Title: Exact Tensor Completion with Sum-of-Squares
Speaker: Aaron Potechin, Princeton University
Date: Wednesday, April 26, 2017 11:00am-12:00pm
Location: CoRE Bldg, Room 301, Rutgers University, Busch Campus, Piscataway, NJ
Abstract:
In the matrix completion problem, we are given some entries of a matrix and we are asked to fill in the remaining entries. A canonical example of this problem is the Netflix challenge, where we are given the ratings of users on some movies and we are asked to predict their ratings on other movies. While the matrix completion problem is impossible to solve in general, it can be solved efficiently if the matrix has the additional structure of being low-rank.
In this talk, I will describe how the nuclear norm minimization method for matrix completion can be viewed in terms of a dual certificate. I will then describe how this view can be generalized to the analogous tensor completion problem via the sum-of-squares hierarchy. No background except for linear algebra will be assumed for this talk.
See: http://www.math.rutgers.edu/~sk1233/theory-seminar/S17/