We consider the LPBoost family of boosting algorithms for the \(\ell _1\) -norm regularized soft margin optimization, where the problem instance is implicitly given as a huge scale LP problem and the goal is to efficiently find an optimal solution with the aid of a certain oracle. Although the optimal solution yields a linear classifier with good generalization ability, the LPBoost family is less popular, since all existing algorithms in the family are either very slow on real data or have no theoretical convergence guarantees. In this chapter, we first show that each algorithm in the LPBoost family can be viewed as an instance of a general scheme for solving convex optimization known as the Frank-Wolfe method, and thus its convergence guarantee can be immediately applied. The Frank-Wolfe method has the feature that the convergence guarantee remains to hold even when arbitrary update rules including the standard one are greedily applied in each iteration. Taking advantage of the feature, we propose an algorithm that performs efficiently on real data while maintaining the same convergence guarantees as the existing methods.