Unsupervised learning aims at learning some underlying hidden structure of the data when we don’t have the labels.

Unsupervised models can be used as a pre step for supervised learning, e.g.:

• reduce the training sample (dimensionality reduction: forward selection, PCA, autoencoders)

• give output for unlabeled data (clustering, autoencoders)

• grow the training sample (generative models)