In Advanced Research Computing, we collaborate closely with the Institute for Data and AI (IDAI) and will be featuring a series of guest posts by Joshua Panteli, who has several roles at the University of Birmingham, as Project Manager for IDAI and the Birmingham Institute for Sustainability and Climate Action (BISCA), as well as being Turing Liaison.
This year, the Institute for Data and AI (IDAI) opened a call for applications as part of our ‘Pump Prime Funding’ initiative. The funding provided the opportunity for researchers to explore new research questions that would incorporate the principled use and advancement of Data Science and AI methods. The call was open to researchers of all disciplines and departments and we were delighted to award funding to 8 Principal Investigators (PI’s), each of whom proposed an innovative and high-impact project.
This post is the first in a series that shines a spotlight on each of these amazing projects in the words of the PI’s.
Below, we discuss three of the projects:
- A radiomics approach to define the brain metastasis-brain parenchyma interface on MRI – led by Dr Vinton Cheng
- Digital Governance Twins: LLM Multi-Agent Systems for Evidence-Based Decision Support’ – led by Dr Christine Sheldon
- Multi-modal hydrological data – causal foundation AI model of hydrological system – led by Dr Shasha Han
We look forward to seeing what exciting discoveries and opportunities these projects will lead on to!
Dr Vinton Cheng – A radiomics approach to define the brain metastasis-brain parenchyma interface on MRI
I am extremely grateful to IDAI for this pump priming award, which will enable our interdisciplinary team to test new MRI analysis tools to allow precise treatment delivery for patients with secondary brain tumours. The data generated from this project will provide a vital boost for leveraging more substantive funding in the future. I am really looking forward to the outputs of this project and, one day, improving the care I can offer for my patients.
Dr Christine Sheldon – Digital governance twins: LLM multi-agent systems for evidence-based decision support
This project, supported by IDAI Pump Prime funding, explores whether artificial intelligence can simulate political decision-making, specifically focusing on how policies are negotiated within the European Council. While AI has been used in many fields, its ability to model complex human negotiations remains largely untested. This research is an exploration, not a guarantee—our goal is to see if AI-driven simulations can realistically reflect the way policy makers debate and make decisions.
A major challenge is ensuring accuracy and reliability. AI models sometimes generate misleading or biased responses, so the project includes designing and applying a rigorous validation process to test its credibility at every step. We will compare AI-generated negotiation scenarios with real-world political discussions, consult governance experts, and refine the system based on evidence. By doing so, we aim to determine whether AI can offer valuable insights into how policies take shape.
If this approach works, it could give policymakers a new tool to experiment with policy ideas before implementing them, helping them anticipate potential challenges and test different negotiation strategies. However, even if the technology falls short, the research will still be valuable—it will clarify AI’s limitations in governance modelling and set the stage for future studies at the intersection of AI and politics. Ultimately, this project is about pushing boundaries—testing the potential of AI in governance while ensuring transparency and critical evaluation. By exploring what AI can and cannot do in political decision-making, we hope to contribute to a deeper, more informed conversation about the role of AI in shaping our political future.
Dr Shasha Han – Multi-model hydrological data – causal foundation AI model of a hydrological system
Over the past few decades in the UK, climate change and human activities have significantly altered the characteristics of hydroclimatic extremes, such as droughts and floods, leading to increased frequency and intensity in many regions. Moreover, there is growing concern that the unprecedented severity and frequency of these extreme events may continue to increase. These extreme events are very costly and have significant negative impacts on the environment, society, and the economy. For instance, the UK 2018 drought resulted in losses exceeding £180 million for farmers. Meanwhile, in England and Wales alone, it is estimated that over 4 million people and over £200 billion in properties are at risk of potential flooding.
To mitigate the impacts of climate change and reduce the risk of hydro disasters, we need an in-depth understanding of hydrological processes, as well as improved simulation and prediction of hydro extreme events. However, these efforts largely depend on data quality, data length, and data resolution. In the hydrology field, many regions lack adequate monitoring stations, particularly in remote and mountainous areas. Historical records frequently contain large gaps or missing values, and the temporal and spatial resolutions are often insufficient. Additionally, there can be considerable inconsistencies in data from different sources.
Foundation AI models have strengths in handling multi-modal data and capturing complex non-linear relationships. By integrating large scale, multi-modal data and leveraging self-supervised and transfer learning, these models demonstrate satisfactory spatial-temporal simulation and predictions across domains, even with limited data. This pump prime funding will be used to support the development of a comprehensive UK national hydrological dataset to be used for training the foundation AI model. Hydrological data from multiple sources with multi-modality (e.g. time series, geospatial data, text, and image) across the UK will be collected, including river flow, precipitation, evapotranspiration, temperature, land cover, groundwater, soil moisture, reservoirs, water abstraction, catchment attributes, etc. All datasets will be quality-checked, then extracted and organised into hydrological units (i.e. catchments) using a consistent format.
By training the foundation AI model with this national-scale dataset and leveraging advanced ML techniques such as self-supervised and transfer learning, we aim to: 1) enable accurate estimations in ungauged basins, 2) improve hydrological forecasting such as floods and droughts, 3) uncover casual relationships within the hydrological systems, and 4) enhance physical understanding of hydrological processes.
Congratulations to you all!
For more information on each of these pump prime funding projects and to hear about new and exciting opportunities to engage with IDAI, please visit our website at: https://www.birmingham.ac.uk/research/centres-institutes/data-and-ai
And follow us on LinkedIn at: https://www.linkedin.com/company/institute-for-data-and-ai/?