Spherical Tree Structured Self-Organizing Map

Abstract

∗In order to speed up the search for winner nodes in a SelfOrganizing Map (SOM), Tree Structured SOM(TS-SOM), which applies the tree search method to the SOM, was proposed. Since TS-SOM has the edges of the map like the SOM, the learning is biased depending on the position of the winner node. Therefore, in this paper, in order to eliminate the edges of TS-SOM’s map, we propose Spherical TS-SOM(S-TS-SOM) in which nodes are placed on spheres and the tree search method is applied. Also, we evaluate the performance of S-TS-SOM.