LocalNewton: Reducing communication rounds for distributed learning

Abstract

To address the communication bottleneck problem in distributed optimization within a master-worker framework, we propose LocalNewton, a distributed second-order algorithm with