Matches in SemOpenAlex for { <https://semopenalex.org/work/W2994779554> ?p ?o ?g. }
- W2994779554 abstract "Mini-batch stochastic gradient methods (SGD) are state of the art for distributed training of deep neural networks. Drastic increases in the mini-batch sizes have lead to key efficiency and scalability gains in recent years. However, progress faces a major roadblock, as models trained with large batches often do not generalize well, i.e. they do not show good accuracy on new data. As a remedy, we propose a emph{post-local} SGD and show that it significantly improves the generalization performance compared to large-batch training on standard benchmarks while enjoying the same efficiency (time-to-accuracy) and scalability. We further provide an extensive study of the communication efficiency vs. performance trade-offs associated with a host of emph{local SGD} variants." @default.
- W2994779554 created "2019-12-26" @default.
- W2994779554 creator A5045519627 @default.
- W2994779554 creator A5050008056 @default.
- W2994779554 creator A5056989506 @default.
- W2994779554 creator A5073756389 @default.
- W2994779554 date "2020-04-30" @default.
- W2994779554 modified "2023-10-04" @default.
- W2994779554 title "Don't Use Large Mini-batches, Use Local SGD" @default.
- W2994779554 cites W114517082 @default.
- W2994779554 cites W1510543252 @default.
- W2994779554 cites W1572016165 @default.
- W2994779554 cites W1677182931 @default.
- W2994779554 cites W1686810756 @default.
- W2994779554 cites W1994616650 @default.
- W2994779554 cites W2026369565 @default.
- W2994779554 cites W2072566913 @default.
- W2994779554 cites W2108598243 @default.
- W2994779554 cites W2117539524 @default.
- W2994779554 cites W2130062883 @default.
- W2994779554 cites W2131613942 @default.
- W2994779554 cites W2138302120 @default.
- W2994779554 cites W2144839971 @default.
- W2994779554 cites W2166706236 @default.
- W2994779554 cites W2194775991 @default.
- W2994779554 cites W2263490141 @default.
- W2994779554 cites W2271840356 @default.
- W2994779554 cites W2301987905 @default.
- W2994779554 cites W2302255633 @default.
- W2994779554 cites W2331143823 @default.
- W2994779554 cites W2401231614 @default.
- W2994779554 cites W2405883473 @default.
- W2994779554 cites W2460815216 @default.
- W2994779554 cites W2513383847 @default.
- W2994779554 cites W2518108298 @default.
- W2994779554 cites W2525332836 @default.
- W2994779554 cites W2617242334 @default.
- W2994779554 cites W2622263826 @default.
- W2994779554 cites W2743945814 @default.
- W2994779554 cites W2749988060 @default.
- W2994779554 cites W2768267830 @default.
- W2994779554 cites W2788463493 @default.
- W2994779554 cites W2789115819 @default.
- W2994779554 cites W2796146910 @default.
- W2994779554 cites W2796196168 @default.
- W2994779554 cites W2799042347 @default.
- W2994779554 cites W2803867449 @default.
- W2994779554 cites W2886404071 @default.
- W2994779554 cites W2899771611 @default.
- W2994779554 cites W2900167092 @default.
- W2994779554 cites W2911863041 @default.
- W2994779554 cites W2921013684 @default.
- W2994779554 cites W2940453604 @default.
- W2994779554 cites W2950928354 @default.
- W2994779554 cites W2951696358 @default.
- W2994779554 cites W2951841718 @default.
- W2994779554 cites W2962747323 @default.
- W2994779554 cites W2962915600 @default.
- W2994779554 cites W2962933129 @default.
- W2994779554 cites W2963069632 @default.
- W2994779554 cites W2963173418 @default.
- W2994779554 cites W2963177640 @default.
- W2994779554 cites W2963408258 @default.
- W2994779554 cites W2963446712 @default.
- W2994779554 cites W2963632834 @default.
- W2994779554 cites W2963664311 @default.
- W2994779554 cites W2963702144 @default.
- W2994779554 cites W2963959597 @default.
- W2994779554 cites W2964072432 @default.
- W2994779554 cites W2964160102 @default.
- W2994779554 cites W2970224333 @default.
- W2994779554 cites W3037047862 @default.
- W2994779554 cites W3102816259 @default.
- W2994779554 cites W3118608800 @default.
- W2994779554 cites W71081281 @default.
- W2994779554 hasPublicationYear "2020" @default.
- W2994779554 type Work @default.
- W2994779554 sameAs 2994779554 @default.
- W2994779554 citedByCount "50" @default.
- W2994779554 countsByYear W29947795542018 @default.
- W2994779554 countsByYear W29947795542019 @default.
- W2994779554 countsByYear W29947795542020 @default.
- W2994779554 countsByYear W29947795542021 @default.
- W2994779554 countsByYear W29947795542022 @default.
- W2994779554 crossrefType "proceedings-article" @default.
- W2994779554 hasAuthorship W2994779554A5045519627 @default.
- W2994779554 hasAuthorship W2994779554A5050008056 @default.
- W2994779554 hasAuthorship W2994779554A5056989506 @default.
- W2994779554 hasAuthorship W2994779554A5073756389 @default.
- W2994779554 hasConcept C111919701 @default.
- W2994779554 hasConcept C119857082 @default.
- W2994779554 hasConcept C120314980 @default.
- W2994779554 hasConcept C134306372 @default.
- W2994779554 hasConcept C154945302 @default.
- W2994779554 hasConcept C177148314 @default.
- W2994779554 hasConcept C206688291 @default.
- W2994779554 hasConcept C26517878 @default.
- W2994779554 hasConcept C2984842247 @default.
- W2994779554 hasConcept C33923547 @default.
- W2994779554 hasConcept C41008148 @default.