Laboratoire de Recherche en Informatique (Université Paris Sud, CNRS, Université Paris Saclay) (France)

Programme

Salle : Petit Amphi (bât. 660- DIGITEO) 

Plan d'accès : https://www.lri.fr/info.pratiques.php

Programme  

9h15 - Accueil des participants

9h40 - Ouverture de la journée

9h45Présentations de l’AFIA, par Yves Demazeau (Président del'AFIA)  et d'EGC par Cyril de Runz (Membre du bureau de l'association EGC) 

10h00 - Conférencier invité : Luis Galarraga (CR INRIA à l’IRISA)  sur le thème de « Interpretability in classifiers: Hybrid techniques based on black-box classifers and interpretable control modules »
 
Résumé
Recent technological advances rely on accurate decision support systems that have been constructed as black boxes. That is, the system's internal logic is not available to the user, either for financial reasons or due to the complexity of system. This lack of explanation can lead to technical, ethical, and legal issues. For example, if the control module of a self-driving car failed at detecting a pedestrian, it becomes crucial to know why the system erred. In some other cases, the decision system may reflect unacceptable biases that can generate distrust. Recently, the European Parliament adopted the General Data Protection Regulation (GDPR), a law that introduces the idea that humans should have the possibility to obtain an explanation of a decision proposed by automated processing, and to challenge that proposed decision. For all these reasons, multiple research approaches provide comprehensible explanations for traditionally accurate but black-box-like machine learning classifiers such as neural networks, random forests, and support vector machines. In this talk I will survey the most relevant research efforts in "opening a black-box classifier" and I will explain why it is challenging task (the trade-off between interpretability and accuracy).
 
 
11h00 - Learning How to Correct a Knowledge Base -Thomas Pellissier Tanon, Camille Bourgaux, Fabian Suchanek    (https://thomas.pellissier-tanon.fr/papers/2019-WWW-corhist.pdf)

11h30 - Towards Interactive Causal Relation Discovery Driven by an Ontology - Melanie Munch, Juliette Dibie-Barthélemy, Pierre-Henri Wuillemin, Cristina Manfredotti  (https://hal.archives-ouvertes.fr/hal-01823862)

12h00-13h30 pause déjeuner (dans la salle BIBLIOTHEQUE - Bâtiment 650-Ada Lovelace

13h30 -  Enhancing the interlinking quality: A structure-based approach - Pierre-Henri Paris, Fayçal Hamdi, Samira Si-Said Cherfi (https://hal.archives-ouvertes.fr/hal-01180919/document)

14h00 - Scalable Schema Discovery for RDF Data - Redouane Bouhamoum - Zoubida Kedad and Stéphane Lopes (https://hal.archives-ouvertes.fr/hal-01757028

14h30 - Revealing the Conceptual Schemas of RDF Datasets Subhi Issa - Pierre-Henri Paris, Fayçal Hamdi and Samira Si-Said Cherfi (https://cedric.cnam.fr/index.php/publis/article/IPH19)

15h00-15h30 - pause

15h30 - Using Redescriptions and Formal Concept Analysis for Mining Definitions in Linked Data - Justine Reynaud, Yannick Toussaint, Amedeo Napoli article

16h00 - Linkex: A Tool for Link Key Discovery Based on Pattern Structure - Nacira Abbas, Jérôme David, Amedeo Napoli Linked

16h30 - Clôture de la journée
 
 
Personnes connectées : 1