| 递阶再励学习中Option的自动发现与生成 |
| |
| 引用本文： | 孟江华,朱纪洪,孙增圻. 递阶再励学习中Option的自动发现与生成[J]. 计算机工程与应用, 2006, 42(33): 34-37 |
| |
| 作者姓名： | 孟江华 朱纪洪 孙增圻 |
| |
| 作者单位： | 清华大学,计算机系,智能技术与系统国家重点实验室,北京,100084;清华大学,计算机系,智能技术与系统国家重点实验室,北京,100084;清华大学,计算机系,智能技术与系统国家重点实验室,北京,100084 |
| |
| 摘 要： | Option的自动发现与生成是递阶再励学习的难点之一,论文提出探索密度检测(ED)法,通过检测状态空间中的探索密度来发现并构建Option。和现有的方法相比具有和任务无关、不需要先验知识等优点;能很好地工作于完全未知的环境中;并且构造出的Option,在同一环境下不同任务间可以直接共享。
|
| 关 键 词： | 递阶再励学习 Option 探索密度检测 |
| 文章编号： | 1002-8331（2006）33-0034-04 |
| 收稿时间： | 2006-04-01 |
| 修稿时间： | 2006-04-01 |
Autonomous Discovery and Creation of Options in Hierarchical Reinforcement Learning |
| |
| MENG Jiang-hua,ZHU Ji-hong,SUN Zeng-qi. Autonomous Discovery and Creation of Options in Hierarchical Reinforcement Learning[J]. Computer Engineering and Applications, 2006, 42(33): 34-37 |
| |
| Authors: | MENG Jiang-hua ZHU Ji-hong SUN Zeng-qi |
| |
| Affiliation: | Department of Computer,Tsinghua University,Beijing 100084 |
| |
| Abstract: | Autonomous discovery and creation of Options is one of the open and difficult problems in research of hierachical reinforcement learning(RL).The new method presented in this article is named "Exploration Density(ED) Inspection",which discovers and creates useful Options through inspecting ED in state space.The method has many advantages such as task-independence,no need of prior knowledge.ED method works well in unknown surroundings and the Options it created can transfer among tasks directly. |
| |
| Keywords: | Option |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
