MARS academics and researchers secure competitive grant funding to tackle complex challenges at the interface of applied mathematics and machine learning. Working with partners in academia, industry and government, our funded projects span environmental science, health, engineering and fundamental AI research.
Recent AI advances, typified by large language models like ChatGPT, have been driven by systems that can generate and analyse vast volumes of data in milliseconds. The next frontier of AI development lies in domains where data are inherently scarce, noisy, and costly to obtain, such as laboratory science and industrial R&D.
In these domains, progress has long relied on sequential experimental design, which uses rigorous mathematical principles to craft batches of experiments that extract the maximum possible insight from each measurement. However, classical mathematical principles of experimental design can no longer keep pace with modern science.
Conservation monitoring has traditionally relied on changes in species numbers or habitat condition. By the time such changes become apparent, ecosystems may already have suffered significant damage.
Animal behaviour often changes much earlier in response to environmental stress, and has the potential to serve as an early warning system. However, behavioural observation is time-consuming and, for this reason, has rarely been used in conservation practice. BEACON addresses this challenge using artificial intelligence.
The atmosphere possesses a natural oxidising capacity. Chemical reactions, driven largely by sunlight over the tropics, break down air pollutants and greenhouse gases. The effectiveness of this process determines how long these gases persist in the atmosphere, with significant implications for both air quality and climate change.
Policy decisions rely on atmospheric models, yet current models are inconsistent in how they simulate oxidising capacity. DeTOX will transform understanding of atmospheric oxidation in the tropics. By uncovering how physical, chemical and meteorological processes interact to control oxidation, DeTOX will provide the first robust attribution of the mechanisms shaping atmospheric composition.
Depression is a leading cause of ill-health and disability worldwide. Different forms of depression are associated with distinct patterns of brain activity and stress hormone levels. Understanding these differences may help to explain why treatments are effective for some patients but not others.
This fellowship will use mathematical modelling to investigate how the body's stress hormone system interacts with brain regions involved in emotion and memory. It will develop computer models of these interconnected systems and analyse them using a mathematical technique known as pseudospectra. This approach reveals how disturbances, such as a stressful event, can produce significant temporary changes within a system.
The West Antarctic Ice Sheet contains enough ice to raise global sea levels by up to five metres. The rate at which it will melt remains one of the greatest uncertainties in climate science, presenting significant challenges for coastal planning and policy.
IceDice aims to improve the reliability of ice melt predictions. Rather than producing a single forecast, the project will estimate the full range of possible outcomes and the likelihood of each, including rare but severe scenarios of rapid sea level rise.
