BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimal Design under Heteroscedasticity for Gaussian Process Emula
 tors with replicated observations - Boukouvalas\, A (Aston)
DTSTART:20110902T110000Z
DTEND:20110902T113000Z
UID:TALK32625@talks.cam.ac.uk
CONTACT:Mustapha Amrani
DESCRIPTION:Computer models\, or simulators\, are widely used in a range o
 f scientific fields to aid understanding of the processes involved and mak
 e predictions. Such simulators are often computationally demanding and are
  thus not amenable to statistical analysis. Emulators provide a statistica
 l approximation\, or surrogate\, for the simulators accounting for the add
 itional approximation uncertainty. \n\nFor random output\, or stochastic\,
  simulators the output dispersion\, and thus variance\, is typically a fun
 ction of the inputs. This work extends the emulator framework to account f
 or such heteroscedasticity by constructing two new heteroscedastic Gaussia
 n process representations and proposes an experimental design technique to
  optimally learn the model parameters. The design criterion is an extensio
 n of Fisher information to heteroscedastic variance models. Replicated obs
 ervations are efficiently handled in both the design and model inference s
 tages. We examine the effect of such optimal designs on both model paramet
 er uncertainty and predictive variance through a series of simulation expe
 riments on both synthetic and real world simulators.\n
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR