Research map / Label-efficient and robust learning / Semi-supervised learning
Triple Generative Adversarial Networks
Chongxuan Li, Kun Xu, Jun Zhu, Jiashuo Liu, Bo Zhang
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 |
|---|---|---|---|
| add the pseudo discriminative loss (alpha_P=3.0 vs 0.0) (ablation) | Triple-GAN-V2 (mean teacher) · CIFAR10 (4000 labels), with augmentation | error rate (%): 13.68 -> 10.01 | Improved |
| add the pseudo discriminative loss (alpha_P=3.0 vs 0.0) (ablation) | Triple-GAN-V2 (mean teacher) · CIFAR10 (4000 labels), no augmentation | error rate (%): 15.90 -> 12.41 | Improved |
| use alpha_P=3.0 instead of 10.0 (ablation) | Triple-GAN-V2 · CIFAR10 (4000 labels), with augmentation | error rate (%): 11.97 -> 10.01 | Improved |
| use alpha_P=1.0 instead of 10.0 (ablation) | Triple-GAN-V2 · CIFAR10 (4000 labels), no augmentation | error rate (%): 13.78 -> 13.10 | Improved |
| add Triple-GAN (G and D) to mean teacher classifier (ablation) | Mean teacher · CIFAR10 (4000 labels), with augmentation | error rate (%): 13.18 -> 10.01 | Improved |
| add Triple-GAN to entropy regularized classifier (ablation) | ER classifier · CIFAR10 (4000 labels), no augmentation | error rate (%): 21.98 -> 15.44 | Improved |
| add Triple-GAN to classifier without unlabeled loss (ablation) | Plain classifier · CIFAR10 (4000 labels), no augmentation | error rate (%): 24.21 -> 17.88 | Improved |
| add standard data augmentation (ablation) | Triple-GAN-V2 · CIFAR10 (4000 labels) | error rate (%): 12.41 -> 10.01 | Improved |
| replace Improved-GAN with Triple-GAN-V1 (same classifier) | Improved-GAN · SVHN (1000 labels) | error rate (%): 8.11 -> 5.77 | Improved |
| add pseudo discriminative loss to Triple-GAN-V1 (ablation) | Triple-GAN-V1 · SVHN | error rate (%): about 7.8% -> 2% better | Improved |
| add G and D to classifier on SVHN (ablation) | Single semi-supervised classifier · SVHN | error rate (%): more than 10% error rate -> 5.77 | Improved |
| replace mean teacher with Triple-GAN-V2 under 26-layer ResNet | MT · CIFAR10 (1000 labels), ResNet | error rate (%): 9.64 -> 8.41 | Improved |
Closest papers by meaning
- Triple Generative Adversarial Nets (NeurIPS 2017 · similarity 0.96)
- Enhancing TripleGAN for Semi-Supervised Conditional Instance Synthesis and Classification (CVPR 2019 · similarity 0.91)
- MCL-GAN: Generative Adversarial Networks with Multiple Specialized Discriminators (NeurIPS 2022 · similarity 0.90)
- Generative Multi-Adversarial Networks (ICLR 2017 · similarity 0.90)
- DO-GAN: A Double Oracle Framework for Generative Adversarial Networks (CVPR 2022 · similarity 0.90)
- Improved Techniques for Training GANs (NeurIPS 2016 · similarity 0.90)
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.
