Learning sketches for low-rank approximation is a problem that many people have been interested in. This paper makes a nice original contribution to the problem. The worst-case bound is interesting, and the empirical results are compelling. I encourage the authors to address reviewer comments in the paper.