Finding density-based clusters and identifying noise points.

Creating labels for supervised learning without any data.

Replacing feature engineering in every problem.

Fitting a straight line to predict a continuous target only.