Abstract:

The aim of this project is the development of improved sampling algorithms for Bayesian inference. To achieve this we build upon the ideas presented in and, to create an efficient and robust sampling scheme, that works well on both high dimensional and multimodal distributions. As an initial application the project will develop a Bayesian model selection tool with an integrated goodness-of-fit test, that reveals how one might improve upon the model to describe the data even better. The goodness of fit test will be build upon the test represented in but using trigonometric functions as an alternative extension, to the exponential family used in.
The hope is that this leads to new avenues of applications for Bayesian inference, due to a sampling algorithm robust and efficient in a wider range of scenarios, such as the improved Bayesian model selection scheme of this project. Allowing the true potential of Bayesian inference to be available for the problems at hand.