Federated Learning: Federated Proximal Implementation and Experiments on different simulations

Project of Blockchain and cryptocurrencies, in this work it has been expanded the original project to the following link:https://github.com/AnaNSi-research/FederatedLearningBlockchain.

Additional features implemented:

• Implemented Federated Proximal [1]
• Insert the simulation of out of battery device (devices which do not send weights)
• Simulation of more devices (greater than 3)
• Adding a new dataset
• Make and analysis different experiments on both dataset

Setup

This setup is just for a simulation

Requirements

• Ganache
• IPFS
• Miniconda
  • eth-brownie
  • cuda
  • tensorflow
  • opencv-python
  • pandas
  • scikit-learn

base deactivate

conda deactivate

environment creation

conda create --name blockchain_project python=3.9

activation

conda activate blockchain_project

pip update

python -m install pip --upgrade pip

cuda installation

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0

tensorflow=2.10 installation

pip install "tensorflow<2.11"

opencv-python installation

pip install opencv-python

pandas installation

pip install pandas==1.5.3

eth-brownie installation

pip install eth-brownie

scikit-learn installation

pip install scikit-learn

ganache installation

https://trufflesuite.com/ganache/

ipfs installation

https://github.com/ipfs/ipfs-desktop/releases

add network brownie

brownie networks add Ethereum fl-local host=http://127.0.0.1:7545 chainid=5777 timeout=3600

check network

brownie networks list

Running

This is just a simulation. For concurruncy problems on training on the same GPU, the collaborator.py script contains a loop that trains the different hospital model instances one at time in sequence. In a real time scenario, with more than one peer, it is possible to run the different learnings at the same time and it works in the same way.

setup first time

It is possible to choose the dataset, inserting the parameter, "brain_tumor" it will be used the brain tumor dataset, if the dataset is not sepcify it will be used the Alzheimer dataset

For Brain Tumor:

brownie run .\scripts\setup.py main brain_tumor --network fl-local

For Alzheimer

brownie run .\scripts\setup.py main --network fl-local

setup after first time

brownie run .\scripts\setup.py --network fl-local

The number of devices is choosen by the constants

run collaborator

brownie run .\scripts\collaborator.py --network fl-local

It is possible to insert the parameter "out" to randomly select the option "devices out of battery". The number of devices out of battery can be selected by the constants, instead the device out of battery and the round which they do not send the weights is randomly select.

brownie run .\scripts\collaborator.py out --network fl-local

Notes: In this configuration you need to wait 3600 s to validate if a device send the weights or not, it possible to change the time to wait, changing the constants TIMEOUT_SECONDS and TIMEOUT_DEVICES for simulation purpose.

run federated_learning

another shell

brownie run .\scripts\manager.py --network fl-local

It is possible to use the parameter FedProx to specify the using of Federated Prox technique.

brownie run .\scripts\manager.py FedProx --network fl-local

Experiments Analysis

In the repo, the notebook experiments_analysis contains all the analysis/plot of the different simulations.

Authors

• Lorenzo Cassano
• Jacopo D'Abramo

References

[1]: Li, Tian, et al. "Federated optimization in heterogeneous networks." Proceedings of Machine learning and systems 2 (2020): 429-450.