Skip to content

Accelerating the evaluation of sign grad. #9

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Lodour wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Accelerating the evaluation of sign grad. #9

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 24 additions & 3 deletions attack/Sign_OPT.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@ def attack_untargeted(self, x0, y0, alpha = 0.2, beta = 0.001, iterations = 1000
if svm == True:
sign_gradient, grad_queries = self.sign_grad_svm(x0, y0, xg, initial_lbd=gg, h=beta)
else:
sign_gradient, grad_queries = self.sign_grad_v1(x0, y0, xg, initial_lbd=gg, h=beta)
sign_gradient, grad_queries = self.sign_grad_v1_fast(x0, y0, xg, initial_lbd=gg, h=beta)

## Line search of the step size of gradient descent
ls_count = 0
Expand Down Expand Up @@ -241,8 +241,29 @@ def sign_grad_v1(self, x0, y0, theta, initial_lbd, h=0.001, D=4, target=None):
# sign_grad *= 0.5*delta

return sign_grad, queries




def sign_grad_v1_fast(self, x0, y0, theta, initial_lbd, h=0.001, D=4, target=None):
# sample K Gaussian noise
k = self.k
u = torch.randn(k, *theta.shape[1:]).cuda()
u /= u.view(k, -1).norm(dim=1)[:, None, None, None]

# compute K new theta's
new_theta = torch.tensor(theta, dtype=torch.float).cuda() + h * u
new_theta /= new_theta.view(k, -1).norm(dim=1)[:, None, None, None]

# predict
pred = self.model.predict_label(x0 + initial_lbd * new_theta)
pred = pred != y0 if target is None else pred == target

# estimate sign grad
sign = (1 - 2 * pred)[:, None, None, None] # (0, 1) -> (+1, -1)
sign_grad = torch.mean(u * sign, dim=0, keepdims=True)

return sign_grad.cpu().numpy(), k


##########################################################################################
def sign_grad_v2(self, x0, y0, theta, initial_lbd, h=0.001, K=200):
"""
Expand Down