Research Activities
My current research lies at the interface of graph based representation of data and the definition of machine learning methods operating on graphs. In my recent works, I focused on :
- learning edit costs for graph edit distance
- improve pooling layer for GNN : pooling
- define methods to alleviate the problem of preimage of graphs
- apply graph based machine learning methods to chemoinformatics
An up to date list of publications can be found at Google Scholar.
All publications are available for download on HAL. If you still need any pdf, please drop a mail.
Most of papers have been implemented. Please check my github.
My ORCID record.
Last papers
A paper on how we can use Normalizing Flows to build pre image free machine learning models operating on graphs -> Paper in GbR
Also a paper in GbR on the combination of MIVS and pooling in GNN -> link to paper
A new open access paper on the stability of graph edit distance heuristics. -> here
Projects
FAMOUS An ANR project on the definition of fair prediction models, with application on graphs. This project includes people from LIS, INT, Laboratoire Hubert Curien and EURONOVA.
CodeGNN ANR project focused on new contributions to Graph Neural Networks both on convolutions and pooling operators. Defining new GNN for temporal graphs is also an objective of this project.
AGAC Application des Graphes À la Chémoinformatique. A regional project with chemical scientists from COBRA to apply machine learning on graphs to chemical problematics.
APi Taming the Beast of the Preimage in Machine Learning for Structured Data: Signal, Image and Graph – APi
Students
I co-advised some PhD :
- Stevan Stanovic (2021-2024) : Apprentissage de la décimation de Graphes pour les GNN
This thesis aims to improve the pooling operator in graph neural networks.
- Clément Gledel (2020-2023) : Preimage Problem for Graph Data
How can we use modern graph generation techniques to improve the computation of graph preimage.
- Linlin Jia (2017-2021) : Machine learning and pattern recognition in chemoinformatics.
Definition of methods based on graph representations for chemoinformatics. Lead to a contribution on the learning of graph edit distance costs and a preimage method based on graph median.
Guillaume Renton (2017-2021) : Réseaux de Neurones sur Graphes : Analyse et Contributions
This thesis bring some contributions to the analysis of spectral and spatial views of GNN’s convolutions. It also define a method to include edge information within the message passing framework.