A0606
Title: Residual projection for quantile regression
Authors: Ye Fan - Capital University of Business and Economics (China)
Nan Lin - Washington University in St. Louis (United States) [presenting]
Abstract: The alternating direction method of multipliers (ADMM) has been a popular solution to the computational challenges for quantile regression in big data. However, its relatively slow convergence can be a bottleneck when communication cost dominates local computational consumption, such as in the Internet of Things (IoT) networks. We propose an alternative technique using residual projection that converges faster. We proved the convergence property of the new technique and further extended it to composite quantile regression (CQR).