Lecture 14: K-Means
Unsupervised learning, clustering, representation learning, K-Means, and choosing useful clusters
Slides
Outline
- Learning without targets: clustering and representation learning
- K-Means: assignment, centroid updates, and prediction
- Initialization, scaling, and feature representation
- Choosing a useful number of clusters with elbow and silhouette diagnostics
- Applying K-Means to image representations
Interactive walkthrough
Explore K-Means: assign, average, repeat
The HTML slides include an interactive walkthrough. The PDF contains a static worked example.