1: \begin{abstract}

2: Recent empirical evidence suggests that the Weston-Watkins support vector machine is among the best performing multiclass extensions of the binary SVM.

3: Current state-of-the-art solvers repeatedly solve a particular subproblem approximately using an iterative strategy.

4: In this work, we propose an algorithm that solves the subproblem exactly using a novel reparametrization of the Weston-Watkins dual problem.

5: For linear WW-SVMs, our solver shows significant speed-up over the state-of-the-art solver when the number of classes is large.

6: Our exact subproblem solver also allows us to prove linear convergence of the overall solver.

7: 

8:   % Existing solvers for the associated optimization problem iteratively solves 

9:   % MSVMPack kernel WW, check it out make sure they don't have solver

10: 

11:   % However to date, there exist no solver for the WW SVM with guaranteed convergence.

12: 

13:   % Shark has GD, but not exact

14: 

15: \end{abstract}

16: