Research map / Label-efficient and robust learning / Semi-supervised learning
Self-paced Multi-view Co-training
Fan Ma, Deyu Meng, Xuanyi Dong, Yi Yang
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 self-paced soft co-regularization instead of hard | SPamCo · Reuters multilingual | accuracy: 73.28 -> 73.83 | Improved |
| use SPamCo(soft) instead of Co-Label | multi-view SSL · Reuters multilingual | accuracy: 72.45 -> 73.83 | Improved |
| draw unlabeled data with replacement in co-training (ablation) | Cotrain · Market-1501 two-view (R50+D121) | MAP: 67.0 -> 67.7 | Improved |
| use SPamCo(soft) instead of Cotrain(Rep) selecting pseudo-labels from all views | co-training with replacement · Market-1501 two-view (R50+D121) | MAP: 67.7 -> 69.5 | Improved |
| add update of pseudo-labels of unlabeled examples (vs fixed labels) with hard term (ablation) | SPamCo(hard) · Market-1501 triple-view | MAP: 64.8 -> 71.3 | Improved |
| add update of pseudo-labels with soft term (ablation) | SPamCo(soft) · Market-1501 triple-view | MAP: 67.3 -> 71.8 | Improved |
| add update of pseudo-labels with hard term (ablation) | SPamCo(hard) · Market-1501 triple-view | rank-1 accuracy: 83.1 -> 87.0 | Improved |
| use parallel instead of serial training of view classifiers (ablation) | SPamCo(soft) · Market-1501 triple-view | MAP: 71.8 -> 71.7 | No clear effect |
| use three views instead of two views | SPamCo(soft) · Market-1501 | MAP: 69.5 -> 71.8 | Improved |
| increase number of added unlabeled instances per iteration (ablation) | SPamCo · Reuters multilingual | accuracy | Mixed |
| use soft instead of hard regularization when labels are fixed (ablation) | SPamCo · Market-1501 triple-view | MAP: 54.7 -> 57.3 | Improved |
| use SPamCo(Psoft) instead of ICT with 4000 labels | semi-supervised CIFAR-10 · CIFAR-10 (4000 labels) | error rate (%): 7.66 -> 7.05 | Improved |
Closest papers by meaning
- Self-Paced Co-training (ICML 2017 · similarity 0.91)
- Semi-supervised Learning with Multi-Head Co-Training (AAAI 2022 · similarity 0.90)
- Co-Labeling for Multi-View Weakly Labeled Learning (TPAMI 2016 · similarity 0.88)
- All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training (CVPR 2021 · similarity 0.88)
- Decouple then Classify: A Dynamic Multi-view Labeling Strategy with Shared and Specific Information (ICML 2024 · similarity 0.85)
- Learning from Synchronization: Self-Supervised Uncalibrated Multi-View Person Association in Challenging Scenes (CVPR 2025 · similarity 0.84)
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.
