BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Sum-Product Networks for Probabilistic Modeling - Robert Peharz (T
 U Graz)
DTSTART:20160524T130000Z
DTEND:20160524T140000Z
UID:TALK66339@talks.cam.ac.uk
CONTACT:Louise Segar
DESCRIPTION:In machine learning and artificial intelligence\, probabilisti
 c graphical models are a principled and widely used approach for dealing w
 ith uncertain knowledge. However\, one of their downsides is that exact in
 ference quickly becomes intractable in practical models. Therefore\, one o
 ften uses simple models\, allowing exact inference\, which however potenti
 ally undermodel a given problem\, or one uses approximate inference method
 s\, whose performance is often hard to assess for particular applications.
  Sum-Product networks (SPNs) are a new avenue for probabilistic modeling\,
  promising a remedy for this problem: Using a deep network structure\, the
 y are able to represent highly complex variable dependencies\, while at th
 e same time many inference scenarios can be solved with computational cost
 s linear in the representation size of the SPN. In this talk\, I give an i
 ntroduction to SPNs\, and discuss basic notions such as completeness\, con
 sistency and decomposability\, which enable tractable inference. I further
  present some recent theoretical and practical results\, together with the
 ir implications on representational properties\, learning and inference. F
 inally\, I present some results of SPNs applied to computer vision and spe
 ech modeling.
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR