Neuro-Symbolic AI for Explainable Artwork Classification

Understanding and classifying artworks represent a fundamental challenge in art analysis, requiring a fine interpretation of complex visual and thematic elements. Notably, this task has been tackled using deep neural networks, including both vision-based methods and hybrid ones. Although these approaches have proved effective in recognizing different characteristics in artworks, they are based on "black-box" models. Therefore, they do not offer the possibility of explaining the model predictions, which could interest experts. In our research, we are working on a novel neuro-symbolic approach, which integrates Graph Neural Networks with Fuzzy Logic to provide an explainable solution for artwork classification based on fuzzy rules. In particular, we are focusing on artwork classification based on genre.