A weighted fuzzy reasoning and its corresponding neural network

This paper presents a weighted fuzzy reasoning method and a fuzzy neural network (FNN) corresponding to the weighted fuzzy reasoning is designed. The back-propagation algorithm for training this FNN, which can be used to refine the weights of weighted fuzzy production rules (WFPRs) so that learning accuracy can be improved considerably is derived. It is demonstrated that the representative power of WFPRs is better than that of fuzzy rules without weights and the time required to consult with domain experts to obtain the weights will greatly be reduced due to the learning capability of the FNN. The proposed backpropagation and weight refinement algorithms are applied to a benchmark problem such as the Iris classification problem and the consequent WFPRs show a kind of optimization feature for learning from examples.