BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Expressivity and Generalisation of Graph Neural Networks
DTSTART:20261116T123000Z
DTEND:20261116T133000Z
DTSTAMP:20261119T172300Z
UID:indico-event-17769@indico.math.cnrs.fr
DESCRIPTION:Speakers: Johannes Lutzeyer (Ecole Polytechnique)\n\nGraph Neu
 ral Networks (GNNs) have celebrated many academic and industrial successes
  in the past years\; providing a rich ground for theoretical analysis and 
 achieving state-of-the-art results in several learning tasks. A major line
  of research in this domain has focused on improving the expressivity of G
 NNs measured by their ability to distinguish non-isomorphic graphs\, which
  has not always led to higher performing models in practice. In this talk\
 , I will give an accessible introduction to GNNs and their expressivity. I
  will then share recent work (ICLR'26) in which we relate the generalisati
 on and expressivity of GNNs. In particular\, we prove a PAC-Bayes bound on
  the generalisation error of expressive GNNs\, which allows us to show tha
 t more expressive GNNs may generalise worse unless their increased complex
 ity is balanced by a sufficiently large training set or reduced distance b
 etween training and test graphs. To conclude\, I will aim to provide a qui
 ck overview of related work we have been doing in recent years on the gene
 ralisation and robustness of GNNs.\n \n\nhttps://indico.math.cnrs.fr/even
 t/17769/
LOCATION:Salle Fokko du Cloux\, bâtiment Braconnier (La Doua)
URL:https://indico.math.cnrs.fr/event/17769/
END:VEVENT
END:VCALENDAR
