SIAs: Beyond MIAs in Federated Learning

This repository contains the code for the experiments of the paper "Source Inference Attacks: Beyond Membership Inference Attacks in Federated Learning"

Requirement

• torch==1.8.1
• numpy==1.18.1
• torchvision==0.9.1
• sklearn==0.22.1

Dataset

We evaluate SIAs on three typical federated learning frameworks, i.e., FedSGD, FedAvg, and FedMD.

We use Synthetic, MNIST, FEMNIST, CIFAR-10, CIFAR-100, and Purchase datasets for the experiments. In particular, we use Synthetic, MNIST, CIFAR-10, and Purchase for evaluating SIAs in FedSGD and FedAvg. We use MNIST, FEMNIST, CIFAR-10, and CIFAR-100 for evaluating SIAs in FedMD.

• For MNIST, CIFAR-10, and CIFAR-100 datasets, you can directly run the code and the dataset will be downloaded automatically.
• For Purchase-100, please first dowanlad it from here, and then put it in the "data" subfolder of the "FedSGD-SIAs" and the "FedAvg-SIAs" folder
• For Synthetic, FEMNIST datasets, we have uploaded them to the repository.

Experiment on FedSGD

You can try different --alpha (non-IID data distribution) to evaluate how non-IID distribution affects the attack performance. For Synthetic and Purchase datasets, we set --model=mlp. For MNIST and CIFAR-10 dataset, we set --model=cnn.

python main_fed.py --dataset=Synthetic --model=mlp --alpha=1 

Experiment on FedAvg

You can try different --alpha (non-IID data distribution), --local_ep (number of local epochs) to see how the attack performance changes. For Synthetic and Purchase datasets, we set --model=mlp. For MNIST and CIFAR-10 dataset, we set --model=cnn.

python main_fed.py --dataset=Synthetic --model=mlp --alpha=1 --local_ep=5

Experiment on FedMD

You can change the paramters in the json file in the "conf" subfolder of the "FedMD-SIAs". Setting different values of alpha and N_private_training_round to change the non-IID data distribution and the local epochs.

python CIFAR.py