Topological maps for mixture densities

Abstract

We propose in this paper a new learning algorithm which uses a probabilistic formalism of topological maps. This algorithm approximates the density distribution of the input set with a mixture of normal distributions. We show that, under certain conditions, the classical Kohonen algorithm is a specific case of this algorithm, therefore allowing a probabilistic interpretation of Kohonen maps.