【作者】 王超； 张旭东；

【Author】 WANG Chao;ZHANG Xudong;Department of Electronic Engineering,Tsinghua University;

【机构】 清华大学电子工程系；

【摘要】 增强学习研究智能体如何基于对环境的认知做出行动来最大化长期收益,是解决智能控制问题的重要方法。近几年人们将深度学习应用到增强学习领域,实现了端到端的增强学习算法。这些算法充分利用了深度学习强大的表示学习能力,将智能体对环境的观测直接映射成为控制指令。首先介绍增强学习的基本理论,然后综述几种常用的增强学习和深度增强学习算法。更多还原

【Abstract】 Reinforcement learning,as an important method for solving intelligent control problems,studies how agents take actions based on their perception of the environment so as to maximize the long-term return.In the past few years,deep learning has been applied to the field of reinforcement learning,creating many end-to-end reinforcement learning algorithms.These algorithms make full use of the powerful representation learning ability of deep learning,and directly map the agent’s observation of the environment into control instructions.This paper first introduces the basic theory of reinforcement learning,and then summarizes several commonly used reinforcement learning and deep reinforcement learning algorithms.更多还原