Matches in SemOpenAlex for { <https://semopenalex.org/work/W3006817794> ?p ?o ?g. }
- W3006817794 abstract "The communication of gradients is costly for training deep neural networks with multiple devices in computer vision applications. In particular, the growing size of deep learning models leads to higher communication overheads that defy the ideal linear training speedup regarding the number of devices. Gradient quantization is one of the common methods to reduce communication costs. However, it can lead to quantization error in the training and result in model performance degradation. In this work, we deduce the optimal condition of both the binary and multi-level gradient quantization for textbf{ANY} gradient distribution. Based on the optimal condition, we develop two novel quantization schemes: biased BinGrad and unbiased ORQ for binary and multi-level gradient quantization respectively, which dynamically determine the optimal quantization levels. Extensive experimental results on CIFAR and ImageNet datasets with several popular convolutional neural networks show the superiority of our proposed methods." @default.
- W3006817794 created "2020-03-06" @default.
- W3006817794 creator A5060016795 @default.
- W3006817794 creator A5063099787 @default.
- W3006817794 creator A5069872293 @default.
- W3006817794 date "2020-02-25" @default.
- W3006817794 modified "2023-10-16" @default.
- W3006817794 title "Optimal Gradient Quantization Condition for Communication-Efficient Distributed Training." @default.
- W3006817794 cites W1686810756 @default.
- W3006817794 cites W1849277567 @default.
- W3006817794 cites W2097117768 @default.
- W3006817794 cites W2117539524 @default.
- W3006817794 cites W2163605009 @default.
- W3006817794 cites W2194775991 @default.
- W3006817794 cites W2267635276 @default.
- W3006817794 cites W2300242332 @default.
- W3006817794 cites W2302255633 @default.
- W3006817794 cites W2405920868 @default.
- W3006817794 cites W2469490737 @default.
- W3006817794 cites W2535838896 @default.
- W3006817794 cites W2586654419 @default.
- W3006817794 cites W2769644379 @default.
- W3006817794 cites W2786548044 @default.
- W3006817794 cites W2803965032 @default.
- W3006817794 cites W2884150179 @default.
- W3006817794 cites W2890924858 @default.
- W3006817794 cites W2907494684 @default.
- W3006817794 cites W2912168260 @default.
- W3006817794 cites W2919115771 @default.
- W3006817794 cites W2920547552 @default.
- W3006817794 cites W2948092338 @default.
- W3006817794 cites W2950826569 @default.
- W3006817794 cites W2963446712 @default.
- W3006817794 cites W2963480671 @default.
- W3006817794 cites W2963540381 @default.
- W3006817794 cites W2963542991 @default.
- W3006817794 cites W2963547822 @default.
- W3006817794 cites W2963664311 @default.
- W3006817794 cites W2963766684 @default.
- W3006817794 cites W2963803379 @default.
- W3006817794 cites W2964004663 @default.
- W3006817794 cites W2964008506 @default.
- W3006817794 cites W2964299589 @default.
- W3006817794 cites W2970971581 @default.
- W3006817794 cites W3118608800 @default.
- W3006817794 hasPublicationYear "2020" @default.
- W3006817794 type Work @default.
- W3006817794 sameAs 3006817794 @default.
- W3006817794 citedByCount "0" @default.
- W3006817794 crossrefType "posted-content" @default.
- W3006817794 hasAuthorship W3006817794A5060016795 @default.
- W3006817794 hasAuthorship W3006817794A5063099787 @default.
- W3006817794 hasAuthorship W3006817794A5069872293 @default.
- W3006817794 hasConcept C101765175 @default.
- W3006817794 hasConcept C11413529 @default.
- W3006817794 hasConcept C154945302 @default.
- W3006817794 hasConcept C173608175 @default.
- W3006817794 hasConcept C28855332 @default.
- W3006817794 hasConcept C2984842247 @default.
- W3006817794 hasConcept C33923547 @default.
- W3006817794 hasConcept C41008148 @default.
- W3006817794 hasConcept C48372109 @default.
- W3006817794 hasConcept C50644808 @default.
- W3006817794 hasConcept C68339613 @default.
- W3006817794 hasConcept C76155785 @default.
- W3006817794 hasConcept C81363708 @default.
- W3006817794 hasConcept C93372532 @default.
- W3006817794 hasConcept C94375191 @default.
- W3006817794 hasConceptScore W3006817794C101765175 @default.
- W3006817794 hasConceptScore W3006817794C11413529 @default.
- W3006817794 hasConceptScore W3006817794C154945302 @default.
- W3006817794 hasConceptScore W3006817794C173608175 @default.
- W3006817794 hasConceptScore W3006817794C28855332 @default.
- W3006817794 hasConceptScore W3006817794C2984842247 @default.
- W3006817794 hasConceptScore W3006817794C33923547 @default.
- W3006817794 hasConceptScore W3006817794C41008148 @default.
- W3006817794 hasConceptScore W3006817794C48372109 @default.
- W3006817794 hasConceptScore W3006817794C50644808 @default.
- W3006817794 hasConceptScore W3006817794C68339613 @default.
- W3006817794 hasConceptScore W3006817794C76155785 @default.
- W3006817794 hasConceptScore W3006817794C81363708 @default.
- W3006817794 hasConceptScore W3006817794C93372532 @default.
- W3006817794 hasConceptScore W3006817794C94375191 @default.
- W3006817794 hasLocation W30068177941 @default.
- W3006817794 hasOpenAccess W3006817794 @default.
- W3006817794 hasPrimaryLocation W30068177941 @default.
- W3006817794 hasRelatedWork W2777406049 @default.
- W3006817794 hasRelatedWork W2799243248 @default.
- W3006817794 hasRelatedWork W2804635205 @default.
- W3006817794 hasRelatedWork W2806978588 @default.
- W3006817794 hasRelatedWork W2891735383 @default.
- W3006817794 hasRelatedWork W2964472257 @default.
- W3006817794 hasRelatedWork W2968683539 @default.
- W3006817794 hasRelatedWork W2970472742 @default.
- W3006817794 hasRelatedWork W3008258049 @default.
- W3006817794 hasRelatedWork W3016081801 @default.
- W3006817794 hasRelatedWork W3089532445 @default.
- W3006817794 hasRelatedWork W3097304224 @default.
- W3006817794 hasRelatedWork W3105131457 @default.
- W3006817794 hasRelatedWork W3123666279 @default.