Matches in SemOpenAlex for { <https://semopenalex.org/work/W4200435900> ?p ?o ?g. }
Showing items 1 to 100 of
100
with 100 items per page.
- W4200435900 endingPage "125" @default.
- W4200435900 startingPage "113" @default.
- W4200435900 abstract "While previous network compression methods achieve great success, most of them rely on the abundant training data which is, unfortunately, often unavailable in practice due to some reasons, e.g., privacy issues, storage constraints, and transmission limitations. A promising way to solve this problem is to perform compression with a few unlabeled data. Proceeding along this way, we propose a novel few-shot network compression framework named Few-Shot Slimming (FSS). FSS follows the student/teacher paradigm, and contains two steps: (1) construct the student by inheriting principal feature maps from the teacher; (2) refine the student feature representation by knowledge distillation with an enhanced mixing data augmentation method called GridMix. Specifically, in the first step, we employ normalized cross correlation to perform the principal feature analysis, and then theoretically construct a new indicator to select the most informative feature maps from the teacher for the student. The indicator is based on the variances of feature maps which can efficiently quantitate the information richness of the input feature maps in a feature-agnostic manner. In the second step, we perform the knowledge distillation for the initialized student in first step with a novel grid-based mixing data augmentation technique which greatly extends the limited sample dataset. In this way, the student is able to refine its feature representation and achieves a better result. Extensive experiments on multiple benchmarks demonstrate the state-of-the-art performance of FSS. For example, by using 0.2% label-free data of full training set, FSS yields a 60% FLOPs reduction for DenseNet-40 on CIFAR-10 with only a loss of 0.8% in top-1 accuracy, achieving a result on par with that obtained by the conventional full-data methods." @default.
- W4200435900 created "2021-12-31" @default.
- W4200435900 creator A5002372483 @default.
- W4200435900 creator A5013413777 @default.
- W4200435900 creator A5030405058 @default.
- W4200435900 creator A5047602732 @default.
- W4200435900 date "2022-03-01" @default.
- W4200435900 modified "2023-10-16" @default.
- W4200435900 title "Towards efficient network compression via Few-Shot Slimming" @default.
- W4200435900 cites W1903029394 @default.
- W4200435900 cites W2089468765 @default.
- W4200435900 cites W2104636679 @default.
- W4200435900 cites W2117539524 @default.
- W4200435900 cites W2194775991 @default.
- W4200435900 cites W2739879705 @default.
- W4200435900 cites W2905102070 @default.
- W4200435900 cites W2924515500 @default.
- W4200435900 cites W2928560789 @default.
- W4200435900 cites W2962851801 @default.
- W4200435900 cites W2963363373 @default.
- W4200435900 cites W2963382930 @default.
- W4200435900 cites W2963446712 @default.
- W4200435900 cites W2964233199 @default.
- W4200435900 cites W2982802130 @default.
- W4200435900 cites W2992308087 @default.
- W4200435900 cites W2997336365 @default.
- W4200435900 cites W3034342078 @default.
- W4200435900 cites W3034513523 @default.
- W4200435900 cites W3034818206 @default.
- W4200435900 cites W3034957837 @default.
- W4200435900 cites W3082366738 @default.
- W4200435900 cites W3184682079 @default.
- W4200435900 cites W3193895134 @default.
- W4200435900 cites W3196865505 @default.
- W4200435900 cites W4214517123 @default.
- W4200435900 doi "https://doi.org/10.1016/j.neunet.2021.12.011" @default.
- W4200435900 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/34999388" @default.
- W4200435900 hasPublicationYear "2022" @default.
- W4200435900 type Work @default.
- W4200435900 citedByCount "2" @default.
- W4200435900 countsByYear W42004359002023 @default.
- W4200435900 crossrefType "journal-article" @default.
- W4200435900 hasAuthorship W4200435900A5002372483 @default.
- W4200435900 hasAuthorship W4200435900A5013413777 @default.
- W4200435900 hasAuthorship W4200435900A5030405058 @default.
- W4200435900 hasAuthorship W4200435900A5047602732 @default.
- W4200435900 hasConcept C119857082 @default.
- W4200435900 hasConcept C124101348 @default.
- W4200435900 hasConcept C138885662 @default.
- W4200435900 hasConcept C153180895 @default.
- W4200435900 hasConcept C154945302 @default.
- W4200435900 hasConcept C177264268 @default.
- W4200435900 hasConcept C17744445 @default.
- W4200435900 hasConcept C199360897 @default.
- W4200435900 hasConcept C199539241 @default.
- W4200435900 hasConcept C2776359362 @default.
- W4200435900 hasConcept C2776401178 @default.
- W4200435900 hasConcept C2780801425 @default.
- W4200435900 hasConcept C41008148 @default.
- W4200435900 hasConcept C41895202 @default.
- W4200435900 hasConcept C59404180 @default.
- W4200435900 hasConcept C94625758 @default.
- W4200435900 hasConceptScore W4200435900C119857082 @default.
- W4200435900 hasConceptScore W4200435900C124101348 @default.
- W4200435900 hasConceptScore W4200435900C138885662 @default.
- W4200435900 hasConceptScore W4200435900C153180895 @default.
- W4200435900 hasConceptScore W4200435900C154945302 @default.
- W4200435900 hasConceptScore W4200435900C177264268 @default.
- W4200435900 hasConceptScore W4200435900C17744445 @default.
- W4200435900 hasConceptScore W4200435900C199360897 @default.
- W4200435900 hasConceptScore W4200435900C199539241 @default.
- W4200435900 hasConceptScore W4200435900C2776359362 @default.
- W4200435900 hasConceptScore W4200435900C2776401178 @default.
- W4200435900 hasConceptScore W4200435900C2780801425 @default.
- W4200435900 hasConceptScore W4200435900C41008148 @default.
- W4200435900 hasConceptScore W4200435900C41895202 @default.
- W4200435900 hasConceptScore W4200435900C59404180 @default.
- W4200435900 hasConceptScore W4200435900C94625758 @default.
- W4200435900 hasFunder F4320322183 @default.
- W4200435900 hasFunder F4320329742 @default.
- W4200435900 hasLocation W42004359001 @default.
- W4200435900 hasLocation W42004359002 @default.
- W4200435900 hasOpenAccess W4200435900 @default.
- W4200435900 hasPrimaryLocation W42004359001 @default.
- W4200435900 hasRelatedWork W2016461833 @default.
- W4200435900 hasRelatedWork W2052253960 @default.
- W4200435900 hasRelatedWork W2147802381 @default.
- W4200435900 hasRelatedWork W2382607599 @default.
- W4200435900 hasRelatedWork W2546942002 @default.
- W4200435900 hasRelatedWork W2554403468 @default.
- W4200435900 hasRelatedWork W2760085659 @default.
- W4200435900 hasRelatedWork W2785535669 @default.
- W4200435900 hasRelatedWork W2998168123 @default.
- W4200435900 hasRelatedWork W4287995534 @default.
- W4200435900 hasVolume "147" @default.
- W4200435900 isParatext "false" @default.
- W4200435900 isRetracted "false" @default.
- W4200435900 workType "article" @default.