Optimal Training Pattern Selection Using a Cluster-Generating Artificial Neural Network

In this piece of research, the problem of optimal pattern selection for artificial neural network training is investigated. Given an initial set of training patterns, the objective is to extract the minimal subset which accurately represents the initial set.The training pattern selection strategy which is presented here is based on the clustering capability of the Harmony Theory simulated-annealing harmony-maximisation artificial neural network [1]. Correction and reduction of the training set is achieved by rectifying the repetition and misclassification errors as well as by organising and unifying the training patterns in terms of their similarities.