Self-organizing fuzzy inference system by Q-learning

The fuzzy inference system (FIS) is an expert system based on if-then rules which are extracted from experts' knowledge. To obtain experts' knowledge, however, is not always easy and may be expensive. Q-learning is one type of reinforcement learning in which the desired sequence of actions can be obtained by trial and error without a priori knowledge about the model. In this paper, the extended rule and the interpolation technique are proposed to combine FIS and Q-learning. The resulting self-organizing fuzzy inference system by Q-learning (SOFIS-Q) has the capability of generating the fuzzy rule base automatically and on-line by trial and error without any experts' knowledge.