Title: Higher-order accurate inference for network moments
Authors: Yuan Zhang - Ohio State University (United States) [presenting]
Abstract: The aim is to derive and use high-order expansions of network moment statistics for exchangeable network models, including the popular stochastic block model for one- and two-sample network inferences, under very mild assumptions. By this approach, we can achieve the following two goals simultaneously (i) higher-order control of the type I error; and (ii) rate-optimal separation condition on the alternative hypothesis for the test to be consistent. Notice that goal (i) was previously only achieved by computationally expensive bootstrap methods with no power guarantees. We also demonstrate our approach's effectiveness in numerical examples.