This talk has moved to Spring 2017 — On the Power of Truncated SVD for General High-rank Matrix Estimation Problems.