The self-organising map learning algorithm with inactive and relative winning frequency of active neurons
Vikas Chaudhary, R. S. Bhatia, Anil Ahlawat · HKIE Transactions · 2014
The self-organizing map (SOM) is applied to data clustering, image analysis, dimension reduction and so forth. During the study of conventional SOM, we identified active and inactive neurons on the map. But the conventional SOM treats all the neurons on the map as active neurons and does learning accordingly. Also, the conventional SOM does not consider the relative winning frequency of active neurons. In this paper, a modified SOM is being proposed, which divides the neurons into active and inactive groups and different learning processes are used for active and inactive neurons. The modified SOM considers the relative winning frequency of active neurons also. The learning efficiency is measured using three well-known parameters. The modified SOM is applied on various standard input data sets. The learning results of the modified SOM are better than that of the conventional SOM. It is observed that the modified SOM is able to obtain the effective map reflecting the distribution of the input data.