gongzhimin / Membership-Inference-Attack-in-Federated-Learning

Membership inference against Federated learning.
Apache License 2.0
8 stars 4 forks source link

Is your attack method a new one or a reproduction of Song's paper? #3

Open 2016312357 opened 2 years ago

2016312357 commented 2 years ago

Hello, I've been interested in membership inference attacks against FL. And I found your code on Github. I'd appreciate that if you could give a more detailed description of your attack method, since the code is a little hard to follow.

gongzhimin commented 2 years ago

Hello, @2016312357. Thank you for your attention! The membership inference attack in this repo is from Shokri's paper [1]. And here the code, which is similar to Song's implementation [2]. I'm sorry there are still some hidden faults in the master branch as I want to refactor Shokri's code. But the branch old-version is okay.

[1] Comprehensive Privacy Analysis of Deep Learning: Passive and Active White-box Inference Attacks against Centralized and Federated Learning (https://arxiv.org/abs/1812.00910). [2] Membership Inference Attack against Machine Learning Models (http://ieeexplore.ieee.org/document/7958568/).

macknight commented 1 year ago

Hi, Is the master branch runnable now?

BR

gongzhimin commented 1 year ago

Hallo, @macknight. The master branch is runnable, but the results are uncorrected yet. >~< Maybe because I defined the structures of models unsuitably. Sorry, I'm writing a paper currently, and unavailable to fix these long-ago faults in this branch. But the old-version branch is ok.

Sincerely

macknight commented 1 year ago

Thanks