Matches in SemOpenAlex for { <https://semopenalex.org/work/W3206219838> ?p ?o ?g. }
- W3206219838 abstract "In distributed learning, local SGD (also known as federated averaging) and its simple baseline minibatch SGD are widely studied optimization methods. Most existing analyses of these methods assume independent and unbiased gradient estimates obtained via with-replacement sampling. In contrast, we study shuffling-based variants: minibatch and local Random Reshuffling, which draw stochastic gradients without replacement and are thus closer to practice. For smooth functions satisfying the Polyak-{L}ojasiewicz condition, we obtain convergence bounds (in the large epoch regime) which show that these shuffling-based variants converge faster than their with-replacement counterparts. Moreover, we prove matching lower bounds showing that our convergence analysis is tight. Finally, we propose an algorithmic modification called synchronized shuffling that leads to convergence rates faster than our lower bounds in near-homogeneous settings." @default.
- W3206219838 created "2021-10-25" @default.
- W3206219838 creator A5045562332 @default.
- W3206219838 creator A5046093647 @default.
- W3206219838 creator A5058767558 @default.
- W3206219838 date "2021-10-19" @default.
- W3206219838 modified "2023-09-25" @default.
- W3206219838 title "Minibatch vs Local SGD with Shuffling: Tight Convergence Bounds and Beyond" @default.
- W3206219838 cites W1971459440 @default.
- W3206219838 cites W2022829242 @default.
- W3206219838 cites W203462006 @default.
- W3206219838 cites W2112269233 @default.
- W3206219838 cites W2405543129 @default.
- W3206219838 cites W2409483509 @default.
- W3206219838 cites W2535838896 @default.
- W3206219838 cites W2541884796 @default.
- W3206219838 cites W2900182564 @default.
- W3206219838 cites W2918941755 @default.
- W3206219838 cites W2962970252 @default.
- W3206219838 cites W2963179579 @default.
- W3206219838 cites W2963933775 @default.
- W3206219838 cites W2970020383 @default.
- W3206219838 cites W2982654255 @default.
- W3206219838 cites W2989763744 @default.
- W3206219838 cites W2995653155 @default.
- W3206219838 cites W3007986655 @default.
- W3206219838 cites W3008438679 @default.
- W3206219838 cites W3034013096 @default.
- W3206219838 cites W3034666238 @default.
- W3206219838 cites W3034697819 @default.
- W3206219838 cites W3034741284 @default.
- W3206219838 cites W3035507578 @default.
- W3206219838 cites W3037674069 @default.
- W3206219838 cites W3038022836 @default.
- W3206219838 cites W3041733661 @default.
- W3206219838 cites W3046380121 @default.
- W3206219838 cites W3098677710 @default.
- W3206219838 cites W3101271792 @default.
- W3206219838 cites W3102089081 @default.
- W3206219838 cites W3106152900 @default.
- W3206219838 cites W3111681398 @default.
- W3206219838 cites W3126295195 @default.
- W3206219838 cites W3132116480 @default.
- W3206219838 cites W3132614086 @default.
- W3206219838 cites W3153737196 @default.
- W3206219838 cites W3170464493 @default.
- W3206219838 cites W3171966211 @default.
- W3206219838 doi "https://doi.org/10.48550/arxiv.2110.10342" @default.
- W3206219838 hasPublicationYear "2021" @default.
- W3206219838 type Work @default.
- W3206219838 sameAs 3206219838 @default.
- W3206219838 citedByCount "0" @default.
- W3206219838 crossrefType "posted-content" @default.
- W3206219838 hasAuthorship W3206219838A5045562332 @default.
- W3206219838 hasAuthorship W3206219838A5046093647 @default.
- W3206219838 hasAuthorship W3206219838A5058767558 @default.
- W3206219838 hasBestOaLocation W32062198381 @default.
- W3206219838 hasConcept C105795698 @default.
- W3206219838 hasConcept C11413529 @default.
- W3206219838 hasConcept C114614502 @default.
- W3206219838 hasConcept C126255220 @default.
- W3206219838 hasConcept C162324750 @default.
- W3206219838 hasConcept C165064840 @default.
- W3206219838 hasConcept C167927819 @default.
- W3206219838 hasConcept C26517878 @default.
- W3206219838 hasConcept C2777303404 @default.
- W3206219838 hasConcept C33923547 @default.
- W3206219838 hasConcept C38652104 @default.
- W3206219838 hasConcept C41008148 @default.
- W3206219838 hasConcept C50522688 @default.
- W3206219838 hasConcept C57869625 @default.
- W3206219838 hasConcept C66882249 @default.
- W3206219838 hasConceptScore W3206219838C105795698 @default.
- W3206219838 hasConceptScore W3206219838C11413529 @default.
- W3206219838 hasConceptScore W3206219838C114614502 @default.
- W3206219838 hasConceptScore W3206219838C126255220 @default.
- W3206219838 hasConceptScore W3206219838C162324750 @default.
- W3206219838 hasConceptScore W3206219838C165064840 @default.
- W3206219838 hasConceptScore W3206219838C167927819 @default.
- W3206219838 hasConceptScore W3206219838C26517878 @default.
- W3206219838 hasConceptScore W3206219838C2777303404 @default.
- W3206219838 hasConceptScore W3206219838C33923547 @default.
- W3206219838 hasConceptScore W3206219838C38652104 @default.
- W3206219838 hasConceptScore W3206219838C41008148 @default.
- W3206219838 hasConceptScore W3206219838C50522688 @default.
- W3206219838 hasConceptScore W3206219838C57869625 @default.
- W3206219838 hasConceptScore W3206219838C66882249 @default.
- W3206219838 hasLocation W32062198381 @default.
- W3206219838 hasOpenAccess W3206219838 @default.
- W3206219838 hasPrimaryLocation W32062198381 @default.
- W3206219838 hasRelatedWork W1586585988 @default.
- W3206219838 hasRelatedWork W1970358763 @default.
- W3206219838 hasRelatedWork W2083219623 @default.
- W3206219838 hasRelatedWork W2128702080 @default.
- W3206219838 hasRelatedWork W2382095754 @default.
- W3206219838 hasRelatedWork W2389459180 @default.
- W3206219838 hasRelatedWork W2540751784 @default.
- W3206219838 hasRelatedWork W2908009812 @default.
- W3206219838 hasRelatedWork W2952048857 @default.
- W3206219838 hasRelatedWork W4295922964 @default.