Secure Federated Learning in Distributed Environments: A Blockchain-Based Approach

We present a novel blockchain-based mechanism for secure and private model updates in a fully distributed system. This mechanism follows Zero Trust principles to allow confidentiality and privacy protection of the shared model updates. Thus, allowing entities to train model on potentially confidential information by allowing plausible deniability, as the aggregation processes are performed in an oblivious manner from the model user point of view. This method is currently being adopted as the preferred mechanism for model sharing within the FLUIDOS MetaOS.