Peruzzi M, Banerjee S, Dunson DB, Finley AO (2021). Grid-Parametrize-Split (GriPS) for Improved Scalable Inference in Spatial Big Data Analysis. arxiv.org/abs/2101.03579
How can one improve the performance of posterior sampling algorithms for spatial regression models for large scale data? We target posterior sampling for Meshed Gaussian Processes and propose a strategy based on gridding and parameter expansion. We show that both computation time and MCMC efficiency in terms of effective sample size can be significantly improved using our methodologies.
Data application: LiDAR data over the Tanana forest in Alaska. Data size approximately 5 million.
