Optimal learning algorithm for multilayer feedforward neural networks

Abstract

A faster new learning algorithm to adjust the weights of multilayer feedforward neural network is proposed. In this new algorithm, the weight matrix (W/sub 2/) of the output layer and the output vector (Y) of the previous layer are treated as two independent variable sets. A optimal solution pair (W/sub 2/*, Y*) is found to minimize the total mean-square-error of the patterns input. Y/sub P/* is then used as the desired output of the previous layer. The optimal weight matrix and layer output vector of the hidden layers in the network will be found with the same method as the output layer. Computer simulation shows that the new algorithm is dominant in converging speed, computing time and node number requirement among the existing learning algorithms.< >