2026-09论文中文摘要
路由应当自负盈亏：面向经济型LLM路由的稀疏监督
Routing Should Pay for Itself: Sparse Supervision for Economical LLM Routing
arXiv 2609.37402 · 2026-09-30AdS不变理论中的量子态
Quantum states in an AdS-invariant theory
arXiv 2609.37401 · 2026-09-30BeatDance：基于层次化时空建模的节拍一致三维舞蹈生成
BeatDance: Generating Beat-Consistent 3D Dance with Hierarchical Spatial-Temporal Modeling
arXiv 2609.37400 · 2026-09-30MEDEM：多引擎深度学习加速器设计方法
MEDEM: Multi-Engine DL Accelerator Design Methodology
arXiv 2609.37399 · 2026-09-30直接经验世界模型优化：超越动作模仿学习世界
Direct Experience World-Model Optimization: Learning the World Beyond Action Imitation
arXiv 2609.37398 · 2026-09-30不可压缩多孔介质方程分层稳态的尖锐稳定性与不稳定性
Sharp stability and instability of stratified steady states for the incompressible porous media equation
arXiv 2609.37397 · 2026-09-30置信度引导的协议中间表示用于大语言模型辅助的安全协议建模
Confidence-Guided Protocol IR for LLM-Aided Security Protocol Modeling
arXiv 2609.37396 · 2026-09-30多体纠缠可以探测二分纠缠无法看到的虫洞模参数
Multipartite Entanglement Can Probe Wormhole Moduli Invisible to Bipartite Entanglement
arXiv 2609.37395 · 2026-09-30多体纠缠能否探测到二分纠缠不可见的虫洞模量？
Can Multipartite Entanglement Probe Wormhole Moduli Invisible to Bipartite Entanglement?
arXiv 2609.37394 · 2026-09-30扩展RIFT 1：将RIFT扩展到第三代引力波分析
Scaling RIFT 1: Extending RIFT to third-generation gravitational-wave analyses
arXiv 2609.37393 · 2026-09-30重新思考扩散语言模型中用于并行解码的软令牌
Rethinking Soft Tokens for Parallel Decoding in Diffusion Language Models
arXiv 2609.37391 · 2026-09-30Elementary Askey-Wilson 函数
Elementary Askey-Wilson Functions
arXiv 2609.37390 · 2026-09-30关于具有显式基本单位的三次域的一个无限族
On an infinite family of cubic fields with explicit fundamental units
arXiv 2609.37389 · 2026-09-30Assouad维数与下维数：具有快速增长频率的Weierstrass型函数图像
Assouad and Lower Dimensions of Graphs of Weierstrass-Type Functions with Rapidly Growing Frequencies
arXiv 2609.37388 · 2026-09-30基于MRI的血管周围间隙扩大负担自动分级的多任务学习
Multi-task learning for the automatic grading of enlarged perivascular space burden using MRI
arXiv 2609.37387 · 2026-09-30解剖感知的3D CT支气管镜可达性预测
Anatomy-Aware Prediction of Bronchoscopic Accessibility from 3D CT
arXiv 2609.37386 · 2026-09-30电离驱动的气相金刚烷团簇脱氢与分子生长
Ionization-Driven Dehydrogenation and Molecular Growth in Gas-Phase Adamantane Clusters
arXiv 2609.37385 · 2026-09-30MoTIF-X：一种用于可解释且可扩展的分子表示学习的多模态标记化框架
MoTIF-X: A Multimodal Tokenized Framework for Interpretable and Extensible Molecular Representation Learning
arXiv 2609.37384 · 2026-09-30中性重整化吸引子的刚性
Rigidity of the attractor of neutral renormalization
arXiv 2609.37383 · 2026-09-30去偏与同时推断：基于残差化Walsh-Hadamard得分的异质性因子效应修饰变量
Debiased and Simultaneous Inference for Heterogeneous Factorial Effect Modifiers via Residualized Walsh-Hadamard Scores
arXiv 2609.37382 · 2026-09-30高维潜在空间中的基于模拟的推断
High-Dimensional Simulation-Based Inference in Latent Spaces
arXiv 2609.37381 · 2026-09-30多阶段库存控制的信息极限：学习、估值与删失
Information Limits of Multistage Inventory Control: Learning, Valuation, and Censoring
arXiv 2609.37380 · 2026-09-30循环Transformer作为优化器
Looped Transformers as Optimizers
arXiv 2609.37379 · 2026-09-30Do-JEPA：从掩蔽到潜在世界模型中的干预
Do-JEPA: From Masking to Intervention in Latent World Models
arXiv 2609.37378 · 2026-09-30超越提示数量：数据如何在同策略蒸馏中塑造迁移
Beyond Prompt Count: How Data Shapes Transfer in On-Policy Distillation
arXiv 2609.37377 · 2026-09-30较短偶环的广义Turán问题
Generalized Turán problems for shorter even cycles
arXiv 2609.37376 · 2026-09-30一致历史与无限张量积上的超度量
Consistent histories and the ultrametric on infinite tensor products
arXiv 2609.37375 · 2026-09-30MG-Thinker：面向多图像推理定位的双轴自反思方法
MG-Thinker: Bi-Axial Self-Reflection for Multi-Image Reasoning Grounding
arXiv 2609.37374 · 2026-09-30多人选举中的策略性投票公理
Strategic Voting Axioms for Multiwinner Elections
arXiv 2609.37373 · 2026-09-30先思考再评分：面向视觉生成的思考奖励模型
Think Before You Score: Thinking Reward Model for Visual Generation
arXiv 2609.37372 · 2026-09-30将学习问题编译为语言模型的适应程序
Compiling Learning Problems into Adaptation Programs for Language Models
arXiv 2609.37371 · 2026-09-30CGC 拟合 HERA 深度非弹性散射与 p+p 碰撞中的前向强子产生
CGC Fit to HERA DIS and Forward Hadron Production in p+p Collisions
arXiv 2609.37370 · 2026-09-30塑造观点：量化自主多智能体LLM交互的心理影响
Shaping Opinion: Quantifying the Psychological Impact of Autonomous Multi-Agent LLM Interactions
arXiv 2609.37369 · 2026-09-30去中心化大语言模型微调中的后门缓解
Backdoor Mitigation in Decentralized LLM Fine-Tuning
arXiv 2609.37367 · 2026-09-30双边与匹配市场中的尖锐次优福利
Sharp Second-Best Welfare in Bilateral and Matching Markets
arXiv 2609.37366 · 2026-09-30偏置补偿的复值双线性滤波
Bias-Compensated Complex-Valued Bilinear Filtering
arXiv 2609.37365 · 2026-09-30周期图算子：所有势函数下色散多项式可约
Periodic graph operators with reducible dispersion polynomials for all potentials
arXiv 2609.37364 · 2026-09-30单位球中Moser-Trudinger型不等式的紧性
The compactness of Moser-Trudinger type inequalities in the unit ball
arXiv 2609.37363 · 2026-09-30预训练一次，随处路由：迈向LLM路由的基础模型
Pretrain Once, Route Anywhere: Towards a Foundation Model for LLM Routing
arXiv 2609.37362 · 2026-09-30SemOPT：通过奖励引导搜索修复基于LLM的优化建模中的语义错误
SemOPT: Fixing Semantic Errors in LLM-based Optimization Modeling via Reward-Guided Search
arXiv 2609.37361 · 2026-09-30数据稀缺下用于模型选择的视觉异常合成
Visual Anomaly Synthesis for Model Selection in Data Scarcity
arXiv 2609.37360 · 2026-09-30Encore：少样本智能体发现操作策略
Encore: Few-Shot Agentic Discovery of Manipulation Strategies
arXiv 2609.37359 · 2026-09-30ALPINE-CRISTAL-JWST巡天：研究高红移星系中星际尘埃、气体和恒星在千秒差距尺度上的作用
The ALPINE-CRISTAL-JWST Survey: Investigating the role of interstellar dust, gas and stars in high-z galaxies at kpc-scales
arXiv 2609.37358 · 2026-09-30Q的加法子群：对数描述与交集构型
Additive subgroups of Q: logarithmic description and intersection configurations
arXiv 2609.37357 · 2026-09-30通过求解器反馈教会大语言模型生成具有挑战性的MILP实例
Teaching LLMs to Generate Challenging MILP Instances via Solver Feedback
arXiv 2609.37356 · 2026-09-30一种自适应分数态将电路机制与视觉层级中的皮层动态联系起来
An adaptive fractional state links circuit mechanisms to cortical dynamics across the visual hierarchy
arXiv 2609.37355 · 2026-09-30先寻后动：视觉语言导航中进度锚定的证据搜寻
Seek Before You Move: Evidence Seeking for Progress Grounding in Vision-Language Navigation
arXiv 2609.37353 · 2026-09-30MCMC收敛图：多峰后验的诊断工具
The MCMC convergence graph: a diagnostic for multimodal posteriors
arXiv 2609.37352 · 2026-09-30端口-哈密顿潜在深思：缓解测试时计算扩展中的深思漂移悬崖
Port-Hamiltonian Latent Deliberation: Mitigating the Deliberation Drift Cliff in Test-Time Compute Scaling
arXiv 2609.37351 · 2026-09-30PCaPaint：通过缓解捷径学习实现前列腺癌图像修复
PCaPaint: Prostate Cancer Inpainting by Mitigating Shortcut Learning
arXiv 2609.37350 · 2026-09-30
