Cooperative Co-evolution of Multilayer Perceptrons

Co-evolution is a posible solution to the problem of simultaneous optimization of artificial neural network and training agorithm parameters, due to its ability to deal with vast search spaces. Moreover, this scheme is recommendable when the optimization problem is decomposable in subcomponents. In this paper an approach to cooperative co-evolutionary optimisation of multilayer perceptrons, that improves the G-Prop genetic back-propagation algorithm, is presented. Obtained results show that this co-evolutionary version of G-Prop obtains similar or better results needing much fewer training epochs and thus using much less time than the sequential versions.