MACHINE LEARNING IN ENGINEERING SCIENCE / FALL 2026

One exercise for every class.

16 classroom exercises and 2 optional examination rehearsals. Week 07 has a separate exercise for each of its three one-hour modules. Holidays have no new exercise. These are formative practice, not additional weekly graded submissions.

Each exercise includes a complete worked notebook, detailed instructions, three prediction/experiment/explanation tasks, self-checks, and an evidence export. All use CPU-friendly free tools.

Course website Exercise manifest

W01 / 2026-09-07 / 35 MIN

One exercise for every class.

Measurements, missing values, and provenance

Leakage-safe preprocessing and feature meaning

Baselines, target types, and model assumptions

Confusion matrices, thresholds, and prevalence

PCA, clusters, and what geometry can establish

LSTM memory, safe windows, and honest comparisons

Attention arithmetic, masks, and interpretation

Trace and test a tiny causal Transformer

Find failures hidden by an aggregate metric

Reproduce a frozen experiment

Optional midterm written and oral rehearsal

Validate model proposals before tool execution

Retrieve evidence and abstain responsibly

Agent budgets, permissions, and failure traces

A traceable bounded analysis request

Build an evaluation set that exposes a bug

Package evidence and defend its scope

Optional final written and oral rehearsal