This thesis presents a novel algorithm for training feed-forward neural networks, addressing the limitations of the widely used back-propagation and Quickprop algorithms. While Quickprop is significantly faster than standard ...