cs.LG（2026-10-23）

📊 共 29 篇论文 | 🔗 1 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (13) 支柱九：具身大模型 (Embodied Foundation Models) (9) 支柱一：机器人控制 (Robot Control) (6 🔗1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (13 篇)

#题目一句话要点标签🔗⭐
1 DT-JEPA: Learning to Predict Future Representations for Decision-Making 提出DT-JEPA以解决离线强化学习中的未来状态预测问题 reinforcement learning policy learning offline RL
2 KnitTrans: Coordinated Trajectory Transfer for Effective and Efficient LLM Reinforcement Learning 提出KnitTrans以解决大语言模型强化学习中的轨迹转移问题 reinforcement learning large language model multimodal
3 PRISM: Promoting Sampling Diversity of Large Language Models with Implicit Stochastic Modulation 提出PRISM框架以解决大语言模型采样多样性不足问题 reinforcement learning large language model
4 Controlled Evaluation of EEG Foundation Models: Random Initialisation, Band-Power and Validation-Selected Probes 评估EEG基础模型的控制方法以提高性能 JEPA foundation model
5 Accordion: Elastic-Depth Inference via Fixed-Prefix Layer Looping 提出Accordion模型以解决多预算推理的性能挑战 reinforcement learning distillation large language model
6 Where Rounding Happens Matters: Rounding-Location Alignment Stabilizes LLM Reinforcement Learning 通过对齐特定部分实现BF16强化学习的稳定性 reinforcement learning large language model
7 Rethinking Representation Geometry in Supervised Contrastive Learning under Long-Tailed Data 提出结构平衡方法以优化长尾数据下的对比学习表现 representation learning contrastive learning
8 Dual-Adaptive Z-Score PPO: Dynamic Clipping Using Action Confidence and Normalized Advantage Magnitude 提出双自适应Z-Score PPO以解决固定剪切阈值问题 reinforcement learning PPO
9 Conditional Flow Models for Offline Maximum Entropy Reinforcement Learning 提出条件流模型以解决离线最大熵强化学习问题 reinforcement learning deep reinforcement learning
10 Knowledge Distillation across Item and Code Spaces for Parallel Generative Recommendation 提出并行生成推荐的知识蒸馏方法以解决标签监督不足问题 distillation
11 Reinforcement Learning with Multi-Step Transition Predictions: A Bellman Theory 提出基于多步转移预测的强化学习方法以提升决策效率 reinforcement learning
12 Internal Discrepancy for Contrastive Learning: Generalization Bounds and Downstream Behaviors 提出内部差异度量以提升对比学习的泛化能力 contrastive learning
13 MS-TTRL: Breaking the Frequency Conflict in Multiple-Solution Test-Time Reinforcement Learning 提出MS-TTRL以解决多解测试时强化学习中的频率冲突问题 reinforcement learning

🔬 支柱九：具身大模型 (Embodied Foundation Models) (9 篇)

#题目一句话要点标签🔗⭐
14 DIGOM: Disentangled Interaction Graphs for Outcome prediction in Multimodal healthcare 提出DIGOM以解决多模态医疗数据的临床结果预测问题 multimodal
15 REACT: Representational Early Analysis and Control of Chain-of-Thought Trajectories 提出REACT以提高链式思维生成的安全性与控制力 chain-of-thought
16 Trusting the Tool: Asymmetric Deference of LLM Vulnerability Auditors to Static-Analysis Evidence 提出静态分析与大型语言模型结合的审计方法以解决证据可信度问题 large language model
17 Sustainability of TSFM Deployment and Pretraining: From Carbon Footprint Analysis to Mitigation via Pruning 提出TSFM模型的碳足迹分析与修剪以提升可持续性 foundation model
18 Quantum-Classical Link Prediction on Dark Web Hyperlink Graphs 提出量子-经典链接预测方法以分析暗网超链接图 multimodal
19 Sharpness-aware Orthogonalized Momentum Algorithms 提出SOMA算法以解决LLM训练中的噪声影响问题 large language model
20 BiLoCo: Binary Low-Rank Corrections for Post-Training Quantization 提出BiLoCo以解决FP4后训练量化的准确性问题 large language model
21 Democratizing LLM Training with the Edge: Challenges, Systems, and the Road Ahead 提出分布式LLM训练框架以解决集中式开发的挑战 large language model
22 Beyond Data Schedule: Fast LLM Pretraining on Data Mixtures by Loss-Reweighting 提出LARE以解决大语言模型预训练中的知识遗忘问题 large language model

🔬 支柱一：机器人控制 (Robot Control) (6 篇)

#题目一句话要点标签🔗⭐
23 Regularised RL improves sim-to-real transfer 提出正则化强化学习以改善仿真到现实的转移问题 sim-to-real reinforcement learning PPO
24 RobustRLlib: A Unified Library and Benchmark for Robust Reinforcement Learning Algorithms 提出RobustRLlib以解决鲁棒强化学习算法的可重用性和评估问题 humanoid reinforcement learning vision-language-action
25 When Does Dynamics Randomization Test Context Inference? 提出动态随机化以测试上下文推断的有效性 dynamics randomization world model world models
26 ZENO: Nonlinear filtering with Zakai Equation Neural Operators 提出ZENO以解决非线性滤波与状态推断问题 locomotion
27 A Tutorial on Physics Constrained Machine Learning for Hybrid Modeling, Optimization, and Control 提出KKT-HardNet以解决物理约束机器学习问题 model predictive control ✅
28 Putting Transformer Theories to the Test 提出七种变换器理论以评估其能力与局限性 MPC

🔬 支柱八：物理动画 (Physics-based Animation) (1 篇)

#题目一句话要点标签🔗⭐
29 When to Trust Memory: Retrieval-Guided Probabilistic Spatiotemporal Forecasting under Distribution Shift 提出REMIND框架以解决时空预测中的分布转移问题 spatiotemporal

⬅️ 返回 cs.LG 首页 · 🏠 返回主页