Federal learning global model training method based on differential privacy and quantification

Abstract

A federated learning global model training method based on differential privacy and quantification comprises the following steps that a central server issues a pre-trained federated learning global model, a local model gradient of each local user is generated, and noise adding, threshold quantification and compression quantification is sequentially performed on the local model gradient of each local user, and the compressed and quantified local model gradient is uploaded to the central server, the central server performs weighted aggregation on the uploaded local model gradient by the central server, updates a global model, issues the global model to each local user, and training is ended when the privacy budget value of each local user is exhausted or the federal learning global model is converged. According to the method, on the premise that the precision of the federal learning global model is not lost, the privacy of local users is protected, the communication overhead in the transmission process is reduced, and the training efficiency of the federal learning global model is improved.

Description

technical field [0001] The invention belongs to the technical field of data processing, and further relates to a federated learning global model training method based on differential privacy and quantization in the technical field of machine learning model gradient data processing. The present invention can be used to train machine learning models from local data scattered in users, and supports the protection of individual user privacy, reduces the transmission scale of user gradient data, and finally reduces the communication overhead of the transmission process to improve the training efficiency of the federated learning global model Purpose. Background technique [0002] Federated learning allows users to jointly obtain a shared global model without storing data centrally. Specifically, users use local data to train their own local models locally, and upload the trained local model gradient data to the central server, which aggregates and updates the global model. Duri...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned existing technologies, and propose a federated learning global model training method based on differential privacy and quantization, which is used to solve the privacy leakage, large communication cost and computational overhead when uploading local model gradients in federated learning big problem

Images