Michael Perrot
MPI - Tubingen
https://mperrot.github.io/
Date(s) : 21/02/2020 iCal
14 h 00 min - 15 h 00 min
We address the problem of learning in a framework where one does not have access to a representation of the examples nor to their pairwise similarities. Instead, we assume that only a set of comparisons between objects is available. These comparisons are statements of the form « objects A and B are more similar than objects C and D » or « object A is more similar to object B than to object C. » Such a scenario is commonly encountered in crowdsourcing applications. In this talk, we study the problems of hierarchical clustering and classification in such a comparison-based framework. In both cases, we propose new algorithms that are able to directly learn from the comparisons and come with theoretical guarantees on their performances.
Slides : https://test.i2m.univ-amu.fr/seminaires_signal_apprentissage/Slides/Marseille_21.02.2020.pdf
Emplacement
Site Nord, CMI, Salle de Séminaire R164 (1er étage)
Catégories