BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:iCalendar-Ruby
BEGIN:VEVENT
CATEGORIES:Conferences/Seminars/Lectures
DESCRIPTION:Yash Deshpande (MIT)\n\nTitle: An information-theoretic analysi
s of the stochastic block model\n\n \n\nAbstract:\n\nThe stochastic block m
odel is a popular model for large networks with latent clustered structure.
Over the last few years\, there has been significant interest in the stati
stical\, computational and robustness aspects of the stochastic block model
. In particular\, insights from statistical physics have been particularly
instrumental in spurring this interest and providing sharp predictions via
non-rigorous methods. In this talk\, we will consider information-theoretic
view of the simplest stochastic block model and rigorously establish much
of the picture provided by statistical physics\, in a certain ‘diverging de
gree’ limit. We will also show how information-theoretic quantities are int
imately related with fundamental limits in estimation.\n\nThe analysis proc
eeds by a universality argument reducing the stochastic block models to (po
ssibly more) familiar low-rank perturbations of GOE matrices. This is estab
lished through the classical Lindeberg swapping trick. Thanks to the Gaussi
an structure\, the low-rank perturbed models are then amenable to a variety
of analysis methods. We will take an algorithmic route using a particular
‘approximate message passing’ scheme to establish the final result.\n\nJoin
t work with Emmanuel Abbe and Andrea Montanari.
DTEND:20171204T221500Z
DTSTAMP:20190425T162220Z
DTSTART:20171204T211500Z
LOCATION:4-153\, 4-153
SEQUENCE:0
SUMMARY:Probability Seminar
UID:tag:localist.com\,2008:EventInstance_3189472
URL:http://calendar.mit.edu/event/probability_seminar_2495
END:VEVENT
END:VCALENDAR