We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.ME

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Methodology

Title: Model comparison for Gibbs random fields using noisy reversible jump Markov chain Monte Carlo

Abstract: The reversible jump Markov chain Monte Carlo (RJMCMC) method offers an across-model simulation approach for Bayesian estimation and model comparison, by exploring the sampling space that consists of several models of possibly varying dimensions. A naive implementation of RJMCMC to models like Gibbs random fields suffers from computational difficulties: the posterior distribution for each model is termed doubly-intractable since computation of the likelihood function is rarely available. Consequently, it is simply impossible to simulate a transition of the Markov chain in the presence of likelihood intractability. A variant of RJMCMC is presented, called noisy RJMCMC, where the underlying transition kernel is replaced with an approximation based on unbiased estimators. Based on previous theoretical developments, convergence guarantees for the noisy RJMCMC algorithm are provided. The experiments show that the noisy RJMCMC algorithm can be much more efficient than other exact methods, provided that an estimator with controlled Monte Carlo variance is used, a fact which is in agreement with the theoretical analysis.
Comments: Accepted for publication in Computational Statistics and Data Analysis
Subjects: Methodology (stat.ME); Computation (stat.CO)
Journal reference: Computational Statistics and Data Analysis 128 (2018) 221-241
DOI: 10.1016/j.csda.2018.07.005
Cite as: arXiv:1712.05358 [stat.ME]
  (or arXiv:1712.05358v3 [stat.ME] for this version)

Submission history

From: Lampros Bouranis [view email]
[v1] Thu, 14 Dec 2017 17:37:52 GMT (61kb)
[v2] Thu, 21 Dec 2017 14:20:51 GMT (61kb)
[v3] Fri, 13 Jul 2018 22:51:33 GMT (65kb)

Link back to: arXiv, form interface, contact.