Method for training privacy protection model and privacy protection method and device

Abstract

Embodiments of the invention provide a privacy protection model training method, a privacy protection method and a privacy protection device. The privacy protection model is used for providing model privacy protection for a pre-trained deep learning model, an output result of the deep learning model is an image, and the training method comprises the steps of obtaining first data, and a first imagegenerated by the deep learning model for the first data; inputting the first image into the privacy protection model to generate a second image; determining an image difference loss based on the first image and the second image; inputting the first data into a substitution model for simulating a deep learning model, determining a first gradient by taking the first image as a label, and determining a second gradient by taking the second image as a label; determining gradient difference loss according to the first gradient and the second gradient; and updating the privacy protection model by taking image difference loss tending to be reduced and gradient difference loss tending to be increased as targets.

Description

technical field [0001] One or more embodiments of this specification relate to the fields of machine learning and data security, and in particular, to a method for training a privacy protection model, a privacy protection method, and a device. Background technique [0002] In recent years, artificial intelligence systems have developed rapidly and quickly entered people's production and life. At the same time, the security of artificial intelligence systems has also received unprecedented attention. The core of the artificial intelligence system is the deep learning model, so the privacy and security of the deep learning model determines the privacy and security level of the entire artificial intelligence system. Therefore, how to prevent the model from being stolen - model privacy protection has become a hot topic in recent years. [0003] The methods of stealing models can be roughly divided into two categories. The first category starts with analyzing the code of the art...

AI Technical Summary

Problems solved by technology

The second type of attack has no effective defense method yet

Images