GL-Coarsener: A Graph representation learning framework to construct coarse grid hierarchy for AMG solvers

  • 2020-11-19 17:49:09
  • Reza Namazi, Arsham Zolanvari, Mahdi Sani, Seyed Amir Ali Ghafourian Ghahramani
  • 9

Abstract

In many numerical schemes, the computational complexity scales non-linearlywith the problem size. Solving a linear system of equations using directmethods or most iterative methods is a typical example. Algebraic multi-grid(AMG) methods are numerical methods used to solve large linear systems ofequations efficiently. One of the main differences between AMG methods is howthe coarser grid is constructed from a given fine grid. There are two mainclasses of AMG methods; graph and aggregation based coarsening methods. Here wepropose an aggregation-based coarsening framework leveraging graphrepresentation learning and clustering algorithms. Our method introduces thepower of machine learning into the AMG research field and opens a newperspective for future researches. The proposed method uses graphrepresentation learning techniques to learn latent features of the graphobtained from the underlying matrix of coefficients. Using these extractedfeatures, we generated a coarser grid from the fine grid. The proposed methodis highly capable of parallel computations. Our experiments show that theproposed method's efficiency in solving large systems is closely comparablewith other aggregation-based methods, demonstrating the high capability ofgraph representation learning in designing multi-grid solvers.

 

Quick Read (beta)

loading the full paper ...