– Start at 10:00
– 9h45 – 10:30 Michèle Sebag “Slides_Explain_Roman”
In critical contexts, the decision makers need to identify the options best accommodating the expert criteria, and they rely on Multi-Criteria Decision Making (MCDM) systems. These systems, most often manually designed by experts, are required to be accountable and explainable to let the decision maker fully appreciate the context and see whether elements of decision are most relevant in this context.
The framework of Choquet integrals is long used in the MCDM setting. The talk will show how to leverage neural nets to automatically learn from data hierarchical Choquet integrals from data, from the construction of utility functions from raw sensors to interactions between criteria.
Joint work: Roman Bresson; Eyke Hullermeier; Johanne Cohen; Christophe Labreuche. (IJCAI 2019; KR 2021).
– 11h10 – 11h45 : Luis Galarraga “S-LIME: Reconciling Locality and Fidelity in Linear Explanations”
[presented at IDA 2022]
https://link.springer.com/chapter/10.1007/978-3-031-01333-1_9
– 11h45 – 12h30 : Victor Guyomar “VCNet: A self-explaining model for realistic counterfactual get neration”hyaiai_pres_victor
– 12h30 Lunch at the Touareg (https://restaurant-letouareg.fr/fr)
– 14h00 – 15h00 (remotely): Georgios Zervakis “Editing Factual Knowledge in Pretrained Language Models: a review” Editing_Knowledge_in_Language_Models__slides___HyAIAI_
– 15h00 – 15h45 : Carlos Cotrini “Privacy preserving negotiations”
References from Miguel:
-Philippe Balbiani, Hans van Ditmarsch, Saúl Fernández González:
Asynchronous Announcements. ACM Trans. Comput. Log. 23(2): 10:1-10:38 (2022)
-Hans van Ditmarsch, Wiebe van der Hoek, Louwe B. Kuijer:
The logic of gossiping. Artif. Intell. 286: 103306 (2020)
– 16h05 – 17h00 : Mohit Mittal “Inspecting and Debugging a VQA Systems via post-hoc explanations”
– 17h00 – 17h15 : Discussion about the challenge
Nothing much to say, no news from the data providers.
