机器学习学术速递[5.13]

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360篇

大模型相关(55篇)

【1】Task-Adaptive Embedding Refinement via Test-time LLM Guidance
标题：通过测试时LLM指导的任务自适应嵌入细化
链接：https://arxiv.org/abs/2605.12487

【2】Learning, Fast and Slow: Towards LLMs That Adapt Continually
标题：学习，快与慢：迈向不断适应的法学硕士
链接：https://arxiv.org/abs/2605.12484

【3】Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training
标题：超越GRPO和策略上蒸馏：一个经验性的稀疏到密集的奖励原则，用于存储模型后训练
链接：https://arxiv.org/abs/2605.12483

【4】Solve the Loop: Attractor Models for Language and Reasoning
标题：解决循环：语言和推理的吸引器模型
链接：https://arxiv.org/abs/2605.12466

【5】Multi-Stream LLMs: Unblocking Language Models with Parallel Streams of Thoughts, Inputs and Outputs
标题：多流LLM：通过并行思想、输入和输出流解锁语言模型
链接：https://arxiv.org/abs/2605.12460

【6】TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection
标题：TextSeal：一个本地化LLM水印，用于源与蒸馏保护
链接：https://arxiv.org/abs/2605.12456

【7】ORCE: Order-Aware Alignment of Verbalized Confidence in Large Language Models
标题：ORCE：大型语言模型中言语可信度的订单感知对齐
链接：https://arxiv.org/abs/2605.12446

【8】OGLS-SD: On-Policy Self-Distillation with Outcome-Guided Logit Steering for LLM Reasoning
标题：OGLS-SD：基于政策的自我蒸馏和结果引导的Logit引导，用于LLM推理
链接：https://arxiv.org/abs/2605.12400

【9】Scalable Token-Level Hallucination Detection in Large Language Models
标题：大型语言模型中的可扩展令牌级幻觉检测
链接：https://arxiv.org/abs/2605.12384

【10】Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models
标题：填补差距：多模式大型语言模型中视觉推理的粒度对齐范式
链接：https://arxiv.org/abs/2605.12374

【11】Grid Games: The Power of Multiple Grids for Quantizing Large Language Models
标题：网格游戏：量化大型语言模型的多个网格的力量
链接：https://arxiv.org/abs/2605.12327

【12】PriorZero: Bridging Language Priors and World Models for Decision Making
标题：PriorZero：连接语言先验和决策的世界模型
链接：https://arxiv.org/abs/2605.12289

【13】Instruction Lens Score: Your Instruction Contributes a Powerful Object Hallucination Detector for Multimodal Large Language Models
标题：指令镜头评分：您的指令为多模式大型语言模型提供了强大的对象幻觉检测器
链接：https://arxiv.org/abs/2605.12258

【14】A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning
标题：多域多任务图对齐指令调优的统一图语言模型
链接：https://arxiv.org/abs/2605.12197

【15】PrivacySIM: Evaluating LLM Simulation of User Privacy Behavior
标题：PrivacySIM：评估用户隐私行为的LLM模拟
链接：https://arxiv.org/abs/2605.12147

【16】Efficient and Adaptive Human Activity Recognition via LLM Backbones
标题：通过LLM主干进行高效且自适应的人类活动识别
链接：https://arxiv.org/abs/2605.12019

【17】The Illusion of Power Capping in LLM Decode: A Phase-Aware Energy Characterisation Across Attention Architectures
标题：LLM解码中的权力上限幻觉：注意力架构中的阶段感知能量特征
链接：https://arxiv.org/abs/2605.11999

【18】Towards Order Fairness: Mitigating LLMs Order Sensitivity through Dual Group Advantage Optimization
标题：迈向订单公平：通过双集团优势优化缓解LLM订单敏感性
链接：https://arxiv.org/abs/2605.11974

【19】Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models
标题：Qwen-Scope：将稀疏功能转化为大型语言模型的开发工具
链接：https://arxiv.org/abs/2605.11887

【20】Beyond Parameter Aggregation: Semantic Consensus for Federated Fine-Tuning of LLMs
标题：超越参数聚合：LLM联邦微调的语义共识
链接：https://arxiv.org/abs/2605.11857

【21】GEAR: Granularity-Adaptive Advantage Reweighting for LLM Agents via Self-Distillation
标题：GEAR：通过自蒸馏对LLM代理进行粒度自适应优势重新加权
链接：https://arxiv.org/abs/2605.11853

【22】ROMER: Expert Replacement and Router Calibration for Robust MoE LLMs on Analog Compute-in-Memory Systems
标题：ROMER：模拟计算机内存系统上稳健的MoE LLM的专家替换和路由器校准
链接：https://arxiv.org/abs/2605.11800

【23】From Token to Token Pair: Efficient Prompt Compression for Large Language Models in Clinical Prediction
标题：从令牌到令牌对：临床预测中大型语言模型的高效提示压缩
链接：https://arxiv.org/abs/2605.11774

【24】Training-Inference Consistent Segmented Execution for Long-Context LLMs
标题：长上下文LLM的训练-推理一致分段执行
链接：https://arxiv.org/abs/2605.11744

【25】U-STS-LLM A Unified Spatio-Temporal Steered Large Language Model for Traffic Prediction and Imputation
标题：U-STS-LLM用于流量预测和归责的统一时空定向大语言模型
链接：https://arxiv.org/abs/2605.11735

【26】Block-R1: Rethinking the Role of Block Size in Multi-domain Reinforcement Learning for Diffusion Large Language Models
标题：Block-R1：重新思考块大小在扩散大型语言模型的多域强化学习中的作用
链接：https://arxiv.org/abs/2605.11726

【27】EPIC: Efficient Predicate-Guided Inference-Time Control for Compositional Text-to-Image Generation
标题：EPIC：用于合成文本到图像生成的高效预测引导推理时间控制
链接：https://arxiv.org/abs/2605.11722

【28】GRAFT: Graph-Tokenized LLMs for Tool Planning
标题：GRAFT：用于工具规划的图形令牌化LLM
链接：https://arxiv.org/abs/2605.11706

【29】Targeted Tests for LLM Reasoning: An Audit-Constrained Protocol
标题：LLM推理的有针对性测试：一种听觉约束协议
链接：https://arxiv.org/abs/2605.11599

【30】AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive
标题：AutoLLMResearch：训练研究代理以自动化LLM实验配置--从廉价、优化昂贵的地方学习
链接：https://arxiv.org/abs/2605.11518

【31】Hierarchical LLM-Driven Control for HAPS-Assisted UAV Networks: Joint Optimization of Flight and Connectivity
标题：HAPS辅助无人机网络的分层LLM驱动控制：飞行和连接性的联合优化
链接：https://arxiv.org/abs/2605.11509

【32】CTFusion: A CTF-based Benchmark for LLM Agent Evaluation
标题：CTFusion：基于CTF的LLM代理评估基准
链接：https://arxiv.org/abs/2605.11504

【33】Breaking $\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning
标题：破$ extit{Winner-Takes-All}$：合作政策优化改进了多元化的LLM推理
链接：https://arxiv.org/abs/2605.11461

【34】Predictive Maps of Multi-Agent Reasoning: A Successor-Representation Spectrum for LLM Communication Topologies
标题：多智能体推理的预测地图：LLM通信布局的内部表示谱
链接：https://arxiv.org/abs/2605.11453

【35】20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone
标题：20/20视觉语言模型：仅通过数据修复即可提供更好的VLM处方
链接：https://arxiv.org/abs/2605.11405

【36】VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference
标题：VERDI：通过分解推理对基于验证的LLM法官进行单次呼叫置信度估计
链接：https://arxiv.org/abs/2605.11334

【37】Curriculum Learning-Guided Progressive Distillation in Large Language Models
标题：大语言模型中课程学习引导的渐进蒸馏
链接：https://arxiv.org/abs/2605.11260

【38】Internalizing Curriculum Judgment for LLM Reinforcement Fine-Tuning
标题：LLM强化微调的课程评价内化
链接：https://arxiv.org/abs/2605.11235

【39】Rethinking LLMOps for Fraud and AML: Building a Compliance-Grade LLM Serving Stack
标题：重新思考针对欺诈和反洗钱的LLMOp：构建合规级LLM服务栈
链接：https://arxiv.org/abs/2605.11232

【40】PIVOT: Bridging Planning and Execution in LLM Agents via Trajectory Refinement
标题：PIDOT：通过轨迹细化连接LLM代理的规划和执行
链接：https://arxiv.org/abs/2605.11225

【41】ADMM-Q: An Improved Hessian-based Weight Quantizer for Post-Training Quantization of Large Language Models
标题：ADMM-Q：一种改进的基于Hessian的权重量化器，用于大型语言模型的训练后量化
链接：https://arxiv.org/abs/2605.11222

【42】Leveraging RAG for Training-Free Alignment of LLMs
标题：利用RAG进行LLM的免训练调整
链接：https://arxiv.org/abs/2605.11217

【43】Measuring Five-Nines Reliability: Sample-Efficient LLM Evaluation in Saturated Benchmarks
标题：衡量五个九的可靠性：饱和基准中的样本高效LLM评估
链接：https://arxiv.org/abs/2605.11209

【44】Continuous Discovery of Vulnerabilities in LLM Serving Systems with Fuzzing
标题：使用Fuzzing连续发现LLM服务系统中的漏洞
链接：https://arxiv.org/abs/2605.11202

【45】CATS: Cascaded Adaptive Tree Speculation for Memory-Limited LLM Inference Acceleration
标题：CATS：用于内存有限LLM推理加速的级联自适应树推测
链接：https://arxiv.org/abs/2605.11186

【46】The Bicameral Model: Bidirectional Hidden-State Coupling Between Parallel Language Models
标题：两院模型：并行语言模型之间的双向隐藏状态耦合
链接：https://arxiv.org/abs/2605.11167

【47】Enabling Performant and Flexible Model-Internal Observability for LLM Inference
标题：实现LLM推理的性能和灵活的模型内部可观察性
链接：https://arxiv.org/abs/2605.11093

【48】MCPShield: Content-Aware Attack Detection for LLM Agent Tool-Call Traffic
标题：MCPShield：LLM代理工具呼叫流量的内容感知攻击检测
链接：https://arxiv.org/abs/2605.11053

【49】Efficient LLM Reasoning via Variational Posterior Guidance with Efficiency Awareness
标题：通过具有效率意识的变分后验引导进行高效LLM推理
链接：https://arxiv.org/abs/2605.11019

【50】LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models
标题：LoopUS：将预训练的LLM重新转换为循环潜在细化模型
链接：https://arxiv.org/abs/2605.11011

【51】Structural Interpretations of Protein Language Model Representations via Differentiable Graph Partitioning
标题：通过可微图划分对蛋白质语言模型表示的结构解释
链接：https://arxiv.org/abs/2605.10985

【52】Steering Without Breaking: Mechanistically Informed Interventions for Discrete Diffusion Language Models
标题：不间断地引导：离散扩散语言模型的机械知情干预
链接：https://arxiv.org/abs/2605.10971

【53】FERA: Uncertainty-Aware Federated Reasoning for Large Language Models
标题：FERA：大型语言模型的不确定性感知联邦推理
链接：https://arxiv.org/abs/2605.10082

【54】MultiSoc-4D: A Benchmark for Diagnosing Instruction-Induced Label Collapse in Closed-Set LLM Annotation of Bengali Social Media
标题：MultiSoc-4D：诊断孟加拉社交媒体封闭式LLM注释中教学引起的标签崩溃的基准
链接：https://arxiv.org/abs/2605.06940

【55】Quantifying the Reconstructability of Astrophysical Methods with Large Language Models and Information Theory: A Case Study in Spectral Reconstruction
标题：用大语言模型和信息论量化天体物理方法的可重建性：光谱重建案例研究
链接：https://arxiv.org/abs/2605.11154

Graph相关(图学习|图神经网络|图优化等)(12篇)

【1】Learning Minimally Rigid Graphs with High Realization Counts
标题：学习具有高实现次数的最小刚性图
链接：https://arxiv.org/abs/2605.12427

【2】SEMIR: Semantic Minor-Induced Representation Learning on Graphs for Visual Segmentation
标题：SEMIR：用于视觉分割的图形上的语义次诱导表示学习
链接：https://arxiv.org/abs/2605.12389

【3】From Message-Passing to Linearized Graph Sequence Models
标题：从消息传递到线性化图序列模型
链接：https://arxiv.org/abs/2605.12358

【4】Transferable Delay-Aware Reinforcement Learning via Implicit Causal Graph Modeling
标题：通过隐式因果图建模的可转移延迟感知强化学习
链接：https://arxiv.org/abs/2605.12312

【5】Approximation of Maximally Monotone Operators : A Graph Convergence Perspective
标题：最大单调运算符的逼近：图收敛的角度
链接：https://arxiv.org/abs/2605.12301

【6】Random-Set Graph Neural Networks
标题：随机集图神经网络
链接：https://arxiv.org/abs/2605.11987

【7】STAGE: Tackling Semantic Drift in Multimodal Federated Graph Learning
标题：阶段：解决多模式联邦图学习中的语义漂移
链接：https://arxiv.org/abs/2605.11919

【8】Crash Assessment via Mesh-Based Graph Neural Networks and Physics-Aware Attention
标题：通过基于网格的图神经网络和物理感知注意力进行碰撞评估
链接：https://arxiv.org/abs/2605.11784

【9】Learning Feature Encoder with Synthetic Anomalies for Weakly Supervised Graph Anomaly Detection
标题：用于弱监督图异常检测的具有合成异常的学习特征编码器
链接：https://arxiv.org/abs/2605.11749

【10】CORE: Cyclic Orthotope Relation Embedding for Knowledge Graph Completion
标题：核心：知识图完成的循环正位关系嵌入
链接：https://arxiv.org/abs/2605.11159

【11】Rank Is Not Capacity: Spectral Occupancy for Latent Graph Models
标题：排名不是容量：潜图模型的光谱占据
链接：https://arxiv.org/abs/2605.11142

【12】Hierarchical Multi-Scale Graph Neural Networks: Scalable Heterophilous Learning with Oversmoothing and Oversquashing Mitigation
标题：分层多尺度图神经网络：具有过度平滑和过度挤压缓解的可扩展异类学习
链接：https://arxiv.org/abs/2605.10975

Transformer(10篇)

【1】Elastic Attention Cores for Scalable Vision Transformers
标题：可扩展视觉Transformer的弹性注意力核心
链接：https://arxiv.org/abs/2605.12491

【2】Lower bounds for one-layer transformers that compute parity
标题：计算宇称的单层转换器的下限
链接：https://arxiv.org/abs/2605.12171

【3】ChunkFlow: Communication-Aware Chunked Prefetching for Layerwise Offloading in Distributed Diffusion Transformer Inference
标题：ChunkFlow：分布式扩散Transformer推理中的通信感知分块预取分层卸载
链接：https://arxiv.org/abs/2605.11335

【4】Latent Chain-of-Thought Improves Structured-Data Transformers
标题：潜在思想链改进结构化数据Transformer
链接：https://arxiv.org/abs/2605.11262

【5】Variational Linear Attention: Stable Associative Memory for Long-Context Transformers
标题：变分线性注意力：长上下文变形者的稳定联想记忆
链接：https://arxiv.org/abs/2605.11196

【6】ACSAC: Adaptive Chunk Size Actor-Critic with Causal Transformer Q-Network
标题：ACSAC：基于因果Transformer Q网络的自适应块大小Actor-Critic算法
链接：https://arxiv.org/abs/2605.11009

【7】RT-Transformer: The Transformer Block as a Spherical State Estimator
标题：RT-Transformer：作为球形状态估计器的Transformer模块
链接：https://arxiv.org/abs/2605.11007

【8】Vertex-Softmax: Tight Transformer Verification via Exact Softmax Optimization
标题：Vertex-Softmax：通过精确Softmax优化进行紧密Transformer验证
链接：https://arxiv.org/abs/2605.10974

【9】Bin Latent Transformer (BiLT): A shift-invariant autoencoder for calibration-free spectral unmixing of turbid media
标题：Bin Latent Transformer（BiLT）：一种位移不变自动编码器，用于对浑浊媒体进行免校准光谱解混
链接：https://arxiv.org/abs/2605.11829

【10】Uniform Scaling Limits in AdamW-Trained Transformers
标题：AdamW训练的Transformer中的统一缩放限制
链接：https://arxiv.org/abs/2605.11059

GAN|对抗|攻击|生成相关(17篇)

【1】AlphaGRPO: Unlocking Self-Reflective Multimodal Generation in UMMs via Decompositional Verifiable Reward
标题：AlphaGRPO：通过分解可验证奖励解锁UMM中的自反射多模式生成
链接：https://arxiv.org/abs/2605.12495

【2】Attacks and Mitigations for Distributed Governance of Agentic AI under Byzantine Adversaries
标题：拜占庭攻击下分布式人工智能治理的攻击与缓解
链接：https://arxiv.org/abs/2605.12364

【3】STRUM: A Spectral Transcription and Rhythm Understanding Model for End-to-End Generation of Playable Rhythm-Game Charts
标题：SEARCH：用于端到端生成可玩节奏游戏图表的光谱转录和节奏理解模型
链接：https://arxiv.org/abs/2605.12135

【4】MULTI: Disentangling Camera Lens, Sensor, View, and Domain for Novel Image Generation
标题：MULTI：解开相机镜头、传感器、视图和领域，以生成新颖的图像
链接：https://arxiv.org/abs/2605.12134

【5】SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces
标题：SkillSafetyBench：评估面向技能的攻击面下的特工安全性
链接：https://arxiv.org/abs/2605.12015

【6】Persona-Conditioned Adversarial Prompting: Multi-Identity Red-Teaming for Adversarial Discovery and Mitigation
标题：角色条件对抗预算：多身份红色团队对抗发现和缓解
链接：https://arxiv.org/abs/2605.11730

【7】Primal-Dual Policy Optimization for Linear CMDPs with Adversarial Losses
标题：具有对抗性损失的线性CMDP的主二元策略优化
链接：https://arxiv.org/abs/2605.11535

【8】Robust Multi-Agent Path Finding under Observation Attacks: A Principled Adversarial-Plus-Smoothing Training Recipe
标题：观察攻击下的稳健多智能体路径查找：原则性的对抗加平滑训练食谱
链接：https://arxiv.org/abs/2605.11469

【9】Test-Time Compute for Dense Retrieval: Agentic Program Generation with Frozen Embedding Models
标题：密集检索的测试时计算：使用冻结嵌入模型的大型程序生成
链接：https://arxiv.org/abs/2605.11374

【10】LPDP: Inference-Time Reward Control for Variable-Length DNA Generation with Edit Flows
标题：LSDP：具有编辑流的可变长度DNA生成的推理时奖励控制
链接：https://arxiv.org/abs/2605.11368

【11】Vision2Code: A Multi-Domain Benchmark for Evaluating Image-to-Code Generation
标题：Vision 2Code：评估图像到代码生成的多域基准
链接：https://arxiv.org/abs/2605.11307

【12】A Theory of Time-Sensitive Language Generation: Sparse Hallucination Beats Mode Collapse
标题：时间敏感语言生成理论：稀疏幻觉击败模式崩溃
链接：https://arxiv.org/abs/2605.11302

【13】Primal Generation, Dual Judgment: Self-Training from Test-Time Scaling
标题：原始一代，双重判断：测试时间缩放的自我训练
链接：https://arxiv.org/abs/2605.11299

【14】ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?
标题：ExploitGym：人工智能代理能否将安全漏洞转化为真正的攻击？
链接：https://arxiv.org/abs/2605.11086

【15】AESOP: Adversarial Execution-path Selection to Overload Deep Learning Pipelines
标题：AESOP：对抗性执行路径选择，以过载深度学习管道
链接：https://arxiv.org/abs/2605.10987

【16】Assessment of cloud and associated radiation fields from a GAN stochastic cloud subcolumn generator
标题：来自GAN随机云子柱发生器的云和相关辐射场的评估
链接：https://arxiv.org/abs/2605.11968

【17】Adversarial Effects on Expressibility and Trainability in Distributed Variational Quantum Algorithms
标题：分布式变分量子算法中的可表达性和可训练性的对抗效应
链接：https://arxiv.org/abs/2605.03629

半/弱/无/有监督|不确定性|主动学习(15篇)

【1】A Semi-Supervised Framework for Speech Confidence Detection using Whisper
标题：使用Whisper的语音置信度检测半监督框架
链接：https://arxiv.org/abs/2605.12387

【2】From Model Uncertainty to Human Attention: Localization-Aware Visual Cues for Scalable Annotation Review
标题：从模型不确定性到人类注意力：可扩展注释审查的本地化视觉线索
链接：https://arxiv.org/abs/2605.12303

【3】Reconstruction of Personally Identifiable Information from Supervised Finetuned Models
标题：基于监督微调模型的个人身份信息重构
链接：https://arxiv.org/abs/2605.12264

【4】Information theoretic underpinning of self-supervised learning by clustering
标题：集群自我监督学习的信息理论基础
链接：https://arxiv.org/abs/2605.11870

【5】Martingale-Consistent Self-Supervised Learning
标题：马丁格一致的自我监督学习
链接：https://arxiv.org/abs/2605.11846

【6】Beyond Prediction: Interval Neural Networks for Uncertainty-Aware System Identification
标题 ：超越预测：用于不确定性感知系统识别的区间神经网络
链接：https://arxiv.org/abs/2605.11460

【7】Epistemic Uncertainty for Test-Time Discovery
标题：测试时间发现的认识不确定性
链接：https://arxiv.org/abs/2605.11328

【8】DenseTRF: Texture-Aware Unsupervised Representation Adaptation for Surgical Scene Dense Prediction
标题：DenseTRF：用于手术场景密集预测的纹理感知无监督表示适应
链接：https://arxiv.org/abs/2605.11265

【9】HEPA: A Self-Supervised Horizon-Conditioned Event Predictive Architecture for Time Series
标题：HEPA：一种用于时间序列的自监督地平线条件事件预测架构
链接：https://arxiv.org/abs/2605.11130

【10】Seeing the Needle in the Haystack: Towards Weakly-Supervised Log Instance Anomaly Localization via Counterfactual Perturbation
标题：大难不死：通过反事实扰动实现监督较弱的日志实例异常定位
链接：https://arxiv.org/abs/2605.10988

【11】Rotation-Preserving Supervised Fine-Tuning
标题：旋转保持监督微调
链接：https://arxiv.org/abs/2605.10973

【12】Self-Supervised Laplace Approximation for Bayesian Uncertainty Quantification
标题：Bayesian不确定性量化的自监督拉普拉斯逼近
链接：https://arxiv.org/abs/2605.12208

【13】Learning U-Statistics with Active Inference
标题：通过主动推理学习U统计
链接：https://arxiv.org/abs/2605.11638

【14】Exact Stiefel Optimization for Probabilistic PLS: Closed-Form Updates, Error Bounds, and Calibrated Uncertainty
标题：概率最大化（Probabilitary）的精确Stiefel优化：封闭式更新、误差界和校准的不确定性
链接：https://arxiv.org/abs/2605.11607

【15】Deploying Self-Supervised Learning for Real Seismic Data Denoising
标题：部署自我监督学习以实现真实地震数据去噪
链接：https://arxiv.org/abs/2605.11109

迁移|Zero/Few/One-Shot|自适应(17篇)

【1】Environment-Adaptive Preference Optimization for Wildfire Prediction
标题：野火预测的环境适应性偏好优化
链接：https://arxiv.org/abs/2605.12435

【2】Trust the Batch, On- or Off-Policy: Adaptive Policy Optimization for RL Post-Training
标题：信任批次、政策开放或非政策：RL训练后的自适应策略优化
链接：https://arxiv.org/abs/2605.12380

【3】Not How Many, But Which: Parameter Placement in Low-Rank Adaptation
标题：不是多少，而是哪个：低等级适应中的参数放置
链接：https://arxiv.org/abs/2605.12207

【4】Learning What Matters: Adaptive Information-Theoretic Objectives for Robot Exploration
标题：学习重要的事情：机器人探索的自适应信息理论目标
链接：https://arxiv.org/abs/2605.12084

【5】Anomaly-Aware Vision-Language Adapters for Zero-Shot Anomaly Detection
标题：用于Zero-Shot异常检测的异常感知视觉语言适配器
链接：https://arxiv.org/abs/2605.12069

【6】Adaptive TD-Lambda for Cooperative Multi-agent Reinforcement Learning
标题：协作多智能体强化学习的自适应TD-Lambda
链接：https://arxiv.org/abs/2605.11880

【7】Unlocking Compositional Generalization in Continual Few-Shot Learning
标题：在连续Few-Shot学习中解锁成分概括
链接：https://arxiv.org/abs/2605.11710

【8】Shaping Zero-Shot Coordination via State Blocking
标题：通过状态阻塞塑造Zero-Shot协调
链接：https://arxiv.org/abs/2605.11688

【9】Adaptive Calibration in Non-Stationary Environments
标题：非平稳环境下的自适应校准
链接：https://arxiv.org/abs/2605.11490

【10】Generative Diffusion Prior Distillation for Long-Context Knowledge Transfer
标题：长上下文知识转移的生成扩散优先蒸馏
链接：https://arxiv.org/abs/2605.11414

【11】fg-expo: Frontier-guided exploration-prioritized policy optimization via adaptive kl and gaussian curriculum
标题：fg-expo：通过自适应kl和高斯课程进行前沿引导的探索优先策略优化
链接：https://arxiv.org/abs/2605.11403

【12】Enforcing Constraints in Generative Sampling via Adaptive Correction Scheduling
标题：通过自适应修正调度在生成采样中强制约束
链接：https://arxiv.org/abs/2605.11214

【13】SURGE: Surrogate Gradient Adaptation in Binary Neural Networks
标题：SURGE：二元神经网络中的替代梯度自适应
链接：https://arxiv.org/abs/2605.10989

【14】Empowering VLMs for Few-Shot Multimodal Time Series Classification via Tailored Agentic Reasoning
标题：通过定制的统计推理为VLM提供Few-Shot多峰时间序列分类
链接：https://arxiv.org/abs/2605.09395

【15】Probing Non-Equilibrium Grain Boundary Dynamics with XPCS and Domain-Adaptive Machine Learning
标题：利用XPCS和域自适应机器学习探索非平衡颗粒边界动力学
链接：https://arxiv.org/abs/2605.12194

【16】Adaptive Policy Learning Under Unknown Network Interference
标题：未知网络干扰下的自适应策略学习
链接：https://arxiv.org/abs/2605.11191

【17】Parameter-Efficient Adaptation of Pre-Trained Vision Foundation Models for Active and Passive Seismic Data Denoising
标题：用于主动和被动地震数据去噪的预训练视觉基础模型的参数高效调整
链接：https://arxiv.org/abs/2605.10953

强化学习(10篇)

【1】Events as Triggers for Behavioral Diversity in Multi-Agent Reinforcement Learning
标题：事件触发多智能体强化学习中行为多样性
链接：https://arxiv.org/abs/2605.12388

【2】Discrete Flow Matching for Offline-to-Online Reinforcement Learning
标题：离线到在线强化学习的离散流匹配
链接：https://arxiv.org/abs/2605.12379

【3】Delay-Empowered Causal Hierarchical Reinforcement Learning
标题：延迟赋权因果分层强化学习
链接：https://arxiv.org/abs/2605.12261

【4】Intrinsic Vicarious Conditioning for Deep Reinforcement Learning
标题：深度强化学习的内在替代条件反射
链接：https://arxiv.org/abs/2605.12224

【5】On the Importance of Multistability for Horizon Generalization in Reinforcement Learning
标题：论多稳定性对强化学习中水平泛化的重要性
链接：https://arxiv.org/abs/2605.12206

【6】Stochastic Minimum-Cost Reach-Avoid Reinforcement Learning
标题：随机最小成本可达避免强化学习
链接：https://arxiv.org/abs/2605.11975

【7】TOPPO: Rethinking PPO for Multi-Task Reinforcement Learning with Critic Balancing
标题：TOPPO：通过批判性平衡重新思考PPO用于多任务强化学习
链接：https://arxiv.org/abs/2605.11473

【8】Quotient-Categorical Representations for Bellman-Compatible Average-Reward Distributional Reinforcement Learning
标题：Bellman相容平均奖励分布强化学习的商范畴表示
链接：https://arxiv.org/abs/2605.11289

【9】Newton's Lantern: A Reinforcement Learning Framework for Finetuning AC Power Flow Warm Start Models
标题：牛顿灯笼：用于微调交流潮流热启动模型的强化学习框架
链接：https://arxiv.org/abs/2605.11102

【10】Trust Region Inverse Reinforcement Learning: Explicit Dual Ascent using Local Policy Updates
标题：信任区域反向强化学习：使用本地政策更新显式双重上升
链接：https://arxiv.org/abs/2605.11020

元学习(1篇)

【1】MetaColloc: Optimization-Free PDE Solving via Meta-Learned Basis Functions
标题：MetaColloc：通过元学习基函数进行免优化的PED求解
链接：https://arxiv.org/abs/2605.12368

符号|符号学习(1篇)

【1】A Comparative Study of Model Selection Criteria for Symbolic Regression
标题：符号回归模型选择标准的比较研究
链接：https://arxiv.org/abs/2605.11233

分层学习(1篇)

【1】Fed-BAC: Federated Bandit-Guided Additive Clustering in Hierarchical Federated Learning
标题：Fed-BAT：分层联邦学习中的联邦盗贼引导添加性集群
链接：https://arxiv.org/abs/2605.11815

医学相关(3篇)

【1】A Proof-of-Concept Simulation-Driven Digital Twin Framework for Decision-Aware Diabetes Modeling
标题：概念验证模拟驱动的数字双胞胎框架，用于决策意识糖尿病建模
链接：https://arxiv.org/abs/2605.11247

【2】Pretraining Strategies and Scaling for ECG Foundation Models: A Systematic Study
标题：心电图基础模型的预训练策略和缩放：系统性研究
链接：https://arxiv.org/abs/2605.12241

【3】NexOP: Joint Optimization of NEX-Aware k-space Sampling and Image Reconstruction for Low-Field MRI
标题：NexOP：用于低场MRI的NEX感知k空间采样和图像重建的联合优化
链接：https://arxiv.org/abs/2605.11583

蒸馏|知识提取(6篇)

【1】From Generic Correlation to Input-Specific Credit in On-Policy Self Distillation
标题：从一般相关性到政策上自我蒸馏中的特定投入信用
链接：https://arxiv.org/abs/2605.11613

【2】Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information
标题：基于点互信息的推理RL的反自蒸馏
链接：https://arxiv.org/abs/2605.11609

【3】Hindsight Hint Distillation: Scaffolded Reasoning for SWE Agents from CoT-free Answers
标题：事后诸葛亮蒸馏：SWE特工从无CoT答案中进行支架推理
链接：https://arxiv.org/abs/2605.11556

【4】MaskTab: Scalable Masked Tabular Pretraining with Scaling Laws and Distillation for Industrial Classification
标题：MaskTab：具有标度定律和蒸馏的可扩展掩蔽表格预训练，用于工业分类
链接：https://arxiv.org/abs/2605.11408

【5】Minimax Rates and Spectral Distillation for Tree Ensembles
标题：树木群落的极小极大速率和光谱蒸馏
链接：https://arxiv.org/abs/2605.11841

【6】Beyond Manual Curation: Augmenting Targeted Protein Degradation Databases via Agentic Literature Extraction Workflows
标题：超越手动腌制：通过统计文献提取工作流程增强目标蛋白质降解数据库
链接：https://arxiv.org/abs/2605.11221

自动驾驶|车辆|车道检测等(3篇)

【1】More Than Meets the Eye: A Semantics-Aware Traffic Augmentation Framework for Generalizable Website Fingerprinting
标题：不仅仅是满足眼睛：用于通用网站指纹识别的语义感知流量增强框架
链接：https://arxiv.org/abs/2605.11402

【2】Physics-Informed Teacher-Student Ensemble Learning for Traffic State Estimation with a Varying Speed Limit Scenario
标题：了解身体状况的师生在不同限速场景下进行交通状态估计学习
链接：https://arxiv.org/abs/2605.11346

【3】MambaNetBurst: Direct Byte-level Network Traffic Classification without Tokenization or Pretraining
标题：MambaNetBurst：无需令牌化或预训练的直接字节级网络流量分类
链接：https://arxiv.org/abs/2605.11034

点云|SLAM|雷达|激光|深度RGBD相关(1篇)

【1】TriBand-BEV: Real-Time LiDAR-Only 3D Pedestrian Detection via Height-Aware BEV and High-Resolution Feature Fusion
标题：TriBand-BEV：通过高度感知BEV和高分辨率特征融合进行实时纯LiDART 3D行人检测
链接：https://arxiv.org/abs/2605.12220

联邦学习|隐私保护|加密(3篇)

【1】COSMOS: Model-Agnostic Personalized Federated Learning with Clustered Server Models and Pseudo-Label-Only Communication
标题：COSMOS：与模型无关的个性化联邦学习，具有可扩展的服务器模型和仅伪标签通信
链接：https://arxiv.org/abs/2605.11165

【2】FedSurrogate: Backdoor Defense in Federated Learning via Layer Criticality and Surrogate Replacement
标题：FedSurrogate：通过层关键性和代理替换在联邦学习中进行后门防御
链接：https://arxiv.org/abs/2605.11122

【3】A Comparative Study of Federated Learning Aggregation Strategies under Homogeneous and Heterogeneous Data Distributions
标题：同质和同质数据分布下联邦学习聚合策略的比较研究
链接：https://arxiv.org/abs/2605.11010

推理|分析|理解|解释(17篇)

【1】KV-Fold: One-Step KV-Cache Recurrence for Long-Context Inference
标题：KV-Fold：用于长上下文推理的一步KV-缓存回归
链接：https://arxiv.org/abs/2605.12471

【2】Why Conclusions Diverge from the Same Observations: Formalizing World-Model Non-Identifiability via an Inference
标题：为什么结论与相同的观察结果不同：通过推理将世界模式的不可识别性形式化
链接：https://arxiv.org/abs/2605.12255

【3】Understanding Sample Efficiency in Predictive Coding
标题：了解预测编码中的样本效率
链接：https://arxiv.org/abs/2605.11911

【4】FIS-DiT: Breaking the Few-Step Video Inference Barrier via Training-Free Frame Interleaved Sparsity
标题：FIS-DiT：通过免训练帧交织稀疏性打破几步视频推理障碍
链接：https://arxiv.org/abs/2605.11869

【5】More Edits, More Stable: Understanding the Lifelong Normalization in Sequential Model Editing
标题：编辑更多，更稳定：理解顺序模型编辑中的终身规范化
链接：https://arxiv.org/abs/2605.11836

【6】Evolutionary Task Discovery: Advancing Reasoning Frontiers via Skill Composition and Complexity Scaling
标题：进化任务发现：通过技能组合和复杂性缩放推进推理前沿
链接：https://arxiv.org/abs/2605.11666

【7】A Controlled Counterexample to Strong Proxy-Based Explanations of OOD Performance: in a Fixed Pretraining-and-Probing Setup
标题：基于强代理的OOD性能描述的受控反例：在固定的预训练和探索设置中
链接：https://arxiv.org/abs/2605.11554

【8】Fast MoE Inference via Predictive Prefetching and Expert Replication
标题：通过预测性预取和专家复制进行快速MoE推理
链接：https://arxiv.org/abs/2605.11537

【9】FERMI: Exploiting Relations for Membership Inference Against Tabular Diffusion Models
标题：FEMI：利用表格扩散模型的成员推理关系
链接：https://arxiv.org/abs/2605.11527

【10】Understanding and Preventing Entropy Collapse in RLVR with On-Policy Entropy Flow Optimization
标题：通过政策上的熵流优化来理解和预防WLVR中的熵崩溃
链接：https://arxiv.org/abs/2605.11491

【11】Drop the Act: Probe-Filtered RL for Faithful Chain-of-Thought Reasoning
标题：放弃法案：忠诚思想链推理的探针过滤RL
链接：https://arxiv.org/abs/2605.11467

【12】Causal Fairness for Survival Analysis
标题：生存分析的因果公平性
链接：https://arxiv.org/abs/2605.11362

【13】Beyond Similarity: Temporal Operator Attention for Time Series Analysis
标题：超越相似性：时间操作员对时间序列分析的关注
链接：https://arxiv.org/abs/2605.11287

【14】When to Ask a Question: Understanding Communication Strategies in Generative AI Tools
标题：何时提问：了解生成性人工智能工具中的沟通策略
链接：https://arxiv.org/abs/2605.11240

【15】FeatMap: Understanding image manipulation in the feature space and its implications for feature space geometry
标题：DeliverMap：了解要素空间中的图像操作及其对要素空间几何的影响
链接：https://arxiv.org/abs/2605.11203

【16】SkillGen: Verified Inference-Time Agent Skill Synthesis
标题：SkillGen：经过验证的推理时间代理技能合成
链接：https://arxiv.org/abs/2605.10999

【17】Post-ADC Inference: Valid Inference After Active Data Collection
标题：ADC后推断：活动数据收集后的有效推断
链接：https://arxiv.org/abs/2605.11511

检测相关(5篇)

【1】Detecting overfitting in Neural Networks during long-horizon grokking using Random Matrix Theory
标题：使用随机矩阵理论在长时间观察期间检测神经网络中的过拟
链接：https://arxiv.org/abs/2605.12394

【2】VNDUQE: Information-Theoretic Novelty Detection using Deep Variational Information Bottleneck
标题：VNDUQE：使用深度变分信息瓶颈的信息论新奇检测
链接：https://arxiv.org/abs/2605.11551

【3】Causal Bias Detection in Generative Artifical Intelligence
标题：生成智力中的因果偏差检测
链接：https://arxiv.org/abs/2605.11365

【4】LEAP: Unlocking dLLM Parallelism via Lookahead Early-Convergence Token Detection
标题：LEAP：通过Lookahead早期收敛令牌检测解锁dLLM并行主义
链接：https://arxiv.org/abs/2605.10980

【5】Trajectory-Agnostic Asteroid Detection in TESS with Deep Learning
标题：利用深度学习在TESS中进行轨迹不可知小行星检测
链接：https://arxiv.org/abs/2605.12391

分类|识别(9篇)

【1】Fair Conformal Classification via Learning Representation-Based Groups
标题：通过学习基于表示的组实现公平的共形分类
链接：https://arxiv.org/abs/2605.12195

【2】EqOD: Symmetry-Informed Stability Selection for PDE Identification
标题：EqOD：用于PCE识别的对称性知情稳定性选择
链接：https://arxiv.org/abs/2605.11524

【3】Diabetic Retinopathy Classification using Downscaling Algorithms and Deep Learning
标题：使用缩减算法和深度学习进行糖尿病视网膜病变分类
链接：https://arxiv.org/abs/2605.11430

【4】A Boundary-Aware Non-parametric Granular-Ball Classifier Based on Minimum Description Length
标题：基于最小描述长度的边界感知非参数颗粒球分类器
链接：https://arxiv.org/abs/2605.11406

【5】$\varepsilon$-Good Action Identification in Fixed-Budget Monte Carlo Tree Search
链接：https://arxiv.org/abs/2605.11324

【6】Multi-Agent System Identification with Nonlinear Sheaf Diffusion
标题：具有非线性铺层扩散的多智能体系统识别
链接：https://arxiv.org/abs/2605.11204

【7】Read, Extract, Classify: A Tool for Smarter Requirements Engineering
标题：阅读、提取、分类：更智能的需求工程工具
链接：https://arxiv.org/abs/2605.11045

【8】SCOPE: Siamese Contrastive Operon Pair Embeddings for Functional Sequence Representation and Classification
标题：范围：用于功能序列表示和分类的连体对比操纵子对嵌入
链接：https://arxiv.org/abs/2605.11022

【9】Testing General Relativity Through Gravitational Wave Classification: A Convolutional Neural Network Framework
标题：通过引力波分类测试广义相对论：卷积神经网络框架
链接：https://arxiv.org/abs/2605.02453

表征(9篇)

【1】Multi-Task Representation Learning for Conservative Linear Bandits
标题：保守线性盗贼的多任务表示学习
链接：https://arxiv.org/abs/2605.12176

【2】Disentangled Sparse Representations for Concept-Separated Diffusion Unlearning
标题：概念分离扩散去学习的解纠缠稀疏表示
链接：https://arxiv.org/abs/2605.12122

【3】WorldComp2D: Spatio-semantic Representations of Object Identity and Location from Local Views
标题：WorldComp2D：来自本地视图的对象身份和位置的空间语义表示
链接：https://arxiv.org/abs/2605.11743

【4】Debiased Model-based Representations for Sample-efficient Continuous Control
标题：基于去偏模型的表示实现样本高效连续控制
链接：https://arxiv.org/abs/2605.11711

【5】A Composite Activation Function for Learning Stable Binary Representations
标题：学习稳定二进制表示的复合激活函数
链接：https://arxiv.org/abs/2605.11558

【6】TRACE: Temporal Routing with Autoregressive Cross-channel Experts for EEG Representation Learning
标题：TRACE：用于脑电表示学习的自回归跨通道专家的时间路由
链接：https://arxiv.org/abs/2605.11380

【7】Optimal Representations for Generalized Contrastive Learning with Imbalanced Datasets
标题：不平衡数据集下广义对比学习的最佳表示
链接：https://arxiv.org/abs/2605.11291

【8】Oversmoothing as Representation Degeneracy in Neural Sheaf Diffusion
标题：过度平滑作为神经片扩散中的表示简并
链接：https://arxiv.org/abs/2605.11178

【9】The Price of Proportional Representation in Temporal Voting
标题：临时投票中比例代表的代价
链接：https://arxiv.org/abs/2605.11157

3D|3D重建等相关(1篇)

【1】PG-3DGS: Optimizing 3D Gaussian Splatting to Satisfy Physics Objectives
标题：PG-3DGS：优化3D高斯飞溅以满足物理目标
链接：https://arxiv.org/abs/2605.11266

优化|敛散性(16篇)

【1】Aligning Flow Map Policies with Optimal Q-Guidance
标题：将流程图策略与最佳Q指导保持一致
链接：https://arxiv.org/abs/2605.12416

【2】SOAR: Scale Optimization for Accurate Reconstruction in NVFP4 Quantization
标题：SOAR：NVFP 4量化中精确重建的规模优化
链接：https://arxiv.org/abs/2605.12245

【3】Expected Batch Optimal Transport Plans and Consequences for Flow Matching
标题：预期批量最佳运输计划和流量匹配的后果
链接：https://arxiv.org/abs/2605.12174

【4】Elicitation-Augmented Bayesian Optimization
标题：启发式增强Bayesian优化
链接 ：https://arxiv.org/abs/2605.12079

【5】Augmented Lagrangian Method for Last-Iterate Convergence for Constrained MDPs
标题：约束MDPs最后迭代收敛的增广拉格朗日方法
链接：https://arxiv.org/abs/2605.11694

【6】Slicing and Dicing: Configuring Optimal Mixtures of Experts
标题：切片和切片：配置最佳专家组合
链接：https://arxiv.org/abs/2605.11689

【7】MuonQ: Enhancing Low-Bit Muon Quantization via Directional Fidelity Optimization
标题：MuonQ：通过方向保真度优化增强低位μ子量化
链接：https://arxiv.org/abs/2605.11396

【8】Gradient-Free Noise Optimization for Reward Alignment in Generative Models
标题：生成模型中奖励一致的无干扰噪音优化
链接：https://arxiv.org/abs/2605.11347

【9】Support-Proximity Augmented Diffusion Estimation for Offline Black-Box Optimization
标题：离线黑匣子优化的支持邻近增广扩散估计
链接：https://arxiv.org/abs/2605.11246

【10】Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training
标题：偏好优化中的伪相关学习：机制、后果和通过领带训练的缓解
链接：https://arxiv.org/abs/2605.11134

【11】TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment
标题：TMPO：轨迹匹配政策优化，以实现多元化和高效的扩散对齐
链接：https://arxiv.org/abs/2605.10983

【12】$ξ$-DPO: Direct Preference Optimization via Ratio Reward Margin
标题：$ð $-DPO：通过比例回报保证金直接偏好优化
链接：https://arxiv.org/abs/2605.10981

【13】QuIDE: Mastering the Quantized Intelligence Trade-off via Active Optimization
标题：QuIDE：通过主动优化掌握量化情报权衡
链接：https://arxiv.org/abs/2605.10959

【14】Optimal Policy Learning under Budget and Coverage Constraints
标题：预算和覆盖范围约束下的最佳政策学习
链接：https://arxiv.org/abs/2605.12235

【15】Sharp feature-learning transitions and Bayes-optimal neural scaling laws in extensive-width networks
标题：宽网络中的尖锐特征学习转变和Bayes最优神经缩放定律
链接：https://arxiv.org/abs/2605.10395

【16】Breaking QAOA's Fixed Target Hamiltonian Barrier: A Fully Connected Quantum Boltzmann Machine via Bilevel Optimization
标题：打破QAOA的固定目标Hamilton障碍：通过二层优化的全连接量子Boltzmann机
链接：https://arxiv.org/abs/2605.07473

预测|估计(9篇)

【1】Predicting Decisions of AI Agents from Limited Interaction through Text-Tabular Modeling
标题：基于文本表格模型的人工智能主体决策预测
链接：https://arxiv.org/abs/2605.12411

【2】ECTO: Exogenous-Conditioned Temporal Operator for Ultra-Short-Term Wind Power Forecasting
标题：ECTO：超短期风电预测的外生条件时态操作员
链接：https://arxiv.org/abs/2605.12196

【3】Decomposing the Generalization Gap in PROTAC Activity Prediction: Variance Attribution and the Inter-Laboratory Ceiling
标题：分解PROTAC活动预测中的一般化差距：方差归因和实验室间上限
链接：https://arxiv.org/abs/2605.11764

【4】Partial Model Sharing Improves Byzantine Resilience in Federated Conformal Prediction
标题：部分模型共享提高联邦保形预测中的拜占庭弹性
链接：https://arxiv.org/abs/2605.11684

【5】EpiCastBench: Datasets and Benchmarks for Multivariate Epidemic Forecasting
标题：EpiCastBench：多元流行病预测的数据集和基准
链接：https://arxiv.org/abs/2605.11598

【6】Three Regimes of Context-Parametric Conflict: A Predictive Framework and Empirical Validation
标题：三种情境-参数冲突机制：一个预测框架与实证检验
链接：https://arxiv.org/abs/2605.11574

【7】Dual-Temporal LSTM with Hybrid Attention for Airline Passenger Load Factor Forecasting: Integrating Intra-Flight and Inter-Flight Booking Dynamics
标题：航空公司乘客负荷率预测混合关注的双时态LSTM：集成机上和机上预订动态
链接：https://arxiv.org/abs/2605.11569

【8】Deep Learning for Protein Complex Prediction and Design
标题：用于蛋白质复合物预测和设计的深度学习
链接：https://arxiv.org/abs/2605.11189

【9】Multi-Variable Conformal Prediction: Optimizing Prediction Sets without Data Splitting
标题：多变量保形预测：在不进行数据拆分的情况下优化预测集
链接：https://arxiv.org/abs/2605.12341

其他神经网络|深度学习|模型|建模(56篇)

【1】Routers Learn the Geometry of Their Experts: Geometric Coupling in Sparse Mixture-of-Experts
标题：路由器学习其专家的几何：稀疏混合专家中的几何耦合
链接：https://arxiv.org/abs/2605.12476

【2】Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space
标题：空间故事：概念信念空间中的情境学习轨迹
链接：https://arxiv.org/abs/2605.12412

【3】EHR-RAGp: Retrieval-Augmented Prototype-Guided Foundation Model for Electronic Health Records
标题：EHR-RAGp：电子健康记录检索增强原型引导基础模型
链接：https://arxiv.org/abs/2605.12335

【4】Autoregressive Learning in Joint KL: Sharp Oracle Bounds and Lower Bounds
标题：联合KL中的自回归学习：尖锐的Oracle界和下界
链接：https://arxiv.org/abs/2605.12316

【5】In-context learning to predict critical transitions in dynamical systems
标题：上下文学习预测动态系统中的关键转变
链接：https://arxiv.org/abs/2605.12308

【6】KAN-CL: Per-Knot Importance Regularization for Continual Learning with Kolmogorov-Arnold Networks
标题：KAN-CL：使用Kolmogorov-Arnold网络进行连续学习的按结重要性正规化
链接：https://arxiv.org/abs/2605.12306

【7】STRABLE: Benchmarking Tabular Machine Learning with Strings
标题：STRAPLE：使用字符串对表格机器学习进行基准测试
链接：https://arxiv.org/abs/2605.12292

【8】What makes a word hard to learn? Modeling L1 influence on English vocabulary difficulty
标题：是什么让一个词很难学？建模一语对英语词汇难度的影响
链接：https://arxiv.org/abs/2605.12281

【9】DriftXpress: Faster Drifting Models via Projected RKHS Fields
标题：DriftXpress：通过投影RKHS场实现更快的漂移模型
链接：https://arxiv.org/abs/2605.12183

【10】Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics
标题：企业系统需要习得的世界模型吗？上下文对推断动态的重要性
链接：https://arxiv.org/abs/2605.12178

【11】On What We Can Learn from Low-Resolution Data
标题：我们可以从低分辨率数据中学到什么
链接：https://arxiv.org/abs/2605.12168

【12】Learning plug-in surrogate endpoints for randomized experiments
标题：学习插件替代随机实验的终点
链接：https://arxiv.org/abs/2605.12051

【13】Scaling Laws and Tradeoffs in Recurrent Networks of Expressive Neurons
标题：表达神经元回归网络中的缩放定律和权衡
链接：https://arxiv.org/abs/2605.12049

【14】Resilient Vision-Tabular Multimodal Learning under Modality Missingness
标题：情态缺失下的弹性视觉表式多模式学习
链接：https://arxiv.org/abs/2605.12031

【15】Limits of Learning Linear Dynamics from Experiments
标题：从实验中学习线性动力学的局限性
链接：https://arxiv.org/abs/2605.12010

【16】Procedural-skill SFT across capacity tiers: A W-Shaped pre-SFT Trajectory and Regime-Asymmetric Mechanism on 0.8B-4B Qwen3.5 Models
标题：跨容量层的过程技能SFT：0.8B-4 B Qwen 3.5模型上的W形前SFT轨迹和状态不对称机制
链接：https://arxiv.org/abs/2605.11907

【17】Incentivizing Truthfulness and Collaborative Fairness in Bayesian Learning
标题：激励Bayesian学习中的真实性和协作公平性
链接：https://arxiv.org/abs/2605.11889

【18】Improving the Performance and Learning Stability of Parallelizable RNNs Designed for Ultra-Low Power Applications
标题：提高专为超低功耗应用设计的可并行化RNN的性能和学习稳定性
链接：https://arxiv.org/abs/2605.11855

【19】Multi-Timescale Conductance Spiking Networks: A Sparse, Gradient-Trainable Framework with Rich Firing Dynamics for Enhanced Temporal Processing
标题：多时间尺度电导尖峰网络：一个具有丰富激发动力学的稀疏、可训练的框架，用于增强的时间处理
链接：https://arxiv.org/abs/2605.11835

【20】Stop Marginalizing My Dreams: Model Inversion via Laplace Kernel for Continual Learning
标题：停止边缘化我的梦想：通过拉普拉斯核进行持续学习的模型倒置
链接：https://arxiv.org/abs/2605.11804

【21】Is Monotonic Sampling Necessary in Diffusion Models?
标题：扩散模型中需要单调抽样吗？
链接：https://arxiv.org/abs/2605.11773

【22】A nonlinear extension of parametric model embedding for dimensionality reduction in parametric shape design
标题：参数化形状设计降维参数化模型嵌入的非线性扩展
链接：https://arxiv.org/abs/2605.11759

【23】One-Step Generative Modeling via Wasserstein Gradient Flows
标题：通过Wasserstein梯度流的一步生成建模
链接：https://arxiv.org/abs/2605.11755

【24】Online Continual Learning with Dynamic Label Hierarchies
标题：具有动态标签分层结构的在线持续学习
链接：https://arxiv.org/abs/2605.11742

【25】Finite Sentence-Interface Control for Learning Bounded-Fan-Out Linear MCFGs under Fixed Monoid Typing
标题：固定单元类型下学习有界扇形线性MCFG的有限句子接口控制
链接：https://arxiv.org/abs/2605.11644

【26】MIST: Reliable Streaming Decision Trees for Online Class-Incremental Learning via McDiarmid Bound
标题：MIST：通过McDiarmid Bound实现在线课堂增量学习的可靠流媒体决策树
链接：https://arxiv.org/abs/2605.11617

【27】SoK: Unlearnability and Unlearning for Model Dememorization
标题：SoK：不可学习和不可学习以实现模型去化
链接：https://arxiv.org/abs/2605.11592

【28】Learning Weakly Communicating Average-Reward CMDPs: Strong Duality and Improved Regret
标题：学习弱沟通平均回报CMDPs：强二元性和改善遗憾
链接：https://arxiv.org/abs/2605.11586

【29】A Mixture Autoregressive Image Generative Model on Quadtree Regions for Gaussian Noise Removal via Variational Bayes and Gradient Methods
标题：基于变分Bayes和梯度方法去除高斯噪音的四叉树区域混合自回归图像生成模型
链接：https://arxiv.org/abs/2605.11585

【30】OUI as a Structural Observable: Towards an Activation-Centric View of Neural Network Training
标题：OUI作为结构可观察对象：走向以激活为中心的神经网络训练观点
链接：https://arxiv.org/abs/2605.11570

【31】Multi-Narrow Transformation as a Single-Model Ensemble: Boundary Conditions, Mechanisms, and Failure Modes
标题：作为单模型集合的多窄变换：边界条件、机制和故障模式
链接：https://arxiv.org/abs/2605.11530

【32】OverNaN: NaN-Aware Oversampling for Imbalanced Learning with Meaningful Missingness
标题：OverNaN：NaN意识过度采样以实现有意义的缺失的不平衡学习
链接：https://arxiv.org/abs/2605.11525

【33】The Evaluation Differential: When Frontier AI Models Recognise They Are Being Tested
标题：评估差异：当前沿人工智能模型认识到他们正在接受测试时
链接：https://arxiv.org/abs/2605.11496

【34】Efficient Adjoint Matching for Fine-tuning Diffusion Models
标题：微调扩散模型的有效伴随匹配
链接：https://arxiv.org/abs/2605.11480

【35】Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models
标题：克服动态盲目：VLA模型的免训练步程修正
链接：https://arxiv.org/abs/2605.11459

【36】Instruct-ICL: Instruction-Guided In-Context Learning for Post-Disaster Damage Assessment
标题：Direct-ICL：灾后损害评估的指导指导下的情境学习
链接：https://arxiv.org/abs/2605.11439

【37】Couple to Control: Joint Initial Noise Design in Diffusion Models
标题：控制耦合：扩散模型中的联合初始噪音设计
链接：https://arxiv.org/abs/2605.11311

【38】Localization Boosting for Growth Markets: Mitigating Cross-Locale Behavioral Bias in Learning-to-Rank
标题：本地化推动成长型市场：缓解学习排名中的跨地区行为偏见
链接：https://arxiv.org/abs/2605.11272

【39】Language Modeling with Hyperspherical Flows
标题：超球形流语言建模
链接：https://arxiv.org/abs/2605.11125

【40】ShardTensor: Domain Parallelism for Scientific Machine Learning
标题：ShardTensor：科学机器学习的领域并行主义
链接：https://arxiv.org/abs/2605.11111

【41】ASD-Bench: A Four-Axis Comprehensive Benchmark of AI Models for Autism Spectrum Disorder
标题：ASD Bench：自闭症谱系障碍人工智能模型的四轴综合基准
链接：https://arxiv.org/abs/2605.11091

【42】A Switching System Theory of Q-Learning with Linear Function Approximation
标题：线性函数逼近的Q学习切换系统理论
链接：https://arxiv.org/abs/2605.11021

【43】Simpson's Paradox in Behavioral Curves: How Aggregation Distorts Parametric Models of User Dynamics
标题：行为曲线中的辛普森悖论：聚集如何扭曲用户动态的参数模型
链接：https://arxiv.org/abs/2605.11017

【44】Finite Volume-Informed Neural Network Framework for 2D Shallow Water Equations: Rugged Loss Landscapes and the Importance of Data Guidance
标题：二维浅水方程的有限体积神经网络框架：坚固损失景观和数据指导的重要性
链接：https://arxiv.org/abs/2605.11001

【45】Stargazer: A Scalable Model-Fitting Benchmark Environment for AI Agents under Astrophysical Constraints
标题：Stargazer：天体物理约束下人工智能代理的可扩展模型匹配基准环境
链接：https://arxiv.org/abs/2604.15664

【46】Model-based Bootstrap of Controlled Markov Chains
标题：受控马尔可夫链的基于模型的Bootstrap
链接：https://arxiv.org/abs/2605.12410

【47】Online Learning-to-Defer with Varying Experts
标题：不同专家的在线延迟学习
链接：https://arxiv.org/abs/2605.12340

【48】Machine Learning for neutron source distributions
标题：用于中子源分布的机器学习
链接：https://arxiv.org/abs/2605.12165

【49】Variance-aware Reward Modeling with Anchor Guidance
标题：具有锚点指导的方差感知奖励建模
链接：https://arxiv.org/abs/2605.11865

【50】Posterior Contraction Rates for Sparse Kolmogorov-Arnold Networks in Anisotropic Besov Spaces
标题：各向异性Besov空间中稀疏Kolmogorov-Arnold网络的后验压缩率
链接：https://arxiv.org/abs/2605.11652

【51】Efficient and provably convergent end-to-end training of deep neural networks with linear constraints
标题：具有线性约束的深度神经网络的高效且可证明收敛的端到端训练
链接：https://arxiv.org/abs/2605.11526

【52】Rethinking external validation for the target population: Capturing patient-level similarity with a generative model
标题：重新思考目标人群的外部验证：用生成模型捕获患者水平的相似性
链接：https://arxiv.org/abs/2605.11284

【53】Interpretable Machine Learning for Spatial Science: A Lie-Algebraic Kernel for Rotationally Anisotropic Gaussian Processes
标题：空间科学的可解释机器学习：旋转各向异性高斯过程的李代数核
链接：https://arxiv.org/abs/2605.11179

【54】From raw data to neutrino candidates: a neural-network pipeline for Baikal-GVD
标题：从原始数据到候选中微子：贝加尔-GVD的神经网络管道
链接：https://arxiv.org/abs/2605.11176

【55】Acceleration of horizontal numerical advection for atmospheric modeling through surrogate modeling with temporal coarse-graining
标题：通过具有时间粗粒度的代理建模加速大气建模的水平数值平流
链接：https://arxiv.org/abs/2605.10956

【56】Measuring Accuracy and Energy-to-Solution of Quantum Fine-Tuning of Foundational AI Models
标题：基础人工智能模型量子微调的测量准确性和能量到解决方案
链接：https://arxiv.org/abs/2605.02798

其他(83篇)

【1】Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation
标题：Pion：一种通过垂直等效变换的光谱保持优化器
链接：https://arxiv.org/abs/2605.12492

【2】MEME: Multi-entity & Evolving Memory Evaluation
标题：MEME：多实体和不断进化的记忆评估
链接：https://arxiv.org/abs/2605.12477

【3】High-arity Sample Compression
标题：高浓度样本压缩
链接：https://arxiv.org/abs/2605.12465

【4】Search Your Block Floating Point Scales!
标题：搜索您的块浮点规模！
链接：https://arxiv.org/abs/2605.12464

【5】Towards Affordable Energy: A Gymnasium Environment for Electric Utility Demand-Response Programs
标题：迈向负担得起的能源：电力公司需求响应计划的健身房环境
链接：https://arxiv.org/abs/2605.12462

【6】ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging
标题：ORBIT：通过起源监管合并保留GenRetrieval中的基础语言能力
链接：https://arxiv.org/abs/2605.12419

【7】Agent-Based Post-Hoc Correction of Agricultural Yield Forecasts
标题：基于代理的农业产量预测事后修正
链接：https://arxiv.org/abs/2605.12375

【8】A New Technique for AI Explainability using Feature Association Map
标题：一种基于特征关联图的人工智能可解释性新技术
链接：https://arxiv.org/abs/2605.12350

【9】Neural-Schwarz Tiling for Geometry-Universal PDE Solving at Scale
标题：神经-施瓦茨切片用于大规模求解几何通用偏出方程
链接：https://arxiv.org/abs/2605.12343

【10】BSO: Safety Alignment Is Density Ratio Matching
标题：BSO：安全一致就是密度比匹配
链接：https://arxiv.org/abs/2605.12339

【11】Manifold Sampling via Entropy Maximization
标题：通过熵最大化进行总管采样
链接：https://arxiv.org/abs/2605.12338

【12】Targeted Neuron Modulation via Contrastive Pair Search
标题：通过对比对搜索进行有针对性的神经元调制
链接：https://arxiv.org/abs/2605.12290

【13】Hypernetworks for Dynamic Feature Selection
标题：用于动态特征选择的超网络
链接：https://arxiv.org/abs/2605.12278

【14】Missingness-MDPs: Bridging the Theory of Missing Data and POMDPs
标题：缺失-MDPs：缺失数据理论与POMDPs的桥梁
链接：https://arxiv.org/abs/2605.12262

【15】TMRL: Diffusion Timestep-Modulated Pretraining Enables Exploration for Efficient Policy Finetuning
标题：TMRL：扩散时步调制预训练能够探索有效的政策微调
链接：https://arxiv.org/abs/2605.12236

【16】No More, No Less: Task Alignment in Terminal Agents
标题：不多不少：终端代理中的任务协调
链接：https://arxiv.org/abs/2605.12233

【17】Investigating simple target-covariate relationships for Chronos-2 and TabPFN-TS
标题：调查Chronos-2和TabPFN-TS的简单目标-协变量关系
链接：https://arxiv.org/abs/2605.12200

【18】Overtrained, Not Misaligned
标题：过度训练，而不是错位
链接：https://arxiv.org/abs/2605.12199

【19】Fused Gromov-Wasserstein Distance with Feature Selection
标题：融合Gromov-Wasserstein距离与特征选择
链接：https://arxiv.org/abs/2605.12161

【20】Missing Old Logits in Asynchronous Agentic RL: Semantic Mismatch and Repair Methods for Off-Policy Correction
标题：非策略性RL中丢失的旧Logits：用于非策略纠正的语义不匹配和修复方法
链接：https://arxiv.org/abs/2605.12070

【21】Hölder Policy Optimisation
标题：赫德政策优化
链接：https://arxiv.org/abs/2605.12058

【22】Approximation Theory of Laplacian-Based Neural Operators for Reaction-Diffusion System
标题：反应扩散系统拉普拉斯神经运算的逼近理论
链接：https://arxiv.org/abs/2605.12025

【23】Estimating Subgraph Importance with Structural Prior Domain Knowledge
标题：利用结构性先验领域知识估计子图重要性
链接：https://arxiv.org/abs/2605.12009

【24】Split the Differences, Pool the Rest: Provably Efficient Multi-Objective Imitation
标题：消除差异，汇集其余：明显有效的多目标模仿
链接：https://arxiv.org/abs/2605.12000

【25】QDSB: Quantized Diffusion Schrödinger Bridges
标题：QDSB：量化扩散薛定汉桥
链接：https://arxiv.org/abs/2605.11983

【26】NOFE -- Neural Operator Function Embedding
标题：NOFE --神经运算符函数嵌入
链接：https://arxiv.org/abs/2605.11970

【27】Delightful Gradients Accelerate Corner Escape
标题：令人愉悦的乘客加速角落逃生
链接：https://arxiv.org/abs/2605.11908

【28】Sobolev Regularized MMD Gradient Flow
标题：Sobolev正规MMD梯度流
链接：https://arxiv.org/abs/2605.11884

【29】LOFT: Low-Rank Orthogonal Fine-Tuning via Task-Aware Support Selection
标题：LOFT：通过任务感知支持选择进行低级别垂直微调
链接：https://arxiv.org/abs/2605.11872

【30】A Fast and Energy-Efficient Latch-Based Memristive Analog Content-Addressable Memory
标题：一种快速、节能的基于锁存的记忆模拟内容可编址存储器
链接：https://arxiv.org/abs/2605.11847

【31】Gradient Clipping Beyond Vector Norms: A Spectral Approach for Matrix-Valued Parameters
标题：超越载体规范的梯度剪裁：矩阵值参数的谱方法
链接：https://arxiv.org/abs/2605.11838

【32】Entropy Polarity in Reinforcement Fine-Tuning: Direction, Asymmetry, and Control
标题：强化微调中的熵极：方向、不对称和控制
链接：https://arxiv.org/abs/2605.11775

【33】Federated Client Selection under Partial Visibility: A POMDP Approach with Spatio-Temporal Attention
标题：部分可见性下的联邦客户选择：具有时空注意力的POMDP方法
链接：https://arxiv.org/abs/2605.11752

【34】Compositional Neural Operators for Multi-Dimensional Fluid Dynamics
标题：多维流体动力学的合成神经运算符
链接：https://arxiv.org/abs/2605.11691

【35】GeomHerd: A Forward-looking Herding Quantification via Ricci Flow Geometry on Agent Interactive Simulations
标题：GeomHerd：在Agent交互模拟上通过Ricci流几何进行前瞻性羊群量化
链接：https://arxiv.org/abs/2605.11645

【36】PRISM: A Geometric Risk Bound that Decomposes Drift into Scale, Shape, and Head
标题：棱镜：将漂移分解为规模、形状和头部的几何风险界限
链接：https://arxiv.org/abs/2605.11608

【37】FedOUI: OUI-Guided Client Weighting for Federated Aggregation
标题：FedOUI：OUI引导的联合聚合客户端加权
链接：https://arxiv.org/abs/2605.11571

【38】Sharpen Your Flow: Sharpness-Aware Sampling for Flow Matching
标题：加快您的流量：用于流量匹配的敏锐度感知采样
链接：https://arxiv.org/abs/2605.11547

【39】Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting
标题：驯服极端代币：具有协方差意识的GRPO，具有高斯核优势重新加权
链接：https://arxiv.org/abs/2605.11538

【40】Controllable User Simulation
标题：可控用户模拟
链接：https://arxiv.org/abs/2605.11519

【41】On the Approximation Complexity of Matrix Product Operator Born Machines
标题：关于矩阵积运算符Born机的逼近复杂性
链接：https://arxiv.org/abs/2605.11471

【42】Deep Minds and Shallow Probes
标题：深层思想和浅层探测
链接：https://arxiv.org/abs/2605.11448

【43】FastUMAP: Scalable Dimensionality Reduction via Bipartite Landmark Sampling
标题：FastUMAP：通过双方地标采样可扩展的模糊性降低
链接：https://arxiv.org/abs/2605.11428

【44】Behavioral Mode Discovery for Fine-tuning Multimodal Generative Policies
标题：微调多模式生成策略的行为模式发现
链接：https://arxiv.org/abs/2605.11387

【45】Causal Algorithmic Recourse: Foundations and Methods
标题：因果关系追索权：基础与方法
链接：https://arxiv.org/abs/2605.11373

【46】The tractability landscape of diffusion alignment: regularization, rewards, and computational primitives
标题：扩散对齐的可处理性景观：规则化、奖励和计算基元
链接：https://arxiv.org/abs/2605.11361

【47】gym-invmgmt: An Open Benchmarking Framework for Inventory Management Methods
标题：gym-invmgMT：库存管理方法的开放基准框架
链接：https://arxiv.org/abs/2605.11355

【48】MLCommons Chakra: Advancing Performance Benchmarking and Co-design using Standardized Execution Traces
标题：MLCommons脉轮：使用标准化执行轨迹推进绩效基准和联合设计
链接：https://arxiv.org/abs/2605.11333

【49】Neural Statistical Functions
标题：神经统计功能
链接：https://arxiv.org/abs/2605.11327

【50】Error whitening: Why Gauss-Newton outperforms Newton
标题：错误白化：为什么高斯-牛顿优于牛顿
链接：https://arxiv.org/abs/2605.11316

【51】Extending Kernel Trick to Influence Functions
标题：扩展核心技巧以影响功能
链接：https://arxiv.org/abs/2605.11239

【52】DeconDTN-Toolkit: A Library for Evaluation and Enhancement of Robustness to Provenance Shift
标题：DeconDTN-Tools：评估和增强源转移稳健性的图书馆
链接：https://arxiv.org/abs/2605.11237

【53】LiBaGS: Lightweight Boundary Gap Synthesis for Targeted Synthetic Data Selection
标题：LiBaGS：用于有针对性的合成数据选择的轻量级边界间隙合成
链接：https://arxiv.org/abs/2605.11231

【54】The Scaling Law of Evaluation Failure: Why Simple Averaging Collapses Under Data Sparsity and Item Difficulty Gaps, and How Item Response Theory Recovers Ground Truth Across Domains
标题：评估失败的比例定律：为什么简单平均在数据稀疏和项目难度差距下崩溃，以及项目反应理论如何跨领域恢复基本真相
链接：https://arxiv.org/abs/2605.11205

【55】Exploring Token-Space Manipulation in Latent Audio Tokenizers
标题：探索潜在音频令牌器中的令牌空间操纵
链接 ：https://arxiv.org/abs/2605.11192

【56】Muon is Not That Special: Random or Inverted Spectra Work Just as Well
标题：μ子并不那么特别：随机或倒置光谱也同样有效
链接：https://arxiv.org/abs/2605.11181

【57】Optimistic Dual Averaging Unifies Modern Optimizers
标题：乐观的双重平均统一现代优化器
链接：https://arxiv.org/abs/2605.11172

【58】Unlearning with Asymmetric Sources: Improved Unlearning-Utility Trade-off with Public Data
标题：利用不对称来源消除学习：利用公共数据改进的消除学习与效用权衡
链接：https://arxiv.org/abs/2605.11170

【59】Interpretability Can Be Actionable
标题：可解释性可以采取行动
链接：https://arxiv.org/abs/2605.11161

【60】Decomposing Evolutionary Mixture-of-LoRA Architectures: The Routing Lever, the Lifecycle Penalty, and a Substrate-Conditional Boundary
标题：分解进化混合LoRA架构：路由杠杆、超时惩罚和基片条件边界
链接：https://arxiv.org/abs/2605.11153

【61】Debiasing Message Passing to Mitigate Popularity Bias in GNN-based Collaborative Filtering
标题：消除消息传递的偏见以减轻基于GNN的协同过滤中的流行偏见
链接：https://arxiv.org/abs/2605.11145

【62】Control Charts for Multi-agent Systems
标题：多智能体系统的控制图
链接：https://arxiv.org/abs/2605.11135

【63】Steerable Neural ODEs on Homogeneous Spaces
标题：齐次空间上的可操纵神经ODE
链接：https://arxiv.org/abs/2605.11133

【64】GRAFT-ATHENA: Self-Improving Agentic Teams for Autonomous Discovery and Evolutionary Numerical Algorithms
标题：GRAFT-ATHENA：自主发现和进化数值算法的自我改进的抽象团队
链接：https://arxiv.org/abs/2605.11117

【65】LatentHDR: Decoupling Exposure from Diffusion via Conditional Latent-to-Latent Mapping for Text/Image-to-Panoramic HDR
标题：潜伏HDR：通过文本/图像到全景HDR的条件潜伏到潜伏映射将暴露与扩散脱钩
链接：https://arxiv.org/abs/2605.11115

【66】The first global agricultural field boundary map at 10m resolution
标题：第一张10 m分辨率的全球农田边界图
链接：https://arxiv.org/abs/2605.11055

【67】On Problems of Implicit Context Compression for Software Engineering Agents
标题：软件工程代理的隐式上下文压缩问题
链接：https://arxiv.org/abs/2605.11051

【68】Backbone-Equated Diffusion OOD via Sparse Internal Snapshots
标题：通过稀疏内部快照的骨架等效扩散OOD
链接：https://arxiv.org/abs/2605.11014

【69】When and How to Canonize: A Generalization Perspective
标题：何时以及如何封圣：概括的角度
链接：https://arxiv.org/abs/2605.11008

【70】DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe Parallelism
标题：DisagMoE：通过分解的AF管道分类的计算-通信重叠的MoE训练
链接：https://arxiv.org/abs/2605.11005

【71】Test-Time Personalization: A Diagnostic Framework and Probabilistic Fix for Scaling Failures
标题：测试时个性化：扩展失败的诊断框架和概率修复
链接：https://arxiv.org/abs/2605.10991

【72】Interpretable EEG Microstate Discovery via Variational Deep Embedding: A Systematic Architecture Search with Multi-Quadrant Evaluation
标题：通过变分深度嵌入可解释的脑电微状态发现：具有多象限评估的系统架构搜索
链接：https://arxiv.org/abs/2605.10947

【73】A proximal gradient algorithm for composite log-concave sampling
标题：复合双凹抽样的近端梯度算法
链接：https://arxiv.org/abs/2605.12461

【74】Enabling AI-Native Mobility in 6G: A Real-World Dataset for Handover, Beam Management, and Timing Advance
标题：在6G中实现AI原生移动性：用于切换、射束管理和定时提前的真实数据集
链接：https://arxiv.org/abs/2605.12453

【75】Information-Theoretic Generalization Bounds for Sequential Decision Making
标题：顺序决策的信息论推广界
链接：https://arxiv.org/abs/2605.12190

【76】Keeping Score: Efficiency Improvements in Neural Likelihood Surrogate Training via Score-Augmented Loss Functions
标题：保持分数：通过分数增强损失函数提高神经似然代理训练的效率
链接：https://arxiv.org/abs/2605.12118

【77】Rethink the Role of Neural Decoders in Quantum Error Correction
标题：重新思考神经解码器在量子错误纠正中的作用
链接：https://arxiv.org/abs/2605.12046

【78】Constrained Stochastic Spectral Preconditioning Converges for Nonconvex Objectives
标题：非凸目标的约束随机谱预处理收敛
链接：https://arxiv.org/abs/2605.11850

【79】Generative climate downscaling enables high-resolution compound risk assessment by preserving multivariate dependencies
标题：生成性气候缩减通过保留多元依赖性来实现高分辨率复合风险评估
链接：https://arxiv.org/abs/2605.11531

【80】Spatial Adapter: Structured Spatial Decomposition and Closed-Form Covariance for Frozen Predictors
标题：空间适配器：冻结预测的结构化空间分解和封闭形式协方差
链接：https://arxiv.org/abs/2605.11394

【81】Operator Spectroscopy of Trained Lattice Samplers
标题：经过训练的格子采样器的操作员光谱
链接：https://arxiv.org/abs/2605.11199

【82】AlphaEarth Satellite Embeddings for Modelling Climate Sensitive Diseases Towards Global Health Resilience
标题：AlphaEarth卫星嵌入用于模拟气候敏感性疾病以实现全球健康弹性
链接：https://arxiv.org/abs/2605.10949

【83】Interpretable rainfall modelling reveals rapid reorganisation of Amazonian rainfall under vegetation loss
标题：可解释降雨模型揭示了亚马逊地区降雨在植被丧失下的快速重组
链接：https://arxiv.org/abs/2605.10948

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