AI Blog

AI Blog

by Michele Laurelli

VAE (Variational Autoencoder)

/viː eɪ iː/
Architecture
Definition

A generative model that learns a probabilistic latent space representation of data.

VAEs encode data into a distribution in latent space (typically Gaussian) and decode samples back to data space. They enable controllable generation and interpolation in latent space.

Examples

1

Image generation

2

Anomaly detection

3

Data compression

Michele Laurelli - AI Research & Engineering