An novel neural network training based on hybrid DE and BP

Abstract

This paper proposes a new approach for training FNN by hybrid DE and BP. It combines the advantages of the global search performed by DE over the FNN parameter space and the local search of BP. Using a function approximation as an illustration, we compare the HDEBP and BP for effectiveness and efficiency for training FNN. It shows that the use of new method can provide better results than BP. © 2005 by International Federation for Information Processing.