Matches in SemOpenAlex for { <https://semopenalex.org/work/W3133281092> ?p ?o ?g. }
- W3133281092 abstract "Instance discriminative self-supervised representation learning has been attracted attention thanks to its unsupervised nature and informative feature representation for downstream tasks. In practice, it commonly uses a larger number of negative samples than the number of supervised classes. However, there is an inconsistency in the existing analysis; theoretically, a large number of negative samples degrade classification performance on a downstream supervised task, while empirically, they improve the performance. We provide a novel framework to analyze this empirical result regarding negative samples using the coupon collector's problem. Our bound can implicitly incorporate the supervised loss of the downstream task in the self-supervised loss by increasing the number of negative samples. We confirm that our proposed analysis holds on real-world benchmark datasets." @default.
- W3133281092 created "2021-03-01" @default.
- W3133281092 creator A5019732992 @default.
- W3133281092 creator A5060421432 @default.
- W3133281092 date "2021-02-13" @default.
- W3133281092 modified "2023-10-03" @default.
- W3133281092 title "Understanding Negative Samples in Instance Discriminative Self-supervised Representation Learning" @default.
- W3133281092 cites W104184427 @default.
- W3133281092 cites W1486649854 @default.
- W3133281092 cites W1836465849 @default.
- W3133281092 cites W1897801964 @default.
- W3133281092 cites W2005830269 @default.
- W3133281092 cites W2011301426 @default.
- W3133281092 cites W2025768430 @default.
- W3133281092 cites W2101234009 @default.
- W3133281092 cites W2108598243 @default.
- W3133281092 cites W2121949863 @default.
- W3133281092 cites W2122925692 @default.
- W3133281092 cites W2138204974 @default.
- W3133281092 cites W2148349024 @default.
- W3133281092 cites W2153579005 @default.
- W3133281092 cites W2163922914 @default.
- W3133281092 cites W2169071224 @default.
- W3133281092 cites W2194775991 @default.
- W3133281092 cites W2251658415 @default.
- W3133281092 cites W2308529009 @default.
- W3133281092 cites W2321533354 @default.
- W3133281092 cites W2326925005 @default.
- W3133281092 cites W2526800167 @default.
- W3133281092 cites W2558661413 @default.
- W3133281092 cites W2601450892 @default.
- W3133281092 cites W2757910899 @default.
- W3133281092 cites W2798991696 @default.
- W3133281092 cites W2804935296 @default.
- W3133281092 cites W2842511635 @default.
- W3133281092 cites W2883725317 @default.
- W3133281092 cites W2913939497 @default.
- W3133281092 cites W2917551568 @default.
- W3133281092 cites W2962742544 @default.
- W3133281092 cites W2962772361 @default.
- W3133281092 cites W2962971234 @default.
- W3133281092 cites W2963012544 @default.
- W3133281092 cites W2963263347 @default.
- W3133281092 cites W2963341956 @default.
- W3133281092 cites W2963420272 @default.
- W3133281092 cites W2963626623 @default.
- W3133281092 cites W2963644595 @default.
- W3133281092 cites W2970241862 @default.
- W3133281092 cites W2970971581 @default.
- W3133281092 cites W2971155163 @default.
- W3133281092 cites W2995480165 @default.
- W3133281092 cites W3034978746 @default.
- W3133281092 cites W3035164673 @default.
- W3133281092 cites W3035524453 @default.
- W3133281092 cites W3037428746 @default.
- W3133281092 cites W3046882683 @default.
- W3133281092 cites W3086105743 @default.
- W3133281092 cites W3091555878 @default.
- W3133281092 cites W3095121901 @default.
- W3133281092 cites W3096655658 @default.
- W3133281092 cites W3098903812 @default.
- W3133281092 cites W3100859887 @default.
- W3133281092 cites W3101821705 @default.
- W3133281092 cites W3104618597 @default.
- W3133281092 cites W3106485021 @default.
- W3133281092 cites W3118062200 @default.
- W3133281092 cites W3118608800 @default.
- W3133281092 cites W3120044914 @default.
- W3133281092 cites W3122325173 @default.
- W3133281092 cites W3131755153 @default.
- W3133281092 cites W3132004818 @default.
- W3133281092 cites W3149173402 @default.
- W3133281092 cites W3150635270 @default.
- W3133281092 cites W3166872541 @default.
- W3133281092 cites W3169212847 @default.
- W3133281092 cites W3171007011 @default.
- W3133281092 cites W3171229070 @default.
- W3133281092 cites W3175214727 @default.
- W3133281092 cites W3213189520 @default.
- W3133281092 doi "https://doi.org/10.48550/arxiv.2102.06866" @default.
- W3133281092 hasPublicationYear "2021" @default.
- W3133281092 type Work @default.
- W3133281092 sameAs 3133281092 @default.
- W3133281092 citedByCount "2" @default.
- W3133281092 countsByYear W31332810922021 @default.
- W3133281092 crossrefType "posted-content" @default.
- W3133281092 hasAuthorship W3133281092A5019732992 @default.
- W3133281092 hasAuthorship W3133281092A5060421432 @default.
- W3133281092 hasBestOaLocation W31332810921 @default.
- W3133281092 hasConcept C119857082 @default.
- W3133281092 hasConcept C127413603 @default.
- W3133281092 hasConcept C13280743 @default.
- W3133281092 hasConcept C136389625 @default.
- W3133281092 hasConcept C138885662 @default.
- W3133281092 hasConcept C153180895 @default.
- W3133281092 hasConcept C154945302 @default.
- W3133281092 hasConcept C17744445 @default.
- W3133281092 hasConcept C185798385 @default.
- W3133281092 hasConcept C199539241 @default.
- W3133281092 hasConcept C201995342 @default.