Lecturer Guidance

One-page Summary for Invited Lecturers

These briefs are intended to help invited lecturers prepare sessions at a consistent level and in a consistent style across the week.

The audience is technically strong, numerate, and able to code, but new to modern machine learning. Lectures should therefore be intuitive first, practically grounded, explicit about limitations, and closely linked to engineering use in nuclear thermal hydraulics.


General Guidance

Item Guidance
Audience Researchers and engineers in nuclear thermal hydraulics with strong technical backgrounds but limited prior ML experience.
Teaching level Conceptual and formula-aware, but not proof-heavy. Assume participants can run and edit Jupyter notebooks, but do not assume ML jargon or advanced statistics.
Morning structure Two 50-minute lecture blocks plus a worked example / pre-lab ramp.
Afternoon structure Guided lab, checkpoint, partially independent work, extension task, consolidation, and wrap-up.
Lecture block Normally around 50 minutes. Where relevant, lecturers may also contribute to the worked example / pre-lab ramp.
Lab block Full afternoon practical session. Labs should be scaffolded, stable, and runnable on standard laptops without heavy compute.
Common expectation Explain why the method matters, how it works, what can go wrong, and how it should be used responsibly.
Avoid Overly theoretical derivations, excessive scope drift, unstable code, large compute requirements, and unnecessary new frameworks.

Programme Context

The summer school focuses on:

The main running example is:

Critical Heat Flux (CHF) prediction / structured thermal-hydraulics data

The school is not intended to be a broad AI survey.


Session Briefs

Day 0 — Self-learning Technical Onboarding

Format: Self-learning preparation pack before Day 1, with virtual organiser support.
Expected time: 60–90 minutes.

Purpose

Ensure every participant can run the computing environment, open the notebooks, load the dataset, and understand terminology used throughout the week.

In Scope

Outputs


Day 1 AM — Framing Engineering ML Problems

Purpose

Establish what kind of ML the school covers, where it fits in NTH, and what correct workflow discipline looks like.

In Scope

Learning Outcomes

Participants should be able to frame an engineering prediction problem as supervised learning, explain leakage, choose suitable regression metrics, and distinguish in-domain from out-of-domain use.


Day 1 PM — Lab 1: Baseline CHF Regression Model

Purpose

Build the first correct end-to-end workflow before introducing more complex models.

Outputs


Day 2 — Tree Ensembles

In Scope

Outputs


Day 3 — Surrogate Modelling and Gaussian Processes

In Scope

Outputs


Day 4 — Neural Networks and Physics-based Guardrails

In Scope

Outputs


Day 5 — VVUQ, Decision Support, and Optimisation

In Scope

Learning Outcomes

Participants should understand how uncertainty can inform decisions, recognise the risks of optimisation without validity checks, and communicate model limitations responsibly.


Common Guidance for All Lecturers

Every Lecture Should Explain

Every Lab Should Include

Preferred Technical Style