Using regularization and second order information in outer approximation for convex MINLP