Supervised, edge adaptive maps

Abstract

This paper describes a supervised, edge adaptive map. The map is defined similar to a SOM consisting of neurons connected by edges. In contradiction to a SOM it is not the positions of the neurons which are adapted, rather it is the distance between them. Thus, after the learning process the neurons represent a mapping of the input data while the neurons are soleley trained using the distances between the input data points. A supervised, edge adaptive map can be used to learn lower dimensional representations of high dimensional input data. Furthermore, it can be applied to determine the positions of landmarks when only the distances between the landmarks are known.