Tight convex relaxations for sparse matrix factorization