| Prof Darren Wilkinson | Professor of Statistics |
| Department of Mathematical Sciences | |
| Durham University |
Please share with appropriately qualified students
A fully funded PhD studentship is available in the Department of Mathematical Sciences of Durham University, UK.
The studentship covers tuition fees and standard UK PhD stipend for 3.5 years, but is available only to students eligible for home fees (typically, applicants from the UK).
Closing date for applications: 24th March, 2023
Fully Bayesian inference for large and complex statistical models typically relies on some kind of Monte Carlo approach, but it is challenging to develop efficient algorithms in high-dimensional settings. Machine learning researchers have developed a range of interesting programming languages, libraries and frameworks for scalable gradient-based optimisation of objectives defined in terms of arrays. These frameworks are often differentiable, and generate highly-optimised code for very fast parallel execution on multiple CPUs or GPUs. Although these frameworks are not typically designed with Monte Carlo-based fully Bayesian inference in mind, and their use in this context is under-explored, some are in fact highly suitable for this purpose. This project will examine the effectiveness of several different frameworks for fully Bayesian inference, and the adaptation of algorithms for effective exploitation. Applications to large hierarchical models and discretely observed stochastic processes will be used to assess scalability in practice.
For informal enquiries, please contact Darren Wilkinson (darren.j.wilkinson@durham.ac.uk).
To apply, follow these instructions, being sure to specifically mention the title/supervisor of this project.
|
|
| darrenjw.github.io |