A Methodology for High-efficient Federated-learning with Consortium Blockchain

Abstract

This paper we propose a distributed computing architecture, the federated Learning based on the consortium blockchain. With decentralized distributed training, the federated learning modeling process is more robust, it can solve the current fairness and security problems of federated learning. We research the modeling efficiency problems in the Consortium Blockchain Federated Learning (CBFL) architecture at model training process, using model compression to improve the modeling efficiency, and we analyze Practical Byzantine Fault Tolerance consensus algorithm commonly used in the consortium chain, and propose the improvement of PBFT algorithm on the consensus efficiency and mechanism fault tolerance. The experiment results show that the improved CBFL has better practicability.