Research map / Image generation / GANs
Language GANs Falling Short
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 |
|---|---|---|---|
| Reduce MLE inference temperature from 1.0 to 0.001 (ablation) | MLE · Synthetic task | NLL_oracle: 9.40 -> 4.58 | Helped |
| Use temperature-tuned MLE instead of the strongest listed GAN result | Synthetic task · Synthetic task | NLL_oracle: 6.91 -> 4.58 | Helped |
| Tune MLE inference temperature | MLE · EMNLP News 2017 test | BLEU-5: 0.27 -> 0.32 | Helped |
| Tune MLE inference temperature | MLE · EMNLP News 2017 test | Self-BLEU-5: 0.30 -> 0.41 | Hurt |
| Use MLE instead of MaliGAN | Image COCO · Image COCO test | Self-BLEU-5: 0.44 -> 0.38 | Helped |
| Raise temperature during training (ablation) | GAN generator · EMNLP News 2017 | diversity | Hurt |
| Use stochastic beam search or generator rejection sampling instead of temperature tuning (ablation) | MLE decoding · EMNLP News 2017 | quality-diversity trade-off | Helped |
| Increase generator rejection threshold (ablation) | MLE decoding · EMNLP News 2017 | decoding time | Hurt |
| Increase stochastic beam size (ablation) | MLE decoding · EMNLP News 2017 | quality-diversity control | Hurt |
| Begin adversarial updates after MLE training (ablation) | GAN generator · Synthetic task | NLL_test | Hurt |
Closest papers by meaning
- Training Language GANs from Scratch (NeurIPS 2019 · similarity 0.90)
- ColdGANs: Taming Language GANs with Cautious Sampling Strategies (NeurIPS 2020 · similarity 0.90)
- Are GANs overkill for NLP? (NeurIPS 2022 · similarity 0.89)
- SimpleStrat: Diversifying Language Model Generation with Stratification (NeurIPS 2025 · similarity 0.87)
- Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · similarity 0.87)
- Enhancing NLU in Large Language Models Using Adversarial Noisy Instruction Tuning (AAAI 2025 · similarity 0.86)
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.