Learning Hierarchical Priors in VAEs

Abstract

We propose to learn a hierarchical prior in the context of variational autoencoders to avoid the over-regularisation resulting from a standard normal prior distribution. To incentivise an informative latent representation of the data, we formulate the learning problem as a constrained optimisation problem by extending the Taming VAEs framework to two-level hierarchical models. We introduce a graph-based interpolation method, which shows that the topology of the learned latent representation corresponds to the topology of the data manifold---and present several examples, where desired properties of latent representation such as smoothness and simple explanatory factors are learned by the prior.

Published at NeurIPS 2019 (spotlight)

References in corpus (9)

• Auto-Encoding Variational Bayes
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Variational Inference with Normalizing Flows
• Deep Unsupervised Clustering with Gaussian Mixture Variational Autoencoders
• InfoVAE: Information Maximizing Variational Autoencoders
• Variational Lossy Autoencoder
• Reinterpreting Importance-Weighted Autoencoders
• Tackling Over-pruning in Variational Autoencoders
• Taming VAEs

Cited by in corpus (4)

• Generative Latent Flow
• Hierarchical Variational Auto-Encoding for Unsupervised Domain Generalization
• MaskAAE: Latent space optimization for Adversarial Auto-Encoders
• Multi-Task Neural Processes