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:Bayesian optimal design for Gaussian process model
  - Maria Adamou (University of Southampton)
DTSTART;TZID=Europe/London:20180208T160000
DTEND;TZID=Europe/London:20180208T170000
UID:TALK100225AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/100225
DESCRIPTION:<span>Co-author: Dave Woods		(University of Southa
 mpton)        <br></span><br>Data collected from c
 orrelated processes arise in many diverse applicat
 ion areas including both computer and physical exp
 eriments\, and studies in environmental science. O
 ften\, such data are used for prediction and optim
 isation of the process under study. For example\, 
 we may wish to construct an emulator of a computat
 ionally expensive computer model\, or simulator\, 
 and then use this emulator to find settings of the
  controllable variables that maximise the predicte
 d response.  The design of the experiment from whi
 ch the data are collected may strongly influence t
 he quality of the model fit and hence the precisio
 n and accuracy of subsequent predictions and decis
 ions. We consider Gaussian process models that are
  typically defined by a correlation structure that
  may depend upon unknown parameters. This parametr
 ic uncertainty may affect the choice of design poi
 nts\, and ideally should be taken into account whe
 n choosing a design. We consider a decision-theore
 tic Bayesian design for Gaussian process models wh
 ich is usually computationally challenging as it r
 equires the optimization of an analytically intrac
 table expected loss function over high-dimensional
  design space. We use a new approximation to the e
 xpected loss to find decision-theoretic optimal de
 signs. The resulting designs are illustrated throu
 gh a number of simple examples.
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR