Variational Autoencoder (VAE)
Definition
Variational autoencoders learn compressed representations from which new variants can be generated.
A latent generative model that learns an approximate posterior and optimizes reconstruction with regularization towards the prior distribution.
Simply explained
The model describes examples in a simplified internal form and learns to reconstruct similar examples from it.
Why it matters in practice
The term helps to select, develop, evaluate AI systems in a well-founded manner and realistically assess their technical risks.
Example from business
A manufacturer learns a latent space for controlled variants of simple component images.
Limits and pitfalls
- The ELBO is not the exact likelihood; posterior collapse and blurry reconstructions are possible issues.
Origin & history
Kingma and Welling introduced the reparameterized variational autoencoder in 2013.
Use in mid-sized companies
- Possible application: A manufacturer learns a latent space for controlled variants of simple component images.
Frequently asked questions
What does Variational Autoencoder (VAE) mean?
A latent generative model that learns an approximate posterior and optimizes reconstruction with regularization towards the prior distribution. What should one watch out for? The ELBO is not the exact likelihood; posterior collapse and blurry reconstructions are possible issues.
Sources
- Auto-Encoding Variational Bayes · Kingma and Welling, Auto-Encoding Variational Bayes
Source checked on 10.10.2026. Editorial application examples are hypothetical; no proof of real customer results.
Editorially reviewed on: