Multilayer Artificial Neural Network Design and Architecture Optimization for the Pattern Recognition and Prediction of EEG Signals Based on Hénon Map Chaotic System

Abstract

This paper investigates the training performances of multilayer artificial neural network (ANN) architectures for the implementation of chaotic systems. The designed ANN models can be employed for pattern recognition and prediction of dynamic chaotic systems, in order to simulate and analyze brain activities captured by Electroencephalogram (EEG). Previous research shows that EEG signals demonstrate chaotic features. Chaotic systems can be represented by a set of mathematical equations, which can be used to generate the target outputs for training ANN. In this research, the Henon map is selected as an example for ANN-based chaotic system design. The optimization of ANN architecture is important for improving the performance of hardware implementation. ANN architectures with up to 3 hidden layers combined with different number of hidden neurons are compared by measuring the training performance using the mean square errors (MSE). The ANN training are carried out using three training algorithms: Levenberg-Marquardt, Bayesian Regulation and Scaled Conjugated Gradient. Nonlinear autoregressive (NAR) model is used for ANN architectures design. The training results demonstrate that the training performance can not be improved simply by increasing the complexity of the ANN architecture in terms of the number of hidden layers and hidden neurons. It is therefore necessary to optimize the ANN architecture on a case-by-case basis in order to improve the efficiency of the ANN implementation for specified applications.