The Evolving Fuzzy Neural Network (EFuNN) has been proposed as a neurocomputing system that evolves and grows to model a specific problem either for prediction or classification. The mechanism by which it currently learns is largely based on principles where it only evolves its structure to accommodate new data examples that are "different" to what it has already been presented.
This measure of difference is one of the primary drivers that governs the resulting structure of the EFuNN. But is this measure the only means by which this learning system can evolve?
Recently I have been experimenting with an alternative approach that employs the use of an entropy criterion to determine how the EFuNN will evolve it structure. I will present my approach, discuss how it differs from the original EFuNN learning algorithm, and present some preliminary findings on its performance.
This page is maintained by the seminar list administrator.