Machine Learning

What it covers

• Generative models: LDA/QDA, nearest mean, Parzen windows, k-NN, Naive Bayes
• Discriminative models: logistic regression, SVM, decision trees, basic MLPs
• Model evaluation: cross-validation, ROC curves, the bias-variance tradeoff, gradient descent
• Unsupervised learning: k-means, hierarchical clustering, PCA
• A dedicated module on algorithmic bias and ML ethics

Assessment

100% a single on-campus computer exam on WebLab, closed-book except for Python/NumPy documentation. Mixes theory questions (multiple-choice, short-answer, mirroring the weekly WebLab exercises) with implementation questions in Python (mirroring the lab notebooks). Pass mark is 5.8; one resit per year.

Worth knowing

• Heavy reliance on first-year calculus (gradients), linear algebra (eigenvectors for PCA) and probability (Bayes) — a shaky foundation there hurts here immediately.
• The weekly Jupyter/WebLab exercises are optional and ungraded, but the exam mirrors their style directly — skipping them means skipping exam prep, not saving time.
• Bishop's textbook is the authoritative source for notation and derivations; lecture slides follow it closely.