| 多代理最优响应Q学习及收敛性证明 |
| |
| 引用本文： | 张化祥 黄上腾. 多代理最优响应Q学习及收敛性证明[J]. 计算机科学, 2004, 31(4): 96-98 |
| |
| 作者姓名： | 张化祥 黄上腾 |
| |
| 作者单位： | 上海交通大学计算机科学与工程系,上海,200030;上海交通大学计算机科学与工程系,上海,200030 |
| |
| 摘 要： | 在分析了多代理强化学习的基础上，提出了一种基于对手策略假设的代理最优响应强化学习规则，并证明了当对手策略满足一定条件时，基于该学习规则的Q值收敛。实验结果与理论证明相一致。
|
| 关 键 词： | 强化学习 Markov对策 收敛 |
Multiagent Optimal Response Q-learning and its Convergence |
| |
| ZHANG Hua-Xiang HUANG Shang-Teng. Multiagent Optimal Response Q-learning and its Convergence[J]. Computer Science, 2004, 31(4): 96-98 |
| |
| Authors: | ZHANG Hua-Xiang HUANG Shang-Teng |
| |
| Abstract: | Based on analysis of multiagent reinforcement learning, an agent optimal response learning rule is proposed provided the assumptions of opponents' policy. Q values have been proved to be convergent if opponents' policy satisfies certain restrictions, and experimental results of grid games are consistent with the convergence proof. |
| |
| Keywords: | Reinforcement learning Markov games Convergence |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
点击此处可从《计算机科学》下载全文 |
