Stochastic Block Models of Mixed Membership

• Activity Number: 124
• Type: Invited
• Date/Time: Monday, July 30, 2007 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Edoardo M. Airoldi*+ and David Blei and Stephen Fienberg and Eric Xing
• Companies: Princeton University and Princeton University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: Hierarchical Bayes ; Latent variables ; Mean-field approximation ; Statistical network analysis ; Protein interaction networks ; Social networks

Abstract

Observations consisting of measurements on pairs of objects arise in a variety of biological settings, with collections of author-recipient email, and in social networks. Analyses of such data typically aim at clustering the objects of study, or situating them in a low dimensional space, and at estimating relational structures among the clusters. For example, given protein interaction networks we want to estimate the memberships of individual proteins to stable protein complexes (i.e., clusters of proteins), how stable protein complexes interact with one another, and how many there are. In this talk we introduce stochastic block models of mixed membership, which support such integrated data analyses within a hierarchical Bayesian framework. A variational scheme for fast, approximate inference is presented. The methodology is demonstrated on social and protein interaction networks.