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


Current browse context:


Change to browse by:

References & Citations


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

Mathematics > Statistics Theory

Title: Random forward models and log-likelihoods in Bayesian inverse problems

Abstract: We consider the use of randomised forward models and log-likelihoods within the Bayesian approach to inverse problems. Such random approximations to the exact forward model or log-likelihood arise naturally when a computationally expensive model is approximated using a cheaper stochastic surrogate, as in Gaussian process emulation (kriging), or in the field of probabilistic numerical methods. We show that the Hellinger distance between the exact and approximate Bayesian posteriors is bounded by moments of the difference between the true and approximate log-likelihoods. Example applications of these stability results are given for randomised misfit models in large data applications and the probabilistic solution of ordinary differential equations.
Comments: 25 pages
Subjects: Statistics Theory (math.ST); Numerical Analysis (math.NA); Probability (math.PR); Methodology (stat.ME)
MSC classes: 62F15, 62G08, 65C99, 65D05, 65D30, 65J22, 68W20
Journal reference: ASA/SIAM Journal of Uncertainty Quantification (2018)
DOI: 10.1137/18M1166523
Cite as: arXiv:1712.05717 [math.ST]
  (or arXiv:1712.05717v5 [math.ST] for this version)

Submission history

From: Tim Sullivan [view email]
[v1] Fri, 15 Dec 2017 15:39:26 GMT (30kb)
[v2] Mon, 22 Jan 2018 10:08:59 GMT (30kb)
[v3] Wed, 31 Jan 2018 14:16:50 GMT (31kb)
[v4] Wed, 27 Jun 2018 10:53:11 GMT (37kb)
[v5] Fri, 28 Sep 2018 09:02:26 GMT (37kb)

Link back to: arXiv, form interface, contact.