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:Information Engineering Division seminar list
SUMMARY:Bayesian Optimization for Probabilistic Programs -
  Tom Rainforth\, University of Oxford
DTSTART;TZID=Europe/London:20170315T110000
DTEND;TZID=Europe/London:20170315T120000
UID:TALK71612AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/71612
DESCRIPTION:Probabilistic programming systems (PPS) allow prob
 abilistic models to be represented in the form of 
 a generative model and statements for conditioning
  on data. Their core philosophy is to decouple mod
 el specification and inference\, the former corres
 ponding to the user-specified program code and the
  latter to an inference engine capable of operatin
 g on arbitrary programs.  Removing the need for us
 ers to write inference algorithms significantly re
 duces the burden of developing new models and make
 s effective statistical methods accessible to non-
 experts.\n\nIn this talk I will present BOPP: a ge
 neral purpose framework for (type II) maximum like
 lihood and marginal maximum a posteriori estimatio
 n of probabilistic program variables. By using a s
 eries of code transformations\, the evidence of an
 y probabilistic program\, and therefore of any gra
 phical model\, can be optimized with respect to an
  arbitrary subset of its sampled variables.  To ca
 rry out this optimization\, we develop the first B
 ayesian optimization package to directly exploit t
 he source code of its target\, leading to innovati
 ons in problem-independent hyperpriors\, unbounded
  optimization\, and implicit constraint satisfacti
 on\; delivering significant performance improvemen
 ts over prominent existing packages.  We present a
 pplications of our method to a number of tasks inc
 luding engineering design and parameter optimizati
 on.\n\n3 minute version of the talk: http://youtu.
 be/gVzV-NxKa9U\n\nAssociated paper: http://papers.
 nips.cc/paper/6421-bayesian-optimization-for-proba
 bilistic-programs\n\nCode: http://github.com/probp
 rog/bopp
LOCATION: Cambridge University Engineering Department\, CBL
  Seminar room BE4-38
CONTACT:
END:VEVENT
END:VCALENDAR