Supporting GNG-based clustering with local input space histograms.
Jochen Kerdels, Gabriele Peters · 2014
Abstract. This paper presents an extension to the growing neural gas (GNG) algorithm that allows to capture local characteristics of the input space. Using these characteristics clustering schemes based on the GNG network can be improved by discarding uncertain edges of the network and identifying edges that span discontinuous regions of input space. We ap-plied the described approach to different two-dimensional data sets found in the literature and obtained comparable results. 1