A1030
Title: A pathwise coordinate descent algorithm for penalized quantile regression
Authors: Sumanta Basu - Cornell University (United States) [presenting]
Abstract: A fast pathwise coordinate descent algorithm is introduced for penalized quantile regression. A closed-form update of the coordinate-wise minimization problem is derived, strategies for fast computation in high-dimension by leveraging underlying sparsity structure are discussed, and the benefit of a random perturbation is shown to help the algorithm avoid getting stuck along the regularization path. Computational efficiency gain over existing alternatives is demonstrated on simulated and real data sets.
