Community Detection Using Gaussian Mixture Models

Community detection is a common theme across many fields. In particular, a lot of work has focused on recoverability guarantees for the stochastic block model. Recent work has focused on modeling the spectral embedding of a network using Gaussian mixture models (GMMs). This work focuses on regimes where ability to recover community labels improves as the network grows, however this is not very realistic. We present methodology for the non-vanishing noise regime by using shrinkage and truncation to use GMMs.