Improving the Back-Propagation Algorithm Using Evolutionary Strategy

In order to overcome the existence of the local minimum in the multilayer perceptron (MLP) implemented with back-propagation (BP) algorithm, the evolutionary strategy (ES) is proposed. Introducing the factors of the chromosome and gene mutation rates, one can enhance the flexibility of the mutation. The bounds of the chromosome and gene mutation rates are derived. Simulation results are shown to verify the theoretical calculations and also suggest appropriate strategy parameter values. The theoretical results are studied using the MLP-based decision feedback equalizer (MLP DFE) scenarios. The results indicate that the evolutionary strategy outperforms the BP algorithm