What is a Variational Autoencoder (VAE)?

A VAE is a probabilistic autoencoder that learns a distribution over latent space.

Unlike standard autoencoders that learn deterministic mappings, VAEs learn a probability distribution (mean and variance) over latent variables. The encoder outputs parameters of a distribution, and sampling is done using the reparameterization trick. It optimizes reconstruction loss + KL divergence to enforce structured latent space.