Restricted gaussian process for predicting latent functions

In this paper, we evaluate the gaussian process (GP) as a powerful toolkit for nonparametric classification, and regression. Unlike traditional parametric methods, GPs provide a distribution over functional spaces to model the uncertainty in predictions. The relationship between GP and input correlation kernel functions are illustrated, and some different kernels are introduced. Moreover, practical applications of GP for large scale problems using the Nyström approximation have been studied, and several numerical examples have been provided to verify the validity and efficiency of the proposed method. The implementation codes have been executed in Python using Scikit-learn library.

Keywords:

Gaussian process , Predicting latent functions , global optimization , Nystrom approximation