IMPROVEMENT OF THE NEURAL NETWORK TRAINING

ROTAR DAN

UNIVERSITY OF BACĂU

Abstract

The paper shows a new method for improvement the neural network training. Based on the drawings programs this method allows in advance determination of weight and bias of the neural network. This is very useful for the neural network design, selection and optimization and the time reduce of the neural network training. The method described is based on graphical representation of the input domain of data and the vector representation of the decision surfaces.

Keywords

neural network graphic representation and vector network design back propagation algorithm.