The widespread adoption and success of Machine Learning (ML) technologies depend on thorough testing of the resilience and robustness to adversarial attacks. The testing should focus on both the model and the data. It is necessary to build robust and resilient systems to withstand disruptions and remain functional despite the action of adversaries, specifically in the security-sensitive Nuclear Industry (NI), where consequences can be fatal in terms of both human lives and assets. We analyse ML-based research works that have investigated adversaries and defence strategies in the NI. We then present the progress in the adoption of ML techniques, identify use cases where adversaries can threaten the ML-enabled systems, and finally identify the progress on building Resilient Machine Learning (rML) systems entirely focusing on the NI domain.

机器学习 (ML)技术的广泛采用和成功取决于对对抗性攻击的弹性和稳健性的彻底测试。测试应重点关注模型和数据。有必要建立强大且有弹性的系统，以承受干扰并在对手采取行动的情况下保持功能，特别是在安全敏感的核工业（NI）中，其后果可能对人员生命和资产造成致命影响。我们分析了基于机器学习的研究工作，这些研究工作调查了NI中的对手和防御策略。然后，我们介绍 ML 技术采用的进展，确定对手可以威胁支持 ML 的系统的用例，最后确定完全专注于 NI 领域的弹性机器学习 (rML)系统的构建进展。