Boundary-Seeking Generative Adversarial Networks
This addresses a key limitation in GANs for researchers and practitioners working with discrete data, such as in natural language processing, by enabling stable training and generation.
The paper tackles the problem of training generative adversarial networks (GANs) on discrete data, which normally fails due to differentiability issues, by introducing boundary-seeking GANs (BGANs) that use importance weights from the discriminator to provide a policy gradient for the generator, demonstrating effectiveness on discrete image and character-based natural language generation tasks.
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.