Parallelization in combining the SOM and Sammon's mapping

In this paper, we propose a parallel algorithm for multidimensional data visualization combining the neural network (the self-organizing map-SOM) and Sammon’s mapping. Here n-dimensional vectors are projected onto the plane by using Sammon’s mapping taking into account the learning flow of the SOM....

Full description

Bibliographic Details
Main Authors: Gintautas Dzemyda, Olga Kurasova, Virginijus Marcinkevičius
Format: Article
Language:English
Published: Vilnius University Press 2003-12-01
Series:Lietuvos Matematikos Rinkinys
Online Access:https://www.journals.vu.lt/LMR/article/view/32403
Description
Summary:In this paper, we propose a parallel algorithm for multidimensional data visualization combining the neural network (the self-organizing map-SOM) and Sammon’s mapping. Here n-dimensional vectors are projected onto the plane by using Sammon’s mapping taking into account the learning flow of the SOM. It is necessary to investigate some important factors that influence the efficiency of the parallel algorithm. The results of investigation allow us to optimize the number of the SOM training epochs, the number of the SOM training blocks, and the number of Sammon’s iterations.
ISSN:0132-2818
2335-898X