How to test madrylab's defense methods?

[SoEzreal]

https://github.com/MadryLab/mnist_challenge

Hello, I am trying to attack MadryLab's defense strategy with your project （advGAN）. However, it did not get the 92.76% effect mentioned in the article （Generating Adversarial Examples with Adversarial Networks）. Since I just started this research work, I don't know whether it is the problem of parameter setting or the wrong operations.

[mathcbc]

It is not the official implementation and I didn't test the code in the Madrylab's defense strategy. On 11/26/2019 15:08, SoEzreal wrote: https://github.com/MadryLab/mnist_challenge Hello, I am trying to attack MadryLab's defense strategy with your project （advGAN）. However, it did not get the 92.76% effect mentioned in the article （Generating Adversarial Examples with Adversarial Networks）. Since I just started this research work, I don't know whether it is the problem of parameter setting or the wrong operations. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.