cs.LG（2026-09-29）

📊 共 46 篇论文 | 🔗 3 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (22) 支柱九：具身大模型 (Embodied Foundation Models) (20 🔗2) 支柱一：机器人控制 (Robot Control) (2 🔗1) 支柱四：生成式动作 (Generative Motion) (1) 支柱三：空间感知与语义 (Perception & Semantics) (1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (22 篇)

#题目一句话要点标签🔗⭐
1 MoTIF-X: A Multimodal Tokenized Framework for Interpretable and Extensible Molecular Representation Learning 提出MoTIF-X框架以解决分子表示学习中的多模态整合问题 representation learning contrastive learning multimodal
2 Dr. OPD: Learning What to Follow for Optimal On-Policy Distillation of Large Language Models 提出Dr. OPD以优化大语言模型的在线蒸馏训练 distillation large language model
3 Lucid Dreaming for World Models: Learning to Doubt Imagination and Decide by Trust 提出LucidWM以解决世界模型中的不确定性问题 policy learning world model world models
4 Multi-Agent Flow Matching with Decoupled Generative Guidance 提出DeGG-Flow以解决多智能体生成中的约束满足问题 flow matching affordance multimodal
5 Unlocking the Critic: Reward-Free Policy Optimization for LLM Post-Training 提出奖励无关的策略优化方法以提升LLM后训练效率 reinforcement learning PPO large language model
6 Diffusion Policy Improvement with Proposal-Conditioned Refinement Flows 提出Proposal-Conditioned Refinement Flows以解决离线强化学习中的行为建模问题 reinforcement learning offline reinforcement learning diffusion policy
7 Interpolated Policy Distillation: A Controllable Continuum Between Off-Policy and On-Policy Distillation 提出插值策略蒸馏以解决蒸馏轨迹质量与可学习性平衡问题 distillation multimodal
8 Beam Search as Test-Time Self-Distillation via Counterfactual Contexts 提出测试时自蒸馏方法以提升推理性能 policy learning distillation
9 LeapQuant: Efficient Linear Attention with Accurate Recurrent State Quantization 提出LeapQuant以解决长上下文处理中的量化问题 linear attention
10 Explore Broadly, Reason Sharply: Push Small Models toward the Frontier via Sampling 提出并行功率温度化以解决推理过程中的探索与利用问题 reinforcement learning large language model
11 Improving Function Space Flow Matching with Kernel Optimal Transport 提出核函数最优传输以改进函数空间流匹配 flow matching
12 Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning 提出OASIS以解决大规模模型自蒸馏中的监督问题 distillation
13 Privy to the Foil: Recasting Value Estimation with a Self-Privileged Critic for RLVR 提出$π$PPO框架以改进稀疏奖励下的价值估计问题 PPO large language model
14 Optimizer-dependent training dynamics converge to the same one-third optimal data scaling 提出优化器依赖的训练动态以实现统一的1/3最优数据缩放 teacher-student large language model
15 Graph-Conditioned On-Policy Agent Distillation from Off-the-Shelf Teachers 提出图条件的在线策略代理蒸馏以解决教师反馈不足问题 distillation
16 Probabilistic Symbolic-Distillation Model of Droplet Collision for Spray Simulation at High Ambient Pressures 提出概率符号蒸馏模型以解决高压环境下液滴碰撞问题 distillation
17 Identifying ODEs from Unstructured Data with Causal Representation Learning 提出SPEED-AE以从无结构数据中识别常微分方程 representation learning
18 High-Resolution Dynamic Functional Connectivity Generation with Graph-Variate Flow Matching 提出GVD-CFM以解决高分辨率动态功能连接生成问题 flow matching
19 Group-Marginalized Self-Rewarding RL Drives Zero-Label Self-Evolving 提出群体边际自奖励强化学习以解决无标签自我演化问题 reinforcement learning large language model
20 Act First, Reason Later: Accelerating On-Policy Distillation for Multi-Turn Agents via Reference-Conditioned Inverse Dynamics 提出ActFirst-OPD以加速多回合智能体的在线蒸馏训练 distillation
21 Triadic Linear Attention: Three-Dimensional Recurrent States for Long-Context Sequence Modeling 提出三元线性注意力以解决长上下文序列建模问题 linear attention
22 NowcastDiT: Diffusion Transformers are Effective Precipitation Nowcasters 提出NowcastDiT以解决降水短期预报问题 reinforcement learning spatiotemporal

🔬 支柱九：具身大模型 (Embodied Foundation Models) (20 篇)

#题目一句话要点标签🔗⭐
23 Multimodal Detection of Higher-Order Behavioral Constructs: Self-Compassion in Structured Reflective Interaction 提出多模态检测方法以解决自我同情建构的建模问题 multimodal
24 A foundation model for energy and radiation systems built on heterogeneous scientific interfaces 提出GEODE模型以解决异构科学接口下的能量与辐射系统问题 foundation model
25 Latent Inference-Time Guidance of Time Series Foundation Models 提出潜在推理时指导方法以增强时间序列基础模型的预测能力 foundation model
26 TabFM-Auto: Self-Evolving Pipelines for Tabular Foundation Models 提出TabFM-Auto以优化表格基础模型的数据处理流程 foundation model
27 TabFM: A Zero-Shot Foundation Model for Tabular Data 提出TabFM以解决表格数据预测的零-shot学习问题 foundation model
28 Hidden Reasoning Must Leak, but Need Not Be Readable: Fundamental Opportunities and Limits for Chain-of-Thought Monitoring 提出链式思维监控的新理论框架以解决信息泄露问题 chain-of-thought
29 Loss-Guided Pretraining Data Selection for Time-Series Foundation Models 提出静态数据选择框架以优化时间序列基础模型的预训练数据 foundation model
30 UnlearningSoup: Is Repeated Tuning Necessary for Large Language Model Unlearning? 提出UnlearningSoup以解决大语言模型遗忘效率低下问题 large language model
31 Replay the Curvature: Accurate and Scalable NVFP4 Quantization for Large Language Model Inference 提出Schur Replay算法以解决大语言模型推理中的量化问题 large language model
32 $S^3$: Spectral Null-Space Swap Makes Reasoning Models Efficient 提出$S^3$以提升推理模型的效率 multimodal chain-of-thought
33 Hierarchical Compression of Vision-Language Model Benchmarks 提出PRIMEBench以降低视觉语言模型基准评估成本 multimodal
34 Storage Is Not Strategy: State-Conditioned Support Control for LLM Unlearning 提出状态条件支持控制以优化LLM的去学习策略 large language model
35 Delta-Matching: Closing the Final Gap of Native 8-bit Training for LLMs 提出Delta-Matching以解决原生8位训练中的FP8注意力问题 large language model
36 Scaling Influence Functions in LLMs through Eigenbasis-Corrected One-Bit Gradient Projection 提出EOGP以压缩LLM梯度存储成本 large language model
37 Weights Read and Write Features: Scalable Parameter Decomposition Grounded in Activation Space 提出激活支持的参数分解方法以提升模型可解释性 large language model
38 SafeLLM4SE: Statistical Evaluation and Reporting for LLM-based Software Engineering Systems 提出SafeLLM4SE以解决LLM软件工程评估中的不确定性问题 large language model
39 Trajectory Soup: Pushing the Compute-Scaling Frontier of LLM Mid-training via Diverse Trajectories 提出Trajectory Soup以提升大语言模型中期训练的计算效率 large language model
40 Iterative Exact Discrete Guidance for Energy-Based Sampling 提出迭代精确离散引导以解决大离散状态空间采样问题 multimodal ✅
41 CHAIN: Calibrated LLM Forecasting via Causal-Temporal Hypergraph Inference 提出CHAIN以解决大语言模型预测中的校准偏差问题 large language model ✅
42 Invariant Atoms: Sparse Coordinates of Local Semantic Geometry in Language Model Representations 提出不变原子以解决语言模型局部语义几何问题 large language model

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
43 Looped Actor: Depth-Recurrent Reasoning Models for Reinforcement Learning 提出Looped Actor以解决强化学习中的动态计算问题 manipulation reinforcement learning offline reinforcement learning ✅
44 Differentiating Bisimulation Metrics: A Framework for Parametric Markov Chain Fitting via Bicausal Optimal Transport 提出可微分双因果最优传输以解决序列决策问题 sim-to-real imitation learning world model

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
45 On Trajectory-Aware Training for Masked Diffusion Language Models 提出PUMBA框架以解决掩蔽扩散语言模型的训练与推理不一致问题 MDM

🔬 支柱三：空间感知与语义 (Perception & Semantics) (1 篇)

#题目一句话要点标签🔗⭐
46 Procedural Core: A Compact Recurrent Initialization for Vision Transformers 提出Procedural Core以解决Transformer初始化效率问题 depth estimation

⬅️ 返回 cs.LG 首页 · 🏠 返回主页