Training Data Provenance Verification on CIFAR10
100Avg AUCCREDIT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CREDITBackbone=ResNet2026.02 | 100 | — | — | |
| CREDITBackbone=VGG2026.02 | 100 | — | — | |
| CREDITBackbone=DenseNet2026.02 | 100 | — | — | |
| CREDITBackbone=GoogLeNet2026.02 | 100 | — | — | |
| TrainProVeBackbone=ResNet182025.03 | 99.8 | 99.7 | 99.2 | |
| TrainProVeBackbone=Swin-B2025.03 | 99.8 | 99.7 | 99.2 | |
| TrainProVeBackbone=ConvNeXt-B2025.03 | 99.7 | 99.5 | 98.9 | |
| TrainProVe-Ent2025.03 | 90.2 | 86.8 | 77.4 | |
| TrainProVe-Sim2025.03 | 82.1 | 81.2 | 64.7 | |
| BackdooringBackbone=GoogLeNet2026.02 | 80.25 | — | — | |
| UAPBackbone=GoogLeNet2026.02 | 79.63 | — | — | |
| BackdooringBackbone=DenseNet2026.02 | 74.07 | — | — | |
| Han et al.'s Work2025.03 | 73.1 | 76.8 | 57.9 | |
| UAPBackbone=DenseNet2026.02 | 72.22 | — | — | |
| IPGuardBackbone=DenseNet2026.02 | 67.9 | — | — | |
| UAPBackbone=VGG2026.02 | 67.28 | — | — | |
| EWEBackbone=VGG2026.02 | 62.96 | — | — | |
| EWEBackbone=GoogLeNet2026.02 | 62.35 | — | — | |
| IPGuardBackbone=VGG2026.02 | 61.11 | — | — | |
| BackdooringBackbone=ResNet2026.02 | 58.64 | — | — | |
| UAPBackbone=ResNet2026.02 | 52.47 | — | — | |
| BackdooringBackbone=VGG2026.02 | 51.85 | — | — | |
| EWEBackbone=ResNet2026.02 | 51.24 | — | — | |
| IPGuardBackbone=GoogLeNet2026.02 | 50.62 | — | — | |
| Random2025.03 | 50 | 50 | 28.6 | |
| EWEBackbone=DenseNet2026.02 | 48.77 | — | — | |
| IPGuardBackbone=ResNet2026.02 | 40.74 | — | — |