Title: Finite-graph superpopulation inference for random graphs with complex dependence
Authors: Michael Schweinberger - Department of Statistics, Rice University (United States) [presenting]
Abstract: In practice, network scientists are often interested in superpopulation inference for random graphs with complex topological structures. In other words, there is a finite population of nodes and a population graph is generated by a population probability model capturing complex topological structures, including various forms closure in networks. We consider a finite population of nodes partitioned into subpopulations, e.g., armed forces partitioned into units of armed forces or school populations partitioned into school classes. Such data are increasingly widely collected in network science, and are called multilevel network data. We present non-asymptotic concentration and consistency results, assuming the population probability model restricts dependence to subpopulations, the number of subpopulations is large relative to the size of the largest subpopulation, and the subpopulation graphs are governed by curved exponential families with geometrically weighted terms capturing sensible forms of transitive closure.