Scaling Up Inductive Logic Programming by Learning from Interpretations