The Benefits of Pairwise Discriminators for Adversarial Training
This addresses instability issues in adversarial training for generative modeling, offering a more robust approach for researchers and practitioners.
The paper tackles the instability problem in adversarial training where sub-optimal discriminators can disrupt alignment even with perfect generators, by introducing pairwise discriminators that preserve alignment regardless of discriminator quality. The result includes theoretical convergence conditions and empirical demonstrations showing improved high-resolution image generation.
Adversarial training methods typically align distributions by solving two-player games. However, in most current formulations, even if the generator aligns perfectly with data, a sub-optimal discriminator can still drive the two apart. Absent additional regularization, the instability can manifest itself as a never-ending game. In this paper, we introduce a family of objectives by leveraging pairwise discriminators, and show that only the generator needs to converge. The alignment, if achieved, would be preserved with any discriminator. We provide sufficient conditions for local convergence; characterize the capacity balance that should guide the discriminator and generator choices; and construct examples of minimally sufficient discriminators. Empirically, we illustrate the theory and the effectiveness of our approach on synthetic examples. Moreover, we show that practical methods derived from our approach can better generate higher-resolution images.