Joint learning method, device and system based on parameter expansion

Abstract

The invention relates to the technical field of joint learning, in particular to a joint learning method, device and system based on parameter expansion. The method comprises the steps: training locally a global model, and recording gradient values generated in the local training process; improving a k anonymity algorithm by using the gradient value set and adopting an expansion method, and expanding a parameter set; sending the parameter sets to a server, performing single-point aggregation on the received parameter sets by the server, and solving an average value in each parameter set; solving new parameters of the average values by adopting weighted average; constructing a new global model by utilizing the new parameters, testing whether the new global model meets an iteration stoppingcondition or not, stopping the training process if the new global model meets the iteration stopping condition, and otherwise, sending the global model to the computing equipment by the server to continue the training. According to the method disclosed in the scheme, the reconstruction attack problem in the joint learning process can be solved, the parameter concealment is enhanced, and the data security is protected.

Description

technical field [0001] The present invention relates to the technical field of joint learning, in particular to a joint learning method, device and system based on parameter expansion. Background technique [0002] In recent years, machine learning has become the core tool of image processing, natural language recognition and other technologies, and has achieved breakthrough applications in image recognition, automatic driving and other fields. The application of machine learning algorithms is inseparable from the available data. Large-scale data collection can improve the performance of machine learning applications, but at the same time, many personal privacy information contained in the data set will face leakage as the data set is shared and applied. risks of. For example, in the field of Internet of Vehicles, the current Internet of Vehicles adopts a network architecture centered on cloud computing. There are two serious problems in the process of data interaction betw...

AI Technical Summary

Problems solved by technology

[0009]Aiming at the deficiencies of the existing technology, the present invention proposes a joint learning method, device and system based on parameter expansion, and solves the reconstruction faced by joint learning through anonymization technology As for the attack problem, the research shows that the privacy protection technology based on data anonymization is relatively balanced in terms of privacy protection degree, algorithm complexity, data validity and algorithm scalability, which makes the calculation overhead and information loss of privacy protection less

Images