Week 9 Seminar: Model audit

Running Wren's stroke-level classifier against the semester's shortlist, then trying to break it.

3 May 2027

Before the seminar

No reading. Bring a laptop if you have one; the seminar shares a small number if not.

In the seminar

Wren runs the stroke-level classifier from lecture live against the semester’s shortlisted paintings and a few securely attributed controls. Working in pairs, try to construct an input — a cropped detail, a mirrored image, a securely attributed painting from outside the training set — that makes the model confidently wrong, and note what that reveals about what it actually learned.

Afterwards

Keep your “broke the model” example. If your final project touches on computational methods at all, it needs to state their limits as concretely as this seminar just demonstrated one, not as a general disclaimer.

The spec

What you bring to the session. Written so a reader can tell whether it has been met without asking you.

you can state what the classifier's training set actually contained, and one way that could bias its output
you have found at least one input that makes the classifier confidently wrong

Teaching team

Wren Okafor — guest teacher, week 9