Matches in SemOpenAlex for { <https://semopenalex.org/work/W3168771896> ?p ?o ?g. }
Showing items 1 to 89 of
89
with 100 items per page.
- W3168771896 abstract "Training models with discrete latent variables is challenging due to the high variance of unbiased gradient estimators. While low-variance reparameterization gradients of a continuous relaxation can provide an effective solution, a continuous relaxation is not always available or tractable. Dong et al. (2020) and Yin et al. (2020) introduced a performant estimator that does not rely on continuous relaxations; however, it is limited to binary random variables. We introduce a novel derivation of their estimator based on importance sampling and statistical couplings, which we extend to the categorical setting. Motivated by the construction of a stick-breaking coupling, we introduce gradient estimators based on reparameterizing categorical variables as sequences of binary variables and Rao-Blackwellization. In systematic experiments, we show that our proposed categorical gradient estimators provide state-of-the-art performance, whereas even with additional Rao-Blackwellization, previous estimators (Yin et al., 2019) underperform a simpler REINFORCE with a leave-one-out-baseline estimator (Kool et al., 2019)." @default.
- W3168771896 created "2021-06-22" @default.
- W3168771896 creator A5048032272 @default.
- W3168771896 creator A5052175108 @default.
- W3168771896 creator A5079370148 @default.
- W3168771896 date "2021-06-15" @default.
- W3168771896 modified "2023-09-27" @default.
- W3168771896 title "Coupled Gradient Estimators for Discrete Latent Variables" @default.
- W3168771896 cites W1516111018 @default.
- W3168771896 cites W1602773783 @default.
- W3168771896 cites W1921523184 @default.
- W3168771896 cites W1959608418 @default.
- W3168771896 cites W1979432659 @default.
- W3168771896 cites W2046765929 @default.
- W3168771896 cites W2119717200 @default.
- W3168771896 cites W2123094878 @default.
- W3168771896 cites W2187922941 @default.
- W3168771896 cites W2194321275 @default.
- W3168771896 cites W2291809032 @default.
- W3168771896 cites W2547875792 @default.
- W3168771896 cites W2548228487 @default.
- W3168771896 cites W2602076750 @default.
- W3168771896 cites W2750384547 @default.
- W3168771896 cites W2909637611 @default.
- W3168771896 cites W2943262294 @default.
- W3168771896 cites W2962897886 @default.
- W3168771896 cites W2963851840 @default.
- W3168771896 cites W2966628687 @default.
- W3168771896 cites W3036990844 @default.
- W3168771896 cites W3133702157 @default.
- W3168771896 cites W3156891177 @default.
- W3168771896 hasPublicationYear "2021" @default.
- W3168771896 type Work @default.
- W3168771896 sameAs 3168771896 @default.
- W3168771896 citedByCount "2" @default.
- W3168771896 countsByYear W31687718962021 @default.
- W3168771896 crossrefType "posted-content" @default.
- W3168771896 hasAuthorship W3168771896A5048032272 @default.
- W3168771896 hasAuthorship W3168771896A5052175108 @default.
- W3168771896 hasAuthorship W3168771896A5079370148 @default.
- W3168771896 hasConcept C105795698 @default.
- W3168771896 hasConcept C121955636 @default.
- W3168771896 hasConcept C144133560 @default.
- W3168771896 hasConcept C185429906 @default.
- W3168771896 hasConcept C196083921 @default.
- W3168771896 hasConcept C28826006 @default.
- W3168771896 hasConcept C33923547 @default.
- W3168771896 hasConcept C48372109 @default.
- W3168771896 hasConcept C51167844 @default.
- W3168771896 hasConcept C5274069 @default.
- W3168771896 hasConcept C94375191 @default.
- W3168771896 hasConceptScore W3168771896C105795698 @default.
- W3168771896 hasConceptScore W3168771896C121955636 @default.
- W3168771896 hasConceptScore W3168771896C144133560 @default.
- W3168771896 hasConceptScore W3168771896C185429906 @default.
- W3168771896 hasConceptScore W3168771896C196083921 @default.
- W3168771896 hasConceptScore W3168771896C28826006 @default.
- W3168771896 hasConceptScore W3168771896C33923547 @default.
- W3168771896 hasConceptScore W3168771896C48372109 @default.
- W3168771896 hasConceptScore W3168771896C51167844 @default.
- W3168771896 hasConceptScore W3168771896C5274069 @default.
- W3168771896 hasConceptScore W3168771896C94375191 @default.
- W3168771896 hasLocation W31687718961 @default.
- W3168771896 hasOpenAccess W3168771896 @default.
- W3168771896 hasPrimaryLocation W31687718961 @default.
- W3168771896 hasRelatedWork W1482063688 @default.
- W3168771896 hasRelatedWork W2229796326 @default.
- W3168771896 hasRelatedWork W2248939373 @default.
- W3168771896 hasRelatedWork W2608754646 @default.
- W3168771896 hasRelatedWork W2775219695 @default.
- W3168771896 hasRelatedWork W2799256851 @default.
- W3168771896 hasRelatedWork W2890397555 @default.
- W3168771896 hasRelatedWork W2900693706 @default.
- W3168771896 hasRelatedWork W2952549473 @default.
- W3168771896 hasRelatedWork W2959652366 @default.
- W3168771896 hasRelatedWork W2962887762 @default.
- W3168771896 hasRelatedWork W3026313505 @default.
- W3168771896 hasRelatedWork W3122364754 @default.
- W3168771896 hasRelatedWork W3129542174 @default.
- W3168771896 hasRelatedWork W3141101630 @default.
- W3168771896 hasRelatedWork W3148747106 @default.
- W3168771896 hasRelatedWork W3151203257 @default.
- W3168771896 hasRelatedWork W3183513855 @default.
- W3168771896 hasRelatedWork W3212081099 @default.
- W3168771896 hasRelatedWork W1651121560 @default.
- W3168771896 isParatext "false" @default.
- W3168771896 isRetracted "false" @default.
- W3168771896 magId "3168771896" @default.
- W3168771896 workType "article" @default.