Score function-based natural language generation (NLG) approaches such as REINFORCE, in general, suffer from low sample efficiency and training instability problems. This is mainly due to the non-differentiable nature of the discrete space sampling and thus these methods have to treat the discriminator as a black box and ignore the gradient information. To improve the sample efficiency and reduce the variance of REINFORCE, we propose a novel approach, TaylorGAN, which augments the gradient estimation by off-policy update and the first-order Taylor expansion. This approach enables us to train NLG models from scratch with smaller batch size -- without maximum likelihood pre-training, and outperforms existing GAN-based methods on multiple metrics of quality and diversity. The source code and data are available at https://github.com/MiuLab/TaylorGAN
@article{arxiv.2011.13527,
title = {TaylorGAN: Neighbor-Augmented Policy Update for Sample-Efficient Natural Language Generation},
author = {Chun-Hsing Lin and Siang-Ruei Wu and Hung-Yi Lee and Yun-Nung Chen},
journal= {arXiv preprint arXiv:2011.13527},
year = {2020}
}
Comments
34th Conference on Neural Information Processing Systems (NeurIPS 2020), Vancouver, Canada