智能系统全生命周期安全测试理论与方法

人工智能技术在公共、国防安全领域得到了广泛应用,然而智能系统的安全性面临极大挑战。如何有效、全面、深入地对智能系统进行安全测试成为解决当前智能系统安全问题的重要途径。近年来,国内外高度重视智能系统的安全性问题,开展了大量的安全测试理论方法研究并出台了大量相关政策文件。针对智能系统面临的安全问题,本文详细阐述了面向智能系统全生命周期的安全测试理论与方法。首先,本文说明了智能系统的特点、安全内涵及安全机理;接着,本文结合模型训练、模型推理、模型部署三个主要生命周期阶段,详细地阐述了智能系统面临的安全性挑战和测试理论方法;最后,本文从标准、平台等角度阐明了构建智能系统安全测试支撑体系的方式,分析了自动驾驶典型场景下的智能系统安全测试案例,并给出了未来展望。开展智能系统安全测试理论和方法体系的建设,可有效规避系统潜在风险和质量缺陷,是实现人工智能算法可解释、可信赖的基本路径,对于确保人工智能技术安全、可靠、可控具有重要意义。

Theory and Method of Full Life Cycle Security Testing for Intelligent Systems

Although artificial intelligence technology has been widely used in the field of public security and national defense security, the security of these intelligent systems faces great challenges. How to effectively, comprehensively, and deeply test the security of intelligent systems has become a crucial problem to enhance the security of current intelligent systems. In recent years, researchers and institutions attached great importance to the security of intelligent systems, carried out a large number of studies on security testing theories and methods, and issued a large number of relevant policy documents. Aiming at the security problems in intelligent systems, this research elaborates the theory and method of security testing from the perspective of the full life cycle ideology. Firstly, it explains the characteristics, security connotation and security mechanism of intelligent system; Then, combining the critical life cycle stages of model training, model inference, and model deployment, it elaborates the security challenges faced by the intelligent system and the testing theory and methods in detail; Finally, it clarifies the way to build an security test support hierarchy from the perspective of standards and platforms, analyzes the intelligent system security test cases under typical scenarios of automatic driving, and gives the prospects of the future security testing. It concludes that the construction of intelligent system security testing theory and method system can effectively avoid potential risks and quality defects; it is a basic path to achieve an interpretable and trustworthy artificial intelligence algorithm and is of great significance to ensure the safety, reliability and controllability of artificial intelligence technology.