1: \begin{abstract}

2: %\boldmath

3: Boosting    is a learning scheme that combines weak prediction rules

4: to produce a strong composite estimator, with the underlying

5: intuition that one can obtain accurate prediction rules by combining

6: ``rough'' ones. Although boosting is proved to be  consistent and

7: overfitting-resistant, its numerical convergence rate

8:   is relatively slow. The aim of this paper is to develop a new

9: boosting strategy, called the re-scale boosting (RBoosting), to

10: accelerate the numerical convergence rate and, consequently, improve

11: the learning performance of boosting. Our studies show that

12: RBoosting possesses the almost optimal numerical convergence rate in

13: the sense that, up to a logarithmic factor, it can reach the minimax

14: nonlinear approximation rate. We then use RBoosting to tackle both

15: the  classification and regression  problems,   and deduce a tight

16: generalization error estimate. The theoretical and experimental

17: results show that RBoosting outperforms boosting in terms of

18: generalization.

19: 

20: 

21: \end{abstract}

22: