Matches in SemOpenAlex for { <https://semopenalex.org/work/W3107355992> ?p ?o ?g. }
- W3107355992 abstract "Contrastive divergence is a popular method of training energy-based models, but is known to have difficulties with training stability. We propose an adaptation to improve contrastive divergence training by scrutinizing a gradient term that is difficult to calculate and is often left out for convenience. We show that this gradient term is numerically significant and in practice is important to avoid training instabilities, while being tractable to estimate. We further highlight how data augmentation and multi-scale processing can be used to improve model robustness and generation quality. Finally, we empirically evaluate stability of model architectures and show improved performance on a host of benchmarks and use cases,such as image generation, OOD detection, and compositional generation." @default.
- W3107355992 created "2020-12-07" @default.
- W3107355992 creator A5014193325 @default.
- W3107355992 creator A5041777043 @default.
- W3107355992 creator A5061084605 @default.
- W3107355992 creator A5071093940 @default.
- W3107355992 date "2020-12-02" @default.
- W3107355992 modified "2023-09-25" @default.
- W3107355992 title "Improved Contrastive Divergence Training of Energy Based Models" @default.
- W3107355992 cites W1505878979 @default.
- W3107355992 cites W189596042 @default.
- W3107355992 cites W2108598243 @default.
- W3107355992 cites W2116064496 @default.
- W3107355992 cites W2129069237 @default.
- W3107355992 cites W2267126114 @default.
- W3107355992 cites W2272361785 @default.
- W3107355992 cites W2336509789 @default.
- W3107355992 cites W2416272112 @default.
- W3107355992 cites W2423557781 @default.
- W3107355992 cites W2478027467 @default.
- W3107355992 cites W2527798464 @default.
- W3107355992 cites W2531327146 @default.
- W3107355992 cites W2594103415 @default.
- W3107355992 cites W2606757878 @default.
- W3107355992 cites W2728358341 @default.
- W3107355992 cites W2766363782 @default.
- W3107355992 cites W2770567114 @default.
- W3107355992 cites W2783047733 @default.
- W3107355992 cites W2788289427 @default.
- W3107355992 cites W2804123849 @default.
- W3107355992 cites W2804821933 @default.
- W3107355992 cites W2805470269 @default.
- W3107355992 cites W2922772346 @default.
- W3107355992 cites W2938608316 @default.
- W3107355992 cites W2950873402 @default.
- W3107355992 cites W2953327099 @default.
- W3107355992 cites W2962760235 @default.
- W3107355992 cites W2962879692 @default.
- W3107355992 cites W2963140844 @default.
- W3107355992 cites W2963373786 @default.
- W3107355992 cites W2963684088 @default.
- W3107355992 cites W2963836885 @default.
- W3107355992 cites W2963948515 @default.
- W3107355992 cites W2963981733 @default.
- W3107355992 cites W2964121744 @default.
- W3107355992 cites W2970747737 @default.
- W3107355992 cites W2970803184 @default.
- W3107355992 cites W2971034910 @default.
- W3107355992 cites W2978120197 @default.
- W3107355992 cites W2994434574 @default.
- W3107355992 cites W2997843561 @default.
- W3107355992 cites W3005680577 @default.
- W3107355992 cites W3006210322 @default.
- W3107355992 cites W3009318622 @default.
- W3107355992 cites W3018305985 @default.
- W3107355992 cites W3021677849 @default.
- W3107355992 cites W3030947604 @default.
- W3107355992 cites W3035166812 @default.
- W3107355992 cites W3036167779 @default.
- W3107355992 cites W3106068426 @default.
- W3107355992 cites W3107632401 @default.
- W3107355992 cites W3119691539 @default.
- W3107355992 cites W3128546090 @default.
- W3107355992 cites W2156775093 @default.
- W3107355992 cites W2523655942 @default.
- W3107355992 hasPublicationYear "2020" @default.
- W3107355992 type Work @default.
- W3107355992 sameAs 3107355992 @default.
- W3107355992 citedByCount "9" @default.
- W3107355992 countsByYear W31073559922020 @default.
- W3107355992 countsByYear W31073559922021 @default.
- W3107355992 crossrefType "posted-content" @default.
- W3107355992 hasAuthorship W3107355992A5014193325 @default.
- W3107355992 hasAuthorship W3107355992A5041777043 @default.
- W3107355992 hasAuthorship W3107355992A5061084605 @default.
- W3107355992 hasAuthorship W3107355992A5071093940 @default.
- W3107355992 hasConcept C104317684 @default.
- W3107355992 hasConcept C112972136 @default.
- W3107355992 hasConcept C119857082 @default.
- W3107355992 hasConcept C121332964 @default.
- W3107355992 hasConcept C138885662 @default.
- W3107355992 hasConcept C139807058 @default.
- W3107355992 hasConcept C153294291 @default.
- W3107355992 hasConcept C154945302 @default.
- W3107355992 hasConcept C15744967 @default.
- W3107355992 hasConcept C169760540 @default.
- W3107355992 hasConcept C185592680 @default.
- W3107355992 hasConcept C207390915 @default.
- W3107355992 hasConcept C2777211547 @default.
- W3107355992 hasConcept C41008148 @default.
- W3107355992 hasConcept C41895202 @default.
- W3107355992 hasConcept C51632099 @default.
- W3107355992 hasConcept C55493867 @default.
- W3107355992 hasConcept C61797465 @default.
- W3107355992 hasConcept C62520636 @default.
- W3107355992 hasConcept C63479239 @default.
- W3107355992 hasConceptScore W3107355992C104317684 @default.
- W3107355992 hasConceptScore W3107355992C112972136 @default.
- W3107355992 hasConceptScore W3107355992C119857082 @default.
- W3107355992 hasConceptScore W3107355992C121332964 @default.