Variational Autoencoder (VAE)

VAEs combine neural networks with probabilistic inference, learning a continuous latent space where similar data points are grouped together. They are used in drug discovery (generating novel molecular structures), image synthesis, and anomaly detection. Stable Diffusion uses a VAE as part of its architecture to compress images into a latent space before applying diffusion. While VAEs produce less sharp images than GANs or diffusion models, they provide a mathematically principled framework for learning data distributions and offer smooth interpolation between generated samples.