1 On functional equivalence of certain fuzzy controllers and RBF type approximation schemes ? László

Both general fuzzy systems and most neural networks are universal approximators in the sense that they are capable of approximating any continuous function with arbitrary accuracy with respect to, e.g., the supremum norm. It means that these techniques share approximation capabilities. However, the way they captures the underlying transfer function is different. Fuzzy systems operating with if-then rules have the advantage of easy linguistic interpretability, while neural networks can adapt learning methods to improve their performance according to a training data set. We point out in this paper that several fuzzy controllers implement one of the typical neural networks (having radial basis type activation functions), and hence, their combination may alloy the the advantageous properties of the two techniques.