Model training node selection method for hierarchical federal edge learning

Abstract

The invention provides a hierarchical federal edge learning-oriented model training node selection method, which comprises the following steps that: a cloud server proposes conditions such as a reputation threshold value required by training to a reputation block chain, and the reputation block chain selects an edge server and terminal equipment which meet the conditions and transmits a selection result to the cloud server; model training is carried out according to the selection result, and the cloud server and the edge server carry out quality evaluation on the edge model and the local model respectively and reject unreliable edge servers and terminal devices; and performing weighted calculation on the historical reputation and the reputation obtained through model training to obtain the latest reputation, and updating the latest reputation to the alliance block chain. According to the method, three indexes of reputation, time delay and energy consumption are introduced to cooperatively enhance the reliability of model training, and a calculation method for weighted updating of historical reputation and current reputation is designed to improve the accuracy of reputation evaluation of a hierarchical federated edge learning system on terminal equipment and an edge server.

Description

technical field [0001] The invention relates to the technical fields of wireless communication and federated machine learning, in particular to a model training node selection method for layered federated edge learning. Background technique [0002] Federated Learning (FL for short) is an emerging machine learning technology that uses local data sets of nodes (such as edge servers and terminal devices) for distributed model training. Compared with traditional machine learning that uploads raw data of nodes on remote cloud servers, nodes in FL only share model parameters without uploading raw data, so it can provide privacy protection for network nodes. [0003] Edge Computing (EC for short) is an emerging technology that sinks mobile computing, network control and storage to the edge of the network (such as base stations and access points), which enables terminal devices with limited resources to run computing Intensive and delay-sensitive applications, and improve the priv...

AI Technical Summary

Problems solved by technology

[0005] The existing research on hierarchical federated edge learning technology mainly focuses on designing more advanced hierarchical federated edge learning algorithms to achieve better learning performance, and it is usually assumed that in the process of hierarchical federated edge learning, terminal devices and edge servers Unconditional contribution of computing resources, unconditional provision of real local data for model training, and unconditional willingness to own model training, while there are few studies on the incentive mechanism and selection method of terminal equipment participating in training, which hinders stratification Development and application of federated edge learning

Images