Making Deep Neural Networks Robust to Label Noise: a Loss Correction Approach

Benchmark Model Rank Results
image-classification-on-clothing1m-usingForward#2Accuracy: 80.27
image-classification-on-mini-webvision-1-0F-Correction (Inception-ResNet-v2)#33Top-1 Accuracy: 61.12Top-5 Accuracy: 82.68
learning-with-noisy-labels-on-cifar-100nBackward-T#17Accuracy (mean): 57.14
learning-with-noisy-labels-on-cifar-100nForward-T#19Accuracy (mean): 57.01
learning-with-noisy-labels-on-cifar-10nForward-T#22Accuracy (mean): 88.24
learning-with-noisy-labels-on-cifar-10nBackward-T#23Accuracy (mean): 88.13
learning-with-noisy-labels-on-cifar-10n-1Backward-T#22Accuracy (mean): 87.14
learning-with-noisy-labels-on-cifar-10n-1Forward-T#23Accuracy (mean): 86.88
learning-with-noisy-labels-on-cifar-10n-2Backward-T#20Accuracy (mean): 86.28
learning-with-noisy-labels-on-cifar-10n-2Forward-T#21Accuracy (mean): 86.14
learning-with-noisy-labels-on-cifar-10n-3Forward-T#20Accuracy (mean): 87.04
learning-with-noisy-labels-on-cifar-10n-3Backward-T#21Accuracy (mean): 86.86
learning-with-noisy-labels-on-cifar-10n-worstForward-T#22Accuracy (mean): 79.79
learning-with-noisy-labels-on-cifar-10n-worstBackward-T#23Accuracy (mean): 77.61