Distributed Second Order Methods with Fast Rates
                and Compressed Communication

                   Rustem Islamov 1 2 Xun Qian 1 Peter Richtarik 1

              Abstract              1. Introduction
                                The prevalent paradigm for training modern supervised ma-
                                chine learning models is based on (regularized) empirical
   We develop several new communication-efficient      risk minimization (ERM) (Shalev-Shwartz and Ben-D ...