Efficiency research of batch and single pattern MLP parallel training algorithms

The development of parallel algorithms for batch and single pattern back propagation training of a multilayer perceptron and the research of their efficiency on a general-purpose parallel computer are presented in this paper. The multilayer perceptron model and the sequential batch and single pattern training algorithms are theoretically described. An algorithmic description of the parallel versions of the batch and single pattern training methods are specified. The efficiencies of the developed parallel algorithms are investigated by progressively increasing the dimension of the parallelized problem on a general-purpose parallel computer NEC TX-7.