2026-09论文中文摘要
MemAgent：学习管理异构记忆提供者以支持LLM智能体
MemAgent: Learning to Manage Heterogeneous Memory Providers for LLM Agents
arXiv 2609.32521 · 2026-09-29当用户改变主意：测量与修复LLM智能体中的意图漂移
When Users Change Their Minds: Measuring and Repairing Intent Drift in LLM Agents
arXiv 2609.32520 · 2026-09-29STR：用于减少转向副作用的监督式转码器替换
STR: Supervised Transcoder Replacement for Reducing Steering Side Effects
arXiv 2609.32519 · 2026-09-29UnStep：无需蒸馏训练即可用更少步数加速因果视频扩散
UnStep: Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation
arXiv 2609.32518 · 2026-09-29LocalProp：神经局部化的内存高效反向传播
LocalProp: Neuro-Localized Memory-Efficient Backpropagation
arXiv 2609.32517 · 2026-09-29REFINE：面向安全运营中心智能企业告警分诊的弹性进化框架
REFINE: A Resilient Evolution Framework for Intelligent Enterprise Alert Triage in Security Operations Centers
arXiv 2609.32516 · 2026-09-29可退化量子信道上私有通信的指数强逆定理
An exponential strong converse for private communication over degradable quantum channels
arXiv 2609.32515 · 2026-09-29从异常到失败：构建用于智能体轨迹诊断的因果错误图
From Anomalies to Failures: Constructing Causal Error Graphs for Agentic Trace Diagnosis
arXiv 2609.32514 · 2026-09-29跨域少样本书写者自适应用于真实世界手写数学表达式识别
Cross-Domain Few-Shot Writer Adaptation for Real-World Handwritten Mathematical Expression Recognition
arXiv 2609.32513 · 2026-09-29潜在预测车辆表征保留了什么？测量状态、几何与局部响应
What Do Latent Predictive Vehicle Representations Retain? Measuring State, Geometry, and Local Response
arXiv 2609.32512 · 2026-09-29从他人处学习，为你而行动：面向LLM智能体的跨用户记忆共享
Learning from Others, Acting for You: Cross-User Memory Sharing for LLM Agents
arXiv 2609.32511 · 2026-09-29GeoCR：从异构观测中学习通用云去除先验
GeoCR: Learning a Generalist Cloud Removal Prior from Heterogeneous Observations
arXiv 2609.32510 · 2026-09-29关于签名 $(r,r,p)$ 的费马型方程
On Fermat-type equations of signature $(r,r,p)$
arXiv 2609.32509 · 2026-09-29随机域问题的随机伽辽金近似的深度学习方法
Deep learning methods for stochastic Galerkin approximations of random domain problems
arXiv 2609.32508 · 2026-09-29低跨度Jones多项式的纽结的纯美容手术
Purely cosmetic surgeries on knots with low-span Jones polynomials
arXiv 2609.32507 · 2026-09-29复空间形式中的双调和拉格朗日 $H$-脐子流形
Biharmonic Lagrangian $H$-umbilical submanifolds in complex space forms
arXiv 2609.32506 · 2026-09-29通过极分解在GPU上实现快速可微SVD
Fast Differentiable SVD on GPU via Polar Decomposition
arXiv 2609.32505 · 2026-09-29面向人类对齐的语音情感相似度评判
Toward Human-Aligned Judgement of Speech Emotion Similarity
arXiv 2609.32504 · 2026-09-29TreeRef-BFN：基于二维拓扑与内部三维几何的无等变性从头分子生成
TreeRef-BFN: Equivariance-Free De Novo Molecule Generation based on 2D Topology and Internal 3D Geometry
arXiv 2609.32502 · 2026-09-29非局域双线超材料中的非对称负折射
Asymmetric Negative Refraction in Nonlocal Double-Wire Metamaterials
arXiv 2609.32501 · 2026-09-29Rondo：无监督发现重复出现的时间结构
Rondo: Unsupervised Discovery of Recurring Temporal Structure
arXiv 2609.32500 · 2026-09-29学习锚定提示空间用于大型语言模型的持续适应
Learning an Anchored Prompt Space for Continual Adaptation of Large Language Models
arXiv 2609.32499 · 2026-09-29DAAF：从故障定位到LLM智能体中的可编辑系统资产
DAAF: From Failure Localization to Editable System Assets in LLM Agents
arXiv 2609.32498 · 2026-09-29厄米矩阵模型中谱几何的有理探针
Rational Probes of Spectral Geometry in Hermitian Matrix Models
arXiv 2609.32497 · 2026-09-29局部可靠，全局不足：多跳推理中的局部-全局差距
Locally Sound, Globally Insufficient: The Local-Global Gap in Multi-Hop Reasoning
arXiv 2609.32496 · 2026-09-29传闻：审计员能否信任已部署智能体框架所写入的记录？
Hearsay: Can an Auditor Trust the Record a Deployed Agent Harness Writes?
arXiv 2609.32495 · 2026-09-29Kerr-Bertotti-Robinson黑洞的能量提取：极区坍缩与负能量质子俘获
Energy extraction from Kerr--Bertotti--Robinson black holes: polar region collapse and the capture of negative energy protons
arXiv 2609.32494 · 2026-09-29SoFT：面向可泛化大语言模型微调的软目标
SoFT: Soft Targets for Generalizable LLM Fine-Tuning
arXiv 2609.32493 · 2026-09-29超越提示或技能？基于归因的模块化语言程序优化
Beyond Prompt or Skill? Attribution-Guided Optimization of Modular LLM Programs
arXiv 2609.32492 · 2026-09-29用词汇蕴含解释文本蕴含：利用LLM为形式证明提供词汇关系
Explaining Textual Entailment with Lexical Entailments: Using LLMs to Supply Lexical Relations for Formal Proofs
arXiv 2609.32491 · 2026-09-29当有帮助的文本有害时：视觉语言模型中的选项重定向偏差
When Helpful Text Hurts: Option-Redirecting Bias in Vision-Language Models
arXiv 2609.32489 · 2026-09-29稠密检索何时需要非对称几何？共享与双投影的偏差-方差理论
When Does Dense Retrieval Need Asymmetric Geometry? A Bias-Variance Theory of Shared and Dual Projections
arXiv 2609.32488 · 2026-09-29神经动力学作为量化单元的组合
Neural Dynamics as the Composition of Quantized Units
arXiv 2609.32487 · 2026-09-29弹性选择性频谱混合模型：用于一次训练、多次导出的预算推理
Elastic Selective Spectral Hybrids for Train-Once, Export-Many Budgeted Inference
arXiv 2609.32486 · 2026-09-29联邦LoRA应该共享什么？通过输入感知子空间对齐实现的FedSAIL
What Should Federated LoRA Share? FedSAIL via Input-aware Subspace Alignment
arXiv 2609.32485 · 2026-09-29面向语言模型预训练的可扩展数据多样化：基于杠杆分数采样
Towards Scalable Data Diversification for Language Model Pretraining via Leverage Score Sampling
arXiv 2609.32484 · 2026-09-29将诊断与疾病表征分离：神经动力学引导变形的双视角脑电学习
Separating Diagnosis from Disease Representation: Dual-View EEG Learning with Neural-Dynamics-Guided Deformation
arXiv 2609.32483 · 2026-09-29从结果到策略：学习数学推理的策略效用
From Outcomes to Strategies: Learning Strategy Utility for Mathematical Reasoning
arXiv 2609.32482 · 2026-09-29JEPA 学习掩码无法恢复的内容
JEPA Learns What the Mask Leaves Unrecoverable
arXiv 2609.32481 · 2026-09-29重尾位移的杀死分支随机游走的尖锐相变
A Sharp Phase Transition for Killed Branching Random Walks with Heavy-Tailed Displacements
arXiv 2609.32480 · 2026-09-29对抗性响应下的不可知平滑在线回归
Agnostic Smoothed Online Regression with Adversarial Responses
arXiv 2609.32478 · 2026-09-29从清晰回溯：自我学习远距离文本识别
Back-Tracking from Clarity: Self-Learning to See Text from Afar
arXiv 2609.32477 · 2026-09-29自主智能体与机器人的经济治理：要素来源核算与社会自动化基金
Economic Governance of Autonomous Agents and Robots: Factor-Origin Accounting and Social Automation Funds
arXiv 2609.32476 · 2026-09-29关于秩五的 hit 问题
On the hit problem of rank five
arXiv 2609.32475 · 2026-09-29PC-SubMax：通过正则化子模最大化实现高效提示压缩
PC-SubMax: Efficient Prompt Compression via Regularized Submodular Maximization
arXiv 2609.32474 · 2026-09-29VPEvolve：用于计算光刻的自进化虚拟工艺工程师
VPEvolve: A Self-Evolving Virtual Process Engineer for Computational Lithography
arXiv 2609.32473 · 2026-09-29论言语化置信度先验在大型推理模型校准中的陷阱
On the Pitfalls of Verbalized Confidence Priors for Calibrating Large Reasoning Models
arXiv 2609.32470 · 2026-09-29PULSE：利用稀疏自编码器识别示范效用特征
PULSE: Identifying Demonstration-Utility Features with Sparse Autoencoders
arXiv 2609.32469 · 2026-09-29物质点法在涉及大变形的低速小行星碰撞中的应用与评估
Application and Evaluation of the Material Point Method for Low-Velocity Asteroid Collisions Involving Large Deformations
arXiv 2609.32468 · 2026-09-29Bison：跨数据集学习用于未见化合物扰动预测
Bison: Cross-Dataset Learning for Unseen-Compound Perturbation Prediction
arXiv 2609.32467 · 2026-09-29
