Improving back-propagation learning using auxiliary neural networks

Multi-layered perceptrons with the back-propagation learning algorithm represent an emerging tool in non-linear systems modelling and control. One of the main drawbacks of the traditional back-propagation algorithm is its slow rate of convergence. A new method to improve the speed of the learning phase, involving the use of a suitable number of additional neural networks, is proposed. The auxiliary networks work concurrently to the principal network without slowing down the procedure. In this paper, it is shown how to choose the structure of the auxiliary networks and how these have to be trained. Several examples confirm the suitability of the proposed procedure