{\rtf1\ansi\ansicpg1252\cocoartf1404\cocoasubrtf130
{\fonttbl\f0\fnil\fcharset0 Calibri;\f1\fnil\fcharset0 Cambria;\f2\froman\fcharset0 TimesNewRomanPSMT;
}
{\colortbl;\red255\green255\blue255;}
\margl1440\margr1440\vieww10800\viewh8400\viewkind0
\deftab720
\pard\pardeftab720\partightenfactor0

\f0\fs34 \cf0 \expnd0\expndtw0\kerning0
Dimension-Independent MCMC algorithms for PDE-contrained BIPs
\f1\fs32 \kerning1\expnd0\expndtw0 \
\pard\pardeftab720\ri720\partightenfactor0
\cf0 \
Bayesian inverse problems often involve sampling probability\'a0distributions on functions. Traditional MCMC algorithms fail under mesh\'a0refinement. Recently, a variety of dimension-independent MCMC methods\'a0have emerged, but few of them take the geometry of the posterior into\'a0account.\'a0 In this work, we try to blend recent developments in finite\'a0dimensional manifold samplers with dimension-independent MCMC. The goal of such a marriage is to speed up the mixing of the Markov chain by using\'a0geometry, whilst remaining robust under mesh-refinement. The key idea is to employ the manifold methods on an appropriately chosen finite dimensional subspace.
\f2\fs24 \
}