A new visualization scheme for self-organizing neural networks
Cristián J. Figueroa, P. A. Estévez · 2005
A new on-line visualization scheme for self-organizing neural networks is presented. The proposed updating rule for position vectors is applied to the Kohonen's SOM, the neural gas (NG) and the growing neural gas (GNG) neural networks, to create the enhanced versions TOPSOM, TOPNG and TOPGNG, respectively. The proposed models are tested on the visualization of benchmark and real-world datasets, and compared with DIPOL-SOM, as well as the off-line combination of SOM, NG and GNG with the Sammon's non-linear mapping. The results obtained with TOPSOM and TOPNG are better than that of DIPOL-SOM, and similar to those obtained with off-line strategies, in terms of distance and topology preservation measures.