Accelerating Bayesian computations with generative diffusion models

Generative diffusion models are transforming fields ranging from the creative industries to computational chemistry. This PhD project will explore how these powerful ideas can be adapted to accelerate Bayesian computation for largescale mathematical models, where rigorous uncertainty quantification is essential but existing methods can be prohibitively expensive. This project will develop fast, accurate computational methods with the potential to transform natural-hazard response, including real-time decision-making for search, surveillance and rescue operations. More broadly, the techniques developed will have impact across a wide range of applications where high-stakes decisions must be made quickly and reliably under uncertainty.

What you will work on

• Contribute to developing new diffusion-guided methods (Score Generative models, Diffusion Schrodinger Bridge models) for Bayesian sampling.
• Undertake theoretical and numerical work at the intersection of Bayesian computation, stochastic processes, Monte Carlo methods and modern machine learning.
• Investigate how newly developed methods can improve environmental modelling for flood and bushfire risk management. One strategic application you will work on is drone surveillance for early bushfire detection, where flight paths must account for uncertain fire-danger information and update as new sensor data are gathered.
• Liaise with ANU’s Scout Drone project when working with real drone-based sensor data.

About you

• A strong interest in stochastic processes, computational mathematics, Bayesian computation and machine learning.
• Interest in developing both theory and practical algorithms.
• Prior experience or interest in Monte Carlo methods, stochastic differential equations, neural networks, probabilistic modelling, scientific computing, environmental modelling, reinforcement learning or path-planning will be advantageous.
• Open to International and Domestic candidates with Hons Class I or Masters with research thesis in the areas of Statistics, Mathematics, or related computational disciplines.

Scholarship Details

• $43,533 per annum (tax free) with potential further allowances for conference travel and research visits.
• Based at UNSW School of Mathematics and Statistics, supervised by Dr. Sahani Pathiraja (UNSW), Dr. Thang Bui (ANU) and Prof. Sumeetpal Singh (Uni Wollongong). You will also collaborate with the project’s external partner investigator, Prof Pierre Del Moral, who is a highly distinguished professor based at INRIA (France).
• Funding available through the ARC Discovery Project DP260101187 Accelerating Bayesian Computations with Generative Diffusion Models.
• Expected start date: 2027 (see application deadline below)

How to apply

To apply, send your full CV, university transcripts, and a cover letter describing your interest in this PhD position to Sahani Pathiraja at s.pathiraja@unsw.edu.au

Application deadline: Applications will be received until the Scholarship is filled. The PhD is expected to start at UNSW in Term 2. 

Shortlisted candidates will be required to submit a formal application for the PhD program at UNSW and apply for an RTP stipend at the time of application: https://www.unsw.edu.au/research/hdr/application.