Auxiliary Offloading Optimization Method for Hierarchical Federated Learning

• Published in: Proceedings of the 12th International Conference on Computer Engineering and Networks pp. 146 - 155
• Main Authors: Qiu, Yingzhao, Guo, Shaoyong, Shen, Tao, Feng, Yan, Bai, Fenhua
• Format: Book Chapter
• Language: English
• Published: Singapore Springer Nature Singapore
• Series: Lecture Notes in Electrical Engineering
• Subjects: Computation offloading; Delay; Edge computing; Federated learning; Materials genetic data engineering
• ISBN: 9811969000; 9789811969003
• ISSN: 1876-1100; 1876-1119
• DOI: 10.1007/978-981-19-6901-0_16

Materials genetic data engineering is booming, and with it comes the problem of data islands and other issues faced by the materials genetic database. While federated learning solves the problem of material genetic data engineering, it also has the problems of low model accuracy and high delay. In this paper, an auxiliary offloading optimization method for hierarchical federated learning is designed. Under the hierarchical federated learning system model based on cloud-edge-end, the data offloading algorithm under unbalanced data balances the data volume of each client, reduces the local gradient difference and the model accuracy of hierarchical federated learning is improved; under the condition of limited resources, a low-delay-based hierarchical federated learning offloading strategy is designed to computing offloading, which reduces the delay of hierarchical federated learning in the training and updating process. Through experiment analysis, the auxiliary offload optimization method of hierarchical federated learning proposed in this paper has higher model accuracy and lower delay than traditional federated learning methods.