variational autoencoder
autoencoder
semi-supervised & self-organised︎ unlabelled dataset
︎ automatic feature extraction
variational
encodes data as vectors, not points︎ space between encoded samples is meaningful
︎ ability to generate new data
latent space
compressed statistical representation of dataset︎ efficient, fast calculations (eg. interpolation)
The specific VAE I am training at the moment is the Oobleck VAE by Stability AI.
- Evans et al, 2024