Hello! I have questions on your paper #4
Description
I read your paper with interest! Currently, I have a plan to develop your research.
I have some question in your paper. Please answering below three questions
(1) In paper, I confirm that you evaluated 100 PGD attack on test data. Then, setting test_numsteps=100 when evaluating is same with this?
(2) In finetuning.py, I found attack_CW (not ```attach_pgd``). Can you explain what this attack means? and What is paper result correspoding this attack?
(3) Do you have full result l2-norm bound pgd attack? In your paper, l_infinity-norm bound is used.
I am curious about l2-norm result!