The minimax concave penalty （MCP） has been demonstrated theoretically and practical- ly to be effective in nonconvex penalization for variable selection and parameter estimation. In this paper, we develop an efficie...The minimax concave penalty （MCP） has been demonstrated theoretically and practical- ly to be effective in nonconvex penalization for variable selection and parameter estimation. In this paper, we develop an efficient alternating direction method of multipliers （ADMM） with continuation algorithm for solving the MCP-penalized least squares problem in high dimensions. Under some mild conditions, we study the convergence properties and the Karush-Kuhn-Tucker （KKT） optimality con- ditions of the proposed method. A high-dimensional BIC is developed to select the optimal tuning parameters. Simulations and a real data example are presented to illustrate the efficiency and accuracy of the proposed method.展开更多
The seamless-L0 （SELO） penalty is a smooth function on [0, ∞） that very closely resembles the L0 penalty, which has been demonstrated theoretically and practically to be effective in nonconvex penalization for var...The seamless-L0 （SELO） penalty is a smooth function on [0, ∞） that very closely resembles the L0 penalty, which has been demonstrated theoretically and practically to be effective in nonconvex penalization for variable selection. In this paper, we first generalize SELO to a class of penalties retaining good features of SELO, and then propose variable selection and estimation in linear models using the proposed generalized SELO （GSELO） penalized least squares （PLS） approach. We show that the GSELO-PLS procedure possesses the oracle property and consistently selects the true model under some regularity conditions in the presence of a diverging number of variables. The entire path of GSELO-PLS estimates can be efficiently computed through a smoothing quasi-Newton （SQN） method. A modified BIC coupled with a continuation strategy is developed to select the optimal tuning parameter. Simulation studies and analysis of a clinical data are carried out to evaluate the finite sample performance of the proposed method. In addition, numerical experiments involving simulation studies and analysis of a microarray data are also conducted for GSELO-PLS in the high-dimensional settings.展开更多
Coordinate descent method is a unconstrained optimization technique. When it is applied to support vector machine （SVM）, at each step the method updates one component of w by solving a one-variable sub-problem while...Coordinate descent method is a unconstrained optimization technique. When it is applied to support vector machine （SVM）, at each step the method updates one component of w by solving a one-variable sub-problem while fixing other components. All components of w update after one iteration. Then go to next iteration. Though the method converges and converges fast in the beginning, it converges slow for final convergence. To improve the speed of final convergence of coordinate descent method, Hooke and Jeeves algorithm which adds pattern search after every iteration in coordinate descent method was applied to SVM and a global Newton algorithm was used to solve one-variable subproblems. We proved the convergence of the algorithm. Experimental results show Hooke and Jeeves＇ method does accelerate convergence specially for final convergence and achieves higher testing accuracy more quickly in classification.展开更多
基金Supported by the National Natural Science Foundation of China(Grant Nos.11571263,11501579,11701571 and41572315)the Fundamental Research Funds for the Central Universities,China University of Geosciences(Wuhan)(Grant No.CUGW150809)
文摘The minimax concave penalty （MCP） has been demonstrated theoretically and practical- ly to be effective in nonconvex penalization for variable selection and parameter estimation. In this paper, we develop an efficient alternating direction method of multipliers （ADMM） with continuation algorithm for solving the MCP-penalized least squares problem in high dimensions. Under some mild conditions, we study the convergence properties and the Karush-Kuhn-Tucker （KKT） optimality con- ditions of the proposed method. A high-dimensional BIC is developed to select the optimal tuning parameters. Simulations and a real data example are presented to illustrate the efficiency and accuracy of the proposed method.
基金Supported by the National Natural Science Foundation of China(11501578,11501579,11701571,41572315)the Fundamental Research Funds for the Central Universities(CUGW150809)
文摘The seamless-L0 （SELO） penalty is a smooth function on [0, ∞） that very closely resembles the L0 penalty, which has been demonstrated theoretically and practically to be effective in nonconvex penalization for variable selection. In this paper, we first generalize SELO to a class of penalties retaining good features of SELO, and then propose variable selection and estimation in linear models using the proposed generalized SELO （GSELO） penalized least squares （PLS） approach. We show that the GSELO-PLS procedure possesses the oracle property and consistently selects the true model under some regularity conditions in the presence of a diverging number of variables. The entire path of GSELO-PLS estimates can be efficiently computed through a smoothing quasi-Newton （SQN） method. A modified BIC coupled with a continuation strategy is developed to select the optimal tuning parameter. Simulation studies and analysis of a clinical data are carried out to evaluate the finite sample performance of the proposed method. In addition, numerical experiments involving simulation studies and analysis of a microarray data are also conducted for GSELO-PLS in the high-dimensional settings.
基金supported by the National Natural Science Foundation of China (6057407560705004)
文摘Coordinate descent method is a unconstrained optimization technique. When it is applied to support vector machine （SVM）, at each step the method updates one component of w by solving a one-variable sub-problem while fixing other components. All components of w update after one iteration. Then go to next iteration. Though the method converges and converges fast in the beginning, it converges slow for final convergence. To improve the speed of final convergence of coordinate descent method, Hooke and Jeeves algorithm which adds pattern search after every iteration in coordinate descent method was applied to SVM and a global Newton algorithm was used to solve one-variable subproblems. We proved the convergence of the algorithm. Experimental results show Hooke and Jeeves＇ method does accelerate convergence specially for final convergence and achieves higher testing accuracy more quickly in classification.
基金Supported by National Natural Science Foundation of China(1150157811501579+2 种基金1170157141572315)Fundamental Research Funds for the Central Universities,China University of Geosciences(Wuhan)(CUGW150809)
