BEGIN:VCALENDAR
PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3-
M3//EN
VERSION:2.0
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20180107T095906Z
DTSTART:20180214T160000Z
DTEND:20180214T173000Z
SUMMARY:Mitchell Centre Seminar Series
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}b5s-jb6w9s0
i-mqb2en
DESCRIPTION:Lampros Bouranis\, School of Mathematics and Statistics & Ins
ight Centre for Data Analytics\, University College Dublin\, Ireland\n\n
Bayesian model selection for exponential random graph models via adjuste
d pseudolikelihoods\n\nThis talk is concerned with the issue of model ch
oice for the well-known exponential random graph model\, widely used in
social network analysis. This amounts to choosing which network statisti
cs to include in the ERG model. However this task is complicated by the
fact that the ERG likelihood is intractable. We approach this problem fr
om a Bayesian perspective and propose a tractable approximation of the l
ikelihood using the pseudolikelihood function popularised by Strauss and
Ikeda. In particular\, we propose a novel adjustment of the pseudolikel
ihood function so that it is closely matches the ERG likelihood\, in a c
ertain sense. This\, in turn\, allows us to implement widely used comput
ational methods for Bayesian model selection in the context of exponenti
al random graph models for the analysis of real-world social networks. E
mpirical comparisons to existing methods for various experiments show th
at our procedure yields similar results to existing methods\, but at a f
raction of the computational cost. This is joint work with Nial Friel an
d Florian Maire.\n
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:A114\, Samuel Alexander Building\, Manchester
END:VEVENT
END:VCALENDAR