BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Hyper and structural Markov laws for graphical models - Simon Byrn
 e (Statistical Laboratory\, University of Cambridge)
DTSTART:20111020T130000Z
DTEND:20111020T143000Z
UID:TALK33961@talks.cam.ac.uk
CONTACT:Konstantina Palla
DESCRIPTION:My talk will be based on the concept of _hyper Markov laws_\, 
 introduced\nby Dawid and Lauritzen (1993):\nhttp://projecteuclid.org/eucli
 d.aos/1176349260\n\nThe general idea is to use distributions on the parame
 ters (termed\n"laws" in the paper) that have analogous conditional indepen
 dence\nproperties to those of the Markov distributions. These commonly ari
 se\nin two circumstances: as the sampling distributions of estimators\n(e.
 g. maximum likelihood estimators)\, and as priors and posteriors for\nBaye
 sian inference. As with message passing algorithms for\nmarginalisation\, 
 we can exploit the conditional independence\nproperties to perform calcula
 tions locally at certain points of\nthe graph.\n\nSecondly\, I'll introduc
 e my own work on extending these ideas to the\ncase where the structure of
  the graph itself is unknown (commonly\ncalled structural learning)\, by d
 efining what I term _structural Markov properties_. These characterise an 
 exponential family over the\nset of graphs\, and form a conjugate prior al
 lowing convenient prior to\nposterior updating.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR