军事智能数据安全问题:对抗攻击威胁  

The data security of military intelligence:adversarial attacks

作　　者：陆正之 黄希宸 彭勃 Lu Zhengzhi;Huang Xichen;Peng Bo(Test Center,National University of Defense Technology,Xi′an 710106,China;College of Electronic Science and Technology,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]国防科技大学试验训练基地,陕西西安710106 [2]国防科技大学电子科学学院,湖南长沙410073

出　　处：《网络安全与数据治理》2024年第11期23-28,共6页CYBER SECURITY AND DATA GOVERNANCE

基　　金：长沙市杰出创新青年培养计划(kq2107002)。

摘　　要：人工智能技术已深入军事作战的各个领域,对现代战争形态进行了全面革新。数据作为军事智能模型的核心驱动力,为模型的有效运转提供了保障。然而,由于深度学习的不可解释性,对抗攻击技术的存在给当前军事智能模型带来了严峻的数据安全问题。这种威胁在智能系统的训练和推理过程中均可能产生,形式多样,难以防范。同时,受到对抗样本干扰的军事数据类型多样,敌方采取的欺骗手段也日趋复杂。因此,分析军事智能数据安全风险样态,并进一步给出军事智能数据风险的防范措施,希望能够为增强军事智能数据的安全性提供有益的参考和借鉴。Artificial intelligence technology has now been deeply applied in various fields of military operations,comprehensively changing the shape of modern warfare.Data is the core driving force of military intelligence models,providing a guarantee for the effective operation of the models.However,due to the non-interpretability of deep learning,the existence of adversarial attack techniques has brought serious data security problems to current military intelligence models.On the one hand,such security threats come in various forms and can be affected during the full life cycle of training and reasoning of intelligent systems.On the other hand,the types of military data interfered by adversarial samples are complicated,and the means of implementing deception show a diversified trend.Therefore,this paper will analyse the security risk pattern of military intelligent data,and further give specific measures on how to prevent the risk of military intelligent data in the hope that it can provide certain references and lessons for improving the security of military intelligent data.

关 键 词：军事人工智能 数据安全 对抗攻击 物理对抗攻击 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]