Aspects of Bayesian inverse problems

Doctoral thesis English OPEN
Agapiou, Sergios
  • Subject: QA

The goal of this thesis is to contribute to the formulation and understanding\ud of the Bayesian approach to inverse problems in function space. To this end\ud we examine two important aspects of this approach: the frequentist asymptotic\ud properties of the posterior, and the extraction of information from the posterior\ud via sampling. We work in a separable Hilbert space setting and consider Gaussian\ud priors on the unknown in conjugate Gaussian models. In the first part of this\ud work we consider linear inverse problems with Gaussian additive noise and study\ud the contraction in the small noise limit of the Gaussian posterior distribution to\ud a Dirac measure centered on the true parameter underlying the data. In a wide\ud range of situations, which include both mildly and severely ill-posed problems, we\ud show how carefully calibrating the scaling of the prior as a function of the size of\ud the noise, based on a priori known information on the regularity of the truth, yields\ud optimal rates of contraction. In the second part we study the implementation in\ud RN of hierarchical Bayesian linear inverse problems with Gaussian noise and priors,\ud and with hyper-parameters introduced through the scalings of the prior and noise\ud covariance operators. We use function space intuition to understand the large N\ud behaviour of algorithms designed to sample the posterior and show that the two\ud scaling hyper-parameters evolve under these algorithms in contrasting ways: as N\ud grows the prior scaling slows down while the noise scaling speeds up. We propose\ud a reparametrization of the prior scaling which is robust with respect to the increase\ud in dimension. Our theory on the slowing down of the evolution of the prior scaling\ud extends to hierarchical approaches in more general conjugate Gaussian settings,\ud while our intuition covers other parameters of the prior covariance operator as well.\ud Throughout the thesis we use a blend of results from measure theory and probability\ud theory with tools from the theory of linear partial differential equations and\ud numerical analysis.
  • References (21)
    21 references, page 1 of 3

    [1] R. J. Adler. An introduction to continuity, extrema, and related topics for general Gaussian processes. Institute of Mathematical Statistics Lecture Notes| Monograph Series, 12. Institute of Mathematical Statistics, Hayward, CA, 1990.

    [2] S. Agapiou, J. M. Bardsley, O. Papaspiliopoulos, and A. M. Stuart. Analysis of the Gibbs sampler for hierarchical inverse problems. Arxiv preprint: 1311.1138.

    [3] S. Agapiou, S. Larsson, and A. M. Stuart. Posterior contraction rates for the Bayesian approach to linear ill-posed inverse problems. Stoch. Proc. Apl., 123(10):3828{3860, 2013.

    [4] S. Agapiou, A. M. Stuart, and Y. X. Zhang. Bayesian posterior contraction rates for linear severely ill-posed inverse problems. Arxiv preprint: 1210.1563 (to appear, J. Inverse Ill-posed Probl.).

    [5] H. Attouch, G. Buttazzo, and G. Michaille. Variational Analysis in Sobolev and BV Spaces: Applications to PDEs and Optimization. SIAM, 2006.

    [11] A. Beskos, G. Roberts, and A. Stuart. Optimal scalings for local MetropolisHastings chains on nonproduct targets in high dimensions. Ann. Appl. Probab., 19(3):863{898, 2009.

    [12] N. Bochkina. Consistency of the posterior distribution in generalized linear inverse problems. Inverse Problems, 29(9), 2013.

    [13] V. I. Bogachev. Gaussian measures, volume 62. Amer Mathematical Society, 1998.

    [14] L Cavalier. Nonparametric statistical inverse problems. Inverse Problems, 24(3):034004, 2008.

    [17] G. Da Prato. An introduction to in nite-dimensional analysis. Springer, 2005.

  • Metrics
    views in OpenAIRE
    views in local repository
    downloads in local repository

    The information is available from the following content providers:

    From Number Of Views Number Of Downloads
    Warwick Research Archives Portal Repository - IRUS-UK 0 42
Share - Bookmark