BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A Bayesian approach to network modularity: inferring the structure
  and scale of modular networks - Jake Hofman (Columbia University)
DTSTART:20080306T110000Z
DTEND:20080306T120000Z
UID:TALK10992@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:We present an efficient\, principled\, and interpretable techn
 ique for\ninferring module assignments and identifying the optimal number 
 of\nmodules in relational data. Our approach is based on a generative\nmod
 el equivalent to an infinite-range spin-glass Potts model on the\nirregula
 r lattice defined by a given network\; the problem of\nidentifying modules
  is then tantamount to inferring posterior\ndistributions over both the la
 tent module assignments (i.e. spin\nstates) and the model parameters (i.e.
  coupling constants) while also\nidentifying the number of modules (i.e. n
 umber of occupied spin\nstates) in the network. Using the variational Baye
 s framework we\nderive a mean-field free energy\, the minimization of whic
 h provides\ncontrolled approximations to the distributions of interest.  W
 e show\nhow several existing methods for finding modules can be described 
 as\nvariant\, special\, or limiting cases of our work\, and how related\nm
 ethods for complexity control -- identification of the true number of\nmod
 ules -- are outperformed.  We apply the technique to synthetic and\nreal n
 etworks and outline how the method naturally allows for model\nselection a
 mong competing network models.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR