基于污点分析的移动端深度学习模型泄露自动分析方法

打开文本图片集

关键词：动态分析；污点分析；深度学习；Android应用

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）08-025-2437-09

doi：10.19734/j.issn.1001-3695.2024.10.0501

Automatic method to analyze deep-learning model leakage on mobile systems using taint analysis

Zhu Wentian，Lin Jingqiang† （SchoolofCyberScience&Technology，UniversityofScience&TechnologyofChina，Hefei230O27，China）

Abstract：Toanalyze dep-learning modelleakageonmobilesystems，existing methods require todynamicallyrun mobile applicationsand manually trigerdeep-learningfunctions，whichareunstableand needalotof manualparticipations，limiting the widespread useof such analysis methods.To automatically analyze more deep-learing models on mobile systems，this paper proposed an automatic method to analyze deep-learning modelleakage on mobiledevices using taint analysis.This method combined keyword matching and entropy evaluation to extract modelfiles inmobile applications，usedataint-analysis method basedonsimulated execution totrackthefunctionaddress of modeldecryption，andfinallcalledthedecryption functionto obtaintheplaintextmodel.Basedonthisdesign，thispaperimplementedanautomatedanalysis toolcaledModelDec.The analysis esults on mainstream APP stores show thatthe model leakage detectionrateand theanalysis speed are better than existing methods，which demonstrate the effectiveness of the proposed method.

Key words：dynamic analysis； taint analysis；deep learning model；Android applications

0 引言

随着人工智能技术的飞速发展和广泛应用，深度学习算法正逐渐被集成到智能手机等移动设备上。（剩余24436字）