City Research Online

Abstract

A central theme of Neurosymbolic AI involves combining subsymbolic learning with symbolic reasoning. Our research explores this theme using OWL ontologies and knowledge graphs to provide symbolic reasoning services within neurosymbolic systems. Our central research questions are: (i) how can we combine subsymbolic learning with symbolic OWL reasoning, and (ii) what are the effects or benefits of doing so?

Our research has three threads. Thread one involves the creation of NeSy4VRD, a unique dataset resource that combines an image dataset and high-quality visual relationship annotations with a well-aligned, custom-designed, common sense OWL ontology called VRD-World.

Thread two uses NeSy4VRD to explore ways of combining neural network-based subsymbolic learning with symbolic OWL reasoning in neurosymbolic systems for detecting visual relationships in images.

Thread three focuses on our notion of a tensor knowledge graph --- a binary tensor representation for arbitrary symbolic OWL-based knowledge graphs. We describe this representation and show how OWL reasoning (and hence description logic reasoning, more generally) can be emulated using matrix algebra techniques based on relational operations defined within relational mathematics. We then explore applications of this representation, and associated reasoning techniques, within neurosymbolic systems.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software
Departments:	School of Science & Technology > Department of Computer Science School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads