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 applications: BayesDB and
stochastic digital circuits - Vikash K. Mansinghka
(MIT)
DTSTART;TZID=Europe/London:20140404T110000
DTEND;TZID=Europe/London:20140404T120000
UID:TALK51687AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/51687
DESCRIPTION:This talk consists of two shorter talks on specifi
c probabilistic computing systems:\n\n1. BayesDB\,
a Bayesian database table\, lets users query many
of the probable implications of their tabular dat
a as naturally as SQL lets them query the data its
elf. With the Bayesian Query Language (BQL)\, a do
main-specific probabilistic programming language f
or data tables\, users with no statistics training
can solve basic data science problems\, such as d
etecting predictive relationships between variable
s\, inferring missing values\, simulating probable
observations\, and identifying statistically simi
lar database entries. BayesDB is based on a nonpar
ametric Bayesian machine learning technique for di
rectly estimating the full multivariate (joint) di
stribution underlying high-dimensional\, heterogen
eously typed data. I will illustrate BayesDB and g
ive an overview of current applications to dataset
s from econometrics and sociology. I will also dis
cuss the potential for using BayesDB to improve th
e quality of the empirical reasoning performed by
non-experts and begin to mitigate the shortage of
analysts with expertise in statistics.\n\n2. The b
rain interprets ambiguous sensory information fast
er and more reliably than modern computers\, using
neurons that are slower and less reliable than lo
gic gates. But Bayesian inference\, which underpin
s many computational models of perception and cogn
ition\, appears computationally challenging even g
iven modern transistor speeds and energy budgets.
I will show how to build fast Bayesian computing m
achines using intentionally stochastic\, digital p
arts\, narrowing this efficiency gap by multiple o
rders of magnitude.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR