The paper “STARdom: an architecture for trusted and secure human-centered manufacturing systems” has been accepted for presentation at the IFIP International conference on “Advances in Production Management Systems” (APMS) that will take place 5-9 September 2021 in Nantes, France. Dr Sofianna Menesidou, Dr Dimitris Papamartzivanos, and Dr Thanassis Giannetsos (members of UBITECH’s Digital Security and Trusted Computing Research Group), and their co-authors propose an architecture fortrusted and human-centered AI systems in pro[1]duction environments. The architecture aims to serve as a blueprint for realizing manufacturing systems that support AI models, Explainable AI (XAI), and Active Learning (AL), while providing security guarantees. It provides structuring principles for integrating these mod[1]ules. It creates synergies between them, aligned with some of the widely-adopted reference models, such as the Big Data Value Association (BDVA) reference architecture model and the Industrial Internet Security Framework (IISF).
Specifically, the proposed architecture leverages layering concepts and com ponents from these architectures towards providing a blueprint for developing trusted AI systems. Leveraging on these blueprints, they also illustrate how cybersecurity components for industrial AI systems are combined with XAI and AL to enable trusted, human-centered AI. The architecture is instantiated and validated on a real-world use case regarding demand forecasting. The proposed modular architecture facilitates efficient industrial systems that combine advanced AI systems with appropriate cyber defense strategies. The architecture integrates predictions, gathers insights on relevant features, incorporates domain knowledge and context to each prediction, and provides a forecast explanation and decision-making options to the end-user.
