HamadJassem / flowersoup

Federated Learning Simulation

This project is a simulation of federated learning using Flower. In this project, a FedSoup adaptation for flower is trialed and compared with FedAvg. The model leveraged in this simulation is ResNet18. The adaptations are contained in the client.py file.

Requirements

• Python 3.7 or higher
• PyTorch
• Flower
• Numpy

Usage

To run the simulation, use the following command:

python main.py --num_clients 5 --total_rounds 40 --alpha 0.75

Client Changes

The following lines specify the position of the code we provide to adapt FedAvg to FedSoup

client.py

fit method:

1. Lines 65-90: obtaining parameters from server
2. Line 92: Local training
3. Lines 93-155 FedSoup Extension

get_parameters method:

1. returns client parameters

evaluate method:

1. Sets parameters with global parameters
2. evaluates model on local test set
3. returns results