Furong Huang, Candidate, UC Irvine MLconf NYC

Discovery of Latent Factors in Highdimensional Data Using Tensor Methods: Learning latent variable mixture models in highdimension is applicable in numerous domains where lowdimensional latent factors out of the highdimensional observations are desired. Popular likelihood based methods optimize over a nonconvex likelihood which is computationally challenging due to the highdimensionality of the data, and it is usually not guaranteed to converge to a global or even local optima without additional assum