Efficient and Fast Fitting of Block Models

• Activity Number: 50
• Type: Invited
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Aiyou Chen*+ and Arash Amini and Peter Bickel and Liza Levina
• Companies: Google and University of Michigan and University of California at Berkeley and University of Michigan
• Keywords: social network ; spectral clustering ; degree-corrected block models ; pseudo likelihood ; regularization

Abstract

Network community detection has become an interesting topic in statistics recently, and various random graph models such as stochastic block model and its variation have been proposed. Most work in the literature has focused on the case where graphs are relatively dense, and not much work available on the case of relatively sparse graphs. We look into the latter case and propose a couple of techniques useful in general: 1) a pseudo likelihood approach derived from block models, and 2) regularized spectral clustering. A few use cases are shown by simulations and data examples.