Data model training method and device

Abstract

The invention provides a data model training method. The method includes the steps of acquiring a training set which includes multiple pieces of training data, and conducting first privacy protectionprocessing on the training set to obtain an encrypted training set; using the encrypted training set for training a data model to obtain model parameters; conducting second privacy protection processing on the obtained model parameters to obtain encryption model parameters, and generating an encrypted data model based on the encryption model parameters. Through the first privacy protection processing, the encrypted training set is different from the training set, and training results of the encrypted training set and the training set are the same; through the second privacy protection processing, the encryption model parameters are different from the model parameters, and the encrypted data model generated based on the encryption model parameters and the data model generated based on the model parameters have the same output results under the same input. The invention further provides a data model training device and computer equipment.

Description

technical field [0001] The present disclosure relates to a data model training method and device. Background technique [0002] In recent years, with the forced rise of artificial intelligence, people have realized the great potential of artificial intelligence technology. In artificial intelligence technology, data is the carrier, intelligence is the goal, and machine learning is the technical path from data to intelligence. The essence of machine learning is a process in which a computer uses existing data to train a certain data model and uses this data model to make predictions. [0003] In the process of implementing the disclosed concept, the inventors found that in the process of data model training based on machine learning or deep learning in the prior art, the training data and the model parameters obtained after the training are often public, which is convenient for attackers to obtain. The insecure state that can be obtained is easily exploited maliciously by at...

AI Technical Summary

Problems solved by technology

[0003] In the process of implementing the disclosed concept, the inventors found that in the process of data model training based on machine learning or deep learning in the prior art, the training data and the model parameters obtained after the training are often public, which is convenient for attackers to obtain. Obtainable insecure state, easily exploited maliciously by attackers

Images