Overview
This summer school introduces machine learning as a careful engineering tool for nuclear thermal hydraulics, with emphasis on problem framing, correct workflow, validation, uncertainty, domain validity, and safe use.
Structured Data ML
Supervised learning workflows for thermal-hydraulics tabular datasets, including baselines and tree-based models.
Surrogate Modelling
Fast emulators for expensive simulations, including Gaussian Processes and safe-domain reasoning.
Decision Support
Uncertainty quantification, model validity, and optimisation workflows for engineering decision-making.
Core Case Study
Who Should Attend?
The school is intended for researchers, PhD students, and practical engineers in nuclear thermal hydraulics with strong modelling or computational backgrounds but limited prior experience in modern machine learning.
Participants should be comfortable with numerical modelling and basic Python/Jupyter notebooks. Prior deep learning, advanced statistics, or GPU programming experience is not required.
Weekly Structure
Day 1
Problem framing, supervised ML, workflow discipline, and baseline models.
Day 2
Random Forest, Gradient Boosting, model comparison, and regime-wise diagnostics.
Day 3
Surrogate modelling, Gaussian Processes, uncertainty, and extrapolation tests.
Day 4
Neural networks for tabular data and lightweight physics-guided guardrails.
Day 5
VVUQ, uncertainty-aware decisions, optimisation, guest talks, and closing discussion.
Day 0
Self-learning onboarding pack to verify environment, notebooks, and datasets.
Delivery Notes
- Runs on standard laptops / PCs.
- No national HPC resources are required.
- Pre-configured environment and notebooks will be provided.
- All notebooks should be tested end-to-end before the school.
Logos and Sponsors
Replace the placeholder logo in assets/logo.svg, or add additional logos below.
Contacts
- Professor Michael Bluck — m.bluck@imperial.ac.uk
- Dr Wei Wang — wei.wang@stfc.ac.uk
- Dr Yu Duan — yu.duan@sheffield.ac.uk / y.duan@imperial.ac.uk