install.packages(c("tidyverse", "tidymodels", "here"))AI and Application
Five modules, building from “what is a model” to “should I have used one”.
About this section
Section 1 argued that method choice follows from the question. This section teaches the methods themselves — enough to use them, and enough to know when not to.
Every module uses real data. Where a method is a poor fit for the data we have, we say so rather than pretending otherwise.
The five modules
| Module | Core idea | |
|---|---|---|
| 1 | Foundations of Machine Learning | Train/test splits, overfitting, why a baseline comes first |
| 2 | Supervised Learning | Classification, and why accuracy misleads |
| 3 | Unsupervised Learning | Clustering and dimensionality reduction; finding structure that isn’t there |
| 4 | Deep Learning & Sequence Models | What neural networks buy you, and what they cost |
| 5 | LLMs in Research | Using AI assistants without damaging your science |
How to read this section
The modules are cumulative. Module 1’s train/test discipline is assumed everywhere after it.
ImportantA running theme
Each module ends by asking whether a simpler method would have done as well. Surprisingly often the answer is yes. That is a real result, not a failure.
Prerequisites
R and the tidyverse. If you worked through the exercise in Section 1, you have the conceptual grounding you need.
NoteStatus
Module content is under active development for this workshop. The structure and learning objectives below are settled; worked examples are being finalised against the workshop datasets.