Application of BSP-based computational cost model to predict parallelization efficiency of MLP training algorithm

Abstract

The development of a computational cost model of parallel batch pattern back propagation training algorithm of a multilayer perceptron is presented in this paper. The model is developed using Bulk Synchronous Parallelism approach. The concrete parameters of the computational cost model are obtained. The developed model is used for the theoretical prediction of a parallelization efficiency of the algorithm. The predicted and real parallelization efficiencies are compared for different parallelization scenarios on two parallel high performance systems. © 2010 Springer-Verlag Berlin Heidelberg.