Improving the Generalization of Supervised Models