Federated2Fog

Please note that the code here is corresponding to the equations in v4 of paper on ArXiv. for versions > v4 of the the paper, we have updated equations which tighten the bounds of corresponding equations in v4. Similar plots can be obtained using the updated equations by adjusting the value of $\chi$ as mentioned on page 12 of updated manuscripts.

• ArXiv: Multi-Stage Hybrid Federated Learning over Large-Scale Wireless Fog Networks
• Github: Federated2Fog
• Data and Code: Drive Link

Details

• alpha in the code represents $\sigma$'s in the equations
• omega in the code represents $M$ in the equations
• learning rate (lr) and $\eta$ are two different parameters

Hyperparameters

• Hyperparamters are stored in separate files
  • hyperparameters.txt for 125 node experiments
  • hyperparameters_625.txt for 625 node experiments
• These need to be replaced in arguments.py __init__ function call to regenerate the plots

Plots

• The list of selected plots is summarized in plots.txt
• Download the data from the drive link provided to access the generated plots.

Misc.

Also available in the drive link (above) provided:

• Pretrained models
• Training histories
• Training logs
• Preprocessed datasets
• etc.

Citation

If you find Federated2Fog useful, please cite the following paper

@article{hosseinalipour2020multi,
  title={Multi-Stage Hybrid Federated Learning over Large-Scale D2D-Enabled Fog Networks},
  author={Hosseinalipour, S and Azam, SS and Brinton, CG and Michelusi, N and Aggarwal, V and Love, DJ and Dai, H},
  journal={arXiv preprint arXiv:2007.09511},
  year={2020}
}