BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Isaac Newton Institute Seminar Series
SUMMARY:Optimal Bayesian experimental design: focused obje
ctives and observation selection strategies - Yous
sef Marzouk (Massachusetts Institute of Technology
)
DTSTART;TZID=Europe/London:20180410T090000
DTEND;TZID=Europe/London:20180410T100000
UID:TALK103570AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/103570
DESCRIPTION:I will discuss two complementary efforts in Bayesi
an optimal experimental design for inverse problem
s. The first focuses on evaluating an experiment
al design objective: we describe a new computation
al approach for ``focused'\;'\; optimal Baye
sian experimental design with nonlinear models\, w
ith the goal of maximizing expected information ga
in in targeted subsets of model parameters. Our ap
proach considers uncertainty in the full set of mo
del parameters\, but employs a design objective th
at can exploit learning trade-offs among different
parameter subsets. We introduce a layered multipl
e importance sampling scheme that provides consist
ent estimates of expected information gain in this
focused setting\, with significant reductions in
estimator bias and variance for a given computatio
nal effort. The second effort focuses on optimiza
tion of information theoretic design objectives---
in particular\, from the combinatorial perspective
of observation selection. Given many potential ex
periments\, one may wish to choose a most informat
ive subset thereof. Even if the data have in princ
iple been collected\, practical constraints on sto
rage\, communication\, and computational costs may
limit the number of observations that one wishes
to employ. We introduce methods for selecting near
-optimal subsets of the data under cardinality con
straints. Our methods exploit the structure of lin
ear inverse problems in the Bayesian setting\, and
can be efficiently implemented using low-rank app
roximations and greedy strategies based on modular
bounds. This is joint work with Chi Feng and Jay
anth Jagalur-Mohan.
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:info@newton.ac.uk
END:VEVENT
END:VCALENDAR