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:Stochastic control as an inference problem - Prof.
Bert Kappen (University of Nijmegen)
DTSTART;TZID=Europe/London:20090217T140000
DTEND;TZID=Europe/London:20090217T150000
UID:TALK15879AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/15879
DESCRIPTION:Stochastic optimal control theory deals with the p
roblem to compute an\noptimal set of actions to at
tain some future goal. Examples are found\nin many
contexts such as motor control tasks for robotics
\, planning and\nscheduling tasks or managing a fi
nancial portfolio. The computation of\nthe optimal
control is typically very difficult due to the si
ze of the\nstate space and the stochastic nature o
f the problem.\n\nWe introduce a class of stochast
ic optimal control problems that can\nbe mapped on
to a probabilistic inference problem. This duality
between\ncontrol and inference is well-known. The
novel aspect of the present\nformulation is that
the optimal solution is given by the minimum of a\
nfree energy and the link to graphical model infer
ence. We can thus apply principled approximations
such as the belief propagation or the Cluster Vari
ation method to obtain efficient approximations.\n
We will illustrate the method for the task stackin
g blocks. If time permits we will discuss distribu
ted (agent) solutions and comment on the partial o
bservable case.
LOCATION:Cambridge University Engineering Department\, LR5
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR