Seminar Abstract

Monday, December 01, 2008 • 12:15 PM • Medium Conference Room, SFI

Benjamin P. Olding Harvard University

A Selected Survey of Inference Procedures for Graph and Network Data

Past decades have witnessed a variety of advances in the treatment of graphs and networks as combinatoric or algebraic objects. However, while many of the resultant tools have been adopted by practitioners in the service of inference, the community as a whole still lacks a unifying framework for the statistical analysis of network data; indeed, many analogs to classical statistics in the context of network inference are not yet well established, even for apparently simple cases. Here we illustrate this point by considering the seemingly straightforward problem of detecting network structure by way of hypothesis testing, for which we invoke a generalized likelihood ratio framework. We demonstrate how stochastic computation, by way of importance sampling, can be applied to estimate the statistical power of such tests, and detail its computational complexity. This work is joint with P. J. Wolfe at Harvard (sisl.seas.harvard.edu).

Host: Nathan Eagle