A Bayesian Nonparametric Approach to Multiview Clustering

Multiview data consist of various types of information about the same subjects, and multiview clustering aims to infer separate but interdependent clustering structures across these views. The challenge lies in defining models that can range from completely dependent partitions, where the clusters are identical across views, to independent partitions that treat each view separately. Taking inspiration from a recent work of Dombowsky and Dunson, we introduce a Bayesian nonparametric hierarchical model for multiview data, relying on the Pitman-Yor process. We propose a novel Chinese restaurant metaphor that facilitates the development of a sampling scheme to address Bayesian inference. The performance of our model is tested on different simulated scenarios that illustrate various dependence structures among the view-specific partitions.