1: \begin{abstract}

2: 

3:     In this work,  we propose a new optimization framework

4:     for multiclass boosting learning.

5:     In the literature,

6:     AdaBoost.MO and AdaBoost.ECC are the two successful multiclass boosting

7:     algorithms, which can use binary weak learners.

8:     We explicitly derive these two algorithms' Lagrange dual problems

9:     based on their regularized loss functions.

10:     We show that the Lagrange dual formulations enable us to

11:     design totally-corrective multiclass

12:     algorithms by using the primal-dual optimization technique.

13:     Experiments on benchmark data sets suggest that

14:     our multiclass boosting

15:     can achieve a comparable generalization capability with state-of-the-art,

16:     but the convergence speed is much faster than stage-wise gradient descent boosting.

17:     In other words, the new totally corrective algorithms

18:     can maximize the margin more aggressively.

19: 

20: 

21: \end{abstract}

22: