# Improving Conditional Sequence Generative Adversarial Networks by Stepwise Evaluation

arXiv: https://arxiv.org/abs/1808.05599 · HTML: https://mlsanity.com/paper/1808.05599/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 16 Aug 2018.

## Built on by (most cited first, 6 of 6)

- [Towards Safety and Helpfulness Balanced Responses via Controllable Large Language Models](https://mlsanity.com/paper/2404.01295/) (2024)
- [Towards Cost-Effective Reward Guided Text Generation](https://mlsanity.com/paper/2502.04517/) (2025)
- [Adversarial Machine Learning in Text Analysis and Generation](https://mlsanity.com/paper/2101.08675/) (2021)
- [CausalDialogue: Modeling Utterance-level Causality in Conversations](https://mlsanity.com/paper/2212.10515/) (2022)
- [A Gradient Analysis Framework for Rewarding Good and Penalizing Bad Examples in Language Models](https://mlsanity.com/paper/2408.16751/) (2024)
- [A Conditional Generative Chatbot using Transformer Model](https://mlsanity.com/paper/2306.02074/) (2023)

## Abstract (arXiv)

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.