[Nets-seminars] EE talk 29th October. 4pm GS/302

[Richard G. Clegg]

A new method for generating network topologies -- Richard G. Clegg

Since 1999 there has been huge research interest in network
topologies and the internet topology in particular -- this
relates to the famous ``power law" or ``scale free" nature
of many networks.  Barabasi and Albert's influential work on preferential
attachment models showed how simple statistical processes could
grow networks with the scale-free properties exhibited by many
real-life networks (biological, electronic, computer and social
networks).  Since then the subject has progressed by investigating
new measurements (assortivity, k-core, rich-club coefficient, clustering
coefficient amongst others) and these measures have, in turn,
produced a new generation of network topology generation models
(generalised linear preference, truncated random walks, positive
feedback preference and others) which attempt to create networks
which match the measurements made on real networks.

In this talk a new approach is outlined which allows the fitting
of network topology generation models directly to the observed network
data.  A method is described which directly allows candidate models
to be assessed for goodness of fit to the nodes connected in the real
network rather than attempting to replicate a basket of different
statistical measures on the topology in question.  The FETA (Framework
for Evolving Topology Analysis) approach can also automatically tune
model parameters to fit real network data.  It allows the creation of
hybrid models (part PFP, part preferential attachment for example).
The approach is demonstrated here by recovering known model parameters
on simulated topologies and by analysis of the internet autonomous
system topology.

-- 
Richard G. Clegg,
Dept of Elec. Eng.,
University College London
http://www.richardclegg.org/