Failure Modes of Variational Autoencoders and Their Effects on Downstream Tasks
Variational Auto-encoders (VAEs) are deep generative latent variable models that are widely used for a number of downstream tasks. While it has been demonstrated that VAE training can suffer from a number of pathologies, existing literature lacks characterizations of exactly when these pathologies o...
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Artículo |
| Publicado: |
2020
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2007.07124 http://arxiv.org/abs/2007.07124 |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|