Many likelihood models are defined up to an intractable normalising constant, especially Markov random fields, such as those that arise in spatial statistics and statistical network analysis. These models are termed doubly intractable because the marginal likelihood or evidence of the corresponding posterior distribution is also unavailable.
There is a large body of literature dealing with the problem of parameter inference for doubly intractable distributions, including auxiliary variable methods and approximate Bayesian computation. This talk will explore how some of these methods can be extended and adapted to the problem of computing the evidence, thereby allowing statements about the probability of the model to be made.
Keywords: Double intractable distributions; Markov random fields; Auxiliary variable methods
Biography: Nial Friel is an associate professor in the school of Mathematical Sciences at University College Dublin. Prior to that he was a lecturer and then reader at the University of Glasgow. His research interests are in statistical network analysis, spatial statistics and Markov chain Monte Carlo methods.