机器学习学术速递[5.19]

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504篇

大模型相关(56篇)

【1】Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation
标题：Vision-OPD：通过策略自蒸馏学习查看多模式LLM的精细细节
链接：https://arxiv.org/abs/2605.18740

【2】Predictable Confabulations: Factual Recall by LLMs Scales with Model Size and Topic Frequency
标题：可预测的虚构：LLM量表的事实回忆，具有模型大小和主题频率
链接：https://arxiv.org/abs/2605.18732

【3】Forecasting Downstream Performance of LLMs With Proxy Metrics
标题：使用代理收件箱预测LLM的下游性能
链接：https://arxiv.org/abs/2605.18607

【4】Prompt2Fingerprint: Plug-and-Play LLM Fingerprinting via Text-to-Weight Generation
标题：Atlant2 Fingerprint：通过文本到权重生成的即插即用LLM指纹识别
链接：https://arxiv.org/abs/2605.18474

【5】Text2CAD-Bench: A Benchmark for LLM-based Text-to-Parametric CAD Generation
标题：文本2CAD-Bench：基于LLM的文本转参数CAD生成的基准
链接：https://arxiv.org/abs/2605.18430

【6】Prune, Update and Trim: Robust Structured Pruning for Large Language Models
标题：修剪、更新和修剪：大型语言模型的稳健结构化修剪
链接：https://arxiv.org/abs/2605.18331

【7】Alignment Dynamics in LLM Fine-Tuning
标题：LLM微调中的对齐动态
链接：https://arxiv.org/abs/2605.18309

【8】Elastic-dLLM: Position Preserving Context Compression and Augmentation of Diffusion LLMs
标题：Elastic-dLLM：位置保持上下文压缩和扩散LLM的增强
链接：https://arxiv.org/abs/2605.18165

【9】LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning
标题：协作多智能体强化学习的LLM引导通信
链接：https://arxiv.org/abs/2605.18077

【10】LogRouter: Adaptive Two-Level LLM Routing for Log Question Answering in Big Data Systems
标题：LogRouter：用于大数据系统中日志问题响应的自适应两级LLM路由
链接：https://arxiv.org/abs/2605.18015

【11】Enhancing the Code Reasoning Capabilities of LLMs via Consistency-based Reinforcement Learning
标题：通过基于一致性的强化学习增强LLM的代码推理能力
链接：https://arxiv.org/abs/2605.17958

【12】HydroAgent: Closing the Gap Between Frontier LLMs and Human Experts in Hydrologic Model Calibration via Simulator-Grounded RL
标题：HydroAgent：通过模拟器接地RL缩小前沿LLM和人类专家在水文模型校准方面的差距
链接：https://arxiv.org/abs/2605.17792

【13】Revisiting the Adam-SGD Gap in LLM Pre-Training: The Role of Large Effective Learning Rates
标题：重新审视LLM预训练中的Adam-Singapore差距：高有效学习率的作用
链接：https://arxiv.org/abs/2605.17787

【14】LLMForge: Multi-Backend Hardware-Aware Neural Architecture Search with Infinite-Head Attention for Edge Language Models
标题：LLMForge：边缘语言模型的多后台硬件感知神经架构搜索
链接：https://arxiv.org/abs/2605.17653

【15】VeriCache: Turning Lossy KV Cache into Lossless LLM Inference
标题：Veriache：将有损的KV缓存转变为有损的LLM推理
链接：https://arxiv.org/abs/2605.17613

【16】How Off-Policy Can GRPO Be? Mu-GRPO for Efficient LLM Reinforcement Learning
标题：GRPO有多不符合政策？Mu-GRPO用于高效的LLM强化学习
链接：https://arxiv.org/abs/2605.17570

【17】Self-Supervised On-Policy Distillation for Reasoning Language Models
标题：推理语言模型的自我监督按策略提炼
链接：https://arxiv.org/abs/2605.17497

【18】DyGRO-VLA: Cross-Task Scaling of Vision-Language-Action Models via Dynamic Grouped Residual Optimization
标题：DyCRO-VLA：通过动态分组剩余优化实现视觉-语言-动作模型的跨任务缩放
链接：https://arxiv.org/abs/2605.17486

【19】WinQ: Accelerating Quantization-Aware Training of Language Models Around Saddle Points
标题：WinQ：加速鞍点周围语言模型的量化感知训练
链接：https://arxiv.org/abs/2605.17471

【20】DP-SelFT: Differentially Private Selective Fine-Tuning for Large Language Models
标题：DP-SelFT：针对大型语言模型的差异化私有选择性微调
链接：https://arxiv.org/abs/2605.17432

【21】Compress the Context, Keep the Commitments: A Formal Framework for Verifiable LLM Context Compression
标题：压缩上下文，遵守承诺：可验证LLM上下文压缩的正式框架
链接：https://arxiv.org/abs/2605.17304

【22】DISA: Offline Importance Sampling for Distribution-Matching LLM-RL
标题：DISA：分布匹配LLM-RL的离线重要性抽样
链接：https://arxiv.org/abs/2605.17295

【23】Step-wise Rubric Rewards for LLM Reasoning
标题：LLM推理的分步规则奖励
链接：https://arxiv.org/abs/2605.17291

【24】LEAP: Learnable End-to-End Adaptive Pruning of Large Language Models
标题：LEAP：大型语言模型的可学习端到端自适应修剪
链接：https://arxiv.org/abs/2605.17289

【25】Charon: A Unified and Fine-Grained Simulator for Large-Scale LLM Training and Inference
标题：Charon：用于大规模LLM训练和推理的统一细粒度模拟器
链接：https://arxiv.org/abs/2605.17164

【26】Contrastive Conceptor Activation Steering (COAST): Unlocking Vision-Language-Action Models through Hidden States
标题：对比概念激活引导（COAST）：通过隐藏状态解锁视觉-语言-动作模型
链接：https://arxiv.org/abs/2605.17144

【27】Capturing LLM Capabilities via Evidence-Calibrated Query Clustering
标题：通过证据校准查询集群捕获LLM能力
链接：https://arxiv.org/abs/2605.17110

【28】HyDRA: Hybrid Dynamic Routing Architecture for Heterogeneous LLM Pools
标题：Hyspel：用于异类LLM收件箱的混合动态路由架构
链接：https://arxiv.org/abs/2605.17106

【29】Scale Determines Whether Language Models Organize Representation Geometry for Prediction
标题：规模决定语言模型是否组织表示几何以进行预测
链接：https://arxiv.org/abs/2605.17084

【30】S-Bus: Automatic Read-Set Reconstruction for Multi-Agent LLM State Coordination
标题：S-Bus：用于多代理LLM状态协调的自动读集重建
链接：https://arxiv.org/abs/2605.17076

【31】The Range Shrinks, the Threat Remains: Re-evaluating LLM Package Hallucinations on the 2026 Frontier-Model Cohort
标题：范围缩小，威胁依然存在：重新评估2026年前沿模型队列的LLM包幻觉
链接：https://arxiv.org/abs/2605.17062

【32】BoLT: A Benchmark to Democratize Black-box Optimization Research for Expensive LLM Tasks
标题：BoLT：将昂贵的LLM任务的黑匣子优化研究民主化的基准
链接：https://arxiv.org/abs/2605.17000

【33】Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation
标题：KL和轨迹脱钩：LLM蒸馏中SFT、DAger、离线RL和OPD的统一视角
链接：https://arxiv.org/abs/2605.16826

【34】Confidence Geometry Reveals Trace-Level Correctness in Large Language Model Reasoning
标题：置信度几何揭示大型语言模型推理中的踪迹级正确性
链接：https://arxiv.org/abs/2605.16824

【35】The Unlearnability Phenomenon in RLVR for Language Models
标题：语言模型RL VR中的不可学习现象
链接：https://arxiv.org/abs/2605.16787

【36】Lever: Speculative LLM Inference on Smartphones
标题：杠杆：关于智能手机的推测性LLM推理
链接：https://arxiv.org/abs/2605.16786

【37】Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning
标题：可区分删除：统一大型语言模型取消学习的知识擦除和拒绝
链接：https://arxiv.org/abs/2605.16776

【38】State Contamination in Memory-Augmented LLM Agents
标题：内存增强LLM代理中的状态污染
链接：https://arxiv.org/abs/2605.16746

【39】Scalable Knowledge Editing for Mixture-of-Experts LLMs via Tensor-Structured Updates
标题：通过张量结构化更新进行专家混合LLM的可扩展知识编辑
链接：https://arxiv.org/abs/2605.16686

【40】Right Predictions, Misleading Explanations: On the Vulnerability of Vision-Language Model Explanations
标题：正确的预测，误导性的解释：论视觉语言模型解释的脆弱性
链接：https://arxiv.org/abs/2605.16651

【41】Peak-Detector: Explainable Peak Detection via Instruction-Tuned Large Language Models in Physiological Sign
标题：Peak-Detector：可解释的生理信号峰值检测
链接：https://arxiv.org/abs/2605.16452

【42】Membership Inference Attacks on Discrete Diffusion Language Models
标题：离散扩散语言模型的隶属度推理攻击
链接：https://arxiv.org/abs/2605.16445

【43】A Theory of Training Profit-Optimal LLMs
标题：训练利润最佳法学硕士的理论
链接：https://arxiv.org/abs/2605.16430

【44】Reducing Hallucination in Vision-Language Models via Stage-wise Preference Optimization under Distribution Shift
标题：分布移位下基于阶段偏好优化的视觉-语言模型中的幻觉抑制
链接：https://arxiv.org/abs/2605.16411

【45】Multilingual OCR-Aware Fine-Tuning and Prompt-Guided Chain-of-Thought Reasoning for Multimodal Large Language Models
标题：多模式大型语言模型的多语言OCR感知微调和预算引导思维链推理
链接：https://arxiv.org/abs/2605.16409

【46】Mixing Times of Glauber Dynamics on Masked Language Models
标题：掩蔽语言模型上Glauber动力学的混合时间
链接：https://arxiv.org/abs/2605.16378

【47】CheckSupport: A Local LLM-Powered Tool for Automated Manuscript Submission Checklist Selection and Completion
标题：CheckSupport：一个本地LLM-Powered工具，用于自动选择和完成Mannipt提交检查表
链接：https://arxiv.org/abs/2605.16377

【48】ProxyKV: Cross-Model Proxy Pruning for Efficient Long-Context LLM Inference
标题：ProxyKN：跨模型代理修剪，以实现高效的长上下文LLM推理
链接：https://arxiv.org/abs/2605.16360

【49】Augmenting Human Evaluation with LLM Judges: How Many Human Reviews Do You Need?
标题：通过LLM评委加强人性评估：您需要多少个人性评估？
链接：https://arxiv.org/abs/2605.16354

【50】HPC-LLM: Practical Domain Adaptation and Retrieval-Augmented Generation for HPC Support
标题：HPC-LLM：用于高性能计算支持的实用领域适应和检索增强生成
链接：https://arxiv.org/abs/2605.16347

【51】PropGuard: Safeguarding LLM-MAS via Propagation-Aware Exploration and Remediation
标题：PropGuard：通过认知探索和修复保护LLM-MAS
链接：https://arxiv.org/abs/2605.16346

【52】Goal-Conditioned Supervised Learning for LLM Fine-Tuning
标题：LLM微调的目标条件监督学习
链接：https://arxiv.org/abs/2605.16345

【53】DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language Models
标题：DACA-GRPO：扩散语言模型中强化学习的去感知信用分配
链接：https://arxiv.org/abs/2605.16342

【54】ANNEAL: Adapting LLM Agents via Governed Symbolic Patch Learning
标题：安妮：通过管辖符号补丁学习适应LLM代理
链接：https://arxiv.org/abs/2605.16309

【55】Helping Customers in Distress: An LLM-powered Agent that Converses, Probes, and Routes
标题：帮助陷入困境的客户：一个由LLM支持的代理，能够对话、探测和路线
链接：https://arxiv.org/abs/2605.16268

【56】Estimating Item Difficulty with Large Language Models as Experts
标题：以大型语言模型作为专家估计项目难度
链接：https://arxiv.org/abs/2605.18562

Graph相关(图学习|图神经网络|图优化等)(17篇)

【1】An Approximation Algorithm for Graph Label Selection
标题：图标号选择的一个近似算法
链接：https://arxiv.org/abs/2605.18623

【2】S2Aligner: Pair-Efficient and Transferable Pre-Training for Sparse Text-Attributed Graphs
标题：S2 Aligner：稀疏文本属性图的成对高效且可转移的预训练
链接：https://arxiv.org/abs/2605.18579

【3】Graph Hierarchical Recurrence for Long-Range Generalization
标题：用于长期推广的图分层回归
链接：https://arxiv.org/abs/2605.18387

【4】Dynamic Elliptical Graph Factor Models via Riemannian Optimization with Geodesic Temporal Regularization
标题：基于测地时间正则化黎曼优化的动态椭圆图因子模型
链接：https://arxiv.org/abs/2605.18316

【5】FLAG: Foundation model representation with Latent diffusion Alignment via Graph for spatial gene expression prediction
标题：FLAG：通过图形进行潜在扩散对齐的基础模型表示，用于空间基因表达预测
链接：https://arxiv.org/abs/2605.18055

【6】RL4RLA: Teaching ML to Discover Randomized Linear Algebra Algorithms Through Curriculum Design and Graph-Based Search
标题：RL 4 RLA：通过课程设计和基于图的搜索教授ML发现随机线性代数算法
链接：https://arxiv.org/abs/2605.18004

【7】Function graph transformers universally approximate operators between function spaces
标题：函数图变换函数空间之间的普适逼近运算符
链接：https://arxiv.org/abs/2605.17968

【8】Heterogeneous Information-Bottleneck Coordination Graphs for Multi-Agent Reinforcement Learning
标题：多智能体强化学习的异类信息瓶颈协调图
链接：https://arxiv.org/abs/2605.17393

【9】Learning Fill-in Reduction Ordering via Graph Policy Optimization for Sparse Matrices
标题：通过稀疏矩阵的图策略优化学习填充约简排序
链接：https://arxiv.org/abs/2605.17362

【10】Tensor Channel Equivariant Graph Neural Networks for Molecular Polarizability Prediction
标题：分子极化率预测的张量通道等变图神经网络
链接：https://arxiv.org/abs/2605.16891

【11】Plan First, Diffuse Later: Extrinsic Graph Guidance for Long-Horizon Diffusion Planning
标题：先计划，后扩散：长期扩散规划的外部图指南
链接：https://arxiv.org/abs/2605.16863

【12】Universal Graph Backdoor Defense: A Feature-based Homophily Perspective
标题：通用图后门防御：基于冲突的同质性观点
链接：https://arxiv.org/abs/2605.16815

【13】Informative Graph Structure Learning
标题：信息性图结构学习
链接：https://arxiv.org/abs/2605.16809

【14】GraViti: Graph-Level Variational Autoencoders with Relaxed Permutation Invariance
标题：GraViti：具有宽松排列不变性的图形级变分自动编码器
链接：https://arxiv.org/abs/2605.16668

【15】LARGER: Lexically Anchored Repository Graph Exploration and Retrieval
标题：LARGER：词汇锚定知识库图形探索和检索
链接：https://arxiv.org/abs/2605.16352

【16】Bi-Level Chaotic Fusion Based Graph Convolutional Network for Stock Market Prediction Interval
标题：基于双层混乱融合的图卷积网络的股市预测区间
链接：https://arxiv.org/abs/2605.16324

【17】AdaGraph: A Graph-Native Clustering Algorithm That Overcomes the Curse of Dimensionality and Enables Scientific Discovery
标题：AdaShape：一种图形原生集群算法，克服了抽象性的诅咒并实现科学发现
链接：https://arxiv.org/abs/2605.16320

Transformer(14篇)

【1】The Expressive Power of Low Precision Softmax Transformers with (Summarized) Chain-of-Thought
标题：具有（总结）思想链的低精度SoftmaxTransformer的表现力
链接：https://arxiv.org/abs/2605.18079

【2】InfoFlow: A Framework for Multi-Layer Transformer Analysis
标题：InfoFlow：多层Transformer分析框架
链接：https://arxiv.org/abs/2605.17930

【3】AdaptiveLoad: Towards Efficient Video Diffusion Transformer Training
标题：AdaptiveLoad：实现高效的视频扩散Transformer训练
链接：https://arxiv.org/abs/2605.17923

【4】One Model, Two Roles: Emergent Specialization in a Shared Recurrent Transformer
标题：一个模型，两个角色：共享循环Transformer的紧急专业化
链接：https://arxiv.org/abs/2605.17811

【5】FishBack: Pullback Fisher Geometry for Optimal Activation Steering in Transformers
标题：FishBack：回撤Fisher几何，在Transformer中实现最佳激活转向
链接：https://arxiv.org/abs/2605.17231

【6】OPTNet: Ordering Point Transformer Network for Post-disaster 3D Semantic Segmentation
标题：OPTNet：用于灾后三维语义分割的有序点Transformer网络
链接：https://arxiv.org/abs/2605.17197

【7】Transformer-Based MCS Prediction for 5G Multicast-Broadcast Services (MBS)
标题：5G多播-广播服务（MBS）基于转换器的CS预测
链接：https://arxiv.org/abs/2605.16735

【8】DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers
标题：DiRotQ：4位扩散Transformer的旋转感知量化
链接：https://arxiv.org/abs/2605.16732

【9】Where Pretraining writes and Alignment reads: the asymmetry of Transformer weight space
标题：Pretraining写入和对齐读取：Transformer重量空间的不对称性
链接：https://arxiv.org/abs/2605.16600

【10】Attention-Aware Transformer-Based Aggregation Network for Video Periocular Recognition
标题：基于注意力感知转换器的聚合网络用于视频眼周识别
链接：https://arxiv.org/abs/2605.16550

【11】Inducing Spatial Locality in Vision Transformers through the Training Protocol
标题：通过训练协议在视觉变形器中诱导空间局部性
链接：https://arxiv.org/abs/2605.16390

【12】LoopQ: Quantization for Recursive Transformers
标题：LoopQ：回归Transformer的量化
链接：https://arxiv.org/abs/2605.16343

【13】Forecasting Medium-Horizon Alzheimer's Disease Progression: Residual Gap-Aware Transformers for 24-Month CDR-SB Change from ADNI Clinical and Biomarker Histories
标题：预测中期阿尔茨海默病进展：ADNI临床和生物标志物历史中24个月CDR-SB变化的残留差距意识变形者
链接：https://arxiv.org/abs/2605.16319

【14】Training Infinitely Deep and Wide Transformers
标题：训练无限深宽Transformer
链接：https://arxiv.org/abs/2605.17660

GAN|对抗|攻击|生成相关(27篇)

【1】A No-Defense Defense Against Gradient-Based Adversarial Attacks on ML-NIDS: Is Less More?
标题：针对ML-NIDS基于攻击的无防御：少是多吗？
链接：https://arxiv.org/abs/2605.18666

【2】KairosHope: A Next-Generation Time-Series Foundation Model for Specialized Classification via Dual-Memory Architecture
标题：KairosHope：通过双内存架构进行专业分类的下一代时间序列基础模型
链接：https://arxiv.org/abs/2605.18657

【3】Generative Adversarial Learning from Deterministic Processes
标题：来自确定性过程的生成性对抗学习
链接：https://arxiv.org/abs/2605.18425

【4】Generating Physically Consistent Molecules with Energy-Based Models
标题：用基于能量的模型生成物理一致的分子
链接：https://arxiv.org/abs/2605.18381

【5】Attacking the First-Principle: A Black-Box, Query-Free Targeted Mimicry Attack on Binary Function Classifiers
标题：攻击第一原则：对二进制函数分类器的黑匣子、无查询有针对性的模仿攻击
链接：https://arxiv.org/abs/2605.18231

【6】Interaction-Breaking Adversarial Learning Framework for Robust Multi-Agent Reinforcement Learning
标题：用于鲁棒多智能体强化学习的打破交互的对抗学习框架
链接：https://arxiv.org/abs/2605.18024

【7】Universal Adversarial Triggers
标题：普遍对抗触发
链接：https://arxiv.org/abs/2605.17936

【8】DCFold: Efficient Protein Structure Generation with Single Forward Pass
标题：DCFold：高效的蛋白质结构生成，单次正向传递
链接：https://arxiv.org/abs/2605.17899

【9】Generating Pretraining Tokens from Organic Data for Data-Bound Scaling
标题：从有机数据生成预训练令牌以进行数据绑定扩展
链接：https://arxiv.org/abs/2605.17849

【10】TabKDE: Simple and Scalable Tabular Data Generation with Kernel Density Estimates
标题：TabTEK：具有核密度估计的简单且可扩展的表格数据生成
链接：https://arxiv.org/abs/2605.17642

【11】SynVA: A Modular Toolkit for Vessel Generation and Aneurysm Editing
标题：SynVA：血管生成和动脉瘤编辑的模块化工具包
链接：https://arxiv.org/abs/2605.17620

【12】PFlow-T: A Persistence-Driven Forward Process for Topology-Controlled Generation
标题：PFlow-T：一种持久驱动的用于布局控制生成的正向过程
链接：https://arxiv.org/abs/2605.17555

【13】CasualSynth: Generating Structurally Sound Synthetic Data
标题：CasualSynth：生成结构合理的合成数据
链接：https://arxiv.org/abs/2605.17528

【14】The Silent Brush: Evaluating Artistic Style Leakage in AI Art Generation
标题：无声的画笔：评估人工智能艺术生成中的艺术风格泄露
链接：https://arxiv.org/abs/2605.17500

【15】A2RBench: An Automatic Paradigm for Formally Verifiable Abstract Reasoning Benchmark Generation
标题：A2RBench：一个自动生成形式可验证抽象推理基准的范例
链接：https://arxiv.org/abs/2605.17278

【16】Latency-Aware Deep Learning Benchmark for Real-Time Cyber-Physical Attack and Fault Classification in Inverter-Dominated Power Grids
标题：用于逆变器主导电网中实时网络物理攻击和故障分类的延迟感知深度学习基准
链接：https://arxiv.org/abs/2605.17256

【17】Privacy-Preserving Generation Fraud Detection for Distributed Photovoltaic Systems: A Solar Irradiance-Fused Federated Learning Framework
标题：分布式太阳能电池系统的保护隐私发电欺诈检测：太阳辐射融合联邦学习框架
链接：https://arxiv.org/abs/2605.17039

【18】Privacy Policy Enforcement Guardrails for Data-Sensitive Retrieval-Augmented Generation
标题：数据敏感检索增强一代的隐私政策执行护栏
链接：https://arxiv.org/abs/2605.17034

【19】Adversarial Fragility and Language Vulnerability in Clinical AI: A Systematic Audit of Diagnostic Collapse Under Imperceptible Perturbations and Cross-Lingual Drift in Low-Resource Healthcare Settings
标题：临床人工智能中的对抗脆弱性和语言脆弱性：对低资源医疗保健环境中难以察觉的扰动和跨舌漂移下诊断崩溃的系统审计
链接：https://arxiv.org/abs/2605.16993

【20】AIM: Adversarial Information Masking for Faithfulness Evaluation of Saliency Maps
标题：目的：对抗信息掩蔽显着性地图的可信度评估
链接：https://arxiv.org/abs/2605.16905

【21】Genflow Ad Studio: A Compound AI Architecture for Brand-Aligned, Self-Correcting Video Generation
标题：GenFlow Ad Studio：用于品牌一致、自我纠正视频生成的复合人工智能架构
链接：https://arxiv.org/abs/2605.16748

【22】Compositional Adversarial Training for Robust Visual Watermarking
标题：鲁棒视觉水印的合成对抗训练
链接：https://arxiv.org/abs/2605.16720

【23】Physics-Guided Geometric Diffusion for Macro Placement Generation
标题：用于宏布局生成的物理引导几何扩散
链接：https://arxiv.org/abs/2605.16451

【24】Video Reconstruction using Diffusion-based Image-to-Video Generation with Trajectory Guidance
标题：基于扩散的轨迹制导图像到视频生成的视频重建
链接：https://arxiv.org/abs/2605.16420

【25】When Actions Disappear: Adversarial Action Removal in Self-Play Reinforcement Learning
标题：当动作消失时：自玩强化学习中的对抗动作去除
链接：https://arxiv.org/abs/2605.16312

【26】Accelerating Redshift-Conditioned Galaxy Image Synthesis with One-step Generative Modeling
标题：通过一步生成建模加速红移条件下的银河图像合成
链接：https://arxiv.org/abs/2605.17546

【27】HYVINT: Intensity-Driven Hypergraph Generation with Variational Representations
标题：HyVint：具有变分表示的强度驱动超图生成
链接：https://arxiv.org/abs/2605.16836

半/弱/无/有监督|不确定性|主动学习(19篇)

【1】An Assessment of Human vs. Model Uncertainty in Soft-Label Learning and Calibration
标题：软标签学习和校准中人类与模型不确定性的评估
链接：https://arxiv.org/abs/2605.18648

【2】Self-supervised local learning rules learn the hidden hierarchical structure of high-dimensional data
标题：自我监督的本地学习规则学习多维数据的隐藏分层结构
链接：https://arxiv.org/abs/2605.18557

【3】Lost in the Folds: When Cross-Validation Is Not a Deep Ensemble for Uncertainty Estimation
标题：迷失在折叠中：当交叉验证不是不确定性估计的深度包围时
链接：https://arxiv.org/abs/2605.18329

【4】Uncertainty Reliability Under Domain Shift: An Investigation for Data-Driven Blood Pressure Estimation in Photoplethysmography
标题：域转移下的不确定性可靠性：光体积脉搏成像中数据驱动血压估计的研究
链接：https://arxiv.org/abs/2605.18008

【5】When Accuracy Is Not Enough: Uncertainty Collapse between Noisy Label Learning and Out-of-Distribution Detection
标题：当精度不够时：噪声标签学习和分布外检测之间的不确定性崩溃
链接：https://arxiv.org/abs/2605.17795

【6】Uncertainty-Calibrated Recommendations for Low-Active Users
标题：针对低活跃用户的不确定性校准建议
链接：https://arxiv.org/abs/2605.17788

【7】Multi-task learning on partially labeled datasets via invariant/equivariant semi-supervised learning
标题：通过不变/等变半监督学习在部分标记数据集上进行多任务学习
链接：https://arxiv.org/abs/2605.17624

【8】Learning Displacement-Robust Representations for Landslide Early Warning under Rainfall Forecast Uncertainty
标题：降雨预测不确定性下滑坡预警的学习位移鲁棒表示
链接：https://arxiv.org/abs/2605.17419

【9】Self-Supervised Learning for Sparse Matrix Reordering
标题：稀疏矩阵重排序的自我监督学习
链接：https://arxiv.org/abs/2605.17403

【10】UNR-Explainer: Counterfactual Explanations for Unsupervised Node Representation Learning Models
标题：对象-解释者：无监督节点表示学习模型的反事实解释
链接：https://arxiv.org/abs/2605.17285

【11】D$^2$Evo: Dual Difficulty-Aware Self-Evolution for Data-Efficient Reinforcement Learning
标题：D $' 2$Evo：双重困难感知自我进化，实现数据高效强化学习
链接：https://arxiv.org/abs/2605.17037

【12】VolTA-3D: Self-Supervised Learning for Brain MRI using 3D Volumetric Token Alignment
标题：VolTA-3D：使用3D体积标记对齐的脑MRI自我监督学习
链接：https://arxiv.org/abs/2605.16775

【13】Automatic Unsupervised Ensemble Outlier Model Selection--Extended Version
标题：自动无监督参与离群值模型选择--扩展版本
链接：https://arxiv.org/abs/2605.16567

【14】Multiscale Supervised Unbalanced Optimal Transport Flow Matching
标题：多尺度有监督的不平衡最优交通流匹配
链接：https://arxiv.org/abs/2605.16529

【15】Lost or Hidden? A Concept-Level Forgetting in Supervised Continual Learning
标题：丢失还是隐藏？监督式持续学习中的概念层面遗忘
链接：https://arxiv.org/abs/2605.16374

【16】Cross-Source Supervision for Bone Infection Segmentation in Dual-Modality PET-CT
标题：双模式PET-CT中骨感染分割的跨源监督
链接：https://arxiv.org/abs/2605.16373

【17】Federated Nested Learning: Collaborative Training of Self-Referential Memories for Test-Time Adaptation
标题：联合巢式学习：自我参考记忆的协作训练以适应测试时间
链接：https://arxiv.org/abs/2605.16350

【18】A Structural Threshold in Decision Capacity Governs Collapse in Self-Play Reinforcement Learning
标题：决策能力的结构性阈值导致自玩强化学习的崩溃
链接：https://arxiv.org/abs/2605.16315

【19】Diffusion-Based Stochastic Operator Networks for Uncertainty Quantification in Stochastic Partial Differential Equations
标题：基于扩散的随机运营商网络用于随机偏微方程中不确定性量化
链接：https://arxiv.org/abs/2605.17107

迁移|Zero/Few/One-Shot|自适应(18篇)

【1】DashAttention: Differentiable and Adaptive Sparse Hierarchical Attention
标题：DashAttention：可区分且自适应的稀疏分层注意力
链接：https://arxiv.org/abs/2605.18753

【2】Adaptive Experimentation for Censored Survival Outcomes
标题：审查生存结果的适应性实验
链接：https://arxiv.org/abs/2605.18459

【3】MARR: Module-Adaptive Residual Reconstruction for Low-Bit Post-Training Quantization
标题：MARR：用于低位训练后量化的模块自适应残余重建
链接：https://arxiv.org/abs/2605.17997

【4】Transfer Learning for Customized Car Racing Environments
标题：定制赛车环境的迁移学习
链接：https://arxiv.org/abs/2605.17928

【5】AMO: Adaptive Muon Orthogonalization
标题：AMO：自适应μ子同步化
链接：https://arxiv.org/abs/2605.17806

【6】Adaptive Generate-Rank-Verify: Inference-Time Search with Costly Verification
标题：自适应生成-排名-验证：具有昂贵验证的推断时搜索
链接：https://arxiv.org/abs/2605.17609

【7】Q-LocalAdam: Memory-Efficient Client-Side Adaptive Optimization for Edge Federated Learning
标题：Q-LocalAdam：边缘联邦学习的内存高效客户端自适应优化
链接：https://arxiv.org/abs/2605.17552

【8】Few-Shot Network Intrusion Detection Using Online Triplet Mining
标题：使用在线三重挖掘的Few-Shot网络入侵检测
链接：https://arxiv.org/abs/2605.17530

【9】\textsc{MasFACT}: Continual Multi-Agent Topology Learning via Geometry-Aware Posterior Transfer
标题： extSC{MasFACT}：通过几何感知后验传输进行连续多代理拓扑学习
链接：https://arxiv.org/abs/2605.17361

【10】Towards Principled Test-Time Adaptation for Time Series Forecasting
标题：时间序列预测的测试时间自适应原则
链接：https://arxiv.org/abs/2605.17250

【11】Anytime and Difficulty-Adaptive PAC-Bayes for Constrained Density-Ratio Network with Continual Learning Guarantees
标题：具有连续学习保证的约束密度比网络的随时和困难自适应PAC-Bayes
链接：https://arxiv.org/abs/2605.17212

【12】UB-SMoE: Universally Balanced Sparse Mixture-of-Experts for Resource-adaptive Federated Fine-tuning of Foundation Models
标题：UB-SMoE：普遍平衡的稀疏专家混合，用于基础模型的资源自适应联邦微调
链接：https://arxiv.org/abs/2605.16690

【13】Sustainable Intelligence for the Wild: Democratizing Ecological Monitoring via Knowledge-Adaptive Edge Expert Agents
标题：野生可持续智能：通过知识自适应边缘专家代理实现生态监测民主化
链接：https://arxiv.org/abs/2605.16671

【14】How Few-Shot Examples Add Up: A Causal Decomposition of Function Vectors in In-Context Learning
标题：很少的例子如何相加：上下文学习中功能载体的因果分解
链接：https://arxiv.org/abs/2605.16591

【15】Strategic Over-Parameterization for Generalizable Low-Rank Adaptation
标题：可推广低等级适应的战略过度参数化
链接：https://arxiv.org/abs/2605.16470

【16】CADS: Conformal Adaptive Decision System for Cost-Efficient Image Classification
标题：CADS：具有成本效益的图像分类的保形自适应决策系统
链接：https://arxiv.org/abs/2605.16401

【17】Can Adaptive Gradient Methods Converge under Heavy-Tailed Noise? A Case Study of AdaGrad
标题：自适应梯度方法在重尾噪音下能否收敛？AdaGrad案例研究
链接：https://arxiv.org/abs/2605.18694

【18】Longwang: Zero-Shot Global Spatiotemporal Precipitation Downscaling with a Latent Generative Prior
标题：龙旺：Zero-Shot全球时空降水量与潜在的生成先验性缩减
链接：https://arxiv.org/abs/2605.17603

强化学习(17篇)

【1】General Preference Reinforcement Learning
标题：一般偏好强化学习
链接：https://arxiv.org/abs/2605.18721

【2】AMARIS: A Memory-Augmented Rubric Improvement System for Rubric-Based Reinforcement Learning
标题：AMARIS：一个用于基于条目的强化学习的内存增强条目改进系统
链接：https://arxiv.org/abs/2605.18592

【3】Scheduling That Speaks: An Interpretable Programmatic Reinforcement Learning Framework
标题：说话的调度：一个可解释的程序强化学习框架
链接：https://arxiv.org/abs/2605.18454

【4】Modelling Customer Trajectories with Reinforcement Learning for Practical Retail Insights
标题：利用强化学习对客户轨迹建模以实现实际零售洞察
链接：https://arxiv.org/abs/2605.18449

【5】Heterogeneous Tasks Offloading in Vehicular Edge Computing: A Federated Meta Deep Reinforcement Learning Approach
标题：车辆边缘计算中的异类任务卸载：联邦Meta深度强化学习方法
链接：https://arxiv.org/abs/2605.18437

【6】Beyond Inference-Time Search: Reinforcement Learning Synthesizes Reusable Solvers
标题：超越推理时搜索：强化学习合成可重复使用的求解器
链接：https://arxiv.org/abs/2605.18374

【7】ISEP: Implicit Support Expansion for Offline Reinforcement Learning via Stochastic Policy Optimization
标题：ISEP：基于随机策略优化的离线强化学习隐式支持度扩展
链接：https://arxiv.org/abs/2605.18320

【8】Privacy Preserving Reinforcement Learning with One-Sided Feedback
标题：具有单边反馈的隐私保护强化学习
链接：https://arxiv.org/abs/2605.18246

【9】ClaHF: A Human Feedback-inspired Reinforcement Learning Framework for Improving Classification Tasks
标题：ClaHF：一个基于人类反馈的强化学习框架，用于改进分类任务
链接：https://arxiv.org/abs/2605.17458

【10】Progressive Generalization Augmentation with Deeply Coupled RND-PPO and Domain-Prioritized Noise Injection for Robust Crop Management Reinforcement Learning
标题：深度耦合RND-PPO和领域优先级噪音注入的渐进概括增强，用于鲁棒的作物管理强化学习
链接：https://arxiv.org/abs/2605.17428

【11】Leveraging Error Diversity in Group Rollouts for Reinforcement Learning
标题：利用群组滚动中的错误多样性进行强化学习
链接：https://arxiv.org/abs/2605.17333

【12】From Imitation to Interaction: Mastering Game of Schnapsen with Shallow Reinforcement Learning
标题：从模仿到互动：用浅层强化学习掌握施纳普森游戏
链接：https://arxiv.org/abs/2605.17162

【13】Ranking-Aware Calibration for Reliable Multimodal Reinforcement Learning
标题：可靠的多模式强化学习的排名感知校准
链接：https://arxiv.org/abs/2605.16999

【14】QuantFPFlow: Quantum Amplitude Estimation for Fokker--Planck Policy Optimisation in Continuous Reinforcement Learning
标题：QuantFPFlow：Fokker的量子幅度估计--连续强化学习中的普朗克政策优化
链接：https://arxiv.org/abs/2605.16429

【15】Investigating Action Encodings in Recurrent Neural Networks in Reinforcement Learning
标题：研究强化学习中的回归神经网络中的动作编码
链接：https://arxiv.org/abs/2605.16318

【16】CPMobius: Iterative Coach-Player Reasoning for Data-Free Reinforcement Learning
标题：CPMobius：用于无数据强化学习的迭代教练-球员推理
链接：https://arxiv.org/abs/2602.02979

【17】Deep Reinforcement Learning Framework for Diversified Portfolio Management Across Global Equity Markets
标题：全球股票市场多元化投资组合管理的深度强化学习框架
链接：https://arxiv.org/abs/2605.17307

元学习(4篇)

【1】Efficient Bilevel Optimization for Meta Label Correction in Noisy Label Learning
标题：噪声标签学习中Meta标签校正的有效双层优化
链接：https://arxiv.org/abs/2605.17833

【2】Verifier-Guided Code Translation via Meta-Step Decoding
标题：通过元步解码的验证器引导代码翻译
链接：https://arxiv.org/abs/2605.17626

【3】Boundedly Rational Meta-Learning in Sequential Consumer Choice
标题：消费者顺序选择中的有界理性元学习
链接：https://arxiv.org/abs/2605.16532

【4】Seeking the Unfamiliar but Memorable: Conceptual Creativity as Meta-Learning
标题：寻找陌生但难忘的事物：作为元学习的概念创造力
链接：https://arxiv.org/abs/2605.16477

符号|符号学习(2篇)

【1】Concise and Logically Consistent Conformal Sets for Neuro-Symbolic Concept-Based Models
标题：基于神经符号概念的模型的简洁且逻辑一致的保形集
链接 ：https://arxiv.org/abs/2605.18202

【2】Quantitative Linear Logic for Neuro-Symbolic Learning and Verification
标题：用于神经符号学习和验证的定量线性逻辑
链接：https://arxiv.org/abs/2605.13845

分层学习(1篇)

【1】Learning Multi-Timescale Abstractions for Hierarchical Combinatorial Planning
标题：分层组合规划的多时标抽象学习
链接：https://arxiv.org/abs/2605.17058

医学相关(11篇)

【1】Beyond Morphology: Quantifying the Diagnostic Power of Color Features in Cancer Classification
标题：超越形态学：量化颜色特征在癌症分类中的诊断能力
链接：https://arxiv.org/abs/2605.18522

【2】SIREM: Speech-Informed MRI Reconstruction with Learned Sampling
标题：SIREM：使用习得采样的语音知情MRI重建
链接：https://arxiv.org/abs/2605.18221

【3】Domain Incremental Learning for Pandemic-Resilient Chest X-Ray Analysis
标题：用于抗流行性胸部X射线分析的领域增量学习
链接：https://arxiv.org/abs/2605.17729

【4】How Do Electrocardiogram Models Scale?
标题：心电图模型如何缩放？
链接：https://arxiv.org/abs/2605.17276

【5】Extending Pretrained 10-Second ECG Foundation Models to Longer Horizons
标题：将预训练的10秒心电图基础模型扩展到更长的视野
链接：https://arxiv.org/abs/2605.16975

【6】PhysioSeq2Seq: A Hybrid Physiological Digital Twin and Sequence-to-Sequence LSTM for Long-Horizon Glucose Forecasting in Type 1 Diabetes
标题：PhysoSeq 2Seq：混合生理数字双胞胎和序列到序列LSTM，用于1型糖尿病的长视野血糖预测
链接：https://arxiv.org/abs/2605.16860

【7】MedMIX: Modality-Internal Expert Fusion for Multimodal Medical Diagnosis
标题：MedMIX：多模式医疗诊断的模式内部专家融合
链接：https://arxiv.org/abs/2605.16639

【8】Symphony for Speech-to-Text: Supporting Real-Time Medical Voice Interfaces
标题：语音转文本交响曲：支持实时医疗语音接口
链接：https://arxiv.org/abs/2605.16545

【9】DeepArrhythmia: Segment-Contextualized ECG Arrhythmia Classification via Selective Evidence Acquisition
标题：DeepArrhythmia：通过选择性证据获取进行分段背景化心电图心律失常分类
链接：https://arxiv.org/abs/2605.16441

【10】ReTAMamba: Reliability-Aware Temporal Aggregation with Mamba for Irregular Clinical Time Series Prediction
标题：ReTAMamba：使用Mamba进行可靠性感知时间聚集，用于不规则临床时间序列预测
链接：https://arxiv.org/abs/2605.16380

【11】Deep Learning for MRI Slice Interpolation: The Critical Role of Problem Formulation
标题：MRI切片插值的深度学习：问题制定的关键作用
链接：https://arxiv.org/abs/2605.16476

蒸馏|知识提取(10篇)

【1】Distilling Tabular Foundation Models for Structured Health Data
标题：提取结构化健康数据的表格基础模型
链接：https://arxiv.org/abs/2605.18702

【2】Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees
标题：袖珍基金会模型：将TFM提炼成可供处理器支持的树
链接：https://arxiv.org/abs/2605.18654

【3】Post-Trained MoE Can Skip Half Experts via Self-Distillation
标题：经过训练的MoE可以通过自我蒸馏跳过一半专家
链接：https://arxiv.org/abs/2605.18643

【4】FedSDR: Federated Self-Distillation with Rectification
标题：FedSDR：联合自蒸馏和整流
链接：https://arxiv.org/abs/2605.18028

【5】SAS: Semantic-aware Sampling for Generative Dataset Distillation
标题：SAS：生成式数据集蒸馏的语义感知采样
链接：https://arxiv.org/abs/2605.18012

【6】HINT-SD: Targeted Hindsight Self-Distillation for Long-Horizon Agents
标题：HINT-SD：针对长期代理的有针对性的事后诸葛亮自我蒸馏
链接：https://arxiv.org/abs/2605.17873

【7】$\boldsymbol{f}$-OPD: Stabilizing Long-Horizon On-Policy Distillation with Freshness-Aware Control
标题：$oldSymbol{f}$-OPD：通过新鲜度感知控制稳定长期政策蒸馏
链接：https://arxiv.org/abs/2605.17862

【8】Balancing Knowledge Distillation for Imbalance Learning with Bilevel Optimization
标题：用二层优化平衡知识提炼以解决不平衡学习
链接：https://arxiv.org/abs/2605.17839

【9】Agentic Cost-Aware Query Planning with Knowledge Distillation for Big Data Analytics
标题：面向大数据分析的基于知识蒸馏的成本感知查询规划
链接：https://arxiv.org/abs/2605.17831

【10】Self-Distillation is Optimal Among Spectral Shrinkage Estimators in Spiked Covariance Models
标题：尖峰协方差模型中谱收缩估计器中自蒸馏是最优的
链接：https://arxiv.org/abs/2605.17778

推荐(2篇)

【1】Learning Variable-Length Tokenization for Generative Recommendation
标题：学习生成式推荐的变长令牌化
链接：https://arxiv.org/abs/2605.17779

【2】A Production-Ready RL Framework for Personalized Utility Tuning with Pareto Sweeping in Pinterest Recommender Systems
标题：一个生产就绪RL框架，用于在Pinterest推荐系统中使用帕累托扫描进行个性化实用程序调优
链接：https://arxiv.org/abs/2605.16344

聚类(1篇)

【1】When Fireflies Cluster; Enhancing Automatic Clustering via Centroid-Guided Firefly Optimization
标题：当萤火虫聚集时;通过重心引导的萤火虫优化增强自动聚集
链接：https://arxiv.org/abs/2605.18460

超分辨率|去噪|去模糊|去雾(4篇)

【1】Fine-tuning Pocket-Aware Diffusion Models via Denoising Policy Optimization
标题：通过去噪政策优化微调口袋感知扩散模型
链接：https://arxiv.org/abs/2605.17693

【2】Forward-Learned Discrete Diffusion: Learning how to noise to denoise faster
标题：向前学习的离散扩散：学习如何更快地降噪
链接：https://arxiv.org/abs/2605.18204

【3】Wasserstein bounds for denoising diffusion probabilistic models via the Föllmer process
标题：基于Föllmer过程的去噪扩散概率模型的Wasserstein界
链接：https://arxiv.org/abs/2605.18069

【4】A note on connections between the Föllmer process and the denoising diffusion probabilistic model
标题：关于Föllmer过程和去噪扩散概率模型之间联系的注释
链接：https://arxiv.org/abs/2605.18040

自动驾驶|车辆|车道检测等(2篇)

【1】UniAlign: A Model-Agnostic Framework for Robust Network Traffic Classification under Distribution Shifts
标题：UniAlign：分布变化下稳健网络流量分类的模型不可知框架
链接：https://arxiv.org/abs/2605.17575

【2】CLAP: Contrastive Latent-space Prompt Optimization for End-to-end Autonomous Driving
标题：CLAP：端到端自动驾驶的对比潜在空间即时优化
链接：https://arxiv.org/abs/2605.17284

点云|SLAM|雷达|激光|深度RGBD相关(1篇)

【1】A Systematic Survey on Deep Learning Architectures for Point Cloud Classification and Segmentation
标题：点云分类和分割深度学习架构的系统调查
链接：https://arxiv.org/abs/2605.17131

联邦学习|隐私保护|加密(5篇)

【1】Federated Learning by Utility-Constrained Stochastic Aggregation for Improving Rational Participation
标题：基于效用约束随机聚集的联邦学习
链接：https://arxiv.org/abs/2605.18020

【2】BESplit: Bias-Compensated Split Federated Learning with Evidential Aggregation
标题：BESplit：具有证据聚集的偏差补偿分裂联邦学习
链接：https://arxiv.org/abs/2605.17508

【3】Byzantine-Resilient Federated Learning via QUBO-Based Client Selection on Quantum Annealers
标题：通过基于QUBO的客户端选择在量子Annealers上进行拜占庭弹性联邦学习
链接：https://arxiv.org/abs/2605.16438

【4】M$^2$FedAQI: Multimodal Federated Learning for Air Quality Prediction on Heterogeneous Edge Devices
标题：M $' 2$FedAQI：用于异类边缘设备上空气质量预测的多模式联邦学习
链接：https://arxiv.org/abs/2605.16375

【5】Statistical Limits and Efficient Algorithms for Differentially Private Federated Learning
标题：差异私有联邦学习的统计限制和有效算法
链接：https://arxiv.org/abs/2605.18656

推理|分析|理解|解释(28篇)

【1】Learning Quantifiable Visual Explanations Without Ground-Truth
标题：在没有基本真相的情况下学习可量化的视觉解释
链接：https://arxiv.org/abs/2605.18681

【2】XCTFormer: Leveraging Cross-Channel and Cross-Time Dependencies for Enhanced Time-Series Analysis
标题：XCTFormer：利用跨渠道和跨时间的重复性增强时间序列分析
链接：https://arxiv.org/abs/2605.18534

【3】What is Holding Back Latent Visual Reasoning?
标题：什么是阻碍潜在视觉推理？
链接：https://arxiv.org/abs/2605.18445

【4】Lightweight Gaussian Process Inference in C++ on Metal and CUDA
标题：C++中Metal和CUDA上的高斯轻量级过程推理
链接：https://arxiv.org/abs/2605.17898

【5】CoX-MoE: Coalesced Expert Execution for High-Throughput MoE Inference with AMX-Enabled CPU-GPU Co-Execution
标题 ：CoX-MoE：合并专家执行，用于高吞吐量MoE推理，并采用AMX支持的CPU-图形处理器联合执行
链接：https://arxiv.org/abs/2605.17889

【6】SNLP: Layer-Parallel Inference via Structured Newton Corrections
标题：SNLP：通过结构化牛顿修正的层并行推理
链接：https://arxiv.org/abs/2605.17842

【7】Counterfactual Explanations Under Concept Drift
标题：概念漂移下的反事实解释
链接：https://arxiv.org/abs/2605.17651

【8】TriAxialKV: Toward Extreme Low-Precision KV-Cache Quantization for Agentic Inference Tasks
标题：TriAxialKN：迈向用于统计推理任务的极低精度KV-缓存量化
链接：https://arxiv.org/abs/2605.17170

【9】Principal Component Analysis for Lunar Crater Detection
标题：月球陨石坑探测的主成分分析
链接：https://arxiv.org/abs/2605.17125

【10】Visual Timelines of Police Encounters in Body-Worn Camera Footage: Operational Context and Activity Cataloging for Training and Analysis in OpenBWC
标题：随身摄像机镜头中警察遭遇的视觉时间线：OpenBWC中训练和分析的操作背景和活动编目
链接：https://arxiv.org/abs/2605.17095

【11】Mechanism Learning: Prototype-Anchored Mechanism Inference for Scientific Forecasting
标题：机制学习：科学预测的原型锚定机制推理
链接：https://arxiv.org/abs/2605.17091

【12】Why Do Reasoning Models Lose Coverage? The Role of Data and Forks in the Road
标题：为什么推理模型失去覆盖范围？数据和叉子在道路上的作用
链接：https://arxiv.org/abs/2605.17026

【13】When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited
标题：当动态变化时，稳健任务推理获胜：重新审视行为基础模型的离线模仿学习
链接：https://arxiv.org/abs/2605.17017

【14】Cross-Domain Molecular Relational Learning: Leveraging Chemical Structure-Activity Analysis
标题：跨领域分子关系学习：利用化学结构-活性分析
链接：https://arxiv.org/abs/2605.16799

【15】Multi-Object Tracking Consistently Improves Wildlife Inference
标题：多目标跟踪持续改善野生动物推理
链接：https://arxiv.org/abs/2605.16672

【16】Public-Decay Homomorphic State Space Models for Private Sequence Inference
标题：私有序列推理的公开衰变同胚状态空间模型
链接：https://arxiv.org/abs/2605.16647

【17】Reducing Credit Assignment Variance via Counterfactual Reasoning Paths
标题：通过反事实推理路径减少信用分配差异
链接：https://arxiv.org/abs/2605.16302

【18】Systematic Optimization of Real-Time Diffusion Model Inference on Apple M3 Ultra
标题：苹果M3 Ultra实时扩散模型推理的系统优化
链接：https://arxiv.org/abs/2605.16259

【19】Can machine learning for quantum-gas experiments be explainable?
标题：量子气实验的机器学习可以解释吗？
链接：https://arxiv.org/abs/2605.18689

【20】PACE: Geometry-Aware Bridge Transport for Single-Cell Trajectory Inference
标题 ：PACE：用于单细胞轨迹推断的几何感知桥传输
链接：https://arxiv.org/abs/2605.18587

【21】Generalized Functional ANOVA in Closed-Form: A Unified View of Additive Explanations
标题：封闭形式的广义函数方差分析：加性解释的统一观点
链接：https://arxiv.org/abs/2605.18422

【22】Subject-Specific Analysis of Self-Initiated Attention Shifts from EEG with Controlled Internal and External Attention Conditions
标题：控制内部和外部注意力条件下自我启动注意力与脑电的转移的受试者特定分析
链接：https://arxiv.org/abs/2605.18251

【23】Buffer-Parameterized Machine Learning Surrogate Models for Cross-Technology Signal Integrity Analysis and Optimization
标题：用于跨技术信号完整性分析和优化的缓冲区参数化机器学习代理模型
链接：https://arxiv.org/abs/2605.18170

【24】Long-horizon prediction of three-dimensional wall-bounded turbulence with CTA-Swin-UNet and resolvent analysis
标题：利用CTA-Swin-UNet和解决方案分析对三维壁边界湍流的长视界预测
链接：https://arxiv.org/abs/2605.17888

【25】Simple Approximation and Derivative Free Inference-Time Scaling for Diffusion Models via Sequential Monte Carlo on Path Measures
标题：扩散模型的简单逼近和无导推理-时间标度，基于路径测量的序列蒙特卡罗
链接：https://arxiv.org/abs/2605.17850

【26】Topological Data Analysis combined with Machine Learning for Predicting Permeability of Porous Media
标题：结合机器学习预测多孔介质渗透率的Topics数据分析
链接：https://arxiv.org/abs/2605.17581

【27】FEG-Pro: Forecast-Error Growth Profiling for Finite-Horizon Instability Analysis of Nonlinear Time Series
标题：FEG-Pro：非线性时间序列超时空不稳定性分析的预测误差增长剖析
链接：https://arxiv.org/abs/2605.17282

【28】Dimension-Uniform Discretization Analysis of Preconditioned Annealed Langevin Dynamics for Multimodal Gaussian Mixtures
标题：多峰高斯混合物预处理后的Anglevin动力学的均匀离散化分析
链接：https://arxiv.org/abs/2605.16473

检测相关(12篇)

【1】UTOPYA: A Multimodal Deep Learning Framework for Physics-Informed Anomaly Detection and Time-Series Prediction
标题：UTOPYA：用于物理信息异常检测和时间序列预测的多模式深度学习框架
链接：https://arxiv.org/abs/2605.18188

【2】Guard: Scalable Straggler Detection and Node Health Management for Large-Scale Training
标题：Guard：用于大规模训练的可扩展掉队者检测和节点健康管理
链接：https://arxiv.org/abs/2605.17879

【3】Is Complex Training Necessary for Long-Tailed OOD Detection? A Re-think from Feature Geometry
标题：长尾OOD检测需要复杂的训练吗？特征几何的重新思考
链接：https://arxiv.org/abs/2605.17799

【4】MV-Gate: Insider Threat Detection via Multi-View Behavioral Statistics and Semantic Modeling
标题：MV-Gate：通过多视图行为统计和语义建模进行内部威胁检测
链接：https://arxiv.org/abs/2605.17761

【5】ADR: An Agentic Detection System for Enterprise Agentic AI Security
标题：ADR：一个针对企业大型人工智能安全的大型检测系统
链接：https://arxiv.org/abs/2605.17380

【6】An Efficient Machine Learning-based Framework for Detection and Prevention of Frauds in Telecom Networks
标题：基于机器学习的高效电信网络欺诈检测和预防框架
链接：https://arxiv.org/abs/2605.17245

【7】Integration of AI in Cybersecurity: Current Trends with a Focused Look at Intrusion Detection Applications
标题：人工智能在网络安全中的集成：当前趋势，重点关注入侵检测应用
链接：https://arxiv.org/abs/2605.17219

【8】Filter-then-Verify: A Multiphase GNN and ModernBERT Framework for Social Engineering Detection in Email Networks
标题：过滤然后验证：用于电子邮件网络中社会工程检测的多阶段GNN和ModernBERT框架
链接：https://arxiv.org/abs/2605.17201

【9】On-Device Interpretable Tsetlin Machine-Based Intrusion Detection for Secure IoMT
标题：基于设备可解释Tsetlin机器的安全IoMT入侵检测
链接：https://arxiv.org/abs/2605.16707

【10】Preference Instability in Reward Models: Detection and Mitigation via Sparse Autoencoders
标题：奖励模型中的偏好不稳定性：通过稀疏自编码器检测和缓解
链接：https://arxiv.org/abs/2605.16339

【11】Causal Anomaly Detection for Lithium-Ion Battery Degradation
标题：锂离子电池劣化原因异常检测
链接：https://arxiv.org/abs/2605.17334

【12】Toward Near-Real-Time Marine Oil Spill Detection in SAR Imagery using Quantum-Assisted SVM
标题：基于量子支持向量机的SAR图像近实时海上溢油检测
链接：https://arxiv.org/abs/2605.17217

分类|识别(5篇)

【1】Efficient and Noise-Tolerant PAC Learning of Multiclass Linear Classifiers
标题：多类线性分类器的高效且耐噪PAC学习
链接：https://arxiv.org/abs/2605.18662

【2】Protein Fold Classification at Scale: Benchmarking and Pretraining
标题：大规模蛋白质折叠分类：基准和预训练
链接：https://arxiv.org/abs/2605.18552

【3】Modality vs. Morphology: A Framework for Time Series Classification for Biological Signals
标题：模态与形态学：生物信号时间序列分类的框架
链接：https://arxiv.org/abs/2605.18483

【4】Content-Style Identification via Differential Independence
标题：通过差异独立性的内容风格识别
链接：https://arxiv.org/abs/2605.17827

【5】The Neural Tangent Kernel for Classification
标题：用于分类的神经切核
链接：https://arxiv.org/abs/2605.17606

表征(14篇)

【1】Learning Normal Representations for Blood Biomarkers
标题：学习血液生物标志物的正常表示
链接：https://arxiv.org/abs/2605.18701

【2】Probing for Representation Manifolds in Superposition
标题：叠加中的表现形式的探索
链接：https://arxiv.org/abs/2605.18537

【3】Improved Baselines with Representation Autoencoders
标题：使用表示自动编码器改进的基线
链接：https://arxiv.org/abs/2605.18324

【4】DARE-EEG: A Foundation Model for Mining Dual-Aligned Representation of EEG
标题：DARE-EEG：一种挖掘脑电双对齐表示的基础模型
链接：https://arxiv.org/abs/2605.18298

【5】AURORA: Contextual Orthogonalization for Geometric Representation Learning in Healthcare Foundation Models
标题：AURORA：医疗保健基金会模型中几何表示学习的上下文同步化
链接：https://arxiv.org/abs/2605.17765

【6】Mind the Gap: Learning Modality-Agnostic Representations with a Cross-Modality UNet
标题：Mind the Gap：Learning Modality-Agnostic Representations with a Cross-Modality UNet（英语：Mind the Gap：Learning Modality-Agnostic Representation with a Cross-Modality UNet）
链接：https://arxiv.org/abs/2605.16887

【7】Learning Relative Representations for Fine-Grained Multimodal Alignment with Limited Data
标题：在有限数据下学习细粒度多模式对齐的相对表示
链接：https://arxiv.org/abs/2605.16834

【8】Atoms as Language: VQ-Atom: Semantic Discretization for Molecular Representation Learning
标题：原子作为语言：VQ-Atom：分子表示学习的语义离散化
链接：https://arxiv.org/abs/2605.16823

【9】Jacobian-Guided Anisotropic Noise Reshaping for Enhancing Representation Utility under Local Differential Privacy
标题：局部差异隐私下雅可比引导各向异性噪音重塑以增强表示效用
链接：https://arxiv.org/abs/2605.16812

【10】Structure-Aware Masking for Protein Representation Learning
标题：蛋白质表示学习的结构感知掩蔽
链接：https://arxiv.org/abs/2605.16581

【11】SwordBench: Evaluating Orthogonality of Steering Image Representations
标题：SworthBench：评估转向图像表示的共变性
链接：https://arxiv.org/abs/2605.16372

【12】PIMSM: Physics-Informed Multi-Scale Mamba for Stable Neural Representations under Distribution Shift
标题：PIMSM：基于物理知识的多尺度Mamba，用于分布变化下的稳定神经表示
链接：https://arxiv.org/abs/2605.16351

【13】Patchwork: A compact representation for 3D polygonal shapes
标题：补丁：3D多边形形状的紧凑表示
链接：https://arxiv.org/abs/2605.16266

【14】A Unified Framework for Structured Flow Modeling: From Continuous Fields to Data-Driven Representations
标题：结构化流建模的统一框架：从连续字段到数据驱动表示
链接：https://arxiv.org/abs/2605.18250

编码器(5篇)

【1】Are Sparse Autoencoder Benchmarks Reliable?
标题：稀疏自动编码器基准可靠吗？
链接：https://arxiv.org/abs/2605.18229

【2】t-gems: text-guided exit modules for decreasing clip image encoder
标题：t-gem：文本引导退出模块，用于减少剪辑图像编码器
链接：https://arxiv.org/abs/2605.17499

【3】Mechanistically Interpretable Neural Encoding Reveals Fine-Grained Functional Selectivity in Human Visual Cortex
标题：机械可解释神经编码揭示人类视觉皮质中细粒度的功能选择性
链接：https://arxiv.org/abs/2605.16468

【4】Sparse Mamba Decoder for Quantum Error Correction: Efficient Defect-Centric Processing of Surface Code Syndromes
标题：用于量子错误纠正的稀疏Mamba解码器：表面代码Syndrome的高效以缺陷为中心处理
链接：https://arxiv.org/abs/2605.17156

【5】Kelvin v1.0: A Neural Pre-Encoder for H.264: A standards-compliant learned preprocessor with -27.62% BD-VMAF on UVG
标题：Kelvin v1.0：适用于H.264的神经预编码器：符合标准的学习型预处理器，UVG上具有-27.62%BD-VAPM
链接：https://arxiv.org/abs/2605.16376

优化|敛散性(15篇)

【1】COOPO: Cyclic Offline-Online Policy Optimization Algorithm
标题：COOPO：循环离线-在线策略优化算法
链接：https://arxiv.org/abs/2605.18675

【2】Spherical Harmonic Optimal Transport: Application to Climate Models Comparisons
标题：球调和最优输运：在气候模型比较中的应用
链接：https://arxiv.org/abs/2605.18389

【3】Proximal basin hopping: global optimization with guarantees
标题：近端盆地跳跃：有保证的全球优化
链接：https://arxiv.org/abs/2605.18364

【4】Dimension-Free Convergence of Discrete Diffusion Models: Adjoint Equations Induce the Right Space
标题：离散扩散模型的无边界收敛：伴随方程引入正确的空间
链接：https://arxiv.org/abs/2605.17232

【5】Evolutionary Extreme Learning Machine of ab-initio Energy Landscapes for Crystal Structure Prediction using Manta Ray Optimization with Levy Flight
标题：从头算能量景观的进化极端学习机，用于使用Manta Ray优化和Levy Flight进行晶体结构预测
链接：https://arxiv.org/abs/2605.17148

【6】Differentiable Optimization Layers for Guaranteed Fairness in Deep Learning
标题：可区分的优化层以保证深度学习的公平性
链接：https://arxiv.org/abs/2605.17118

【7】Decision-Aware Proximal Bridge Learning for Optimal Treatment Selection
标题：决策感知的近端桥学习以实现最佳治疗选择
链接：https://arxiv.org/abs/2605.16989

【8】AgentKernelArena: Generalization-Aware Benchmarking of GPU Kernel Optimization Agents
标题：AgentKernelArena：图形处理器核心优化代理的通用感知基准测试
链接：https://arxiv.org/abs/2605.16819

【9】Global Convergence of Sampling-Based Nonconvex Optimization through Diffusion-Style Smoothing
标题：基于抽样的非凸优化通过扩散式平滑的全局收敛
链接：https://arxiv.org/abs/2605.16520

【10】Orth-Dion: Eliminating Geometric Mismatch in Distributed Low-Rank Spectral Optimization
标题：Orth-Dion：消除分布式低阶谱优化中的几何不匹配
链接：https://arxiv.org/abs/2605.16341

【11】SignMuon: Communication-Efficient Distributed Muon Optimization
标题：SignMuon：通信高效的分布式Muon优化
链接：https://arxiv.org/abs/2605.16311

【12】Scale-Invariant Neural Network Optimization: Norm Geometry and Heavy-Tailed Noise
标题：尺度不变神经网络优化：范数几何和重尾噪声
链接：https://arxiv.org/abs/2605.18528

【13】Geometric Dictionary Learning of Dynamical Systems with Optimal Transport
标题：具有最优传输的动态系统的几何字典学习
链接：https://arxiv.org/abs/2605.18276

【14】StatQAT: Statistical Quantizer Optimization for Deep Networks
标题：StatQAT：深度网络的统计量化器优化
链接：https://arxiv.org/abs/2605.17745

【15】Scalable Bi-causal Optimal Transport via KL Relaxation and Policy Gradients
标题：基于KL松弛和策略约束的可扩展双因果最优运输
链接：https://arxiv.org/abs/2605.17271

预测|估计(34篇)

【1】Data Presentation Over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models
标题：架构上的数据呈现：使用表格基础模型进行信用风险预测的恢复策略
链接：https://arxiv.org/abs/2605.18635

【2】TabH2O: A Unified Foundation Model for Tabular Prediction
标题：TabH2O：表格预测的统一基础模型
链接：https://arxiv.org/abs/2605.18383

【3】Dynamic robotic cloth folding with efficient Koopman operator-based model predictive control
标题：动态机器人布料折叠，具有高效的基于Koopman操作员的模型预测控制
链接：https://arxiv.org/abs/2605.18373

【4】Decoupled Conformal Optimisation: Efficient Prediction Sets via Independent Tuning and Calibration
标题：脱钩共形优化：通过独立调整和校准实现高效预测集
链接：https://arxiv.org/abs/2605.18354

【5】Foundation Models for Credit Risk Prediction: A Game Changer?
标题：信用风险预测的基础模型：游戏规则改变者？
链接：https://arxiv.org/abs/2605.18147

【6】DAD4TS: Data-Augmentation-Oriented Diffusion Model for Time-Series Forecasting with Small-Scale Data
标题：DAD 4TS：面向数据增强的扩散模型，用于小规模数据时间序列预测
链接：https://arxiv.org/abs/2605.17866

【7】L-Drive: Beyond a Single Mapping-Latent Context Drives Time Series Forecasting
标题：L-Drive：超越单一映射潜在上下文驱动时间序列预测
链接：https://arxiv.org/abs/2605.17730

【8】PEIRA: Learning Predictive Encoders through Inter-View Regressor Alignment
标题：PEIRA：通过视图间回归量对齐学习预测编码器
链接：https://arxiv.org/abs/2605.17671

【9】When a Zero-Shooter Cheats: Improving Age Estimation via Activation Steering
标题：当零射手作弊时：通过激活引导提高年龄估计
链接：https://arxiv.org/abs/2605.17658

【10】A Feature-Driven Framework for Software Fault Prediction
标题：一种软件故障预测框架
链接：https://arxiv.org/abs/2605.17611

【11】Scale-Equivariant Generative Forecasting: Weight-Tied Dilated Convolutions, Wavelet Scattering Inputs, and Spectral-Consistency Training for Self-Similar Time Series
标题：尺度等变生成预测：自相似时间序列的加权捆绑扩张卷积、子波散布输入和谱一致性训练
链接：https://arxiv.org/abs/2605.17582

【12】Human-Flow Digital Twin for Predicting the Effects of Mobility Introduction on Visitor Circulation
标题：用于预测流动性引入对游客循环的影响的人流数字双胞胎
链接：https://arxiv.org/abs/2605.17426

【13】Position: Age Estimation Models Do Not Process Biometric Data
标题：位置：年龄估计模型不处理生物识别数据
链接：https://arxiv.org/abs/2605.17347

【14】Active Budget Allocation for Efficient Scaling Law Estimation via Surrogate-Guided Pruning
标题：通过代理引导修剪进行有效的标度定律估计的主动预算分配
链接：https://arxiv.org/abs/2605.17234

【15】Weighted Flow Matching and Physics-Informed Nonlinear Filtering for Parameter Estimation in Digital Twins
标题：加权流匹配和物理信息非线性过滤用于数字双胞胎中的参数估计
链接：https://arxiv.org/abs/2605.17146

【16】Empirical evaluation of Time Series Foundation Models for Day-ahead and Imbalance Electricity Price Forecasting in Belgium
标题：比利时前一天和不平衡电价预测时间序列基础模型的实证评估
链接：https://arxiv.org/abs/2605.17045

【17】Cross-modal Affinity-aligned Multimodal Learning Analytics for Predicting Student Collaboration Satisfaction in Game-Based Learning
标题：跨模式亲和力对齐的多模式学习分析，用于预测基于游戏的学习中的学生合作满意度
链接：https://arxiv.org/abs/2605.16806

【18】FIM-LoRA: Task-Informative Rank Allocation for LoRA via Calibration-Time Gradient-Variance Estimation
标题：FIM-LoRA：通过校准时间一致性方差估计的LoRA任务信息等级分配
链接：https://arxiv.org/abs/2605.16800

【19】PULSE: Generative Phase Evolution for Non-Stationary Time Series Forecasting
标题：PulSE：非平稳时间序列预测的生成阶段进化
链接：https://arxiv.org/abs/2605.16793

【20】In-context learning enables continental-scale subsurface temperature prediction from sparse local observations
标题：上下文学习能够根据稀疏的本地观测结果预测大陆规模的地下温度
链接：https://arxiv.org/abs/2605.16665

【21】QuChaTeR: A Hybrid Quantum-Chaotic Temporal Framework for Earthquake Prediction
标题：QuChaTeR：地震预测的混合量子-混乱时间框架
链接：https://arxiv.org/abs/2605.16454

【22】PESD-TSF: A Period-Aware and Explicit Structured Decomposition Framework for Long-Term Time Series Forecasting
标题：PESD-TSF：用于长期时间序列预测的周期感知和显式结构化分解框架
链接：https://arxiv.org/abs/2605.16449

【23】Nested Spatio-Temporal Time Series Forecasting
标题：嵌套时空时间序列预测
链接：https://arxiv.org/abs/2605.16447

【24】Hierarchical Two-Stage Framework for Environment-Aware Long-Horizon Vessel Trajectory Prediction
标题：环境感知长地平线船舶轨迹预测的分层两阶段框架
链接：https://arxiv.org/abs/2605.16442

【25】GPU-Accelerated Deep Learning for Heatwave Prediction and Urban Heat Risk Assessment
标题 ：用于热浪预测和城市高温风险评估的GOP加速深度学习
链接：https://arxiv.org/abs/2605.16435

【26】TailedTS: Benchmark Dataset for Heavy-Tailed Time Series Prediction and Periodicity Quantification
标题：TailedTS：重尾时间序列预测和周期性量化的基准数据集
链接：https://arxiv.org/abs/2605.16361

【27】LEAF: A Living Benchmark for Event-Augmented Forecasting
标题：LEAF：事件增强预测的动态基准
链接：https://arxiv.org/abs/2605.16358

【28】MCQ Difficulty Prediction via Modeling Learner Heterogeneity Using Data-Driven Cognitive Profiling
标题：使用数据驱动认知剖析通过对学习者异源建模来预测MCQ难度
链接：https://arxiv.org/abs/2605.16290

【29】Learned Memory Attenuation in Sage-Husa Kalman Filters for Robust UAV State Estimation
标题：用于鲁棒无人机状态估计的Sage-Husa卡尔曼过滤器中的学习记忆衰减
链接：https://arxiv.org/abs/2605.18704

【30】QLIF-CAST: Quantum Leaky-Integrate-and-Fire for Time-Series Weather Forecasting
标题：QLIF-cast：用于时间序列天气预报的量子泄漏集成和火灾
链接：https://arxiv.org/abs/2605.18333

【31】Online Conformal Prediction for Non-Exchangeable Panel Data
标题：不可交换面板数据的在线共形预测
链接：https://arxiv.org/abs/2605.17705

【32】A Hybrid Gaussian Process Regression Framework for Stable Volatility-Covariance Estimation: Evidence from Global Equity Indices
标题：稳定波动率协方差估计的混合高斯过程回归框架：来自全球股票指数的证据
链接：https://arxiv.org/abs/2605.17275

【33】Prediction-Intervention Games and Invariant Sets
标题：预测干预游戏和不变集
链接：https://arxiv.org/abs/2605.16828

【34】A Machine Learning Framework for EEG-Based Prediction of Treatment Efficacy in Chronic Neck Pain
标题：基于脑电预测慢性颈部疼痛治疗效果的机器学习框架
链接：https://arxiv.org/abs/2605.16326

其他神经网络|深度学习|模型|建模(61篇)

【1】Ensembling Tabular Foundation Models - A Diversity Ceiling And A Calibration Trap
标题：整合表格基础模型-多样性上限和校准陷阱
链接：https://arxiv.org/abs/2605.18696

【2】Better Together: Evaluating the Complementarity of Earth Embedding Models
标题：更好地在一起：评估地球嵌入模型的互补性
链接：https://arxiv.org/abs/2605.18667

【3】GIM: Evaluating models via tasks that integrate multiple cognitive domains
标题：GIM：通过整合多个认知领域的任务评估模型
链接：https://arxiv.org/abs/2605.18663

【4】Learning to Look Benign: Targeted Evasion of Malware Detectors via API Import Injection
标题：学会看起来友善：通过API导入注入有针对性地规避恶意软件检测器
链接：https://arxiv.org/abs/2605.18624

【5】Stochastic Penalty-Barrier Methods for Constrained Machine Learning
标题：约束机器学习的随机罚障方法
链接：https://arxiv.org/abs/2605.18618

【6】Pointwise Generalization in Deep Neural Networks
标题：深度神经网络中的逐点推广
链接：https://arxiv.org/abs/2605.18598

【7】DiPRL: Learning Discrete Programmatic Policies via Architecture Entropy Regularization
标题：DiSPL：通过架构熵正规化学习离散程序政策
链接：https://arxiv.org/abs/2605.18508

【8】GAMMA: Global Bit Allocation for Mixed-Precision Models under Arbitrary Budgets
标题：GABA：任意预算下混合精度模型的全球比特分配
链接：https://arxiv.org/abs/2605.18475

【9】Beyond Square Roots: Explicit Memory-Efficient Factorization for Multi-Epoch Private Learning
标题：超越平方根：多时代私人学习的显式记忆高效分解
链接：https://arxiv.org/abs/2605.18379

【10】The Symmetries of Three-Layer ReLU Networks
标题：三层ReLU网络的对称性
链接：https://arxiv.org/abs/2605.18319

【11】PH-Dreamer: A Physics-Driven World Model via Port-Hamiltonian Generative Dynamics
标题：PH-Dreamer：通过波特-汉密尔顿生成动力学的物理驱动世界模型
链接：https://arxiv.org/abs/2605.18303

【12】Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network
标题：双速率扩散：利用交错的高光网络加速扩散模型
链接：https://arxiv.org/abs/2605.18190

【13】pyforce-1.0.0: Python Framework for data-driven model Order Reduction of multi-physiCs problEms
标题：pyforce-1.0.0：用于数据驱动模型的Python框架多物理问题的降阶
链接：https://arxiv.org/abs/2605.18082

【14】Scalable Decision-Focused Learning through Cost-Sensitive Regression
标题：通过成本敏感回归进行可扩展的以决策为中心的学习
链接：https://arxiv.org/abs/2605.18005

【15】SAFE-SVD: Sensitivity-Aware Fidelity-Enforcing SVD for Physics Foundation Models
标题：SAFE-SVD：用于物理基础模型的灵敏度感知保真度增强SVD
链接：https://arxiv.org/abs/2605.17985

【16】Training data attribution in diffusion models via mirrored unlearning and noise-consistent skew
标题：通过镜像取消学习和噪音一致性倾斜训练扩散模型中的数据属性
链接：https://arxiv.org/abs/2605.17938

【17】Learning over Positive and Negative Edges with Contrastive Message Passing
标题：通过对比信息传递来学习积极和消极的边缘
链接：https://arxiv.org/abs/2605.17854

【18】GenTS: A Comprehensive Benchmark Library for Generative Time Series Models
标题：GenTS：生成式时间序列模型的综合基准库
链接：https://arxiv.org/abs/2605.17804

【19】Toy Combinatorial Interpretability Models Reveal Lottery Tickets in Early Feature Space
标题：玩具组合解释模型揭示早期特征空间中的彩票
链接：https://arxiv.org/abs/2605.17704

【20】Exact Convex Reformulations of Linear Neural Networks via Completely Positive Lifting
标题：基于完全正提升的线性神经网络精确凸重构
链接：https://arxiv.org/abs/2605.17692

【21】Venom: A PyTorch Generative Modeling Toolkit
标题：Venom：PyTorch生成建模工具包
链接：https://arxiv.org/abs/2605.17605

【22】AutoRubric-T2I: Robust Rule-Based Reward Model for Text-to-Image Alignment
标题：AutoRubric-T2 I：用于文本与图像对齐的稳健基于规则的奖励模型
链接：https://arxiv.org/abs/2605.17602

【23】Stable Routing for Mixture-of-Experts in Class-Incremental Learning
标题：类增量学习中混合专家的稳定路由
链接：https://arxiv.org/abs/2605.17571

【24】Structured Neural Marked Point Processes for Interpretable Event Interaction Modeling
标题：可解释事件交互建模的结构化神经标记点过程
链接：https://arxiv.org/abs/2605.17568

【25】Beyond Accuracy: Robustness, Interpretability and Expressiveness of EEG Foundation Models
标题：超越准确性：脑电基础模型的稳健性、可解释性和表现性
链接：https://arxiv.org/abs/2605.17562

【26】Beyond Linear Superposition: Discovering Climate Features in AI Weather Models with KAN-SAE
标题：超越线性叠加：利用KAN-SAGE发现人工智能天气模型中的气候特征
链接：https://arxiv.org/abs/2605.17493

【27】Radial-Angular Geometry for Reliable Update Diagnosis in Noisy-Label Learning
标题：噪声标记学习中用于可靠更新诊断的径向-角几何
链接：https://arxiv.org/abs/2605.17429

【28】Olivia: Harmonizing Time Series Foundation Models with Power Spectral Density
标题：奥利维亚：协调时间序列基础模型与功率谱密度
链接：https://arxiv.org/abs/2605.17340

【29】Bridging the Gap between Sparse Matrix Reordering and Factorization: A Deep Learning Framework for Fill-in Reduction
标题：弥合稀疏矩阵重排序和因式分解之间的差距：填充减少的深度学习框架
链接：https://arxiv.org/abs/2605.17339

【30】Learning Higher-Order Structure from Incomplete Spatiotemporal Data: Multi-Scale Hypergraph Laplacians with Neural Refinement
标题：从不完整时空数据中学习更高级结构：具有神经细化的多尺度超图拉普拉斯
链接：https://arxiv.org/abs/2605.17316

【31】Iterative Chow Filtering for Learning with Distribution Shift
标题：迭代Chow过滤用于分布转移的学习
链接：https://arxiv.org/abs/2605.17251

【32】Learning in Position-Aware Multinomial Logit Bandits: From Multiplicative to General Position Effects
标题：位置感知的多项Logit Bandits中的学习：从相乘到一般位置效应
链接：https://arxiv.org/abs/2605.17238

【33】Stress-Testing Neural Network Verifiers with Provably Robust Instances
标题：具有可证明鲁棒性的压力测试神经网络验证器
链接：https://arxiv.org/abs/2605.17153

【34】Learning-Zone Energy: Online Data Selection for Efficient RL Post-Training
标题：学习区能源：在线数据选择以实现高效RL后训练
链接：https://arxiv.org/abs/2605.17003

【35】Emulating the Forced Response of Climate Models with Flow Matching
标题：利用流量匹配模拟气候模型的强迫响应
链接：https://arxiv.org/abs/2605.16929

【36】Learning Unbiased Permutations via Flow Matching
标题：通过流匹配学习无偏排列
链接 ：https://arxiv.org/abs/2605.16755

【37】Isolating Nonlinear Independent Sources in fMRI with $β$-TCVAE Models
标题：用$β$-TCVAE模型隔离fMRI中的非线性独立源
链接：https://arxiv.org/abs/2605.16708

【38】Your SaaS Is an Insurance Product: A Modeling Framework
标题：您的SaaS是保险产品：建模框架
链接：https://arxiv.org/abs/2605.16699

【39】Identify Then Project: Contrastive Learning of Latent Dynamics from Partial Observations with Port-Hamiltonian Structure
标题：识别然后项目：来自具有波特-汉密尔顿结构的部分观测的潜在动力学对比学习
链接：https://arxiv.org/abs/2605.16682

【40】Learning How to Cube
标题：学习如何立方体
链接：https://arxiv.org/abs/2605.16632

【41】MLReplicate: Benchmarking Autonomous Research Systems for Machine Learning Reproducibility
标题：MLCopy：为机器学习再现性进行基准测试
链接：https://arxiv.org/abs/2605.16616

【42】Learning What Evaluators Value: A Reliable Approach to Modeling Evaluator Preferences
标题：了解评估者的价值：评估者偏好建模的可靠方法
链接：https://arxiv.org/abs/2605.16615

【43】fPINN-DeepONet: A Physics-Informed Operator Learning Framework for Multi-term Time-fractional Mixed Diffusion-wave Equations
标题：fPINN-DeepONet：多项时间分数混合扩散波方程的物理信息操作员学习框架
链接：https://arxiv.org/abs/2605.16594

【44】World Model-Enabled Causal Digital Twins for Semantic Communications in Physical AI Systems
标题：物理人工智能系统中语义通信的世界模型启用因果数字孪生
链接：https://arxiv.org/abs/2605.16547

【45】Identifiable Token Correspondence for World Models
标题：世界模特的可识别代币通信
链接：https://arxiv.org/abs/2605.16457

【46】Two-Valued Symmetric Circulant Matrices: Applications in Deep Learning
标题：二值对称循环矩阵：深度学习中的应用
链接：https://arxiv.org/abs/2605.16443

【47】Edge-AI-Driven Learning-to-Rank for Decentralized Task Allocation in Circular Smart Manufacturing
标题：边缘人工智能驱动的循环智能制造中去中心化任务分配学习
链接：https://arxiv.org/abs/2605.16433

【48】Diffusion Models, Denoiser Architecture and Creativity
标题：扩散模型、降噪建筑和创造力
链接：https://arxiv.org/abs/2605.16415

【49】OrbiSim: World Models as Differentiable Physics Engines for Embodied Intelligence
标题：OrbiSim：世界模型作为均衡智能的差异物理引擎
链接：https://arxiv.org/abs/2605.16395

【50】Machine Learning-Based Pre-Test Risk Stratification for PCR-Confirmed Chlamydia Using Patient-Reported Data and Urine Biomarkers
标题：使用患者报告的数据和尿液生物标志物对PCR确诊的衣钵进行基于机器学习的检测前风险分层
链接：https://arxiv.org/abs/2605.16365

【51】Flow-Direct: Feedback-Efficient and Reusable Guidance for Flow Models via Non-Parametric Guidance Field
标题：Flow-Direct：通过非参数引导场为流量模型提供反馈高效且可重复使用的引导
链接：https://arxiv.org/abs/2605.16348

【52】Phase Transitions in Driven Informational Systems: A Two-Field Perspective on Learning Theory and Non-Equilibrium Chemistry
标题：驱动信息系统中的阶段转变：学习理论和非平衡化学的两场视角
链接：https://arxiv.org/abs/2605.16325

【53】Shallow ReLU$^s$ Networks in $L^p$-Type and Sobolev Spaces: Approximation and Path-Norm Controlled Generalization
标题：$L ' p$-型和Sobolev空间中的浅ReLU$' s$网络：逼近和路径范控制推广
链接：https://arxiv.org/abs/2605.18468

【54】Canonical Regularisation of Wide Feature-Learning Neural Networks
标题：广泛学习神经网络的典型正规化
链接：https://arxiv.org/abs/2605.18180

【55】Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent
标题：线性宽度两层网络中的特征学习：两步与一步梯度下降
链接：https://arxiv.org/abs/2605.17767

【56】How does feature learning reshape the function space?
标题：特征学习如何重塑功能空间？
链接：https://arxiv.org/abs/2605.17718

【57】On Gaussian approximation for entropy-regularized Q-learning with function approximation
标题：基于函数逼近的函数逼近的信息量正化Q学习的高斯逼近
链接：https://arxiv.org/abs/2605.17678

【58】High-dimensional Limit of SGD for Diagonal Linear Networks
标题：对角线性网络的BCD的高维极限
链接：https://arxiv.org/abs/2605.17177

【59】A Fourier perspective on the learning dynamics of neural networks: from sample complexities to mechanistic insights
标题：神经网络学习动力学的傅立叶观点：从样本复杂性到机械见解
链接：https://arxiv.org/abs/2605.16913

【60】Isotonic Survival Regression: Calibrated Survival Distributions from Deep Cox Models
标题：等张生存回归：来自Deep Cox模型的校准生存分布
链接：https://arxiv.org/abs/2605.16571

【61】Overcoming the Intrinsic Performance Limitations of MEMS IMU via Diffusion-Based Generative Learning
标题：通过基于扩散的生成式学习克服MEMS IMU的固有性能限制
链接：https://arxiv.org/abs/2605.16391

其他(119篇)

【1】A Readiness-Driven Runtime for Pipeline-Parallel Training under Runtime Variability
标题：任务可变条件下的ready驱动任务组
链接：https://arxiv.org/abs/2605.18750

【2】ESI-Bench: Towards Embodied Spatial Intelligence that Closes the Perception-Action Loop
标题：ESI-Bench：迈向封闭感知-行动循环的有序空间智能
链接：https://arxiv.org/abs/2605.18746

【3】PIXLRelight: Controllable Relighting via Intrinsic Conditioning
标题：PIXLRelight：通过固有调节进行可控重新点燃
链接：https://arxiv.org/abs/2605.18735

【4】EnvFactory: Scaling Tool-Use Agents via Executable Environments Synthesis and Robust RL
标题：EnvFactory：通过可执行环境合成和稳健RL扩展工具使用代理
链接：https://arxiv.org/abs/2605.18703

【5】Position: Weight Space Should Be a First-Class Generative AI Modality
标题：立场：重量空间应该成为一流的生成人工智能模式
链接：https://arxiv.org/abs/2605.18632

【6】Aligned Training: A Parameter-Free Method to Improve Feature Quality and Stability of Sparse Autoencoders (SAE)
标题：对齐训练：一种提高稀疏自动编码器（AE）特征质量和稳定性的无参数方法
链接：https://arxiv.org/abs/2605.18629

【7】CATA: Continual Machine Unlearning via Conflict-Averse Task Arithmetic
标题：CATA：通过预算厌恶任务算法的连续机器取消学习
链接：https://arxiv.org/abs/2605.18610

【8】Perfect Parallelization in Mini-Batch SGD with Classical Momentum Acceleration
标题：具有经典动量加速的小批量SGD中的完美并行化
链接：https://arxiv.org/abs/2605.18609

【9】Physics-Aligned Canonical Equivariant Fourier Neural Operator under Symmetry-Induced Shifts
标题：对称性诱导位移下的物理对齐典型等变傅里叶神经运算符
链接：https://arxiv.org/abs/2605.18606

【10】Randomized Advantage Transformation (RAT): Computing Natural Policy Gradients via Direct Backpropagation
标题：随机优势转换（RAT）：通过直接反向传播计算自然政策因素
链接：https://arxiv.org/abs/2605.18591

【11】When Outcome Looks Right But Discipline Fails: Trace-Based Evaluation Under Hidden Competitor State
标题：当结果看起来正确但纪律失败时：隐藏竞争者状态下的基于追溯的评估
链接：https://arxiv.org/abs/2605.18580

【12】scHelix: Asymmetric Dual-Stream Integration via Explicit Gene-Level Disentanglement
标题：scHSYS：通过显式基因级解纠缠的非对称双流集成
链接：https://arxiv.org/abs/2605.18576

【13】GUT-IS: A Data-Driven Approach to Integrating Constructs and Their Relations in Information Systems
标题：GUT-IS：一种在信息系统中集成结构及其关系的数据驱动方法
链接：https://arxiv.org/abs/2605.18567

【14】Federated Martingale Posterior Samping
标题：联邦马丁格后验抽样
链接：https://arxiv.org/abs/2605.18554

【15】Beyond Scaling: Agents Are Heading to the Edge
标题：超越扩展：代理商正在走向边缘
链接：https://arxiv.org/abs/2605.18535

【16】Continuous Diffusion Scales Competitively with Discrete Diffusion for Language
标题：语言的连续扩散规模与离散扩散竞争
链接：https://arxiv.org/abs/2605.18530

【17】Offline Contextual Bandits in the Presence of New Actions
标题：新动作下的离线背景盗贼
链接：https://arxiv.org/abs/2605.18509

【18】DBES: A Systematic Benchmark and Metric Suite for Evaluating Expert Specialization in Large-Scale MoEs
标题：DBES：用于评估大型教育部专家专业化的系统基准和指标套件
链接：https://arxiv.org/abs/2605.18498

【19】EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective
标题：EvoMemBench：从自我进化的角度对代理内存进行基准测试
链接：https://arxiv.org/abs/2605.18421

【20】Temporal Task Diversity: Inductive Biases Under Non-Stationarity in Synthetic Sequence Modelling
标题：时间任务多样性：合成序列建模中非平稳性下的归纳偏差
链接：https://arxiv.org/abs/2605.18281

【21】From Volume to Value: Preference-Aligned Memory Construction for On-Device RAG
标题：从数量到价值：设备上RAG的优先级一致存储器构建
链接：https://arxiv.org/abs/2605.18271

【22】A Simplex Witness Certificate for Constant Collapse in Variational Autoencoders
标题：变分自动编码器常数崩溃的单纯形见证证书
链接：https://arxiv.org/abs/2605.18224

【23】Ringmaster LMO: Asynchronous Linear Minimization Oracle Momentum Method
标题：Ringmaster LMO：异步线性最小化Oracle动量方法
链接：https://arxiv.org/abs/2605.18174

【24】Equilibrium Selection in Multi-Agent Policy Gradients via Opponent-Aware Basin Entry
标题：通过对手意识盆地进入的多主体政策因素的均衡选择
链接：https://arxiv.org/abs/2605.18078

【25】Improving Spatio-Temporal Residual Error Propagation by Mitigating Over-Squashing
标题：通过缓解过压缩来改善时空残留误差传播
链接：https://arxiv.org/abs/2605.18068

【26】The MixCount Dataset: Bridging the Data Gap for Open-Vocabulary Object Counting
标题：MixCount数据集：弥合开放词汇对象计数的数据差距
链接：https://arxiv.org/abs/2605.18063

【27】Protection Is (Nearly) All You Need: Structural Protection Dominates Scoring in Globally Capped KV Eviction
标题：保护（几乎）就是你所需要的一切：结构性保护在全球范围内的KV驱逐中占据主导地位
链接：https://arxiv.org/abs/2605.18053

【28】New Insight of Variance reduce in Zero-Order Hard-Thresholding: Mitigating Gradient Error and Expansivity Contradictions
标题：零阶硬保持中方差减小的新认识：消除梯度误差和可扩展性矛盾
链接：https://arxiv.org/abs/2605.18035

【29】Unveiling Memorization-Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise
标题：揭开简化-概括共存的面纱：具有标签噪音的算术任务案例研究
链接：https://arxiv.org/abs/2605.18022

【30】A More Word-like Image Tokenization for MLLMs
标题：MLLM的更类似单词的图像代币化
链接：https://arxiv.org/abs/2605.17954

【31】SVFSearch: A Multimodal Knowledge-Intensive Benchmark for Short-Video Frame Search in the Gaming Vertical Domain
标题：SVFSearch：游戏垂直领域短视频帧搜索的多模式知识密集型基准
链接：https://arxiv.org/abs/2605.17946

【32】Domain Transfer Becomes Identifiable via a Single Alignment
标题：域名转移可通过单一对齐识别
链接：https://arxiv.org/abs/2605.17918

【33】Attention Sinks and Outliers in Attention Residuals
标题：注意力下沉和注意力残留中的异常值
链接：https://arxiv.org/abs/2605.17887

【34】Multi-site PPG: An In-the-Wild Physiological Dataset from Emerging Multi-site Wearables
标题：多站点PPV：来自新兴多站点可穿戴设备的野外生理数据集
链接：https://arxiv.org/abs/2605.17859

【35】A Unified Framework for Data-Free One-Step Sampling via Wasserstein Gradient Flows
标题：通过Wasserstein梯度流进行无数据一步采样的统一框架
链接：https://arxiv.org/abs/2605.17808

【36】Memisis: Orchestrating and Evaluating Synthetic Data for Tabular Health Datasets
标题：Memisis：描绘和评估表格健康数据集的合成数据
链接：https://arxiv.org/abs/2605.17758

【37】OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization
标题：OTAR：用于2位KV缓存量化的离线光谱协方差感知旋转
链接：https://arxiv.org/abs/2605.17757

【38】Testable and Actionable Calibration for Full Swap Regret
标题：完全交换遗憾的可测试和可操作校准
链接：https://arxiv.org/abs/2605.17749

【39】Divergence-Suppressing Couplings for Rectified Flow
标题：整流用的防发散联轴器
链接：https://arxiv.org/abs/2605.17733

【40】Agent Bazaar: Enabling Economic Alignment in Multi-Agent Marketplaces
标题：Agent Bazaar：实现多主体市场的经济协调
链接：https://arxiv.org/abs/2605.17698

【41】Bug or Feature$^2$: Weight Drift, Activation Sparsity, and Spikes
标题：错误或缺陷$#2 $：重量漂移、激活稀疏和尖峰
链接：https://arxiv.org/abs/2605.17659

【42】Form and Function: Machine Unlearning as a Problem of Misaligned States
标题：形式与功能：机器忘记学习是失调状态的问题
链接：https://arxiv.org/abs/2605.17590

【43】Evaluating Deep Research Agents on Expert Consulting Work: A Benchmark with Verifiers, Rubrics, and Cognitive Traps
标题：评估深度研究代理的专家咨询工作：具有验证者、条目和认知陷阱的基准
链接：https://arxiv.org/abs/2605.17554

【44】Coordinate Heterogeneity Governs Binary Quantization: From InfoNCE to Recall
标题：协调异因支配二进制量化：从InfoNSO到召回
链接：https://arxiv.org/abs/2605.17524

【45】Residual Semantic Decomposition of Word Embeddings
标题：词嵌入的剩余语义分解
链接：https://arxiv.org/abs/2605.17482

【46】TriOpt: A Scalable Algorithm for Linear Causal Discovery
标题：TriOpt：线性因果发现的可扩展算法
链接：https://arxiv.org/abs/2605.17465

【47】A semantic mutation metric for metamorphic relation adequacy in scientific computing programs
标题：科学计算程序中变性关系充分性的语义突变指标
链接：https://arxiv.org/abs/2605.17437

【48】MATE: Solving Contextual Markov Decision Processes with Memory of Accumulated Transition Embeddings
标题 ：MATE：用累积转移嵌入的记忆来解决上下文马尔科夫决策过程
链接：https://arxiv.org/abs/2605.17431

【49】IVF-TQ: Streaming-Robust Approximate Nearest Neighbor Search via a Codebook-Free Residual Layer
标题：IVF-TQ：通过无码本残留层流媒体鲁棒的大约最近邻搜索
链接：https://arxiv.org/abs/2605.17415

【50】MiniGPT: Rebuilding GPT from First Principles
标题：MiniGPT：从首要原则重建GPT
链接：https://arxiv.org/abs/2605.17398

【51】NOETHER: A Constructive Framework for Metamorphic Pattern Discovery from Operator Algebras
标题：NOETHER：从运算子代数发现变形模式的结构性框架
链接：https://arxiv.org/abs/2605.17390

【52】FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics
标题：FML-长凳：从搜索动力学角度对人工智能研究代理策略的对照研究
链接：https://arxiv.org/abs/2605.17373

【53】Weak-to-Strong Elicitation via Mismatched Wrong Drafts
标题：通过不匹配的错误草稿从弱到强的激发
链接：https://arxiv.org/abs/2605.17314

【54】State-of-the-Art Claims Require State-of-the-Art Evidence
标题：最先进的索赔需要最先进的证据
链接：https://arxiv.org/abs/2605.17273

【55】Calibeating for general proper losses: A Bregman divergence approach
标题：一般适当损失的校准：布雷格曼分歧方法
链接：https://arxiv.org/abs/2605.17269

【56】When Molecular Similarity Works: Property Cliffs Reveal Hidden Errors
标题：当分子相似性发挥作用时：财产悬崖揭示隐藏的错误
链接：https://arxiv.org/abs/2605.17265

【57】Fidelity Probes for Specification--Code Alignment
标题：规范的富达探针--代码对齐
链接：https://arxiv.org/abs/2605.17246

【58】Drift Flow Matching
标题：漂移流匹配
链接：https://arxiv.org/abs/2605.17244

【59】The Geometry of Projection Heads: Conditioning, Invariance, and Collapse
标题：投影头的几何形状：条件反射、不变性和崩溃
链接：https://arxiv.org/abs/2605.17180

【60】Why Do Safety Guardrails Degrade Across Languages?
标题：为什么安全护栏会在不同语言中退化？
链接：https://arxiv.org/abs/2605.17173

【61】OpenJarvis: Personal AI, On Personal Devices
标题：OpenJarvis：个人人工智能，在个人设备上
链接：https://arxiv.org/abs/2605.17172

【62】Factorized Latent Dynamics for Video JEPA: An Empirical Study of Auxiliary Objectives
标题：视频JEPA的因子化潜在动态：辅助目标的实证研究
链接：https://arxiv.org/abs/2605.17165

【63】When Bits Break Recourse: Counterfactual-Faithful Quantization
标题：当比特中断追索权时：反事实忠实量化
链接：https://arxiv.org/abs/2605.17160

【64】An Analytical Multiple Criteria Framework for Temporal and Dynamic Business-to-Business Customer Segmentation in Manufacturing
标题：制造业中时态和动态企业对企业客户细分的分析多标准框架
链接：https://arxiv.org/abs/2605.17151

【65】DynMuon: A Dynamic Spectral Shaping View of Muon
标题：DynMuon：Muon的动态光谱整形视图
链接：https://arxiv.org/abs/2605.17109

【66】Parallel Recursive LSTM
标题：并行回归LSTM
链接：https://arxiv.org/abs/2605.17108

【67】Taming Audio VAEs via Target-KL Regularization
标题：通过Target-KL规范化驯服音频VAE
链接：https://arxiv.org/abs/2605.17085

【68】1GC-7RC: One Graphic Card -- Seven Research Challenges! How Good Are AI Agents at Doing Your Job?
标题：1GC-7 RC：一张图形卡--七大研究挑战！人工智能代理在完成您的工作方面有多好？
链接：https://arxiv.org/abs/2605.17046

【69】Reliability and Effectiveness of Autonomous AI Agents in Supply Chain Management
标题：供应链管理中自主人工智能代理的可靠性和有效性
链接：https://arxiv.org/abs/2605.17036

【70】Topo-GS: Continuous Volumetric Embedding of High-Dimensional Data via Topological Gaussian Splatting
标题：Topo-GS：通过Topological Gaussian Splatting实现多维数据的连续体积嵌入
链接：https://arxiv.org/abs/2605.17011

【71】SHED: Style-Homogenized Embedding Alignment for Domain Generalization
标题：SHED：用于领域概括的风格同质化嵌入对齐
链接：https://arxiv.org/abs/2605.16973

【72】ArtifactLinker: Linking Scientific Artifacts for Automatic State-of-the-Art Discovery
标题：ArtifactLinker：连接科学文物以自动发现最新技术水平
链接：https://arxiv.org/abs/2605.16902

【73】SE-GA: Memory-Augmented Self-Evolution for GUI Agents
标题：SE-GA：面向图形用户界面代理的内存增强自我进化
链接：https://arxiv.org/abs/2605.16883

【74】Thinking with Patterns: Breaking the Perceptual Bottleneck in Visual Planning via Pattern Induction
标题：用模式思考：通过模式归纳打破视觉规划的感知瓶颈
链接：https://arxiv.org/abs/2605.16848

【75】TIER: Trajectory-Invariant Execution Rewards for Multi-Step Tool Composition
标题：TIER：多步骤工具组合的轨迹不变执行奖励
链接：https://arxiv.org/abs/2605.16790

【76】Propagation of Chaos in Contextual Flow Maps
标题：上下文流图中的混乱传播
链接：https://arxiv.org/abs/2605.16747

【77】EmoMind: Decoding Affective Captions from Human Brain fMRI
标题：大脑功能核磁共振成像解码情感字幕
链接：https://arxiv.org/abs/2605.16739

【78】Convex Dataset Valuation for Post-Training
标题：训练后的凸数据集估值
链接：https://arxiv.org/abs/2605.16704

【79】EfficientTDMPC: Improved MPC Objectives for Sample-Efficient Continuous Control
标题：EfficientTDMPC：改进MPC目标，实现样本高效的连续控制
链接：https://arxiv.org/abs/2605.16692

【80】The Score Kalman Filter
标题：得分卡尔曼过滤器
链接：https://arxiv.org/abs/2605.16644

【81】Provably Shorter Scratchpads in Hybrid DeltaNet-Attention Decoders
标题：混合Delta Net-Attention解码器中明显更短的键盘
链接：https://arxiv.org/abs/2605.16640

【82】Does Weight Decay Enhance Training Stability?
标题：体重下降会增强训练稳定性吗？
链接：https://arxiv.org/abs/2605.16622

【83】SCOUT: Cyclic Causal Discovery Under Soft Interventions with Unknown Targets
标题：SCOT：目标未知的软干预下的循环因果发现
链接：https://arxiv.org/abs/2605.16620

【84】Tensor Cookbook: Mastering Tensors through Diagrams
标题：张量食谱：通过图表掌握张量
链接：https://arxiv.org/abs/2605.16610

【85】To MRL or not to MRL: Text Embeddings are Robust to Truncation Without Matryoshka Embeddings, Except In Heavy Truncation Scenarios
标题：无论是否适用于MRL：文本嵌入在没有Matryoshka嵌入的情况下对截断具有鲁棒性，除非在严重截断的情况下
链接：https://arxiv.org/abs/2605.16608

【86】R2V Agent: Teaching SLMs When to Ask for Help
标题：R2 V代理：教LM何时寻求帮助
链接：https://arxiv.org/abs/2605.16604

【87】Attend Locally, Remember Linearly: Linear Attention as Cross-Frame Memory for Autoregressive Video Diffusion
标题：局部注意，线性记忆：线性注意作为自回归视频扩散的跨帧记忆
链接：https://arxiv.org/abs/2605.16579

【88】Voice ''Cloning'' is Style Transfer
标题：声音“克隆”是风格转移
链接：https://arxiv.org/abs/2605.16578

【89】Wavelet Flow Matching for Multi-Scale Physics Emulation
标题：多尺度物理仿真的小波流匹配
链接：https://arxiv.org/abs/2605.16573

【90】Hypergraph Pattern Machine: Compositional Tokenization for Higher-Order Interactions
标题：Hypergraph Pattern Machine：高阶交互的组合标记化
链接：https://arxiv.org/abs/2605.16527

【91】SeamCam: Quantifying Seamless Camouflage via Multi-Cue Visual Detectability
标题：DeliverCam：通过多线索视觉检测量化无缝伪装
链接：https://arxiv.org/abs/2605.16515

【92】Avoiding Structural Failure Modes in Tabular Fair SSL: Online Primal-Dual Allocation under Confidence Gating
标题：避免表格公平SSL中的结构性故障模式：置信门控下的在线主二元分配
链接：https://arxiv.org/abs/2605.16446

【93】Stable and Near-Reversible Diffusion ODE Solvers for Image Editing
标题：用于图像编辑的稳定且近可逆的扩散ODE解算器
链接：https://arxiv.org/abs/2605.16399

【94】Beyond MMSE: Enhancing PnP Restoration with ProxiMAP
标题：超越MBE：使用ProbiMAP增强激情恢复
链接：https://arxiv.org/abs/2605.16396

【95】An Information-Theoretic Criterion for Efficient Data Synthesis
标题：有效数据合成的信息论准则
链接：https://arxiv.org/abs/2605.16379

【96】ORACLE: Anticipating Scams from Partial Trajectories in Streaming App Usage
标题：Oracle：预计流媒体应用程序使用中的部分轨迹会出现诈骗
链接：https://arxiv.org/abs/2605.16363

【97】When Is Rank-1 Steering Cheap? Geometry, Granularity, and Budgeted Search
标题：一级转向什么时候便宜？几何、粒度和默认搜索
链接：https://arxiv.org/abs/2605.16362

【98】Geometric Asymmetry in MoE Specialization: Functional Decorrelation and Representational Overlap
标题：MoE专业化中的几何不对称：功能去相关与表征重叠
链接：https://arxiv.org/abs/2605.16349

【99】Language Game: Talking to Non-Human Systems
标题：语言游戏：与非人类系统对话
链接：https://arxiv.org/abs/2605.16321

【100】MANTA: Multi-turn Assessment for Nonhuman Thinking & Alignment
标题：MANTA：非人类思维和一致性的多回合评估
链接：https://arxiv.org/abs/2605.16301

【101】Mirror Descent-Type Algorithms for the Variational Inequality Problem with Functional Constraints
标题：具有函数约束的变分不等式问题的镜像下降型算法
链接：https://arxiv.org/abs/2605.16262

【102】SURGE: Approximation-free Training Free Particle Filter for Diffusion Surrogate
标题：SURGE：用于扩散代理的免逼近免训练颗粒过滤器
链接：https://arxiv.org/abs/2605.18745

【103】AI4BayesCode: From Natural Language Descriptions to Validated Modular Stateful Bayesian Samplers
标题：AI 4 BayesCode：从自然语言描述到经过验证的模块化状态Bayesian采样器
链接：https://arxiv.org/abs/2605.18476

【104】Flowing with Confidence
标题：充满信心
链接：https://arxiv.org/abs/2605.18472

【105】On Stability and Decomposition of Sample Quantiles under Heavy-Tailed Distributions
标题：重尾分布下样本分位数的稳定性与分解
链接：https://arxiv.org/abs/2605.18370

【106】Hybrid Quantum-Classical Neural Architecture Search
标题：混合量子-经典神经架构搜索
链接：https://arxiv.org/abs/2605.18345

【107】Robust Player-Conditional Champion Ranking for League of Legends: Style Similarity, Mastery Priors, and Archetype-Constrained Discovery
标题：英雄联盟稳健的球员条件冠军排名：风格相似性、掌握经验和原型限制的发现
链接：https://arxiv.org/abs/2605.18338

【108】Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers
标题：优化器设计的对称兼容原则：嵌入、LM头、SwiGLU MLP和MoE路由器
链接：https://arxiv.org/abs/2605.18106

【109】Real-time Multi-instrument Autonomous Discovery of Novel Phase-change Memory Materials
标题：新型相变化存储材料的实时多仪器自主发现
链接：https://arxiv.org/abs/2605.18033

【110】Sequential Structure in Intraday Futures Data: LSTM vs Gradient Boosting on MNQ
标题：日内期货数据的序列结构：LSTM与MNQ上的梯度提升
链接：https://arxiv.org/abs/2605.17724

【111】ML-based Fast Simulation of FARICH Responses
标题：基于ML的FARICH响应快速模拟
链接：https://arxiv.org/abs/2605.17635

【112】Maximum Likelihood Decoding of Quantum Error Correction Codes
标题：量子错误纠正码的最大似然解码
链接：https://arxiv.org/abs/2605.17230

【113】Sample efficient inductive matrix completion with noise and inexact side information
标题：具有噪音和不精确边信息的高效归纳矩阵完成样本
链接：https://arxiv.org/abs/2605.17189

【114】Multi-task Linear Regression without Eigenvalue Lower Bounds: Adaptivity, Robustness and Safety
标题：无特征值下限的多任务线性回归：适应性、稳健性和安全性
链接：https://arxiv.org/abs/2605.17126

【115】$\mathcal{O}(n)$ alternative to Quantum Fourier Transform with efficient neural net classical post-processing
链接：https://arxiv.org/abs/2605.16998

【116】CAST: Causal Anchored Simplex Transport for Distribution-Valued Time Series
标题：演员：分布值时间序列的因果锚定单形传输
链接：https://arxiv.org/abs/2605.16919

【117】Statistical Unlearning of Distributions: A Hypothesis Testing Approach
标题：分布的统计非学习：一种假设检验方法
链接：https://arxiv.org/abs/2605.16645

【118】StAD: Stein Amortized Divergence for Fast Likelihoods with Diffusion and Flow
标题：StAD：扩散和流动快速似然的Stein分摊散度
链接：https://arxiv.org/abs/2605.16486

【119】Bridging the Modality Bottleneck in Pathology MIL through Virtual Molecular Staining
标题：通过虚拟分子染色弥合病理学MIL中的情态瓶颈
链接：https://arxiv.org/abs/2605.16392

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