A practical comparison between Quickprop and back-propagation

A number of studies have questioned the validity of the Quickprop algorithm and the activation function derivative offset as appropriate methods of speeding backpropagation style training. This paper follows on from those results with a practical study using generated data sets. The generation method is briefly presented, along with the results of comparisons which indicate, at least on these data sets, that the performance of Quickprop is not as good as standard back-propagation, and that the activation function derivative offset is detrimental to learning.