. Author manuscript; available in PMC: 2017 Apr 1.

Published in final edited form as: Ann Stat. 2016 Apr;44(2):629–659. doi: 10.1214/15-AOS1380

Table 2.

Comparison between MIPGO and the gradient methods. The numbers in parenthesis are the standard errors. GM₁ and GM₂ stand for the gradient methods proposed by Loh and Wainwright (2015) and Wang, Liu and Zhang (2014), respectively.

Method	ρ = 0.5, n = 20
Method	AD	FP	FN	Gap	Time
MIPGO	0.188 (0.016)	0.230 (0.042)	0 (0)	0 (0)	29.046 (5.216)

GM₁	2.000 (0.000)	0 (0)	2 (0)	25.828 (0.989)	0.002 (0.001)

GM₂	0.847 (0.055)	5.970 (0.436)	0 (0)	1.542 (0.119)	0.504 (0.042)

	ρ = 0.1, n = 35
MIPGO	0.085 (0.005)	0.020 (0.141)	0 (0)	0 (0)	27.029 (4.673)

GM₁	2.000 (0.000)	0 (0)	2 (0)	31.288 (1.011)	0.002 (0.000)

GM₂	0.936 (0.044)	6.000 (0.348)	0 (0)	4.179 (0.170)	0.524 (0.020)