On the Role of Regularization Parameters in Fitness Functions for Evolutionary Designed Artificial Neural Networks

A main criterion for the accuracy of solutions of Artificial Neural Networks (ANNs) for classification tasks is the architecture. In order to find problem-adapted topologies of ANNs, we adopted the evolutionary approach to ANN design by employing a Genetic Algorithm (GA) to evolve ANNs which are represented using a direct encoding method. The role of fitness functions used by the GA is investigated, especially, the impact of a fitness function expressing both, the learning error and a toplogy-dependent regularization term, is studied. A parallel system - the netGEN system - which has been implemented by the authors is generating problem-adapted Feed-Forward ANNs being trained by Error-Back-Propagation. Empirical results on a real world problem taken from the PROBEN1 ANN benchmark suite are presented.
Helmut A. Mayer <helmut@cosy.sbg.ac.at>
Last modified: Mon Jan 5 1998