Parallel Implementation of Backpropagation on Master Slave Architecture

Back propagation is one of the simplest and most widely used methods for supervised training of multi layer neural networks, which is an extension to LMS (least mean square) algorithm for linear systems. In this paper we present parallel implementation of multiplayer perceptron (MLP) networks using backpropagation on master-slave architecture. The performance parameters speed-up, optimal number of processors and processing time are evaluated for both sequential implementation and parallel implementation. A standard XOR problem is solved by using both parallel and sequential implementations. Analytical and experimental results are also presented.