Fix a bug in universal attack.

DSE-MSU / DeepRobust

A pytorch adversarial library for attack and defense methods on images and graphs

MIT License

957 stars 191 forks source link

Closed HelloWorldLTY closed 3 months ago

HelloWorldLTY commented 4 months ago

I found a bug in the original implementation of universal attack, that is, a undefined variable was referred before.

Moreover, I also found that torch.autograd dropped the function of zero_gradients, so I also replace it with a new function.

Could you please review it and give some feedback? Thanks.

HelloWorldLTY commented 3 months ago

I see, thanks.