1: \begin{abstract}

2: In this paper a robust second-order method is developed for the solution of strongly convex $\ell_1$-regularized problems. 

3: The main aim is to make the proposed method as inexpensive as possible, while even difficult problems can be efficiently solved.

4: The proposed approach is a primal-dual Newton Conjugate Gradients (pdNCG) method. 

5: Convergence properties of pdNCG are studied and worst-case iteration complexity is established. 

6: Numerical results are presented on synthetic sparse least-squares problems and real world machine learning problems.

7: 

8: \keywords{$\ell_1$-regularization $\cdot$ Strongly convex optimization $\cdot$ Second-order methods $\cdot$ Iteration Complexity $\cdot$ Newton Conjugate-Gradients method}

9: 

10: \subclass{68W40, 65K05, 90C06, 90C25, 90C30, 90C51 }

11: 

12: \end{abstract}

13: