Parallel implementation of the feedforward back-propagation algorithm on pyramid networks

In this paper, we present a parallel implementation of the FeedForward BackPropagation algorithm on pyramids. The proposed algorithm is quite simple and easy to implement. For an ANN of L layers and a maximum of N nodes per layer, our algorithm requires a pyramid of (4N(2)-1)/3 processors to perform backpropagation algorithm in O(LN) time complexity. Note that our solution requires less than the number of nodes required by the best known solution for a mesh of appendixed trees topology of 3N(2) processors. In addition, the algorithm can be adapted to quadtrees with slight modifications. The technique used in our implementation can also be applied to the Hopfield network.