Model file protection

By converting the model file into a binary file and performing source code processing on the server side, the security risks of the artificial intelligence model file on the terminal device are resolved, and high security and privacy protection of the model file are achieved.

WO2025201476A1PCT designated stage Publication Date: 2025-10-02ALIPAY (HANGZHOU) INFORMATION TECH CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2025/085432
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2024-03-29
Filing Date
2025-03-27
Publication Date
2025-10-02

AI Technical Summary

Technical Problem

Artificial intelligence model files are at risk of being attacked or stolen on terminal devices, and existing technologies are difficult to effectively prevent reverse operations, affecting the security and privacy protection of model files.

Method used

By using a preset compiler on the server to convert the model file into a binary file, such as ELF, AXF, or BIN format, combined with the parsing and compilation of the source code file, the difficulty of reverse analysis is increased, and it is run on the terminal device to ensure the security of the model file.

Benefits of technology

The security and privacy protection of model files on terminal devices are significantly improved, the difficulty of theft and reverse analysis is reduced, and the equivalence of normal operation of model files on terminal devices is ensured.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2025085432_02102025_PF_FP_ABST
    Figure CN2025085432_02102025_PF_FP_ABST
Patent Text Reader

Abstract

Disclosed in one or more embodiments of the present description are a model file protection method and apparatus. The method comprises: firstly, acquiring a model file corresponding to a target model to be protected, the model file being constructed on the basis of a model graph formed by a plurality of operators; and then, on the basis of the acquired model file, compiling the target model by means of a preset compiler to obtain a binary file corresponding to the target model, and sending the binary file corresponding to the target model to a terminal device, the binary file corresponding to the target model being used for triggering the terminal device to run the target model by running the binary file.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Processing unit, system on chip and processing method

    CN114647612A

  • Diversified compiling method and system based on random confusion

    CN116627421A

  • Model operation method and device for secure multi-party computing, electronic equipment and medium

    CN116821931A

  • Model file protection method and device

    CN118194352A

  • End-side deep learning model protection method and system based on code obfuscation

    CN118349967A