Research map / Label-efficient and robust learning / Semi-supervised learning
Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effect
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 |
|---|---|---|---|
| Add the consistency term to GP-WGAN | CT-GAN · CIFAR-10, small CNN, 1,000 images | Inception score: 2.98±0.11 -> 5.13±0.12 | Helped |
| Add the consistency term to GP-WGAN | CT-GAN · ImageNet 64×64 | Inception score: 9.85±0.17 -> 10.27±0.15 | Helped |
| Add the consistency term to semi-supervised GAN training | CT-GAN · CIFAR-10, 4,000 labeled | Test error: 16.78±1.80 -> 9.98±0.21 | Helped |
| Add the consistency term (ablation) | CT-GAN · CIFAR-10, 4,000 labeled | Test error: 14.98±0.43 -> 9.98±0.21 | Helped |
| Add GAN training (ablation) | CT-GAN semi-supervised learning · CIFAR-10, 4,000 labeled | Test error: 11.98±0.32 -> 9.98±0.21 | Helped |
| Add second-to-last-layer regularization (ablation) | CT-GAN · CIFAR-10, 4,000 labeled | Test error: 10.70±0.24 -> 9.98±0.21 | Helped |
| Replace weight normalization with batch normalization (ablation) | CT-GAN · CIFAR-10, 4,000 labeled | Test error | Hurt |
| Add dropout layers without the consistency term (ablation) | GP-WGAN · CIFAR-10, small CNN, 1,000 images | Inception score: 2.98±0.11 -> 4.29±0.12 | Helped |
| Add the consistency term beyond adding dropout (ablation) | CT-GAN · CIFAR-10, small CNN, 1,000 images | Inception score: 4.29±0.12 -> 5.13±0.12 | Helped |
Closest papers by meaning
- Improved Training of Wasserstein GANs (NeurIPS 2017 · similarity 0.93)
- A Two-Step Computation of the Exact GAN Wasserstein Distance (ICML 2018 · similarity 0.92)
- Wasserstein Generative Adversarial Networks (ICML 2017 · similarity 0.92)
- Towards Generalized Implementation of Wasserstein Distance in GANs (AAAI 2021 · similarity 0.92)
- Lipschitz Generative Adversarial Nets (ICML 2019 · similarity 0.92)
- CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 2024 · similarity 0.89)
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.