Mixed-Integer Evolutionary Optimization of Artificial Neural Networks

Abstract

A novel application to the optimization of artificial neural networks (ANNs) is presented in this paper. Here, the weight and architecture optimization of ANNs can be formulated as a mixed-integer optimization problem. And then a mixed-integer evolutionary algorithm (Mixed-Integer Hybrid Differential Evolution, MIHDE) is used to optimize the ANN. Finally, the optimized ANN is applied to the prediction of chaotic time series. The satisfactory results are achieved, and demonstrate that the optimized ANN by MIHDE can effectively predict the chaotic time series.