An Overview of Mixed Membership Models: Some History and a General Formulation

• Activity Number: 54
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Elena Erosheva* and Steven Fienberg
• Companies: University of Washington and Carnegie Mellon University
• Keywords: Bayesian analysis ; Hierarchical models ; Grade of membership models

Abstract

Mixed membership models have emerged over the past 20 years as a flexible cluster-like modeling tool for unsupervised analyses of high-dimensional multivariate data where the assumption that each unit belongs to one and only one group or cluster no longer holds. We give a brief history of examples of mixed membership arising in diverse application domains, and then lay out a general framework for such models and illustrate how it relates to specific applications including the study of disability via repeated surveys, population genetics, and the analysis of text data.