Unit 5 · Assessed seriously
Finding structure without labels, and the obligations that come with deploying any of this. The responsibility material is assessed with the same seriousness as the technical content.
What a strong answer looks like
A strong Unit 5 answer names a specific group, a specific mechanism, and a specific consequence rather than judging a technology in general.
Know
Clustering groups similar examples without labels. k-means needs the cluster count in advance and converges to a local optimum, so runs differ.
Apply
Choose k with a method such as the elbow or silhouette, and run several times.
Watch out
Treating cluster labels as meaningful categories. Any names are assigned by a human afterwards.
Study move
Explain why two k-means runs on the same data can differ.
Know
Reducing dimensions keeps most of the variation in fewer features, which speeds training and enables visualisation at the cost of some information.
Apply
Use it to visualise high-dimensional data or reduce noise before modelling.
Watch out
Expecting reduced components to be interpretable features. They are combinations, not columns.
Study move
Say what is gained and what is lost when dimensions are reduced.
Know
A model trained on historical decisions reproduces the bias in those decisions. The cause is the data and the framing, not the arithmetic.
Apply
Evaluate performance per group, not only in aggregate.
Watch out
Assuming that removing a sensitive attribute removes bias. Correlated features carry it anyway.
Study move
Explain how a hiring model can discriminate without ever seeing the protected attribute.
Know
A decision affecting someone must be explainable and appealable. Documentation of training data tells a user where predictions are unreliable.
Apply
Provide a route for human review, and document what the model was built on.
Watch out
Deploying an unexplainable model in a high-stakes setting because it scored slightly higher.
Study move
For a described deployment, name the safeguard you would insist on and why.
Connections and techniques that receive extra attention in this unit.
Related topics some schools attach to this unit and others leave out. Covered on request rather than assumed.
Modeling drill
A school proposes a model predicting which students need intervention. Name who benefits, who bears risk, what you would measure per group, and what safeguard you would require.
