Behavior of Fatigable SOM and its Application to Clustering

The self-organizing map (SOM) is popular algorithm for unsupervised learning and visualization introduced by Teuvo Kohonen. One of the most attractive applications of SOM is clustering and several algorithms for various kinds of clustering problems have been reported and investigated. In this study, we propose a new type of SOM algorithm, which is called Fatigable SOM (FSOM) algorithm. The important feature of FSOM is that the neurons are fatigable, namely, the neurons which have become a winner can not become a winner during a certain period of time. Because of this feature, FSOM tends to self-organize only in the area where input data are concentrated. We investigate the behavior of FSOM and apply FSOM to clustering problems. Further, we introduce the fatigue level to FSOM to increase its flexibility for various kinds of clustering problems. The efficiencies of FSOM and the fatigue level are confirmed by several simulation results.