Research map / Label-efficient and robust learning / Semi-supervised learning
ReMixMatch: Semi-Supervised Learning with Distribution Matching and Augmentation Anchoring
David Berthelot, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Kihyuk Sohn, Han Zhang, Colin Raffel
Read the paperCheck 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 |
|---|---|---|---|
| Add distribution alignment to MixMatch (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 5.94 -> 7.28 | Improved |
| Add the pre-mixup unlabeled loss (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 5.94 -> 6.66 | Improved |
| Add the rotation loss (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 5.94 -> 6.08 | Improved |
| Use cross-entropy instead of L2 for the unlabeled loss (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 17.28 -> 5.94 | Improved |
| Add strong augmentation anchoring (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 12.51 -> 5.94 | Improved |
| Add the weakly augmented anchor (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 29.36 -> 5.94 | Improved |
| Increase the number of augmentations from K=1 to K=8 (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 7.32 -> 5.94 | Improved |
| Increase the number of augmentations from K=2 to K=8 (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 6.74 -> 5.94 | Improved |
| Increase the number of augmentations from K=4 to K=8 (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 6.21 -> 5.94 | Improved |
| Increase the number of augmentations from K=8 to K=16 (ablation) | ReMixMatch · CIFAR-10, 250-label split | error rate: 5.94 -> 5.93 | No clear effect |
| Increase the number of augmentations from K=1 to K=4 (ablation) | ReMixMatch · STL-10, 1000-label splits | error rate: $6.77 \pm 1.66$ -> $6.18 \pm 1.24$ | Improved |
| Use ReMixMatch rather than MixMatch | CIFAR-10, 250 labels · CIFAR-10, 250 labels | error rate: 11.08$\pm$0.87 -> 6.27$\pm$0.34 | Improved |
Later papers that compared with this method
23 later accepted papers report ReMixMatch as a baseline in their results tables; 14 of them report a better number on most of the results they share with it.
- Learnable Logit Adjustment for Imbalanced Semi-Supervised Learning under Class Distribution Mismatch (ICCV 2025 · better on 3 of 3 shared results)
- Leveraging Local Variance for Pseudo-Label Selection in Semi-supervised Learning (AAAI 2024 · better on 0 of 1 shared result)
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · better on 1 of 1 shared result)
- OTMatch: Improving Semi-Supervised Learning with Optimal Transport (ICML 2024 · better on 0 of 1 shared result)
- FreeMatch: Self-adaptive Thresholding for Semi-supervised Learning (ICLR 2023 · better on 0 of 2 shared results)
- SoftMatch: Addressing the Quantity-Quality Tradeoff in Semi-supervised Learning (ICLR 2023 · better on 0 of 2 shared results)
- FlatMatch: Bridging Labeled Data and Unlabeled Data with Cross-Sharpness for Semi-Supervised Learning (NeurIPS 2023 · better on 1 of 1 shared result)
- Enhancing Sample Utilization through Sample Adaptive Augmentation in Semi-Supervised Learning (ICCV 2023 · better on 0 of 1 shared result)
15 more not listed.
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MixMatch — MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020 · similarity 0.90)
- ConMatch: Semi-Supervised Learning with Confidence-Guided Consistency Regularization (ECCV 2022 · similarity 0.88)
- RegMixMatch: Optimizing Mixup Utilization in Semi-Supervised Learning (AAAI 2025 · similarity 0.88)
- Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning (NeurIPS 2024 · similarity 0.88)
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · similarity 0.85)
- OTMatch: Improving Semi-Supervised Learning with Optimal Transport (ICML 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.
