Title: Spectral inference for large stochastic blockmodels with nodal covariates
Authors: Angelo Mele - Johns Hopkins University (United States) [presenting]
Joshua Cape - University of Michigan (United States)
Carey Priebe - Johns Hopkins University (United States)
Lingxin Hao - Johns Hopkins University (United States)
Abstract: Spectral methods are studied for inference in large stochastic blockmodels with observed nodal covariates. We formulate the estimation problem as recovery of latent positions in the Generalized Random Dot Product Graph (GRDPG) model, thereby extending recent advances in spectral methods to provide an algorithm that simultaneously estimates the block assignments and parameters for observed covariates. The spectral estimator is asymptotically normal and computationally fast, when compared to a standard variational EM algorithm. The results provide a foundation to estimate the effect of observed covariates as well as unobserved latent community structure on the probability of link formation in massive networks.