[Ml-stat-talks] Fwd: PACM Colloquium Monday 2/3/14

David Blei blei at CS.Princeton.EDU
Wed Jan 29 15:45:42 EST 2014


chris is a great speaker.  this talk is highly recommended.  it will
be of interest for those of you into graphs/networks, bayesian
statistics, variational methods, or seeing great talks :)


---------- Forwarded message ----------
From: Valerie Marino <vmarino at math.princeton.edu>
Date: Wed, Jan 29, 2014 at 12:46 PM
Subject: PACM Colloquium Monday 2/3/14

DATE: Monday, February 3, 2014

PLACE: 214 Fine Hall

TIME: 4:30 pm

SPEAKER:Chris Wiggins - Columbia University


Connections among disparate approaches to graph partitioning may be made
by reinterpreting the problem as a special case of one of either of two
more general and well-studied problems in machine learning: inferring
latent variables in a generative model or estimating an
(information-theoretic) optimal encoding in rate distortion theory. In
either approach, setting in a more general context shows how to unite and
generalize a number of approaches. As an encoding problem, we see how
heuristic measures such as the normalized cut may be derived from
information theoretic quantities. As an inference problem, we see how
variational Bayesian methods lead naturally to an efficient and
interpretable approach to identifying ``communities" in graphs as well as
revealing the natural scale (or number of communities) via Bayesian
approaches to complexity control.

More information about the Ml-stat-talks mailing list