Network Seminar on "Reducing degeneracy in maximum entropy models of networks" by Szabolcs Horváth

Dr. Szabolcs Horváth (SBRI-Inserm Lyon)

 

Abstract: Exponential Random Graph (ERG) models are maximum entropy models of graphs, and can describe network ensembles that satisfy given constraints.  They are the graph theory analog of the canonical ensemble of statistical mechanics.  ERG models are used to predict network properties based on limited available information, and are popular in social network analysis. However, their use is often hindered by the so-called degeneracy problem characterized by spontaneous symmetry breaking, where predictions fail.  After a brief introduction to ERG models, we shall look at when and why the degeneracy problem arises, and offer a practical recipe to avoid it. Finally we look at how degeneracy can appear in non-network related maximum entropy models as well.