Decoder. plain-English AI glossary

VAE

● Core

Also called Variational Autoencoder

An autoencoder that compresses data into a smooth, organized space you can sample new things from.

Think of it like

Like a librarian who not only files every book but arranges the shelves so similar books sit together — so you can wander to a nearby spot and find something plausible.

Example

Most modern image diffusion models actually run inside a VAE’s compressed space to save compute, not on raw pixels.

How it actually works

It squeezes input into a probability distribution over a small latent space, then decodes samples back out. The “variational” part forces that latent space to be smooth and continuous, so nearby points decode to similar outputs and you can generate new samples by sampling the space. Outputs tend to be blurrier than GANs or diffusion, but the tidy latent space is the payoff.

For product teams

Often the invisible compression layer that makes big image models affordable to run.

For engineers

An autoencoder with a KL-regularized latent distribution, trained on a reconstruction-plus-divergence objective.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome