General purpose MIMD computers and neural networks: three case studies

The paper presents three simple applications of general purpose MIMD parallel computers to enhance the performance of neural network systems. The first two approaches aim at speeding up the training of multilayer perceptrons through competition of neural networks with different parameters or distribution of patterns among processing elements. The third approach performs a rather "brute force" search of optimal parameters for the energy function of Hopfield neural networks applied to optimization problems. The effectiveness of the proposed methods is demonstrated through experiments with a binary tree and a mesh parallel computers.