强化学习研究综述

在未知环境中，关于agent的学习行为是一个既充满挑战又有趣的问题，强化学习通过试探与环境交互获得策略的改进，其学习和在线学习的特点使其成为机器学习研究的一个重要分支。介绍了强化学习在理论、算法和应用研究三个方面最新的研究成果，首先介绍了强化学习的环境模型和其基本要素；其次介绍了强化学习算法的收敛性和泛化有关的理论研究问题；然后结合最近几年的研究成果，综述了折扣型回报指标和平均回报指标强化学习算法；最后列举了强化学习在非线性控制、机器人控制、人工智能问题求解、多agent 系统问题等若干领域的成功应用和未来的发展方向。

Reinforcement learning: survey of recent work

The problem of agent learning to act in an unknown world is both challenging and interesting. Reinforcement lear-ning has been successful at finding optimal control policies through trial-and-error interaction with dynamic environment. Its properties of self-improving and online learning make reinforcement learning become one of most important machine learning methods. The goal of this paper was to provide a comprehensive review of reinforcement learning about theory, algorithms and applications. First of all, this paper surveyed the foundation, model of environment of reinforcement learning. Discussed the convergence and generalization of the algorithms in the next. Then deeply discussed two representative selection of these algorithm, including discounted reward and average reward. Finally, provided some applications of reinforcement learning, and pointed out some challenges and problems of reinforcement learning.