Bipartite Graph Matching on GPU over Complete or Local Grid Neighborhoods

Abstract

Several schedule and assignment tasks can be modeled as a bipartite graph matching optimization, aiming to retrieve an optimal set of pairs connecting elements from two distinct sets. In this paper we investigate how to compute a weighted bipartite graph matching on Graphics Processing Units (GPUs) inspired by its low cost and increasing parallel processing power. We propose a data-parallel approach to be computed using GPUs processing kernels based on The Auction Algorithm, and data structures that allow it to be applied to problems modeled over complete bipartite graphs and also over huge bipartite graphs with connections across the neighborhood systems from two sets of 1D, 2D or 3D data grids.KeywordsBipartite GraphGraphic Processing UnitComplete Bipartite GraphLocal IndexMatching CandidateThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.