Factors influencing the choice of a learning rate for a backpropagation neural network

Neural networks have been used effectively in a number of applications. Most of these applications have used the backpropagation algorithm as the learning algorithm. One of the major problems with this algorithm is that its convergence time is usually very long since the training set must be presented many times to the network. The learning rate has to be selected very carefully. If the learning rate is too low, the network will take longer to converge. On the other hand, if the learning rate is too high, the network may be unstable and may never converge. Up to now, designers of neural network applications had to find an appropriate learning rate for their systems by trial and error. In this paper, a new method to compute dynamically the optimal learning rate is proposed. By using this method, a range of good static learning rates could be found. Furthermore, some studies have been done on factors influencing learning rates