论文信息 - Discussion: The Dantzig selector: Statistical estimation when p is much larger than n

Discussion: The Dantzig selector: Statistical estimation when p is much larger than n

n log p, where s is the dimension of the sparsest model. These are, respectively, the conditions of this paper using the Dantzig selector and those of Bunea, Tsybakov and Wegkamp [2] and Meinshausen and Yu [9] using the Lasso. Strictly speaking, Bunea, Tsybakov and Wegkamp consider only prediction, not l2 loss, but in a paper in preparation with Ritov and Tsybakov we show that the spirit of their conditions is applicable for l2 loss as well. Since these authors emphasize different points and use different normalizations, I thought it would be useful to present them together. Write the model as Y = Xn×pβ + e,

P. Bickel

[1] I. Johnstone,et al. Ideal spatial adaptation by wavelet shrinkage , 1994 .

[2] Wenjiang J. Fu,et al. Asymptotics for lasso-type estimators , 2000 .

[3] Adam Krzyzak,et al. A Distribution-Free Theory of Nonparametric Regression , 2002, Springer series in statistics.

[4] D. Donoho. For most large underdetermined systems of linear equations the minimal 𝓁1‐norm solution is also the sparsest solution , 2006 .

[5] Jianqing Fan,et al. Sure independence screening for ultrahigh dimensional feature space , 2006, math/0612857.

[6] P. Bickel,et al. Some Theory for Generalized Boosting Algorithms , 2006, J. Mach. Learn. Res..

[7] A. Tsybakov,et al. Aggregation for Gaussian regression , 2007, 0710.3654.

[8] Terence Tao,et al. The Dantzig selector: Statistical estimation when P is much larger than n , 2005, math/0506081.

[9] N. Meinshausen,et al. LASSO-TYPE RECOVERY OF SPARSE REPRESENTATIONS FOR HIGH-DIMENSIONAL DATA , 2008, 0806.0145.