Refining linear fuzzy rules by reinforcement learning
H.R. Berenji, P.S. Khedkar, A. Malkani · Proceedings of IEEE 5th International Fuzzy Systems · 2002
We present an algorithm that refines a set of linear fuzzy rules, which use ellipsoidal radial basis functions in their antecedents and have multiple linear outputs in their consequents (similar to TSK rules), using reinforcement learning. We show how this learning algorithm can be used to refine the performances of controllers for a typical cart-pole balancing system.