Out-of-distribution Detection on CIFAR-10 (ID) vs LSUN (OOD)
99.96AUROCDeep Abstaining Classifier
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Deep Abstaining ClassifierBackbone=Wide ResNet 40x22021.05 | 99.96 | 0.1 | |
| Deep Abstaining ClassifierBackbone=ResNet 342021.05 | 99.91 | 0.3 | |
| Ensemble of Leave-out ClassifiersBackbone=Wide ResNet 28x102021.05 | 99.7 | 0.88 | |
| Deep Abstaining ClassifierBackbone=Wide ResNet 28x102021.05 | 99.61 | 1.5 | |
| Deep Mahalanobis DetectorBackbone=ResNet 342021.05 | 99.5 | 1.9 | |
| LoRot-EForwarded Samples=5M2022.07 | 98.7 | — | |
| LoRot-IForwarded Samples=5M2022.07 | 98.6 | — | |
| CSI+LoRot-IForwarded Samples=280M2022.07 | 98.3 | — | |
| CSI+LoRot-EForwarded Samples=280M2022.07 | 98 | — | |
| Outlier ExposureBackbone=Wide ResNet 40x22021.05 | 97.6 | 12.1 | |
| RotationsReproduced=true, Forwarded Samples=25M2022.07 | 97.3 | — | |
| CSIForwarded Samples=280M2022.07 | 96.3 | — | |
| ODINBackbone=Wide ResNet 28x102021.05 | 95.4 | 17.6 | |
| FedRoD+FOOGDalpha=0.12024.10 | 94.56 | 36.17 | |
| SupCLRForwarded Samples=70M2022.07 | 92.8 | — | |
| CutmixReproduced=true, Forwarded Samples=5M2022.07 | 92.5 | — | |
| OpenMaxBackbone=ResNet 342021.05 | 92.16 | 18.68 | |
| FedAvg+FOOGDalpha=0.12024.10 | 91.22 | 37.55 | |
| Cross EntropyReproduced=true, Forwarded Samples=5M2022.07 | 90.9 | — | |
| SLA+SDReproduced=true, Forwarded Samples=20M2022.07 | 90.7 | — | |
| FedTHEalpha=0.12024.10 | 83.55 | 49.98 | |
| FOSTERalpha=0.12024.10 | 82.95 | 41.8 | |
| FedRoDalpha=0.12024.10 | 82.83 | 48.73 | |
| FedAvgalpha=0.12024.10 | 80.05 | 62.1 | |
| FedIIRalpha=0.12024.10 | 76.82 | 85.39 | |
| FedLNalpha=0.12024.10 | 62.14 | 76.29 | |
| FedATOLalpha=0.12024.10 | 62.01 | 76.03 | |
| FedICONalpha=0.12024.10 | 43.72 | 76.29 |