Skip to content

Fix gradient tracking order #208

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
cossu555 wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix gradient tracking order #208

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e719da6
Fix gradient tracker
cossu555 Jul 15, 2026
3215cf8
Fix gradient tracker
cossu555 Jul 15, 2026
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
5 changes: 3 additions & 2 deletions src/secmlt/adv/evasion/modular_attacks/modular_attack_fixed_eps.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,7 +90,6 @@ def _run_loop(
# (untargeted: +loss, targeted: -loss), so initialize at -inf.
best_losses = torch.zeros(samples.shape[0]).fill_(-torch.inf)
best_delta = torch.zeros_like(samples)
grad_before_processing = torch.zeros_like(delta)

for i in range(self.num_steps):
x_adv.data, delta.data = self.manipulation_function(
Expand All @@ -107,6 +106,9 @@ def _run_loop(
target=target,
multiplier=multiplier,
)

grad_before_processing = delta.grad.detach().clone()

# keep perturbation with best transformed objective value
best_delta.data = torch.where(
atleast_kd(losses.detach().cpu() > best_losses, len(samples.shape)),
Expand All @@ -130,7 +132,6 @@ def _run_loop(
grad_before_processing.detach().cpu().data,
)

grad_before_processing = delta.grad.data
delta.grad.data = self.gradient_processing(delta.grad.data)
optimizer.step()
scheduler.step()
Expand Down
Loading