Presentations

July 23, 2026 - Trustworthy and Responsible AI

16h30-18h00 CEST
A Metrics-Driven Approach to Trust in the Gen²KGBot and Q²Forge Projects
Yousouf Taghzouti, Postdoctoral researcher, Côte d’Azur University (France), and member of the Wimmics team at Inria (France)

Abstract: Integrating Large Language Models into Question Answering over Knowledge Graphs bridges the gap between natural language (NL) and structured data, but the non-deterministic nature of generative AI often compromises query reliability. This presentation introduces a unified, open-source ecosystem developed within the Wimmics research team to generate, refine, and quantitatively evaluate Text-to-SPARQL translations. We showcase Gen²KGBot¹, a modular framework that translates NL queries into SPARQL using progressive context injection and self-correction loops; Q²Forge², an interactive companion web application for minting competency questions and refining query-question training datasets; and T2S-Metrics³, a standardized Python evaluation toolkit that decouples metric specification from implementation to measure lexical, structural, and execution-based system performance (such as BLEU and AnswerSetF1). Together, this triad of tools demonstrates how trust in generative semantic query systems can be systematically benchmarked and verified, transforming “trust” from a subjective goal into a measurable, reproducible metric.

July 24, 2026 - Industry Use Cases

13h00-13h45 CEST
Trustworthy Industrial AI : Welding and anomaly detection use cases
Yann Gasté, Secretary-General European Trustworthy AI Association (France), and Product Owner IRT SystemX (France)

13h45-14h15 CEST
Non-road mobile machines: their data, digital twins, and autonomy
Victor Zhidchenko, Postdoctoral Researcher, D.Sc. (Tech.), Laboratory of Intelligent Machines, LUT University (Finland)

Abstract: Non-road mobile machinery (NRMM) represents a distinct category of machines with interesting properties. These are complex and expensive devices that often work 24/7 in an unstructured environment and harsh conditions. They usually operate in remote locations with poor (if any) network connectivity and perform the tasks that are difficult to formalize. Since the machines are big, the cost of any misoperation can be substantial. These features complicate NNRM digitalization and automation. On the other hand, NNRM has similarities with other types of devices, which makes research findings transferable to different domains. From this presentation, you will know about research conducted at LUT University on creating the methods that facilitate machine design, operation, automation, maintenance, and sustainability. You can get a new perspective on what the term “robot”, “end user”, “digital twin”, or “autonomous agent” can mean. You will discover new research challenges and collaboration opportunities.

14h15-14h45 CEST
Knowledge Representation for Artificial Intelligence A Metascience Agenda
Paola Di Maio, Research Lead, Center for Systems, Knowledge Representation & Neuroscience, Ronin Institute, and Chair, W3C AI Knowledge Representation Community Group