Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/87579
PIRA download icon_1.1View/Download Full Text
Title: A constructive approach to L0 penalized regression
Authors: Huang, J 
Jiao, Y 
Liu, Y 
Lu, X 
Issue Date: 2018
Source: Journal of machine learning research, Aug. 2018, v. 19, p. 1-37
Abstract: We propose a constructive approach to estimating sparse, high-dimensional linear regression models. The approach is a computational algorithm motivated from the KKT conditions for the `0-penalized least squares solutions. It generates a sequence of solutions iteratively, based on support detection using primal and dual information and root _nding. We refer to the algorithm as SDAR for brevity. Under a sparse Riesz condition on the design matrix and certain other conditions, we show that with high probability, the `2 estimation error of the solution sequence decays exponentially to the minimax error bound in O(log(RpJ)) iterations, where J is the number of important predictors and R is the relative magnitude of the nonzero target coe_cients; and under a mutual coherence condition and certain other conditions, the `1 estimation error decays to the optimal error bound in O(log(R)) iterations. Moreover the SDAR solution recovers the oracle least squares estimator within a _nite number of iterations with high probability if the sparsity level is known. Computational complexity analysis shows that the cost of SDAR is O(np) per iteration. We also consider an adaptive version of SDAR for use in practical applications where the true sparsity level is unknown. Simulation studies demonstrate that SDAR outperforms Lasso, MCP and two greedy methods in accuracy and e_ciency.
Keywords: Geometrical convergence
KKT conditions
Nonasymptotic error bounds
Publisher: MIT Press
Journal: Journal of machine learning research 
ISSN: 1532-4435
EISSN: 1533-7928
Rights: © 2018 Jian Huang, Yuling Jiao, Yanyan Liu, and Xiliang Lu.
License: CC-BY 4.0, see https://creativecommons.org/licenses/by/4.0/. Attribution requirements are provided at http://jmlr.org/papers/v19/17-194.html.
The following publication Huang, J., Jiao, Y., Liu, Y., & Lu, X. (2018). A constructive approach to L0 penalized regression. Journal of Machine Learning Research, 19, 1-37 is available at http://www.jmlr.org/papers/v19/17-194.html
Appears in Collections:Journal/Magazine Article

Files in This Item:
File Description SizeFormat 
Huang_L0_Penalized_Regression.pdf543.08 kBAdobe PDFView/Open
Open Access Information
Status open access
File Version Version of Record
Access
View full-text via PolyU eLinks SFX Query
Show full item record

Page views

67
Last Week
0
Last month
Citations as of May 5, 2024

Downloads

243
Citations as of May 5, 2024

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.