Approche hybride basée sur l'apprentissage automatique pour la réduction de graphes
Abstract
In this paper, we focus on graph reduction using Machine Learning techniques. Recently, the arrival of Deep Learning on graphs led to fruitful results on optimization problems with graphs. The objective of graph reduction is to obtain smaller and simpler graphs, without losing too much information, by grouping similar nodes or edges, or by sparsifying the graph by removing less important edges for the downstream task. We proposed a two steps hybrid method. The first step sparsifies the graph and in the second step, we group the nodes into supernodes while preserving the distances in the graph. Finally, we make a comparison with
the existing methods.