Accelerating Bayesian Computations with Generative Diffusion Models. In many real-world situations it is necessary to make decisions based on uncertain outcomes. The bottleneck however is executing th

Description

Accelerating Bayesian Computations with Generative Diffusion Models. In many real-world situations it is necessary to make decisions based on uncertain outcomes. The bottleneck however is executing the highly complex calculations to find the possible outcomes and their probabilities, as it entails interrogating scientific models with thousands of variables. This project will develop a new computational approach for rapid and accurate prediction for highly-granular real-world scientific models. To do so, it will exploit state-of-the-art diffusion based generative models, which are immensely effective in similar highly granular image generation tasks. This project will then address a decision problem of national importance, namely optimising flight paths for drone-based early detection of bushfires.. Scheme: Discovery Projects. Field: 4611 - Machine Learning. Lead: Prof Sumeetpal Singh