Variational Autoencoders with Riemannian Brownian Motion Priors
| Autor: | Dimitrios Kalatzis, David Eklund, Georgios Arvanitidis, Søren Hauberg |
|---|---|
| Přispěvatelé: | Daume, Hal, Singh, Aarti |
| Rok vydání: | 2020 |
| Předmět: | |
| Zdroj: | Technical University of Denmark Orbit Kalatzis, D, Eklund, D, Arvanitidis, G & Hauberg, S 2020, Variational Autoencoders with Riemannian Brownian Motion Priors . in H Daume & A Singh (eds), Proceedings of the 37 th International Conference on Machine Learning . vol. 119, International Machine Learning Society (IMLS), pp. 5020-5033, 37 th International Conference on Machine Learning, Virtual, Online, 13/07/2020 . Proceedings of the 37th International Conference on Machine Learning (ICML 2020) Proceedings of Machine Learning Research (PMLR) |
| DOI: | 10.48550/arxiv.2002.05227 |
| Popis: | Variational Autoencoders (VAEs) represent the given data in a low-dimensional latent space, which is generally assumed to be Euclidean. This assumption naturally leads to the common choice of a standard Gaussian prior over continuous latent variables. Recent work has, however, shown that this prior has a detrimental effect on model capacity, leading to subpar performance. We propose that the Euclidean assumption lies at the heart of this failure mode. To counter this, we assume a Riemannian structure over the latent space, which constitutes a more principled geometric view of the latent codes, and replace the standard Gaussian prior with a Riemannian Brownian motion prior. We propose an efficient inference scheme that does not rely on the unknown normalizing factor of this prior. Finally, we demonstrate that this prior significantly increases model capacity using only one additional scalar parameter. Comment: Published in ICML 2020 |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|