Statistics Seminar: STA 290
Thursday, April 5th, 2012 at 4.10pm, MSB 1147 (Colloquium Room)
Speaker: Liza Levina (University of Michigan)
Title: Consistency of community detection and probability models in networks
Abstract: Analysis of networks and in particular discovering communities in networks has been a focus of recent work in several fields, with diverse applications including social networks, food webs, and internet security. Nonetheless, there is a certain amount of disconnect between the many algorithms proposed for community detection, the probability models for random graphs that are simple enough to be tractable, and the complex features we observe in real networks. The talk discusses a number of methods and models for community detection, including the recently proposed degree-corrected block model which allows for more flexible network structures. A general asymptotic framework under this model is derived, which allows us to evaluate and compare many community detection methods in terms of consistency. Various algorithms for fitting the models and empirical results on a number of artificial and real networks will also be discussed.