Regular fuzzy neural network as universal approximatorJZ]of fuzzy valued function
Puyin Liu · Kongzhi yu juece · 2003
The systematic analysis on approximation capability of four-layer feedforward regular fuzzy neural network(FNN) is presented. This is done by fuzzy valued multivariate Bernstein polynomial whoseapproximationtofuzzyvalued functions is guaranteed. Such a four-layer FNN constitutes a universal approximator of fuzzy valued function. That is, each continuous fuzzy valued function defined on any compact set of Euclidean space can be approximated by the FNN to any degree of accuracy. A numerical example shows the realization process of the approximation.