BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Variational inference in graphical models: The view from the margi
 nal polytope - David Knowles and Richard Turner (University of Cambridge)
DTSTART:20100204T140000Z
DTEND:20100204T153000Z
UID:TALK22276@talks.cam.ac.uk
CONTACT:Shakir Mohamed
DESCRIPTION:In last week's RCC\, we saw that loopy belief propagation coul
 d be connected to a constrained variational free energy optimisation. The 
 constraints ensured that the beliefs normalised and that they were locally
  consistent. This week\, we'll describe an alternate view of this optimisa
 tion which separately considers the constraints (the domain being called t
 he marginal polytope) and the free-energy. The optimization takes place ov
 er the lower dimensional space of generalised exponential family mean para
 meters. This representation clarifies that there are two distinct componen
 ts to variational inference algorithms: (a) an approximation to the entrop
 y function\; and (b) an approximation to the marginal polytope. This viewp
 oint clarifies the essential ingredients of known variational methods\, an
 d also suggests novel relaxations. Taking the “zero-temperature limit”
  recovers a variational representation for MAP computation as a linear pro
 gram (LP) over the marginal polytope.\n\nThe material we hope to cover (an
 d probably some extra) is covered on slides 1-13 and 23-39 of this "tutori
 al":http://www.eecs.berkeley.edu/~wainwrig/icml08/Wainwright_ICML08.pdf\n\
 nIf you feel inclined to delve into the theory a little more\, refer to "t
 his paper":http://www.cs.berkeley.edu/~jordan/papers/WaiJor_Aller03.ps\n
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR