Journal Article

Scalable spatiotemporal prediction with Bayesian neural fields

Hoffman, Matthew, Saad, Feras, Burnim, Jacob, Carroll, Colin, Köster, Urs, A. Saurous, Rif, Patton, Brian

2024

Overview

Spatiotemporal datasets, which consist of spatially-referenced time series, are ubiquitous in diverse applications, such as air pollution monitoring, disease tracking, and cloud-demand forecasting. As the scale of modern datasets increases, there is a growing need for statistical methods that are flexible enough to capture complex spatiotemporal dynamics and scalable enough to handle many observations. This article introduces the Bayesian Neural Field (B ayes NF), a domain-general statistical model that infers rich spatiotemporal probability distributions for data-analysis tasks including forecasting, interpolation, and variography. B ayes NF integrates a deep neural network architecture for high-capacity function estimation with hierarchical Bayesian inference for robust predictive uncertainty quantification. Evaluations against prominent baselines show that B ayes NF delivers improvements on prediction problems from climate and public health data containing tens to hundreds of thousands of measurements. Accompanying the paper is an open-source software package ( https://github.com/google/bayesnf ) that runs on GPU and TPU accelerators through the Jax machine learning platform. Spatiotemporal data consisting of measurements gathered at different times and locations is challenging to analyse due to variability and noise impact across different scales. The authors propose a statistical approach that delivers models of large-scale spatiotemporal datasets applicable to data-analysis tasks of forecasting and interpolation.

• Publisher: Nature Publishing Group UK,Nature Publishing Group,Nature Portfolio
• Subject: 639/705/1042 / 639/705/117 / 639/705/531 / Air monitoring / Air pollution / Artificial neural networks / Bayesian analysis / Climate prediction / Data analysis / Data processing / Datasets / Forecasting / Humanities and Social Sciences / Impact analysis / Interpolation / Machine learning / Mathematical models / multidisciplinary / Neural networks / Noise prediction / Open source software / Outdoor air quality / Pollution monitoring / Public health / Science / Science (multidisciplinary) / Spatiotemporal data / Statistical analysis / Statistical inference / Statistical methods / Statistical models / Task complexity