19: low rank approximations motivation: main use of SVD in practice is to approx data by a low rank matrix