One of the most important goals of unsupervised learning is to discover meaningful clusters in data. Clustering algorithms strive to discover groups, or clusters, of data points which belong together because they are in some way similar. The research presented in this thesis focuses on using Bayesian statistical techniques to cluster data. We take a model-based Bayesian approach to defining a c...

Decision tree based context clustering [Young; '94] ・ Construct a parameter tying structure ・ Can estimate robust parameter ・ Can generate unseen context dependent models ・ Minimum description length (MDL) criterion [Shinoda; '97] Bayesian approach ・ Variational Bayesian (VB) method [Attias; '99] ⇒ Applied to speech recognition [Watanabe; '04] ・ Can use prior information ⇒ Affect context cluste...

Clustering for the analysis of the gene expression profiles has been used for identifying the functions of the genes and of unknown genes. Since the genes usually belong to multiple functional families, fuzzy clustering methods are more appropriate than the conventional hard clustering methods. However, it is still required to devise natural way to measure the quality of the cluster partitions ...

We present two new algorithms for fast Bayesian Hierarchical Clustering on large data sets. Bayesian Hierarchical Clustering (BHC) [1] is a method for agglomerative hierarchical clustering based on evaluating marginal likelihoods of a probabilistic model. BHC has several advantages over traditional distancebased agglomerative clustering algorithms. It defines a probabilistic model of the data a...

In this work, we propose a novel nonparametric Bayesian method for clustering of data with spatial interdependencies. Specifically, we devise a novel normalized Gamma process, regulated by a simplified (pointwise) Markov random field (Gibbsian) distribution with a countably infinite number of states. As a result of its construction, the proposed model allows for introducing spatial dependencies...

Co-clustering has emerged as an important technique for mining contingency data matrices. However, almost all existing coclustering algorithms are hard partitioning, assigning each row and column of the data matrix to one cluster. Recently a Bayesian co-clustering approach has been proposed which allows a probability distribution membership in row and column clusters. The approach uses variatio...

We have proposed a novel speaker clustering method based on a hierarchically structured utterance-oriented Dirichlet process mixture model. In the proposed method, the number of speakers can be determined from the given data using a nonparametric Bayesian manner and intra-speaker variability is successfully handled by multi-scale mixture modeling. Experimental result showed that the proposed me...