Ensemble of structure-adaptive self-organizing maps for high performance classification

Combining multiple models has been recently exploited for the development of reliable neural networks. This paper introduces a structure-adaptive self-organizing map (SOM) which can adapt the structure as well as the weights, and presents a method to improve the performance by combining the multiple maps. The structure-adaptive SOM places the nodes of prototype vectors into the pattern space properly so as to make the decision boundaries as close to the class boundaries as possible. In order to show the performance of the proposed method, experiments with the unconstrained handwritten digit database of Concordia University in Canada have been conducted. Ó 2000 Elsevier Science Inc. All rights reserved.