FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3087788237> ?p ?o ?g. }

• W3087788237 abstract "As the class size grows, maintaining a balanced dataset across many classes is challenging because the data are long-tailed in nature; it is even impossible when the sample-of-interest co-exists with each other in one collectable unit, e.g., multiple visual instances in one image. Therefore, long-tailed classification is the key to deep learning at scale. However, existing methods are mainly based on re-weighting/re-sampling heuristics that lack a fundamental theory. In this paper, we establish a causal inference framework, which not only unravels the whys of previous methods, but also derives a new principled solution. Specifically, our theory shows that the SGD momentum is essentially a confounder in long-tailed classification. On one hand, it has a harmful causal effect that misleads the tail prediction biased towards the head. On the other hand, its induced mediation also benefits the representation learning and head prediction. Our framework elegantly disentangles the paradoxical effects of the momentum, by pursuing the direct causal effect caused by an input sample. In particular, we use causal intervention in training, and counterfactual reasoning in inference, to remove the bad while keep the good. We achieve new state-of-the-arts on three long-tailed visual recognition benchmarks: Long-tailed CIFAR-10/-100, ImageNet-LT for image classification and LVIS for instance segmentation." @default.
• W3087788237 created "2020-10-01" @default.
• W3087788237 creator A5042324027 @default.
• W3087788237 creator A5044120384 @default.
• W3087788237 creator A5079865276 @default.
• W3087788237 date "2020-09-28" @default.
• W3087788237 modified "2023-09-23" @default.
• W3087788237 title "Long-Tailed Classification by Keeping the Good and Removing the Bad Momentum Causal Effect" @default.
• W3087788237 cites W104184427 @default.
• W3087788237 cites W1501375624 @default.
• W3087788237 cites W1522301498 @default.
• W3087788237 cites W1861492603 @default.
• W3087788237 cites W1980287119 @default.
• W3087788237 cites W2003905926 @default.
• W3087788237 cites W2023060295 @default.
• W3087788237 cites W2036193982 @default.
• W3087788237 cites W2049910836 @default.
• W3087788237 cites W2107091653 @default.
• W3087788237 cites W2117539524 @default.
• W3087788237 cites W2136903812 @default.
• W3087788237 cites W2140310134 @default.
• W3087788237 cites W2141231045 @default.
• W3087788237 cites W2148143831 @default.
• W3087788237 cites W2148385623 @default.
• W3087788237 cites W2161914416 @default.
• W3087788237 cites W2194775991 @default.
• W3087788237 cites W2339172597 @default.
• W3087788237 cites W2486285194 @default.
• W3087788237 cites W2518108298 @default.
• W3087788237 cites W2549139847 @default.
• W3087788237 cites W2565639579 @default.
• W3087788237 cites W2752128027 @default.
• W3087788237 cites W2795282075 @default.
• W3087788237 cites W2796346823 @default.
• W3087788237 cites W2798836702 @default.
• W3087788237 cites W2799269579 @default.
• W3087788237 cites W2903571680 @default.
• W3087788237 cites W2939117106 @default.
• W3087788237 cites W2948672349 @default.
• W3087788237 cites W2950141105 @default.
• W3087788237 cites W2953106684 @default.
• W3087788237 cites W2962858109 @default.
• W3087788237 cites W2963351448 @default.
• W3087788237 cites W2963399829 @default.
• W3087788237 cites W2963403868 @default.
• W3087788237 cites W2963691377 @default.
• W3087788237 cites W2963703618 @default.
• W3087788237 cites W2963741406 @default.
• W3087788237 cites W2964050365 @default.
• W3087788237 cites W2964241181 @default.
• W3087788237 cites W2970711760 @default.
• W3087788237 cites W2970941190 @default.
• W3087788237 cites W2995197345 @default.
• W3087788237 cites W3009499121 @default.
• W3087788237 cites W3010512657 @default.
• W3087788237 cites W3012424925 @default.
• W3087788237 cites W3013774879 @default.
• W3087788237 cites W3033665206 @default.
• W3087788237 cites W3034369739 @default.
• W3087788237 cites W3034601242 @default.
• W3087788237 cites W3035017890 @default.
• W3087788237 cites W3035103424 @default.
• W3087788237 cites W3035342001 @default.
• W3087788237 cites W3035797737 @default.
• W3087788237 cites W3095707208 @default.
• W3087788237 cites W3102146042 @default.
• W3087788237 cites W3102714038 @default.
• W3087788237 hasPublicationYear "2020" @default.
• W3087788237 type Work @default.
• W3087788237 sameAs 3087788237 @default.
• W3087788237 citedByCount "4" @default.
• W3087788237 countsByYear W30877882372020 @default.
• W3087788237 countsByYear W30877882372022 @default.
• W3087788237 crossrefType "posted-content" @default.
• W3087788237 hasAuthorship W3087788237A5042324027 @default.
• W3087788237 hasAuthorship W3087788237A5044120384 @default.
• W3087788237 hasAuthorship W3087788237A5079865276 @default.
• W3087788237 hasConcept C10138342 @default.
• W3087788237 hasConcept C108650721 @default.
• W3087788237 hasConcept C111919701 @default.
• W3087788237 hasConcept C119857082 @default.
• W3087788237 hasConcept C127705205 @default.
• W3087788237 hasConcept C149782125 @default.
• W3087788237 hasConcept C154945302 @default.
• W3087788237 hasConcept C15744967 @default.
• W3087788237 hasConcept C158600405 @default.
• W3087788237 hasConcept C162324750 @default.
• W3087788237 hasConcept C2776214188 @default.
• W3087788237 hasConcept C33923547 @default.
• W3087788237 hasConcept C41008148 @default.
• W3087788237 hasConcept C60718061 @default.
• W3087788237 hasConcept C77805123 @default.
• W3087788237 hasConceptScore W3087788237C10138342 @default.
• W3087788237 hasConceptScore W3087788237C108650721 @default.
• W3087788237 hasConceptScore W3087788237C111919701 @default.
• W3087788237 hasConceptScore W3087788237C119857082 @default.
• W3087788237 hasConceptScore W3087788237C127705205 @default.
• W3087788237 hasConceptScore W3087788237C149782125 @default.
• W3087788237 hasConceptScore W3087788237C154945302 @default.
• W3087788237 hasConceptScore W3087788237C15744967 @default.

• next