cs.LG（2025-02-13）

📊 共 29 篇论文 | 🔗 6 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (14 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (11 🔗3) 支柱八：物理动画 (Physics-based Animation) (2) 支柱一：机器人控制 (Robot Control) (1) 支柱四：生成式动作 (Generative Motion) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (14 篇)

#题目一句话要点标签🔗⭐
1 Improving Acoustic Side-Channel Attacks on Keyboards Using Transformers and Large Language Models 利用Transformer和LLM提升键盘声学侧信道攻击性能 large language model
2 Fine-Tuning Foundation Models with Federated Learning for Privacy Preserving Medical Time Series Forecasting 提出基于联邦学习的微调方法，用于保护隐私的医疗时间序列预测。 foundation model
3 AnomalyGFM: Graph Foundation Model for Zero/Few-shot Anomaly Detection 提出AnomalyGFM，用于零/少样本图异常检测的图基础模型 foundation model
4 RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models 提出RoSTE算法，实现高效的大语言模型量化感知监督微调 large language model ✅
5 Escaping Collapse: The Strength of Weak Data for Large Language Model Training 提出基于Boosting理论的LLM训练框架，解决合成数据训练中的性能崩塌问题 large language model
6 Task Generalization With AutoRegressive Compositional Structure: Can Learning From $D$ Tasks Generalize to $D^{T}$ Tasks? 基于自回归组合结构的任务泛化：从D个任务学习能否泛化到D^T个任务？ large language model chain-of-thought
7 Language in the Flow of Time: Time-Series-Paired Texts Weaved into a Unified Temporal Narrative 提出TaTS框架，利用时序配对文本增强数值时序预测与插补任务性能 large language model multimodal ✅
8 Harnessing Vision Models for Time Series Analysis: A Survey 综述：利用视觉模型进行时间序列分析，弥补序列建模研究的不足。 large language model
9 NestQuant: Nested Lattice Quantization for Matrix Products and LLMs NestQuant：基于嵌套格量化的矩阵乘法和LLM高效量化方案 large language model
10 LoRA Training Provably Converges to a Low-Rank Global Minimum or It Fails Loudly (But it Probably Won't Fail) LoRA训练理论分析：证明收敛至低秩全局最小或显著失败 foundation model
11 Language Agents as Digital Representatives in Collective Decision-Making 提出基于语言代理的数字代表，用于群体决策模拟与机制设计。 large language model
12 GoRA: Gradient-driven Adaptive Low Rank Adaptation 提出GoRA以解决LoRA在适应性和初始化上的不足问题 large language model ✅
13 One-shot Federated Learning Methods: A Practical Guide 综述单轮联邦学习方法，分析挑战与权衡，为未来研究提供指导 large language model
14 End-to-End triplet loss based fine-tuning for network embedding in effective PII detection 提出基于三元组损失的端到端微调方法以提升PII检测效果 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (11 篇)

#题目一句话要点标签🔗⭐
15 Convex Is Back: Solving Belief MDPs With Convexity-Informed Deep Reinforcement Learning 提出凸性指导的深度强化学习方法，解决信念MDP中的值函数学习问题 reinforcement learning deep reinforcement learning DRL ✅
16 When Do Neural Networks Learn World Models? 理论分析神经网络在多任务学习中学习世界模型的能力 world model large language model
17 Digi-Q: Learning Q-Value Functions for Training Device-Control Agents Digi-Q：学习Q值函数训练设备控制Agent，提升离线策略学习效果 reinforcement learning policy learning foundation model ✅
18 Reevaluating Policy Gradient Methods for Imperfect-Information Games 重新评估策略梯度方法在不完美信息博弈中的有效性 reinforcement learning deep reinforcement learning DRL ✅
19 A Survey of Reinforcement Learning for Optimization in Automation 综述：强化学习在自动化优化中的应用，聚焦制造、能源和机器人领域 reinforcement learning
20 Variational Rectified Flow Matching 提出变分校正流匹配，通过建模多模态速度向量场提升生成模型性能。 flow matching
21 Navigating the Social Welfare Frontier: Portfolios for Multi-objective Reinforcement Learning 提出α-近似投资组合以解决多目标强化学习中的社会福利函数选择问题 reinforcement learning
22 Neuro-Symbolic Contrastive Learning for Cross-domain Inference 提出神经符号对比学习框架，提升跨领域推理中逻辑关系的泛化能力。 contrastive learning
23 Beyond Shallow Behavior: Task-Efficient Value-Based Multi-Task Offline MARL via Skill Discovery 提出SD-CQL算法，解决离线多智能体强化学习中的任务泛化与效率问题 CQL conservative q-learning behavior cloning
24 SinSim: Sinkhorn-Regularized SimCLR SinSim：通过Sinkhorn正则化的SimCLR，提升自监督学习表征结构 representation learning contrastive learning
25 Analysis of Off-Policy $n$-Step TD-Learning with Linear Function Approximation 分析线性函数逼近下Off-Policy n步TD学习的收敛性 reinforcement learning policy learning

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

#题目一句话要点标签🔗⭐
26 Neural Spatiotemporal Point Processes: Trends and Challenges 综述：神经时空点过程建模事件，融合深度学习应对复杂时空依赖 spatiotemporal
27 Relational Conformal Prediction for Correlated Time Series 提出CoRel：一种基于关系图深度学习的关联时间序列置信区间预测方法 spatiotemporal

🔬 支柱一：机器人控制 (Robot Control) (1 篇)

#题目一句话要点标签🔗⭐
28 Rolling Ahead Diffusion for Traffic Scene Simulation 提出Rolling Ahead Diffusion模型，用于交通场景中兼顾反应性和效率的模拟。 MPC model predictive control

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
29 Theoretical Benefit and Limitation of Diffusion Language Model 理论分析扩散语言模型的优势与局限性，揭示其在不同指标下的性能表现差异 MDM

⬅️ 返回 cs.LG 首页 · 🏠 返回主页