Ontology Design of Distributed Computing System Based on Hierarchical Federated Learning
Zhengyi Zhong, Ji Wang, Yang Zhang, Gaoyu He, Xueyi Zhang, Hui Yu Xiang · 2021
As a popular distributed learning framework, federated learning is widely used in the training process of distributed models. And as the intelligent core of the distributed system, the intelligent inference model of each computing node must have the ability to continuously evolve with the real scenario. Under the traditional federated learning framework, whether the model participates in the update process depends on whether it is selected by the server, which is contrary to reality. Therefore, focusing on real-world needs, this paper constructs a distributed intelligent computing ontology based on hierarchical federated learning, which regarding each computing device in hierarchically distributed systems as an ontology node, and on this basis builds a distributed computing node resource model. In view of the heterogeneous resources of each node, the ontology can adaptively choose whether to participate in the training process of federated learning according to the resource state of the node itself, which improves the model update and resource utilization efficiency.