Awesome RL with Human Feedback / 2024 / arXiv

Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs

作者：Arash Ahmadian, Chris Cremer, Matthias Gallé, Marzieh Fadaee, Julia Kreutzer, Olivier Pietquin, Ahmet Üstün, Sara Hooker · 单位：Cohere For AI · 来源：arXiv · 发布日期：2024

偏好学习策略优化策略梯度人类反馈强化学习在线人类反馈强化学习样本效率
新标签打开 AlphaXiv
Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs figure
AlphaXiv 中文概览（可滚动查看）