Abstract:
A common problem in training artificial neural networks is determining the optimal network architecture for the particular problem. An ontology is presented that allows for the generalization of the multi-dimensional differential evolution (MD-DE) algorithm in training an ensemble of neural networks with heterogeneous architectures. This generalized algorithm is referred to as the multi-dimensional evolutionary algorithm (MD-EA) since it provides a framework for performing any of the evolutionary optimization algorithms such as differential evolution or genetic algorithms on neural networks with heterogeneous architectures.
