Title: fastkqr: A fast algorithm for kernel quantile regression
Authors: Qian Tang - University of Iowa (United States)
Yuwen Gu - University of Connecticut (United States)
Boxiang Wang - University of Iowa (United States) [presenting]
Abstract: Quantile regression is a powerful tool to model certain quantiles of the response and has been widely used in many application areas, including economics, finance, social sciences, and engineering, among others. The computation of quantile regression is typically expensive due to its nonsmooth loss function. We propose a major advance to the computation of quantile regression in reproducing kernel Hilbert spaces. We develop a novel and efficient algorithm called fastkqr for computing the exact solution path of kernel quantile regression. To improve the computation speed, we develop a fast implementation strategy to carefully reuse the matrix computations in fastkqr. Extensive simulation studies and benchmark applications demonstrate orders of magnitude speedup of fastkqr over the existing algorithms of quantile regression with almost the same algorithm accuracy.