[seminar] [DANAS] Tiago P. Peixoto, 18.05. (ponedjeljak) 11h (s.t.=TOCNO), Predavaona I krila IRB

[Kornelija Passek-Kumericki]

                       SEMINAR  TEORIJSKE  FIZIKE


   (Zajednički seminari Fizičkog odsjeka PMF-a te Zavoda za teorijsku
    fiziku i Zavoda za eksperimentalnu fiziku IRB-a)

----------------------------------------------------------------------


Inferring the mesoscale structure of layered, edge-valued and
time-varying networks


Tiago P. Peixoto University Bremen


   Datum:  ponedjeljak, 18. svibnja 2015.
   Vrijeme: 11 sati s.t.
   Mjesto: IRB, predavaona I krila


Sazetak:

The structural properties of large-scale complex networks are often a
result of unknown generative processes that cannot be directly observed,
and need to be inferred only from their final outcome. Of particular
importance are the so-called large or mesoscale structures, often
represented by modules --- groups of nodes with similar topological
patterns --- for which general formative mechanisms (or even unified
descriptions) have not yet been fully identified. More recently, it has
been increasingly recognized that most network systems are in fact
composed of different types of interactions (represented as layers or
attributes on the edges) and change in time, and that these features
cannot be neglected when attempting to identify mechanisms of network
formation.  Since these elaborations increase the effective dimension of
the network description, they are a double-edged sword: On one hand, the
inclusion of layered or temporal structure can reveal important patterns
that are otherwise obscured, while on the other hand the uncontrolled
incorporation of many uncorrelated variables can in fact hide patterns
which would otherwise be detected. In this talk, I propose a robust and
principled method to tackle this problem, by defining general generative
models of modular network structure, incorporating layered, attributed
and time-varying properties, together with alternative generative
processes incorporating hierarchical structure, degree correction and
overlapping groups, as well as a Bayesian methodology to infer the
parameters from data and select between model variants. I show that the
method is capable of revealing hidden structure in layered, edge-valued
and time-varying networks, and that the most appropriate level of
granularity with respect to the added dimensions can be reliably
identified. I illustrate our approach on a variety of empirical
systems, including a social network of physicians, the voting
correlations of deputies in the Brazilian national congress, the global
airport network, and a proximity network of high-school students.

Voditeljica seminara: Kornelija Passek-Kumericki (passek at irb dot hr)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Tiago120515.pdf
Type: application/pdf
Size: 122996 bytes
Desc: 
URL: <http://www.phy.pmf.unizg.hr/pipermail/seminar/attachments/20150518/9e025084/attachment-0001.pdf>
-------------- next part --------------
_______________________________________________

http://www.phy.pmf.unizg.hr/cgi-bin/mailman/listinfo/seminar