Abstract

In this paper we investigate the supervised backpropagation training ofmultilayer neural networks from a dynamical systems point of view. We discusssome links with the qualitative theory of differential equations and introducethe overfly algorithm to tackle the local minima problem. Our approach is basedon the existence of first integrals of the generalised gradient system withbuild-in dissipation.

 

Quick Read (beta)