About the convergence of adversarial training.

[anonymous530]

Hello. Thanks for your great work. May I ask, how to judge the convergence of adversarial training? Just according to the curve of loss?
As training going on, models become more robust. What if attack fails when training?