Optimizing neural network weights using genetic algorithms: A case study

Abstract

It has been demonstrated that genetic algorithms (GAs) can help search the global (or near global) optimum weights for neural networks of relatively small sizes. For larger networks, classical genetic algorithms cannot work effectively any more as too many parameters have to be optimized simultaneously. However, in this paper, it is shown that the combination of the techniques of hidden node redundancy elimination, hidden layer redundancy elimination and the use of adaptive probabilities of crossover and mutation can be used to find a satisfactory solution.