cs.LG（2024-08-21）

📊 共 17 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (7) 支柱二：RL算法与架构 (RL & Architecture) (7) 支柱七：动作重定向 (Motion Retargeting) (2 🔗1) 支柱八：物理动画 (Physics-based Animation) (1 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (7 篇)

#题目一句话要点标签🔗⭐
1 Sliding Window Training -- Utilizing Historical Recommender Systems Data for Foundation Models 提出滑动窗口训练，利用历史推荐系统数据提升基础模型对长期用户偏好的学习。 foundation model
2 MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models MARLIN：在大语言模型上实现混合精度自回归并行推理，提升批量推理效率。 large language model
3 SORSA: Singular Values and Orthonormal Regularized Singular Vectors Adaptation of Large Language Models 提出SORSA，一种基于奇异值分解和正交正则化的高效参数微调方法，加速大语言模型收敛。 large language model
4 Design Principle Transfer in Neural Architecture Search via Large Language Models 提出基于大语言模型的神经架构搜索设计原则迁移框架，提升搜索效率。 large language model
5 Mixed Sparsity Training: Achieving 4$\times$ FLOP Reduction for Transformer Pretraining 提出混合稀疏训练（MST），在Transformer预训练中实现4倍FLOPs降低。 large language model
6 Data-Centric Machine Learning for Earth Observation: Necessary and Sufficient Features 针对地球观测，提出数据中心机器学习方法，探究必要和充分特征集。 multimodal
7 FedMoE: Personalized Federated Learning via Heterogeneous Mixture of Experts 提出FedMoE：一种基于异构专家混合的个性化联邦学习框架，用于解决FedLLM中的数据异构性问题。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (7 篇)

#题目一句话要点标签🔗⭐
8 Using Part-based Representations for Explainable Deep Reinforcement Learning 提出一种非负训练方法，用于深度强化学习中可解释的基于部分的策略模型。 reinforcement learning deep reinforcement learning
9 Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction 提出基于对象中心抽象的高效探索和判别世界模型学习方法 reinforcement learning world model
10 Critique-out-Loud Reward Models 提出Critique-out-Loud奖励模型，提升LLM在RLHF中的偏好建模能力 reinforcement learning RLHF large language model
11 Optimizing Interpretable Decision Tree Policies for Reinforcement Learning 提出DTPO算法，直接优化强化学习中可解释决策树策略 reinforcement learning imitation learning
12 Representation Learning of Complex Assemblies, An Effort to Improve Corporate Scope 3 Emissions Calculation 提出基于半监督学习的企业电子硬件替代部件识别框架，提升Scope 3排放计算准确性。 representation learning
13 Offline Policy Learning via Skill-step Abstraction for Long-horizon Goal-Conditioned Tasks 提出GLvSA框架，通过技能步抽象进行离线策略学习，解决长程目标条件任务。 policy learning
14 Estimated Audio-Caption Correspondences Improve Language-Based Audio Retrieval 提出基于估计音频-文本对应关系的语言音频检索方法，提升检索性能。 contrastive learning distillation

🔬 支柱七：动作重定向 (Motion Retargeting) (2 篇)

#题目一句话要点标签🔗⭐
15 Time Series Foundation Models and Deep Learning Architectures for Earthquake Temporal and Spatial Nowcasting 提出MultiFoundationQuake模型以解决地震实时预测问题 spatial relationship foundation model
16 ST-USleepNet: A Spatial-Temporal Coupling Prominence Network for Multi-Channel Sleep Staging 提出ST-USleepNet，通过时空耦合显著性网络实现多通道睡眠分期 spatial relationship ✅

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

#题目一句话要点标签🔗⭐
17 FATE: Focal-modulated Attention Encoder for Multivariate Time-series Forecasting 提出FATE：一种焦点调制注意力编码器，用于多元时间序列预测。 spatiotemporal ✅

⬅️ 返回 cs.LG 首页 · 🏠 返回主页