1: \begin{abstract}

2: We propose a communication- and computation-efficient distributed

3: optimization algorithm using second-order information for solving

4: empirical risk minimization (ERM) problems with a nonsmooth

5: regularization term.

6: Our algorithm is applicable to both the primal and the dual ERM problem.

7: Current second-order and quasi-Newton methods for this

8: problem either do not work well in the distributed setting or

9: work only for specific regularizers. Our algorithm uses successive

10: quadratic approximations of the smooth part, and we describe how to

11: maintain an approximation of the (generalized) Hessian and solve

12: subproblems efficiently in a distributed manner.

13: When applied to the distributed dual ERM problem, unlike state of the

14: art that takes only the block-diagonal part of the Hessian, our

15: approach is able to utilize global curvature information and is thus

16: magnitudes faster.

17: The proposed method enjoys global linear convergence for

18: a broad range of non-strongly convex problems that includes the most

19: commonly used ERMs, thus requiring lower communication complexity.

20: It also converges on non-convex problems, so has the

21: potential to be used on applications such as deep learning.

22: Computational results demonstrate that our method significantly

23: improves on communication cost and running time over the current

24: state-of-the-art methods.

25: 

26: \end{abstract}

27: