Communication-efficient estimation of high-dimensional quantile regression

Distributed estimation has received increasing attention in the last several years and is particularly useful in the big data setting. Both mean regression and quantile regression has been investigated recently. In this paper, we consider distributed quantile regression with high dimension using a lasso penalty for sparse modeling. We extend a previous communication-efficient approach resulting in a method for distributed quantile regression without the need to smooth the loss or the gradient of the loss. The method is simple to implement and we present some numerical studies with encouraging performances.