Page 295 - 《软件学报》2026年第4期

1736 软件学报 2026 年第 37 卷第 4 期

征向量, 最后利用异类识别网络降低甚至消除多模态特征中受到对抗攻击干扰的模态特征对最终恶意软件预测的
影响, 从而提高其抵御对抗攻击的鲁棒性. 实验结果表明, 相较于现有基线检测方法展现了更高的有效性和鲁棒
性, 即不仅可以有效地检测一般性的安卓恶意软件, 而且可以检测出攻击者恶意生成的对抗性安卓恶意软件.
在下一步研究工作中, 本文计划从以下两方面进行研究探索. 首先, 尽管本文已经选取当前最先进的 3 种对抗
攻击方法来评估的鲁棒性, 但是由于目前没有任何针对安卓恶意软件中控制流图特征的对抗攻击, 更没有任何针
对安卓恶意软件中多模态特征的对抗攻击, 因此无法评估控制流图模态特征以及多模态特征的鲁棒性. 后续的研
究工作将重点探索如何设计并提出针对基于控制流图的安卓恶意软件检测的对抗攻击方法, 并进一步提出针对多
模态特征的安卓恶意软件检测的对抗攻击方法, 从而进一步评估最新安卓恶意软件检测方法及工具的鲁棒性. 其
次, 尽管本文利用已有的对抗攻击方法评估安卓恶意软件检测方法的经验鲁棒性, 但是无法准确地评估该检测方
法针对任何对抗攻击方法的理论鲁棒性. 事实上, 可验证鲁棒性是对抗机器学习领域的一个重要研究方向, 旨在利
用形式化验证等方法来证明机器学习或者深度学习模型在一定范围内对对抗攻击是鲁棒的. 然而, 现有的可验证
鲁棒性研究大量集中在计算机视觉和自然语言处理领域, 因此亟需针对基于学习的安卓恶意软件检测方法进行可
验证鲁棒性研究, 构建可信的安卓恶意软件检测系统, 对保障智能移动终端用户的信息安全具有重要意义.

References
[1] TechInsights. Global smartphone sales forecast by operating system for 88 countries: 2007 to 2029. 2025. https://www.techinsights.com/
blog/global-smartphone-sales-forecast-operating-system-88-countries-2007-2029
[2] XuebaoCaijingshe. Institution: The global smartphone sales are expected to decline by 5% YoY in 2023, with an approximate Android
market share of 80%. 2023 (in Chinese). https://www.speed-daily.com/newsflashes.html#23981
[3] Kaspersky. The mobile malware threat landscape in 2022. 2023. https://securelist.com/mobile-threat-report-2022/108844/
[4] Kaspersky. Harly: Another Trojan subscriber on Google Play. 2023. https://www.kaspersky.com/blog/harly-trojan-subscriber/45573/
[5] Liu Y, Tantithamthavorn C, Li L, Liu YP. Deep learning for Android malware defenses: A systematic literature review. ACM Computing
Surveys, 2022, 55(8): 153. [doi: 10.1145/3544968]
[6] Google for Developers. Google Play protect. 2025. https://developers.google.com/android/play-protect
[7] Norton Inc. Norton mobile security for Android. 2025. https://uk.norton.com/products/mobile-security-for-android?srsltid=AfmBOopv_
qRZtkOMHkt4fsIWHlAqGcEOrJ4gLJ5GCkl1-PykY5fl6Gv7
[8] Tencent Inc. Tencent mobile manager. 2025 (in Chinese). https://m.qq.com/
[9] Qihoo 360 Inc. 360 Security for Mobile Devices 9.0. 2025 (in Chinese). https://shouji.360.cn/v6/index.html
[10] Guarnieri C, Tanasi A, Bremer J, Schloesser M, Houtman K, van Zutphen R, Graaff B. Cuckoo sandbox. 2019. https://cuckoosandbox.org/
community-sub/index
[11] Leavitt N. Malicious code moves to mobile devices. Computer, 2000, 33(12): 16–19.
[12] Feng Y, Anand S, Dillig I, Aiken A. Apposcopy: Semantics-based detection of Android malware through static analysis. In: Proc. of the
22nd ACM SIGSOFT Int’l Symp. on Foundations of Software Engineering. Hong Kong: ACM, 2014. 576–587. [doi: 10.1145/2635868.
2635869]
[13] Faruki P, Bharmal A, Laxmi V, Ganmoor V, Gaur MS, Conti M, Rajarajan M. Android security: A survey of issues, malware penetration,
and defenses. IEEE Communications Surveys & Tutorials, 2015, 17(2): 998–1022. [doi: 10.1109/COMST.2014.2386139]
[14] Ye YF, Li T, Adjeroh D, Iyengar SS. A survey on malware detection using data mining techniques. ACM Computing Surveys, 2017,
50(3): 41. [doi: 10.1145/3073559]
[15] Pan WW, Wang XY, Song ML, Chen C. Survey on generating adversarial examples. Ruan Jian Xue Bao/Journal of Software, 2020,
31(1): 67–81 (in Chinese with English abstract). http://www.jos.org.cn/1000-9825/5884.htm [doi: 10.13328/j.cnki.jos.005884]
[16] Ling X, Wu LF, Zhang JY, Qu ZQ, Deng W, Chen X, Qian YG, Wu CM, Ji SL, Luo TY, Wu JZ, Wu YJ. Adversarial attacks against
windows pe malware detection: A survey of the state-of-the-art. Computers & Security, 2023, 128: 103134. [doi: 10.1016/j.cose.2023.
103134]
[17] Zhao KF, Zhou H, Zhu YL, Zhan X, Zhou K, Li JF, Yu L, Yuan W, Luo XP. Structural attack against graph based Android malware
detection. In: Proc. of the 2021 ACM SIGSAC Conf. on Computer and Communications Security. ACM, 2021. 3218–3235. [doi: 10.1145/
3460120.3485387]
[18] Wu YM, Li XD, Zou DQ, Yang W, Zhang X, Jin H. MalScan: Fast market-wide mobile malware scanning by social-network centrality