A fully automatic data privacy protection system for user data

Abstract

The invention discloses a fully automatic data privacy protection system for user data. The system can search for intelligent modules specified by users, segment them, and repackage them into intelligent applications that can run and test independently. For this repackaged smart application, an IM-specific data protector is adaptively generated, so as to effectively generate a data protection scheme for different smart applications that can balance the protection of user data rights and the quality of deep learning reasoning. The data protection provided by this system has little impact on App DL service quality, i.e., the inference result of IM on protected user input is almost the same as that on unprotected image, but it can protect user data from such large Harm of Privacy Leakage Caused by Scale Face Matching.

Description

technical field [0001] The invention relates to a fully automatic data privacy protection system for user data, belonging to the technical field of data privacy protection. Background technique [0002] Mobile intelligent applications (Intelligent Apps) equipped with powerful deep learning (DL) services have rapidly become popular in recent years. These smart applications provide a variety of high-quality smart inference services, such as photo beautification, augmented reality, and video cropping, by deploying deep learning models on a smartphone's local or remote cloud platform. While enjoying the improvement of work and life brought by these mobile smart applications, users are also deeply concerned about the protection of the rights and interests of the data they provide to smart applications. For example, when users submit photos of family reunions or friends gatherings to greedy smart application providers, these application providers may easily infer the user's ident...

AI Technical Summary

Problems solved by technology

[0005] When using an application, although it is said that the user's control of the input data on the client side is legally compliant and reasonable in design, it is challenging to truly realize autonomous input data control in reality

This is because when evaluating how the data should be provided to the smart application at the user end, the user wants to provide the smart application with as little data access as possible while minimizing the impact on the quality of service of the DL application used. Specifically, the challenge As follows: First, users do not have the source code of the smart application and any help from the smart application provider, so it is difficult to find the DL model and its related usage code in the smart application of the App for data rights protection analysis; secondly, even Locate the deep learning models and related codes in the application. These models are also black boxes for users, so it is impossible to use techniques such as backpropagation to search for suitable data permission control methods, not to mention that users usually do not have deep learning skills. and hacking techniques

Finally, users have limited resources for generating data rights protection solutions. Most users do not have GPU servers at home, and may not be willing to spend a lot of time looking for rights protection solutions

Images