Advances, Systems and Applications
Table 2 The classification accuracy for clean images and adversarial examples generated by black-box attack methods on five pretrained models. The black-box attack are based on MI-FGSM
From: A secure data interaction method based on edge computing
Attack | Inc-v3 | VGG16 | Res50 | Inc-v4 | IncRes-v2 |
|---|---|---|---|---|---|
Clean | 98.9% | 90.8% | 95.6% | 99.7% | 99.0% |
SIM | 44.0% | 21.9% | 33.9% | 46.6% | 51.4% |
Admix | 38.0% | 16.3% | 29.7% | 42.5% | 48.4% |