High-performance simulation of neural networks

Abstract

Artificial neural networks have been used for a wide range of problems in a variety of areas. The back-propagation algorithm is frequently used to train the network, but is time consuming when implemented on generalpurpose computers. This paper examines methods of simulating back-propagation neural networks on parallel systems to achieve high performance. The training of artificial neural networks consists of updating the weights in several nested loops. Parallel simulation methods may be classified based on which of these loops are executed in parallel. These methods are discussed and example implementations of these methods are described.