Matches in SemOpenAlex for { <https://semopenalex.org/work/W3048045781> ?p ?o ?g. }
- W3048045781 abstract "Deep learning algorithms are well-known to have a propensity for fitting the training data very well and often fit even outliers and mislabeled data points. Such fitting requires memorization of training data labels, a phenomenon that has attracted significant research interest but has not been given a compelling explanation so far. A recent work of Feldman (2019) proposes a theoretical explanation for this phenomenon based on a combination of two insights. First, natural image and data distributions are (informally) known to be long-tailed, that is have a significant fraction of rare and atypical examples. Second, in a simple theoretical model such memorization is necessary for achieving close-to-optimal generalization error when the data distribution is long-tailed. However, no direct empirical evidence for this explanation or even an approach for obtaining such evidence were given. In this work we design experiments to test the key ideas in this theory. The experiments require estimation of the influence of each training example on the accuracy at each test example as well as memorization values of training examples. Estimating these quantities directly is computationally prohibitive but we show that closely-related subsampled influence and memorization values can be estimated much more efficiently. Our experiments demonstrate the significant benefits of memorization for generalization on several standard benchmarks. They also provide quantitative and visually compelling evidence for the theory put forth in (Feldman, 2019)." @default.
- W3048045781 created "2020-08-13" @default.
- W3048045781 creator A5016054994 @default.
- W3048045781 creator A5054675941 @default.
- W3048045781 date "2020-08-09" @default.
- W3048045781 modified "2023-09-27" @default.
- W3048045781 title "What Neural Networks Memorize and Why: Discovering the Long Tail via Influence Estimation" @default.
- W3048045781 cites W2034603029 @default.
- W3048045781 cites W2097117768 @default.
- W3048045781 cites W2145147745 @default.
- W3048045781 cites W2194775991 @default.
- W3048045781 cites W2535690855 @default.
- W3048045781 cites W2597603852 @default.
- W3048045781 cites W2753211788 @default.
- W3048045781 cites W2810365967 @default.
- W3048045781 cites W2890457888 @default.
- W3048045781 cites W2903996579 @default.
- W3048045781 cites W2921113176 @default.
- W3048045781 cites W2923764619 @default.
- W3048045781 cites W2963096987 @default.
- W3048045781 cites W2963446712 @default.
- W3048045781 cites W2963739978 @default.
- W3048045781 cites W2963742538 @default.
- W3048045781 cites W2964034630 @default.
- W3048045781 cites W2964161291 @default.
- W3048045781 cites W2966596228 @default.
- W3048045781 cites W2970290137 @default.
- W3048045781 cites W2970716886 @default.
- W3048045781 cites W2982467521 @default.
- W3048045781 cites W2994848047 @default.
- W3048045781 cites W3006856964 @default.
- W3048045781 cites W3014316192 @default.
- W3048045781 cites W3018252856 @default.
- W3048045781 cites W3035261884 @default.
- W3048045781 cites W3100191342 @default.
- W3048045781 cites W3137695714 @default.
- W3048045781 cites W3172093261 @default.
- W3048045781 hasPublicationYear "2020" @default.
- W3048045781 type Work @default.
- W3048045781 sameAs 3048045781 @default.
- W3048045781 citedByCount "8" @default.
- W3048045781 countsByYear W30480457812020 @default.
- W3048045781 countsByYear W30480457812021 @default.
- W3048045781 crossrefType "posted-content" @default.
- W3048045781 hasAuthorship W3048045781A5016054994 @default.
- W3048045781 hasAuthorship W3048045781A5054675941 @default.
- W3048045781 hasConcept C111472728 @default.
- W3048045781 hasConcept C11413529 @default.
- W3048045781 hasConcept C119857082 @default.
- W3048045781 hasConcept C134306372 @default.
- W3048045781 hasConcept C138885662 @default.
- W3048045781 hasConcept C145420912 @default.
- W3048045781 hasConcept C154945302 @default.
- W3048045781 hasConcept C177148314 @default.
- W3048045781 hasConcept C30038468 @default.
- W3048045781 hasConcept C33923547 @default.
- W3048045781 hasConcept C41008148 @default.
- W3048045781 hasConcept C50335755 @default.
- W3048045781 hasConcept C50644808 @default.
- W3048045781 hasConcept C79337645 @default.
- W3048045781 hasConceptScore W3048045781C111472728 @default.
- W3048045781 hasConceptScore W3048045781C11413529 @default.
- W3048045781 hasConceptScore W3048045781C119857082 @default.
- W3048045781 hasConceptScore W3048045781C134306372 @default.
- W3048045781 hasConceptScore W3048045781C138885662 @default.
- W3048045781 hasConceptScore W3048045781C145420912 @default.
- W3048045781 hasConceptScore W3048045781C154945302 @default.
- W3048045781 hasConceptScore W3048045781C177148314 @default.
- W3048045781 hasConceptScore W3048045781C30038468 @default.
- W3048045781 hasConceptScore W3048045781C33923547 @default.
- W3048045781 hasConceptScore W3048045781C41008148 @default.
- W3048045781 hasConceptScore W3048045781C50335755 @default.
- W3048045781 hasConceptScore W3048045781C50644808 @default.
- W3048045781 hasConceptScore W3048045781C79337645 @default.
- W3048045781 hasLocation W30480457811 @default.
- W3048045781 hasOpenAccess W3048045781 @default.
- W3048045781 hasPrimaryLocation W30480457811 @default.
- W3048045781 hasRelatedWork W2194775991 @default.
- W3048045781 hasRelatedWork W2566079294 @default.
- W3048045781 hasRelatedWork W2787075019 @default.
- W3048045781 hasRelatedWork W2787999646 @default.
- W3048045781 hasRelatedWork W2926417365 @default.
- W3048045781 hasRelatedWork W2963622428 @default.
- W3048045781 hasRelatedWork W2984272647 @default.
- W3048045781 hasRelatedWork W3011428993 @default.
- W3048045781 hasRelatedWork W3034791248 @default.
- W3048045781 hasRelatedWork W3088155926 @default.
- W3048045781 hasRelatedWork W3095273266 @default.
- W3048045781 hasRelatedWork W3100352836 @default.
- W3048045781 hasRelatedWork W3118608800 @default.
- W3048045781 hasRelatedWork W3135893734 @default.
- W3048045781 hasRelatedWork W3148050268 @default.
- W3048045781 hasRelatedWork W3167352803 @default.
- W3048045781 hasRelatedWork W3173849000 @default.
- W3048045781 hasRelatedWork W3206275068 @default.
- W3048045781 hasRelatedWork W862523747 @default.
- W3048045781 hasRelatedWork W2182011477 @default.
- W3048045781 isParatext "false" @default.
- W3048045781 isRetracted "false" @default.
- W3048045781 magId "3048045781" @default.