BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Multiresolution network models - Tyler McCormick (University of Wa
 shington)
DTSTART:20160726T123000Z
DTEND:20160726T130000Z
UID:TALK66851@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:<span>Social networks exhibit two key topological features: <i
 >global sparsity</i> and <i>local&nbsp\;</i></span><span><i>density</i>.&n
 bsp\; That is\, overall the propensity for interaction between any&nbsp\;t
 wo randomly selected actors is infinitesimal\, but for any given&nbsp\;ind
 ividual there is massive variability in the propensity to interact&nbsp\;w
 ith others in the network.&nbsp\; Further\, the relevant scientific&nbsp\;
 questions typically differ depending on the scale of analysis.&nbsp\; In t
 his talk\, we&nbsp\;propose a class of multiresolution statistical models 
 that model&nbsp\;network structures on multiple scales to enable inference
  about&nbsp\;relevant population-level parameters.&nbsp\; We capture globa
 l graph&nbsp\;structure using a mixture over projective models that captur
 e local&nbsp\;graph structures. This&nbsp\;approach&nbsp\;is advantageous 
 as it allows us&nbsp\;to&nbsp\;</span>differentially invest modeling effor
 t&nbsp\;within subgraphs of high density\, while maintaining a parsimoniou
 s&nbsp\;structure between such subgraphs. We&nbsp\;illustrate&nbsp\;the ut
 ility of our method using simulation and data on household relations&nbsp\
 ;from Karnataka\, India. &nbsp\;This is joint work with Bailey Fosdick (CS
 U) and Ted Westling (UW). &nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR