Activity Number:
|
198
- SPEED: Nonparametric Statistics: Estimation, Testing, and Modeling
|
Type:
|
Contributed
|
Date/Time:
|
Monday, July 30, 2018 : 11:35 AM to 12:20 PM
|
Sponsor:
|
Section on Nonparametric Statistics
|
Abstract #332621
|
|
Title:
|
Approximate Inference for Large Non-Gaussian Spatial Data
|
Author(s):
|
Daniel Zilber* and Matthias Katzfuss
|
Companies:
|
Texas A&M University and Texas A&M University
|
Keywords:
|
Gaussian Process;
Laplace;
Vecchia;
Approximation;
Generalized Linear Model
|
Abstract:
|
Spatially correlated non-Gaussian observations can be effectively modeled with latent Gaussian processes, but this approach can be analytically intractable and pose computational challenges due to the inversion of the GP covariance matrix. We present an approximation that circumvents these issues and makes analysis with large data sets feasible. The first component consists of a Laplace approximation to the non-Gaussian likelihood, while the second component is a sparse general Vecchia approximation to the GP covariance. We demonstrate the advantages of this Vecchia-Laplace technique via numerical simulation and compare to a popular low-rank alternative.
|
Authors who are presenting talks have a * after their name.