Matches in SemOpenAlex for { <https://semopenalex.org/work/W2963456262> ?p ?o ?g. }
- W2963456262 endingPage "684" @default.
- W2963456262 startingPage "675" @default.
- W2963456262 abstract "Mini-batch stochastic gradient descent and variants thereof have become standard for large-scale empirical risk minimization like the training of neural networks. These methods are usually used with a constant batch size chosen by simple empirical inspection. The batch size significantly influences the behavior of the stochastic optimization algorithm, though, since it determines the variance of the gradient estimates. This variance also changes over the optimization process; when using a constant batch size, stability and convergence is thus often enforced by means of a (manually tuned) decreasing learning rate schedule. We propose a practical method for dynamic batch size adaptation. It estimates the variance of the stochastic gradients and adapts the batch size to decrease the variance proportionally to the value of the objective function, removing the need for the aforementioned learning rate decrease. In contrast to recent related work, our algorithm couples the batch size to the learning rate, directly reflecting the known relationship between the two. On popular image classification benchmarks, our batch size adaptation yields faster optimization convergence, while simultaneously simplifying learning rate tuning. A TensorFlow implementation is available." @default.
- W2963456262 created "2019-07-30" @default.
- W2963456262 creator A5022872779 @default.
- W2963456262 creator A5047338615 @default.
- W2963456262 creator A5070528616 @default.
- W2963456262 date "2017-12-01" @default.
- W2963456262 modified "2023-09-23" @default.
- W2963456262 title "Coupling Adaptive Batch Sizes with Learning Rates" @default.
- W2963456262 cites W1842094663 @default.
- W2963456262 cites W1994616650 @default.
- W2963456262 cites W2061570747 @default.
- W2963456262 cites W2107438106 @default.
- W2963456262 cites W2112796928 @default.
- W2963456262 cites W2120420045 @default.
- W2963456262 cites W2132968152 @default.
- W2963456262 cites W2146502635 @default.
- W2963456262 cites W2162287622 @default.
- W2963456262 cites W2335728318 @default.
- W2963456262 cites W2622379787 @default.
- W2963456262 cites W2950946134 @default.
- W2963456262 cites W2964121744 @default.
- W2963456262 cites W3118608800 @default.
- W2963456262 cites W3140968660 @default.
- W2963456262 hasPublicationYear "2017" @default.
- W2963456262 type Work @default.
- W2963456262 sameAs 2963456262 @default.
- W2963456262 citedByCount "31" @default.
- W2963456262 countsByYear W29634562622017 @default.
- W2963456262 countsByYear W29634562622018 @default.
- W2963456262 countsByYear W29634562622019 @default.
- W2963456262 countsByYear W29634562622020 @default.
- W2963456262 countsByYear W29634562622021 @default.
- W2963456262 crossrefType "proceedings-article" @default.
- W2963456262 hasAuthorship W2963456262A5022872779 @default.
- W2963456262 hasAuthorship W2963456262A5047338615 @default.
- W2963456262 hasAuthorship W2963456262A5070528616 @default.
- W2963456262 hasConcept C111919701 @default.
- W2963456262 hasConcept C112972136 @default.
- W2963456262 hasConcept C11413529 @default.
- W2963456262 hasConcept C119857082 @default.
- W2963456262 hasConcept C121955636 @default.
- W2963456262 hasConcept C126255220 @default.
- W2963456262 hasConcept C144133560 @default.
- W2963456262 hasConcept C147764199 @default.
- W2963456262 hasConcept C153258448 @default.
- W2963456262 hasConcept C154945302 @default.
- W2963456262 hasConcept C162324750 @default.
- W2963456262 hasConcept C172658912 @default.
- W2963456262 hasConcept C194387892 @default.
- W2963456262 hasConcept C196083921 @default.
- W2963456262 hasConcept C199360897 @default.
- W2963456262 hasConcept C206688291 @default.
- W2963456262 hasConcept C26517878 @default.
- W2963456262 hasConcept C2776502983 @default.
- W2963456262 hasConcept C2777027219 @default.
- W2963456262 hasConcept C2777303404 @default.
- W2963456262 hasConcept C33923547 @default.
- W2963456262 hasConcept C38652104 @default.
- W2963456262 hasConcept C41008148 @default.
- W2963456262 hasConcept C50522688 @default.
- W2963456262 hasConcept C50644808 @default.
- W2963456262 hasConcept C57869625 @default.
- W2963456262 hasConcept C68387754 @default.
- W2963456262 hasConceptScore W2963456262C111919701 @default.
- W2963456262 hasConceptScore W2963456262C112972136 @default.
- W2963456262 hasConceptScore W2963456262C11413529 @default.
- W2963456262 hasConceptScore W2963456262C119857082 @default.
- W2963456262 hasConceptScore W2963456262C121955636 @default.
- W2963456262 hasConceptScore W2963456262C126255220 @default.
- W2963456262 hasConceptScore W2963456262C144133560 @default.
- W2963456262 hasConceptScore W2963456262C147764199 @default.
- W2963456262 hasConceptScore W2963456262C153258448 @default.
- W2963456262 hasConceptScore W2963456262C154945302 @default.
- W2963456262 hasConceptScore W2963456262C162324750 @default.
- W2963456262 hasConceptScore W2963456262C172658912 @default.
- W2963456262 hasConceptScore W2963456262C194387892 @default.
- W2963456262 hasConceptScore W2963456262C196083921 @default.
- W2963456262 hasConceptScore W2963456262C199360897 @default.
- W2963456262 hasConceptScore W2963456262C206688291 @default.
- W2963456262 hasConceptScore W2963456262C26517878 @default.
- W2963456262 hasConceptScore W2963456262C2776502983 @default.
- W2963456262 hasConceptScore W2963456262C2777027219 @default.
- W2963456262 hasConceptScore W2963456262C2777303404 @default.
- W2963456262 hasConceptScore W2963456262C33923547 @default.
- W2963456262 hasConceptScore W2963456262C38652104 @default.
- W2963456262 hasConceptScore W2963456262C41008148 @default.
- W2963456262 hasConceptScore W2963456262C50522688 @default.
- W2963456262 hasConceptScore W2963456262C50644808 @default.
- W2963456262 hasConceptScore W2963456262C57869625 @default.
- W2963456262 hasConceptScore W2963456262C68387754 @default.
- W2963456262 hasOpenAccess W2963456262 @default.
- W2963456262 hasRelatedWork W1994616650 @default.
- W2963456262 hasRelatedWork W2061570747 @default.
- W2963456262 hasRelatedWork W2083842231 @default.
- W2963456262 hasRelatedWork W2108598243 @default.
- W2963456262 hasRelatedWork W2146502635 @default.
- W2963456262 hasRelatedWork W2194775991 @default.
- W2963456262 hasRelatedWork W230490465 @default.