Diederik P Kingma & Max Welling (2013)
arXiv.
DOI: https://doi.org/10.48550/arxiv.1312.6114
Abstract. Introduces the Variational Autoencoder (VAE), which imposes a probabilistic structure on the latent space of an autoencoder and trains it by optimising the evidence lower bound. The reparameterisation trick enables end-to-end backpropagation through sampling.
Tags: generative vae variational-inference