Matches in SemOpenAlex for { <https://semopenalex.org/work/W2994068082> ?p ?o ?g. }
- W2994068082 abstract "Recently, an extensive amount of research has been focused on compressing and accelerating Deep Neural Networks (DNN). So far, high compression rate algorithms require part of the training dataset for a low precision calibration, or a fine-tuning process. However, this requirement is unacceptable when the data is unavailable or contains sensitive information, as in medical and biometric use-cases. We present three methods for generating synthetic samples from trained models. Then, we demonstrate how these samples can be used to calibrate and fine-tune quantized models without using any real data in the process. Our best performing method has a negligible accuracy degradation compared to the original training set. This method, which leverages intrinsic batch normalization layers' statistics of the trained model, can be used to evaluate data similarity. Our approach opens a path towards genuine data-free model compression, alleviating the need for training data during model deployment." @default.
- W2994068082 created "2019-12-13" @default.
- W2994068082 creator A5001778496 @default.
- W2994068082 creator A5025123281 @default.
- W2994068082 creator A5057489819 @default.
- W2994068082 creator A5087518651 @default.
- W2994068082 date "2019-12-03" @default.
- W2994068082 modified "2023-09-27" @default.
- W2994068082 title "The Knowledge Within: Methods for Data-Free Model Compression" @default.
- W2994068082 cites W1821462560 @default.
- W2994068082 cites W1836465849 @default.
- W2994068082 cites W2117539524 @default.
- W2994068082 cites W2194775991 @default.
- W2994068082 cites W2242818861 @default.
- W2994068082 cites W2524428287 @default.
- W2994068082 cites W2525778437 @default.
- W2994068082 cites W2765407302 @default.
- W2994068082 cites W2766966408 @default.
- W2994068082 cites W2809624076 @default.
- W2994068082 cites W2884805522 @default.
- W2994068082 cites W2890616119 @default.
- W2994068082 cites W2893749619 @default.
- W2994068082 cites W2898422183 @default.
- W2994068082 cites W2899771611 @default.
- W2994068082 cites W2914611487 @default.
- W2994068082 cites W2946830201 @default.
- W2994068082 cites W2950297794 @default.
- W2994068082 cites W2950726407 @default.
- W2994068082 cites W2951004968 @default.
- W2994068082 cites W2951574208 @default.
- W2994068082 cites W2963163009 @default.
- W2994068082 cites W2963207607 @default.
- W2994068082 cites W2963373786 @default.
- W2994068082 cites W2963446712 @default.
- W2994068082 cites W2963981733 @default.
- W2994068082 cites W2964137095 @default.
- W2994068082 cites W2997609712 @default.
- W2994068082 cites W3034957837 @default.
- W2994068082 cites W3118608800 @default.
- W2994068082 hasPublicationYear "2019" @default.
- W2994068082 type Work @default.
- W2994068082 sameAs 2994068082 @default.
- W2994068082 citedByCount "8" @default.
- W2994068082 countsByYear W29940680822020 @default.
- W2994068082 countsByYear W29940680822021 @default.
- W2994068082 crossrefType "posted-content" @default.
- W2994068082 hasAuthorship W2994068082A5001778496 @default.
- W2994068082 hasAuthorship W2994068082A5025123281 @default.
- W2994068082 hasAuthorship W2994068082A5057489819 @default.
- W2994068082 hasAuthorship W2994068082A5087518651 @default.
- W2994068082 hasConcept C105795698 @default.
- W2994068082 hasConcept C111919701 @default.
- W2994068082 hasConcept C119857082 @default.
- W2994068082 hasConcept C124101348 @default.
- W2994068082 hasConcept C127413603 @default.
- W2994068082 hasConcept C136886441 @default.
- W2994068082 hasConcept C144024400 @default.
- W2994068082 hasConcept C153180895 @default.
- W2994068082 hasConcept C154945302 @default.
- W2994068082 hasConcept C159985019 @default.
- W2994068082 hasConcept C165838908 @default.
- W2994068082 hasConcept C171146098 @default.
- W2994068082 hasConcept C177264268 @default.
- W2994068082 hasConcept C180016635 @default.
- W2994068082 hasConcept C19165224 @default.
- W2994068082 hasConcept C192562407 @default.
- W2994068082 hasConcept C199360897 @default.
- W2994068082 hasConcept C25797200 @default.
- W2994068082 hasConcept C33923547 @default.
- W2994068082 hasConcept C41008148 @default.
- W2994068082 hasConcept C50644808 @default.
- W2994068082 hasConcept C511840579 @default.
- W2994068082 hasConcept C51632099 @default.
- W2994068082 hasConcept C58489278 @default.
- W2994068082 hasConcept C73555534 @default.
- W2994068082 hasConcept C78548338 @default.
- W2994068082 hasConcept C98045186 @default.
- W2994068082 hasConceptScore W2994068082C105795698 @default.
- W2994068082 hasConceptScore W2994068082C111919701 @default.
- W2994068082 hasConceptScore W2994068082C119857082 @default.
- W2994068082 hasConceptScore W2994068082C124101348 @default.
- W2994068082 hasConceptScore W2994068082C127413603 @default.
- W2994068082 hasConceptScore W2994068082C136886441 @default.
- W2994068082 hasConceptScore W2994068082C144024400 @default.
- W2994068082 hasConceptScore W2994068082C153180895 @default.
- W2994068082 hasConceptScore W2994068082C154945302 @default.
- W2994068082 hasConceptScore W2994068082C159985019 @default.
- W2994068082 hasConceptScore W2994068082C165838908 @default.
- W2994068082 hasConceptScore W2994068082C171146098 @default.
- W2994068082 hasConceptScore W2994068082C177264268 @default.
- W2994068082 hasConceptScore W2994068082C180016635 @default.
- W2994068082 hasConceptScore W2994068082C19165224 @default.
- W2994068082 hasConceptScore W2994068082C192562407 @default.
- W2994068082 hasConceptScore W2994068082C199360897 @default.
- W2994068082 hasConceptScore W2994068082C25797200 @default.
- W2994068082 hasConceptScore W2994068082C33923547 @default.
- W2994068082 hasConceptScore W2994068082C41008148 @default.
- W2994068082 hasConceptScore W2994068082C50644808 @default.
- W2994068082 hasConceptScore W2994068082C511840579 @default.
- W2994068082 hasConceptScore W2994068082C51632099 @default.