Function Space Metropolis-Hastings Algorithms with Non-Gaussian Priors
Metropolis-Hastings (MH) algorithms are one of the most widely used methods for inference.
However, the convergence properties of these algorithms often deteriorate in high-dimensions
making them unsuitable for Bayesian inverse problems and non-parametric inference where,
in principle, the problem is defined on an infinite-dimensional Hilbert or Banach space.
In this talk we discuss some ideas for designing new MH algorithms that are reversible with
dimension-independent convergence properties. We present a new class of algorithms called
RCAR, that is tailored to priors based on the gamma distribution. We present an application of
RCAR for a deconvolution inverse problem and consider its convergence properties and consistent
I am a von Karman instructor in the Department of Computing and Mathematical Sciences at California Institute of Technology, sponsored by Prof. Andrew Stuart. Prior to that I received my Ph.D. in Applied and Computational Mathematics in the Department of Mathematics at Simon Fraser University with of Profs. Nilima Nigam and John Stockie. I work on problems at the interface of probability, statistics and applied mathematics with a particular focus on the analysis, development and application of computational methods for estimating parameters and quantifying uncertainty.