The integration of symbolic and sub-symbolic techniques (or, neuro-symbolic integration) represents nowadays a popular approach in AI, used to mitigate the issues of neural networks in terms of decision processes, interpretability, and the like.

However, precisely measuring properties such as trustworthiness and interpretability is a complex issue that most neuro-symbolic integration techniques today do not address, by focusing instead on raw predictive performance.

In this talk, we explore two popular paradigms mixing sub-symbolic and symbolic techniques – namely, symbolic knowledge extraction (SKE) and injection (SKI) – and highlight current results and open perspectives in the study of trustworthiness metrics for neuro-symbolic integration, focusing on features such as explainability, robustness, and interpretability.

Download keynote slides (PDF)

Andrea Omicini (Scholar citations 10495, h-index 51 / March 2023) is a Full Professor of the Alma Mater Studiorum – Università di Bologna, and holds a PhD in Computer & Electronic Engineering. He published some hundreds of articles on computational logic, coordination, multi-agent systems, intelligent systems, machine learning, programming languages, simulation, software engineering, pervasive systems, self-organisation, autonomy; on the same subjects, he edited more than fifty volumes (books & special issues). He organised and chaired several international conferences and workshops, and was a committee member of hundreds of international conferences, workshops, and symposia; he will co-host ECAI 2025 in Bologna. He worked as the Chair of the SIG on Agents and Multi-Agent Systems of the Italian Association for Artificial Intelligence (AI*IA), and the ACM Representative in the IFIP "Artificial Intelligence" TC; currently, he is Emeritus Member of the Board of Director of the European Association for Multi-Agent Systems (EURAMAS).

Prof. Omicini's Google Scholar page

Personal homepage