Data privacy protection method and system in machine learning

Abstract

The invention relates to a data privacy protection method and system in machine learning. The method is characterized by comprising the following steps of: 1) selecting a to-be-used encryption algorithm and system parameters to generate a secret key; 2) encrypting original data to generate corresponding cyphertext data; 3) carrying out training and parameter adjustment on a to-be-used machine learning model by using the cyphertext data so as to obtain an optimal machine learning model; and 4) encrypting to-be-predicted or classified original data by using the secret key in the step 1) by adoption of the method in the step 2) , and inputting the to-be-predicted or classified original data into the optimal machine learning model to obtain a prediction or classification result. According to the method and system, an order preserving / distribution property preserving encryption algorithm and the machine learning model are combined, so that the original data and the machine learning model can be protected. The swelling degree of cyphertext output by the order preserving / distribution property preserving encryption algorithm is far lower than that of a full-homomorphic encryption algorithm, and certain distribution features in plaintext features can be kept, so that the machine learning is relatively high in efficiency and has relatively good expansibility.

Description

technical field [0001] The invention relates to the field of data security, in particular to a data privacy protection method and system in machine learning. Background technique [0002] In recent years, with the continuous development of information technology, machine learning technology has become the cornerstone of technology in the era of big data. Machine learning technology explores the existing data, discovers potential connections in the data, and classifies or predicts according to the obtained model. Numerous service providers use machine learning models as a resource to provide services to the public, bringing many conveniences to people. However, current machine learning-based services ignore public privacy concerns. In order to be able to train models, service providers collect a large amount of user information, which even includes user privacy data. Users lose control over the data after uploading the data, and cannot guarantee whether the data will be mi...

AI Technical Summary

Problems solved by technology

[0004] 1) The current security protection of statistical data mainly focuses on protecting individual data from being extracted from group data, but lacks the protection of the data content itself;

[0005] 2) Existing schemes propose to use fully homomorphic encryption technology to process data and apply it to machine learning, but the length of ciphertext generated by fully homomorphic encryption is much larger than the length of plaintext, resulting in a serious drop in training efficiency. Poor in aspect;

[0006] 3) When applying the processing method based on fully homomorphic encryption, the corresponding machine learning algorithm needs to be modified accordingly, which increases the accuracy loss of the model and also reduces the scalability of the method

Images