Evolving complex neural networks that age

The combination of the broad problem-searching capabilities of a genetic algorithm with the local maxima location capabilities of a hill-climbing algorithm can be a powerful technique for solving classification problems. Producing a number of specialist artificial neural networks, each an expert on one category, can be beneficial when solving problems in which the categories are distinct. This paper describes combining genetic algorithms, hill climbing and sets of specialist artificial neural networks to solve a difficult character recognition problem. It also describes a method by which the effects of a large "elite" sub-population can be counter-balanced by using an aging coefficient in the fitness calculation.