How does Disagreement Help Generalization against Label Corruption?

Benchmark Model Rank Results
learning-with-noisy-labels-on-cifar-100nCo-Teaching+#15Accuracy (mean): 57.88
learning-with-noisy-labels-on-cifar-10n-random1Co-Teaching+#15Accuracy (mean): 89.70
learning-with-noisy-labels-on-cifar-10n-random2Co-Teaching+#15Accuracy (mean): 89.47
learning-with-noisy-labels-on-cifar-10n-random3Co-Teaching+#15Accuracy (mean): 89.54
learning-with-noisy-labels-on-cifar-10n-worstCo-Teaching+#15Accuracy (mean): 83.26
learning-with-noisy-labels-on-cifar-10n-aggregateCo-Teaching+#19Accuracy (mean): 90.61