Parallelization of neural network decoding algorithm for IFS codes on a hypercube

This paper provides the parallelization of neural network algorithm on a hypercube to speed-up the execution time for generating images by using IFS (iterated function systems) codes. We parallelize the sequential neural network by two stages: the decomposition of neural network and the communication of neural states. At the stage of neural network decomposition, the neural network is decomposed into nodes of hypercube by the neurons of the output layer. For the communication of neural states, the gray-scales of neurons in the output layer at each iteration are updated and are communicated to other nodes which have these neurons as the input layer in the next iteration. This technique can also be applied to the fractal image compression using neural networks.