Enhancing clustering performance of feature maps using randomness

This paper presents an enhancement made to a high dimensional variant of a growing self organizing map called the High Dimensional Growing Self Organizing Map (HDGSOM) that enhances the clustering of the algorithm. The enhancement is based on randomness that expedites the self organizing process by moving the inputs out from local minima producing better clusters within a shorter training time. The enhancement is described in detail and several experiments on very large text datasets illustrating the effect of the enhancement are also presented.