Early-Learning Regularization Prevents Memorization of Noisy Labels

Benchmark Model Rank Results
image-classification-on-clothing1mELR+#13Accuracy: 74.81%
image-classification-on-mini-webvision-1-0ELR+ (Inception-ResNet-v2)#21Top-1 Accuracy: 77.78Top-5 Accuracy: 91.68
learning-with-noisy-labels-on-cifar-100nELR+#6Accuracy (mean): 66.72
learning-with-noisy-labels-on-cifar-100nELR#12Accuracy (mean): 58.94
learning-with-noisy-labels-on-cifar-10nELR+#8Accuracy (mean): 94.83
learning-with-noisy-labels-on-cifar-10nELR#11Accuracy (mean): 92.38
learning-with-noisy-labels-on-cifar-10n-1ELR+#7Accuracy (mean): 94.43
learning-with-noisy-labels-on-cifar-10n-1ELR#9Accuracy (mean): 91.46
learning-with-noisy-labels-on-cifar-10n-2ELR+#5Accuracy (mean): 94.20
learning-with-noisy-labels-on-cifar-10n-2ELR#6Accuracy (mean): 91.61
learning-with-noisy-labels-on-cifar-10n-3ELR+#5Accuracy (mean): 94.34
learning-with-noisy-labels-on-cifar-10n-3ELR#7Accuracy (mean): 91.41
learning-with-noisy-labels-on-cifar-10n-worstELR+#8Accuracy (mean): 91.09
learning-with-noisy-labels-on-cifar-10n-worstELR#13Accuracy (mean): 83.58