LipingYi / QSFL

QSFL: A Two-Level Uplink Communication Optimization Framework for Federated Learning (ICML'22)

QSFL: A Two-Level Uplink Communication Optimization Framework for Federated Learning

This is an official implementation of QSFL: A Two-Level Uplink Communication Optimization Framework for Federated Learning paper.

Installation

• Create a virtual environment with virtualenv
• Clone the repo
• Run: cd <PATH_TO_THE_CLONED_REPO>
• Run: pip install -r requirements.txt to install necessary packages.

---

Reproduce Paper Results

• Server executes script: run_server.sh
• Clients execute script: run.sh
• Stop all threads and end FL workflow: stop.sh

---

Citation

If you find QSFL to be useful in your own research, please consider citing the following bibtex:

@inproceedings{QSFL,
  author    = {Liping Yi and
               Wang Gang and
               Xiaoguang Liu},
  title     = {{QSFL:} {A} Two-Level Uplink Communication Optimization Framework
               for Federated Learning},
  booktitle = {International Conference on Machine Learning, {ICML} 2022, 17-23 July
               2022, Baltimore, Maryland, {USA}},
  series    = {Proceedings of Machine Learning Research},
  volume    = {162},
  pages     = {25501--25513},
  publisher = {{PMLR}},
  year      = {2022},
}