BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Probabilistic programming for experimental design - Jesse Mu (Univ
 ersity of Cambridge\, Computer Laboratory)
DTSTART:20171107T184500Z
DTEND:20171107T193000Z
UID:TALK94603@talks.cam.ac.uk
CONTACT:Jesse Mu
DESCRIPTION:Scientists run experiments to distinguish between competing hy
 potheses\, but how do we select the best experiment to run? The answer is 
 often non-trivial\, as there are usually many possible experiments but lim
 ited time and resources. I describe a system for Bayesian optimal experime
 nt design (OED) based on probabilistic programming languages (PPLs): given
  hypotheses encoded as PPL models and an explicit definition of the experi
 ment space\, OED automates the search for experiments with high expected i
 nformation gain. Additionally\, I describe "adaptive OED"\, a framework fo
 r active learning: by updating our prior beliefs on hypotheses with the ob
 served response to an experiment\, OED can suggest further experiments to 
 continue to tease apart hypotheses. I apply this system to two domains in 
 cognitive psychology—sequence prediction and causal knowledge—and demo
 nstrate that adaptive OED performs better than standard OED and other naiv
 e experiment selection procedures. 
LOCATION:Bevin Room\, Churchill College
END:VEVENT
END:VCALENDAR