Interacting individually and collectively treated neurons for improved visualization

Abstract

In this paper, we propose a new type of learning method in which neurons are treated individually and collectively. In addition, the collectivity is defined in terms of distance and similarity between neurons. We applied the method to the self-organizing maps, because our method makes it possible to control flexibly a process of cooperation between neurons. Then, we applied the method with the self-organizing maps to the visualization of the pound-yen exchange rates. We succeeded in producing clearer class structure. The entire period of the exchange rates was divided into three distinct periods. © Springer-Verlag Berlin Heidelberg 2013.