Artificial neural networks implementation on vectorial supercomputers

We present the results from the implementation of a multilayer perceptron with the backpropagation algorithm on a FUJITSU VP-2400/10 vectorial supercomputer. The programming methodology employed, tries to obtain the maximum performance in this kind of machines based on: input/output structures, treatment of conditional statements, do-loops vectorization and compiler directives. In this work, computing times for the VU (vectorial unit) and CPU are presented. These times indicate the possibility of real time operation in applications which demand artificial neural networks with a high structural complexity.<
>