Matches in SemOpenAlex for { <https://semopenalex.org/work/W3174319688> ?p ?o ?g. }
Showing items 1 to 85 of
85
with 100 items per page.
- W3174319688 abstract "We propose a theoretical approach towards the training numerical stability of Variational AutoEncoders (VAE). Our work is motivated by recent studies empowering VAEs to reach state of the art generative results on complex image datasets. These very deep VAE architectures, as well as VAEs using more complex output distributions, highlight a tendency to haphazardly produce high training gradients as well as NaN losses. The empirical fixes proposed to train them despite their limitations are neither fully theoretically grounded nor generally sufficient in practice. Building on this, we localize the source of the problem at the interface between the model's neural networks and their output probabilistic distributions. We explain a common source of instability stemming from an incautious formulation of the encoded Normal distribution's variance, and apply the same approach on other, less obvious sources. We show that by implementing small changes to the way we parameterize the Normal distributions on which they rely, VAEs can securely be trained." @default.
- W3174319688 created "2021-07-05" @default.
- W3174319688 creator A5059183826 @default.
- W3174319688 creator A5059942051 @default.
- W3174319688 date "2021-06-25" @default.
- W3174319688 modified "2023-09-27" @default.
- W3174319688 title "Re-parameterizing VAEs for stability." @default.
- W3174319688 cites W1710476689 @default.
- W3174319688 cites W1959608418 @default.
- W3174319688 cites W2911290044 @default.
- W3174319688 cites W2962770929 @default.
- W3174319688 cites W2963135265 @default.
- W3174319688 cites W2963857374 @default.
- W3174319688 cites W2971612319 @default.
- W3174319688 cites W3036520878 @default.
- W3174319688 cites W3041956526 @default.
- W3174319688 cites W3048815401 @default.
- W3174319688 cites W3106570356 @default.
- W3174319688 cites W3118608800 @default.
- W3174319688 hasPublicationYear "2021" @default.
- W3174319688 type Work @default.
- W3174319688 sameAs 3174319688 @default.
- W3174319688 citedByCount "0" @default.
- W3174319688 crossrefType "posted-content" @default.
- W3174319688 hasAuthorship W3174319688A5059183826 @default.
- W3174319688 hasAuthorship W3174319688A5059942051 @default.
- W3174319688 hasConcept C112972136 @default.
- W3174319688 hasConcept C113843644 @default.
- W3174319688 hasConcept C11413529 @default.
- W3174319688 hasConcept C115961682 @default.
- W3174319688 hasConcept C119857082 @default.
- W3174319688 hasConcept C121955636 @default.
- W3174319688 hasConcept C129307140 @default.
- W3174319688 hasConcept C144133560 @default.
- W3174319688 hasConcept C154945302 @default.
- W3174319688 hasConcept C157915830 @default.
- W3174319688 hasConcept C173608175 @default.
- W3174319688 hasConcept C196083921 @default.
- W3174319688 hasConcept C39890363 @default.
- W3174319688 hasConcept C41008148 @default.
- W3174319688 hasConcept C49937458 @default.
- W3174319688 hasConcept C50644808 @default.
- W3174319688 hasConceptScore W3174319688C112972136 @default.
- W3174319688 hasConceptScore W3174319688C113843644 @default.
- W3174319688 hasConceptScore W3174319688C11413529 @default.
- W3174319688 hasConceptScore W3174319688C115961682 @default.
- W3174319688 hasConceptScore W3174319688C119857082 @default.
- W3174319688 hasConceptScore W3174319688C121955636 @default.
- W3174319688 hasConceptScore W3174319688C129307140 @default.
- W3174319688 hasConceptScore W3174319688C144133560 @default.
- W3174319688 hasConceptScore W3174319688C154945302 @default.
- W3174319688 hasConceptScore W3174319688C157915830 @default.
- W3174319688 hasConceptScore W3174319688C173608175 @default.
- W3174319688 hasConceptScore W3174319688C196083921 @default.
- W3174319688 hasConceptScore W3174319688C39890363 @default.
- W3174319688 hasConceptScore W3174319688C41008148 @default.
- W3174319688 hasConceptScore W3174319688C49937458 @default.
- W3174319688 hasConceptScore W3174319688C50644808 @default.
- W3174319688 hasLocation W31743196881 @default.
- W3174319688 hasOpenAccess W3174319688 @default.
- W3174319688 hasPrimaryLocation W31743196881 @default.
- W3174319688 hasRelatedWork W2557513777 @default.
- W3174319688 hasRelatedWork W2681861910 @default.
- W3174319688 hasRelatedWork W2795026356 @default.
- W3174319688 hasRelatedWork W2891401260 @default.
- W3174319688 hasRelatedWork W2900208959 @default.
- W3174319688 hasRelatedWork W2905266595 @default.
- W3174319688 hasRelatedWork W2949342585 @default.
- W3174319688 hasRelatedWork W2963142510 @default.
- W3174319688 hasRelatedWork W2965141076 @default.
- W3174319688 hasRelatedWork W2971246770 @default.
- W3174319688 hasRelatedWork W2992535490 @default.
- W3174319688 hasRelatedWork W3091324939 @default.
- W3174319688 hasRelatedWork W3115985185 @default.
- W3174319688 hasRelatedWork W3122023385 @default.
- W3174319688 hasRelatedWork W3159019720 @default.
- W3174319688 hasRelatedWork W3166037492 @default.
- W3174319688 hasRelatedWork W3169779692 @default.
- W3174319688 hasRelatedWork W3178129852 @default.
- W3174319688 hasRelatedWork W3213436811 @default.
- W3174319688 hasRelatedWork W3214266893 @default.
- W3174319688 isParatext "false" @default.
- W3174319688 isRetracted "false" @default.
- W3174319688 magId "3174319688" @default.
- W3174319688 workType "article" @default.