Some Basic Issues in Network Models and Fitting

Peter Bickel – University of California, Berkeley

Abstract

After giving some broader examples, we briefly review Aldous-Hoover and Stochastic Blockmodels for undirected graphs without covariates and some structural features. We discuss some strengths and weaknesses of such models, as well as ML and spectral types of methods of fitting and their strengths and weaknesses, statistical and computational. We introduce hierarchical block models, the properties, computational and statistical of an old fitting principle and a real world application of this model. We end with a general discussion.