A privacy protection method and system based on cloud-edge-end edge computing system

Abstract

The present invention relates to a privacy protection method and system of a cloud-edge-end edge computing system based on federated learning, including: establishing a convergence model and a privacy leakage model based on the privacy budget of user equipment; based on the convergence model and the privacy leakage model, Establish the cloud server utility function, edge server utility function and user equipment utility function; optimize the edge server utility function and user equipment utility function based on the algorithm of optimal control theory, and obtain the edge server incentive strategy and user equipment privacy policy; based on gradient ascent The algorithm optimizes the utility function of the cloud server to obtain the incentive strategy of the cloud server; the federated learning of the image classification model is carried out based on the privacy policy of the user equipment, the incentive strategy of the edge server and the incentive strategy of the cloud server, so as to realize the protection of the private data of the user equipment during the federated learning process, and at the same time It can improve the classification accuracy of the image classification model.

Description

Technical field [0001] The present invention relates to the technical field of protecting user privacy data, in particular to a privacy protection method and system of a federal learning-based cloud-edge-ended edge computing system. Background technique [0002] Federal learning based on edge computing and differential privacy, by training models on user equipment (not need to upload raw data to server), and to maximize user data privacy leakage to maximize user data privacy. Among them, the applied noise will affect the accuracy of the image classification model. As the noise scale increases, the accuracy of the image classification model will decrease. Therefore, we need to weigh privacy and image classification model, which can be implemented by incentive mechanism. Through the quantified privacy leak model, there is a small number of users who have a privacy, and the noise scale is small, and the revenue has higher revenue, we can give high return incentives. The design probl...

AI Technical Summary

Problems solved by technology

However, the design problem of the three-tier incentive mechanism based on edge computing is more complicated and needs to be solved.

The main goal of existing scheme design incentive mechanism is to encourage user equipment to provide more data volume and computing resources, but it does not consider the issue of privacy leakage, and cannot guarantee that users who are more sensitive to privacy can actively participate in federated learning tasks, and cannot obtain high-precision Model

Images