Project Level: PhD

Statistical inference is one of the most important tools used for scientific investigation. When dealing with data, the Bayesian paradigm is very appealing since it allows to incorporate prior knowledge into a proposed model, provides a well-structured inference method (conditional on the newly observed information), does not rely on asymptotic approximation, provides interpretable answers, and implements a straight-forward framework for model comparison and hypothesis testing. While these merits often come with high computational costs, continuing progress in the available computing resources allowed Bayesian statistics to rise to greater eminence in many scientific fields such as natural science, econometrics, social science, and engineering. However, despite recent advances, many real-life inference problems are still beyond the reach of classical Bayesian methods. Specifically, for many practical models, the evaluation of the likelihood function, a critical component of the Bayesian analysis, is either intractable or computationally prohibitive. In this project, you will investigate a number of methods such as the Pseudo-Marginal, the Integrated Nested Laplace, the Bayesian Synthetic Likelihood, the Variational Bayes, and the Approximate Bayesian Computation.

Project members

Dr Slava Vaisman

Lecturer
Mathematics