Research map / Label-efficient and robust learning / Semi-supervised learning
MaxMatch: Semi-Supervised Learning With Worst-Case Consistency
Yangbangyan Jiang, Xiaodan Li, Yuefeng Chen, Yuan He, Qianqian Xu, Zhiyong Yang, Xiaochun Cao, Qingming Huang
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 worst-case consistency over K augmented variants to FixMatch | FixMatch · ImageNet-1k (10% labels, from scratch) | top-1 error (%): 28.74 -> 27.83 | Improved |
| add worst-case consistency to FixMatch with MoCo v2 pretraining | FixMatch · ImageNet-1k (MoCo v2 pretrain) | top-1 error (%): 28.25 -> 28.09 | Improved |
| add worst-case consistency to FixMatch with MoCo v2 and EMAN | FixMatch+EMAN · ImageNet-1k (MoCo v2 + EMAN) | top-1 error (%): 26.56 -> 26.09 | Improved |
| raise initial learning rate from 0.01 to 0.03 for FixMatch (ablation) | FixMatch · ImageNet-1k | top-1 error (%): 29.70 -> 28.74 | Improved |
| raise initial learning rate from 0.02 to 0.03 for MaxMatch (ablation) | MaxMatch · ImageNet-1k | top-1 error (%): 27.83 -> 28.48 | Hurt |
| MaxMatch advantage over FixMatch at lr 0.01 (ablation) | MaxMatch · ImageNet-1k (lr 0.01) | top-1 error (%): 29.70 -> 29.31 | Improved |
| MaxMatch advantage over FixMatch at lr 0.03 (ablation) | MaxMatch · ImageNet-1k (lr 0.03) | top-5 error (%): 10.94 -> 10.93 | No clear effect |
| use RandAugment instead of CTAugment in MaxMatch (ablation) | MaxMatch · SVHN@40 | error rate (%): 1.92 -> 1.75 | Improved |
| use RandAugment instead of CTAugment in FixMatch (ablation) | FixMatch · SVHN@40 | error rate (%): 7.65 -> 3.96 | Improved |
| add worst-case consistency to FixMatch (RandAugment) with 40 labels (ablation) | FixMatch (RA) · SVHN@40 | error rate (%): 3.96 -> 1.75 | Improved |
| add worst-case consistency to FixMatch (CTAugment) with 40 labels (ablation) | FixMatch (CTA) · SVHN@40 | error rate (%): 7.65 -> 1.92 | Improved |
| add worst-case consistency to FixMatch (RandAugment) with 1000 labels (ablation) | FixMatch (RA) · SVHN@1000 | error rate (%): 2.28 -> 1.71 | Improved |
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.
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 10 of 10 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
- AlphaMatch: Improving Consistency for Semi-Supervised Learning With Alpha-Divergence (CVPR 2021 · similarity 0.92)
- ConMatch: Semi-Supervised Learning with Confidence-Guided Consistency Regularization (ECCV 2022 · similarity 0.92)
- Boosting Semi-Supervised Learning by Exploiting All Unlabeled Data (CVPR 2023 · similarity 0.91)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.89)
- RegMixMatch: Optimizing Mixup Utilization in Semi-Supervised Learning (AAAI 2025 · similarity 0.89)
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · similarity 0.89)
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.
