Matches in SemOpenAlex for { <https://semopenalex.org/work/W3116136715> ?p ?o ?g. }
- W3116136715 abstract "How can we efficiently compress Convolutional Neural Network (CNN) while retaining their accuracy on classification tasks? Depthwise Separable Convolution (DSConv), which replaces a standard convolution with a depthwise convolution and a pointwise convolution, has been used for building lightweight architectures. However, previous works based on depthwise separable convolution are limited when compressing a trained CNN model since 1) they are mostly heuristic approaches without a precise understanding of their relations to standard convolution, and 2) their accuracies do not match that of the standard convolution. In this paper, we propose FALCON, an accurate and lightweight method to compress CNN. FALCON uses GEP, our proposed mathematical formulation to approximate the standard convolution kernel, to interpret existing convolution methods based on depthwise separable convolution. By exploiting the knowledge of a trained standard model and carefully determining the order of depthwise separable convolution via GEP, FALCON achieves sufficient accuracy close to that of the trained standard model. Furthermore, this interpretation leads to developing a generalized version rank-k FALCON which performs k independent FALCON operations and sums up the result. Experiments show that FALCON 1) provides higher accuracy than existing methods based on depthwise separable convolution and tensor decomposition, and 2) reduces the number of parameters and FLOPs of standard convolution by up to a factor of 8 while ensuring similar accuracy. We also demonstrate that rank-k FALCON further improves the accuracy while sacrificing a bit of compression and computation reduction rates." @default.
- W3116136715 created "2021-01-05" @default.
- W3116136715 creator A5024696928 @default.
- W3116136715 creator A5031726289 @default.
- W3116136715 creator A5074561505 @default.
- W3116136715 creator A5083273971 @default.
- W3116136715 date "2019-09-25" @default.
- W3116136715 modified "2023-09-27" @default.
- W3116136715 title "FALCON: Lightweight and Accurate Convolution" @default.
- W3116136715 cites W1798945469 @default.
- W3116136715 cites W2115120746 @default.
- W3116136715 cites W2163605009 @default.
- W3116136715 cites W2172166488 @default.
- W3116136715 cites W2211192759 @default.
- W3116136715 cites W2382313035 @default.
- W3116136715 cites W2515144511 @default.
- W3116136715 cites W2531409750 @default.
- W3116136715 cites W2588860837 @default.
- W3116136715 cites W2612445135 @default.
- W3116136715 cites W2808868252 @default.
- W3116136715 cites W2883780447 @default.
- W3116136715 cites W2891967877 @default.
- W3116136715 cites W2944779197 @default.
- W3116136715 cites W2962835968 @default.
- W3116136715 cites W2962965870 @default.
- W3116136715 cites W2962988160 @default.
- W3116136715 cites W2963048316 @default.
- W3116136715 cites W2963114950 @default.
- W3116136715 cites W2963125010 @default.
- W3116136715 cites W2963163009 @default.
- W3116136715 cites W2963287528 @default.
- W3116136715 cites W2963424132 @default.
- W3116136715 cites W2963547822 @default.
- W3116136715 cites W2963674932 @default.
- W3116136715 cites W2963813662 @default.
- W3116136715 cites W2963935227 @default.
- W3116136715 cites W2964164354 @default.
- W3116136715 cites W2964299589 @default.
- W3116136715 cites W2964350391 @default.
- W3116136715 cites W2967733054 @default.
- W3116136715 cites W2970657225 @default.
- W3116136715 cites W2982083293 @default.
- W3116136715 cites W3016232661 @default.
- W3116136715 cites W3034797615 @default.
- W3116136715 cites W3089769400 @default.
- W3116136715 hasPublicationYear "2019" @default.
- W3116136715 type Work @default.
- W3116136715 sameAs 3116136715 @default.
- W3116136715 citedByCount "0" @default.
- W3116136715 crossrefType "posted-content" @default.
- W3116136715 hasAuthorship W3116136715A5024696928 @default.
- W3116136715 hasAuthorship W3116136715A5031726289 @default.
- W3116136715 hasAuthorship W3116136715A5074561505 @default.
- W3116136715 hasAuthorship W3116136715A5083273971 @default.
- W3116136715 hasConcept C11413529 @default.
- W3116136715 hasConcept C114614502 @default.
- W3116136715 hasConcept C118615104 @default.
- W3116136715 hasConcept C134306372 @default.
- W3116136715 hasConcept C154945302 @default.
- W3116136715 hasConcept C164226766 @default.
- W3116136715 hasConcept C33923547 @default.
- W3116136715 hasConcept C41008148 @default.
- W3116136715 hasConcept C45347329 @default.
- W3116136715 hasConcept C50644808 @default.
- W3116136715 hasConcept C70710897 @default.
- W3116136715 hasConcept C74193536 @default.
- W3116136715 hasConcept C81363708 @default.
- W3116136715 hasConceptScore W3116136715C11413529 @default.
- W3116136715 hasConceptScore W3116136715C114614502 @default.
- W3116136715 hasConceptScore W3116136715C118615104 @default.
- W3116136715 hasConceptScore W3116136715C134306372 @default.
- W3116136715 hasConceptScore W3116136715C154945302 @default.
- W3116136715 hasConceptScore W3116136715C164226766 @default.
- W3116136715 hasConceptScore W3116136715C33923547 @default.
- W3116136715 hasConceptScore W3116136715C41008148 @default.
- W3116136715 hasConceptScore W3116136715C45347329 @default.
- W3116136715 hasConceptScore W3116136715C50644808 @default.
- W3116136715 hasConceptScore W3116136715C70710897 @default.
- W3116136715 hasConceptScore W3116136715C74193536 @default.
- W3116136715 hasConceptScore W3116136715C81363708 @default.
- W3116136715 hasLocation W31161367151 @default.
- W3116136715 hasOpenAccess W3116136715 @default.
- W3116136715 hasPrimaryLocation W31161367151 @default.
- W3116136715 hasRelatedWork W1501755821 @default.
- W3116136715 hasRelatedWork W1586217608 @default.
- W3116136715 hasRelatedWork W1604736418 @default.
- W3116136715 hasRelatedWork W1902041153 @default.
- W3116136715 hasRelatedWork W1980531051 @default.
- W3116136715 hasRelatedWork W2244934222 @default.
- W3116136715 hasRelatedWork W2524238209 @default.
- W3116136715 hasRelatedWork W2790966157 @default.
- W3116136715 hasRelatedWork W2891432604 @default.
- W3116136715 hasRelatedWork W2923155842 @default.
- W3116136715 hasRelatedWork W2950162667 @default.
- W3116136715 hasRelatedWork W2951400455 @default.
- W3116136715 hasRelatedWork W2963681088 @default.
- W3116136715 hasRelatedWork W2966491199 @default.
- W3116136715 hasRelatedWork W2993495376 @default.
- W3116136715 hasRelatedWork W3003293616 @default.
- W3116136715 hasRelatedWork W3009123798 @default.