Research map / Label-efficient and robust learning / Semi-supervised learning
There Are Many Consistent Explanations of Unlabeled Data: Why You Should Average
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 |
|---|---|---|---|
| Add fast-SWA weight averaging to Mean Teacher (ablation) | Mean Teacher · CIFAR-10, 4k labels, CNN | test error: 11.41 ± 0.27 -> 9.05 ± 0.21 | Helped |
| Average iterates with fast-SWA rather than SWA (ablation) | Mean Teacher · CIFAR-10, 4k labels, CNN, 1200 schedule | test error: 9.38 ± 0.28 -> 9.05 ± 0.21 | Helped |
| Use a longer fast-SWA schedule (ablation) | Mean Teacher · CIFAR-10, 4k labels, CNN | test error: 10.67 ± 0.18 -> 9.86 ± 0.27 -> 9.05 ± 0.21 | Helped |
| Add fast-SWA weight averaging to the Π model (ablation) | Π model · CIFAR-10, 4k labels, CNN | test error: 12.64 ± 0.11 -> 10.07 ± 0.27 | Helped |
| Average iterates with fast-SWA rather than SWA (ablation) | Π model · CIFAR-10, 4k labels, CNN, 1200 schedule | test error: 10.73 ± 0.39 -> 10.07 ± 0.27 | Helped |
| Use fast-SWA rather than leaving VAT + EntMin unaveraged (ablation) | VAT + EntMin · CIFAR-10, 4k labels | test error: 11.26 -> 10.97 | Helped |
| Add SWA to VAT (ablation) | VAT · CIFAR-10, 4k labels | test error: 11.99 -> 11.16 | Helped |
| Use fast-SWA rather than SWA for the domain adaptation model | Consistency-based CIFAR-10-to-STL adaptation · CIFAR-10 to STL | test error: 19.9% -> 16.8% | Helped |
| Use fast-SWA, averaging multiple points per cycle, rather than SWA's low-learning-rate points (ablation) | Consistency-based semi-supervised models · CIFAR-10, 4k labels, CNN | epochs to reach similar test error: 200 -> 350 | Helped |
| Use Π instead of Mean Teacher with fast-SWA (ablation) | CIFAR-10 Shake-Shake · 4k, 10k, and 50k labels | test error | Mixed |
| Add unlabeled data to Π training (ablation) | Π model · CIFAR-10 | test accuracy | Helped |
| Use EMA weight averaging on the Π student (ablation) | Π model · CIFAR-10, CNN | test error: 12.64 ± 0.11 -> 12.52 ± 0.16 | Helped |
Closest papers by meaning
- Semi-Supervised Learning with Variational Bayesian Inference and Maximum Uncertainty Regularization (AAAI 2021 · similarity 0.88)
- Time-Consistent Self-Supervision for Semi-Supervised Learning (ICML 2020 · similarity 0.87)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.87)
- Generalization Analysis of Stochastic Weight Averaging with General Sampling (ICML 2024 · similarity 0.86)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · similarity 0.85)
- When, Where and Why to Average Weights? (ICML 2025 · 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.