ProMix: Combating Label Noise via Maximizing Clean Sample Utility

Benchmark Model Rank Results
learning-with-noisy-labels-on-cifar-100nProMix#2Accuracy (mean): 73.39
learning-with-noisy-labels-on-cifar-10nProMix#1Accuracy (mean): 97.39
learning-with-noisy-labels-on-cifar-10n-1ProMix#1Accuracy (mean): 96.97
learning-with-noisy-labels-on-cifar-10n-worstProMix#1Accuracy (mean): 96.16