Abstract:This paper presents a new approach for Gaussian process (GP) regression for large datasets. The approach involves partitioning the regression input domain into multiple local regions with a different local GP model fitted in each region. Unlike existing local partitioned GP approaches, we introduce a technique for patching together the local GP models nearly seamlessly to ensure that the local GP models for two neighboring regions produce nearly the same response prediction and prediction error variance on the boundary between the two regions. This largely mitigates the well-known discontinuity problem that degrades the boundary accuracy of existing local partitioned GP methods. Our main innovation is to represent the continuity conditions as additional pseudo-observations that the differences between neighboring GP responses are identically zero at an appropriately chosen set of boundary input locations. To predict the response at any input location, we simply augment the actual response observations with the pseudo-observations and apply standard GP prediction methods to the augmented data. In contrast to heuristic continuity adjustments, this has an advantage of working within a formal GP framework, so that the GP-based predictive uncertainty quantification remains valid. Our approach also inherits a sparse block-like structure for the sample covariance matrix, which results in computationally efficient closed-form expressions for the predictive mean and variance. In addition, we provide a new spatial partitioning scheme based on a recursive space partitioning along local principal component directions, which makes the proposed approach applicable for regression domains having more than two dimensions. Using three spatial datasets and three higher dimensional datasets, we investigate the numerical performance of the approach and compare it to several state-of-the-art approaches.

Fast Bayesian inference of block Nearest Neighbor Gaussian models for large data

Hierarchical Nearest-Neighbor Gaussian Process Models for Large Geostatistical Datasets

Bayesian inference for high-dimensional nonstationary Gaussian processes

spNNGP R package for Nearest Neighbor Gaussian Process models

Amortized Bayesian Local Interpolation NetworK: Fast covariance parameter estimation for Gaussian Processes

Fast Bayesian inference in large Gaussian graphical models

Block Vecchia Approximation for Scalable and Efficient Gaussian Process Computations

Multivariate nearest‐neighbors Gaussian processes with random covariance matrices

Nearest-Neighbor Mixture Models for Non-Gaussian Spatial Processes

Composite Inference for Gaussian Processes

High-Dimensional Bayesian Geostatistics

Bayesian Nonstationary Spatial Modeling for Very Large Datasets

Fast Bayesian estimation of spatial count data models

Spatial Bayesian Neural Networks

A Scalable Partitioned Approach to Model Massive Nonstationary Non-Gaussian Spatial Datasets

Bayesian Geostatistical Modeling for Discrete-Valued Processes

Patchwork Kriging for Large-scale Gaussian Process Regression

Variational Nearest Neighbor Gaussian Process

Integrated Nested Laplace Approximations for Large-Scale Spatial-Temporal Bayesian Modeling

Efficient Large-scale Nonstationary Spatial Covariance Function Estimation Using Convolutional Neural Networks

smashGP: Large-Scale Spatial Modeling via Matrix-Free Gaussian Processes