cs.LG（2025-04-17）

📊 共 18 篇论文 | 🔗 4 篇有代码

🎯 兴趣领域导航

支柱二：RL算法与架构 (RL & Architecture) (9 🔗2) 支柱九：具身大模型 (Embodied Foundation Models) (8 🔗2) 支柱四：生成式动作 (Generative Motion) (1)

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

#	题目	一句话要点	标签	🔗	⭐
1	Representation Learning for Tabular Data: A Comprehensive Survey	表格数据表示学习综述：系统性地回顾了基于深度神经网络的表格数据表示学习方法。	representation learning foundation model multimodal	✅
2	GraphOmni: A Comprehensive and Extendable Benchmark Framework for Large Language Models on Graph-theoretic Tasks	提出GraphOmni框架以评估大语言模型在图论任务上的推理能力	reinforcement learning large language model	✅
3	Recursive Deep Inverse Reinforcement Learning	提出递归深度逆强化学习(RDIRL)，用于在线高效地推断对手目标。	reinforcement learning inverse reinforcement learning
4	PSG-MAE: Robust Multitask Sleep Event Monitoring using Multichannel PSG Reconstruction and Inter-channel Contrastive Learning	提出PSG-MAE，通过多通道脑电重建和对比学习实现鲁棒的多任务睡眠事件监测。	MAE contrastive learning
5	A Collaborative Platform for Soil Organic Carbon Inference Based on Spatiotemporal Remote Sensing Data	WALGREEN平台：利用时空遥感数据协同推断土壤有机碳含量	predictive model spatiotemporal
6	An Optimal Discriminator Weighted Imitation Perspective for Reinforcement Learning	提出IDRL，通过最优判别器加权模仿学习视角解决离线强化学习问题	reinforcement learning behavior cloning
7	Can Masked Autoencoders Also Listen to Birds?	针对鸟鸣声识别，提出Bird-MAE，实现全流程自适应，刷新BirdSet数据集SOTA。	masked autoencoder MAE
8	On the minimax optimality of Flow Matching through the connection to kernel density estimation	通过核密度估计，证明Flow Matching在生成建模中的极小极大最优性	flow matching
9	RL-PINNs: Reinforcement Learning-Driven Adaptive Sampling for Efficient Training of PINNs	提出RL-PINNs，通过强化学习驱动的自适应采样高效训练物理信息神经网络	reinforcement learning

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

#	题目	一句话要点	标签	🔗	⭐
10	Uncertainty-Aware Trajectory Prediction via Rule-Regularized Heteroscedastic Deep Classification	提出SHIFT框架，结合不确定性建模与规则先验，提升轨迹预测在复杂场景下的泛化能力。	large language model	✅
11	VLMGuard-R1: Proactive Safety Alignment for VLMs via Reasoning-Driven Prompt Optimization	提出VLMGuard-R1，通过推理驱动的提示优化实现VLM的主动安全对齐	multimodal
12	It's All Connected: A Journey Through Test-Time Memorization, Attentional Bias, Retention, and Online Optimization	提出Miras框架，通过可定制的记忆、注意力偏置和遗忘机制，设计高性能序列模型。	foundation model
13	NNTile: a machine learning framework capable of training extremely large GPT language models on a single node	NNTile框架：在单节点上训练超大型GPT语言模型	large language model
14	Non-Uniform Class-Wise Coreset Selection for Vision Model Fine-tuning	提出非均匀类感知 Coreset 选择方法 NUCS，用于高效微调视觉模型。	foundation model
15	A Theoretical Framework for OOD Robustness in Transformers using Gevrey Classes	利用Gevrey类，为Transformer在OOD泛化中的鲁棒性提供理论框架	chain-of-thought
16	Tilus: A Tile-Level GPGPU Programming Language for Low-Precision Computation	Tilus：一种面向低精度计算的Tile级GPGPU编程语言	large language model	✅
17	Harmony: A Unified Framework for Modality Incremental Learning	Harmony：提出统一模态增量学习框架，解决持续演进模态序列中的知识获取与保留问题。	multimodal

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#	题目	一句话要点	标签	🔗	⭐
18	Hierarchical Vector Quantized Graph Autoencoder with Annealing-Based Code Selection	提出基于退火码选择的分层向量量化图自编码器，提升图拓扑结构捕获能力。	VQ-VAE

⬅️ 返回 cs.LG 首页 · 🏠 返回主页