Thesis defense “Deep learning techniques for graph embedding at different scales”

Speaker : Alexis Galland
Date: 17/12/2020
Time: 2:00 pm - 5:00 pm


In many scientific fields, studied data have an underlying graph or manifold structure such as communication networks (whether social or technical), knowledge graphs or molecules. A graph is composed of nodes, also called vertices, con- nected together by edges. Recently, deep learning algorithms have become state-of-the-art models in many fields and in particular in natural language processing and image analysis. It led the way to a great line of studies to generalize deep learning models to graphs. In particular, several formulations of convolutional neural networks were proposed and research is carried to develop new layers and network architectures to graphs. Those models aim at solving different tasks such as node classification, link prediction or graph classification. In this work, we study node, subgraph or graph embeddings produced by graph neural networks. These embeddings at different scales encode hierarchical represen- tations of graphs. Based on these embedding techniques, we propose new deep learning architectures to tackle node classification or graph classification tasks. We study several applications of these new techniques. For example, we study the problem of having a graph embedding invariant by node permutation and the interpretability of graph neural networks.

Here’s the details to follow it on GoToMeeting:

Participez à ma réunion depuis votre ordinateur, tablette ou smartphone. 

Vous pouvez aussi appeler à l’aide de votre téléphone. 
États-Unis: +1 (872) 240-3311 

Code d’accès: 962-144-589 

Rejoignez la réunion depuis une salle ou un système de vidéoconférence. 
Composez ou tapez : ou 
ID réunion: 962 144 589 
Ou appelez directement: 962144589@ ou 

Vous n’utilisez pas encore GoToMeeting ? Téléchargez l’application dès maintenant et soyez prêt pour votre première réunion :