Unsupervised learning for hierarchical clustering using statistical information

This paper proposes a novel hierarchical clustering method that can classify given data without specified knowledge of the number of classes. In this method, at each node of a hierarchical classification tree, log-linearized Gaussian mixture networks [2] are utilized as classifiers to divide data into two subclasses based on statistical information, which are then classified into secondary subclasses and so on. Also, unnecessary structure of the tree can be avoided by training in a cross-validation manner. Validity of the proposed method is demonstrated with classification experiments on artificial data.