Bounded Regression with Gaussian Process Projection

Abstract: Examples with bound information on the regression function and density abound in many real applications. We propose a novel approach for estimating such functions by incorporating the prior knowledge on the bounds. Specially, a Gaussian process is first imposed on the regression function whose posterior distribution is then projected onto the bounded space. The resulting projected measure is then used for inference. The projected sample path has closed form which facilitates efficient computations. In particular, our projection approach maintains a comparable computational efficiency with that of the original GP. The proposed method yield predictions that respects bound constraints everywhere, while allows varying bounds across the input domain. An extensive simulation study is carried out which demonstrates that the performance of our approach dominates that of the competitors. An application to real data set is also considered.