Improving Conditional Sequence Generative Adversarial Networks by Stepwise Evaluation
This work addresses computational efficiency and performance issues in conditional sequence generation for applications like dialogue systems, but it is incremental as it builds on existing SeqGAN methods.
The paper tackled the problem of stabilizing training in conditional sequence generative adversarial networks (SeqGAN) by proposing StepGAN, a method that modifies the discriminator to assign scores for each subsequence at every generation step, resulting in significantly lower computational costs than Monte Carlo tree search (MCTS) and outperforming previous GAN-based methods in synthetic and chit-chat dialogue generation experiments.
Sequence generative adversarial networks (SeqGAN) have been used to improve conditional sequence generation tasks, for example, chit-chat dialogue generation. To stabilize the training of SeqGAN, Monte Carlo tree search (MCTS) or reward at every generation step (REGS) is used to evaluate the goodness of a generated subsequence. MCTS is computationally intensive, but the performance of REGS is worse than MCTS. In this paper, we propose stepwise GAN (StepGAN), in which the discriminator is modified to automatically assign scores quantifying the goodness of each subsequence at every generation step. StepGAN has significantly less computational costs than MCTS. We demonstrate that StepGAN outperforms previous GAN-based methods on both synthetic experiment and chit-chat dialogue generation.