A Contrastive Learning Approach for Training Variational Autoencoder Priors

V ariational autoencoders (V AEs) are one of the powerful likelihood-based generative models with applications in many domains.