An Analysis of Chaotic Noise Injected to Backpropagation Algorithm in Feedforward Neural Network

There have been much interest in applying noise to neural networks in order to observe their effect on network performance. In our previous research, we have proposed a new modified backpropagation learning algorithm, in which chaotic noise is added into weight update process. By computer simulations, we confirmed that the presence of chaotic noise during weight update process in feedforward neural network can give a better convergence rate and can find a good solution in early time. Hence, in this study, we extend these results by introducing other difficult learning example and analyzing the effect of noise parameters such as noise amplitude and control parameter of chaos to the learning performance.