∆ -Learning for Coarse-Grained Potentials