Seminar by Dr Richard G. Clegg "Analysing time varying graphs"

Dr Richard G. Clegg is Senior Research Fellow in Electronic Engineering at UCL. Dr Clegg and a renowned expert in the statistical nature of network traffic, complex network analysis and overlay networks.

Abstract: Models which construct graphs have attracted a great deal of research interest.  Erdos-Reyni (random graphs), Watts-Strogatz (small worlds) and Barabasi-Albert (preferential attachment) all propose tractable mathematical models which produce graphs with different properties. It has been found that many real-life graphs have properties which are also properties of some of these models. For example, scale free behaviour is produced by the preferential attachment model and observed in social networks, computer networks, authorship networks and many others.  However, the exact statistical nature of these networks differs and there is a need to distinguish which model best fits a given "target" graph.

This talk describes the Framework for Evolving Topology Analysis (FETA) which uses combined models to produce an exact likelihood measure of whether a particular target graph has evolved using a hypothesised model.  The framework is shown to recover known parameters from artificial models and also to be able to predict goodness-of-fit of models on real data.  Example networks are drawn from the Internet, authorship networks and online social networks