1: \begin{abstract}

2: In nonparametric regression problems involving multiple

3: predictors, there is typically interest in estimating the multivariate regression

4: surface in the important predictors while discarding the unimportant ones.

5: Our focus is on defining a Bayesian procedure that leads to the minimax

6: optimal rate of posterior contraction (up to a log factor) adapting to the unknown

7: dimension and anisotropic smoothness of the true surface.  We propose such an

8: approach based on a Gaussian process prior with dimension-specific scalings, which

9: are assigned carefully-chosen hyperpriors.  

10: %The prior also leads to consistent

11: %Bayesian variable selection. 

12: We additionally show that using a homogenous Gaussian

13: process with a single bandwidth leads to a sub-optimal rate in anisotropic cases.

14: %Gaussian processes are widely used as priors on function spaces in a variety of non-parametric Bayesian methods including density estimation, regression, classification among others. \cite{van2009adaptive} showed that rescaling a homogeneous smooth Gaussian field with an appropriate prior on the scaling parameter leads to a minimax-optimal rate of contraction of the posterior distribution that also adapts to the unknown smoothness of the functional parameter. In multidimensional problems, practitioners often use dimension specific scalings to allow for anisotropy with a point mass mixture prior on these scales to allow a subset of variables to drop out from the covariance kernel. Although such methods have been empirically successful, there hasn't been a theoretical study of such procedures in a Bayesian framework. In this article, we propose a joint prior on the multiple scales via a hierarchical Bayesian framework that simultaneously adapts to all anisotropy and reduced dimensions. We additionally show that one obtains a sub-optimal rate of convergence using a single-bandwidth in such cases.

15: \end{abstract}

16: