Research map / Label-efficient and robust learning / Semi-supervised learning
Tangent-Normal Adversarial Regularization for Semi-Supervised Learning
Bing Yu, Jingfeng Wu, Jinwen Ma, Zhanxing Zhu
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 |
|---|---|---|---|
| Use tangent and normal adversarial regularization together (ablation) | TNAR-VAE · CIFAR-10, 4,000 labels, no augmentation | classification error (%): 13.87\pm 0.32 -> 12.06\pm 0.35; 15.91\pm 0.09 -> 12.06\pm 0.35 | Improved |
| Add tangent adversarial regularization to normal-only regularization (ablation) | TNAR-VAE · FashionMNIST, 100 labels | classification error (%): 27.83 -> 23.35 | Improved |
| Add normal adversarial regularization to tangent-only regularization (ablation) | TNAR-VAE · FashionMNIST, 100 labels | classification error (%): 26.45 -> 23.35 | Improved |
| Replace the VAE-estimated manifold with a localized-GAN-estimated manifold | TNAR · FashionMNIST, 100 labels | classification error (%): 23.35 -> 23.65 | Hurt |
| Use TNAR with an oracle manifold rather than VAT | Two-rings artificial dataset · Two-rings artificial dataset | classification error (%): 23.80 -> 0 | Improved |
| Use TNAR-VAE rather than VAT | FashionMNIST · FashionMNIST, 1,000 labels | classification error (%): 14.51 -> 12.86 | Improved |
| Use TNAR-VAE rather than VAT (large) | TNAR-VAE (large) · SVHN, 1,000 labels, no augmentation | classification error (%): 4.28\pm 0.10 -> 3.80\pm 0.12 | Improved |
| Use TNAR-VAE rather than VAT (large) | TNAR-VAE (large) · CIFAR-10, 4,000 labels, with augmentation | classification error (%): 10.55\pm 0.05 -> 8.85\pm 0.03 | Improved |
| Add entropy regularization (ablation) | TNAR-Manifold · Two-rings artificial dataset | classification error (%): 9.90 -> 0 | Improved |
| Add entropy regularization (ablation) | VAT · Two-rings artificial dataset | classification error (%) | Hurt |
Closest papers by meaning
- Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning (TPAMI 2018 · similarity 0.91)
- Adversarial Transformations for Semi-Supervised Learning (AAAI 2020 · similarity 0.89)
- Enhancing Adversarial Robustness in Low-Label Regime via Adaptively Weighted Regularization and Knowledge Distillation (ICCV 2023 · similarity 0.87)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.86)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.86)
- Efficient local linearity regularization to overcome catastrophic overfitting (ICLR 2024 · similarity 0.85)
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.
