problems in backdoor_isolation.py

Question

problems in backdoor_isolation.py

Closed this issue 3 years ago · 3 comments

GuanZihan commented 3 years ago

#237 line in backdoor_isolation.py

I thought the parameter here should be "opt" rather than "opt.lr" ?

Answer 1 · 2021-12-13T02:55:23.000Z

Hi, could you further explain why and what’s the difference between them?

Answer 2 · 2022-02-23T05:36:05.000Z

Hi, I also have this problem, change the opt.lr to opt in backdoor_isolation.py. And if you want to run the backdoor_unlearning.py, you should add a parser for check_point in config.py.

How to get the 10 backdoor attacks code in this paper? I find the create_poisoned_data.py is empty.

Answer 3 · 2022-02-23T06:50:46.000Z

Hi, thank you for the kind reminder for the error of opt. The implementation for most of the input-space attacks such as BadNets, Trojan, SIG, etc. has already been contained in our data_loader.py. As described in the readme, not all of the attacks can include in the data_loader.py easily, so for the other attacks (e.g. Dynamic, FC) we left the reference code of generating the backdoored model, which is implemented similar to our work.