Fast-Training Algorithm for Hidden-Layer Forward-Feed Neural Networks.

Abstract : A new training algorithm, based on the matrix pseudoinverse method, is shown to train hidden-layer, forward-feed neural networks with high accuracy in a short time for nonlinear time series prediction. The algorithm is applied to chaotic time series generated from the logistics map and the Mackey-Glass delay differential equation and compared to corresponding results generated using a backpropagation training algorithm. We demonstrate orders of magnitude shorter training time and comparable accuracy with the new algorithm and show forecasting and self-generation for these systems.