Models that partition the dataset into semantically meaningful clusters without having access to the ground truth labels.
<span style="color:grey; opacity: 0.6"> Image credit: ImageNet clustering results of SCAN: Learning to Classify Images without Labels (ECCV 2020) </span>