Research map / Label-efficient and robust learning / Semi-supervised learning
Shrinking Class Space for Enhanced Certainty in Semi-Supervised Learning
Lihe Yang, Zhen Zhao, Lei Qi, Yu Qiao, Yinghuan Shi, Hengshuang Zhao
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 ShrinkMatch uncertain-sample loss to FixMatch+DA baseline (ablation) | FixMatch+DA · CIFAR-100 (400 labels) | accuracy (%): 62.46 -> 64.64 | Improved |
| add ShrinkMatch to baseline (ablation) | FixMatch+DA · STL-10 (40 labels) | accuracy (%): 84.85 -> 85.98 | Improved |
| add both reweighting principles to shrunk loss (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 63.16 -> 64.24 | Improved |
| replace adaptive reweighting with linear scheduling of loss weight (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 64.24 -> 61.23 | Hurt |
| use auxiliary head instead of main head for shrunk-space loss (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 60.65 -> 64.24 | Improved |
| shrink class space instead of using original space with aux head (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 61.69 -> 64.24 | Improved |
| use hard label rather than soft label in shrunk space (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 61.34 -> 64.24 | Improved |
| learn uncertain samples directly in original space with main head (ablation) | FixMatch+DA baseline · CIFAR-100 (400 labels) | accuracy (%): 60.96 -> 52.39 | Hurt |
| raise confidence threshold from 0.95 to 0.98 (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 65.00 -> 65.80 | Improved |
| lower confidence threshold from 0.95 to 0.70 (ablation) | ShrinkMatch · CIFAR-100 (400 labels) | accuracy (%): 65.00 -> 59.58 | Hurt |
| ShrinkMatch vs SimMatch with 2500 labels | ShrinkMatch · CIFAR-100 (2500 labels) | accuracy (%): 74.93 -> 74.83 | No clear effect |
| ShrinkMatch vs FlexMatch with 250 labels | ShrinkMatch · STL-10 (250 labels) | accuracy (%): 91.77 -> 91.55 | 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 2 on most shared results.
- FlexMatch — FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling (NeurIPS 2021) · better on 1 of 2 shared results
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 1 of 1 shared result
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
- HyperMatch: Noise-Tolerant Semi-Supervised Learning via Relaxed Contrastive Constraint (CVPR 2023 · similarity 0.90)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.89)
- ConMatch: Semi-Supervised Learning with Confidence-Guided Consistency Regularization (ECCV 2022 · similarity 0.88)
- MaxMatch: Semi-Supervised Learning With Worst-Case Consistency (TPAMI 2022 · similarity 0.88)
- SoftMatch: Addressing the Quantity-Quality Tradeoff in Semi-supervised Learning (ICLR 2023 · similarity 0.88)
- OwMatch: Conditional Self-Labeling with Consistency for Open-World Semi-Supervised Learning (NeurIPS 2024 · similarity 0.86)
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.
