Aproximaciones entre las redes neuronales y los operadores borrosos

Abstract

This paper describes the main results concerning a comparative study between simple architectures of trained neural networks using the logistic function as activation function, and the performance of the classic Lukasiewicz operators. The paper shows how those operators can explain trained neural networks. Other fuzzy operators are tested and the results explained. In all cases the results obtained promise an advance in the explanation of trained neural networks using fuzzy logic.