Sampling nodes on graphs using Determinantal Point Processes

08.11.2023, 13:00  –  Haus 9, Raum 0.17 und Zoom
Forschungsseminar Diskrete Spektraltheorie

Simon Barthelmé

Abstract: Joint work with Nicolas Tremblay, Pierre-Olivier Amblard

Determinantal Point Processes (DPPs) are an important class of models of random sets that arise in many areas of mathematics and physics. DPPs produce point sets with repulsion, corresponding for instance to models of particles that move randomly while repelling one another. In this talk I'll explain how DPPs can be used to sample nodes that are well-distributed over a given graph. This is useful for instance when one wishes to measure a signal at a subset of nodes, while being able to reconstruct the signal globally over the graph. If time allows I'll talk about how DPPs on graphs relate to DPPs in space in certain classes of random graphs.