jeudi 11 juin 2026

par BUTEL Nathalie

Pierre Monnin, chercheur au sein du pôle SPARKS dans l'équipe Wimmics, donnera un séminaire jeudi 11 juin 2026 à 14h00 dans la salle 007 (Algorithmes - Bâtiment Euclide B).

 

Courte biographie : 

My research focuses on neuro-symbolic AI with and for knowledge graphs. I investigate interactions between domain knowledge in knowledge graphs and different forms of reasoning in a neuro-symbolic perspective (e.g., injection of domain knowledge in Machine Learning models, analogical reasoning). In particular, I study such interactions in the context of the lifecycle of knowledge graphs (construction, matching, refinement, mining, knowledge discovery), and their usage in downstream applications (e.g., recommender systems, explainable AI). My work involves both theoretical and applied perspectives, often in interdisciplinary settings (e.g., biomedical, educational domains). 

 

Titre : Neurosymbolic AI for and with Knowledge Graphs

 

Résumé : 

Neuro-symbolic AI is often regarded as the 3rd wave of AI, aiming to integrate symbolic and neural approaches to combine their strengths and address their respective limitations. Stemming from symbolic AI, knowledge graphs (KGs) are a key representation for structured knowledge on the Web. They also constitute a natural meeting point between symbolic reasoning and machine learning, with their logic-based framework and the availability of efficient Machine Learning models to manipulate them. In this talk, I present an overview of my research on neuro-symbolic AI for and with Knowledge Graphs, spanning the full KG lifecycle. In particular, I review past work on KG construction and refinement, including analogy-based pruning methods for scalable KG bootstrapping, and knowledge-aware evaluation metrics and loss functions for link prediction. I conclude by outlining future directions that position Knowledge Graphs as a pivot structure for neuro-symbolic AI, enabling the integration of heterogeneous data, AI methods, and human-AI interactions.