The Geodesic Self-Organizing Map and Its Error Analysis

Wu, Y. and Takatsuka, M.

The Self-Organizing Map (SOM) is one of the popular Artificial Neural Networks which is a useful in clustering and visualizing complex high dimensional data. Conventional SOMs are based on the two-dimensional (2D) grid structure, which usually results in less accurate representation of the data. Several SOMs using spherical data structures have been proposed to remove the 'border effect'. In this paper, we compared our proposed Geodesic SOM (GeoSOM) with the 2D hexagonal SOM by experiments. The result shows that the GeoSOM not only runs as fast as the conventional 2D SOM, but also represents the data more accurately within fewer training epochs.

Cite as: Wu, Y. and Takatsuka, M. (2005). The Geodesic Self-Organizing Map and Its Error Analysis. In Proc. Twenty-Eighth Australasian Computer Science Conference (ACSC2005), Newcastle, Australia. CRPIT, 38. Estivill-Castro, V., Ed. ACS. 343-352.

(from crpit.com) (local if available)