Research Fellow in Multimodal Physics-Informed Foundation Model for Air Quality Applications

Do you want to build the first-ever Foundation Model for air quality? We are looking for a world-class AI researcher to join the FMAI4Air project, a high-stakes mission to revolutionise how we predict air pollution, evaluating its impacts on people's health and protecting millions of lives.

This isn't just another incremental AI project; we are pioneering a Physics-Informed Multimodal High-Resolution Foundation Model. You will be at the intersection of deep learning and atmospheric sciences, working within the prestigious and world-leading ecosystem of Surrey's PAI, CVSSP and GCARE, under the leadership of Assoc Prof Erick Sperandio.

Key responsibilities:

- Architect the Future of AI for Environmental Sciences: Design and pre-train a novel, task-agnostic, multimodal, high-resolution and physics-informed Foundation Model (FM) for air quality prediction that fuses heterogeneous data, from satellite imagery and surface sensors to meteorological grids and emissions inventories, and evaluates the impacts of air pollution on people's health.

- Bridge AI and Atmospheric Physics: Move beyond the 'black box' by embedding physical constraints (such as advection-diffusion equations) directly into the model architecture to ensure scientific consistency, interpretability and trustworthiness.

- Push the Resolution Limit: Implement state-of-the-art super-resolution and downscaling techniques to translate coarse global data into street-level, actionable insights.

- Solve Real-World Grand Challenges: Fine-tune the pretrained FM for critical downstream applications, including early-warning systems for wildfires and simulating urban air-quality mitigation scenarios to inform government policy through a web-based application for FMAI4Air proof-of-concept.

- Leverage High-Performance Computing: Pretrain and fine-tune the models on world-leading HPC infrastructure, including Surrey's AI@Surrey and the national Isambard-AI HPC.

- Collaborate Nationally and Globally: Work alongside a powerhouse consortium including Imperial College London, the University of Manchester, and NTNU (Norway), as well as with consortium partners from academia including IAG-USP (Brazil), NC Climate Office at NCSU (USA), Chiba University (Japan), from industry including NVIDIA and WSP Global, and from government such as Guildford Borough Council.

About you:

- Expertise in designing and training large-scale Deep Learning architectures (e.g., Transformers, Mamba, or similar spatiotemporal models).

- Proven experience in multimodal data fusion (handling a mix of time-series, imagery, and tabular data).

- Strong proficiency in PyTorch and TensorFlow, and experience deploying models on HPC clusters and on the edge for fast and efficient inference.

- Expertise in developing and deploying web applications that will be used to operationalise FMAI4Air models for policy and decision making.

- A solid mathematical foundation in calculus and linear algebra, with the ability to implement physics-informed constraints into neural networks' loss functions.

- A track record of publishing high-quality research in top-tier AI or environmental science conferences/journals.

- Commitment to deliver high-quality research and innovation outcomes, including publishing papers in top-tier venues, and support research, dissemination, communication and management activities.