Self Organization Map in Real-Number Coordinate Plane

Abstract

SOM(Self-Organizing Map) by T.Kohonen is one of the methods for classifying data of a large number of attributes automatically, where output units are restrictedly placed on the grid points. We propose a real-number SOM(RSOM), where output units are randomly placed on the real-number coordinates and visualized as a Voronoi diagram. Voronoi diagram partitions the plane into many areas that contains the points that are closest to the corresponding output unit. RSOM can visualize more expressively and place output units on curved surfaces easily. To make more understandable diagram, we also propose two methods to move output units to the boundary from the best matching units and a method to remove the units that do not affect the boundary.