Abstract
Kohonen's Self-Organizing Map (SOM) is a kind of neural networks that learns the feature of multi input data without supervision. Therefore, it seems difficult to apply the SOM, in the case that the feature of input data changes from learning example data. In this paper, the authors proposed that "Automatic Incremental Learning System of Self-Organizing Map" is effective for its solution. This system consists of two techniques. The one works automatic learning of the input data feature, when the input data has partly been shifted from features of learning example, and the other is "semi-automatic labeling" on new SOM feature map that adjust territory labels by using information from the former map.
