Based on these bounds, we further propose a new Upper Bound based Greedy (UBG in short) algorithm which can significantly reduce the number of estimations in greedy-based algorithms, especially at the initial step.