An efficient mapping of multilayer perceptron with backpropagation ANNs on hypercubes

This paper proposes a parallel structure, the mesh-of-appendixed-trees (MAT), for efficient implementation of artificial neural networks (ANNs). Algorithms to implement both the recall and the training phases of the multilayer perceptron and backpropagation ANN model are provided. A recursive procedure for embedding the MAT structure into the hypercube topology is used as the basis for an efficient mapping technique to map ANN computations on general purpose massively parallel hypercube systems. In addition, based on the mapping scheme, a fast special purpose parallel architecture for ANNs is developed. The major advantage of our technique is high performance. Unlike the other techniques presented in the literature which require O(N) time, where N is the size of the largest layer, our implementation requires only O(log N) time. Moreover, it allows the pipelining of more than one input pattern and thus further improves the performance