Clustering spherical shells by a mini-max information algorithm

Abstract

We focus on spherical shells clustering by a mini-max information (MMI) clustering algorithm based on mini-max optimization of mutual information (MI). The minimization optimization leads to a mass constrained deterministic annealing (DA) approach, which is independent of the choice of the initial data configuration and has the ability to avoid poor local optima. The maximization optimization provides a robust estimation of probability soft margin to phase out outliers. Furthermore, a novel cluster validity criteria is estimated to determine an optimal cluster number of spherical shells for a given set of data. The effectiveness of MMI algorithm for clustering spherical shells is demonstrated by experimental results. © Springer-Verlag Berlin Heidelberg 2006.