Auto-encoder

Pre-trained self-supervised learning

Reconstruction(cycle GAN)

Dimension reduction

De-noising auto-encoder

Feature disentangle

Different dimension represent different information

Voice conversion

Generator

Anomaly detection