Research map / Label-efficient and robust learning / Semi-supervised learning
Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning
Takeru Miyato, Shin‐ichi Maeda, Masanori Koyama, Shin Ishii
This paper tried a common change: Add data augmentation
add data augmentation to VAT+EntMin · CIFAR-10 semi-supervised · test error rate: 13.15 -> 10.55 Improved
What 689 accepted papers report for this change
- 560 improved
- 44 no clear effect
- 26 mixed
- 59 hurt
- Understanding Multimodal Learning: A Loss Landscape Smoothness Perspective ICML 2026 · add data augmentation (jitter, crop, flip) · UPMC-Food101 · accuracy: 92.84 -> 94.93Improved
- AsyncBEV: Cross-modal flow alignment in Asynchronous 3D Object Detection ICLR 2026 · add EMC + asynchronous data augmentation training · nuScenes val, 0 s offset · NDS: 66.7 -> 63.1Hurt
- Towards Understanding Continual Factual Knowledge Acquisition of Language Models: From Theory to Algorithm ICML 2026 · add data augmentation (5 biographies per individual) in pre-training · Biography PT test, unseen templates · EM: 2.71 -> 81.27Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 virtual adversarial perturbation instead of random perturbation (RPT) (ablation) | VAT · MNIST supervised | test error rate: 0.84 -> 0.64 | Improved |
| use virtual adversarial perturbation instead of random perturbation (RPT) (ablation) | VAT · CIFAR-10 supervised | test error rate: 6.30 -> 5.81 | Improved |
| use virtual adversarial perturbation instead of random perturbation (RPT) (ablation) | VAT · MNIST semi-supervised N_l=100 | test error rate: 6.81 -> 1.36 | Improved |
| use virtual adversarial perturbation instead of random perturbation (RPT) (ablation) | VAT · CIFAR-10 semi-supervised Conv-Small | test error rate: 18.56 -> 14.87 | Improved |
| use VAT instead of L2-constrained adversarial training (ablation) | supervised MLP · MNIST supervised | test error rate: 0.71 -> 0.64 | Improved |
| add VAT regularization to baseline (ablation) | supervised MLP · MNIST supervised | test error rate: 1.11 -> 0.64 | Improved |
| add conditional entropy minimization to VAT (ablation) | Conv-Large · SVHN semi-supervised | test error rate: 5.77 -> 4.28 | Improved |
| add conditional entropy minimization to VAT (ablation) | Conv-Large · CIFAR-10 semi-supervised | test error rate: 14.18 -> 13.15 | Improved |
| add data augmentation to VAT+EntMin (ablation) | Conv-Large · CIFAR-10 semi-supervised | test error rate: 13.15 -> 10.55 | Improved |
| increase power iterations K from 1 to 4 (ablation) | VAT · CIFAR-10 semi-supervised | test error rate: 14.18 -> 14.25 | No clear effect |
| average 9 random perturbation gradients in RPT to match VAT gradient variance (ablation) | RPT · MNIST supervised | test error rate | No clear effect |
| use larger epsilon beyond optimum (ablation) | VAT · SVHN/CIFAR-10 semi-supervised | test error rate | Hurt |
Closest papers by meaning
- Tangent-Normal Adversarial Regularization for Semi-Supervised Learning (CVPR 2019 · similarity 0.91)
- Adversarial Transformations for Semi-Supervised Learning (AAAI 2020 · similarity 0.90)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.87)
- Training Deep Neural Networks with Virtual Smoothing Classes (AAAI 2025 · similarity 0.87)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.86)
- Weakly Supervised Contrastive Adversarial Training for Learning Robust Features from Semi-supervised Data (CVPR 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.
