Improving the Performance of Multilayer Backpropagation Neural Networks with Adaptive Leaning Rate

Abstract

In this paper a technique is developed to improve performance of multi-layer neural networks in their learning process. Learning rate is one of the important parameters in learning process of neural networks. Inappropriate learning rate reduces the convergence rate. We have developed a technique to adaptively set the learning rate during the learning process based on error curve changes. The results on several datasets indicate superiority of the developed technique in learning process and hence accuracy of the neural network compared to nonadaptive approaches.