Distributed quantile regression in decentralized optimization.

When dealing with massive data that is distributed across multiple servers, it is particularly important to solve the distributed learning problem while minimizing the communication cost between servers. In this paper, we investigate an estimation procedure based on the group alternating direction method of multipliers (GADMM) algorithm for computing distributed quantile regression models. Numerical experiments show that our proposed method has competitive performance in both communication cost and statistical computational efficiency. We also provide a real-world data application to demonstrate the superiority of our method.