This research fellow post sits within the FMAI4Air project at the University of Surrey, which seeks a foundation model for predicting air pollution and its health effects. The work centres on designing and pre-training a physics-informed model that fuses satellite, sensor, meteorological and emissions data, then adapting it to wildfire early warning and urban mitigation scenarios. The role draws on national and international HPC facilities and a consortium of universities, industry and local government. Fit depends on large-scale deep learning experience and a record of top-tier publications.
What the job involves
- Design and pre-train a task-agnostic, multimodal, high-resolution, physics-informed foundation model for air quality prediction
- Fuse satellite imagery, surface sensor, meteorological grid and emissions inventory data within the model
- Evaluate the impacts of air pollution on people's health
- Embed physical constraints such as advection-diffusion equations into the model architecture
- Apply super-resolution and downscaling to turn coarse global data into street-level insights
- Fine-tune the pretrained model for wildfire early-warning systems and urban air-quality mitigation scenarios
- Build a web-based proof-of-concept application to inform government policy
- Pretrain and fine-tune models on HPC infrastructure including AI@Surrey and Isambard-AI
- Collaborate with academic, industry and government consortium partners nationally and globally
- Publish research in top-tier venues and support research, dissemination, communication and management activities
What’s required
Experience
- Designing and training large-scale deep learning architectures such as Transformers, Mamba or similar spatiotemporal models
- Proven multimodal data fusion across time-series, imagery and tabular data
- Developing and deploying web applications to operationalise models for policy and decision making
- Track record of publishing high-quality research in top-tier AI or environmental science venues
Skills & knowledge
- Strong proficiency in PyTorch and TensorFlow
- Deploying models on HPC clusters and at the edge for fast, efficient inference
- Solid grounding in calculus and linear algebra
- Implementing physics-informed constraints in neural network loss functions
#J-18808-Ljbffr
Contact Details:
The Methodist Church in Britain Recruitment Team
