Quasi-Newton updating for large-scale distributed learning

Abstract Distributed computing is critically important for modern statistical analysis. Herein, we develop a distributed quasi-Newton (DQN) framework with excellent statistical, computation, and communication efficiency. In the DQN method, no Hessian matrix inversion or needed. This considerably reduces computation complexity of proposed method. Notably, related existing methods only analyse numerical convergence require diverging number iterations to converge. However, investigate properties method theoretically demonstrate that resulting estimator statistically efficient over small under mild conditions. Extensive analyses finite sample performance.