AI Glossary

Variational autoencoder

Also called VAE

What does “Variational autoencoder” mean?

Variational autoencoder: Compresses input into a compact latent representation and then reconstructs it, while keeping that latent space smooth and organized. Sampling from the latent space lets it generate new data similar to what it was trained on.

Variational autoencoder visual guide: Compresses input into a compact latent representation and then reconstructs it, while keeping that latent space smooth and organized. Sampling from the latent space lets it generate new data similar to what it was trained on.
Visual guide Open full size (new tab)

What is an example of “Variational autoencoder”?

Latent diffusion systems such as Stable Diffusion use a VAE to shrink images into a smaller latent form, making generation far cheaper than working on full-size pixels.

Keep learning

See also

Latent spaceTermLatent diffusionTermEncoderTermDecoderTerm

Same topic

Diffusion modelTermGenerative adversarial networkTermVision transformerTermTransformerTermVision-language modelTerm