A fast splitting method tailored for Dantzig selector

被引：0
作者：
Hongjin He
Xingju Cai
Deren Han
机构：
[1] Hangzhou Dianzi University,Department of Mathematics, School of Science
[2] Nanjing Normal University,School of Mathematical Sciences, Jiangsu Key Laboratory for NSLSCS
来源：
Computational Optimization and Applications | 2015年 / 62卷
关键词：
Fast splitting method; Dantzig selector; Linear regression; Alternating direction method of multipliers;
D O I：
暂无
中图分类号：
学科分类号：
摘要：
In this paper, we introduce a splitting method for solving Dantzig selector problem, a new linear regression model that was extensively studied in the literature in the past few years. The new method is very simple in the sense that, per iteration, it only performs a projection onto a box, and does some matrix-vector products. We prove the global convergence of the method and report some promising numerical results, which demonstrate that the new method is competitive with some state-of-the-art methods recently developed in the literature.
引用
收藏
页码：347 / 372
页数：25
相关论文
共 81 条
[1]  
Becker S(2011)Templates for convex cone problems with applications to sparse signal recovery Math. Program. Comput. 3 165-218
[2]  
Candés E(2007)Discussion: the dantzig selector: statistical estimation when Ann. Stat. 35 2352-2357
[3]  
Grant M(2010) is much larger than Found. Trends Mach. Learn. 3 1-122
[4]  
Bickel P(1995)Distributed optimization and statistical learning via the alternating direction method of multipliers Technometrics 37 373-384
[5]  
Boyd S(2007)Better subset regression using the non-negative garrote Ann. Stat. 35 2365-2368
[6]  
Parikh N(2007)Discussion: the dantzig selector: statistical estimation when Ann. Stat. 35 2313-2351
[7]  
Chu E(2007) is much larger than Ann. Stat. 35 2392-2404
[8]  
Peleato B(2011)The Dantzig selector: statistical estimation when SIAM J. Imaging Sci. 4 807-826
[9]  
Eckstein J(2012) is much larger than IMA J. Numer. Anal. 32 227-245
[10]  
Breiman L(2009)Rejoinder: the dantzig selector: statistical estimation when Biostatistics 10 621-639
123456789