Graph signal coarsening: Dimensionality reduction in irregular domain

Abstract

Graph signal coarsening is a kind of dimensionality reduction in irregular domain. Given a graph signal, it aims to simultaneously obtain a coarser version of the graph and a coarsened signal on the new graph. In this work, we explore the design space for the graph signal coarsening problem and show that solutions can be split into four categories. We propose an effective method that uses a successive approach and spectral-domain-based signal coarsening for solving the problem, which is the first that falls into one of the four categories. Experiments are conducted to show the effectiveness of the proposed method.