Our Dapp, FedEthML, shows how the blockchain technology can be adopted to address two of the main issues of Federated Learning. The first is centralization: the central server is replaced by the blockchain, and any adversarial behavior of the central node is therefore prevented. The second issue is represented by the lazy/adversarial behavior of training nodes. This problem is (partially) solved by introducing an incentive mechanism based on the proof-of-stake upon which the Ethereum environment relies
Different workers with different datasets will participate in the training, which will be coordinated by a smart contract deployed on an Ethereum blockchain.
For further details check the Report.
USAGE:
'cd' into the root folder.
If it's the first time you open this project, run: 'npm install' to install all the dependencies.
'npx hardhat node' to spin up a local developement blockchain.
In another terminal:
npx hardhat compileto compile the contract inside the contracts folder.
npx hardhat run --network hardhat ./scripts/deploy.jsto deploy the smart contract.
python ./scripts/python/number_oracle.pystarts and oracle sending random numbers to the smart contract (they are needed for some stuff).
In another terminal:
npm run devstarts the web app.
IMPORTANT: This project uses IPFS from Pinata. To correctly create and visualize the tasks you need to correctly place your Pinata credentials in the ipfsFunctions.jsx file.
