A communication efficient distributed one-step estimation.

Distributed statistical inference has received enormous attracted attention in recent years. In this paper, a novel communication efficient one-step estimator on a distributed system is presented, which only needs one round of communication, effectively lowers the communication cost and sufficiently reduces the local computation complexity compared with the existing one-step method. The resulting estimator is statistically efficient as that the entire dataset is analyzed on one machine. Under mild conditions, a risk upper bound is established and the consistency together with asymptotic normality is demonstrated for the proposed estimator. Finally, numerical simulations are carried out to conform the theoretical result. [ABSTRACT FROM AUTHOR]