Data sharing method based on collaborative deep learning

Abstract

The invention discloses a data sharing method based on collaborative deep learning. The system comprises two servers and a plurality of participation mechanisms. Each participation mechanism has biomedical data in the same format. The server is responsible for updating and maintaining the model parameters uploaded by the participation mechanism; each participation mechanism only needs to downloadthe latest parameter secret share from the two servers. Secrete reconstruction and recovery are carried out locally to obtain a latest parameter; and then deep learning model training is carried out locally by utilizing the parameters and private biomedical data of the two servers, the updated model parameters are split into two secret shares through a secret sharing scheme after training is completed, and the two servers carry out parameter updating operation on the secret shares respectively for downloading of the participating institutions. The cooperative deep learning method and the secret sharing scheme are used, data privacy in the biomedical data sharing process is protected, and great significance is achieved for promoting sharing of biomedical data.

Description

technical field [0001] The invention relates to the field of privacy protection data sharing, in particular to a data sharing method based on collaborative deep learning. Background technique [0002] With the maturity of artificial intelligence technology in recent years, the application of artificial intelligence in the medical field has greatly promoted the degree of medical intelligence, especially for the direction of disease diagnosis based on artificial intelligence, there have been a lot of research results. Artificial intelligence has enormous potential to revolutionize the diagnosis and management of diseases by analyzing and classifying vast amounts of data that is difficult for human experts to accomplish. [0003] At present, there are many examples of using deep learning methods to train various disease diagnosis models, and by extracting features from biomedical data, a diagnosis can be made on the patient's physical health. However, while the artificial inte...

AI Technical Summary

Problems solved by technology

However, they did not take into account that the shared parameters in the collaborative learning process may also leak local data privacy. For example, Hitaj et al. proposed using the generative adversarial learning method to implement inference attacks on the data of other participants in the collaborative learning process to obtain the privacy of other participants. training data

Images