Kantorovich Strikes Back! Wasserstein GANs are not Optimal Transport?Download PDF

04 Jun 2022, 21:36 (modified: 14 Oct 2022, 13:03)NeurIPS 2022 Datasets and Benchmarks Readers: Everyone
Keywords: wasserstein gans, optimal transport, neural dual solvers, ray monotone transport plans
Abstract: Wasserstein Generative Adversarial Networks (WGANs) are the popular generative models built on the theory of Optimal Transport (OT) and the Kantorovich duality. Despite the success of WGANs, it is still unclear how well the underlying OT dual solvers approximate the OT cost (Wasserstein-1 distance, W1) and the OT gradient needed to update the generator. In this paper, we address these questions. We construct 1-Lipschitz functions and use them to build ray monotone transport plans. This strategy yields pairs of continuous benchmark distributions with the analytically known OT plan, OT cost and OT gradient in high-dimensional spaces such as spaces of images. We thoroughly evaluate popular WGAN dual form solvers (gradient penalty, spectral normalization, entropic regularization, etc.) using these benchmark pairs. Even though these solvers perform well in WGANs, none of them faithfully compute W1 in high dimensions. Nevertheless, many provide a meaningful approximation of the OT gradient. These observations suggest that these solvers should not be treated as good estimators of W1 but to some extent they indeed can be used in variational problems requiring the minimization of W1.
Supplementary Material: pdf
URL: https://github.com/justkolesov/Wasserstein1Benchmark
Dataset Url: https://github.com/justkolesov/Wasserstein1Benchmark
Dataset Embargo: N/A
License: MIT, see the provided link.
Author Statement: Yes
Contribution Process Agreement: Yes
In Person Attendance: No
22 Replies