FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3005304850> ?p ?o ?g. }

• W3005304850 abstract "Variational Bayesian Inference is a popular methodology for approximating posterior distributions over Bayesian neural network weights. Recent work developing this class of methods has explored ever richer parameterizations of the approximate posterior in the hope of improving performance. In contrast, here we share a curious experimental finding that suggests instead restricting the variational distribution to a more compact parameterization. For a variety of deep Bayesian neural networks trained using Gaussian mean-field variational inference, we find that the posterior standard deviations consistently exhibit strong low-rank structure after convergence. This means that by decomposing these variational parameters into a low-rank factorization, we can make our variational approximation more compact without decreasing the models' performance. Furthermore, we find that such factorized parameterizations improve the signal-to-noise ratio of stochastic gradient estimates of the variational lower bound, resulting in faster convergence." @default.
• W3005304850 created "2020-02-14" @default.
• W3005304850 creator A5000291599 @default.
• W3005304850 creator A5021577029 @default.
• W3005304850 creator A5022871230 @default.
• W3005304850 creator A5025839770 @default.
• W3005304850 creator A5033851429 @default.
• W3005304850 creator A5036302820 @default.
• W3005304850 creator A5044982268 @default.
• W3005304850 creator A5065197523 @default.
• W3005304850 creator A5067993987 @default.
• W3005304850 creator A5082512329 @default.
• W3005304850 date "2020-02-07" @default.
• W3005304850 modified "2023-10-05" @default.
• W3005304850 title "The k-tied Normal Distribution: A Compact Parameterization of Gaussian Mean Field Posteriors in Bayesian Neural Networks" @default.
• W3005304850 cites W1522301498 @default.
• W3005304850 cites W1533861849 @default.
• W3005304850 cites W1534417711 @default.
• W3005304850 cites W1677182931 @default.
• W3005304850 cites W1719489212 @default.
• W3005304850 cites W1826234144 @default.
• W3005304850 cites W2047229728 @default.
• W3005304850 cites W2064675550 @default.
• W3005304850 cites W2108677974 @default.
• W3005304850 cites W2111051539 @default.
• W3005304850 cites W2112796928 @default.
• W3005304850 cites W2113459411 @default.
• W3005304850 cites W2129869373 @default.
• W3005304850 cites W2149991584 @default.
• W3005304850 cites W2150807068 @default.
• W3005304850 cites W2153455783 @default.
• W3005304850 cites W2156387975 @default.
• W3005304850 cites W2158840489 @default.
• W3005304850 cites W2194775991 @default.
• W3005304850 cites W2269892441 @default.
• W3005304850 cites W2302053044 @default.
• W3005304850 cites W2552233335 @default.
• W3005304850 cites W2557283755 @default.
• W3005304850 cites W2579661004 @default.
• W3005304850 cites W2592505114 @default.
• W3005304850 cites W2725061391 @default.
• W3005304850 cites W2731982447 @default.
• W3005304850 cites W2774412855 @default.
• W3005304850 cites W2785606994 @default.
• W3005304850 cites W2807776404 @default.
• W3005304850 cites W2891952778 @default.
• W3005304850 cites W2894356164 @default.
• W3005304850 cites W2907020378 @default.
• W3005304850 cites W2951004968 @default.
• W3005304850 cites W2951266961 @default.
• W3005304850 cites W2962851448 @default.
• W3005304850 cites W2963090522 @default.
• W3005304850 cites W2963173382 @default.
• W3005304850 cites W2963736577 @default.
• W3005304850 cites W2964052395 @default.
• W3005304850 cites W2970859221 @default.
• W3005304850 cites W2970861023 @default.
• W3005304850 cites W2970971315 @default.
• W3005304850 cites W3034669169 @default.
• W3005304850 cites W3037238545 @default.
• W3005304850 cites W3103982962 @default.
• W3005304850 cites W3104819538 @default.
• W3005304850 cites W3118608800 @default.
• W3005304850 cites W3148198191 @default.
• W3005304850 cites W759726671 @default.
• W3005304850 hasPublicationYear "2020" @default.
• W3005304850 type Work @default.
• W3005304850 sameAs 3005304850 @default.
• W3005304850 citedByCount "9" @default.
• W3005304850 countsByYear W30053048502020 @default.
• W3005304850 countsByYear W30053048502021 @default.
• W3005304850 crossrefType "posted-content" @default.
• W3005304850 hasAuthorship W3005304850A5000291599 @default.
• W3005304850 hasAuthorship W3005304850A5021577029 @default.
• W3005304850 hasAuthorship W3005304850A5022871230 @default.
• W3005304850 hasAuthorship W3005304850A5025839770 @default.
• W3005304850 hasAuthorship W3005304850A5033851429 @default.
• W3005304850 hasAuthorship W3005304850A5036302820 @default.
• W3005304850 hasAuthorship W3005304850A5044982268 @default.
• W3005304850 hasAuthorship W3005304850A5065197523 @default.
• W3005304850 hasAuthorship W3005304850A5067993987 @default.
• W3005304850 hasAuthorship W3005304850A5082512329 @default.
• W3005304850 hasConcept C105795698 @default.
• W3005304850 hasConcept C107673813 @default.
• W3005304850 hasConcept C110121322 @default.
• W3005304850 hasConcept C11413529 @default.
• W3005304850 hasConcept C114614502 @default.
• W3005304850 hasConcept C121332964 @default.
• W3005304850 hasConcept C126255220 @default.
• W3005304850 hasConcept C134306372 @default.
• W3005304850 hasConcept C154945302 @default.
• W3005304850 hasConcept C160234255 @default.
• W3005304850 hasConcept C162324750 @default.
• W3005304850 hasConcept C163716315 @default.
• W3005304850 hasConcept C164226766 @default.
• W3005304850 hasConcept C177769412 @default.
• W3005304850 hasConcept C2777303404 @default.
• W3005304850 hasConcept C28826006 @default.
• W3005304850 hasConcept C33923547 @default.
• W3005304850 hasConcept C41008148 @default.

• next