On the Use of Transputers to Implement Neural Networks

Abstract

Neural networks are a class of computational models that has recently raised the interest of the Artificial Intelligence community. One of the drawbacks of neural networks is the slow rate of convergence. This is even more true when we want to use them in significant applications, where the number of units has to be very high. Even the most powerful sequential computers haven’t capacity enough to solve in reasonable time most of real problems. Nevertheless, neural networks inherently posses a high degree of parallelism that allows them to be implemented easily on parallel architectures. In our work we investigated the possibility of mapping a generic neural network on a transputer system.