TRAINING FEED-FORWARD ARTIFICIAL NEURAL NETWORKS FOR PATTERN-CLASSIFICATION USING THE HARMONY SEARCH ALGORITHM

The Harmony Search algorithm is relatively a young stochastic meta-heuristic that was inspired from the improvisation process of musicians. HS has been successfully applied as an optimization method in many scientific and engineering fields and was reported to be competitive alternative to many rivals. In this work a new framework is presented for adapting the HS algorithm as a method for the supervised training of feed-forward artificial neural networks with fixed architectures. Implementation considers a number of pattern classification benchmarking problems and comparisons are made against the traditional Back Propagation training method and an evolutionary based genetic algorithm training method. Results show that the proposed Harmony Search based method has attained results that are on par or better than those of Back Propagation and Genetic Algorithm. However BP seems to have better fine-tuning capabilities than the proposed HS-based method but might take longer overall training time.