Matches in SemOpenAlex for { <https://semopenalex.org/work/W3144139771> ?p ?o ?g. }
- W3144139771 abstract "Neural networks trained with class-imbalanced data are known to perform poorly on minor classes of scarce training data. Several recent works attribute this to over-fitting to minor classes. In this paper, we provide a novel explanation of this issue. We found that a neural network tends to first under-fit the minor classes by classifying most of their data into the major classes in early training epochs. To correct these wrong predictions, the neural network then must focus on pushing features of minor class data across the decision boundaries between major and minor classes, leading to much larger gradients for features of minor classes. We argue that such an under-fitting phase over-emphasizes the competition between major and minor classes, hinders the neural network from learning the discriminative knowledge that can be generalized to test data, and eventually results in over-fitting. To address this issue, we propose a novel learning strategy to equalize the training progress across classes. We mix features of the major class data with those of other data in a mini-batch, intentionally weakening their features to prevent a neural network from fitting them first. We show that this strategy can largely balance the training accuracy and feature gradients across classes, effectively mitigating the under-fitting then over-fitting problem for minor class data. On several benchmark datasets, our approach achieves the state-of-the-art accuracy, especially for the challenging step-imbalanced cases." @default.
- W3144139771 created "2021-04-13" @default.
- W3144139771 creator A5018066903 @default.
- W3144139771 creator A5065180062 @default.
- W3144139771 creator A5077697772 @default.
- W3144139771 date "2021-04-05" @default.
- W3144139771 modified "2023-09-26" @default.
- W3144139771 title "Procrustean Training for Imbalanced Deep Learning" @default.
- W3144139771 cites W1544092585 @default.
- W3144139771 cites W1861492603 @default.
- W3144139771 cites W2097117768 @default.
- W3144139771 cites W2108598243 @default.
- W3144139771 cites W2118978333 @default.
- W3144139771 cites W2136903812 @default.
- W3144139771 cites W2163605009 @default.
- W3144139771 cites W2187089797 @default.
- W3144139771 cites W2194775991 @default.
- W3144139771 cites W2250384498 @default.
- W3144139771 cites W2277195237 @default.
- W3144139771 cites W2339172597 @default.
- W3144139771 cites W2440599146 @default.
- W3144139771 cites W2469885745 @default.
- W3144139771 cites W2515770085 @default.
- W3144139771 cites W2519584884 @default.
- W3144139771 cites W2549139847 @default.
- W3144139771 cites W2551429935 @default.
- W3144139771 cites W2566079294 @default.
- W3144139771 cites W2590780569 @default.
- W3144139771 cites W2600610494 @default.
- W3144139771 cites W2737691244 @default.
- W3144139771 cites W2753211788 @default.
- W3144139771 cites W2753300133 @default.
- W3144139771 cites W2766890262 @default.
- W3144139771 cites W2767106145 @default.
- W3144139771 cites W2775447965 @default.
- W3144139771 cites W2781292787 @default.
- W3144139771 cites W2797977484 @default.
- W3144139771 cites W2798381792 @default.
- W3144139771 cites W2798869704 @default.
- W3144139771 cites W2800788706 @default.
- W3144139771 cites W2910580498 @default.
- W3144139771 cites W2913592748 @default.
- W3144139771 cites W2921861056 @default.
- W3144139771 cites W2936503027 @default.
- W3144139771 cites W2943560353 @default.
- W3144139771 cites W2950220847 @default.
- W3144139771 cites W2962835968 @default.
- W3144139771 cites W2962895364 @default.
- W3144139771 cites W2962933664 @default.
- W3144139771 cites W2963070905 @default.
- W3144139771 cites W2963078860 @default.
- W3144139771 cites W2963081269 @default.
- W3144139771 cites W2963212406 @default.
- W3144139771 cites W2963351448 @default.
- W3144139771 cites W2963371670 @default.
- W3144139771 cites W2963399829 @default.
- W3144139771 cites W2963446712 @default.
- W3144139771 cites W2963453204 @default.
- W3144139771 cites W2963476860 @default.
- W3144139771 cites W2963702891 @default.
- W3144139771 cites W2963703197 @default.
- W3144139771 cites W2964050365 @default.
- W3144139771 cites W2970711760 @default.
- W3144139771 cites W2986604550 @default.
- W3144139771 cites W2991497298 @default.
- W3144139771 cites W2995197345 @default.
- W3144139771 cites W2997580736 @default.
- W3144139771 cites W3009883635 @default.
- W3144139771 cites W3010512657 @default.
- W3144139771 cites W3021199811 @default.
- W3144139771 cites W3034601242 @default.
- W3144139771 cites W3034711780 @default.
- W3144139771 cites W3034858314 @default.
- W3144139771 cites W3034933032 @default.
- W3144139771 cites W3035552357 @default.
- W3144139771 cites W3041497966 @default.
- W3144139771 cites W3106060611 @default.
- W3144139771 cites W3118608800 @default.
- W3144139771 cites W3126272270 @default.
- W3144139771 cites W2551176409 @default.
- W3144139771 doi "https://doi.org/10.48550/arxiv.2104.01769" @default.
- W3144139771 hasPublicationYear "2021" @default.
- W3144139771 type Work @default.
- W3144139771 sameAs 3144139771 @default.
- W3144139771 citedByCount "0" @default.
- W3144139771 crossrefType "posted-content" @default.
- W3144139771 hasAuthorship W3144139771A5018066903 @default.
- W3144139771 hasAuthorship W3144139771A5065180062 @default.
- W3144139771 hasAuthorship W3144139771A5077697772 @default.
- W3144139771 hasBestOaLocation W31441397711 @default.
- W3144139771 hasConcept C119857082 @default.
- W3144139771 hasConcept C121332964 @default.
- W3144139771 hasConcept C13280743 @default.
- W3144139771 hasConcept C138885662 @default.
- W3144139771 hasConcept C153180895 @default.
- W3144139771 hasConcept C153294291 @default.
- W3144139771 hasConcept C154945302 @default.
- W3144139771 hasConcept C16910744 @default.
- W3144139771 hasConcept C17744445 @default.
- W3144139771 hasConcept C185798385 @default.