This code includes 2 algorithms —— SPFL and PLGP, which are Synchronous Federated Learning and Asynchronous Federated Learning, we do personalization FL on both of them.

Environment

• python 3.7

install packages

requirements.txt is basically the exact environment of mine. Let's create a conda environment and replicate the envorinment as follows.

conda create -n Fede python==3.7
conda activate Fede
pip install -r requirements.txt

DataSet

For both of the two experiments, We use four datasets: EMNIST，MNIST，CIFAR10，CIFAR100. The split methods of datasets is in " PLGA/SPFL/dataset/ ".

SPFL

This is a code for Synchronous Federated Learning, and we add personalization for the FL. We named this algorithm SPFL.

Running the SPFL experiments

sh run_all.sh

It will run all the experiments (FedAvg(Async), FedAsync, PerFedAvg, LGA, PLGA ).

PLGP

This is a code for Asynchronous Federated Learning, and we also add personalization for the FL. We named this algorithm PLGA.

Running the PLGA experiments

sh run.sh

It will run all the experiments (FedAvg, FedUpdate, PerFedAvg, SPFL-w(1-step), SPFL-w, SPFL(1-step), SPFL ).