Matches in SemOpenAlex for { <https://semopenalex.org/work/W2912205739> ?p ?o ?g. }
- W2912205739 abstract "We propose a general yet simple theorem describing the convergence of SGD under the arbitrary sampling paradigm. Our theorem describes the convergence of an infinite array of variants of SGD, each of which is associated with a specific probability law governing the data selection rule used to form mini-batches. This is the first time such an analysis is performed, and most of our variants of SGD were never explicitly considered in the literature before. Our analysis relies on the recently introduced notion of expected smoothness and does not rely on a uniform bound on the variance of the stochastic gradients. By specializing our theorem to different mini-batching strategies, such as sampling with replacement and independent sampling, we derive exact expressions for the stepsize as a function of the mini-batch size. With this we can also determine the mini-batch size that optimizes the total complexity, and show explicitly that as the variance of the stochastic gradient evaluated at the minimum grows, so does the optimal mini-batch size. For zero variance, the optimal mini-batch size is one. Moreover, we prove insightful stepsize-switching rules which describe when one should switch from a constant to a decreasing stepsize regime." @default.
- W2912205739 created "2019-02-21" @default.
- W2912205739 creator A5008334471 @default.
- W2912205739 creator A5014152561 @default.
- W2912205739 creator A5025477585 @default.
- W2912205739 creator A5036598221 @default.
- W2912205739 creator A5040348769 @default.
- W2912205739 creator A5075390126 @default.
- W2912205739 date "2019-06-09" @default.
- W2912205739 modified "2023-10-02" @default.
- W2912205739 title "SGD: General Analysis and Improved Rates" @default.
- W2912205739 cites W1505731132 @default.
- W2912205739 cites W1516903196 @default.
- W2912205739 cites W1916951228 @default.
- W2912205739 cites W1992208280 @default.
- W2912205739 cites W1994616650 @default.
- W2912205739 cites W2032395696 @default.
- W2912205739 cites W2112269233 @default.
- W2912205739 cites W2153635508 @default.
- W2912205739 cites W2154682027 @default.
- W2912205739 cites W2156779765 @default.
- W2912205739 cites W2162287622 @default.
- W2912205739 cites W2164075197 @default.
- W2912205739 cites W2195435876 @default.
- W2912205739 cites W2205007824 @default.
- W2912205739 cites W2293647857 @default.
- W2912205739 cites W2622263826 @default.
- W2912205739 cites W2624091502 @default.
- W2912205739 cites W2625685405 @default.
- W2912205739 cites W2774062531 @default.
- W2912205739 cites W2780752111 @default.
- W2912205739 cites W2799870997 @default.
- W2912205739 cites W2889801698 @default.
- W2912205739 cites W2896222303 @default.
- W2912205739 cites W2912598912 @default.
- W2912205739 cites W2922223813 @default.
- W2912205739 cites W2949699060 @default.
- W2912205739 cites W2951781666 @default.
- W2912205739 cites W2953000681 @default.
- W2912205739 cites W2962712496 @default.
- W2912205739 cites W2962990180 @default.
- W2912205739 cites W2963228337 @default.
- W2912205739 cites W2963244042 @default.
- W2912205739 cites W2963248893 @default.
- W2912205739 cites W2963433607 @default.
- W2912205739 cites W2963434703 @default.
- W2912205739 cites W2963589953 @default.
- W2912205739 cites W2963794891 @default.
- W2912205739 cites W2963830980 @default.
- W2912205739 cites W2963948233 @default.
- W2912205739 cites W3141595720 @default.
- W2912205739 cites W604534808 @default.
- W2912205739 hasPublicationYear "2019" @default.
- W2912205739 type Work @default.
- W2912205739 sameAs 2912205739 @default.
- W2912205739 citedByCount "27" @default.
- W2912205739 countsByYear W29122057392018 @default.
- W2912205739 countsByYear W29122057392019 @default.
- W2912205739 countsByYear W29122057392020 @default.
- W2912205739 countsByYear W29122057392021 @default.
- W2912205739 crossrefType "proceedings-article" @default.
- W2912205739 hasAuthorship W2912205739A5008334471 @default.
- W2912205739 hasAuthorship W2912205739A5014152561 @default.
- W2912205739 hasAuthorship W2912205739A5025477585 @default.
- W2912205739 hasAuthorship W2912205739A5036598221 @default.
- W2912205739 hasAuthorship W2912205739A5040348769 @default.
- W2912205739 hasAuthorship W2912205739A5075390126 @default.
- W2912205739 hasBestOaLocation W29122057391 @default.
- W2912205739 hasConcept C102634674 @default.
- W2912205739 hasConcept C105795698 @default.
- W2912205739 hasConcept C106131492 @default.
- W2912205739 hasConcept C111472728 @default.
- W2912205739 hasConcept C11413529 @default.
- W2912205739 hasConcept C121955636 @default.
- W2912205739 hasConcept C126255220 @default.
- W2912205739 hasConcept C127162648 @default.
- W2912205739 hasConcept C129848803 @default.
- W2912205739 hasConcept C134306372 @default.
- W2912205739 hasConcept C138885662 @default.
- W2912205739 hasConcept C14036430 @default.
- W2912205739 hasConcept C140779682 @default.
- W2912205739 hasConcept C144133560 @default.
- W2912205739 hasConcept C154945302 @default.
- W2912205739 hasConcept C162324750 @default.
- W2912205739 hasConcept C196083921 @default.
- W2912205739 hasConcept C199360897 @default.
- W2912205739 hasConcept C2777027219 @default.
- W2912205739 hasConcept C2777303404 @default.
- W2912205739 hasConcept C2780586882 @default.
- W2912205739 hasConcept C28826006 @default.
- W2912205739 hasConcept C31258907 @default.
- W2912205739 hasConcept C31972630 @default.
- W2912205739 hasConcept C33923547 @default.
- W2912205739 hasConcept C41008148 @default.
- W2912205739 hasConcept C50522688 @default.
- W2912205739 hasConcept C57869625 @default.
- W2912205739 hasConcept C77553402 @default.
- W2912205739 hasConcept C78458016 @default.
- W2912205739 hasConcept C81917197 @default.
- W2912205739 hasConcept C86803240 @default.