Effect of chaos noise on the learning ability of back propagation algorithm in feed forward neural network

Abstract

In the area of artificial neural networks, the Back Propagation (BP) learning algorithm has proved to be efficient in many engineering applications especially in pattern recognition, signal processing and system control. Although the BP learning has been a significant research area of neural network, it has also been known as an algorithm with a poor convergence rate. Many attempts have been made on the learning algorithm to improve the performance on convergence speed and learning efficiency. In this study, we propose a new modified BP learning algorithm by adding chaotic noise into weight update process during error propagation. The chaotic noise is generated using various chaotic maps such as Logistic map, Skew Tent map and Bernoulli Shift map. By computer simulations, we confirm that our proposed algorithm can give a better convergence rate and can find a good solution in early time compared to the conventional BP learning algorithm. Weight update position, noise amplitude and control parameter of chaos can give a big effect on the learning ability of feed forward neural network.