An Efficient Consensus Algorithm for Blockchain-Based Federated Learning

Machine learning techniques have proved to be an inevitable tool for data analysis and prediction in various domains. Federated learning has enhanced it by allowing localized training of datasets where the owners need not share their data. The participating parties only need to share their locally trained model weights and biases to accomplish a federated learning task. This sharing mechanism can become significantly secure and tamper-proof on a blockchain network. In this paper, we have proposed a consensus algorithm to ensure efficient and economical utilization of blockchain to perform federated learning activities. The proposed model is tested in a simulated test bed to present a practical implementation of our algorithm and to reflect upon its efficiency.