An improved probability-one homotopy map for tracking constrained clustering solutions

Layne T. Watson, Naren Ramakrishnan

Abstract

This paper proposes a new homotopy map for use with constrained clustering problems, improving over previously introduced maps through the introduction of a K-Means approximation and the use of the Kreisselmeier-Steinhauser function to provide more efficient computation, as well as demonstrating through experimentation the power of this new map.