Automatic task decomposition for the neuroevolution of augmenting topologies (NEAT) algorithm

Abstract

Neuroevolution, the process of creating artificial neural networks through simulated evolution, can become impractical for arbitrarily complex problems requiring large or intricate neural network architectures. The modular feed forward neural network (MFFN) architecture decomposes a problem among a number of independent task specific neural networks, and is suggested here as a means of managing neuroevolution for complex problems. We present an algorithm for evolving MFFN architectures based on the NeuroEvolution of Augmenting Topologies (NEAT) algorithm. The algorithm proposed here, denoted MFF-NEAT, outlines an approach to automatically evolving, attributing fitness values and combining the task specific networks in a principled manner. © 2012 Springer-Verlag.