A Novel Communication-Efficient and Blockchain-Secured Distributed Federated Learning

Machine learning is crucial in transforming large volumes of data into useful information and decisions. However, using traditional machine learning methods raises critical data confidentiality and security issues, such as data privacy and single point of failure. Hence, this paper adopts a secure average computation (SAC) peer-to-peer federated learning (P2-PFL) approach supported by a clustering strategy and blockchain technology for operation in a wireless environment. Specifically, the objective is to secure model update exchanges during the distributed P2P learning process, while reducing communication costs and providing full traceability of the FL process. Through experimentation, we evaluated the performance of our approach in terms of FL accuracy, precision, F1-score, FL round time, size of transiting data, and block addition time. The obtained results confirm the feasibility of blockchain-assisted SAC P2P-FL and their importance for future research in the distributed federated learning field.