Matches in SemOpenAlex for { <https://semopenalex.org/work/W3114371569> ?p ?o ?g. }
Showing items 1 to 93 of
93
with 100 items per page.
- W3114371569 endingPage "7640" @default.
- W3114371569 startingPage "7632" @default.
- W3114371569 abstract "Stochastic neural networks with discrete random variables are an important class of models for their expressiveness and interpretability. Since direct differentiation and backpropagation is not possible, Monte Carlo gradient estimation techniques are a popular alternative. Efficient stochastic gradient estimators, such Straight-Through and Gumbel-Softmax, work well for shallow stochastic models. Their performance, however, suffers with hierarchical, more complex models. We focus on stochastic networks with Boolean latent variables. To analyze such networks, we introduce the framework of harmonic analysis for Boolean functions to derive an analytic formulation for the bias and variance in the Straight-Through estimator. Exploiting these formulations, we propose emph{FouST}, a low-bias and low-variance gradient estimation algorithm that is just as efficient. Extensive experiments show that FouST performs favorably compared to state-of-the-art biased estimators and is much faster than unbiased ones." @default.
- W3114371569 created "2021-01-05" @default.
- W3114371569 creator A5002625178 @default.
- W3114371569 creator A5079743569 @default.
- W3114371569 creator A5083922034 @default.
- W3114371569 date "2020-11-21" @default.
- W3114371569 modified "2023-09-23" @default.
- W3114371569 title "Low Bias Low Variance Gradient Estimates for Boolean Stochastic Networks" @default.
- W3114371569 cites W1891181203 @default.
- W3114371569 cites W1899006432 @default.
- W3114371569 cites W1909320841 @default.
- W3114371569 cites W1959608418 @default.
- W3114371569 cites W2026799324 @default.
- W3114371569 cites W2042194938 @default.
- W3114371569 cites W2095546965 @default.
- W3114371569 cites W2097268041 @default.
- W3114371569 cites W2119717200 @default.
- W3114371569 cites W2295021037 @default.
- W3114371569 cites W2547875792 @default.
- W3114371569 cites W2602076750 @default.
- W3114371569 cites W2753738274 @default.
- W3114371569 cites W2767002724 @default.
- W3114371569 cites W2952165242 @default.
- W3114371569 cites W2952264928 @default.
- W3114371569 cites W2963341924 @default.
- W3114371569 cites W2963408680 @default.
- W3114371569 cites W2963799213 @default.
- W3114371569 cites W2964122153 @default.
- W3114371569 hasPublicationYear "2020" @default.
- W3114371569 type Work @default.
- W3114371569 sameAs 3114371569 @default.
- W3114371569 citedByCount "3" @default.
- W3114371569 countsByYear W31143715692021 @default.
- W3114371569 crossrefType "proceedings-article" @default.
- W3114371569 hasAuthorship W3114371569A5002625178 @default.
- W3114371569 hasAuthorship W3114371569A5079743569 @default.
- W3114371569 hasAuthorship W3114371569A5083922034 @default.
- W3114371569 hasConcept C105795698 @default.
- W3114371569 hasConcept C11413529 @default.
- W3114371569 hasConcept C121955636 @default.
- W3114371569 hasConcept C126255220 @default.
- W3114371569 hasConcept C134444547 @default.
- W3114371569 hasConcept C144133560 @default.
- W3114371569 hasConcept C154945302 @default.
- W3114371569 hasConcept C185429906 @default.
- W3114371569 hasConcept C187455244 @default.
- W3114371569 hasConcept C196083921 @default.
- W3114371569 hasConcept C2781067378 @default.
- W3114371569 hasConcept C33923547 @default.
- W3114371569 hasConcept C41008148 @default.
- W3114371569 hasConceptScore W3114371569C105795698 @default.
- W3114371569 hasConceptScore W3114371569C11413529 @default.
- W3114371569 hasConceptScore W3114371569C121955636 @default.
- W3114371569 hasConceptScore W3114371569C126255220 @default.
- W3114371569 hasConceptScore W3114371569C134444547 @default.
- W3114371569 hasConceptScore W3114371569C144133560 @default.
- W3114371569 hasConceptScore W3114371569C154945302 @default.
- W3114371569 hasConceptScore W3114371569C185429906 @default.
- W3114371569 hasConceptScore W3114371569C187455244 @default.
- W3114371569 hasConceptScore W3114371569C196083921 @default.
- W3114371569 hasConceptScore W3114371569C2781067378 @default.
- W3114371569 hasConceptScore W3114371569C33923547 @default.
- W3114371569 hasConceptScore W3114371569C41008148 @default.
- W3114371569 hasLocation W31143715691 @default.
- W3114371569 hasOpenAccess W3114371569 @default.
- W3114371569 hasPrimaryLocation W31143715691 @default.
- W3114371569 hasRelatedWork W174438472 @default.
- W3114371569 hasRelatedWork W2035149996 @default.
- W3114371569 hasRelatedWork W2082321879 @default.
- W3114371569 hasRelatedWork W2092789011 @default.
- W3114371569 hasRelatedWork W2101092845 @default.
- W3114371569 hasRelatedWork W2135817792 @default.
- W3114371569 hasRelatedWork W2165609874 @default.
- W3114371569 hasRelatedWork W2219117298 @default.
- W3114371569 hasRelatedWork W2292875544 @default.
- W3114371569 hasRelatedWork W2415850724 @default.
- W3114371569 hasRelatedWork W2883664406 @default.
- W3114371569 hasRelatedWork W3002223934 @default.
- W3114371569 hasRelatedWork W3004162305 @default.
- W3114371569 hasRelatedWork W3022485013 @default.
- W3114371569 hasRelatedWork W3040431967 @default.
- W3114371569 hasRelatedWork W3041640539 @default.
- W3114371569 hasRelatedWork W3097797625 @default.
- W3114371569 hasRelatedWork W3126797316 @default.
- W3114371569 hasRelatedWork W3165777227 @default.
- W3114371569 hasRelatedWork W778777784 @default.
- W3114371569 hasVolume "119" @default.
- W3114371569 isParatext "false" @default.
- W3114371569 isRetracted "false" @default.
- W3114371569 magId "3114371569" @default.
- W3114371569 workType "article" @default.