Model Stealing Attack on CIFAR10
96.7AgreementCont-Steal
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Cont-StealTarget Model=ViT2022.01 | 96.7 | — | |
| Cont-StealTarget Model=CLIP2022.01 | 88.9 | — | |
| Cont-StealTarget Encoder training set=CIFAR10, Downstream Classifier training set=CIFAR102022.01 | 84.5 | 85.4 | |
| INFONCETarget Encoder training set=CIFAR10, Downstream Classifier training set=CIFAR102022.01 | 82.6 | 82.8 | |
| ConventionalTarget Model=CLIP2022.01 | 81.5 | — | |
| StolenEncoderTarget Encoder training set=CIFAR10, Downstream Classifier training set=CIFAR102022.01 | 81.1 | 80.8 | |
| BaselineTarget Encoder training set=CIFAR10, Downstream Classifier training set=CIFAR102022.01 | 78.5 | 79 | |
| ConventionalTarget Model=ViT2022.01 | 74.5 | — | |
| Cont-StealTarget Model=MAE2022.01 | 71.2 | — | |
| ConventionalTarget Model=MAE2022.01 | 55.5 | — | |
| KL DivergenceTarget Encoder training set=CIFAR10, Downstream Classifier training set=CIFAR102022.01 | 21.3 | 20.3 |