A Machine Learning Tour in Network Science
Résumé
Graphs, also known as networks, are widely used data structures for modeling complex
systems in various fields, from the social sciences to biology and engineering. The
strength lies in their ability to represent relationships between entities, such as friendships in social networks or protein interactions in biological networks. In addition to their modeling capabilities, graphs offer a mathematical framework to analyze, understand, and make predictions from real-world datasets. This HDR manuscript presents part of my research contributions to the field of graph representation learning and its applications in network science, focusing on the work conducted after joining CentraleSupélec, Université Paris-Saclay in 2017. The first part of the manuscript explores structure-preserving node embedding techniques that leverage random walks. The second part addresses the challenge of developing graph representation learning models for multilayer and heterogeneous graphs, with a specific focus on applications arising from the domain of computational biology. The third part delves into the design of expressive and explainable graph neural network models. Finally, the last part investigates the application of graph representation learning to tackle the well-studied problems of social influence learning and maximization in complex networks.
Origine | Fichiers produits par l'(les) auteur(s) |
---|