New communication methods to enable federated learning for bushfire prediction using LEO satellites
This project expects to generate new knowledge in satellite communications. Expected outcomes include new transmission designs and resource management schemes that enable distributed machine learning, called federated learning, over satellite links with intermittent connectivity and limited resources. This should provide significant benefits for Australia to quickly and accurately predict and prevent natural disasters, minimising their devastating consequences to wildlife, the environment, and people. Unlike conventional approaches, this project leverages federated learning to enable distributed bushfire prediction within LEO satellite constellations. It will eliminate the need for massive transmission bandwidth and central imagery data storage, while significantly reducing transmission delays. The proposed transmission designs will support a massive number of LEO satellite links with short and intermittent connectivity patterns. By comparison, existing methods are designed for much fewer satellites in higher earth orbits with more stable connectivity. By employing a rigorous optimisation framework, the proposed resource management solutions will provide theoretical guarantees of convergence, optimality, and stability, which are largely absent in current methods.
$32,000 per annum (2023 rate) indexed annually. For a PhD candidate, the living allowance scholarship is for 3.5 years and the tuition fee scholarship is for four years. Scholarships also include up to $1,500 relocation allowance.
A/Prof Duy Ngo, A/Prof Lawrence Ong, A/Prof Nguyen Tran
Application closing date
08 August 2023
If you are interested in applying for this scholarship, please contact me.