Activity Number:
|
229
- Geostatistical Computing on Modern Parallel Architectures
|
Type:
|
Topic Contributed
|
Date/Time:
|
Tuesday, August 9, 2022 : 8:30 AM to 10:20 AM
|
Sponsor:
|
Section on Statistical Computing
|
Abstract #323343
|
|
Title:
|
Distributed Inference for a Spatial Bayesian Network with Application to Natural Hazard Risk Assessment
|
Author(s):
|
Christopher Krapu* and Nolan Hayes and Robert Stewart and Amy Rose and Alexandre Sorokine and Kuldeep Kurte
|
Companies:
|
Oak Ridge National Laboratory and Oak Ridge National Laboratory and Oak Ridge National Laboratory and Oak Ridge National Laboratory and Oak Ridge National Laboratory and Oak Ridge National Laboratory
|
Keywords:
|
MCMC;
Distributed inference;
Spatial statistics;
Bayesian networks
|
Abstract:
|
Major challenges for modeling opportunistically sampled real-world data are a high degree of missingness, strong sampling bias, as well as inherent spatial autocorrelation. To address these issues, we propose a novel graphical model for a spatial Bayesian network which combines a dimension-reduced latent Gaussian spatial field with parameters enforcing a DAG-derived cross-variable covariance structure which is amenable to usage of prior information derived from expert elicitation. To perform inference using large datasets with frequent missing data, we implement a distributed Gibbs sampling scheme composed of alternating steps of data augmentation and Hamiltonian Monte Carlo in PyMC3. We present a case study on modeling the properties of buildings for natural hazard risk assessment in Washington, D.C.
|
Authors who are presenting talks have a * after their name.