BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimal Bayesian experimental design: focused objectives and obser
 vation selection strategies - Youssef Marzouk (Massachusetts Institute of 
 Technology)
DTSTART:20180410T080000Z
DTEND:20180410T090000Z
UID:TALK103567@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:I will discuss two complementary efforts in Bayesian optimal e
 xperimental design for inverse problems.   The first focuses on evaluating
  an experimental design objective: we describe a new computational approac
 h for ``focused&#39\;&#39\; optimal Bayesian experimental design with nonl
 inear models\, with the goal of maximizing expected information gain in ta
 rgeted subsets of model parameters. Our approach considers uncertainty in 
 the full set of model parameters\, but employs a design objective that can
  exploit learning trade-offs among different parameter subsets. We introdu
 ce a layered multiple importance sampling scheme that provides consistent 
 estimates of expected information gain in this focused setting\, with sign
 ificant reductions in estimator bias and variance for a given computationa
 l effort.  The second effort focuses on optimization of information theore
 tic design objectives---in particular\, from the combinatorial perspective
  of observation selection. Given many potential experiments\, one may wish
  to choose a most informative subset thereof. Even if the data have in pri
 nciple been collected\, practical constraints on storage\, communication\,
  and computational costs may limit the number of observations that one wis
 hes to employ. We introduce methods for selecting near-optimal subsets of 
 the data under cardinality constraints. Our methods exploit the structure 
 of linear inverse problems in the Bayesian setting\, and can be efficientl
 y implemented using low-rank approximations and greedy strategies based on
  modular bounds.  This is joint work with Chi Feng and Jayanth Jagalur-Moh
 an.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR