Training of Cc4 Neural Network Using Unary and Spread Unary Inputs

Abstract

This thesis presents results on training the corner classification (CC4) feed forward neural networks with the recently proposed spread unary inputs. We show the performance of this network is quite like that of the original CC4 network. The spread unary CC4 network is tested on pattern classification data. It is also used in time series prediction which is illustrated using the Mackey-Glass time series, sunspot predictions and the prediction of d-sequences.