Tuesday, October 6, 2026, Grenoble
As part of the Neurosymbolic Methods and Their Applications in Data Science workshop, the CNRS GDRs MaDICS and RADIA, through their joint working group RECAST, together with the Computer Science Laboratory of Grenoble (LIG), the MIAI NSRL Chair, and LabEx PERSYVAL-Lab, are co-organizing a one-day conference dedicated to neuro-symbolic methods and their applications in data science.
Open to all researchers, faculty members, PhD students, and students, this event will be an opportunity to discover recent advances in the field and exchange with international experts.
Date:
Tuesday, October 6, 2026, from 9:00 AM to 5:00 PM
Location:
Amphitheater, IMAG Building
Grenoble University Campus
150 place du Torrent
38400 Saint-Martin-d'Hères, France
The program will feature four keynote talks, a panel discussion, and a poster session:
The detailed schedule for the day will be available soon.
Confirmed speakers:
Floris Geerts (University of Antwerp, Belgium): Relational Neural Networks
Abstract. Relational data capture rich, structured dependencies that traditional learning methods struggle to exploit. Relational neural networks aim to model such data by integrating the relational reasoning principles underlying graph neural networks and graph transformers. This talk explores the theoretical foundations of these models and how their learning behaviour interacts with data heterogeneity, temporal dynamics, and schema structure. We question how to characterise what relational architectures can represent and how they learn. Finally, we outline prospects for principled foundation models that unify representation, reasoning, and transfer across diverse relational domains.
Bio. Floris Geerts is professor at the University of Antwerp, Belgium. Previously, he was a senior research fellow at the University of Edinburgh and a postdoctoral researcher at the University of Helsinki. He received his PhD in 2001 from the University of Hasselt, Belgium. His research interests include the theory and practice of databases, the study of data quality, the interaction between linear algebra, relational databases and graph neural networks, and quantum computing. He has written a book on data quality and published over 130 technical papers. His awards include three best paper awards, the PODS Alberto O. Mendelzon Test-of-Time award, an ACM SIGMOD Research Highlight Award and an ICLR outstanding paper award. He is an ACM Distinguished Member, was program chair of PODS and ICDT, the general chair of EDBT/ICDT and is currently the general chair of PODS. He served on the editorial boards of ACM TODS and IEEE TKDE, and was editor of various proceedings and special journal issues in the area of databases.
Thomas Schiex (INRAE, France): A neuro-symbolic architecture learns how to solve NP-hard puzzles: from logical games and discrete optimization to molecular design
Abstract. Deep learning in general, and Large Language Models specifically, have extreme difficulties in reliably solving hard reasoning problems, even when provided with millions of solved instances for training. Even then, introducing side constraints is usually difficult, when possible. In this talk, I will introduce a (deep) neuro-symbolic architecture that we initially designed to efficiently learn how to play one-player decision NP-complete games that reside in 2D or 3D space such as Sudoku, Futoshiki or, more seriously, computational protein design. Thanks to a dedicated loss function, this architecture is able to efficiently end-to-end learn how to predict an explicit probabilistic (or deterministic) model, conditioned by the provided 2D/3D information. This explicit representation can a posteriori be sampled, optimized or constrained by a symbolic tool, without retraining but with a possibly high inference compute cost, related to the NP-complete nature of the underlying problems.
Bio. Thomas Schiex is a Senior Research Director (DR1) at INRAE within the Applied Mathematics and Computer Science unit (MIAT) in Toulouse, France. He is a Fellow of both the European Association for Artificial Intelligence (EurAI) and the Association for the Advancement of Artificial Intelligence (AAAI). Dr. Schiex’s research bridges automated reasoning, constraint programming, and machine learning. Renowned for foundational contributions to Valued and Weighted Constraint Satisfaction Problems (Cost Function Networks) and the development of the discrete graphical model optimization solver toulbar2, his current work centers on neuro-symbolic AI. By marrying the pattern-recognition capabilities of deep neural networks with the rigorous guarantees of discrete optimization, his group develops hybrid architectures capable of solving hard reasoning problems and sampling constrained solutions. These neuro-symbolic methods are applied prominently to computational biology, notably for de novo protein and macromolecular design.
Mehwish Alam (Télécom Paris, France): Language Models and Symbolic AI
Abstract. This talk examines the evolving synergy between language models and symbolic artificial intelligence, focusing on their complementary strengths in representing and reasoning over structured knowledge. Recent advances in how language models and symbolic frameworks can benefit from each other for tasks such as knowledge completion, taxonomy refinement, and fact verification will be discussed during this talk. Particular attention is given to how these approaches enable more flexible and scalable handling of semi-structured and structured data. We further highlight applications in domain-specific settings, including biomedicine and cultural heritage, where data complexity and ambiguity pose unique challenges.
Giuseppe Marra (KU Leuven, Belgium): Neurosymbolic Concept Based Models
Abstract. As deep learning systems become increasingly powerful, understanding and controlling how they reach their predictions remains a central challenge. Concept-based models offer an alternative to post-hoc explanations by designing models whose internal representations are aligned with high-level concepts that are meaningful to humans. This makes concepts more than explanatory labels: they provide an intermediate vocabulary through which models can be inspected, corrected, and ultimately reason about their predictions. This talk will introduce the main principles behind concept-based machine learning and discuss how different architectural choices determine the interpretability, expressivity, and intervenability of these models. Particular attention will be given to the connection between concept-based learning and neurosymbolic AI. Concepts provide a natural interface between neural perception and structured reasoning, enabling models whose predictions can be expressed through compositional, rule-based, or otherwise structured mechanisms. Such structures may encode prior knowledge, but they can also be learned from data, opening a path toward systems that combine the flexibility of deep learning with explicit and human-grounded reasoning.
Bio. Giuseppe Marra is an Assistant Professor in the Declarative Languages and Artificial Intelligence (DTAI) research group at KU Leuven. His research focuses on the integration of neural computation and automated reasoning, with an emphasis on formal and probabilistic methods for neurosymbolic AI.
The event is open to researchers and students interested in artificial intelligence, data science, and neuro-symbolic approaches. Participation is free, but registration is mandatory:
Poster submission:
Participants wishing to present a poster are invited to check the corresponding box during online registration. They will be contacted later with practical details regarding the submission.
This event is co-organized by the CNRS GDRs MaDICS and RADIA, through their joint working group RECAST, together with the Computer Science Laboratory of Grenoble (LIG), the MIAI NSRL Chair, and LabEx PERSYVAL-Lab.