Research map / Label-efficient and robust learning / Semi-supervised learning
Unsupervised Data Augmentation for Consistency Training
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 |
|---|---|---|---|
| Replace simple input noise with advanced task-specific augmentation in consistency training | UDA on CIFAR-10 and Yelp-5 · CIFAR-10 and Yelp-5 | error rate: Cutout: 5.43; RandAugment: 4.32. Switchout: 43.38; Back-translation: 41.35 | Helped |
| Add RandAugment instead of simpler cropping and flipping | UDA on CIFAR-10 · CIFAR-10 (4k labeled, 50k unlabeled) | error rate: 10.94 -> 4.32 | Helped |
| Add back-translation instead of no augmentation | UDA on Yelp-5 · Yelp-5 (2.5k labeled) | error rate: 50.80 -> 41.35 | Helped |
| Add UDA consistency training | BERT fine-tuned on in-domain unlabeled data · IMDb (20 labeled) | error rate: 6.50 -> 4.20 | Helped |
| Add unlabeled examples to UDA training (ablation) | UDA · CIFAR-10 | error rate: With 250 labeled: 35.41 at 5,000 unlabeled; 5.43 ± 0.96 at 50,000 unlabeled | Helped |
| Use a richer set of possible RandAugment transformations (ablation) | UDA on CIFAR-10 · CIFAR-10 | error rate | Helped |
| Add training signal annealing (ablation) | UDA on Yelp-5 · Yelp-5 (2.5k labeled, 6m unlabeled) | error rate: 50.81 -> 41.35 | Helped |
| Use an exponential TSA schedule instead of log or linear schedules (ablation) | UDA on Yelp-5 · Yelp-5 (2.5k labeled, 6m unlabeled) | error rate: log-schedule 49.06; linear-schedule 45.41; exp-schedule 41.35 | Helped |
Closest papers by meaning
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · similarity 0.88)
- SimPLE: Similar Pseudo Label Exploitation for Semi-Supervised Classification (CVPR 2021 · similarity 0.88)
- Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning (NeurIPS 2024 · similarity 0.88)
- Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling (NeurIPS 2026 · similarity 0.87)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.87)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · similarity 0.87)
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.