Early Warning Systems: an approach via Self Organizing Maps with applications to emergent markets
Proceedings of the 2009 conference on New Directions in Neural Networks: 18th Italian Workshop on Neural Networks: WIRN 2008
Hi-index | 0.00 |
Kohonen's self-organizing map (SOM) network isone of the most important network architecturesdeveloped during the 1980's. The main function of SOMnetworks is to map the input data from an n-dimensionalspace to a lower dimensional (usually one or two dimensional) plot while maintaining the originaltopological relations. A well known limitation of theKohonen network is the "boundary effect" of nodes on ornear the edge of the network. The boundary effect isresponsible for retaining the undue influence of initialrandom weights assigned to the nodes of the networkleading to ineffective topological representations. Toovercome this limitation, we introduce and evaluate amodified, "circular" weight adjustment procedure. Thisprocedure is applicable to a class of problems where theactual coordinates of the output map do not need tocorrespond to the original input topology. We tested thecircular method with an example problem from thedomain of group technology, typical of such class ofproblems.