Séminaire Image: Kirchhoff Forests: Theory and Applications par Yusuf Yigit Pilavci

Nous aurons le plaisir d’écouter Yusuf Yigit Pilavci , Postdoc a l’Institut Mines Telecom Nord Europe.
Il donnera un séminaire IMAGE le mardi 05 mai 2026 à 11h en salle de séminaire F-200.

Titre : « Kirchhoff Forests: Theory and Applications »

Résumé : Many machine learning problems involve data structured as graphs, from social networks and recommendation systems to traffic and biological networks. A key tool for analyzing these graphs is the graph Laplacian, which captures important structural information but becomes expensive to compute at large scale.
In this presentation, I will introduce a probabilistic tool called Kirchhoff forests for accelerating expensive operations involving graph Laplacians, such as computing their inverse, estimating the trace of the inverse, and estimating effective resistances. This probabilistic method is based on a random process that allows us to generate graph sketches strongly connected to the graph Laplacian through various theoretical guarantees. In turn, this approach leads to efficient algorithms for applications in graph signal processing and machine learning.