BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Efficient Bayesian inference of multi-scale network structures - T
 iago Peixoto (ISI Foundation\; Universität Bremen )
DTSTART:20160714T153000Z
DTEND:20160714T160000Z
UID:TALK66760@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:A principled approach to characterize the hidden structure of 
 networks is to formulate generative models\, and then infer their paramete
 rs from data. When the desired structure is composed of modules or "commun
 ities"\, a popular choice for this task is the stochastic block model\, wh
 ere nodes are divided into groups\, and the placement of edges is conditio
 ned on the group memberships. In this talk\, we will present a nonparametr
 ic Bayesian inference framework based on a <i>microcanonical</i> formulati
 on of the stochastic block model. We show how this simple model variation 
 allows simultaneously for two important improvements over more traditional
  inference approaches: 1. Deeper Bayesian hierarchies\, with noninformativ
 e priors replaced by sequences of priors and hyperpriors\, that not only r
 emove limitations that seriously degrade the inference of large networks\,
  but also reveal structures at multiple scales\; 2. A very efficient infer
 ence algorithm that scales well not only for networks with a large number 
 of nodes and edges\, but also with an <i>unlimited</i> <i>number of groups
 </i>. We show also how this approach can be used to sample group hierarchi
 es from the posterior distribution\, perform model selection\, and how it 
 can be easily be generalized to networks with edge covariates and&nbsp\; n
 ode annotations. <br>
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR