Research map / Image generation / GANs
SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace MLE training with discriminator-guided policy-gradient training and Monte Carlo rollouts | LSTM sequence generator · Synthetic sequences | NLL_oracle: 9.038 -> 8.736 | Helped |
| Use discriminator feedback instead of BLEU rewards in policy-gradient training | LSTM sequence generator · Synthetic sequences | NLL_oracle: 8.946 -> 8.736 | Helped |
| Use discriminator-guided policy-gradient training | LSTM sequence generator · Chinese poems | BLEU-2: 0.6670 -> 0.7389 | Helped |
| Use discriminator-guided policy-gradient training | LSTM sequence generator · Chinese poems | Human score: 0.4165 -> 0.5356 | Helped |
| Use discriminator-guided policy-gradient training | LSTM sequence generator · Obama political speeches | BLEU-3: 0.519 -> 0.556 | Helped |
| Use discriminator-guided policy-gradient training | LSTM sequence generator · Obama political speeches | BLEU-4: 0.416 -> 0.427 | Helped |
| Use discriminator-guided policy-gradient training | LSTM sequence generator · Nottingham music | BLEU-4: 0.9210 -> 0.9406 | Helped |
| Use discriminator-guided policy-gradient training | LSTM sequence generator · Nottingham music | MSE: 22.38 -> 20.62 | Helped |
| Limit generator MLE pre-training to 20 epochs rather than training to convergence (ablation) | SeqGAN · Synthetic sequences | NLL_oracle | Hurt |
| Train the generator for many updates before updating the discriminator (ablation) | SeqGAN · Synthetic sequences | NLL_oracle | Mixed |
| Increase discriminator training epochs (ablation) | SeqGAN · Synthetic sequences | NLL_oracle | Helped |
| Update the discriminator after one generator epoch, before it is fooled (ablation) | SeqGAN · Synthetic sequences | NLL_oracle | Helped |
Closest papers by meaning
- Generative Cooperative Networks for Natural Language Generation (ICML 2022 · similarity 0.88)
- Verification of the Implicit World Model in a Generative Model via Adversarial Sequences (ICLR 2026 · similarity 0.87)
- DO-GAN: A Double Oracle Framework for Generative Adversarial Networks (CVPR 2022 · similarity 0.87)
- DASCO: Dual-Generator Adversarial Support Constrained Offline Reinforcement Learning (NeurIPS 2022 · similarity 0.87)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.86)
- GFlowNet Training by Policy Gradients (ICML 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.