A Virtual Software Environment for Artificial Neural Networks on General Purpose Parallel Architectures

Abstract

Parallel hardware is an essential requirement for the implementation and study of true-size and real-time neural network applications. General purpose parallel machines, although expected to be slower than special purpose neurocomputers, are generally much more commonly available. For this reason, they present an interesting alternative for supporting neural computations. In the present work we propose a software environment for neural network computing on general purpose parallel machines. A description of the general environment is presented, based on concepts of the GALATEA neurocomputing project. Furthermore, a specific implementation on the PADMAVATI machine is discussed in some details.