Advances, Systems and Applications
Table 3 The experimental results about attacking pairwise combinations of five pretrained models. The attacks are based on MI-FGSM. The value of adversarial examples is represents the number of images that are able to mislead the two models at the same time. The last two columns in the table represents the amount of images which two models give the same predictions and different predictions
From: A secure data interaction method based on edge computing
Attack | Inc-v3 | VGG16 | Res50 | Inc-v4 | IncRes-v2 | Adversarial Examples | Same Predictions | Different Predictions |
|---|---|---|---|---|---|---|---|---|
SIM | \(\checkmark\) | \(\checkmark\) | Â | Â | Â | 516 | 131 | 385 |
\(\checkmark\) | Â | \(\checkmark\) | Â | Â | 490 | 148 | 342 | |
\(\checkmark\) | Â | Â | \(\checkmark\) | Â | 447 | 154 | 293 | |
\(\checkmark\) | Â | Â | Â | \(\checkmark\) | 413 | 144 | 269 | |
| Â | \(\checkmark\) | \(\checkmark\) | Â | Â | 620 | 183 | 437 | |
| Â | \(\checkmark\) | Â | \(\checkmark\) | Â | 493 | 120 | 373 | |
| Â | \(\checkmark\) | Â | Â | \(\checkmark\) | 457 | 110 | 347 | |
| Â | Â | \(\checkmark\) | \(\checkmark\) | Â | 465 | 135 | 330 | |
| Â | Â | \(\checkmark\) | Â | \(\checkmark\) | 434 | 132 | 302 | |
| Â | Â | Â | \(\checkmark\) | \(\checkmark\) | 411 | 148 | 263 | |
Admix | \(\checkmark\) | \(\checkmark\) | Â | Â | Â | 588 | 136 | 452 |
\(\checkmark\) | Â | \(\checkmark\) | Â | Â | 548 | 162 | 386 | |
\(\checkmark\) | Â | Â | \(\checkmark\) | Â | 492 | 150 | 342 | |
\(\checkmark\) | Â | Â | Â | \(\checkmark\) | 460 | 163 | 297 | |
| Â | \(\checkmark\) | \(\checkmark\) | Â | Â | 677 | 196 | 481 | |
| Â | \(\checkmark\) | Â | \(\checkmark\) | Â | 553 | 125 | 428 | |
| Â | \(\checkmark\) | Â | Â | \(\checkmark\) | 496 | 93 | 403 | |
| Â | Â | \(\checkmark\) | \(\checkmark\) | Â | 519 | 143 | 376 | |
| Â | Â | \(\checkmark\) | Â | \(\checkmark\) | 477 | 138 | 339 | |
| Â | Â | Â | \(\checkmark\) | \(\checkmark\) | 450 | 156 | 294 |