Current browse context:
math.ST
Change to browse by:
References & Citations
Mathematics > Probability
Title: Well-posed Bayesian inverse problems and heavy-tailed stable quasi-Banach space priors
(Submitted on 19 May 2016 (v1), last revised 30 Jun 2017 (this version, v5))
Abstract: This article extends the framework of Bayesian inverse problems in infinite-dimensional parameter spaces, as advocated by Stuart (Acta Numer. 19:451--559, 2010) and others, to the case of a heavy-tailed prior measure in the family of stable distributions, such as an infinite-dimensional Cauchy distribution, for which polynomial moments are infinite or undefined. It is shown that analogues of the Karhunen--Lo\`eve expansion for square-integrable random variables can be used to sample such measures on quasi-Banach spaces. Furthermore, under weaker regularity assumptions than those used to date, the Bayesian posterior measure is shown to depend Lipschitz continuously in the Hellinger metric upon perturbations of the misfit function and observed data.
Submission history
From: Tim Sullivan [view email][v1] Thu, 19 May 2016 11:44:51 GMT (356kb)
[v2] Mon, 30 May 2016 17:53:09 GMT (1609kb)
[v3] Tue, 31 May 2016 09:58:14 GMT (1611kb)
[v4] Fri, 18 Nov 2016 15:34:41 GMT (218kb,D)
[v5] Fri, 30 Jun 2017 13:38:06 GMT (218kb,D)
Link back to: arXiv, form interface, contact.