Researchers from the University of Washington and Google have Developed Distilling Step-by-Step Technology to Train a Dedicated Small Machine Learning Model with Less Data