Research map / Large language models / Language, safety and interpretability
Incorporating Discriminator in Sentence Generation: a Gibbs Sampling Method
Jinyue Su, Jiacheng Xu, Xipeng Qiu, Xuanjing Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use Gibbs Sampling instead of Beam Search | Constrained sentence generation · SST-2 | BLEU-4 (w/ penalty): 1.290 -> 2.060 | Improved |
| Use Gibbs Sampling instead of Beam Search | Constrained sentence generation · SST-2 | BLEU-4 (w/o penalty): 6.059 -> 8.302 | Improved |
| Use Gibbs Sampling instead of Reject Sampling | Constrained sentence generation · SST-2 | BLEU-4 (w/ penalty): 0.384 -> 2.060 | Improved |
| Use Gibbs Sampling instead of Reject Sampling | Constrained sentence generation · SST-2 | BLEU-4 (w/o penalty): 4.029 -> 8.302 | Improved |
| Use Gibbs Sampling instead of Beam Search | Constrained sentence generation · Product | BLEU-4 (w/ penalty): 0.218 -> 1.476 | Improved |
| Use Gibbs Sampling instead of Beam Search | Constrained sentence generation · Product | BLEU-4 (w/o penalty): 4.556 -> 4.750 | Improved |
| Use Gibbs Sampling instead of Reject Sampling | Constrained sentence generation · Product | BLEU-4 (w/ penalty): 0.369 -> 1.476 | Improved |
| Use Gibbs Sampling instead of Reject Sampling | Constrained sentence generation · Product | BLEU-4 (w/o penalty): 3.323 -> 4.750 | Improved |
| Use Gibbs Sampling instead of Random generation | Constrained sentence generation · SST-2 | BLEU-4 (w/ penalty): 4.338 -> 2.060 | Hurt |
| Use Gibbs Sampling instead of Random generation | Constrained sentence generation · SST-2 | BLEU-4 (w/o penalty): 5.888 -> 8.302 | Improved |
| Use Gibbs Sampling instead of Random generation | Constrained sentence generation · Product | BLEU-4 (w/ penalty): 2.093 -> 1.476 | Hurt |
| Use Gibbs Sampling instead of Random generation | Constrained sentence generation · Product | BLEU-4 (w/o penalty): 2.854 -> 4.750 | Improved |
Closest papers by meaning
- CGMH: Constrained Sentence Generation by Metropolis-Hastings Sampling (AAAI 2019 · similarity 0.88)
- Controlling Global Statistics in Recurrent Neural Network Text Generation (AAAI 2018 · similarity 0.86)
- Show Me How To Revise: Improving Lexically Constrained Sentence Generation with XLNet (AAAI 2021 · similarity 0.86)
- Multiresolution Recurrent Neural Networks: An Application to Dialogue Response Generation (AAAI 2017 · similarity 0.86)
- Recurrent Hierarchical Topic-Guided RNN for Language Generation (ICML 2020 · similarity 0.85)
- Generating Sentences Using a Dynamic Canvas (AAAI 2018 · similarity 0.84)
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.
