Downloaded: 1353 times
© 2021 Entso Veliou | Dimitrios Papamartzivanos | Sofia Anna Menesidou | Panagiotis Gouvas | Thanassis Giannetsos
This chapter aims to review, from the security standpoint, the artificial intelligence solutions used to empower smart manufacturing environments. Our analysis will focus on the adversarial models utilized by malevolent entities in order to cause malfunctions to AI-powered systems both during the training process, but also during the inferencing mode of the leveraged machine learning models. Such attacks can have significant impact to the operation of the manufacturing supply chain ecosystem, as they can affect not only the business continuity, but more importantly, the integrity of safety-critical operations of systems. Towards this direction, this chapter reviews the state-of-the-art in technical approaches to secure machine-learning models and pave the way towards the safe adoption of such measures in the manufacturing field. The focus is on new generation of artificial intelligence setups using at their core deep neural network structures. In addition, the chapter elaborates on attestation-based provenance mechanisms that guarantee the trustworthiness of data streams feeding AI systems. The goal is to highlight the need for robust solutions against adversarial machine learning attacks for such environments and to provide additional insights on the appropriate mitigation strategies against such intelligent aggressors.