Intervenants

Fabian M. Suchanek (Télécom Paris)  - "A hitchhiker’s guide to Ontology", le 27 mars 2024, de 10h à 11h.

Résumé:

Language Models have brought major breakthroughs in natural language processing. Notwithstanding this success, I will show that certain applications still need symbolic representations. I will then show how different methods (language models and others) can be harnessed to build such symbolic representations. I will also introduce our main project in this direction, the YAGO knowledge base. I will then talk about the incompleteness of knowledge bases. We have developed several techniques to estimate how much data is missing in a knowledge base, as well as rule mining methods to derive that data. I will then present our work on efficient querying of knowledge bases. Finally, I will talk about applications of knowledge bases in the domain of speech analysis and the digital humanities, as well as about our methods for explainable AI.

Biographie: 

Fabian M. Suchanek is a full professor at the Telecom Paris University in France. He obtained   his PhD at the Max-Planck Institute for Informatics under the supervision of Gerhard Weikum.   In his thesis, Fabian developed inter alia the knowledge base YAGO, one of the largest public   general-purpose knowledge bases, which earned him a honorable mention of the SIGMOD   dissertation award. Fabian was a postdoc at Microsoft Research in Silicon Valley (reporting to   Rakesh Agrawal) and at INRIA Saclay/France (reporting to Serge Abiteboul). He continued as   the leader of the Otto Hahn Research Group “Ontologies” at the Max-Planck Institute for   Informatics in Germany. Since 2013, he is an associate professor at Télécom Paris University   in France, and since 2016 a full professor. Fabian teaches classes on the Semantic Web,   Information Extraction and Knowledge Representation in France, in Germany, and in Senegal.   With his students, he works on information extraction, rule mining, ontology matching, and   other topics related to large knowledge bases. He has published more than 100 scientific   articles, among others at ISWC, VLDB, SIGMOD, WWW, CIKM, ICDE, and SIGIR, and his   work has been cited more than 16,000 times. His 2007 paper on YAGO won the Test of Time   Award of the WWW 2018 conference.

Gabriele Ciravegna (DBDMG team, Politecnico di Torino) - "Concept-based Explainable AI (C-XAI): a paradigm shift in XAI", le 27 mars de 14h30 à 15H30.

Résumé :

The field of eXplainable Artificial Intelligence (XAI) emerged in response to the growing need for more transparent and reliable models. However, using raw features to provide explanations has been discussed in several works lately, advocating for more user-understandable explanations.  To address this issue, a wide range of papers proposed Concept-based XAI (C-XAI) methods have been published in recent years. Nevertheless, a unified categorization and precise field definition are still missing. In this talk, we will try to fill the gap through a review of C-XAI approaches [1]. We will identify and define what is a concept and a concept-based explanation. We will then provide guidelines for selecting a suitable category based on the application context. Additionally, we will analyse two interesting proposals focusing on supervised concept-based model [2,3], proposing to employ embedded representation of the concepts to provide interpretable predictions.

Biographie:

Gabriele Ciravegna is a researcher at Politecnico di Torino working on Trustworthy and Explainable AI. Previously, he was a postDoc researcher at Inria and Univeristé Côte d'Azur. He got his PhD with Prof. Marco Gori at Univeristà di Siena.  His research area focuses on Trustworthy and Explainable AI. He is in the programming commitee of the AAAI, IJCAI and COLLAs conferences and serves as reviewer for the Artificial Intelligence Journal and the IEEE Transaction on Neural Networks and Learning Systems.  E-mail: gabriele.ciravegna@polito.it, website: https://dbdmg.polito.it/dbdmg_web/gabriele-ciravegna/, Google scholar: https://scholar.google.com/citations?user=-k0H_DIAAAAJ. 

Wendy Mackay (Inria Saclay, Université Paris-Saclay)- "Les partenariats humain-machine : interagir avec l’intelligence artificielle", le 28 mars de 09h à 10h. 

Résumé :

Comment pouvons-nous concevoir des « partenariats humain-machine » qui tirent le meilleur parti des compétences humaines et des capacités des systèmes ? La recherche en intelligence artificielle est généralement mesurée par l'efficacité des algorithmes, alors que la recherche en interaction humain-machine se concentre sur l'amélioration des compétences humaines. Dans cette présentation je défend l'idée que de meilleurs algorithmes d'intelligence artificielle ne sont ni nécessaires ni suffisants pour créer des systèmes humain-machine plus efficaces. Nous devons plutôt nous concentrer sur les détails de l'interaction et réussir à trouver un équilibre entre la simplicité de l’interaction et la puissance d’expression du système. Après avoir présenté nos travaux sur les « théories génératives de l'interaction » je décrirai plusieurs projets qui illustrent de nouveaux systèmes interactifs intelligents que les utilisateurs trouvent découvrables, expressifs et appropriables. 

Biographie:

Wendy Mackay est Directrice de Recherche, Classe Exceptionnelle, à Inria Saclay et Professeur Attaché à l’Université Paris-Saclay. Elle dirige l'équipe de recherche Ex-Situ en Interaction Humain-Machine, au Laboratoire Interdisciplinaire en Sciences du Numérique (LISN) commune commune à Inria Saclay et l'Université Paris-Saclay et CNRS. Elle est titulaire de la Chaire Annuelle d’Informatique et Sciences Numériques 2021-2022 du Collège de France. Elle a obtenu son doctorat au MIT et est docteur Honoris Causa de l'Université d’Aarhus, ACM Fellow et membre de l'ACM CHI Academy. Elle est lauréate d’une ERC Advanced Grant sur les partenariats humain-machine. Elle a publié plus de 200 articles de recherche dans le domaine de l'interaction humain-machine. Ses travaux combinent des contributions théoriques, empiriques et conceptuelles, avec pour objectif de repenser l'interaction entre les utilisateurs humains et les systèmes intelligents. Elle a introduit de nombreuses méthodes de conception et d'évaluation multidisciplinaires et étudie actuellement comment concevoir des systèmes où les utilisateurs et les agents intelligents se partagent le contrôle, à la fois de manière interactive et sur de longues périodes, afin d'éviter la déqualification et d'accroître au contraire les capacités humaines. Les domaines d'application actuels vont du travail avec des professionnels de la création (chorégraphes, designers et musiciens) à des environnements critiques en matière de sécurité (cockpits intelligents et salles de contrôle d'urgence).

Pierre Marquis (Cril, Lens) - "Knowledge Representation for eXplainable Artificial Intelligence", le 28 mars de 14h30 à 15h30.

Résumé :

The rise of Machine Learning (ML) through its various high-stake applications has stimulated the development of eXplainable Artificial Intelligence (XAI) for the past five years. The main goal of XAI is to make ML models less opaque and more trustable. In this lecture, I will show how Knowledge Representation (KR) can prove useful to reach this goal. In particular, I will show how concepts and methods developed in the KR field can be leveraged to address some XAI issues.

Biographie:

Pierre Marquis (https://www.cril.fr/~marquis/) is a Professor of Computer Science at Artois University and an honorary member of Institut Universitaire de France. His research topics concern knowledge representation and automated reasoning. Currently, Pierre Marquis is the head of the CRIL lab at Lens (CRIL - Lens Computer Science Research Center - is a joint lab associated with Artois University and CNRS). Pierre Marquis also heads a research and teaching chair in AI, centered on XAI issues.