alanpeng0897 / defend_mi Goto Github PK

Can you explain what the reg and vib in the defends parameter means, thank you!

Nevermind, I found them :)

Can you provide no defense and MID model training code also to compare?
Thank you.

Hey,

thank you for providing code for your defense. I am currently trying to evaluate the models with VMI. However, the code seems currently missing the args.json file loaded here:

Without the config, the code seems not to work properly. Could you please provide the missing config file?

Best,
Lukas

Hi,

Thank you for sharing amazing work.

I am currently working on your code and curious about the difference between models training with BiDO (lambda_x=0 and lambda_y=0) and No. Def model.

Based on my understanding, it should be similar? In your code, I noticed that the difference comes from hyper-parameters (learning rate, learning rate scheduler, etc)?

Thank you very much.

Why in the file: https://github.com/AlanPeng0897/Defend_MI/blob/master/GMI/attack.py
line 169, you use g_path? should it be d_path?

Thanks.

I could not get the adorable DMI attack accuracy and recovery performance on MNIST. I used the same running setting to train the targetor model and launch the DMI attack. could you tell me some possible points causing the bad performance?