| 强化学习中的混合探索方案 |
| |
| 引用本文： | 李春贵,陈波. 强化学习中的混合探索方案[J]. 计算机工程与设计, 2006, 27(9): 1595-1597 |
| |
| 作者姓名： | 李春贵 陈波 |
| |
| 作者单位： | 广西工学院,计算机系,广西,柳州,545006;广西工学院,计算机系,广西,柳州,545006 |
| |
| 基金项目： | 广西自然科学基金;广西工学院校科研和教改项目 |
| |
| 摘 要： | 对强化学习中的探索方案进行了研究,描述了间接探索和直接探索两种方案各自的特点.综合它们的优点,提出了一种集直接探索和间接探索为一体的混合探索方案.该方案在学习的初始阶段,由于对环境的经验知识较少,侧重于直接探索;在获得比较多的经验后,侧重于间接探索,使得行动选择渐渐趋向于最优策略.实验表明该方案比纯粹的间接探索-greedy方案有更高的学习效率.
|
| 关 键 词： | 强化学习 Markov决策过程 探索 利用 Q学习 |
| 文章编号： | 1000-7024（2006）09-1595-03 |
| 收稿时间： | 2005-03-10 |
| 修稿时间： | 2005-03-10 |
Hybrid exploration strategy for reinforcement learning |
| |
| LI Chun-gui,CHEN Bo. Hybrid exploration strategy for reinforcement learning[J]. Computer Engineering and Design, 2006, 27(9): 1595-1597 |
| |
| Authors: | LI Chun-gui CHEN Bo |
| |
| Affiliation: | Department of Computer, Guangxi University of Technology, Liuzhou545006, China |
| |
| Abstract: | Exploration strategies in reinforcement learning were discussed.And characters of direct or indirect exploration strategies were introduced.By integrating both of the strategies' strongpoint,a hybrid exploration strategy was presented.In this strategy,direct strategy is emphasized because oflacking forenvironment knowledge at the beginning stage,and after alot of such knowledge is acquired,indirect strategy will be adopted instead of the direct one,that will ensure the policy converging to the optimal policy.Experiment result show that the new strategy is more effective than-greedy exploration strategy. |
| |
| Keywords: | reinforcement learning Markov decision process exploration exploitation Q learning |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
