Résumé : There is a growing need for adapting data analysis and machine
learning methods to graphs, arising in different fields such as
chemistry, biology, social science or image processing. Defining a
graph kernel makes it possible to apply a whole spectrum of machine
learning algorithms to graphs. These kernels have to respect the
structure and node/edge labels of graphs and, importantly, they have
to be efficient to compute in order to be applicable to large graphs.
This talk will give an overview of different graph comparison methods
and present a family of kernels for large graphs with discrete node
labels. Particular instances of this family scale only linearly in the
number of
edges of the graphs, and outperform state-of-the-art graph kernels on
several graph classification benchmark data sets in terms of accuracy
and runtime. At the end of the talk we will discuss the next
challenges in graph comparison.

Cet exposé se tiendra en salle C20-13, 20ème étage, Université
Paris 1, Centre Pierre Mendès-France, 90 rue de Tolbiac, 75013 Paris
(métro : Olympiades).