|
Activity Number:
|
274
- Advances in Scalable Bayesian Methods for Spatial Data
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 9, 2022 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
International Society for Bayesian Analysis (ISBA)
|
|
Abstract #322107
|
|
|
Title:
|
Spatial Meshing and Manifold Preconditioning for Big Multivariate Models
|
|
Author(s):
|
Michele Peruzzi* and David Dunson
|
|
Companies:
|
Duke University and Duke University
|
|
Keywords:
|
spatial;
Gaussian process;
Bayesian;
MCMC;
gradient-based;
software
|
|
Abstract:
|
Quantifying spatial associations in multivariate geolocated data of different types is achievable via random effects in a Bayesian hierarchical model, but severe computational bottlenecks arise when spatial dependence is encoded as a latent Gaussian process (GP) in the increasingly common large scale data settings on which we focus. The scenario worsens in non-Gaussian models because the reduced analytical tractability leads to additional hurdles to computational efficiency. We introduce methodologies for efficiently computing multivariate Bayesian models of spatially referenced non-Gaussian data. First, we outline spatial meshing as a tool for building scalable processes using patterned directed acyclic graphs. Then, we introduce a novel Langevin method based on manifold preconditioning which achieves superior sampling performance with non-Gaussian multivariate data that are common in studying species' communities. Applications showcase the flexibility of the proposed methodologies and the 'meshed' package for R.
|
Authors who are presenting talks have a * after their name.
