with Gradient Penalized Latent Dynamics
标题：通过梯度惩罚潜在动力学实现平稳梦想和高效采样
链接：https://arxiv.org/abs/2605.23089

【20】ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention
标题：ThriftAttention：长上下文FP4注意力的选择性混合精确度
链接：https://arxiv.org/abs/2605.23081

【21】Orbax: Distributed Checkpointing with JAX
标题：Orbax：使用JAX的分布式检查点
链接：https://arxiv.org/abs/2605.23066

【22】Millimeter-wave Imaging for Anthropometric Body Measurement
标题：毫米波成像用于人体测量
链接：https://arxiv.org/abs/2605.23064

【23】Decomposing and Measuring Evaluation Awareness
标题：分解和衡量评估意识
链接：https://arxiv.org/abs/2605.23055

【24】The TIME Machine: On The Power of Motion for Efficient Perception
标题：时间机器：运动的力量促进高效感知
链接：https://arxiv.org/abs/2605.23045

【25】The Deterministic Horizon: Impossibility Results as Design Specifications for Trustworthy AI Systems
标题：确定性地平线：不可能结果作为值得信赖的人工智能系统的设计规范
链接：https://arxiv.org/abs/2605.23024

【26】Test-Time Training Undermines Safety Guardrails
标题：考试时间训练破坏了安全护栏
链接：https://arxiv.org/abs/2605.22984

【27】Memorization Dynamics of Fill-in-the-Middle Pretraining
标题：中间填充预训练的小型化动力学
链接：https://arxiv.org/abs/2605.22981

【28】A mathematical theory of balancing relational generalization and memorization
标题：平衡关系概括和记忆的数学理论
链接：https://arxiv.org/abs/2605.22972

【29】SCRIPT: Scalable Diffusion Policy with Multi-stage Training for Language-driven Physics-Based Humanoid Control
标题：可扩展扩散策略与多阶段训练用于存储驱动的基于物理的类人控制
链接：https://arxiv.org/abs/2605.22894

【30】Pointwise Metrics Mislead: An Evaluation Protocol for Multimodal Inverse Problems
标题：逐点搜索误导：多峰逆问题的评估协议
链接：https://arxiv.org/abs/2605.22891

【31】Resilience Characterization of AI-Native Wireless Receivers via Persistent Homology
标题：通过持久同源性对AI原生无线接收器进行弹性表征
链接：https://arxiv.org/abs/2605.22886

【32】Energy per Successful Goal: Goal-Level Energy Accounting for Agentic AI Systems
标题：每个成功目标的能源：大型人工智能系统的目标级能源核算
链接：https://arxiv.org/abs/2605.22883

【33】RMA: an Agentic System for Research-Level Mathematical Problems
标题：LMA：研究级数学问题的统计系统
链接：https://arxiv.org/abs/2605.22875

【34】MedExpMem: Adapting Experience Memory for Differential Diagnosis
标题：MedExpMem：调整经验记忆进行鉴别诊断
链接：https://arxiv.org/abs/2605.22872

【35】FuRA: Full-Rank Parameter-Efficient Fine-Tuning with Spectral Preconditioning
标题：FuRA：具有光谱预处理的全等级参数高效微调
链接：https://arxiv.org/abs/2605.22869

【36】BOHM: Zero-Cost Hierarchical Attribution for Compound AI Systems
标题：BOHM：复合人工智能系统的零成本分层归因
链接：https://arxiv.org/abs/2605.22866

【37】From Simulation to Discovery: AI Enabled Probabilistic Emulation of Mechanistic Crop Systems
标题：从模拟到发现：人工智能支持机械作物系统的概率模拟
链接：https://arxiv.org/abs/2605.22848

【38】On the Stability of Spherical Hellinger-Kantorovich Flows and Their Implications for Differential Privacy
标题：论球形Hellinger-Kantorovich流的稳定性及其对差异隐私的影响
链接：https://arxiv.org/abs/2605.23879

【39】Move on Muon : A Hamiltonian probability gradient flow perspective of Muon optimizer
标题：继续μ on：μ on优化器的汉密尔顿概率梯度流观点
链接：https://arxiv.org/abs/2605.23871

【40】Asymmetric Scaling Laws from Sparse Features
标题：稀疏特征的不对称缩放定律
链接：https://arxiv.org/abs/2605.23591

【41】Accelerating ground state search of spatial photonic Ising machines with genetic-simulated annealing hybrid algorithm
标题：利用遗传模拟模拟算法加速空间量子伊辛机的接地状态搜索
链接：https://arxiv.org/abs/2605.23295

【42】Discontinuous Galerkin Neural Operator for Pathology Defocus Deblurring
标题：用于病理散焦去模糊的不连续Galerkin神经运算器
链接：https://arxiv.org/abs/2605.23282

【43】Coupled Training with Privileged Information and Unlabeled Data
标题：将训练与特权信息和未标记数据相结合
链接：https://arxiv.org/abs/2605.23268

【44】Active Sensing Subserves Task-Level Control
标题：主动感知辅助任务级控制
链接：https://arxiv.org/abs/2605.22988

【45】L-FAME: Longitudinal Focused Attention Meditation EEG Dataset and Benchmark
标题：L-FAME：纵向聚焦注意力冥想脑电数据集和基准
链接：https://arxiv.org/abs/2605.22893

【46】Is TabPFN the Silver Bullet for Insurance Pricing?
标题：TabPFN是保险定价的灵丹妙药吗？
链接：https://arxiv.org/abs/2605.22892

【47】Topological Signal Processing: An Application-Oriented Tutorial
标题：布局信号处理：一种面向应用的工作空间
链接：https://arxiv.org/abs/2605.22853

【48】Evaluating PhaseNet on Teleseismic Data with MsPASS
标题：使用MsPass评估Telesephone数据上的PhaseNet
链接：https://arxiv.org/abs/2605.22837

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