References & Citations
Mathematics > Statistics Theory
Title: Adaptive Bayesian nonparametric regression using kernel mixture of polynomials with application to partial linear model
(Submitted on 22 Oct 2017 (v1), last revised 13 Sep 2018 (this version, v3))
Abstract: We propose a kernel mixture of polynomials prior for Bayesian nonparametric regression. The regression function is modeled by local averages of polynomials with kernel mixture weights. We obtain the minimax-optimal rate of contraction of the full posterior distribution up to a logarithmic factor that adapts to the smoothness level of the true function by estimating metric entropies of certain function classes. We also provide a frequentist sieve maximum likelihood estimator with a near-optimal convergence rate. We further investigate the application of the kernel mixture of polynomials to the partial linear model and obtain both the near-optimal rate of contraction for the nonparametric component and the Bernstein-von Mises limit (i.e., asymptotic normality) of the parametric component. The proposed method is illustrated with numerical examples and shows superior performance in terms of computational efficiency, accuracy, and uncertainty quantification compared to the local polynomial regression, DiceKriging, and the robust Gaussian stochastic process.
Submission history
From: Fangzheng Xie [view email][v1] Sun, 22 Oct 2017 21:28:09 GMT (252kb,D)
[v2] Mon, 5 Mar 2018 02:04:30 GMT (490kb,D)
[v3] Thu, 13 Sep 2018 20:13:07 GMT (585kb,D)
Link back to: arXiv, form interface, contact.