Boundary-Seeking Generative Adversarial Networks

TL;DR

Boundary-Seeking GANs (BGANs) introduce a novel training method for GANs that effectively handles discrete data by using discriminator-based importance weights, improving training stability and performance.

Contribution

The paper proposes a boundary-seeking approach for GANs that enables training with discrete data and enhances stability for continuous data.

Findings

Effective training of GANs with discrete data demonstrated on images and language.

Improved training stability for continuous data on Celeba, LSUN, and Imagenet.

Boundary-seeking objective enhances discriminator-guided importance weighting.

Abstract

Generative adversarial networks (GANs) are a learning framework that rely on training a discriminator to estimate a measure of difference between a target and generated distributions. GANs, as normally formulated, rely on the generated samples being completely differentiable w.r.t. the generative parameters, and thus do not work for discrete data. We introduce a method for training GANs with discrete data that uses the estimated difference measure from the discriminator to compute importance weights for generated samples, thus providing a policy gradient for training the generator. The importance weights have a strong connection to the decision boundary of the discriminator, and we call our method boundary-seeking GANs (BGANs). We demonstrate the effectiveness of the proposed algorithm with discrete image and character-based natural language generation. In addition, the boundary-seeking objective extends to continuous data, which can be used to improve stability of training, and we demonstrate this on Celeba, Large-scale Scene Understanding (LSUN) bedrooms, and Imagenet without conditioning.