Improvement method for visual stability of SOM feature map by initial value assignment

MOMOI Shinji
MIYOSHI Tsutomu

Abstract: The location of the node or the distance between nodes on SOM feature 
map is important factor to determine feature of individual data. In many 
applications the judgment is made by a person depending on a location on the 
feature map. In conventional method, initial value of feature map has been 
decided at random, so it can be said that it lacks visual stability. In this 
paper, we focused on visual stability of SOM feature map, and we proposed new 
initialization method with less processing or less analysis of leaning data. By 
two types of experiments, we revealed that, proposed method is visually stable 
than conventional method in the point of feature map location, the 
computational complexity, and the standard deviation of winning node location.