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:Machine Learning @ CUED
SUMMARY:Probabilistic computing for Bayesian inference - V
ikash K. Mansinghka (MIT)
DTSTART;TZID=Europe/London:20140402T133000
DTEND;TZID=Europe/London:20140402T143000
UID:TALK51686AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/51686
DESCRIPTION:Probabilistic modeling and Bayesian inference prov
ide a unifying theoretical framework for uncertain
reasoning. They have become central tools for eng
ineering machine intelligence\, modeling human cog
nition\, and analyzing structured and unstructured
data. However\, they often seem far less unified\
, complete and expressive in practice than they ar
e in theory\, and can require significant interdis
ciplinary expertise to apply. Domains such as robo
tics and statistics involve diverse modeling idiom
s\, speed/accuracy requirements\, dataset sizes\,
and approximation techniques. Inference in simple
latent variable models can be computationally chal
lenging\, while state-of-the-art models do not fit
within standard formalisms and can be cumbersome
to specify\, let alone use.\n\n\nIn this talk\, I
will describe probabilistic computing systems that
address several of these challenges and that fit
together into a mathematically coherent software a
nd hardware stack for Bayesian inference and intel
ligent computation.\n\nI will focus on Venture\, a
new\, Turing-complete probabilistic programming p
latform descended from the Church probabilistic pr
ogramming language. In Venture\, models are repres
ented by executable code\, with random choices cor
responding to latent variables. Inference from dat
a is done via automatic but reprogrammable mechani
sms that cover a broad class of approximation stra
tegies\, including novel hybrids of Markov chain\,
sequential Monte Carlo and variational techniques
. I will describe applications in text analysis\,
high-dimensional statistics and computer vision th
at yield a 100x savings in lines of code versus st
andard approaches.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR