Designing a Federated Learning Architecture for the Data Middle Platform
Rongrong Zhang, Zhiqiang Gao · 2024
Since entering the information age, data owners have been increasingly concerned about data privacy, which has led to isolated and dispersed storage of their data. In particular, it has become difficult to collect data from multiple data owners for model training. Therefore, how to break the “data silo” formed by isolated and scattered data and achieve model training on the basis of data privacy and security has become an urgent problem in the context of information technology era. In this paper, the Data middle platform architecture is redesigned to adapt to the federated learning system, and according to the characteristics of different types of federated learning, the new Data middle platform architecture is integrated into various federated learning processes to form a data centre-oriented federated learning architecture. We also design a backup system and scheme for the possible problems of the new architecture to ensure the normal operation of the whole architecture system. Then, based on the data centre-oriented federal learning architecture, we design a “cloud-edge-end” data sharing and flow scheduling application scenario covering the whole domain. It provides a new and more reliable solution to solve the problems of data silos and data privacy and security.