Speaker：Guoxin Zuo (Central China Normal University)

Time：2023-11-29 10:00

Location：Conference Room 686 at the 6th floor of Shuli Building at Haiyun Campus

Abstract:

The prior likelihood of generalized Bayes framework for probabilistic clustering through general discrete random structure is developed, so that the framework can contain the background of both K-fixed and K-adaptive clustering methods, such as K-means and Gibbs partition. The relative derivations including the classifier, posterior inference and detailed framework adjustments are deduced. A modified algorithm combining the ideas of K-dissimilarity algorithm and reversible jump Markov chain Monte Carlo method is proposed to search for global optimization. By iterations on re-clustering through Gibbs sampling method, the stationary distribution of clustering likelihood is obtained for uncertainty quantification, where the choice of K can be controlled by QAIC and parametric setting in different K adaptive methods. Finally, it can be shown that the clustering results in application excellently divide the samples into groups with the satisfying number of categories and high accuracy.