科研之友
微信
新浪微博
Facebook
分享链接
摘要
The high-dimensional linear regression model has attracted much attention in areas like information technology, biology, chemometrics, economics, finance and other scientific fields. In this paper, we use smoothing techniques to deal with high-dimensional sparse models via quantile regression with the nonconvex l(p), penalty (0 < p < 1). We introduce two kinds of smoothing functions and give the estimation of approximation by our different smoothing functions. By smoothing the quantile function, we derive two types of lower bounds for any local solution of the smoothing quantile regression with the nonconvex l(p) penalty. Then with the help of l(1) regularization, we propose a smoothing iterative method for the smoothing quantile regression with the weighted l(1) penalty and establish its global convergence, whose efficient performance is illustrated by the numerical experiments.
