General properties of the generalised Lotto-type competitive learning
Anson M. Y. Luk, Sandra Lien · 2002
We highlight the difficulty of detecting small and rare clusters. Although in theory it is possible for the algorithms to follow the source density function, we note that in experiment it is very difficult to locate these smaller clusters, especially if the prototype set is limited. We propose to train the network iteratively with the same prototype set. This simple method enable one to locate these smaller clusters, albeit with the possibility of over-training.