Using Deep Learning to Correct Theoretically-derived Models