Qu'est-ce que mon GNN capture vraiment ? Exploration des représentations internes d'un GNN

From LRDE

Abstract

While existing GNN's explanation methods explain the decision by studying the output layer, we propose a method that analyzes the hidden layers to identify the neurons that are co-activated for a class. We associate to them a graph.


Bibtex (lrde.bib)

@InProceedings{	  veyrin-forrer.22.egc,
  author	= {Luca {Veyrin-Forrer} and Ataollah Kamal and Stefan Duffner
		  and Marc Plantevit and C\'eline Robardet},
  title		= {Qu'est-ce que mon {GNN} capture vraiment ? {E}xploration
		  des repr\'esentations internes d'un {GNN}},
  booktitle	= {Extraction et Gestion des Connaissances, {EGC} 2022,
		  Blois, France, 24 au 28 janvier 2022},
  pages		= {159--170},
  year		= {2022},
  opturl	= {http://editions-rnti.fr/?inprocid=1002725},
  note		= {In French, Best paper award},
  category	= {national},
  abstract	= {While existing GNN's explanation methods explain the
		  decision by studying the output layer, we propose a method
		  that analyzes the hidden layers to identify the neurons
		  that are co-activated for a class. We associate to them a
		  graph.}
}