Statistical learning on large scale graphs

statistical learning on LARge scale GRaphs (LARGR)

9-10 Mar 2023 Lille (France)

Outline

In many problems in science and engineering, we are given access to data in the form of pairwise relationships between a set of n objects. These pairwise relationships naturally lead to an underlying graph, with nodes corresponding to the objects, and edges encoding the pairs of objects for which information is available. The goal, then, is to learn an underlying latent structure associated with the objects using the available pairwise data. Such problems arise in a wide range of applications, such as computer vision, recommendation systems, sports tournaments, biology, social sciences, to name a few. In many cases, the data is high dimensional with n of the order of thousands, or even millions (e.g. social networks), thus rendering the algorithmic task challenging from a practical point of view. The past decade has witnessed an impressive array of results – both from a theoretical and practical perspective – with several foundational results, as well as recent breakthroughs achieved by Graph Neural Networks in many important applications.

This workshop will cover topics under the scope of such learning problems on large scale networks, including topics such as clustering, ranking, graph matching, graph neural networks, graph signal processing and dynamic processes on networks. It will feature a combination of long talks (1 hour each, by invited speakers), and contributed talks (20 mins each, by PhD students/post-docs/young researchers). The talks will include a general introduction to make them accessible to a wider audience.

Online user: 2

Privacy