Generative & Distribution Matching
Generative models learn the distribution of data and can generate new samples from it. This base brings several families together: GANs / WGAN, VAE, WAE, AAE, and EVIA — the entropic-OT autoencoder — plus self-supervised learning, the toolbox that turns missing-wedge restoration into a generative problem.
Articles in this base 6 articles
GANs and the Wasserstein GAN
Training a generator against a critic, and replacing the classifier with a Wasserstein-distance critic for stable gradients.
Variational Autoencoder (VAE)
A latent-variable generative model trained by maximizing a variational lower bound on the data likelihood, with an amortized encoder and the reparameterization trick.
Wasserstein Autoencoder (WAE)
An autoencoder derived from the optimal-transport view, matching the aggregated posterior to the prior with an MMD or adversarial penalty rather than a per-sample KL.
Adversarial Autoencoder (AAE)
An autoencoder that matches its aggregated posterior to a prior with an adversarial discriminator instead of a KL penalty, placing it between GANs and autoencoders.
Entropic Variational Inference Auto-encoding (EVIA)
Matching the aggregated posterior with entropic optimal transport to obtain a Boltzmann posterior and a soft barycentric encoder — a stochastic generalization of the WAE
Self-supervised learning
Learning representations and solving inverse problems without labels, by constructing supervision from the data itself.