Image Classification on CIFAR-10 (Varying Data Corruption Noise Levels)
96.83Accuracy (Noise 20%)C2D (ELR+ with SimCLR)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| C2D (ELR+ with SimCLR)Architecture=ResNet-34, Evaluation State=Peak2021.03 | 96.83 | 95.96 | 93.67 | 89.94 | 83.23 | 94.32 | |
| C2D (ELR+ with SimCLR)Architecture=ResNet-34, Evaluation State=Final2021.03 | 96.74 | 95.55 | 93.11 | 89.3 | 80.21 | 93.78 | |
| C2D (DivideMix with SimCLR)Architecture=PreAct ResNet-18, Evaluation State=Peak2021.03 | 96.43 | 95.32 | 94.4 | 93.57 | 89.24 | 93.45 | |
| C2D-DivideMixVal (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 96.3 | 95.2 | 94.4 | 93.5 | — | — | |
| DMLP-DivideMixVal (Use of clean validation set)=true, Accuracy Snapshot=Best2023.02 | 96.3 | 95.8 | 94.5 | 94.3 | — | — | |
| C2D (DivideMix with SimCLR)Architecture=PreAct ResNet-18, Evaluation State=Final2021.03 | 96.23 | 95.15 | 94.3 | 93.42 | 87.72 | 90.75 | |
| C2D-DivideMixVal (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 96.2 | 95.1 | 94.1 | 93.4 | — | — | |
| DMLP-DivideMixVal (Use of clean validation set)=true, Accuracy Snapshot=Last2023.02 | 96.2 | 95.6 | 94.3 | 94 | — | — | |
| DivideMixArchitecture=PreAct ResNet-18, Evaluation State=Peak2021.03 | 96.1 | 94.6 | 93.2 | 76 | — | 93.4 | |
| DivideMixVal (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 96.1 | 94.6 | 93.2 | 76 | — | — | |
| DivideMix*Val (Use of clean validation set)=true, Accuracy Snapshot=Best2023.02 | 96.1 | 94.9 | 93.6 | 77.3 | — | — | |
| DivideMix*Val (Use of clean validation set)=true, Accuracy Snapshot=Last2023.02 | 95.9 | 94.6 | 93 | 76.5 | — | — | |
| ELR+Architecture=ResNet-34, Evaluation State=Final2021.03 | 95.8 | 94.8 | 93.3 | 78.7 | — | 93 | |
| REEDVal (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 95.8 | 95.6 | 94.3 | 93.6 | — | — | |
| DivideMixArchitecture=PreAct ResNet-18, Evaluation State=Final2021.03 | 95.7 | 94.4 | 92.9 | 75.4 | — | 92.1 | |
| REEDVal (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 95.7 | 95.4 | 94.1 | 93.5 | — | — | |
| DivideMixVal (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 95.7 | 94.4 | 92.9 | 75.4 | — | — | |
| Sel-CL+Val (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 95.5 | 93.9 | 89.2 | 81.9 | — | — | |
| Sel-CL+Val (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 95.1 | 93.3 | 88.7 | 81.6 | — | — | |
| CE+mixup with SimCLRArchitecture=ResNet-34, Evaluation State=Peak2021.03 | 94.98 | 93.83 | 91.33 | 87.54 | 73.97 | 91.63 | |
| DMLP-NaiveVal (Use of clean validation set)=true, Accuracy Snapshot=Best2023.02 | 94.7 | 94.2 | 93.5 | 92.8 | — | — | |
| CE+mixup with SimCLRArchitecture=ResNet-34, Evaluation State=Final2021.03 | 94.56 | 93.2 | 90.77 | 85.77 | 72.71 | 90.39 | |
| DMLP-NaiveVal (Use of clean validation set)=true, Accuracy Snapshot=Last2023.02 | 94.2 | 94 | 93.2 | 92 | — | — | |
| MOIT+Val (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 94.1 | 91.8 | 81.1 | 74.7 | — | — | |
| MOIT+Val (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 93.8 | 91.3 | 80.6 | 74 | — | — | |
| MSLCVal (Use of clean validation set)=true, Accuracy Snapshot=Best2023.02 | 93.4 | 89.9 | 69.8 | 56.1 | — | — | |
| MSLCVal (Use of clean validation set)=true, Accuracy Snapshot=Last2023.02 | 93.3 | 89.4 | 68.8 | 55.2 | — | — | |
| Meta-learningArchitecture=PreAct ResNet-32, Evaluation State=Peak2021.03 | 92.9 | 89.3 | 77.4 | 58.7 | — | 89.2 | |
| Meta-LearningVal (Use of clean validation set)=true, Accuracy Snapshot=Best2023.02 | 92.9 | 89.3 | 77.4 | 58.7 | — | — | |
| MLCVal (Use of clean validation set)=true, Accuracy Snapshot=Best2023.02 | 92.6 | 88.1 | 77.4 | 67.9 | — | — | |
| PENCILVal (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 92.4 | 89.1 | 77.5 | 58.9 | — | — | |
| Meta-learningArchitecture=PreAct ResNet-32, Evaluation State=Final2021.03 | 92 | 88.8 | 76.1 | 58.3 | — | 88.6 | |
| PENCILVal (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 92 | 88.7 | 76.5 | 58.2 | — | — | |
| Meta-LearningVal (Use of clean validation set)=true, Accuracy Snapshot=Last2023.02 | 92 | 88.8 | 76.1 | 58.3 | — | — | |
| MLCVal (Use of clean validation set)=true, Accuracy Snapshot=Last2023.02 | 91.8 | 87.5 | 77.1 | 67 | — | — | |
| Co-teaching+Val (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 89.5 | 85.7 | 67.4 | 47.9 | — | — | |
| Co-teaching+Val (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 88.2 | 84.1 | 45.5 | 30.1 | — | — | |
| Cross-Entropy (CE)Val (Use of clean validation set)=false, Accuracy Snapshot=Best2023.02 | 86.8 | 79.1 | 62.9 | 42.7 | — | — | |
| Cross-Entropy (CE)Val (Use of clean validation set)=false, Accuracy Snapshot=Last2023.02 | 82.7 | 57.9 | 26.1 | 16.8 | — | — |