Clustering in a Latent Space

4190.428 Lecture 11 · K-means (k = 10) on 60,000 Fashion-MNIST training images · t-SNE?t-SNE: a technique that finds a better low-dimensional space to look at the data — images that are close in the feature space stay close in the 2-D map (distances between far-apart groups mean little). Not covered in this course.
van der Maaten & Hinton, JMLR 2008 →
of 2,000 of them
K-means on f(x) — accuracy
40%50%60%
Accuracy
NMI
Color
Cluster
click a point