SHARP ORACLE INEQUALITIES FOR LEAST SQUARES ESTIMATORS IN SHAPE RESTRICTED REGRESSION

被引:54
作者
Bellec, Pierre C. [1 ,2 ,3 ]
机构
[1] CNRS, UMR 9194, ENSAE, Paris, France
[2] Rutgers State Univ, Dept Stat & Biostat, 501 Hill Ctr,Busch Campus,110 Frelinghuysen Rd, Piscataway, NJ 08854 USA
[3] ENSAE, 3 Ave Pierre Larousse, F-92245 Malakoff, France
关键词
Shape restricted regression; convexity; minimax rates; Gaussian width; concentration; CONVEX REGRESSION; RISK BOUNDS;
D O I
10.1214/17-AOS1566
中图分类号
O21 [概率论与数理统计]; C8 [统计学];
学科分类号
020208 ; 070103 ; 0714 ;
摘要
The performance of Least Squares (LS) estimators is studied in shape-constrained regression models under Gaussian and sub-Gaussian noise. General bounds on the performance of LS estimators over closed convex sets are provided. These results have the form of sharp oracle inequalities that account for the model misspecification error. In the presence of misspecification, these bounds imply that the LS estimator estimates the projection of the true parameter at the same rate as in the well-specified case. In isotonic and unimodal regression, the LS estimator achieves the non-parametric rate n(-2/3) as well as a parametric rate of order k/n up to logarithmic factors, where k is the number of constant pieces of the true parameter. In univariate convex regression, the LS estimator satisfies an adaptive risk bound of order q/n up to logarithmic factors, where q is the number of affine pieces of the true regression function. This adaptive risk bound holds for any collection of design points. While Guntuboyina and Sen [Probab. Theory Related Fields 163 (2015) 379-411] established that the nonparametric rate of convex regression is of order n(-4/5) for equispaced design points, we show that the nonparametric rate of convex regression can be as slow as n(-2/3) for some worst-case design points. This phenomenon can be explained as follows: Although convexity brings more structure than unimodality, for some worstcase design points this extra structure is uninformative and the nonparametric rates of unimodal regression and convex regression are both n(-2/3). Higher order cones, such as the cone of beta-monotone sequences, are also studied.
引用
收藏
页码:745 / 780
页数:36
相关论文
共 25 条
[1]   Living on the edge: phase transitions in convex programs with random data [J].
Amelunxen, Dennis ;
Lotz, Martin ;
McCoy, Michael B. ;
Tropp, Joel A. .
INFORMATION AND INFERENCE-A JOURNAL OF THE IMA, 2014, 3 (03) :224-294
[2]  
[Anonymous], 1991, ERGEBNISSE MATH IHRE
[3]  
[Anonymous], SLOPE MEETS LASSO IM
[4]  
[Anonymous], SHARP TIME DATA TRAD
[5]   Empirical minimization [J].
Bartlett, PL ;
Mendelson, S .
PROBABILITY THEORY AND RELATED FIELDS, 2006, 135 (03) :311-334
[6]  
BELLEC P. C., 2018, SHARP ORACLE INEQU S, DOI [10.1214/17-AOS1566SUPP, DOI 10.1214/17-AOS1566SUPP]
[7]  
Bellec PC, 2015, J MACH LEARN RES, V16, P1879
[8]  
Boucheron S., 2013, CONCENTRATION INEQUA, DOI DOI 10.1093/ACPROF:OSO/9780199535255.001.0001
[9]   Computational and statistical tradeoffs via convex relaxation [J].
Chandrasekaran, Venkat ;
Jordan, Michael I. .
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2013, 110 (13) :E1181-E1190
[10]   The Convex Geometry of Linear Inverse Problems [J].
Chandrasekaran, Venkat ;
Recht, Benjamin ;
Parrilo, Pablo A. ;
Willsky, Alan S. .
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS, 2012, 12 (06) :805-849