Dan Simpson ( University of Toronto, Department of Statistical Sciences)
Title: Placating pugilistic pachyderms: proper priors prevent poor performance
Modern statistical inference finds itself caught between two charging elephants: an elephant named model complexity and the elephant who answers only to "expressivity". Within the Bayesian framework, prior distributions are a way to try to balance these angry pachyderms. In this talk, I will cover a bunch of methods for specifying and evaluating prior distributions for complex statistical models.