RingFFL: A Ring-Architecture-Based Fair Federated Learning Framework

In the ring-architecture-based federated learning framework, security and fairness are severely compromised when dishonest clients abort training process after obtaining useful information. To solve problem, we propose a Ring- architecture-based Fair Federated Learning framework called RingFFL, in which design penalty mechanism for FL. Before starts each round, all that will participate pay deposits set order record transactions on blockchain to ensure they not tampered with. Subsequently, perform FL process, correctness of models transmitted by is guaranteed HASH algorithm during process. When honestly, client can obtain final model, number digital currencies client’s wallet kept constant; otherwise, who leave halfway be compensated honestly this way, through mechanism, either model or compensated, thus ensuring learning. The analysis experimental results show RingFFL only guarantees accuracy but also fairness.