|
Activity Number:
|
254
- Contributed Poster Presentations: Section on Bayesian Statistical Science
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 29, 2019 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
|
Abstract #302885
|
|
|
Title:
|
A Bayesian Methodology for High-Dimensional Discrete Graphical Models
|
|
Author(s):
|
Anwesha Bhattacharyya*
|
|
Companies:
|
|
|
Keywords:
|
Bayesian;
Ising-Potts;
high-dimension;
parallel computation
|
|
Abstract:
|
This work introduces a Bayesian methodology for fitting large discrete graphical models (Ising and Potts models) that uses the pseudo-likelihood instead of the likelihood. This is a powerful relaxation which allows node-wise parallel computation under separable priors. In high dimensional graphical models we can expect some sparsity in the network. The natural approach is to encode this sparsity using spike-and-slab priors. We propose the use of a weak spike and slab prior in the form of a Gaussian spike for such models. Standard MCMC algorithms can be used for the purpose of evaluating resulting posterior distribution. The use of a Bayesian approach allows us to leverage existing prior information and implement variable selection and estimation simultaneously. We illustrate the method through several examples and real data.
|
Authors who are presenting talks have a * after their name.
