BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A Model-Learner Pattern for Bayesian Reasoning - Andy Gordon\, Mic
 rosoft Research and University of Edinburgh
DTSTART:20121119T130000Z
DTEND:20121119T140000Z
UID:TALK41245@talks.cam.ac.uk
CONTACT:Peter Sewell
DESCRIPTION:A Bayesian model consists of a pair of probability distributio
 ns\, known as the prior and sampling distributions. A wide range of fundam
 ental machine learning tasks\, including regression\, classification\, clu
 stering\, and many others\, can all be seen as Bayesian models. We propose
  a new probabilistic programming abstraction\, a typed Bayesian model\, wh
 ich is a pair of probabilistic functions for the prior and sampling distri
 butions. A sampler for a model is an algorithm to compute synthetic data f
 rom its sampling distribution\, while a learner for a model is an algorith
 m for probabilistic inference on the model.\nModels\, samplers\, and learn
 ers form a generic programming pattern for model-based inference.\nThey su
 pport the uniform expression of common tasks including model testing\, and
  generic compositions such as mixture models\, evidence-based model averag
 ing\, and mixtures of experts. A formal semantics supports reasoning about
  model equivalence and implementation correctness. By developing a series 
 of examples and three learner implementations based on exact inference\, b
 elief-propagation\, and Markov chain Monte Carlo\,\nwe demonstrate the bro
 ad applicability of this new programming pattern.\n\nThe talk is based on 
 joint work with Mihhail Aizatulin (Open University)\, Johannes Borgstroem 
 (Uppsala University)\, Guillaume Claret (MSR)\, Thore Graepel (MSR)\, Adit
 ya Nori (MSR)\, Sriram Rajamani (MSR)\, and Claudio Russo (MSR).\n\nSee ht
 tp://research.microsoft.com/fun
LOCATION:FW26
END:VEVENT
END:VCALENDAR