In recent years, there has been a growing interest in the effects of data poisoning attacks on data-driven control methods. Poisoning attacks are well-known to the Machine Learning community, which, however, make use of assumptions, such as cross-sample independence, that in general do not hold for linear dynamical systems. Consequently, these systems require different attack and detection methods than those developed for supervised learning problems in the i.i.d.\ setting. Since most data-driven control algorithms make use of the least-squares estimator, we study how poisoning impacts the least-squares estimate through the lens of statistical testing, and question in what way data poisoning attacks can be detected. We establish under which conditions the set of models compatible with the data includes the true model of the system, and we analyze different poisoning strategies for the attacker. On the basis of the arguments hereby presented, we propose a stealthy data poisoning attack on the least-squares estimator that can escape classical statistical tests, and conclude by showing the efficiency of the proposed attack.

翻译：近年来,人们对数据中毒袭击对数据驱动控制方法的影响越来越感兴趣。中毒袭击是机器学习界所熟知的,但是,他们利用了跨样数独立等假设,这些假设一般不支持线性动态系统,因此,这些系统要求采用不同于i.d.d.设置中为监督学习问题而开发的进攻和检测方法。由于大多数数据驱动控制算法使用最小估计值,因此我们研究通过统计测试透镜对最小估计值的影响,并询问如何检测数据中毒袭击。我们确定与数据兼容的模型在哪些条件下包括系统的真正模型,我们分析攻击者的不同中毒战略。我们在此提出的论点的基础上,提议对最起码估计值的测算器进行隐性数据中毒袭击,以显示拟议攻击的效率。