Folded concave penalized sparse linear regression: sparsity, statistical performance, and algorithmic theory for local solutions

被引:26
作者
Liu, Hongcheng [1 ]
Yao, Tao [1 ]
Li, Runze [2 ,3 ]
Ye, Yinyu [4 ]
机构
[1] Penn State Univ, Harold & Inge Marcus Dept Ind & Mfg Engn, University Pk, PA 16802 USA
[2] Penn State Univ, Dept Stat, University Pk, PA 16802 USA
[3] Penn State Univ, Methodol Ctr, University Pk, PA 16802 USA
[4] Stanford Univ, Dept Management Sci & Engn, Stanford, CA 94305 USA
基金
美国国家科学基金会;
关键词
Sparse recovery; Non-convex programming; NP-completeness; Folded concave penalty; Lasso; VARIABLE SELECTION; M-ESTIMATORS; LIKELIHOOD; REGULARIZATION; COMPLEXITY; RATES;
D O I
10.1007/s10107-017-1114-y
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
This paper concerns the folded concave penalized sparse linear regression (FCPSLR), a class of popular sparse recovery methods. Although FCPSLR yields desirable recovery performance when solved globally, computing a global solution is NP-complete. Despite some existing statistical performance analyses on local minimizers or on specific FCPSLR-based learning algorithms, it still remains open questions whether local solutions that are known to admit fully polynomial-time approximation schemes (FPTAS) may already be sufficient to ensure the statistical performance, and whether that statistical performance can be non-contingent on the specific designs of computing procedures. To address the questions, this paper presents the following threefold results: (1) Any local solution (stationary point) is a sparse estimator, under some conditions on the parameters of the folded concave penalties. (2) Perhaps more importantly, any local solution satisfying a significant subspace second-order necessary condition (SONC), which is weaker than the second-order KKT condition, yields a bounded error in approximating the true parameter with high probability. In addition, if the minimal signal strength is sufficient, the SONC solution likely recovers the oracle solution. This result also explicates that the goal of improving the statistical performance is consistent with the optimization criteria of minimizing the suboptimality gap in solving the non-convex programming formulation of FCPSLR. (3) We apply (2) to the special case of FCPSLR with minimax concave penalty and show that under the restricted eigenvalue condition, any SONC solution with a better objective value than the Lasso solution entails the strong oracle property. In addition, such a solution generates a model error (ME) comparable to the optimal but exponential-time sparse estimator given a sufficient sample size, while the worst-case ME is comparable to the Lasso in general. Furthermore, to guarantee the SONC admits FPTAS.
引用
收藏
页码:207 / 240
页数:34
相关论文
共 39 条
  • [1] Adamczak R, 2010, J AM MATH SOC, V23, P535
  • [2] LEAST QUANTILE REGRESSION VIA MODERN OPTIMIZATION
    Bertsimas, Dimitris
    Mazumder, Rahul
    [J]. ANNALS OF STATISTICS, 2014, 42 (06) : 2494 - 2525
  • [3] Bian W., 2014, OPTIMALITY CONDITION
  • [4] Complexity analysis of interior point algorithms for non-Lipschitz and nonconvex minimization
    Bian, Wei
    Chen, Xiaojun
    Ye, Yinyu
    [J]. MATHEMATICAL PROGRAMMING, 2015, 149 (1-2) : 301 - 327
  • [5] SIMULTANEOUS ANALYSIS OF LASSO AND DANTZIG SELECTOR
    Bickel, Peter J.
    Ritov, Ya'acov
    Tsybakov, Alexandre B.
    [J]. ANNALS OF STATISTICS, 2009, 37 (04) : 1705 - 1732
  • [6] Decoding by linear programming
    Candes, EJ
    Tao, T
    [J]. IEEE TRANSACTIONS ON INFORMATION THEORY, 2005, 51 (12) : 4203 - 4215
  • [7] Adaptive cubic regularisation methods for unconstrained optimization. Part I: motivation, convergence and numerical results
    Cartis, Coralia
    Gould, Nicholas I. M.
    Toint, Philippe L.
    [J]. MATHEMATICAL PROGRAMMING, 2011, 127 (02) : 245 - 295
  • [8] Chen XJ, 2014, MATH PROGRAM, V143, P371, DOI 10.1007/s10107-012-0613-0
  • [9] LOWER BOUND THEORY OF NONZERO ENTRIES IN SOLUTIONS OF l2-lp MINIMIZATION
    Chen, Xiaojun
    Xu, Fengmin
    Ye, Yinyu
    [J]. SIAM JOURNAL ON SCIENTIFIC COMPUTING, 2010, 32 (05) : 2832 - 2852
  • [10] STRONG ORACLE OPTIMALITY OF FOLDED CONCAVE PENALIZED ESTIMATION
    Fan, Jianqing
    Xue, Lingzhou
    Zou, Hui
    [J]. ANNALS OF STATISTICS, 2014, 42 (03) : 819 - 849