The easiest way to generate adversarial examples remains gradient-based attacks, by reading the first part we could think that GANs are almost the only way to go.
You are correct, it should have been more clear. We were trying to build on a previous issue of the newsletter that covered GANs. The text says one of the most common applications of GANs is for adversarial attacks but didn't try to imply that adversarial attacks can only be done with GANs.
The easiest way to generate adversarial examples remains gradient-based attacks, by reading the first part we could think that GANs are almost the only way to go.
You are correct, it should have been more clear. We were trying to build on a previous issue of the newsletter that covered GANs. The text says one of the most common applications of GANs is for adversarial attacks but didn't try to imply that adversarial attacks can only be done with GANs.