A Blockchained Incentive Architecture for Federated Learning

The naive use of Federated Learning (FL) in a distributed environment exposes it to a risk of corruption, whether intentional or not, during the training phase. It happens because of the lack of monitoring of the training increments and difficulty of checking the quality of the training datasets. A very common type of attack of this type is Model Poisoning. To improve the security of the FL structure, we propose a decentralized FL framework based on blockchain, that is, a blockchain-based FL framework to increment the system security using an incentive mechanism to reward good trainers in the form of tokens. The system modeling will be presented as well as its implementation in the Mininet simulator. The validation tests performed to attest its accuracy were executed using the MNIST dataset.