Edge privacy protection method based on federal learning

Abstract

The invention relates to an edge privacy protection method based on federated learning, and the method comprises the steps: an edge device downloads a model from a server, initializes a federated training model, and employs local data to train the model; adding disturbance to gradient parameters of a local model by using a localized differential privacy technology, and uploading the encrypted parameters to a server; and the server receives the local model parameters subjected to local differential privacy encryption, aggregates the local model parameters and updates global model parameters. According to the method, local differential privacy is applied to federated learning, localized training and model aggregation of data in edge equipment can be completed while safe data sharing of edge users is guaranteed, data does not need to be uploaded in the whole process, privacy safety of the data of the edge users is guaranteed, and the purpose of protecting model parameters is achieved while the accuracy of the model is guaranteed.

Description

technical field [0001] The present invention relates to the field of data security and privacy protection technology, in particular to an edge privacy protection method based on federated learning. Background technique [0002] In the era of big data, IoT devices continue to proliferate, and the data in the network is growing explosively. Cloud computing, which traditionally processes data in a centralized manner, is gradually unable to meet the needs of networked devices. In this context, edge computing for distributed learning emerges as the times require. The combination of edge computing and machine learning has greatly promoted the application and development of edge computing. When a large number of machine learning services become an indispensable part of daily life, more user data needs to be collected, and the security of user privacy data cannot be guaranteed. Therefore, corresponding measures must be taken to protect the privacy of users. But standard machine l...

AI Technical Summary

Problems solved by technology

However, the federated learning algorithm based on differential privacy does not consider privacy protection during the parameter upload stage, that is, when the local model is uploaded to the server, the customer's gradient information may be intercepted by hidden opponents

In the scenario of edge computing, edge users outsource private data through edge servers, resulting in edge users being unable to control the use of outsourced data

At the same time, edge devices that store outsourced data cannot provide reliable security

Images