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:A* Sampling - Chris Maddison (U Toronto)
DTSTART;TZID=Europe/London:20150224T110000
DTEND;TZID=Europe/London:20150224T120000
UID:TALK58164AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/58164
DESCRIPTION:The problem of drawing samples from a discrete dis
tribution can be converted into a discrete optimiz
ation problem. In this work\, we show how sampling
from a continuous distribution can be converted i
nto an optimization problem over continuous space.
Central to the method is a stochastic process rec
ently described in mathematical statistics that we
call the Gumbel process. We present a new constru
ction of the Gumbel process and A-star sampling\,
a practical generic sampling algorithm that search
es for the maximum of a Gumbel process using A-sta
r search. We analyze the correctness and convergen
ce time of A-star sampling and demonstrate empiric
ally that it makes more efficient use of bound and
likelihood evaluations than the most closely rela
ted adaptive rejection sampling-based algorithms.
LOCATION:Engineering Department\, CBL Room BE-438.
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR