The Guelph Darwin Project: the evolution of neural networks by genetic algorithms
Deborah Stacey, Stefan C. Kremer · 2002
Summary form only given, as follows. The Guelph Darwin Project is involved in research into the development of a system based on the principles of Darwinian evolution to improve the learning algorithms of feedforward, backpropagation (BP) neural networks. This selective-type system for artificial neural networks is achieved through the use of genetic algorithms. The project has concentrated on three stages in system development: (1) the development of a genetic code (learning algorithm rule description) which can express the learning algorithm for a neural net, (2) the analysis of the BP algorithm with respect to improvements in the training procedure, and (3) the testing of systems of competing individuals for the solution of particular problems. Initial experience with the GMIPS Darwin III system demonstrates the utility of this approach for the study of controlled evolution of artificial neural networks by genetic algorithms.>