TaylorGAN: Neighbor-Augmented Policy Update for Sample-Efficient Natural Language Generation
This work improves the sample efficiency and stability of GAN-based natural language generation models, which is a problem for researchers and practitioners developing text generation systems.
This paper addresses the low sample efficiency and training instability in score function-based natural language generation (NLG) by introducing TaylorGAN. TaylorGAN improves gradient estimation through off-policy updates and first-order Taylor expansion, enabling training from scratch with smaller batch sizes and outperforming existing GAN-based methods on multiple quality and diversity metrics.
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