Special Sessions

Explainable and Interpretable Machine Learning: From Symbolic Models to Neuro-Symbolic Approaches

Organizer: Dr. Federico Sabbatini, University of Bologna, Italy + University of Urbino, Italy + INFN Section in Florence, Italy (Contact: federico.sabbatini@uniurb.it)

Description and Rationale:

This special session focuses on recent advances in explainable artificial intelligence (XAI) and interpretable machine learning, with particular emphasis on methods relevant to pattern recognition and data-driven modeling. As machine learning systems are increasingly deployed in high-stakes and real-world scenarios, transparency, robustness, fairness, and trustworthiness have become essential requirements.

The session addresses both theoretical and applied aspects of interpretability, including intrinsically interpretable models, post-hoc explanation techniques, and emerging symbolic and neuro-symbolic approaches. Particular attention is devoted to methods that balance predictive performance with interpretability, as well as to rigorous evaluation frameworks.

A key focus is placed on fairness-aware machine learning, including bias detection, mitigation strategies, and the development of models that ensure equitable and accountable decision-making. The session aims to highlight how explainability and fairness can be jointly addressed, especially in sensitive application domains.

Contributions are encouraged on novel methodologies, quantitative evaluation metrics, and real-world applications where interpretability and fairness are critical, fostering cross-disciplinary exchange within the ANNPR community.

Topics of interest include (but are not limited to):

• Explainable artificial intelligence (XAI)
• Interpretable machine learning
• Symbolic and rule-based models
• Neuro-symbolic approaches
• Intrinsic vs post-hoc explainability methods
• Evaluation metrics for interpretability and explanation quality
• Bias detection and mitigation
• Fairness-aware ML models
• Accountability and transparency in AI systems
• Explainability in pattern recognition applications
• Case studies in critical domains (e.g., healthcare, finance, security)

To submit to this special session follow the On-line submission via OpenReview through the ANNPR 2026 submission link and select the special session name in the system: https://openreview.net/group?id=iapr.org/TC3/2026/Workshop/ANNPR&referrer=%5BHomepage%5D(%2F)#tab-your-consoles

Multimodal AI in Archaeology

Organizers: Nevio Dubbini (Chair), University of Pisa, Italy, nevio.dubbini@unipi.it, Sinem Aslan (University of Milan, Italy – sinem.aslan@unimi.it), Michel Mickael (Polish academy of Sciences, Poland m.mickael@igbzpan.pl) Nusret Demir (Akdeniz University, Turkey, nusretdemir@akdeniz.edu.tr), Marek Bundzel (Technical University in Košice, Slovakia, marek.bundzel@tuke.sk)

Recent advances in artificial intelligence (AI), particularly in machine learning, computer vision, and pattern recognition, are profoundly transforming the way archaeological data are collected, processed, and interpreted. The rapid growth in the availability of large-scale and heterogeneous datasets, including satellite and aerial imagery, LiDAR scans, geophysical data, 3D models of artifacts and sites, as well as textual and epigraphic sources, has opened new opportunities for data-driven approaches to Cultural Heritage. In this context, multimodal AI techniques, capable of jointly analyzing and integrating diverse data sources, are emerging as a key enabler for advancing archaeological research beyond traditional methodologies. Archaeology is inherently multimodal: understanding past human activity often requires combining spatial, visual, textual, and contextual information. However, the integration and joint modeling of heterogeneous data sources remain significant challenges. Recent advances in multimodal representation learning, cross-modal retrieval, and foundation models offer principled approaches to address these limitations, enabling richer interpretations and more robust analytical pipelines. This special session aims to address this fragmentation by defining a focused research venue for AI-driven approaches, with an emphasis on multimodal methods in archaeological contexts, enabling structured interaction between computer scientists, data scientists, and domain experts. The proposed session is well aligned with the scope of ANNPR 2026, emphasising artificial neural networks and pattern recognition techniques applied to complex, real-world problems. It will highlight both methodological innovations and practical applications in Archaeology and Cultural Heritage. Submissions are encouraged to demonstrate both methodological and technical contributions and meaningful impact on archaeological research questions, with specific attention to multimodal approaches involving the fusion, alignment, and joint representation of heterogeneous data sources such as imagery, spatial data, 3D models, and textual records.

Topics of interest for this special session include, but are not limited to:

• computer vision techniques for archaeological site detection, mapping, and monitoring;
• remote sensing and satellite imagery analysis for cultural heritage discovery and preservation;
• data fusion across visual, spatial, and textual sources;
• 3D reconstruction, modeling, and analysis of artifacts and archaeological environments;
• pattern recognition and anomaly detection in archaeological datasets;
• AI-based classification, dating, and provenance analysis of artifacts;
• natural language processing for historical documents, inscriptions, and excavation reports;
• cross-modal retrieval and alignment between images, 3D data, and textual records;
• predictive modeling for archaeological site discovery and landscape analysis;
• multimodal AI methods for conservation, restoration, and risk assessment of cultural heritage;
• explainability, uncertainty quantification, and trustworthiness in multimodal AI for archaeology.