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:Microsoft Research Cambridge\, public talks
SUMMARY:Coping with the Intractability of Graphical Models
- Justin Domke
DTSTART;TZID=Europe/London:20141127T100000
DTEND;TZID=Europe/London:20141127T110000
UID:TALK56481AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/56481
DESCRIPTION:Many potential applications of graphical models (s
uch as Conditional Random Fields) are complicated
by the fact that exact inference is intractable.
This talk will describe two strategies for coping
with this situation. The first is based on restri
cting consideration to a tractable set of paramete
rs. Rather than tree-structured parameters\, as i
s common\, I will explore a notion of tractability
where Markov chain Monte Carlo is guaranteed to q
uickly converge to the stationary distribution. T
his can be used both for inference (as a type of g
eneralized mean-field algorithm) and for learning\
, where it gives a FPRAS for maximum likelihood le
arning when restricted to this set. The second\,
more pragmatic\, strategy is based on empirical ri
sk minimization\, where a given approximate infere
nce method is “baked in” to the loss function. In
particular\, I will also discuss a recently relea
sed open-source tool for distributed learning of s
uch models using MPI.
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station R
oad\, Cambridge\, CB1 2FB
CONTACT:Microsoft Research Cambridge Talks Admins
END:VEVENT
END:VCALENDAR