On Positive-Unlabeled Classification in GAN

This paper defines a positive and unlabeled classification problem for standard GANs, which then leads to a novel technique to stabilize the training of the discriminator in GANs. Traditionally, real data are taken as positive while generated data are negative. This positive-negative classification criterion was kept fixed all through the learning process of the discriminator without considering the gradually improved quality of generated data, even if they could be more realistic than real data at times. In contrast, it is more reasonable to treat the generated data as unlabeled, which could be positive or negative according to their quality. The discriminator is thus a classifier for this positive and unlabeled classification problem, and we derive a new Positive-Unlabeled GAN (PUGAN). We theoretically discuss the global optimality the proposed model will achieve and the equivalent optimization goal. Empirically, we find that PUGAN can achieve comparable or even better performance than those sophisticated discriminator stabilization methods.

1. Introduction

Recently, deep generative models have received remarkable achievements in image generation tasks [14], [22], [25], [5]. As a representative generative model, GANs [5] approximated a target distribution via playing a min-max game. In the standard framework of GAN [5], [23], a generator takes noise vectors from a prior distribution (e.g. Gaussian distribution and normal distribution) as the input and tends to produce data that follows the distribution of the reference natural images, while the discriminator aims to distinguish the generated data from the real data. Various GAN methods have been developed in many interesting applications. For example, in the image-to-image translation task, generators in GANs map the input image to output image. Representative methods include Pix2pix [10] over paired training images and cycleGAN [30] in an unsupervised way.