Research map / Label-efficient and robust learning / Recognition and long-tail
On Positive-Unlabeled Classification in GAN
Tianyu Guo, Chang Xu, Jiajun Huang, Yunhe Wang, Boxin Shi, Chao Xu, Dacheng Tao
Check 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 |
|---|---|---|---|
| treat generated samples as unlabeled and train the discriminator to distinguish high- from low-quality generated samples | GANs on MNIST, Fashion, CIFAR-10, CelebA-64, and CAT-64 · MNIST, Fashion, CIFAR-10, CelebA-64, CAT-64 | FID: 27.64 -> 22.32 | Improved |
| use PU learning rather than ordinary real-versus-generated classification | GANs · MNIST | FID: 13.47 -> 13.96 | Hurt |
| use PU learning rather than ordinary real-versus-generated classification | GANs · CAT-64 stability | FID mean: 48.36 -> 16.38 | Improved |
| increase the learning rate to .001 (ablation) | PUSGAN · CIFAR-10 | FID: 31.02 -> 40.93 | Hurt |
| remove batch normalization (ablation) | PUSGAN · CIFAR-10 | FID: 31.02 -> 37.85 | Hurt |
| replace activation functions with Tanh (ablation) | PUSGAN · CIFAR-10 | FID: 31.02 -> 54.70 | Hurt |
| increase the learning rate to .001 (ablation) | PULSGAN · CIFAR-10 | FID: 22.32 -> 37.65 | Hurt |
| remove batch normalization (ablation) | PULSGAN · CIFAR-10 | FID: 22.32 -> 37.34 | Hurt |
| replace activation functions with Tanh (ablation) | PULSGAN · CIFAR-10 | FID: 22.32 -> 51.45 | Hurt |
| start the class prior at 0.3 and increase it by 0.1 every 10k iterations (ablation) | PUSGAN · CAT-64 | FID | Hurt |
| fix the class prior at 0.3 (ablation) | PUSGAN · CAT-64 | FID | Mixed |
| fix the class prior at 0.5 (ablation) | PUSGAN · CAT-64 | FID | Mixed |
Closest papers by meaning
- On Positive-Unlabeled Classification From Corrupted Data in GANs (TPAMI 2025 · similarity 0.98)
- Predictive Adversarial Learning from Positive and Unlabeled Data (AAAI 2021 · similarity 0.90)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.88)
- Mixed batches and symmetric discriminators for GAN training (ICML 2018 · similarity 0.87)
- Discriminative Forests Improve Generative Diversity for Generative Adversarial Networks (AAAI 2024 · similarity 0.86)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.86)
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.
