New parallel algorithms for back-propagation learning
Robinson Alves, J.D. de Melo, Adrião Duarte Dória Neto, Ana Carolina Albuquerque · 2003
It is presented in this work new parallel algorithms to train a multilayer perceptron network using the error backpropagation algorithm. An analysis of the different paralleling strategies used is shown and aspects such as-task definition and communications profile were taken into account. An application in image compression illustrates the capacity of these new procedures when compared to the classical approach and to other parallel implementations.