Implementation Example of Rate Adaptation Algorithm for Decentralized Federated Learning

• This code demonstrates the rate-adaptation algorithm proposed in the following paper:
  • Koya Sato and Daisuke Sugimura, "Rate-Adapted Decentralized Learning Over Wireless Networks," IEEE Trans. Cogn. Commun. Netw., vol.7, no.4, pp.1412-1429, Dec. 2021.
  • https://ieeexplore.ieee.org/document/9410554
• You can evaluate the effect of $\lambda_\mathrm{target}$ on the average transmission rate (related to Fig.11 in the above paper)
• Please adjust simulation parameters based on the desired condition (e.g., this code sets n=4; however, Fig.11 assumes n=6)
• This code can work with the following command:

$python main.py

Note

This code does not contain the learning part. Please implement it with your favorite ML framework (e.g., PyTorch) if necessary.

License

The MIT License (MIT)

Copyright (c) 2022 Koya SATO.