Nicolas Tremblay, Simon Barthelmé
Date(s) : 13/03/2020 iCal
14 h 00 min - 15 h 00 min
Smoothing is one way to estimate an underlying graph signal from noisy measurements. This operation can be explicitly written in terms of a simple graph filtering operation. However, on large graphs,exact filtering becomes prohibitive computationally and approximate methods are necessary. Two classical options are polynomial approximations and conjugate gradient methods. We will discuss a novel approach, based on recent results on random spanning forests,thereby uncovering another elegant link between the spectral information of a graph and random processes defined on it.
Reference: Yusuf Y. Pilavci, Pierre-Olivier Amblard, Simon Barthelmé, Nicolas Tremblay « Smoothing graph signal via random spanning forests » https://arxiv.o