A Region Reproduction Algorithm for Optimization of Neural Networks

Abstract

Among the research of artificial neural networks, the most important problem is how to select the appropriate parameters for an artificial neural network. In this paper, a new evolutionary algorithm called region reproduction algorithm (RRA) is introduced to optimize the parameters of neural networks. The algorithm firstly generates some regions in space and then the offspring in the region is reproduced by the fitness in the superior regions. Because the algorithm is more concerned in the superior regions, it has more probability to find the optimal than traditional algorithms. Experiments for the Apple stock price data and Dell stock price data shows that our proposed RRA-NN model performed better than the traditional GA-NN model and can give much faster learning speed.