Ph.D XXXVII

Supervisor: Vincenzo Della Mea

Phone: +39 0432 558457

Room: Rizzi NS3

Mail: mihaihoria.popescu@uniud.it

Research Project

Integration of symbolic and sub-symbolic techniques for Explainable Artificial Intelligence (XAI)

In the last years, Artificial Intelligence (AI) has achieved a fast and widespread adoption that may deliver the best of expectations over many application sectors across the field. Nowadays, new AI applications often make use of sub-symbolic approaches, such as deep learning (DL) techniques to provide sophisticated features that would otherwise be difficult for human computer scientists to implement. However, despite the unprecedented performance arising from the use of DL, experts and practitioners nowadays recognizing a lack of key features such as inspectability, interpretability, or explainability, given that they are inherently designed and used as black boxes.

Although there is still a long way to go, new research initiatives toward explainable artificial intelligence (XAI) are focused at mitigating the opacity issue of black boxes and pursuing the ultimate objective of constructing comprehensible, accountable, and trustable intelligent systems that allow humans to understand the reasons behind system recommendations or decisions by combining the symbolic and sub-symbolic approaches which are somewhat complementary.

The aim of my PhD is to identify and support possible applications where sub-symbolic techniques are not suitable, and interpretability or explainability are necessary key feature to archive the goal. In other words, is to create a suite of machine learning techniques that:

• Produce more explainable models, while maintaining a high level of learning performance (prediction accuracy); and
• Enable human users to understand, appropriately trust, and effectively manage the emerging generation of artificially intelligent partners.