BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.bham.ac.uk//v3//EN
BEGIN:VEVENT
CATEGORIES:Applied Mathematics Seminar Series
SUMMARY:Surrogate Models in Large-Scale Bayesian Inverse P
roblems - Aretha Teckentrup\, University of Edinbu
rgh
DTSTART:20171130T120000Z
DTEND:20171130T130000Z
UID:TALK2866AT
URL:/talk/index/2866
DESCRIPTION:We are interested in the inverse problem of estima
ting unknown parameters in a mathematical model fr
om observed data. We follow the Bayesian approach\
, in which the solution to the inverse problem is
the distribution of the unknown parameters conditi
oned on the observed data\, the so-called posterio
r distribution. We are particularly interested in
the case where the mathematical model is non-linea
r and expensive to simulate\, for example given by
a partial differential equation. In this case\, t
he solution of the inverse problem quickly becomes
computationally infeasible in practice.\n\nTo all
eviate this problem\, we consider the use of surro
gate models to approximate the Bayesian posterior
distribution. We present a general framework for t
he analysis of the error introduced in the posteri
or distribution\, and discuss particular examples
of surrogate models such as Gaussian process emula
tors and randomised misfit approaches.
LOCATION:Nuffield G17
CONTACT:Meurig Gallagher
END:VEVENT
END:VCALENDAR