Learning by Minimizing the Sum of Ranked Range