共 50 条
Approximation bounds for sparse principal component analysis
被引:0
|作者:
Alexandre d’Aspremont
Francis Bach
Laurent El Ghaoui
机构:
[1] CNRS & D.I.,École Normale Supérieure
[2] UMR 8548,École Normale Supérieure
[3] INRIA,EECS
[4] SIERRA Project-Team & D.I.,undefined
[5] U.C. Berkeley,undefined
来源:
Mathematical Programming
|
2014年
/
148卷
关键词:
62H25;
90C22;
90C27;
D O I:
暂无
中图分类号:
学科分类号:
摘要:
We produce approximation bounds on a semidefinite programming relaxation for sparse principal component analysis. The sparse maximum eigenvalue problem cannot be efficiently approximated up to a constant approximation ratio, so our bounds depend on the optimum value of the semidefinite relaxation: the higher this value, the better the approximation. In particular, these bounds allow us to control approximation ratios for tractable statistics in hypothesis testing problems where data points are sampled from Gaussian models with a single sparse leading component.
引用
收藏
页码:89 / 110
页数:21
相关论文