Matches in SemOpenAlex for { <https://semopenalex.org/work/W3019505787> ?p ?o ?g. }
- W3019505787 abstract "Variational Autoencoder (VAE) is widely used as a generative model to approximate a model's posterior on latent variables by combining the amortized variational inference and deep neural networks. However, when paired with strong autoregressive decoders, VAE often converges to a degenerated local optimum known as posterior collapse. Previous approaches consider the Kullback Leibler divergence (KL) individual for each datapoint. We propose to let the KL follow a distribution across the whole dataset, and analyze that it is sufficient to prevent posterior collapse by keeping the expectation of the KL's distribution positive. Then we propose Batch Normalized-VAE (BN-VAE), a simple but effective approach to set a lower bound of the expectation by regularizing the distribution of the approximate posterior's parameters. Without introducing any new model component or modifying the objective, our approach can avoid the posterior collapse effectively and efficiently. We further show that the proposed BN-VAE can be extended to conditional VAE (CVAE). Empirically, our approach surpasses strong autoregressive baselines on language modeling, text classification and dialogue generation, and rivals more complex approaches while keeping almost the same training time as VAE." @default.
- W3019505787 created "2020-05-01" @default.
- W3019505787 creator A5001469325 @default.
- W3019505787 creator A5022526821 @default.
- W3019505787 creator A5040117777 @default.
- W3019505787 creator A5048465819 @default.
- W3019505787 creator A5078222384 @default.
- W3019505787 creator A5079869432 @default.
- W3019505787 creator A5082289405 @default.
- W3019505787 date "2020-04-27" @default.
- W3019505787 modified "2023-10-18" @default.
- W3019505787 title "A Batch Normalized Inference Network Keeps the KL Vanishing Away" @default.
- W3019505787 cites W1522301498 @default.
- W3019505787 cites W1909320841 @default.
- W3019505787 cites W2064675550 @default.
- W3019505787 cites W2188365844 @default.
- W3019505787 cites W2250539671 @default.
- W3019505787 cites W2557579533 @default.
- W3019505787 cites W2560512785 @default.
- W3019505787 cites W2594538354 @default.
- W3019505787 cites W2605246398 @default.
- W3019505787 cites W2622563070 @default.
- W3019505787 cites W2753738274 @default.
- W3019505787 cites W2784067649 @default.
- W3019505787 cites W2784400615 @default.
- W3019505787 cites W2796335507 @default.
- W3019505787 cites W2798596203 @default.
- W3019505787 cites W2798970679 @default.
- W3019505787 cites W2888960966 @default.
- W3019505787 cites W2906670434 @default.
- W3019505787 cites W2910135751 @default.
- W3019505787 cites W2949117887 @default.
- W3019505787 cites W2950902819 @default.
- W3019505787 cites W2951575317 @default.
- W3019505787 cites W2951670304 @default.
- W3019505787 cites W2952478253 @default.
- W3019505787 cites W2962738009 @default.
- W3019505787 cites W2963018920 @default.
- W3019505787 cites W2963145887 @default.
- W3019505787 cites W2963223306 @default.
- W3019505787 cites W2963275229 @default.
- W3019505787 cites W2963366196 @default.
- W3019505787 cites W2963399222 @default.
- W3019505787 cites W2963522047 @default.
- W3019505787 cites W2963773425 @default.
- W3019505787 cites W2963987720 @default.
- W3019505787 cites W2970810127 @default.
- W3019505787 doi "https://doi.org/10.48550/arxiv.2004.12585" @default.
- W3019505787 hasPublicationYear "2020" @default.
- W3019505787 type Work @default.
- W3019505787 sameAs 3019505787 @default.
- W3019505787 citedByCount "1" @default.
- W3019505787 countsByYear W30195057872021 @default.
- W3019505787 crossrefType "posted-content" @default.
- W3019505787 hasAuthorship W3019505787A5001469325 @default.
- W3019505787 hasAuthorship W3019505787A5022526821 @default.
- W3019505787 hasAuthorship W3019505787A5040117777 @default.
- W3019505787 hasAuthorship W3019505787A5048465819 @default.
- W3019505787 hasAuthorship W3019505787A5078222384 @default.
- W3019505787 hasAuthorship W3019505787A5079869432 @default.
- W3019505787 hasAuthorship W3019505787A5082289405 @default.
- W3019505787 hasBestOaLocation W30195057871 @default.
- W3019505787 hasConcept C101738243 @default.
- W3019505787 hasConcept C107673813 @default.
- W3019505787 hasConcept C11413529 @default.
- W3019505787 hasConcept C138885662 @default.
- W3019505787 hasConcept C149782125 @default.
- W3019505787 hasConcept C153180895 @default.
- W3019505787 hasConcept C154945302 @default.
- W3019505787 hasConcept C159877910 @default.
- W3019505787 hasConcept C177264268 @default.
- W3019505787 hasConcept C177769412 @default.
- W3019505787 hasConcept C199360897 @default.
- W3019505787 hasConcept C207390915 @default.
- W3019505787 hasConcept C2776214188 @default.
- W3019505787 hasConcept C33923547 @default.
- W3019505787 hasConcept C41008148 @default.
- W3019505787 hasConcept C41895202 @default.
- W3019505787 hasConcept C50644808 @default.
- W3019505787 hasConcept C51167844 @default.
- W3019505787 hasConcept C57830394 @default.
- W3019505787 hasConceptScore W3019505787C101738243 @default.
- W3019505787 hasConceptScore W3019505787C107673813 @default.
- W3019505787 hasConceptScore W3019505787C11413529 @default.
- W3019505787 hasConceptScore W3019505787C138885662 @default.
- W3019505787 hasConceptScore W3019505787C149782125 @default.
- W3019505787 hasConceptScore W3019505787C153180895 @default.
- W3019505787 hasConceptScore W3019505787C154945302 @default.
- W3019505787 hasConceptScore W3019505787C159877910 @default.
- W3019505787 hasConceptScore W3019505787C177264268 @default.
- W3019505787 hasConceptScore W3019505787C177769412 @default.
- W3019505787 hasConceptScore W3019505787C199360897 @default.
- W3019505787 hasConceptScore W3019505787C207390915 @default.
- W3019505787 hasConceptScore W3019505787C2776214188 @default.
- W3019505787 hasConceptScore W3019505787C33923547 @default.
- W3019505787 hasConceptScore W3019505787C41008148 @default.
- W3019505787 hasConceptScore W3019505787C41895202 @default.
- W3019505787 hasConceptScore W3019505787C50644808 @default.
- W3019505787 hasConceptScore W3019505787C51167844 @default.
- W3019505787 hasConceptScore W3019505787C57830394 @default.