NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:8347
Title:Sliced Gromov-Wasserstein


		
This paper proposes a "sliced" approximation to the Gromov-Wasserstein distance (i.e., using random projections). The analysis builds on the closed-form solution of the GW problem in one dimension and then using the slicing trick to construct an estimator in the multidimensional case. A computational implementation in PyKeops compares favorably to existing methods. The consensus among the reviewers, taking into account the authors' response, is that this is a solid contribution to the literature on computational optimal transport.