Research map / Image generation / GANs
CatGAN: Category-Aware Generative Adversarial Networks with Hierarchical Evolutionary Learning for Category Text Generation
Zhiyue Liu, Jiahai Wang, Zhiwei Liang
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 CatGAN rather than SentiGAN | category text generation · category synthetic, length 20 and 40 | NLL_oracle: 6.976 -> 6.649 ± 0.097; 6.821 -> 6.498 ± 0.186 | Improved |
| Use CatGAN rather than RelGAN | general text generation · general synthetic, length 20 and 40 | NLL_oracle: 6.680 -> 6.377 ± 0.116; 6.756 -> 6.235 ± 0.131 | Improved |
| Use CatGAN rather than the strongest listed baseline | category text generation · MR | BLEU-4: 0.167 -> 0.194 ± 0.028 | Improved |
| Use CatGAN rather than the strongest listed baseline | category text generation · AR | BLEU-4: 0.691 -> 0.867 ± 0.016 | Improved |
| Use CatGAN rather than the strongest listed baseline | general text generation · EN | BLEU-4: 0.501 -> 0.603 ± 0.010 | Improved |
| Increase λ from 0 to 1 (ablation) | CatGAN · AR | NLL_div: 1.470 -> 1.689 | Improved |
| Increase λ from 0 to 1 (ablation) | CatGAN · AR | BLEU-5: 0.757 -> 0.654 | Hurt |
| Add hierarchical evolutionary learning (ablation) | CatGAN · AR | BLEU-5: 0.613 -> 0.751 | Improved |
| Add temperature mutation strategy (TMS) (ablation) | CatGAN · AR | BLEU-5: 0.703 -> 0.751 | Improved |
| Add temperature mutation strategy (TMS) (ablation) | CatGAN · AR | NLL_div: 1.555 -> 1.539 | No clear effect |
| Use OMS rather than only the CatRa objective (ablation) | CatGAN · AR | BLEU-5: 0.638 -> 0.751 | Improved |
| Use OMS rather than only the CatRa objective (ablation) | CatGAN · AR | NLL_div: 1.618 -> 1.539 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SeqGAN — SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient (AAAI 2017) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Branch-GAN: Improving Text Generation with (not so) Large Language Models (ICLR 2024 · similarity 0.88)
- Self-Adversarial Learning with Comparative Discrimination for Text Generation (ICLR 2020 · similarity 0.88)
- RelGAN: Relational Generative Adversarial Networks for Text Generation (ICLR 2019 · similarity 0.87)
- Adversarial Feature Matching for Text Generation (ICML 2017 · similarity 0.87)
- Long Text Generation via Adversarial Training with Leaked Information (AAAI 2018 · similarity 0.87)
- Adversarial Text to Continuous Image Generation (CVPR 2024 · 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.
