Description:

People love to rank things.  From songs to sportspeople, top 10s are ubiquitous.  So, it is perhaps unsurprising that some of the earliest developments in mathematical statistics relate to ranking.  More surprising, however, is the relatively weak extent to which ranking methods are subject to scientific scrutiny.

I will discuss the problem of ranking things from sets of pairwise comparisons – such as matches between tennis players.  I will present two algorithms for rigorously applying the models: one using belief propagation, another neural networks.  With the flexibility and computational efficiency of these methods, we can begin to question the basic assumptions of our methods by applying Bayesian model selection techniques.

Dr. George Cantwell, Postdoctoral Fellow, Santa Fe Institute

Host: Istvan Kovacs