An evolutionary algorithm for parallel mapping of backpropagation learningon heterogeneous processor networks

This paper presents an evolutionary algorithm for optimal mapping of the ba ck-propagation learning algorithm onto a parallel heterogeneous processor n etwork. Training-set parallelism is used as the paradigm for parallelizing the backpropagation algorithm, and the processor network is a heterogeneous array of transputers connected in a pipelined ring topology. It is known f rom earlier studies that finding the optimal mapping (i.e. optimal allocati on of training patterns among the processors to minimize the time for a tra ining epoch) involved solving a linear Mixed Integer Programming (MIP) prob lem. Solving the MIP using the traditional Branch and Bound (B&B) method ta kes a large amount of computing time. Approaches based on evolutionary algo rithms are then investigated as alternatives to the branch and bound method to solve the pattern allocation problem. It is found that a conventional g enetic algorithms (GAs) search time taken. However when the crossover and m utation probabilities in the GA are varied over a wide range, the best solu tion is obtained by an evolutionary algorithm even though the studies were begun with a conventional GA. A new stopping criterion to detect convergenc e to stop the search is also incorporated in the final algorithm.