Keywords: : Artificial Intelligence, AI Act, EU Regulation, Readiness Assessment, Compliance, Critical Infrastructure
Astract: The integration of Artificial Intelligence (AI) into critical infrastructure presents both opportunities and challenges. Compliance with regulations, such as the EU's Artificial Intelligence Act, is therefore crucial but can be challenging and resource intensive. This thesis aims to address this issue by developing a requirements catalogue for AI system providers in critical infrastructure to assess compliance with the AI Act. The research explored AI's application in infrastructure domains like water, electricity, heating, gas, and road traffic management. The Design Science Research framework, along with a Systematic Literature Review (SLR), the Technology Acceptance Model, and the System Usability Scale, were used as methodological approaches. Interviews and surveys with Information Security experts evaluated the usefulness of the catalogue. The SLR revealed that AI has been integrated into all critical infrastructure domains, with road traffic management showing the most significant concentration. The developed requirements catalogue was well accepted, with high perceived usefulness and ease of use, supported by an average SUS score of 92.9%. It offers several benefits to organisations seeking compliance, including time and resource savings, a clear presentation of requirements, shared understanding among team members and aid in contextualising requirements. Visualisations such as spider graphs and heatmaps enhance the interpretation of requirements and identify areas requiring further action. The catalogue also calculates the maturity level as an indicator of overall compliance. In conclusion, this thesis emphasises the valuable role of the requirements catalogue in navigating AI integration into critical infrastructure, providing clarity and guidance throughout the compliance process.

Keywords: Conceptual Modeling, Graphical Language Server Protocol, Integrated Development Environment, Language Server Protocol, Modeling tools, Software Engineering
Astract: With the introduction of the Language Server Protocol (LSP), a fundamental shift has been observed in the development of language editing support for Integrated Development Environments (IDEs), such as VS Code, the traditional Eclipse IDE, or Eclipse Theia. LSP establishes a uniform protocol that standardizes the communication between a language client (e. g., an IDE like Eclipse) and a language server (e. g., for a programming language like Java). The language client only needs to be able to interpret and understand the protocol instead of the specific programming language. Likewise, the language server can focus on language support and does not need to consider the specifics of a respective IDE. This reduces the complexity of realizing language support on different editors and IDEs and enables smooth transitions from one IDE to another. LSP is an open and community-driven protocol that has been developed within the realm of the VS Code community, initiated and driven by Microsoft. The generic concept and architectural pattern of LSP enables widespread applications that go far beyond the realization of editing support for programming languages. This paper provides an introduction to LSP, describes its evolution and core characteristics, and delineates its potential for revolutionizing not only the IDE market but also other software systems, such as modeling tools.

Keywords: Conceptual Modeling

Keywords: Graph Neural Networks, Ontology-Driven Conceptual models, Representation Learning
Astract: Conceptual Models (CMs) are essential for information systems engineering since they provide explicit and detailed representations of the subject domains at hand. Ontology-driven conceptual modeling (ODCM) languages provide primitives for articulating these domain notions based on the ontological categories put forth by upper-level (or foundational) ontologies. Many existing CMs have been created using ontologically-neutral languages (e.g., UML, ER). Connecting these models to ontological categories would provide better support for meaning negotiation, semantic interoperability, and complexity management. However, given the sheer size of this legacy base, manual stereotyping is a prohibitive task. This paper addresses this problem by proposing an approach based on Graph Neural Networks towards automating the task of stereotyping UML class diagrams with the meta-classes offered by the ODCM language OntoUML. Since these meta-classes (stereotypes) represent ontological distinctions put forth by a foundational ontology, this task is equivalent to ontological category prediction for these classes. To enable this approach, we propose a strategy for representing CM vector embeddings that preserve the model elements’ structure and ontological categorization. Finally, we present an evaluation that shows convincing learning of OntoUML model node embeddings used for OntoUML stereotype prediction.

Keywords: Model-Based Construction