DiSCO: Distributed Optimization for Self-Concordant Empirical Loss

We propose a new distributed algorithm for empirical risk minimization in machine learning. The algorithm is based on an inexact damped Newton method, where the inexact Newton steps are computed by a