FAST 2ND-ORDER LEARNING ALGORITHM FOR FEEDFORWARD MULTILAYER NEURAL NETWORKS AND ITS APPLICATIONS

The paper presents the efficient training program of multilayer feedfo rward neural networks. It is based on the best second order optimizati on algorithms including variable metric and conjugate gradient as well as application of directional minimization in each step. Its efficien cy is proved on the standard rests, including parity, dichotomy, logis tic and two-spiral problems. The application of the algorithm to the s olution of higher dimensionality problems like deconvolution, separati on of sources and identification of nonlinear dynamic plant are also g iven and discussed. It is shown that the appropriately trained neural network can be used for the nonconventional solution of these standard signal processing tasks with satisfactory accuracy. The results of nu merical experiments are included and discussed in the paper. Copyright (C) 1996 Elsevier Science Ltd.