cs.LG（2026-09-28）

📊 共 67 篇论文 | 🔗 6 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (42 🔗4) 支柱九：具身大模型 (Embodied Foundation Models) (18 🔗2) 支柱一：机器人控制 (Robot Control) (3) 支柱八：物理动画 (Physics-based Animation) (2) 支柱五：交互与反应 (Interaction & Reaction) (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) (42 篇)

#题目一句话要点标签🔗⭐
1 Adjoint Guidance Flow: Amortized Critic Guidance for VLA Policies 提出Adjoint Guidance Flow以优化VLA策略的长期任务回报 behavior cloning vision-language-action VLA
2 LRC-JEPA: Disentangling Dynamics and Residual Context for Efficient World Models 提出LRC-JEPA以解决动态与残余上下文纠缠问题 world model world models JEPA
3 Distillation Defenses Easily Break After Reinforcement Learning 提出强化学习后蒸馏攻击的新威胁模型 reinforcement learning distillation large language model
4 KiT: A Foundation Model for Financial Time-Series Forecasting using DiffusionTransformers 提出KiT模型以解决金融时间序列预测中的信号噪声问题 flow matching foundation model ✅
5 Structured Latent Modeling for Supervised Multimodal Information Decomposition 提出结构化潜变量建模以解决多模态信息分解问题 representation learning multimodal
6 On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics 研究蒸馏动态中的策略影响，挑战传统观点 reinforcement learning policy learning distillation
7 TIDE: Teacher-Student Transition via Informative Distillation and Exploration for Agentic RL 提出TIDE以解决多回合强化学习中的教师-学生动态平衡问题 reinforcement learning teacher-student distillation
8 Learning What to Recall: Adaptive Multi-Cue Episodic Memory for World Models 提出未来感知回忆框架以解决世界模型中的记忆选择问题 world model world models
9 Graph World Models for Constrained Epidemic Policy Planning 提出EpiMind以解决区域间疫情政策规划问题 world model world models
10 Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation 提出域归一化多教师在线蒸馏以提升强化学习模型性能 reinforcement learning distillation instruction following
11 NeuronDiscover: Agent-in-Twin for Mechanistic Discovery in Neuronal Microenvironments with World Action Models 提出NeuronDiscover以解决神经微环境中的机制发现问题 world action model world action models
12 Scalable In-Context Reinforcement Learning with Recurrent Algorithm Distillation 提出递归算法蒸馏以解决长时间决策的可扩展性问题 reinforcement learning distillation
13 Teacher-Student Gaps Are Not Enough: Outcome-Guided On-Policy Distillation for Multi-Turn Autonomous Agents 提出结果导向的在线蒸馏方法以提升多回合自主代理的表现 teacher-student distillation
14 BA-DPO: Bias-Adjusted Direct Preference Optimization for Language Model Alignment 提出BA-DPO以解决语言模型偏见对齐问题 DPO direct preference optimization
15 Control-Geometry Straightening for Sampling-Based Latent Planning 提出控制几何直化方法以优化基于采样的潜在规划 world model worldmodel world models
16 An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning 提出最小二乘策略蒸馏以提高样本效率的语言模型推理 reinforcement learning teacher-student distillation
17 d-OPD: Future-Aware On-Policy Distillation for Block Diffusion Language Models 提出d-OPD以解决AR与块扩散模型监督不匹配问题 distillation large language model ✅
18 Not All Rollouts Are Worth Learning: On Trajectory Valuation for Post-Training Reinforcement Learning 提出动态轨迹估值方法以优化强化学习训练 reinforcement learning PPO DPO
19 Provable Benefits of Regularization: Fast Rates for Adversarial Imitation Learning 提出双重正则化算法以提升对抗模仿学习的样本效率 policy learning imitation learning
20 Riccati State Space Models: Non-iterative Parallelization for Nonlinear Sequence Modeling 提出RiccatiSSM以解决非线性序列建模的并行化问题 SSM state space model
21 ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy Learning 提出ReSPO以解决离线学习中的梯度饥饿问题 reinforcement learning policy learning
22 DivOPD: Spread Wide, Look Close for Asynchronous On-Policy Distillation of Multi-turn Agents 提出DivOPD以解决异步多回合代理的蒸馏训练问题 teacher-student distillation ✅
23 Coding Agent Memory Post-training: Unlocking the Memory Potential of Pre-trained File Operations for Long-Horizon Tasks via Reinforcement Learning 提出CAMG以解决长时间任务中的记忆使用问题 reinforcement learning PPO
24 Manifold-Stable Flow Matching 提出流匹配方法以解决流生成中的流形约束问题 flow matching
25 Learning High-Risk High-Precision Motion Control 提出SCOOT算法以解决高风险高精度运动控制问题 reinforcement learning deep reinforcement learning DRL
26 $λ$-JEPA Spectral Anti-Collapse Regularization for Self-Supervised Learning 提出SACReg以解决自监督学习中的表示崩溃问题 JEPA ✅
27 KV-streams for Efficient Compaction in Agentic Reinforcement Learning 提出KV-streams以解决长上下文训练效率问题 reinforcement learning
28 Cartridges++: KV Cache Compression without Off-Context Derailment 提出Cartridges++以解决KV缓存压缩中的上下文偏差问题 distillation large language model
29 Reward-Aligned Reweighting for On-Policy Distillation 提出R²-OPD以解决标准OPD中教师偏好与学生决策不匹配问题 distillation
30 Frontier Learning: Training LLM Reasoners at the Edge of Capability 提出前沿学习以解决LLM推理能力提升问题 reinforcement learning large language model
31 Inductive Feedback for Mixed-Policy Distillation 提出混合策略蒸馏的新方法以解决反馈利用不足问题 distillation
32 SpikeCredit: Temporal Credit Carrier for Reinforcement Learning with Sparse Rewards 提出SpikeCredit以解决稀疏奖励强化学习中的信用分配问题 reinforcement learning
33 Adversarial Consistency-Guided Representation Learning for Multi-view Clustering 提出ACGRL框架以解决多视角聚类中的一致性问题 representation learning
34 FlexiWorld: Learning and Planning via Flexible Action Chunks Across Multiple Time Scales 提出FlexiWorld以解决长时间跨度目标规划问题 world model world models JEPA
35 SymbolicArena: A Unified Infrastructure for Benchmark Distillation and Dynamic Evaluation in Symbolic Regression 提出SymbolicArena以解决符号回归基准评估的有效性问题 distillation
36 GraphHCA: Closed-Form Hindsight Credit Assignment for Long-Horizon LLM Agents 提出GraphHCA以解决长时间跨度LLM代理的信用分配问题 reinforcement learning large language model
37 Cross-Rollout Bellman Closure for Long-Horizon Agentic Reinforcement Learning 提出交叉回滚贝尔曼闭包以解决长时间跨度强化学习问题 reinforcement learning
38 MaPP: A Unified Marginalized Posterior-Predictive Framework for Data-Efficient RLVR 提出MaPP框架以提升数据效率的RLVR方法 reinforcement learning large language model
39 Quality Determines Direction, Length Shapes Magnitude: Length Control for Open-Ended Reinforcement Learning 提出质量引导长度优势塑造以解决开放式强化学习中的长度控制问题 reinforcement learning
40 FORGE: Form-Optimal Routing of Grounded Evidence for Frozen LLM Agents 提出FORGE框架以优化冻结LLM代理的证据路由 distillation foundation model
41 DreamingGoose: Staged Distillation from Autoregressive Transformers to Bidirectional Recurrent Diffusion Language Models 提出DreamingGoose以实现自回归变换器到双向扩散语言模型的蒸馏 distillation
42 An analysis of Mirror-Descent Soft Actor-Critic 提出镜面下降软演员-评论家以提高收敛性 reinforcement learning SAC

🔬 支柱九：具身大模型 (Embodied Foundation Models) (18 篇)

#题目一句话要点标签🔗⭐
43 Conformal Prediction and Conditional Coverage for Tabular Foundation Models 提出C-USIM方法以解决表格基础模型的覆盖问题 foundation model multimodal
44 Narrow Multimodal Fine-Tuning Can Induce Emergent Misalignment 提出多模态微调方法以解决新兴不对齐问题 multimodal
45 Output-aware Residual Stream Pruning for Large Language Models 提出敏感性感知的残差流修剪方法以降低大语言模型推理成本 large language model
46 Large Language Models for Automated Cross-Domain Machine Learning Task Type Identification: A Benchmark Dataset and Evaluation 提出基于大语言模型的跨领域机器学习任务类型识别方法 large language model
47 Weighting Schedules Govern What and When Score-Based Generative Models Learn from Multimodal Data 提出加权调度以优化多模态数据的得分生成模型学习 multimodal
48 A Multimodal Autonomic Sensing Framework for Objective Assessment of Patient Responses to Dental Pulp Stimulation 提出多模态自主感知框架以客观评估牙髓刺激下患者反应 multimodal
49 THEIA: A Multimodal Dataset and Benchmark for Vision-Language Analysis of Layout 提出THEIA以解决模拟电路布局分析问题 multimodal
50 TokenCast: Forecasting Token Consumption During LLM Agent Execution 提出TokenCast以解决LLM代理执行中的令牌消耗预测问题 large language model ✅
51 ScAn-Bench: Evaluating Scaling Analysis Methodology 提出ScAn-Bench以系统评估模型扩展分析方法 foundation model
52 MeqMuon: Matrix-Equilibrating Muon for LLM Pretraining 提出MeqMuon优化器以提升LLM预训练效率 large language model
53 Rethinking Personalized Generation: Test-Time Alignment via Factorized Ranking Models 提出基于分解排名模型的测试时对齐方法以解决个性化生成问题 large language model
54 Beyond Energy: When Sustainability Dimensions Reshape LLM Serving Decisions 提出PRISM框架以优化大型语言模型的可持续性决策 large language model
55 SOLO: Pretraining Billion-Parameter Language Models with Shared-Output Local Learning 提出SOLO以解决大规模语言模型预训练中的更新锁定问题 large language model
56 LLMs are General Asynchronous Agents 提出通用异步代理框架以解决LLMs的顺序交互问题 VLA
57 Collaborative Principle Evolution via Evidence Transfer for Scientific Discovery 提出COEVOLVE以解决科学发现中的协作原则演化问题 large language model
58 A Hierarchy of Entropy-Shapley Games for Multivariate Predictive Uncertainty 提出基于熵-沙普利博弈的层次模型以解决多变量预测不确定性问题 foundation model
59 VEX-Bench: Benchmarking Verification Complexity of LLM-Generated Misinformation 提出VEX-Bench以评估LLM生成虚假信息的验证复杂性 large language model ✅
60 Composable Decoding on the Probability Simplex: Theory and Implementation 提出可组合解码方法以优化语言模型的采样策略 large language model

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
61 X-Reset: Scaling Object-Centric Reinforcement Learning via Cross-Embodiment Resets 提出X-Reset以解决强化学习中的探索问题 manipulation dexterous manipulation sim-to-real
62 Behavioral Foundation Models for Quality Diversity 提出BFM-QD框架以优化行为多样性搜索 locomotion manipulation reinforcement learning
63 Coordinated Lane-Level Variable Speed Limits and Ramp Metering for Successive Weaving Segments Considering Merging/Diverging Risks: A Hybrid Model Predictive Control and Multi-Agent Reinforcement Learning Approach 提出混合模型预测控制与多智能体强化学习以优化城市高速公路交通管理 MPC model predictive control reinforcement learning

🔬 支柱八：物理动画 (Physics-based Animation) (2 篇)

#题目一句话要点标签🔗⭐
64 Addressing Spatial Indistinguishability in Spatiotemporal Prediction via Optimal Transport-Guided Masking 提出STOT以解决时空预测中的空间不可区分性问题 spatiotemporal
65 Simulation-Based Inference for Plate Reverb System Identification 提出基于仿真的推断方法以识别板式混响系统参数 PULSE

🔬 支柱五：交互与反应 (Interaction & Reaction) (1 篇)

#题目一句话要点标签🔗⭐
66 Fiona: Accelerating FHE Inference with Packing-Aware Ternary Weights 提出FIONA以加速全同态加密推理 OMOMO

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

#题目一句话要点标签🔗⭐
67 Attention-based Hierarchical Variational Information Bottleneck for Robust Multi-Agent Communication under Variable Bandwidth 提出AH-VIB以解决有限带宽下多智能体通信问题 occupancy grid

⬅️ 返回 cs.LG 首页 · 🏠 返回主页