All MCQs

Nearest-centroid partitions create convex Voronoi regions.
Elongated, unequal-density, or nonconvex clusters can be poorly represented.

K-means depends directly on geometric distances.
Scaling ensures units do not determine feature importance unintentionally.

K-means++ spreads initial centers across the data.
It usually improves convergence and solution quality over naive random starts.

The objective is not jointly convex over assignments and centers.
Multiple starts improve the chance of finding a better solution.

Updated means minimize squared error within each fixed cluster assignment.
Assignment and update repeat until convergence.

Distance to current centroids determines cluster membership.
This step reduces or preserves the objective for fixed centroids.

Each centroid summarizes the current center of its assigned observations.
The algorithm alternates assignment and centroid updates.

K-means creates exactly the requested number of clusters.
Validation methods can help choose a reasonable k.

K-means seeks compact clusters around arithmetic means.
Its standard objective is within-cluster sum of squared errors.

Incorrect pseudo-labels become training targets and may amplify bias.
Confidence thresholds and iterative monitoring reduce the risk.

A supervised model first predicts unlabeled examples.
Selected predictions are added to training as if they were labels.

Unlabeled observations reveal the geometry and density of the input space.
Labels can then be extended along that structure.