BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Statistics
SUMMARY:Community Detection on the Weighted Stochastic Blo
ck Model - Min Xu (U Penn)
DTSTART;TZID=Europe/London:20161014T160000
DTEND;TZID=Europe/London:20161014T170000
UID:TALK67486AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/67486
DESCRIPTION:Ever since the seminal paper of Decelle et al appe
ared in 2011\, Stochastic Block Model has become t
he most well-studied and well-understood model for
network data with an underlying community structu
re. Yet SBM has a limitation: it assumes that each
network edge is Bernoulli 0/1--either on or off\;
this is restrictive because weighted edges are ub
iquitous and\, when edge weights are present\, it
may be important to incorporate them into a cluste
ring algorithm. In this talk\, we study the weight
ed generalization of the stochastic block model in
which an edge random variable can have a general
mixed distribution rather than Bernoulli\; we prop
ose and analyze an algorithm for the weighted SBM
based a binning procedure for nonparametric densit
y estimation. We show that this procedure has erro
r rate exponential in the information divergence t
hat governs the thresholds for the unweighted Stoc
hastic Block Model--a rate that in many cases have
matching lower bounds. \n\nJoint work with Varun
Jog and Po-Ling Loh from University of Wisconsin M
adison and Zongming Ma from the University of Penn
sylvania.
LOCATION:MR12\, Centre for Mathematical Sciences\, Wilberfo
rce Road\, Cambridge.
CONTACT:Quentin Berthet
END:VEVENT
END:VCALENDAR