Accelerated Spatio-Temporal Bayesian Modeling for Multivariate Gaussian Processes
Additional information
Authors
Gaedke-Merzhäuser L.,
Maillou V.,
Avellaneda F. R.,
Schenk O.,
Luisier M.,
Moraga P.,
Ziogas A. N.,
Rue H.
Type
Article in conference proceedings
Year
2025
Language
English
Abstract
Multivariate Gaussian processes (GPs) offer a powerful probabilistic framework to represent complex interdependent phenomena. They pose, however, significant computational challenges in high-dimensional settings, which frequently arise in spatial-temporal applications. We present PyINLA, a highly scalable framework for performing Bayesian inference tasks on spatio-temporal multivariate GPs, based on the methodology of integrated nested Laplace approximations. Our approach relies on a sparse inverse covariance matrix formulation of the GP, puts forward a GPU-accelerated block-dense approach, and introduces a hierarchical, triple-layer, distributed memory parallel scheme. We showcase weak scaling performance surpassing the state- of-the-art by two orders of magnitude on an 8$\times$ larger model and measure strong scaling speedups of
three orders of magnitude when running on 496 GH200 superchips on the Alps supercomputer. Applying PyINLA to air pollution data from northern Italy over 48 days, we showcase refined spatial resolutions over the aggregated pollutant measurements.
Keywords
omputational earth and atmospheric sciences Improved models, algorithms, performance or scalability of specific applications and respective software Use of uncertainty quantification, statistical, and machine-learning techniques to improve a specific HPC application
Conference proceedings
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Numero ( Mese )
November
Publisher
ACM/IEEE
Meeting name
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Meeting place
St. Louis, MO
Meeting date
16–21 Nov 2025
Pages (or article number)
1-12
