Toward a new three layer neural network with dynamical optimal training performance

Abstract

This paper proposes a revised dynamic optimal training algorithm for a three layer neural network with sigmoid activation function in the hidden layer and linear activation function in the output layer. This three layer neural network can be used for classification problems, such as the classification of Iris data. This revised dynamic optimal training finds optimal learning rate with its upper-bound for next iteration to guarantee optimal convergence of training result. With modification of initial weighting factors and activation functions, revised dynamic optimal training algorithm is more stable and faster than dynamic optimal training algorithm. Excellent improvements of computing time and robustness have been obtained for Iris data set.