Research Article
The Defense of Adversarial Example with Conditional Generative Adversarial Networks
Table 6
Comparisons with other defense methods.
| Attack | Target model | No attack | No defense | Defense-GAN | MagNet | Adv.Tr | Our method |
| FGSM | A | 0.997 | 0.217 | 0.988 | 0.191 | 0.651 | 0.986 | B | 0.962 | 0.022 | 0.956 | 0.082 | 0.060 | 0.958 | C | 0.996 | 0.331 | 0.989 | 0.163 | 0.786 | 0.987 | D | 0.992 | 0.038 | 0.980 | 0.094 | 0.732 | 0.983 |
| CW | A | 0.997 | 0.141 | 0.989 | 0.038 | 0.077 | 0.965 | B | 0.962 | 0.032 | 0.916 | 0.034 | 0.280 | 0.924 | C | 0.996 | 0.126 | 0.989 | 0.025 | 0.031 | 0.968 | D | 0.992 | 0.032 | 0.983 | 0.021 | 0.010 | 0.966 |
|
|