Formalizing Ontologies for AI Models Validation: from OWL to Event-B

摘要

Quality data is of decisive importance for controlling critical cyber-physical systems. Most common real-life systems are driven by unstructured, decentralized, and growing amounts of data, while validation requires coherent data, that is well structured, consistent, and without ambiguities. Often, ontologies are well-suited for capturing domain knowledge data, deriving requirements, providing analysis, and developing applications. Still, ontological representations fall short when it comes to formal verification and validation, especially for large complex systems. In this research, we suggest a fully automated approach to transform ontology axioms, expressed in the Web Ontology Language (OWL), to Event-B predicates. We show, in a practical example, how bridging OWL to Event-B can scale the validation of ontology-driven AI systems.