search

AlanPeng0897 / Defend_MI

[KDD 2022] "Bilateral Dependency Optimization: Defending Against Model-inversion Attacks"
https://arxiv.org/pdf/2206.05483.pdf
MIT License
22 stars 5 forks source link

mnist result #2

Open Jasper-Bai opened 2 years ago

Jasper-Bai commented 2 years ago

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?

hung-z commented 1 year ago

The same comment!

AlanPeng0897 commented 1 year ago

In the setting, the MNIST private dataset only contains 5 classes, so the variance of the attack accuracy tends to be large, my suggestion is to use CelebA for evaluation.