Attack on FE-powered Neural Networks

Running the attack on FE-powered NN

1. Generate first layer weights multiplied by the input samples using the command: python src/gen_wx.py

or with any other method using the NN to be attacked and save the results into data:

• a.data should be the result of WX
• w.data and x.data can be used for analyzing the result.

2. Run python src/linprog.py in order to operate the attack

Running the mitigation client and server

The server should be run before the client

• Server: python src/mitigation1_server.py

• Client python src/mitigation1_client.py

Running the samples

• install charm
• download MNIST/CIFAR10 Dataset