The effect of internal parameters and geometry on the performance of back-propagation neural networks: an empirical study

The results obtained indicate that learning rate, momentum, the gain of the transfer function, epoch size and network geometry have a significant impact on training speed, but not on generalisation ability.