arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1809.02145cs.LGcs.AIstat.ML
收藏

超越发散最小化的GANs

GANs beyond divergence minimization

Lady Davis Institute(戴维斯夫人研究所)

Alexia Jolicoeur-Martineau

摘要:

生成对抗网络(GANs)可以被解释为两个玩家之间的对抗游戏,一个是GAN D,另一个是生成器G,其中D学习从虚假数据中区分真实数据,G通过“愚弄”D认为虚假数据实际上是真实数据来学习生成真实数据。目前,主流观点认为,G实际上是通过最小化发散来学习的,因为当D是最优的时,一般目标函数是发散。然而,由于理论与实践之间的不一致,这一观点受到了挑战。在本文中,我们讨论了与G的大多数损失函数相关的性质(例如,饱和/非饱和f-GAN、LSGAN、WGAN等)。我们证明了这些损失函数不是发散的,并且不具有与发散预期相同的均衡。这表明G不需要最小化与D最大化相同的目标函数,也不需要在用假数据交换真实数据(非饱和GAN)之后最大化D的目标,而是可以使用广泛的可能损失函数来学习生成真实数据。我们通过两个单独且独立的D最大化和G最小化步骤来定义GANs。我们将生成器步骤推广到四类新的损失函数,其中大多数是实际的分歧(而传统的G损失函数不是)。我们测试了各种各样的损失函数,从这四个类的合成数据集和CIFAR-10。我们观察到,大多数损失函数收敛良好,并提供可比的数据生成质量的非饱和GAN,LSGAN和WGAN-GP生成器损失函数,无论我们使用发散或非发散。这些结果表明,GANs不符合发散最小化理论,并且形成了比之前假设的更广泛的模型范围。

英文摘要:

Generative adversarial networks (GANs) can be interpreted as an adversarial game between two players, a discriminator D and a generator G, in which D learns to classify real from fake data and G learns to generate realistic data by "fooling" D into thinking that fake data is actually real data. Currently, a dominating view is that G actually learns by minimizing a divergence given that the general objective function is a divergence when D is optimal. However, this view has been challenged due to inconsistencies between theory and practice. In this paper, we discuss of the properties associated with most loss functions for G (e.g., saturating/non-saturating f-GAN, LSGAN, WGAN, etc.). We show that these loss functions are not divergences and do not have the same equilibrium as expected of divergences. This suggests that G does not need to minimize the same objective function as D maximize, nor maximize the objective of D after swapping real data with fake data (non-saturating GAN) but can instead use a wide range of possible loss functions to learn to generate realistic data. We define GANs through two separate and independent D maximization and G minimization steps. We generalize the generator step to four new classes of loss functions, most of which are actual divergences (while traditional G loss functions are not). We test a wide variety of loss functions from these four classes on a synthetic dataset and on CIFAR-10. We observe that most loss functions converge well and provide comparable data generation quality to non-saturating GAN, LSGAN, and WGAN-GP generator loss functions, whether we use divergences or non-divergences. These results suggest that GANs do not conform well to the divergence minimization theory and form a much broader range of models than previously assumed.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑