Article Contents
Article Contents

# Modeling interactive components by coordinate kernel polynomial models

• * Corresponding author: Xin Guo
The work described in this paper is partially supported by FRCAC of Middle Tennessee State University, and is partially supported by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. PolyU 25301115). All the three authors contributed equally to the paper
• We proposed the use of coordinate kernel polynomials in kernel regression. This new approach, called coordinate kernel polynomial regression, can simultaneously identify active variables and effective interactive components. Reparametrization refinement is found critical to improve the modeling accuracy and prediction power. The post-training component selection allows one to identify effective interactive components. Generalization error bounds are used to explain the effectiveness of the algorithm from a learning theory perspective and simulation studies are used to show its empirical effectiveness.

Mathematics Subject Classification: Primary: 68T05, 41A25; Secondary: 68Q32.

 Citation:

• Table 1.  Variable selection accuracy and average MSE for Example 1

 Algorithm TPR($x_1$) TPR($x_2$) FPR MSE CKPR-L 1.00 1.00 0.000 0.008 (0.000) CKPR-G 1.00 1.00 0.011 0.109 (0.015) LASSO 1.00 0.18 0.040 1.129 (0.015) COSSO 0.90 0.02 0.020 10.879 (8.345) SR-SIR (AIC) 1.00 0.89 0.460 - SR-SIR (BIC) 1.00 0.85 0.181 - SR-SIR (RIC) 1.00 0.75 0.053 -

Table 2.  Average and standard error of MSEs for Example 2

 $m=100$ $m=200$ $m=400$ CKPR-G 0.119 (0.003) 0.054 (0.001) 0.025 (0.0004) COSSO(GCV) 0.358 (0.009) 0.100 (0.003) 0.045 (0.001) COSSO(5CV) 0.378 (0.005) 0.094 (0.004) 0.043 (0.001) MARS 0.239 (0.008) 0.109 (0.003) 0.084 (0.001)

Table 3.  RMSE on three UCI data sets

 Ionosphere Sonar MR Wisc. BC $n$ 351 208 683 $p$ 33 60 9 CKPR-L $0.64 (0.04)$ $0.75 (0.06)$ $0.34 (0.02)$ CKPR-G $0.54 (0.03)$ $0.77 (0.06)$ $0.34 (0.02)$ Best in [5] $0.60 (0.05)$ $0.80 (0.04)$ $0.70 (0.01)$
•  [1] F. R. Bach, Consistency of the group lasso and multiple kernel learning, J. Mach. Learn. Res., 9 (2008), 1179-1225. [2] P. L. Bartlett and S. Mendelson, Rademacher and Gaussian complexities: Risk bounds and structural results, J. Mach. Learn. Res., 3 (2003), 463-482.  doi: 10.1162/153244303321897690. [3] M. D. Buhmann, Radial Basis Functions: Theory and Implementations, Cambridge Monographs on Applied and Computational Mathematics, 12, Cambridge University Press, Cambridge, 2003. doi: 10.1017/CBO9780511543241. [4] C. M. Carvalho, J. Chang, J. E. Lucas, J. R. Nevins, Q. Wang and M. West, High-dimensional sparse factor modeling: Applications in gene expression genomics, J. Amer. Statist. Assoc., 103 (2008), 1438-1456.  doi: 10.1198/016214508000000869. [5] C. Cortes, M. Mohri and A. Rostamizadeh, Learning non-linear combinations of kernels, in Advances in Neural Information Processing Systems, Curran Associates, Inc., (2009), 396–404. [6] J. H. Friedman, Multivariate adaptive regression splines, Ann. Statist., 19 (1991), 1-141.  doi: 10.1214/aos/1176347963. [7] I. Guyon, J. Weston, S. Barnhill and V. Vapnik, Gene selection for cancer classification using support vector machines, Machine Learning, 46 (2002), 389-422.  doi: 10.1023/A:1012487302797. [8] R. Kohavi and G. H. John, Wrappers for feature subset selection, Artificial Intelligence, 97 (1997), 273-324. [9] G. R. G. Lanckriet, N. Cristianini, P. Bartlett, L. El Ghaoui and M. I. Jordan, Learning the kernel matrix with semidefinite programming, J. Mach. Learn. Res., 5 (2003/04), 27-72. [10] S. L. Lauritzen, Graphical Models, Oxford Statistical Science Series, 17, The Clarendon Press, Oxford University Press, New York, 1996. [11] L. Li and X. Yin, Sliced inverse regression with regularizations, Biometrics, 64 (2008), 124-131.  doi: 10.1111/j.1541-0420.2007.00836.x. [12] F. Liang, K. Mao, M. Liao, S. Mukherjee and M. West, Nonparametric Bayesian Kernel Models, Technical report, Department of Statistical Science, Duke University, 2007. [13] Y. Lin and H. Zhang, Component selection and smoothing in multivariate nonparametric regression, Ann. Statist., 34 (2006), 2272-2297.  doi: 10.1214/009053606000000722. [14] C. McDiarmid, On the method of bounded differences, in Surveys in Combinatorics, London Math. Soc. Lecture Note Ser., 141, Cambridge Univ. Press, Cambridge, (1989), 148–188 [15] R. Meir and T. Zhang, Generalization error bounds for Bayesian mixture algorithms, J. Mach. Learn. Res., 4 (2004), 839-860.  doi: 10.1162/1532443041424300. [16] S. Mukherjee and Q. Wu, Estimation of gradients and coordinate covariation in classification, J. Mach. Learn. Res., 7 (2006), 2481-2514. [17] S. Mukherjee and D.-X. Zhou, Learning coordinate covariances via gradients, J. Mach. Learn. Res., 7 (2006), 519-549. [18] M. Pontil and C. Micchelli, Learning the kernel function via regularization, J. Mach. Learn. Res., 6 (2005), 1099-1125. [19] H. Qin and X. Guo, Semi-supervised learning with summary statistics, Anal. Appl. (Singap.), 17 (2019), 837-851.  doi: 10.1142/S0219530519400037. [20] B. Schölkopf, A. Smola and K.-R. Müller, Kernel principal component analysis, in Artificial Neural Networks–ICANN'97, Lecture Notes in Computer Science, 1327, Springer, Berlin, Heidelberg, (1997), 583–588. [21] L. Shi, Distributed learning with indefinite kernels, Anal. Appl. (Singap.), 17 (2019), 947-975.  doi: 10.1142/S021953051850032X. [22] T. P. Speed and H. T. Kiiveri, Gaussian Markov distributions over finite graphs, Ann. Statist., 138–150. doi: 10.1214/aos/1176349846. [23] R. Tibshirani, Regression shrinkage and selection via the lasso, J. Roy. Statist. Soc. Ser. B, 58 (1996), 267-288.  doi: 10.1111/j.2517-6161.1996.tb02080.x. [24] V. N. Vapnik, Statistical Learning Theory, John Wiley & Sons, Inc., New York, 1998. [25] G. Wahba, Spline models for observational data, CBMS-NSF Regional Conference Series in Applied Mathematics, 59, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1990. doi: 10.1137/1.9781611970128. [26] Q. Wang and X. Yin, A nonlinear multi-dimensional variable selection method for high dimensional data: Sparse MAVE, Comput. Statist. Data Anal., 52 (2008), 4512-4520.  doi: 10.1016/j.csda.2008.03.003. [27] Q. Wu, Y. Ying and D.-X. Zhou, Multi-kernel regularized classifiers, J. Complexity, 23 (2007), 108-134.  doi: 10.1016/j.jco.2006.06.007. [28] Y. Xu and H. Zhang, Refinable kernels, J. Mach. Learn. Res., 8 (2007), 2083-2120.  doi: 10.1109/IIHMSP.2010.145. [29] Y. Xu and H. Zhang, Refinement of reproducing kernels, J. Mach. Learn. Res., 10 (2009), 107-140. [30] W. Yao and Q. Wang, Robust variable selection through MAVE, Comput. Statist. Data Anal., 63 (2013), 42-49.  doi: 10.1016/j.csda.2013.01.021. [31] Y. Ying and C. Campbell, Rademacher chaos complexities for learning the kernel problem, Neural Comput., 22 (2010), 2858-2886.  doi: 10.1162/NECO_a_00028. [32] H. H. Zhang, Variable selection for support vector machines via smoothing spline ANOVA, Statist. Sinica, 16 (2006), 659-674. [33] N. Zhang, Z. Yu and Q. Wu, Overlapping sliced inverse regression for dimension reduction, Anal. Appl. (Singap.), 17 (2019), 715-736.  doi: 10.1142/S0219530519400013. [34] H. Zou and T. Hastie, Regularization and variable selection via the elastic net, J. R. Stat. Soc. Ser. B Stat. Methodol., 67 (2005), 301-320.  doi: 10.1111/j.1467-9868.2005.00503.x.

Tables(3)