Matches in SemOpenAlex for { <https://semopenalex.org/work/W3087194612> ?p ?o ?g. }
- W3087194612 abstract "Recent work has explored the possibility of pruning neural networks at initialization. We assess proposals for doing so: SNIP (Lee et al., 2019), GraSP (Wang et al., 2020), SynFlow (Tanaka et al., 2020), and magnitude pruning. Although these methods surpass the trivial baseline of random pruning, they remain below the accuracy of magnitude pruning after training, and we endeavor to understand why. We show that, unlike pruning after training, accuracy is the same or higher when randomly shuffling which weights these methods prune within each layer or sampling new initial values. As such, the per-weight pruning decisions made by these methods can be replaced by a per-layer choice of the fraction of weights to prune. This property undermines the claimed justifications for these methods and suggests broader challenges with the underlying pruning heuristics, the desire to prune at initialization, or both." @default.
- W3087194612 created "2020-09-25" @default.
- W3087194612 creator A5021062973 @default.
- W3087194612 creator A5042384857 @default.
- W3087194612 creator A5078716102 @default.
- W3087194612 creator A5091194125 @default.
- W3087194612 date "2020-09-18" @default.
- W3087194612 modified "2023-09-23" @default.
- W3087194612 title "Pruning Neural Networks at Initialization: Why are We Missing the Mark?" @default.
- W3087194612 cites W1677182931 @default.
- W3087194612 cites W2025713906 @default.
- W3087194612 cites W2114766824 @default.
- W3087194612 cites W2145085734 @default.
- W3087194612 cites W2194775991 @default.
- W3087194612 cites W2401231614 @default.
- W3087194612 cites W2764043458 @default.
- W3087194612 cites W2886851211 @default.
- W3087194612 cites W2904243021 @default.
- W3087194612 cites W2915589364 @default.
- W3087194612 cites W2953488952 @default.
- W3087194612 cites W2956434358 @default.
- W3087194612 cites W2962857907 @default.
- W3087194612 cites W2962965870 @default.
- W3087194612 cites W2963247446 @default.
- W3087194612 cites W2963674932 @default.
- W3087194612 cites W2963809228 @default.
- W3087194612 cites W2963813662 @default.
- W3087194612 cites W2995197005 @default.
- W3087194612 cites W2995492258 @default.
- W3087194612 cites W2995816250 @default.
- W3087194612 cites W2996309822 @default.
- W3087194612 cites W2996577930 @default.
- W3087194612 cites W2996642285 @default.
- W3087194612 cites W3030163527 @default.
- W3087194612 cites W3031139557 @default.
- W3087194612 cites W3034290785 @default.
- W3087194612 cites W3034733718 @default.
- W3087194612 cites W3035081900 @default.
- W3087194612 cites W3035180000 @default.
- W3087194612 cites W3035285516 @default.
- W3087194612 cites W3037960716 @default.
- W3087194612 cites W3038041907 @default.
- W3087194612 cites W3104688113 @default.
- W3087194612 hasPublicationYear "2020" @default.
- W3087194612 type Work @default.
- W3087194612 sameAs 3087194612 @default.
- W3087194612 citedByCount "4" @default.
- W3087194612 countsByYear W30871946122020 @default.
- W3087194612 countsByYear W30871946122021 @default.
- W3087194612 crossrefType "posted-content" @default.
- W3087194612 hasAuthorship W3087194612A5021062973 @default.
- W3087194612 hasAuthorship W3087194612A5042384857 @default.
- W3087194612 hasAuthorship W3087194612A5078716102 @default.
- W3087194612 hasAuthorship W3087194612A5091194125 @default.
- W3087194612 hasConcept C108010975 @default.
- W3087194612 hasConcept C111368507 @default.
- W3087194612 hasConcept C111919701 @default.
- W3087194612 hasConcept C114466953 @default.
- W3087194612 hasConcept C119857082 @default.
- W3087194612 hasConcept C126838900 @default.
- W3087194612 hasConcept C12725497 @default.
- W3087194612 hasConcept C127313418 @default.
- W3087194612 hasConcept C127705205 @default.
- W3087194612 hasConcept C149629883 @default.
- W3087194612 hasConcept C154945302 @default.
- W3087194612 hasConcept C167927819 @default.
- W3087194612 hasConcept C171268870 @default.
- W3087194612 hasConcept C178790620 @default.
- W3087194612 hasConcept C183115368 @default.
- W3087194612 hasConcept C185592680 @default.
- W3087194612 hasConcept C199360897 @default.
- W3087194612 hasConcept C2779227376 @default.
- W3087194612 hasConcept C41008148 @default.
- W3087194612 hasConcept C50644808 @default.
- W3087194612 hasConcept C6557445 @default.
- W3087194612 hasConcept C71924100 @default.
- W3087194612 hasConcept C86803240 @default.
- W3087194612 hasConceptScore W3087194612C108010975 @default.
- W3087194612 hasConceptScore W3087194612C111368507 @default.
- W3087194612 hasConceptScore W3087194612C111919701 @default.
- W3087194612 hasConceptScore W3087194612C114466953 @default.
- W3087194612 hasConceptScore W3087194612C119857082 @default.
- W3087194612 hasConceptScore W3087194612C126838900 @default.
- W3087194612 hasConceptScore W3087194612C12725497 @default.
- W3087194612 hasConceptScore W3087194612C127313418 @default.
- W3087194612 hasConceptScore W3087194612C127705205 @default.
- W3087194612 hasConceptScore W3087194612C149629883 @default.
- W3087194612 hasConceptScore W3087194612C154945302 @default.
- W3087194612 hasConceptScore W3087194612C167927819 @default.
- W3087194612 hasConceptScore W3087194612C171268870 @default.
- W3087194612 hasConceptScore W3087194612C178790620 @default.
- W3087194612 hasConceptScore W3087194612C183115368 @default.
- W3087194612 hasConceptScore W3087194612C185592680 @default.
- W3087194612 hasConceptScore W3087194612C199360897 @default.
- W3087194612 hasConceptScore W3087194612C2779227376 @default.
- W3087194612 hasConceptScore W3087194612C41008148 @default.
- W3087194612 hasConceptScore W3087194612C50644808 @default.
- W3087194612 hasConceptScore W3087194612C6557445 @default.
- W3087194612 hasConceptScore W3087194612C71924100 @default.
- W3087194612 hasConceptScore W3087194612C86803240 @default.