Research map / Image generation / GANs
Generative Cooperative Networks for Natural Language Generation
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 |
|---|---|---|---|
| Use normalized cooperative training with discriminator-guided MCTS rather than a scheduled GAN with MCTS sampling (ablation) | GCN vs GAN · SQuAD test | BLEU: 17.1 -> 18 | Helped |
| Use normalized cooperative training with discriminator-guided MCTS rather than a scheduled GAN with MCTS sampling (ablation) | GCN vs GAN · CNNDM test | BLEU: 11.9 -> 12.4 | Helped |
| Use discriminator-guided MCTS sampling instead of generator sampling (ablation) | GCN · SQuAD test | BLEU: 16.5 -> 18 | Helped |
| Use discriminator-guided MCTS sampling instead of generator sampling (ablation) | GCN · CNNDM test | BLEU: 11.5 -> 12.4 | Helped |
| Use discriminator-guided MCTS sampling instead of nucleus sampling (ablation) | GCN · SQuAD test | BLEU: 17.5 -> 18 | Helped |
| Use discriminator-guided MCTS sampling instead of nucleus sampling (ablation) | GCN · CNNDM test | BLEU: 12 -> 12.4 | Helped |
| Add MCTS decoding at inference (ablation) | GCN MCTS · SQuAD test | BLEU: 18 -> 18.4 | Helped |
| Add MCTS decoding at inference (ablation) | GCN MCTS · CNNDM test | BLEU: 12.4 -> 12.7 | Helped |
| Scale the generator from T5-small to T5-3B (ablation) | GCN MCTS · SQuAD test | BLEU: 18 -> 21.8 | Helped |
| Scale the generator from T5-small to T5-3B (ablation) | GCN MCTS · CNNDM test | BLEU: 12.4 -> 19.2 | Helped |
| Remove self-normalization without a scheduler (ablation) | GAN p · SQuAD test | BLEU: 16.2 -> 9.4 | Hurt |
| Remove self-normalization without a scheduler (ablation) | GAN MCTS · CNNDM test | BLEU: 11.9 -> 8.5 | Hurt |
Closest papers by meaning
- To Beam Or Not To Beam: That is a Question of Cooperation for Language GANs (NeurIPS 2021 · similarity 0.93)
- Meta-CoTGAN: A Meta Cooperative Training Paradigm for Improving Adversarial Text Generation (AAAI 2020 · similarity 0.90)
- CoopInit: Initializing Generative Adversarial Networks via Cooperative Learning (AAAI 2023 · similarity 0.89)
- Branch-GAN: Improving Text Generation with (not so) Large Language Models (ICLR 2024 · similarity 0.89)
- Training Language GANs from Scratch (NeurIPS 2019 · similarity 0.89)
- TextGAIL: Generative Adversarial Imitation Learning for Text Generation (AAAI 2021 · similarity 0.88)
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.