A Federated Learning Method Based on Blockchain and Cluster Training

Federated learning (FL) is an emerging machine method in which all participants can collaboratively train a model without sharing their raw data, thereby breaking down data silos and avoiding privacy issues caused by centralized storage. In practical applications, client are non-independent identically distributed, resulting FL requiring multiple rounds of communication to converge, entails high costs. Moreover, the architecture traditional remains susceptible breaches, network congestion, single-point failures. order solve these problems, this paper proposes framework based on blockchain technology cluster training algorithm, called BCFL. We first improved algorithm odd–even round training, accelerates convergence dividing clients into clusters adopting serialized within each cluster. Meanwhile, compression operations were applied parameters before transmission reduce costs improve efficiency. Then, decentralized was designed developed Inter-Planetary File System (IPFS), where records process IPFS optimizes storage associated with blockchain. The experimental results demonstrate superiority terms accuracy