Invited speaker

Abstract:
In this talk I will describe a Bayesian network model for the analysis of data that commonly arises with some DNA samples recovered from a crime scene, in which it is found that DNA from two or more persons is present --- so called DNA mixtures. Such samples present challenging problems for forensic scientists which will be described. A statistical model that directly describes peak height information obtained from an electropherogram of a forensic DNA sample will be presented. For case specific data, the parameters of the model are estimated by maximum likelihood, even in the presence of multiple unknown contributors, by exploiting the Bayesian network representation for efficient computation. Data from a case taken the literature is used to illustrate the efficacy of the model.

This talk is based on joint work with T. Graversen, S. L. Lauritzen and J. Mortera

Short bio:
Robert Cowell obtained a PhD in Theoretical Physics in 1981 from Essex University. His research since 1989 has focussed on probabilistic graphical models, and in particular Bayesian networks. His collaboration with Phil Dawid, David Spiegelhalter and Steffen Lauritzen whilst based at University College London led to several publications including the research monograph "Probabilistic Networks and Expert Systems" published by Springer in 1999, for which they were awarded the 1st DeGroot Prize in 2002.
In 1995 he moved to City University where he has worked ever since. He was a co-chair of the Tenth International Workshop on Artificial Intelligence and Statistics, which took place in January 2005. His recent research interests have focused on applications of Bayesian networks to forensics genetics.