You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Yes, the pointwise attack is quite slow ~30 min / adversarial on a Titan X.
I implemented some speedups: switched to an attack which requires less iterations and changed the default model to an approximation of the ABS model (n_iter=10 instead of n_iter=50). It should now take ~1-5 min to create an adversarials on a GPU with the provided code.
I'm trying to run the demo in attacks.ipynb, but it is really slow, taking over 30 minutes to run the PointwiseAttack.
I'm using GPU and running the classification was quite fast, but I'm wondering if I may be doing something wrong, or is it normally like this?
Thanks
The text was updated successfully, but these errors were encountered: