Nearest Neighbor Gaussian Processes: A Solution to the Scalability Problem

Nearest Neighbor Gaussian Processes (or NNGPs for short) represent one of the latest solutions for scalability problems within GP-based spatial modeling. This thesis provides a theoretical introduction to NNGPs, beginning with the principles of Bayesian statistics, followed by an overview of Gaussian Processes and their framework. To complement the theoretical discussion, the spNNGP R package is applied to Point Spread Function (or PSF) modeling on a large (10^4 observations) astronomical imaging dataset from SDSS DR20, with the purpose of demonstrating NNGPs’ computational advantages on real-world cases.