Activity Number:
|
125
- Bayesian Methods for Discrete Data Problems
|
Type:
|
Contributed
|
Date/Time:
|
Monday, July 31, 2017 : 8:30 AM to 10:20 AM
|
Sponsor:
|
Section on Bayesian Statistical Science
|
Abstract #324276
|
|
Title:
|
Dual-Purpose Bayesian Design for Parameter Estimation and Model Discrimination of Models with Intractable Likelihoods
|
Author(s):
|
Mahasen Bandara Dehideniya* and James McGree and Christopher Drovandi
|
Companies:
|
Queensland University of Technology and Queensland University of Technology and Queensland University of Technology
|
Keywords:
|
Kullback-Leibler divergence ;
Mutual Information ;
Synthetic likelihood ;
Total entropy
|
Abstract:
|
In this work, we propose a methodology based on total entropy to design dual-purpose experiments of parameter estimation and model discrimination for epidemiological models with computationally intractable likelihoods. Our methodology uses a novel synthetic likelihood approach to approximate the likelihood of discrete observations of a stochastic process and to approximate the total entropy utility function. We consider four stochastic processes namely the death model, the Susceptible-Infected model, the Susceptible-Infected-Recovered model, and the Susceptible-Exposed-Infected-Recovered. The model discrimination and parameter estimation properties of the designs were compared against optimal choices under the mutual information utility for model discrimination and the Kullback-Leibler divergence utility for parameter estimation. The results suggest that these dual-purpose designs perform efficiently under both experimental goals.
|
Authors who are presenting talks have a * after their name.