Metaheuristic Optimization Algorithms for Training Artificial Neural Networks

Abstract

Training neural networks is a complex task that is important for supervised learning. A few metaheuristic optimization techniques have been applied to increase the effectiveness of the training process. The Cuckoo Search (CS) algorithm is a recently developed meta-heuristic optimization algorithm which is suitable for solving optimization problems. In this paper, Cuckoo search is implemented in training a feed forward multilayer Perceptron network (MLP). We then evaluate the trained MLP‟s accuracy by applying four benchmark classification problems. Furthermore, the results obtained are compared to those attained using another competing meta-heuristic which is the Particle Swarm Optimization (PSO). Also, Guaranteed Convergence Particle Swarm Optimization (GCPSO) which is a PSO variant is implemented and its results are compared with CS and PSO. CS proved to be superior to PSO and GCPSO in all benchmark problems.