Efficient neural network training algorithm for the Cray Y-MP supercomputer

An efficient implementation of a quasi-Newton algorithm for feedforward neural network training on a Cray Y-MP is presented. The most time-consuming step of a neural network training using the quasi-Newton algorithm is the computation of the error function and its gradient. We describe in this paper how this step can be implemented so that the neural network training may take full advantage of the Cray vectorization capabilities.