Hands-On Machine Learning with Scikit-Learn and PyTorch

Chapter 18. Autoencoders, GANs, and Diffusion Models

Autoencoders are artificial neural networks capable of learning dense representations of the input data, called latent representations or codings, without any supervision (i.e., the training set is unlabeled). These codings typically have a much lower dimensionality than the input data, making autoencoders useful for dimensionality reduction (see Chapter 7), especially for visualization purposes. Autoencoders also act as feature detectors, and they can be used for unsupervised pretraining of deep neural networks (as we discussed in Chapter 11). They are also commonly used for anomaly detection, as we will see. Lastly, some autoencoders are generative models: they are capable of randomly generating new data that looks very similar to the training data. For example, you could train an autoencoder on pictures of faces, and it would then be able to generate new faces.

Generative adversarial networks (GANs) are also neural nets capable of generating data. In fact, they can generate pictures of faces so convincing that it is hard to believe the people they represent do not exist. You can judge for yourself by visiting https://thispersondoesnotexist.com, a website that shows faces generated by a GAN architecture called StyleGAN. GANs have been widely used for super resolution (increasing the resolution of an image), colorization, powerful image editing (e.g., replacing photo bombers with realistic background), turning simple sketches ...