## Polynomial-time tensor decompositions via sum-of-squares

Tengyu Ma

Princeton University

March 21, 2016

Tensor decompositions have been the key algorithmic components in provable learning of a wide range of hidden variable models such as topic models, Gaussian mixture models, independent component analysis, dictionary learning. Despite its success, one of the challenges in this area is to decompose over-complete 3rd-order tensors robustly.