Skip to main content
Browse by:
GROUP

Hypothesis Testing and Community Detection

Event Image
Wednesday, September 03, 2014
3:30 pm - 4:30 pm
Andrew Nobel Department of Statistics and Operations Research UNC Chapel Hill
Machine Learning Seminar

12noon lunch and lecture ¿Perspectives in Machine Learning¿3:30pm seminar and receptionCommunity detection is a central and well studied problem in the field of network analysis, with applications to the analysis of social, political, and biological networks. Broadly speaking, the goal of community detection is to divide the vertices of an observed network into groups (usually disjoint), called communities, in such a way that the edge density within groups is substantially higher than the edge density between groups. In this talk I will describe an extraction based procedure for community detection that is based on statistical ideas from the theory of multiple testing, in particular, p-values derived from a conditional configuration model. The resulting procedure is able to handle overlapping communities and, importantly, to identify background vertices that do not belong to any community.Comparisons to several existing community detection procedures on real and simulated data will be presented. As time permits, I will say a few words about ongoing theoretical analysis of the procedure.Andrew Nobel received his PhD in Electrical Engineering from Stanford University in 1992, under the direction of Professors Amir Dembo and Thomas Cover. From 1992-1995 he was a Postdoctoral Fellow at the Beckman Institute, U of Illinois, Urbana-Champaign.

Type: LECTURE/TALK
Contact: Sayan Mukherjee