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

Download:

Current browse context:

math.ST

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

Mathematics > Statistics Theory

Title: Statistical guarantees for Bayesian uncertainty quantification in non-linear inverse problems with Gaussian process priors

Abstract: Bayesian inference and uncertainty quantification in a general class of non-linear inverse regression models is considered. Analytic conditions on the regression model $\{\mathscr G(\theta): \theta \in \Theta\}$ and on Gaussian process priors for $\theta$ are provided such that semi-parametrically efficient inference is possible for a large class of linear functionals of $\theta$. A general semi-parametric Bernstein-von Mises theorem is proved that shows that the (non-Gaussian) posterior distributions are approximated by certain Gaussian measures centred at the posterior mean. As a consequence posterior-based credible sets are shown to be valid and optimal from a frequentist point of view. The theory is demonstrated to cover two prototypical applications with PDEs that arise in non-linear tomography problems: the first concerns an elliptic inverse problem for the Schr\"odinger equation, and the second the inversion of non-Abelian $X$-ray transforms. New PDE techniques are developed to show that the relevant Fisher information operators are invertible between suitable function spaces.
Comments: 49 pages, 2 figures
Subjects: Statistics Theory (math.ST); Analysis of PDEs (math.AP)
Cite as: arXiv:2007.15892 [math.ST]
  (or arXiv:2007.15892v1 [math.ST] for this version)

Submission history

From: Gabriel Paternain [view email]
[v1] Fri, 31 Jul 2020 08:04:07 GMT (222kb,D)

Link back to: arXiv, form interface, contact.