Matches in SemOpenAlex for { <https://semopenalex.org/work/W2785456003> ?p ?o ?g. }
- W2785456003 abstract "Due to the substantial computational cost, training state-of-the-art deep neural networks for large-scale datasets often requires distributed training using multiple computation workers. However, by nature, workers need to frequently communicate gradients, causing severe bottlenecks, especially on lower bandwidth connections. A few methods have been proposed to compress gradient for efficient communication, but they either suffer a low compression ratio or significantly harm the resulting model accuracy, particularly when applied to convolutional neural networks. To address these issues, we propose a method to reduce the communication overhead of distributed deep learning. Our key observation is that gradient updates can be delayed until an unambiguous (high amplitude, low variance) gradient has been calculated. We also present an efficient algorithm to compute the variance with negligible additional cost. We experimentally show that our method can achieve very high compression ratio while maintaining the result model accuracy. We also analyze the efficiency using computation and communication cost models and provide the evidence that this method enables distributed deep learning for many scenarios with commodity environments." @default.
- W2785456003 created "2018-02-23" @default.
- W2785456003 creator A5041377522 @default.
- W2785456003 creator A5060600154 @default.
- W2785456003 creator A5084654612 @default.
- W2785456003 date "2018-02-16" @default.
- W2785456003 modified "2023-10-14" @default.
- W2785456003 title "Variance-based Gradient Compression for Efficient Distributed Deep Learning" @default.
- W2785456003 cites W104184427 @default.
- W2785456003 cites W1554378292 @default.
- W2785456003 cites W2097117768 @default.
- W2785456003 cites W2131613942 @default.
- W2785456003 cites W2194775991 @default.
- W2785456003 cites W2405578611 @default.
- W2785456003 cites W2407022425 @default.
- W2785456003 cites W2563343794 @default.
- W2785456003 cites W2622263826 @default.
- W2785456003 cites W2622379787 @default.
- W2785456003 cites W2962835968 @default.
- W2785456003 cites W2964004663 @default.
- W2785456003 cites W2964121744 @default.
- W2785456003 cites W3101036738 @default.
- W2785456003 cites W3118608800 @default.
- W2785456003 doi "https://doi.org/10.48550/arxiv.1802.06058" @default.
- W2785456003 hasPublicationYear "2018" @default.
- W2785456003 type Work @default.
- W2785456003 sameAs 2785456003 @default.
- W2785456003 citedByCount "17" @default.
- W2785456003 countsByYear W27854560032019 @default.
- W2785456003 countsByYear W27854560032020 @default.
- W2785456003 countsByYear W27854560032021 @default.
- W2785456003 crossrefType "posted-content" @default.
- W2785456003 hasAuthorship W2785456003A5041377522 @default.
- W2785456003 hasAuthorship W2785456003A5060600154 @default.
- W2785456003 hasAuthorship W2785456003A5084654612 @default.
- W2785456003 hasBestOaLocation W27854560031 @default.
- W2785456003 hasConcept C108583219 @default.
- W2785456003 hasConcept C111919701 @default.
- W2785456003 hasConcept C113775141 @default.
- W2785456003 hasConcept C11413529 @default.
- W2785456003 hasConcept C119857082 @default.
- W2785456003 hasConcept C120314980 @default.
- W2785456003 hasConcept C121955636 @default.
- W2785456003 hasConcept C127413603 @default.
- W2785456003 hasConcept C144133560 @default.
- W2785456003 hasConcept C154945302 @default.
- W2785456003 hasConcept C171146098 @default.
- W2785456003 hasConcept C196083921 @default.
- W2785456003 hasConcept C25797200 @default.
- W2785456003 hasConcept C26517878 @default.
- W2785456003 hasConcept C2776257435 @default.
- W2785456003 hasConcept C2779960059 @default.
- W2785456003 hasConcept C2984842247 @default.
- W2785456003 hasConcept C31258907 @default.
- W2785456003 hasConcept C38652104 @default.
- W2785456003 hasConcept C41008148 @default.
- W2785456003 hasConcept C45374587 @default.
- W2785456003 hasConcept C50644808 @default.
- W2785456003 hasConcept C511840579 @default.
- W2785456003 hasConcept C81363708 @default.
- W2785456003 hasConceptScore W2785456003C108583219 @default.
- W2785456003 hasConceptScore W2785456003C111919701 @default.
- W2785456003 hasConceptScore W2785456003C113775141 @default.
- W2785456003 hasConceptScore W2785456003C11413529 @default.
- W2785456003 hasConceptScore W2785456003C119857082 @default.
- W2785456003 hasConceptScore W2785456003C120314980 @default.
- W2785456003 hasConceptScore W2785456003C121955636 @default.
- W2785456003 hasConceptScore W2785456003C127413603 @default.
- W2785456003 hasConceptScore W2785456003C144133560 @default.
- W2785456003 hasConceptScore W2785456003C154945302 @default.
- W2785456003 hasConceptScore W2785456003C171146098 @default.
- W2785456003 hasConceptScore W2785456003C196083921 @default.
- W2785456003 hasConceptScore W2785456003C25797200 @default.
- W2785456003 hasConceptScore W2785456003C26517878 @default.
- W2785456003 hasConceptScore W2785456003C2776257435 @default.
- W2785456003 hasConceptScore W2785456003C2779960059 @default.
- W2785456003 hasConceptScore W2785456003C2984842247 @default.
- W2785456003 hasConceptScore W2785456003C31258907 @default.
- W2785456003 hasConceptScore W2785456003C38652104 @default.
- W2785456003 hasConceptScore W2785456003C41008148 @default.
- W2785456003 hasConceptScore W2785456003C45374587 @default.
- W2785456003 hasConceptScore W2785456003C50644808 @default.
- W2785456003 hasConceptScore W2785456003C511840579 @default.
- W2785456003 hasConceptScore W2785456003C81363708 @default.
- W2785456003 hasLocation W27854560031 @default.
- W2785456003 hasOpenAccess W2785456003 @default.
- W2785456003 hasPrimaryLocation W27854560031 @default.
- W2785456003 hasRelatedWork W2279398222 @default.
- W2785456003 hasRelatedWork W2337926734 @default.
- W2785456003 hasRelatedWork W2785456003 @default.
- W2785456003 hasRelatedWork W2915754718 @default.
- W2785456003 hasRelatedWork W2962710991 @default.
- W2785456003 hasRelatedWork W3133861977 @default.
- W2785456003 hasRelatedWork W4299822940 @default.
- W2785456003 hasRelatedWork W4311257506 @default.
- W2785456003 hasRelatedWork W4319994054 @default.
- W2785456003 hasRelatedWork W4366224123 @default.
- W2785456003 isParatext "false" @default.
- W2785456003 isRetracted "false" @default.
- W2785456003 magId "2785456003" @default.