Good Semi-supervised Learning that Requires a Bad GAN
This addresses a fundamental issue in semi-supervised learning for researchers, providing theoretical insights and empirical improvements.
The paper tackles the problem of understanding why semi-supervised learning with GANs requires a bad generator and proposes a theoretical definition for a preferred generator, resulting in state-of-the-art performance on multiple benchmark datasets.
Semi-supervised learning methods based on generative adversarial networks (GANs) obtained strong empirical results, but it is not clear 1) how the discriminator benefits from joint training with a generator, and 2) why good semi-supervised classification performance and a good generator cannot be obtained at the same time. Theoretically, we show that given the discriminator objective, good semisupervised learning indeed requires a bad generator, and propose the definition of a preferred generator. Empirically, we derive a novel formulation based on our analysis that substantially improves over feature matching GANs, obtaining state-of-the-art results on multiple benchmark datasets.