2026-09论文中文摘要
涌现，而非带宽：物理耦合与学习型多智能体通信的极限
Emergence, Not Bandwidth: Physical Coupling and the Limits of Learned Multi-Agent Communication
arXiv 2609.34373 · 2026-09-30Temperley-Lieb 幺半群及其他范畴中的幂等元
Idempotents in the Temperley-Lieb Monoid and Other Categories
arXiv 2609.34282 · 2026-09-30Arachne：在动态异构集群上学习规划并行训练
Heddle: Learning Structural Templates for Parallelism Planning on Heterogeneous GPU Clusters
arXiv 2609.34244 · 2026-09-30开放文本生成的连贯性感知分布评估
Coherence-Aware Distributional Evaluation of Open-Ended Text Generation
arXiv 2609.34240 · 2026-09-30RLE-Bench：面向机器人学习工程师的编码智能体资格考试
RLE-Bench: A Qualifying Exam for Coding Agents as Robot Learning Engineers
arXiv 2609.34210 · 2026-09-30冻结的裁判，移动的智能体：版本依赖的LLM裁判误差与裁判辅助智能体评估的局限性
Frozen Judges, Moving Agents: Version-Dependent LLM-Judge Error and the Limits of Judge-Assisted Agent Evaluation
arXiv 2609.34198 · 2026-09-30基于人类示教的统一视觉-触觉-动作建模用于灵巧操作
Unified Visual-Tactile-Action Modeling from Human Demonstrations for Dexterous Manipulation
arXiv 2609.34182 · 2026-09-30GradLev：通过协态预测实现令牌并行的测试时训练
GradLev: Token-Parallel Test-Time Training Via Costate Prediction
arXiv 2609.34174 · 2026-09-30基于自然图像自编码器的fMRI表征用于特质与状态预测
Natural Image Autoencoder-Based fMRI Representations for Trait and State Prediction
arXiv 2609.34167 · 2026-09-30含时Kohn-Sham反演的基础函数
Basis Functions for Time-Dependent Kohn-Sham Inversion
arXiv 2609.34164 · 2026-09-30GUITAR：通过状态转换对GUI代理进行结构化失败诊断
GUITAR: Structured Failure Diagnosis of GUI Agents via State Transitions
arXiv 2609.34113 · 2026-09-30发现周期为62分钟且具有近正交自转几何结构的长周期暂现源
Discovery of a 62-min long-period transient with near-orthogonal rotator geometry
arXiv 2609.34102 · 2026-09-30使用持久同调分析异质质量资源获取与异质严重性干扰
Using Persistent Homology to Analyze Access to Heterogeneous-Quality Resources and Heterogeneous-Severity Nuisances
arXiv 2609.34090 · 2026-09-30揭示信息流中的非正态性：社交媒体级联的网络结构与动态
Uncovering Non-Normality in Information Flow: Network Structure and Dynamics of Social Media Cascades
arXiv 2609.34026 · 2026-09-30FINGR：学习真实世界魔方解算的灵巧手控制
FINGR: Learning Dexterous Hand Control for Real-World Rubik's Cube Solving
arXiv 2609.33973 · 2026-09-30Greenpixie的AI令牌方法论：评估开放与封闭权重模型的AI令牌对能源、水资源及二氧化碳当量的影响
Greenpixie's AI Token Methodology: Assessing the Energy, Water and CO2-eq Impact of AI Tokens for Open and Closed Weight Models
arXiv 2609.33965 · 2026-09-30关于在分布鲁棒性模糊中纳入决策依赖的相关性研究
On the Relevance of Incorporating Decision Dependence in Distributional Ambiguity
arXiv 2609.33926 · 2026-09-30量化误差是谱平坦的：单个随机探针是一种校准的、无数据的灵敏度估计器，并应用于预算目标混合精度量化
Quantization Error Is Spectrally Flat: A Single Random Probe Is a Calibrated, Data-Free Sensitivity Estimator, with Application to Budget-Targeted Mixed-Precision Quantization
arXiv 2609.33923 · 2026-09-30LLMs在训练预测自身准确性时学习不同形式的元认知
LLMs learn different forms of metacognition when trained to predict their own accuracy
arXiv 2609.33886 · 2026-09-30扩散奖励模型
Diffusion Reward Models
arXiv 2609.33803 · 2026-09-30Skill2Env：面向通用智能体的基于技能的能力导向环境合成
Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents
arXiv 2609.33772 · 2026-09-30翻转可打包性：驯服图类的统一刻画
Flip-packability: uniform characterisations of tame graph classes
arXiv 2609.33705 · 2026-09-30隐式-显式时间积分方案及其基于物理的预处理器在双流体托卡马克边界模拟中的应用
Implicit-Explicit time integration scheme with Physics-based preconditioning for two-fluid tokamak boundary simulations
arXiv 2609.33677 · 2026-09-30谁阻挡谁？美式橄榄球中阻挡者和冲传者的概率性阻挡分配评估
Who Blocks Whom? Probabilistic Pass-Blocking Assignments for Evaluating Blockers and Pass Rushers in American Football
arXiv 2609.33664 · 2026-09-30剪枝CTC用于大词表语音识别训练的内存高效方法
Pruned CTC for Memory-Efficient Large-Vocabulary ASR Training
arXiv 2609.33645 · 2026-09-30量化黑盒语言模型中的行为尾部
Quantifying Behavioral Tails in Black-Box Language Models
arXiv 2609.33638 · 2026-09-30攀登陡坡：基于课程强化学习的前沿模型提示注入红队测试
Climbing the Hill: Prompt Injection Red-Teaming Against Frontier Models with Curriculum Reinforcement Learning
arXiv 2609.33628 · 2026-09-30布尔累积量与更新驱动系统的精确约化描述
Boolean Cumulants and Exact Reduced Descriptions of Renewal-Driven Systems
arXiv 2609.33621 · 2026-09-30SymbolicLM：将语言模型训练为符号回归器
SymbolicLM: Training Language Models as Symbolic Regressors
arXiv 2609.33594 · 2026-09-30共享通信信道上的控制数据调度：一种稀疏且无冲突的机制
Control Data Scheduling over Shared Communication Channels: A Sparse and Collision-Free Mechanism
arXiv 2609.33592 · 2026-09-30量子Magnusian的Hopf代数理论
A Hopf Algebraic Theory of the Quantum Magnusian
arXiv 2609.33587 · 2026-09-30PGL-3D：面向三维视觉查询定位的渐进式几何学习
PGL-3D: Towards Progressive Geometric Learning for 3D Visual Query Localization
arXiv 2609.33558 · 2026-09-30用于机器人关节的采用永磁化的盘形磁弹性扭矩传感器
A Disk-Shaped Magnetoelastic Torque Sensor for Robotic Joints Using Permanent Magnetization
arXiv 2609.33542 · 2026-09-30Maldacena-Milekhin-Popov 虫洞的极向度规/轴向导场扰动
Polar-metric/axial-gauge perturbations of the Maldacena-Milekhin-Popov wormhole
arXiv 2609.33511 · 2026-09-30树状数据交易中的利润再分配机制
Profit Reallocation Mechanisms in Tree-based Data Trading
arXiv 2609.33506 · 2026-09-30SphMind：面向360度相机的鲁棒、免训练VLM空间推理
SphMind: Towards Robust, Training-Free VLM-based Spatial Reasoning with a 360 Camera
arXiv 2609.33462 · 2026-09-30共享前缀所隐藏的问题：用于在线策略蒸馏的轨迹丢弃
What Shared Prefixes Hide: Trajectory Dropout for On-Policy Distillation
arXiv 2609.33455 · 2026-09-30TT-VidT：解耦时间轴以实现高效的以运动为中心的视频预训练
TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining
arXiv 2609.33419 · 2026-09-30任意色散单模光纤中的超高斯脉冲畸变
Supergaussian pulse distortion in single-mode fibers with arbitrary dispersion
arXiv 2609.33404 · 2026-09-30SciGen-Verifier：科学图像生成中可解释验证的多模态推理器
SciGen-Verifier: A Multimodal Reasoner for Explainable Verification in Scientific Image Generation
arXiv 2609.33399 · 2026-09-30PulseQuant：面向4比特视频扩散Transformer的传播引导子空间校正
PulseQuant: Propagation-Guided Subspace Correction for 4-Bit Video Diffusion Transformers
arXiv 2609.33384 · 2026-09-30AquaWAM：面向水下具身智能体的动力学感知世界动作模型
AquaWAM: A Dynamics-aware World Action Model for Underwater Embodied Agents
arXiv 2609.33299 · 2026-09-30来自Dirichlet逆的素数检测恒等式
Prime-Detecting Identities from Dirichlet Inversion
arXiv 2609.33278 · 2026-09-30GTRL：基于时间差分的基础分治价值学习
GTRL: Grounding Divide-and-Conquer Value Learning with Temporal Differences
arXiv 2609.33259 · 2026-09-30PARSEE-VAD：基于命题感知推理与流式证据升级的高效免训练在线视频异常检测
PARSEE-VAD: Efficient Training-Free Online Video Anomaly Detection via Proposition-Aware Reasoning and Streaming Evidence Escalation
arXiv 2609.33236 · 2026-09-30教自己看向何处：用于推理的在线策略注意力自蒸馏
Teach Yourself Where to Look: On-Policy Attention Self-Distillation for Reasoning
arXiv 2609.33200 · 2026-09-30我们应该信任哪些自我改进？智能体复用基准时的可靠自我改进
Which Self-Improvements Should We Trust? Reliable Self-Improvement When Agents Reuse Their Benchmarks
arXiv 2609.33180 · 2026-09-302023年7月彗星12P/Pons-Brooks的爆发：尘埃彗发与弧状结构的观测和建模
The July 2023 Outburst of Comet 12P/Pons-Brooks: Observations and Modeling of Dust Coma and Arc Structure
arXiv 2609.33170 · 2026-09-30超越状态即动作：利用命令-状态差异进行机器人模仿学习
Beyond State-as-Action: Exploiting Command-State Discrepancy for Robot Imitation Learning
arXiv 2609.33145 · 2026-09-30知道不等于选择：显式验证在生成式偏好之外提供了什么
Knowing Is Not Choosing: What Explicit Verification Adds Beyond Generative Preference
arXiv 2609.33142 · 2026-09-30
