Research map / Label-efficient and robust learning / Semi-supervised learning
Revisiting Consistency Regularization for Semi-Supervised Learning
Yue Fan, Anna Kukleva, Dengxin Dai, Bernt Schiele
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 FeatDistLoss (increase feature distance) (ablation) | CR-Match · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 48.89 -> 45.52 | Improved |
| push features apart (cosine similarity) instead of pulling together (cosine distance) (ablation) | FeatDistLoss · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 46.98 -> 45.52 | Improved |
| pull features together (cosine distance) vs none (ablation) | FeatDistLoss · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 48.89 -> 46.98 | Improved |
| add FeatDistLoss to RotNet model (ablation) | CR-Match · CIFAR-100 (4 labels/class) | error rate: 41.48 -> 39.22 | Improved |
| add RotNet to FeatDistLoss model (ablation) | CR-Match · CIFAR-100 (4 labels/class) | error rate: 43.48 -> 39.22 | Improved |
| add FeatDistLoss to RotNet model (ablation) | CR-Match · CIFAR-10 (4 labels/class) | error rate: 10.33 -> 10.92 | Hurt |
| add FeatDistLoss to RotNet model (ablation) | CR-Match · Mini-ImageNet (40 labels/class) | error rate: 34.64 -> 33.82 | Improved |
| add FeatDistLoss to RotNet model (ablation) | CR-Match · SVHN (4 labels/class) | error rate: 2.21 -> 2.09 | Improved |
| use weak-strong pair instead of weak-weak for FeatDistLoss (ablation) | CR-Equiv · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 48.51 -> 45.52 | Improved |
| use weak-strong pair instead of strong-strong for FeatDistLoss (ablation) | CR-Equiv · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 48.05 -> 45.52 | Improved |
| add linear projection before FeatDistLoss (ablation) | CR-Match · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 48.37 -> 45.52 | Improved |
| replace linear projection with MLP (ablation) | CR-Match · CIFAR-100 (4 labels/class, WRN-28-2) | error rate: 45.52 -> 47.52 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FeatMatch — FeatMatch: Feature-Based Augmentation for Semi-Supervised Learning (ECCV 2020) · better on 2 of 2 shared results
- ReMixMatch — ReMixMatch: Semi-Supervised Learning with Distribution Matching and Augmentation Anchoring (ICLR 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
- MaxMatch: Semi-Supervised Learning With Worst-Case Consistency (TPAMI 2022 · similarity 0.90)
- On Data-Augmentation and Consistency-Based Semi-Supervised Learning (ICLR 2021 · similarity 0.90)
- Adaptive Consistency Regularization for Semi-Supervised Transfer Learning (CVPR 2021 · similarity 0.89)
- Image-Feature Weak-to-Strong Consistency: An Enhanced Paradigm for Semi-Supervised Learning (ECCV 2024 · similarity 0.89)
- Not All Classes Stand on Same Embeddings: Calibrating a Semantic Distance with Metric Tensor (CVPR 2024 · similarity 0.88)
- Controller-Guided Partial Label Consistency Regularization with Unlabeled Data (AAAI 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.
