In this week’s BRAG meeting (Thursday @ 11am in P504), Matt Moores will be presenting a talk in our series on Bayesian statistics & Big Data:
Big data & doubly-intractable likelihoods
The widespread availability of high-resolution imagery has created a need for computational methods that scale to millions of pixels. We review and evaluate three Bayesian algorithms for estimating the inverse temperature of a hidden Potts model. This hyperparameter governs the strength of spatial cohesion and therefore has a substantial influence over the resulting model fit. The difficulty arises from the dependence of an intractable normalising constant on the value of the inverse temperature, thus there is no closed form solution for sampling from the distribution directly. A number of methods have been proposed for addressing this issue, including pseudolikelihood, path sampling, and the approximate exchange algorithm. We apply these methods to 3D medical images to assess their performance on datasets of realistic size.
This is joint work with Clair Alston & Kerrie Mengersen