Abstract
Incorporating latent or hidden variables is a crucial aspect of statistical modeling. I will present a statistical and a computational framework for guaranteed learning of a wide range of latent variable models such as topic models, Gaussian mixtures, hidden Markov models, and community models. It involves decomposition of multivariate moment tensors through linear algebraic and multilinear algebraic operations. I will discuss the deployment of these methods for discovering communities in facebook, yelp and DBLP data. The tensor methods are easily parallelizable and are orders of magnitude faster than the state-of-art stochastic variational approach.