Backdoor Federated Learning

This repository hosts the project on the backdoor attack in a Federated learning setting using the FATE framework. This project has been developed during the A.Y. 2021-2022 for the Advanced Topics in Security and Privacy course @ University of Groningen.

The details about our work and the obtained results can be found inside the report.

Installation

First things first, clone this repository and then move into it.

git clone https://github.com/andrea-gasparini/backdoor-federated-learning.git && \
cd backdoor-federated-learning

In order to slightly reduce the possibility of encountering problems we are going to install FATE using Docker. So be sure to have Docker installed and to run the following docker commands as a root user (or to follow the Manage Docker as a non-root user guide).

We suggest to use FATE version 1.6.1, for which we provide the following mirror download.

wget http://gasparini.cloud/FATE/docker_standalone_fate_1.6.1.tar.gz

Other versions can be downloaded from the original repository replacing ${version} below with the real version you want to use. Note: This option could take a lot of time (download is from China), so this is not recommended.

wget https://webank-ai-1251170195.cos.ap-guangzhou.myqcloud.com/docker_standalone_fate_${version}.tar.gz

Once we have the docker archive file we can proceed with the actual installation of the docker image.

tar -xzvf docker_standalone_fate_1.6.1.tar.gz
docker load < docker_standalone_fate_1.6.1/fate.tar

And run a new docker container with the FATE image. The current directory is going to be binded to /fate/backdoor-attack inside the container, so be sure that $(pwd) corresponds to the root of our repository.

docker run -d --name fate -p 8080:8080 \
--mount type=bind,source="$(pwd),target=/fate/backdoor-attack" fate:1.6.1

Now create a new shell to interact with the Docker container and the FATE environment.

CONTAINER_ID=`docker ps -aqf "name=fate"`
docker exec -t -i ${CONTAINER_ID} bash

To conveniently interact with FATE, it is recommended to also install FATE-Client and FATE-Test inside the container (through the new shell created in the previous code block).

pip install fate-client fate-test

And, in order to setup Pipeline FATE jobs, initialize FATE-Flow CLI providing server ip/port information of deployed FATE-Flow.

flow init --ip 127.0.0.1 --port 9380
pipeline init --ip 127.0.0.1 --port 9380

Usage

The following usage examples are supposed to be executed from the docker container shell, inside the /fate/backdoor-attack directory.

In order to run a single experiment you can execute the run_experiment.py script, specifying the poison percentage with the --poison-percentage option.

python src/run_experiment.py --poison-percentage 0.5

To run multiple experiments and reproduce the results shown in the report you can simply execute the run_experiment.py script without any argument.

python src/run_experiment.py

Authors

• Andrea Gasparini
• Arjan Tilstra
• Gerrit Luimstra