Learning fuzzy rules and approximate reasoning in fuzzy neural networks and hybrid systems

The paper considers both knowledge acquisition and knowledge interpretation tasks as tightly connected and continuously interacting processes in a contemporary knowledge engineering system. Fuzzy rules are used here as a framework for knowledge representation. An algorithm REFuNN for fuzzy rules extraction from adaptive fuzzy neural networks (FuNN) is proposed. A case study of Iris classification is chosen to illustrate the algorithm. Interpretation of fuzzy rules is possible by using fuzzy neural networks or by using standard fuzzy inference methods. Both approaches are compared in the paper based on the case example. A hybrid environment FuzzyCOPE which facilitates neural network simulation, fuzzy rules extraction from fuzzy neural networks and fuzzy rules interpretation by using different methods for approximate reasoning is briefly described.