cs.LG（2025-08-21）

📊 共 18 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (8 🔗1) 支柱九：具身大模型 (Embodied Foundation Models) (8 🔗1) 支柱七：动作重定向 (Motion Retargeting) (1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱二：RL算法与架构 (RL & Architecture) (8 篇)

#题目一句话要点标签🔗⭐
1 Intern-S1: A Scientific Multimodal Foundation Model 提出 Intern-S1：一个用于科学领域的多模态基础模型，显著提升专业任务性能。 reinforcement learning foundation model multimodal ✅
2 SafeLLM: Unlearning Harmful Outputs from Large Language Models against Jailbreak Attacks SafeLLM：提出基于遗忘学习的防御框架，对抗大语言模型的越狱攻击 direct preference optimization large language model
3 An Efficient Hybridization of Graph Representation Learning and Metaheuristics for the Constrained Incremental Graph Drawing Problem 提出GL-GRASP算法，融合图表示学习与元启发式算法，高效解决约束增量图绘制问题。 reinforcement learning representation learning
4 Recall-Extend Dynamics: Enhancing Small Language Models through Controlled Exploration and Refined Offline Integration 提出RED方法，通过控制探索和优化离线集成，提升小语言模型的推理能力。 reinforcement learning distillation large language model
5 Learning ECG Representations via Poly-Window Contrastive Learning 提出基于多窗口对比学习的ECG表征方法，提升心电信号分析效率与性能。 representation learning contrastive learning
6 Distributed Detection of Adversarial Attacks in Multi-Agent Reinforcement Learning with Continuous Action Space 提出一种基于局部观测的分布式检测器，用于检测连续动作空间多智能体强化学习中的对抗攻击。 reinforcement learning
7 CITE: A Comprehensive Benchmark for Heterogeneous Text-Attributed Graphs on Catalytic Materials CITE：催化材料异构文本属性图综合基准数据集 representation learning large language model
8 Efficient Identification of Critical Transitions via Flow Matching: A Scalable Generative Approach for Many-Body Systems 提出基于Flow Matching的机器学习框架，高效识别多体系统中的临界跃迁。 flow matching

🔬 支柱九：具身大模型 (Embodied Foundation Models) (8 篇)

#题目一句话要点标签🔗⭐
9 CALR: Corrective Adaptive Low-Rank Decomposition for Efficient Large Language Model Layer Compression CALR：一种用于高效大语言模型层压缩的校正自适应低秩分解方法 large language model
10 MMQ: Multimodal Mixture-of-Quantization Tokenization for Semantic ID Generation and User Behavioral Adaptation 提出多模态混合量化（MMQ）框架，用于生成语义ID并适应用户行为，提升推荐系统性能。 multimodal
11 Tutorial on the Probabilistic Unification of Estimation Theory, Machine Learning, and Generative AI 统一概率框架连接估计理论、机器学习与生成AI，解决不确定性数据分析难题 large language model
12 Communication Efficient LLM Pre-training with SparseLoCo SparseLoCo：一种通信高效的LLM预训练方法，实现极高稀疏度并超越DiLoCo large language model
13 Reliable Unlearning Harmful Information in LLMs with Metamorphosis Representation Projection 提出基于变质表示投影的LLM不可逆卸载方法，提升安全性和防御重学习攻击能力 large language model ✅
14 WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling WISCA：通过权重缩放优化LLM训练，提升模型性能 large language model
15 Deep Think with Confidence DeepConf：利用置信度动态过滤推理轨迹，提升LLM推理效率与准确率 large language model
16 End-to-End On-Device Quantization-Aware Training for LLMs at Inference Cost 提出ZeroQAT，实现端到端、低成本的大语言模型量化感知训练。 large language model

🔬 支柱七：动作重定向 (Motion Retargeting) (1 篇)

#题目一句话要点标签🔗⭐
17 Integrated Sensing, Communication, and Computation for Over-the-Air Federated Edge Learning 提出面向无线联邦边缘学习的集成感知、通信与计算框架，优化模型训练性能。 human motion

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

#题目一句话要点标签🔗⭐
18 End-to-End Analysis of Charge Stability Diagrams with Transformers 利用Transformer端到端分析电荷稳定性图，提升量子点器件控制与调谐的通用性与效率。 PULSE

⬅️ 返回 cs.LG 首页 · 🏠 返回主页