cs.LG（2026-03-03）

📊 共 26 篇论文 | 🔗 3 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (11 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (9 🔗1) 支柱一：机器人控制 (Robot Control) (2) 支柱八：物理动画 (Physics-based Animation) (2) 支柱五：交互与反应 (Interaction & Reaction) (1) 支柱四：生成式动作 (Generative Motion) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (11 篇)

#题目一句话要点标签🔗⭐
1 Information Routing in Atomistic Foundation Models: How Equivariance Creates Linearly Disentangled Representations 提出CPD方法，揭示原子级模型中等变性如何解耦线性表示，提升模型性能。 foundation model
2 Sparse autoencoders reveal organized biological knowledge but minimal regulatory logic in single-cell foundation models: a comparative atlas of Geneformer and scGPT 稀疏自编码器揭示单细胞基础模型中的生物学知识组织，但因果调控逻辑极少 foundation model
3 Adapting Time Series Foundation Models through Data Mixtures 提出MixFT方法，通过数据混合微调时间序列基础模型，提升零样本预测性能。 foundation model
4 An Empirical Analysis of Calibration and Selective Prediction in Multimodal Clinical Condition Classification 揭示多模态临床条件分类中选择性预测的不可靠性，强调校准评估的重要性 multimodal
5 Addressing Missing and Noisy Modalities in One Solution: Unified Modality-Quality Framework for Low-quality Multimodal Data 提出统一模态质量框架UMQ，解决低质量多模态数据中的缺失和噪声问题 multimodal
6 Step-Level Sparse Autoencoder for Reasoning Process Interpretation 提出步级别稀疏自编码器(SSAE)用于分析LLM推理过程，揭示其内部逻辑。 large language model chain-of-thought ✅
7 Eliciting Numerical Predictive Distributions of LLMs Without Autoregression 提出一种无需自回归的LLM数值预测分布提取方法，降低计算成本。 large language model
8 From Heuristic Selection to Automated Algorithm Design: LLMs Benefit from Strong Priors 利用高质量先验算法，提升LLM在黑盒优化中的算法设计能力 large language model
9 Causal Learning Should Embrace the Wisdom of the Crowd 融合群体智慧的因果学习：提出一种基于分布式决策的DAG学习框架 large language model
10 MASPOB: Bandit-Based Prompt Optimization for Multi-Agent Systems with Graph Neural Networks 提出MASPOB，基于Bandit优化图神经网络提示，提升多智能体系统性能 large language model
11 ParEVO: Synthesizing Code for Irregular Data: High-Performance Parallelism through Agentic Evolution ParEVO：通过Agent进化合成非规则数据并行代码，实现高性能计算。 large language model ✅

🔬 支柱二：RL算法与架构 (RL & Architecture) (9 篇)

#题目一句话要点标签🔗⭐
12 Beyond One-Size-Fits-All: Adaptive Subgraph Denoising for Zero-Shot Graph Learning with Large Language Models 提出GraphSSR框架，通过自适应子图去噪提升LLM在零样本图学习中的性能 reinforcement learning large language model
13 Contextual Latent World Models for Offline Meta Reinforcement Learning 提出上下文潜在世界模型，用于离线元强化学习中的泛化任务。 reinforcement learning world model representation learning
14 SaFeR-ToolKit: Structured Reasoning via Virtual Tool Calling for Multimodal Safety SaFeR-ToolKit：通过虚拟工具调用实现多模态安全结构化推理 DPO multimodal ✅
15 CGL: Advancing Continual GUI Learning via Reinforcement Fine-Tuning 提出CGL框架，通过强化微调提升GUI Agent的持续学习能力 reinforcement learning large language model multimodal
16 Breaking the Prototype Bias Loop: Confidence-Aware Federated Contrastive Learning for Highly Imbalanced Clients 提出信心感知的联邦对比学习以解决客户端数据不平衡问题 contrastive learning geometric consistency
17 Next Embedding Prediction Makes World Models Stronger NE-Dreamer：基于Transformer的下一嵌入预测增强世界模型 reinforcement learning world model dreamer
18 Learning Memory-Enhanced Improvement Heuristics for Flexible Job Shop Scheduling 提出基于记忆增强改进搜索的MIStar框架，解决柔性作业车间调度问题 reinforcement learning deep reinforcement learning DRL
19 Heterogeneous Agent Collaborative Reinforcement Learning 提出HACRL框架，通过异构智能体协同强化学习提升样本利用率和知识迁移。 reinforcement learning distillation
20 Reinforcement Learning with Symbolic Reward Machines 提出符号奖励机(SRM)，解决强化学习中奖励函数人工标注问题 reinforcement learning

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
21 Real-Time Generative Policy via Langevin-Guided Flow Matching for Autonomous Driving 提出基于Langevin引导的Flow Matching实时生成策略DACER-F，用于自动驾驶。 humanoid reinforcement learning flow matching
22 Improving Diffusion Planners by Self-Supervised Action Gating with Energies SAGE：通过自监督能量动作门控改进扩散规划器，提升动态一致性。 locomotion manipulation reinforcement learning

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

#题目一句话要点标签🔗⭐
23 Physics-informed post-processing of stabilized finite element solutions for transient convection-dominated problems 提出基于物理信息的后处理方法，提升对流占优瞬态问题稳定有限元解的精度 spatiotemporal
24 SynthCharge: An Electric Vehicle Routing Instance Generator with Feasibility Screening to Enable Learning-Based Optimization and Benchmarking SynthCharge：一种电动汽车路径规划实例生成器，支持学习优化与基准测试。 spatiotemporal

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

#题目一句话要点标签🔗⭐
25 Integrating Homomorphic Encryption and Synthetic Data in FL for Privacy and Learning Quality 提出Alt-FL：结合同态加密与合成数据，提升联邦学习隐私与模型质量 OMOMO

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
26 Bridging Diffusion Guidance and Anderson Acceleration via Hopfield Dynamics 通过Hopfield动态桥接扩散引导与Anderson加速，提升生成质量。 classifier-free guidance

⬅️ 返回 cs.LG 首页 · 🏠 返回主页