Variational Autoencoder
Also called VAE
The full name for a VAE — an autoencoder whose middle is a smooth, sampleable space.
Think of it like
A vending machine that not only stores snacks but lays them out so you can dial to a spot between two and get a sensible blend.
Example
When a latent-diffusion image model “decodes” its result to pixels, that decoder is a VAE’s.
How it actually works
It learns to encode inputs as distributions rather than single points, and a regularization term keeps that latent space continuous and well-behaved. That lets you generate by sampling and interpolate between examples smoothly. The trade-off is softer, less crisp reconstructions than adversarial or diffusion approaches.
For product teams
The quiet workhorse compressing data so pricier generative steps stay cheap.
For engineers
Identical to VAE: variational inference over a latent code with an encoder–decoder pair.
Related
- VAE — The everyday short name.
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