Mathematical methods for explainable AI 

The recent surge of Machine Learning (ML) and, more broadly, of Artificial Intelligence (AI) brings to light old and new open issues, and among them, the so-called eXplainable Artificial Intelligence (XAI) - AI that humans can understand.

AI has recently seen a significant shift in focus towards designing and developing intelligent systems that are interpretable, transparent and explainable. This is due to the complexity of the induced model from data and the legal requirement imposed by various national and international parliaments. Consequently, this has echoed in the research literature and the press, attracting scholars worldwide and a lay audience. In particular, XAI can help solve some of the problems of AI, as highlighted in the Regulation of the European Parliament and The Council (AI ACT), laying down harmonised rules on Artificial Intelligence and amending certain union legislative acts.

The aim of this workshop is to discuss possibilities for designing XAI methods and its industrial applications, focussing on the emerging and fundamental role played by Mathematics in this area.

Mathematical methods which promise to improve the explainability of AI include (but are not limited to) Computational Geometry, Topological Data Analysis, Group Equivariant Non- Expansive Operators (GENEOs), Physics Informed Neural Networks (PINNs), measures of robustness, regularisation methods to reduce computational complexity, distributed and federated optimization.

During the workshop a sequence of topics that deal with both theoretical and practical aspects of the above-mentioned mathematical techniques will be presented and discussed.

A number of relevant industrial case studies, also enhancing the main related open problems, will also be presented.