Scalable Estimation of Exponential Random Graph Models on Large Networks

Many scientific fields study large networks with millions of nodes and edges to identify structural and relational patterns. Exponential Random Graph Models (ERGMs) offer a powerful statistical framework for analyzing network formation by estimating the probability distribution over networks based on local structural features and nodal attributes. However, their application to large networks is hindered by the computational challenges inherent in traditional estimation methods. This research introduces a novel 'divide-and-conquer' approach to partition large networks into smaller, overlapping sub-networks. By estimating ERGMs on these sub-networks and then merging the results, this thesis aims to overcome scalability issues while preserving both local and global network structures.