Matches in SemOpenAlex for { <https://semopenalex.org/work/W3201998753> ?p ?o ?g. }
Showing items 1 to 89 of
89
with 100 items per page.
- W3201998753 endingPage "1788" @default.
- W3201998753 startingPage "1784" @default.
- W3201998753 abstract "Binary neural networks (BNNs), where both activations and weights are radically quantized to be {-1, +1}, can massively accelerate the run-time performance of convolution neural networks (CNNs) for edge devices, by computation complexity reduction and memory footprint saving. However, the non-differentiable binarizing function used in BNNs, makes the binarized models hard to be optimized, and introduces significant performance degradation than the full-precision models. Many previous works managed to correct the backward gradient of binarizing function with various improved versions of straight-through estimation (STE), or in a gradual approximate approach, but the gradient suppression problem was not analyzed and handled. Thus, we propose a novel gradient corrected approximation (GCA) method to match the discrepancy between binarizing function and backward gradient in a gradual and stable way. Our work has two primary contributions: The first is to approximate the backward gradient of binarizing function using a simple leaky-steep function with variable window size. The second is to correct the gradient approximation by standardizing the backward gradient propagated through binarizing function. Experiment results show that the proposed method outperforms the baseline by 1.5% Top-1 accuracy on ImageNet dataset without introducing extra computation cost." @default.
- W3201998753 created "2021-10-11" @default.
- W3201998753 creator A5006382294 @default.
- W3201998753 creator A5026870474 @default.
- W3201998753 creator A5036803265 @default.
- W3201998753 creator A5064762015 @default.
- W3201998753 creator A5071412289 @default.
- W3201998753 creator A5071549426 @default.
- W3201998753 creator A5078417441 @default.
- W3201998753 date "2021-10-01" @default.
- W3201998753 modified "2023-09-23" @default.
- W3201998753 title "Gradient Corrected Approximation for Binary Neural Networks" @default.
- W3201998753 cites W2300242332 @default.
- W3201998753 cites W2886472289 @default.
- W3201998753 cites W2887447938 @default.
- W3201998753 cites W2977456144 @default.
- W3201998753 cites W3004734842 @default.
- W3201998753 doi "https://doi.org/10.1587/transinf.2021edl8026" @default.
- W3201998753 hasPublicationYear "2021" @default.
- W3201998753 type Work @default.
- W3201998753 sameAs 3201998753 @default.
- W3201998753 citedByCount "1" @default.
- W3201998753 countsByYear W32019987532023 @default.
- W3201998753 crossrefType "journal-article" @default.
- W3201998753 hasAuthorship W3201998753A5006382294 @default.
- W3201998753 hasAuthorship W3201998753A5026870474 @default.
- W3201998753 hasAuthorship W3201998753A5036803265 @default.
- W3201998753 hasAuthorship W3201998753A5064762015 @default.
- W3201998753 hasAuthorship W3201998753A5071412289 @default.
- W3201998753 hasAuthorship W3201998753A5071549426 @default.
- W3201998753 hasAuthorship W3201998753A5078417441 @default.
- W3201998753 hasBestOaLocation W32019987531 @default.
- W3201998753 hasConcept C111919701 @default.
- W3201998753 hasConcept C11413529 @default.
- W3201998753 hasConcept C134306372 @default.
- W3201998753 hasConcept C14036430 @default.
- W3201998753 hasConcept C154945302 @default.
- W3201998753 hasConcept C162307627 @default.
- W3201998753 hasConcept C202615002 @default.
- W3201998753 hasConcept C33923547 @default.
- W3201998753 hasConcept C41008148 @default.
- W3201998753 hasConcept C45347329 @default.
- W3201998753 hasConcept C45374587 @default.
- W3201998753 hasConcept C48372109 @default.
- W3201998753 hasConcept C50644808 @default.
- W3201998753 hasConcept C74912251 @default.
- W3201998753 hasConcept C78458016 @default.
- W3201998753 hasConcept C81363708 @default.
- W3201998753 hasConcept C86803240 @default.
- W3201998753 hasConcept C94375191 @default.
- W3201998753 hasConceptScore W3201998753C111919701 @default.
- W3201998753 hasConceptScore W3201998753C11413529 @default.
- W3201998753 hasConceptScore W3201998753C134306372 @default.
- W3201998753 hasConceptScore W3201998753C14036430 @default.
- W3201998753 hasConceptScore W3201998753C154945302 @default.
- W3201998753 hasConceptScore W3201998753C162307627 @default.
- W3201998753 hasConceptScore W3201998753C202615002 @default.
- W3201998753 hasConceptScore W3201998753C33923547 @default.
- W3201998753 hasConceptScore W3201998753C41008148 @default.
- W3201998753 hasConceptScore W3201998753C45347329 @default.
- W3201998753 hasConceptScore W3201998753C45374587 @default.
- W3201998753 hasConceptScore W3201998753C48372109 @default.
- W3201998753 hasConceptScore W3201998753C50644808 @default.
- W3201998753 hasConceptScore W3201998753C74912251 @default.
- W3201998753 hasConceptScore W3201998753C78458016 @default.
- W3201998753 hasConceptScore W3201998753C81363708 @default.
- W3201998753 hasConceptScore W3201998753C86803240 @default.
- W3201998753 hasConceptScore W3201998753C94375191 @default.
- W3201998753 hasIssue "10" @default.
- W3201998753 hasLocation W32019987531 @default.
- W3201998753 hasOpenAccess W3201998753 @default.
- W3201998753 hasPrimaryLocation W32019987531 @default.
- W3201998753 hasRelatedWork W2296411508 @default.
- W3201998753 hasRelatedWork W2513928851 @default.
- W3201998753 hasRelatedWork W2544912866 @default.
- W3201998753 hasRelatedWork W2767899175 @default.
- W3201998753 hasRelatedWork W2796942851 @default.
- W3201998753 hasRelatedWork W2995097845 @default.
- W3201998753 hasRelatedWork W3007781960 @default.
- W3201998753 hasRelatedWork W3019027124 @default.
- W3201998753 hasRelatedWork W4287991338 @default.
- W3201998753 hasRelatedWork W4297812642 @default.
- W3201998753 hasVolume "E104.D" @default.
- W3201998753 isParatext "false" @default.
- W3201998753 isRetracted "false" @default.
- W3201998753 magId "3201998753" @default.
- W3201998753 workType "article" @default.