Towards Real World Federated Learning

Machine Learning and Deep Learning 2023

Politecnico di Torino

Starting code for the Federated Learning project. Some functions are explicitly left blank for students to fill in.

Setup

Environment

If not working on CoLab, install environment with conda (preferred):

conda env create -f mldl23fl.yml

Datasets

The repository supports experiments on the following datasets:

1. FEMNIST (Federated Extended MNIST) from LEAF benchmark [1]
   • Task: image classification on 62 classes
   • 3,500 users
   • Instructions for download and preprocessing in data/femnist/
2. Reduced Federated IDDA from FedDrive [2]
   • Task: semantic segmentation for autonomous driving
   • 24 users

How to run

The main.py orchestrates training. All arguments need to be specified through the args parameter (options can be found in utils/args.py). Example of FedAvg experiment (NB training hyperparameters need to explicitly specified by the students):

• FEMNIST (Image Classification)

python main.py --dataset femnist --model resnet18 --num_rounds 1000 --num_epochs 5 --clients_per_round 10 

• IDDA (Semantic Segmentation)

python main.py --dataset idda --model deeplabv3_mobilenetv2 --num_rounds 200 --num_epochs 2 --clients_per_round 8 

References

[1] Caldas, Sebastian, et al. "Leaf: A benchmark for federated settings." Workshop on Federated Learning for Data Privacy and Confidentiality (2019).

[2] Fantauzzo, Lidia, et al. "FedDrive: generalizing federated learning to semantic segmentation in autonomous driving." 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2022.