Constrained Clustering using Gaussian Processes

Panagiotis A. Traganitis, Georgios B. Giannakis

Abstract: Constrained clustering is an important machine learning, signal processing and data mining tool, for discovering clusters in data, in the presence of additional domain information. The present work introduces a probabilistic scheme for constrained clustering based on the popular Gaussian Process framework. The proposed scheme accommodates pairwise, must-and cannot-link constraints between data, does not require hyperparameter tuning, and enables assessment of the reliability of obtained results. Preliminary results on real data showcase the potential of the proposed approach.