Federated learning using a mixture of experts

This repo contains code for the paper Federated learning using a mixture of experts.

Example

To run the code on cifar100 with 50 clients, run the following line. Results will be saved in /save/results_cifar10.csv.

python main_fed.py --model 'cnn' --dataset 'cifar100' --n_data 100 --num_clients 50 --num_classes 100 --epochs 1500 --local_ep 3 --opt 0 --p 1.0 --gpu 0 --runs 1 --filename results_cifar10.csv

To see all option, run

python main_fed.py -h

If you want to allow for overlap between class labels, pass the argument --overlap.

You can also specify a configuration file and iterate over a parameter by running

python iterator.py --filename config_femnist.json

Note that for the FEMNIST dataset, you will have to generate the data by running the preprocessing script in the leaf/data/femnist subfolder. See FEMNIST dataset for more information on this script.

Results

Results for running above example, for p in [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]. Average accuracy over three runs per value of majority class fraction.

Docker

To run this code in Docker, use the following.

In subdirectory docker run make debug or in the root directory docker run -it --rm -v pwd:/home/user/src martisak/fl-moe:latest bash

You can build this image in the docker directory with make build.

Cite

If you find this work useful, please cite us using the following bibtex:

@article{listozec2020federated,
  title={Federated learning using a mixture of experts},
  author={Listo Zec, Edvin and Mogren, Olof and Martinsson, John and S{\"u}tfeld, Leon Ren{\'e} and Gillblad, Daniel},
  journal={arXiv preprint arXiv:2010.02056},
  year={2020}
}

Acknowledgements

The code developed in this repo was was adapted from https://github.com/shaoxiongji/federated-learning.

martisak / federated-learning-mixture