cs.LG（2025-12-26）

📊 共 14 篇论文 | 🔗 4 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (6 🔗1) 支柱二：RL算法与架构 (RL & Architecture) (5 🔗2) 支柱八：物理动画 (Physics-based Animation) (1) 支柱七：动作重定向 (Motion Retargeting) (1) 支柱四：生成式动作 (Generative Motion) (1 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (6 篇)

#题目一句话要点标签🔗⭐
1 The Effectiveness of Approximate Regularized Replay for Efficient Supervised Fine-Tuning of Large Language Models 提出近似正则化回放方法，解决LoRA微调大语言模型时的能力退化问题 large language model
2 Explainable Multimodal Regression via Information Decomposition 提出基于信息分解的可解释多模态回归框架，提升预测精度与可解释性。 multimodal ✅
3 LLMBoost: Make Large Language Models Stronger with Boosting LLMBoost：通过Boosting方法增强大型语言模型性能 large language model
4 Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration 提出完整超参数转移方法以优化大规模模型训练 large language model
5 Unifying Learning Dynamics and Generalization in Transformers Scaling Law 提出统一学习动态与变换器缩放法则以优化模型性能 large language model
6 Prefill vs. Decode Bottlenecks: SRAM-Frequency Tradeoffs and the Memory-Bandwidth Ceiling 研究SRAM频率权衡与内存带宽瓶颈，优化LLM推理能效 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (5 篇)

#题目一句话要点标签🔗⭐
7 MMCTOP: A Multimodal Textualization and Mixture-of-Experts Framework for Clinical Trial Outcome Prediction MMCTOP：用于临床试验结果预测的多模态文本化与混合专家框架 representation learning multimodal
8 Exploring the Heterogeneity of Tabular Data: A Diversity-aware Data Generator via LLMs 提出DATE框架，利用LLM生成高质量多样性表格数据，提升下游任务性能。 DPO direct preference optimization large language model ✅
9 Semiparametric Preference Optimization: Your Language Model is Secretly a Single-Index Model 提出半参数偏好优化方法，解决LLM偏好对齐中链接函数未知问题 policy learning large language model ✅
10 A Comedy of Estimators: On KL Regularization in RL Training of LLMs 研究KL散度估计器对LLM的RL训练影响，提出无偏梯度配置以提升性能。 reinforcement learning large language model
11 PHANTOM: Physics-Aware Adversarial Attacks against Federated Learning-Coordinated EV Charging Management System PHANTOM：针对联邦学习协调的电动汽车充电管理系统的物理感知对抗攻击 reinforcement learning SAC

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

#题目一句话要点标签🔗⭐
12 LangPrecip: Language-Aware Multimodal Precipitation Nowcasting LangPrecip：提出一种语言感知的多模态降水临近预报框架，有效融合文本信息约束降水演化。 spatiotemporal multimodal

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

#题目一句话要点标签🔗⭐
13 Emotion classification using EEG headset signals and Random Forest 提出一种基于脑电信号和随机森林的情感分类模型，用于识别人类情绪状态。 motion prediction

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
14 GQ-VAE: A gated quantized VAE for learning variable length tokens 提出门控量化VAE（GQ-VAE），用于学习变长token，作为现有tokenizer的即插即用替代方案。 VQ-VAE ✅

⬅️ 返回 cs.LG 首页 · 🏠 返回主页