AFLChain: Blockchain-enabled Asynchronous Federated Learning in Edge Computing Network

Abstract

Edge computing network (ECN), which could process learning tasks at the edge, is considered as a potential solution to release the burden of the cloud. Meanwhile, to protect user privacy, federated learning (FL) is used in the ECN to establish models by multi-party collaborative learning on numbers of edge nodes (ENs). However, due to the frequent data interaction between the cloud server and distributed ENs, the reliability of data transmission and the privacy protection capability of the network cannot be guaranteed. In this paper, a distributed ECN is considered, to improve the learning efficiency in the multi-party FL while ensuring the reliability of ENs, a consortium blockchain enabled asynchronous federated learning (AFLChain) algorithm is proposed, which could dynamically allocate the learning tasks to ENs according to their computing capabilities. Moreover, an entropy weight-based reputation mechanism is introduced for the EN evaluation to further improve the performance of the AFLChain. Finally, the simulation results demonstrate the effectiveness of the proposed algorithms.