MOHAWK: Mobility and Heterogeneity-Aware Dynamic Community Selection for Hierarchical Federated Learning (IoTDI 2023)

Allen-Jasmin Farcas, Myungjin Lee, Ramana Rao Kompella, Hugo Latapie, Gustavo De Veciana, Radu Marculescu

Contact: [email protected]

1. Prepare environment

conda create -n mohawk python==3.10
conda activate mohawk
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install paramiko scp tqdm pandas

2. Prepare mobility data

Download the mobility dataset from here. Unzip the downloaded file under the mobility_data folder such that you have the following structure:

mobility_data
--- data_prepare.py
--- mobility_utils.py
--- RVF_ATX_PID_HZ-2020-05.tsv
--- RVF_ATX_PID_HZ_Places_Lookup.tsv

Then, execute the following:

cd mobility_data
python data_prepare.py

3. Prepare datasets

Run python generate_dataset.py to create the local datasets for all users.

4. Change paths

In utils.py the function get_hw_info contains paths for the files. Change them accordingly. The username and password only really matter if you run experiments on real devices. If you use only simulation you can leave any placeholder text there, but the path for the files needs to be completed.

You need to match the hw_type from get_hw_info(hw_type) with device_type from exp0.bash and if the added path is home/user/MOHAWK/files then in exp0.bash use cloud_path="files".

5. Run experiments

Edit the experiment configuration experiments/exp0.bash. Check the simulation.py for more details on the parameters used.

Run bash experiments/exp0.bash to start the experiment

Citation

@inproceedings{farcas2023mohawk,
  title={MOHAWK: Mobility and Heterogeneity-Aware Dynamic Community Selection for Hierarchical Federated Learning},
  author={Farcas, Allen-Jasmin and Lee, Myungjin and Kompella, Ramana Rao and Latapie, Hugo and De Veciana, Gustavo and Marculescu, Radu},
  booktitle={Proceedings of the 8th ACM/IEEE Conference on Internet of Things Design and Implementation},
  pages={249--261},
  year={2023}
}