Litcius/Paper detail

Two Steps at a Time---Taking GAN Training in Stride with Tseng's Method

Axel Böhm, Michael Sedlmayer, Ernö Robert Csetnek, Radu Ioan Boţ

2022SIAM Journal on Mathematics of Data Science23 citationsDOIOpen Access PDF

Abstract

Motivated by the training of generative adversarial networks (GANs), we study methods for solving minimax problems with additional nonsmooth regularizers. We do so by employing monotone operator theory, in particular the forward-backward-forward method, which avoids the known issue of limit cycling by correcting each update by a second gradient evaluation and does so requiring fewer projection steps compared to the extragradient method in the presence of constraints. Furthermore, we propose a seemingly new scheme which recycles old gradients to mitigate the additional computational cost. In doing so we rediscover a known method, related to optimistic gradient descent ascent. For both schemes we prove novel convergence rates for convex-concave minimax problems via a unifying approach. The derived error bounds are in terms of the gap function for the ergodic iterates. For the deterministic and the stochastic problem we show a convergence rate of $\mathcal{O}({1}/{k})$ and $\mathcal{O}({1}/{\sqrt{k}})$, respectively. We complement our theoretical results with empirical improvements in the training of Wasserstein GANs on the CIFAR10 dataset.

Topics & Concepts

Iterated functionMinimaxErgodic theoryMonotone polygonApplied mathematicsLimit (mathematics)Convergence (economics)Computer scienceMathematical optimizationStochastic gradient descentComplement (music)MathematicsRate of convergenceAlgorithmPure mathematicsArtificial intelligenceMathematical analysisArtificial neural networkChannel (broadcasting)GeometryGenePhenotypeChemistryBiochemistryComplementationComputer networkEconomicsEconomic growthGenerative Adversarial Networks and Image SynthesisStochastic Gradient Optimization TechniquesSparse and Compressive Sensing Techniques