BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.bham.ac.uk//v3//EN
BEGIN:VEVENT
CATEGORIES:Applied Mathematics Seminar Series
SUMMARY:Strategies for Multilevel Monte Carlo - Kody Law\,
University of Manchester
DTSTART:20200130T120000Z
DTEND:20200130T130000Z
UID:TALK3610AT
URL:/talk/index/3610
DESCRIPTION:This talk will concern the problem of inference wh
en the posterior measure involves continuous model
s which require approximation before inference can
be performed. Typically one cannot sample from th
e posterior distribution directly\, but can at bes
t only evaluate it\, up to a normalizing constant.
Therefore one must resort to computationally-inte
nsive inference algorithms in order to construct e
stimators. These algorithms are typically of Monte
Carlo type\, and include for example Markov chain
Monte Carlo\, importance samplers\, and sequentia
l Monte Carlo samplers. The multilevel Monte Carlo
method provides a way of optimally balancing disc
retization and sampling error on a hierarchy of ap
proximation levels\, such that cost is optimized.
Recently this method has been applied to computati
onally intensive inference. This non-trivial task
can be achieved in a variety of ways. This talk wi
ll review 3 primary strategies which have been suc
cessfully employed to achieve optimal (or canonica
l) convergence rates – in other words faster conve
rgence than i.i.d. sampling at the finest discreti
zation level. Some of the specific resulting algor
ithms\, and applications\, will also be presented.
LOCATION:Biosciences 301
CONTACT:Fabian Spill
END:VEVENT
END:VCALENDAR