Supervisor: Patrick Koopmann (p.k.koopmann@vu.nl)
Ontologies are an important representation formalism for symbolic AI systems. Ontologies that are formulated in OWL allow to usage of reasoning systems such as HermiT or ELK to infer implicit information from an ontology, or from a dataset that is combined with the ontology. Different to inferences performed by sub-symbolic AI systems, decisions made by such a reasoner are in a way “explainable by design”, because all inferences can be explained solely based on the information available in the ontology.
But what do you do if the reasoner does not infer the expected result, how do you explain missing entailments? In the research literature, there are currently two main approaches to provide such explanations: 1) counterexamples and 2) abduction. A counterexample explains a missing entailment by showing a scenario that is fully consistent with the ontology, but which makes it very clear that the expected entailment does not hold. Abduction by contrast explains the missing entailment by offering a fix, an extension to the current ontology that would produce the desired inference. In this project, you will choose one of these two techniques, and then develop and evaluate extensions/modifications of current methods for computing such explanations, towards making them more efficient or producing better explanations.
The supervisor will give an introduction to the topic (foundations of description logics, counterexamples, abduction) at the beginning of the project. To get an idea, you can consult the following paper, which explains and discusses abduction-based explanations on an example:
https://lat.inf.tu-dresden.de/research/papers/2023/ABFKK-DL23.pdf
Don’t be afraid to contact the supervisor if you would like to have more information on this project or would like to discuss it in more detail in person.