Convolution and Decimation for Graph Neural Networks – CoDeGNN
Many objects of our today life are discrete and described by sequences of elements (strings) or more complex relationships (e.g. graphs). Inferences from these discrete objects is the field of structural pattern recognition. This research field has been limited for decades by costly (e.g. based on subgraph isomorphism) or poorly efficient metrics usually combined with limited machine learning algorithms (mainly k-nearestneighbours algorithm). Two recent breakthroughs have drastically changed this situation: Graph kernels and more recently Graph Neural Networks (GNN). This project aims at solving the current limitations of the two basic operations of GNN, namely graph convolution and graph decimation. We also plan to investigate inferences on a (yet) less investigated type of objects: sequences of graphs.
Project coordination
Luc BRUN (Groupe de recherche en Informatique, Image, Automatique et Instrumentation de Caen)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Partner
LIFAT Laboratoire d'Informatique Fondamentale et Appliquée de Tours
GREYC Groupe de recherche en Informatique, Image, Automatique et Instrumentation de Caen
LITIS LABORATOIRE D'INFORMATIQUE, DE TRAITEMENT DE L'INFORMATION ET DES SYSTÈMES - EA 4108
Help of the ANR 436,800 euros
Beginning and duration of the scientific project:
October 2021
- 48 Months
