期刊名称:International Journal of Artificial Intelligence and Expert Systems (IJAE)
电子版ISSN:2180-124X
出版年度:2013
卷号:4
期号:1
页码:17-26
出版社:Computer Science Journals
摘要:Typical fuzzy reinforcement learning algorithms take value-function based approaches, such as fuzzy Q-learning in Markov Decision Processes (MDPs), and use constant or linear functions in the consequent parts of fuzzy rules. Instead of taking such approaches, we propose a fuzzy reinforcement learning algorithm in another approach. That is the policy gradient approach. Our method can handle fuzzy sets even in the consequent part and also learn the rule weights of fuzzy rules. Specifically, we derived learning rules of membership functions and rule weights for both cases when input/output variables to/from the control system are discrete and continuous.
关键词:Reinforcement Learning; Policy Gradient Method; Fuzzy Inference; Membership Function