Spotlight Seminars on AI – Spring 2026

L’Associazione Italiana per l’Intelligenza Artificiale (AIxIA) è lieta di annunciare il programma dei Seminari Spotlight sull’IA per la primavera 2026.

L’obiettivo dei seminari è quello di illustrare, esplorare e discutere le sfide scientifiche, le tendenze e le possibilità attuali in tutti i rami del nostro articolato campo di ricerca. I seminari si terranno virtualmente (https://www.youtube.com/c/AIxIAit), con cadenza mensile, e saranno tenuti da ricercatori italiani di spicco e da scienziati internazionali di alto livello. I seminari si rivolgono principalmente a un pubblico ampio interessato alla ricerca sull’IA e sono inclusi anche nel programma di dottorato italiano in Intelligenza Artificiale; AIxIA incoraggia vivamente la partecipazione di giovani ricercatori e dottorandi.

Questo il calendario previsto:

• Prof. Fabrizio Silvestri, University “La Sapienza” Roma
  • Aprile 30 – 5:00 PM (CEST)
  • Title: Learning with Structured Consistency: Sheaf Neural Networks from Theory to Applications
  • Abstract: Modern machine learning on graphs is largely built on the assumption that neighboring nodes should be similar, an inductive bias that breaks down in the presence of heterogeneity, multi-modality, and relational ambiguity. Sheaf Neural Networks offer a principled alternative by replacing this notion of similarity with structured consistency: information is no longer directly compared across nodes, but aligned through learnable transformations defined on edges and, more generally, on higher-order relations.In this talk, I introduce the sheaf-theoretic perspective as a natural and flexible framework for learning over complex relational data, spanning both graphs and hypergraphs. I will motivate the approach through concrete failure modes of standard Graph Neural Networks, and show how sheaves provide a unified solution to modeling directionality, heterophily, and context-dependent interactions, while naturally extending to higher-order structures beyond pairwise connections. I will then present a sequence of theoretical results that characterize the expressivity, stability, and spectral properties of Sheaf Neural Networks, highlighting their role as a strict generalization of classical architectures.Finally, I will illustrate how these ideas translate into practice through applications in recommender systems, where modeling structured inconsistency leads to significant empirical improvements. I conclude by discussing open challenges and future directions, including scalable learning of sheaf structures on hypergraphs and connections to broader geometric deep learning paradigms.
  • Bio: Fabrizio Silvestri is a Full Professor at Sapienza, University of Rome in Italy. Formerly a Research Scientist at Facebook AI in London. His interests are in AI applied to integrity-related problems and Natural Language Processing. In the past, he has worked on web search research, and in particular, his specialization is building systems to interpret better queries from search users. Before Facebook, Fabrizio was a principal scientist at Yahoo, where he had worked on sponsored search and native ads within the Gemini project. Fabrizio holds a Ph.D. in Computer Science from the University of Pisa, Italy, where he studied problems related to Web Information Retrieval with a particular focus on Efficiency-related problems like Caching, Collection Partitioning, and Distributed IR in general.

• Prof. Leslie Pack Kaelbling– MIT
  • Maggio 29 – 3:00 PM (CEST) – in collaboration with FBK academy
  • Title: The Role of Rationality in Modern Robotics
  • Abstract: The classical approach to AI designed systems that were rational at run-time: they had explicit representations of beliefs, goals, and plans and ran inference algorithms, online, to select actions. The rational approach was criticized (by the behaviorists) and modified (by the probabilists), but persisted in some form. More recently, relatively unstructured data-driven end-to-end approaches have demonstrated great success in a wide variety of domains and began to seem like a plausible route to general-purpose intelligent robots. However, most recently, we have begun to see the limits of pure behavior learning, and many practitioners are re-integrating forms of search and explicit reasoning into their approaches.
  • I will revisit the rational-agent approach to the design of intelligent robots, from the perspectives of engineering effort, computational efficiency, cognitive modeling and understandability. I will present some current research focused on understanding the roles of learning in runtime-rational agents with the ultimate aim of constructing general-purpose human-level intelligent robots.
  • Bio: Leslie is a Professor at MIT. She has an undergraduate degree in Philosophy and a PhD in Computer Science from Stanford, and was previously on the faculty at Brown University. She was the founding editor-in-chief of the Journal of Machine Learning Research. Her research agenda is to make intelligent robots using methods including learning, planning, and reasoning about uncertainty.  She was doing agentic AI way before it was cool.

• Prof. Diego Calvanese, Free University Bozen-Bolzano
  • Giugno 30 – 5:00 PM (CEST)
  • Title: Lightweight Description Logics with Heavyweight Impact
  • Abstract: Since the mid-2000s, research in knowledge representation has focused on identifying ontology languages that are expressive enough to model complex real-world domains, yet simple enough to support efficient automated reasoning. Two families of such languages, known as DL-Lite and EL, have played a central role in this development. Both are Description Logics, which provide the logical foundations for OWL 2, the W3C standard ontology language, and offer different, carefully tuned ways of making reasoning scalable. DL-Lite was designed to support efficient ontology-mediated query answering over large data and (virtual) knowledge graphs, by relying on query-rewriting techniques. EL, on the other hand, supports consequence-based reasoning with polynomial-time complexity, a feature that has proven crucial for handling large-scale biomedical ontologies and knowledge graphs, such as SNOMED CT. Over the past two decades, DL-Lite and EL have shaped research in Description Logics, with a profound impact in both theory and practice. They have triggered extensive investigations of the trade-off between expressive power and complexity of inference across a wide range of reasoning tasks beyond satisfiability and query answering. They have also influenced the design of practical ontology languages, providing the basis for the QL and EL profiles of OWL 2, and supporting applications in biomedical knowledge representation, ontology-based data access, data integration, and the semantic enrichment of knowledge graphs. In this talk, we revisit the theoretical foundations of these logics, examine their role in defining tractable fragments of Description Logics, and discuss how their principles continue to influence research and applications at the intersection of ontologies, knowledge graphs, and scalable AI reasoning.
  • Bio: Diego Calvanese is a Full Professor at the Faculty of Engineering, Free University of Bozen-Bolzano (unibz), Italy, where he serves as Head of the Institute of Computer Science and Artificial Intelligence. He carries out foundational and applied research in Artificial Intelligence and Databases, with a focus on Virtual Knowledge Graphs for data management and integration, Semantic Technologies, and Data-aware Processes, and his research is widely cited. He has been Program Chair of PODS 2015, General Chair of ESSLLI in 2016 and 2021, Program co-Chair of KR 2020, and Program Chair of IJCAI-ECAI 2026. He is a Fellow of ACM, EurAI, and AAIA. From 2020 to 2024, he was a Wallenberg Guest Professor in Artificial Intelligence for Data Management at Umeå University, Sweden. He is also the co-founder and past President of Ontopic, the first spin-off company of unibz, developing cutting-edge technologies and solutions for data management and integration, which was established in 2019 and acquired by Digital Science in 2026.