Empirical comparison between two computational strategies for topological self-organization

Abstract

The standard Self-Organizing Map (SOM) achieves self organization through incremental, stepwise update of its weight vectors. However, beginning with a randomly initialized map, the convergence speed of SOM is rather slow. It would be desirable to speed up the self-organization process. For this reason, a new strategy for topological self-organization (Xu, 2002) has been proposed to speed up the self-organization process. This paper empirically compares the two strategies. The new strategy is shown to speed up the self-organization significantly. © Springer-Verlag 2003.