A Novel Approach for Fair Incentive Social Physical Data Based on Blockchain-Federated Learning

Jun Jiat Tiang¹Hung Tran-Huy²Trang Hoang-Thu²Iyas Qaddara³Bong Jun Choi^4*Asma Hasan Alsheri⁵Hui Liu⁶

• ¹Multimedia University, Malacca, Malaysia
• ²Phenikaa University, Hanoi, Vietnam
• ³Al-Ahliyya Amman University, Amman, Jordan
• ⁴Soongsil University, Seoul, Republic of Korea
• ⁵Prince Sattam bin Abdulaziz University, Al Kharj, Saudi Arabia
• ⁶Universitat Bremen, Bremen, Germany

Abstract: A key research focus in FL is the incentive mechanism. To ensure that all data owners actively contribute their data for model training, it is necessary to establish a fair incentive system that encourages them to share useful data. A well-functioning incentive system enables all participants to continuously and effectively train models, which in turn enhances the accuracy of the ultimately trained federated model. This paper proposes a new algorithm for optimizing the incentive mechanism. Initially, clients who possess high-quality data can participate in the training due to their reputation value. The client entrusted local data training to the high-performance fog node by auctioning local training tasks to it. The aim of this action was to improve the efficiency of local training and tackle the problem of differing performance levels among clients. Finally, the global gradient aggregation algorithm removes malicious clients from the local gradient. Results from the simulation demonstrate that the suggested algorithm outperforms current algorithms. Keywords: Social physical data; federated learning; blockchain; encryption algorithm; incentive mechanism.