Visitez notre page

 

 

 

 

 

 


A Consistent Algorithm for Semi-Supervised Graph Learning

Séminaire
Organisme intervenant (ou équipe pour les séminaires internes)
Télécom Paris
Nom intervenant
Thomas Bonald
Résumé

We consider a semi-supervised learning task in a graph, where the labels of the nodes must be predicted from the labels known for a few nodes only. This problem can be solved through a process of heat diffusion, the temperature of each node at equilibrium being used as a score function for each label. In this paper, we prove that this popular algorithm is in fact not consistent unless the temperatures of the nodes at equilibrium are centered before scoring. This crucial step does not only make the algorithm provably consistent but brings significant performance gains on real graphs.

Short bio:
Thomas Bonald is a professor in Computer Science at Télécom Paris, Institut Polytechnique de Paris. He obtained a PhD in Applied Mathematics from Ecole Polytechnique in 1999. He has then worked for more than 10 years on stochastic models and performance evaluation, both at Inria and Orange labs. His current research interests are in graph analysis, knowledge bases and natural language processing. His is the co-founder of scikit-network, an open-source Python library for the analysis of large graphs.
Lieu
Amphi C2.0.037
Date du jour
Date de fin du Workshop