Variational Autoencoder

[JacobGlennAyers]

Train a basic non-convex variational autoencoder. Once it is trained, remove the basic decoder, keep the encoder, and retrain the decoder using a vNN

[JacobGlennAyers]

We could also try this both ways. Train a basic non-convex VAE, remove the encoder, keep the decoder, and then retrain the encoder using a vNN

[JacobGlennAyers]

If this works, we could then generate FID and IS scores.