Griffiths’ Variable Learning Rate Online Sequential Learning Algorithm for Feed-Forward Neural Networks

For online sequential training of deep neural networks, where the training data set is chaotic in nature, it becomes quite challenging for choosing a proper learning rate. This paper presents Griffiths’ variable learning rate algorithm for improved performance of online sequential learning of feed-forward neural networks used for chaotic time-series prediction. Here, the learning rate is varied based on Griffiths’ cross-correlation between input training data and squared error, which facilitates better tracking of time-series data.