Variational autoencoders

tags

Neural networks

resources

(Bishop 1994)

Variational autoencoders (VAEs) are a type of generative Autoencoders.

They use a Bayesian latent encoding for the input dataset.

VAEs vs. GANs

VAEs have fallen out of fashion when GANs became popular, because they were able to get visually interesting results more easily. However, some works a few years later seem to show that they have similar potential (Vahdat, Kautz 2020).

Bibliography

1. Christopher M. Bishop. 1994. "Mixture Density Networks". Aston University.
2. Arash Vahdat, Jan Kautz. July 8, 2020. "NVAE: A Deep Hierarchical Variational Autoencoder". Arxiv:2007.03898 [cs, Stat]. http://arxiv.org/abs/2007.03898.