Gabriel Isheden: Bayesian Hierarchic Sample Clustering

This thesis seminar presents a novel clustering algorithm called Bayesian Sample Clustering, invented by Timo Koski and Jukka Corander. Bayesian Sample Clustering is a hierarchich clustering algorithm that uses single-linkage clustering on predictive distributions of data samples. The predictive distributions are compared using a metric called the Chan-Darwiche distance.
The first half of the seminar explores and provides a background on metric spaces, clustering hierarchies, the Chan-Darwiche distance, and predictive distributions. Based on the given theoretical background Bayesian Sample Clustering is introduced. The second half of the seminar investigates the algorithm and its applications. The curse of dimensionality is discussed and synthetic as well as real-world examples of Bayesian Sample Clustering are presented.