1: \begin{abstract}

2: The hard margin loss function has been at the core of the support vector machine (SVM)

3: research from the very beginning due to its generalization capability.

4: On the other hand, the cardinality constraint has been widely used for feature selection, leading to sparse solutions. This paper studies the sparse SVM with the hard-margin loss (SSVM-HM) that integrates the virtues of both worlds.

5: However, SSVM-HM is one of the most challenging models to solve. In this paper,

6: we cast the problem as a composite optimization with the cardinality constraint.

7: We characterize its local minimizers in terms of {\rm P}-stationarity that well captures the combinatorial 

8: structure of the problem. 

9: We then propose an inexact proximal augmented Lagrangian method (iPAL).

10: The different parts of the inexactness measurements from the {\rm P}-stationarity are controlled at

11: different scales in a way that the generated sequence converges both globally and at a linear rate.

12: This matches the best convergence theory for composite optimization. 

13: To make iPAL practically efficient, we propose a gradient-Newton method in a subspace for the iPAL subproblem. 

14: This is accomplished by detecting active samples and features with the help of the proximal operator of the hard margin loss and the projection of cardinality constraint. 

15: Extensive numerical results on both simulated and real datasets demonstrate that the proposed method is

16: fast, produces sparse solution of high accuracy, and can lead to effective reduction on active samples and features  when compared with several leading solvers.

17: \end{abstract}

18: