Successfully training approximations to full-rank matrices for efficiency in deep learning. - View it on GitHub