Skip to content Skip to navigation

Bayesian and Maximum Likelihood Estimation for Gaussian Processes on an Incomplete Lattice

Event Type: 
Colloquium
Speaker: 
Dr. Jonathan Stroud (George Washington University)
Event Date: 
Friday, October 31, 2014 -
12:00pm to 1:00pm
Location: 
SMLC 356
Audience: 
General PublicFaculty/StaffStudentsAlumni/Friends
Sponsor/s: 
stats group

Event Description: 

This research proposes a new approach for Bayesian and maximum likelihood
 parameter estimation for stationary Gaussian processes observed on a
 large lattice with missing values. We propose an MCMC approach for
 Bayesian inference, and a Monte Carlo EM algorithm for maximum
 likelihood inference. Our approach uses data augmentation and
 circulant embedding of the covariance matrix, and provides exact
 inference for the parameters and the missing data. Using simulated
 data and an application to satellite sea surface temperatures in the
 Pacific Ocean, we show that our method provides accurate inference on
 lattices of sizes up to 512 x 512, and outperforms two popular
 methods: composite likelihood and spectral approximations.

Event Contact

Contact Name: Yan Lu

Contact Phone: 5052772544

Contact Email: luyan@math.unm.edu