Incremental evolution in ANNs: Neural nets which grow

This paper explains the optimisation of neural network topology using Incremental Evolution; that is, by allowing the network to expand by adding to its structure. This method allows a network to grow from a simple to a complex structure until it is capable of fulfilling its intended function. The approach is somewhat analogous to the growth of an embryo or the evolution of a fossil line through time, it is therefore sometimes referred to as an embryology or embryological algorithm. The paper begins with a general introduction, comparing this method to other competing techniques such as The Genetic Algorithm, other Evolutionary Algorithms and Simulated Annealing. A literature survey of previous work is included, followed by an extensive new framework for application of the technique. Finally, examples of applications and a general discussion are presented.