Research map / Label-efficient and robust learning / Semi-supervised learning
ConMatch: Semi-Supervised Learning with Confidence-Guided Consistency Regularization
Jiwon Kim, Youngjo Min, Daehwan Kim, Gyuseong Lee, Junyoung Seo, Kwangrok Ryoo, Seungryong Kim
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 confidence-guided consistency (parametric) on top of FixMatch | FixMatch · CIFAR-10 (40 labels) | error rate (%): 13.81 -> 5.13 | Improved |
| add confidence-guided consistency (parametric) on top of FlexMatch (ablation) | FlexMatch · CIFAR-100 (400 labels) | error rate (%): 39.94 -> 38.89 | Improved |
| use non-parametric instead of parametric confidence on FlexMatch (ablation) | ConMatch w/ FlexMatch · CIFAR-100 (400 labels) | error rate (%): 38.89 -> 44.90 | Hurt |
| add logit-level self-supervision without confidence weighting (ablation) | FixMatch three-branch baseline · CIFAR-10 (40 labels) | error rate (%): 18.11 -> 77.50 | Hurt |
| add confidence weighting (logits input) to logit-level self-supervision (ablation) | ConMatch w/ FixMatch · CIFAR-10 (40 labels) | error rate (%): 77.50 -> 7.05 | Improved |
| feed features as well as logits to the confidence estimator (ablation) | ConMatch w/ FixMatch · CIFAR-10 (40 labels) | error rate (%): 7.05 -> 5.13 | Improved |
| use stage-wise instead of end-to-end training (ablation) | ConMatch w/ FlexMatch · CIFAR-100 (400 labels) | error rate (%): 57.16 -> 38.89 | Improved |
| use stage-wise instead of end-to-end training (ablation) | ConMatch w/ FixMatch · CIFAR-10 (40 labels) | error rate (%): 4.85 -> 5.13 | No clear effect |
| add feature-encoder warm-up of 4000 iterations (ablation) | ConMatch-P · CIFAR-10 (40 labels) | accuracy (%): 93.24 -> 94.87 | Improved |
| increase warm-up from 4000 to 20000 iterations (ablation) | ConMatch-P · CIFAR-10 (40 labels) | accuracy (%): 94.87 -> 92.34 | Hurt |
| reduce estimator channels, add top-k masking and BatchNorm (ablation) | ConMatch-P confidence estimator · CIFAR-10 (40 labels) | accuracy (%): 92.35 -> 94.87 | Improved |
| use cross-entropy rather than L2 norm as non-parametric similarity (ablation) | ConMatch-NP · CIFAR-10 (40 labels) | error rate (%): 4.93 -> 4.89 | No clear effect |
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.
- FlexMatch — FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling (NeurIPS 2021) · 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 0.94)
- MaxMatch: Semi-Supervised Learning With Worst-Case Consistency (TPAMI 2022 · similarity 0.92)
- AlphaMatch: Improving Consistency for Semi-Supervised Learning With Alpha-Divergence (CVPR 2021 · similarity 0.92)
- MarginMatch: Improving Semi-Supervised Learning with Pseudo-Margins (CVPR 2023 · similarity 0.91)
- RegMixMatch: Optimizing Mixup Utilization in Semi-Supervised Learning (AAAI 2025 · similarity 0.91)
- Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · similarity 0.88)
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.
