Priv-FedTL

Code base for our implementation of Priv-FedTL. Follow the instructions below to setup the server and client devices for conducting the experiments.

Environment setup

On both types of devices execute the following commands to create and activate a virtual environment named 'testenv' (you can change this name to your preference):

python3 -m venv testenv 

source activate testenv/bin/activate

Also, install the MedMNIST package:

pip install medmnist

Server - Linux (Debian) :

pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu118

pip3 install h5py

pip3 install -U scikit-learn scipy matplotlib

pip3 install tqdm

pip install tenseal

Client - RaspbianOS : We used the Raspberry Pi 4 Model B as the client device. Follow the steps below to setup the Rasberry Pi:

1. Download custom OS image file and burn it in a 32 GB sd card
2. Uninstall Paddle-Lite if not needed as it takes a lot of storage:

du -s * | sort -nr | head -n10
sudo rm -rf ./Paddle-Lite && rm -rf ./PaddleOCR && rm -rf ./MNN

3. Fix the pytorch lib issue:

cd /usr/local/lib/python3.9/dist-packages/torch/
sudo chmod 777 ./storage.py
nano ./storage.py

Add this in the storage.py file

	if torch.cuda.is_available():
		return torch.load(io.BytesIO(b))
	else:
		return torch.load(io.BytesIO(b), map_location=torch.device('cpu'))

4. Build TenSEAL for Raspberry Pi:

git clone https://github.com/OpenMined/TenSEAL.git
pip install cmake

nano ~/.bashrc
export PATH="$HOME/.local/bin:$PATH"
source ~/.bashrc

cd TenSEAL
sudo python setup.py install

5. Clone this git repo

6. Install other dependencies:

pip3 install -U scikit-learn scipy matplotlib
pip install typing-extensions==4.3.0
pip3 install tqdm