Rational approximant architecture for neural networks
Fabio Massimo Frattale Mascioli, Giovanni Martinelli · European Signal Processing Conference · 1996
A novel approach is proposed for overcoming the multiple minima problem, present in the learning of a supervised neural network. It allows to connect rational function approximations to neural networks and is based on the use of a truncated Fourier expansion for determining: 1) the architecture; 2) the parameters of the net, avoiding local minima in an efficient way.