Secure and Efficient Federated Learning Through Layering and Sharding Blockchain

Federated Learning (FL) has become an essential enabling technology for digital twin in Industrial Internet of Things (IIoT) networks. However, due to the master/slave structure of FL, it is very challenging to resist the single point of failure of the master aggregator and attacks from malicious IIoT devices while guaranteeing model convergence speed and accuracy. Recently, blockchain has been brought into FL systems transforming the paradigm to a decentralized manner thus further improving the system security and learning reliability. Unfortunately, the traditional consensus mechanism and architecture of blockchain systems can hardly handle the large-scale FL task and run on IIoT devices due to the huge resource consumption, limited transaction throughput, and high communication complexity. To address these issues, this paper proposes a two-layer blockchain-driven FL system, called ChainFL, which splits IIoT network into multiple shards as the subchain layer to limit the scale of information exchange, and adopts a Direct Acyclic Graph (DAG)- based mainchain as the mainchain layer to achieve parallel and asynchronous cross-shard validation. Furthermore, FL procedure is customized to deeply integrate with blockchain, and the modified DAG consensus mechanism is proposed to mitigate distortion caused by abnormal models. To provide a proof-of-concept implementation and evaluation, multiple subchains based on Hyperledger Fabric and the self-developed DAG-based mainchain are deployed. The extensive experimental results demonstrated that our proposed ChainFL system outperforms the existing main FL systems in terms of acceptable and fast training efficiency (by up to 14%) and stronger robustness (by up to 3 times).