Research map / Label-efficient and robust learning / Semi-supervised learning
Repetitive Reprediction Deep Decipher for Semi-Supervised Learning
Guo-Hua Wang, Jianxin Wu
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 |
|---|---|---|---|
| Use R2-D2 rather than Dual-View Deep Co-Training | ImageNet validation with ResNet-18 · ImageNet validation, 10% labeled | Top-1 error rate (%): 46.50 -> 41.55 | Improved |
| Use R2-D2 rather than DCT | CIFAR-100 with ConvLarge · CIFAR-100 test, 10,000 labeled | Error rate (%): 34.63 ± 0.14 -> 32.87 ± 0.51 | Improved |
| Use R2-D2 rather than HybridNet | CIFAR-10 with Shake-Shake · CIFAR-10 test, 4,000 labeled | Error rate (%): 6.09 -> 5.72 ± 0.06 | Improved |
| Use R2-D2 rather than DCT with 8 Views | SVHN with ConvLarge · SVHN test, 1,000 labeled | Error rate (%): 3.29 ± 0.03 -> 3.64 ± 0.20 | Hurt |
| Repeat the second stage without reprediction or learning-rate reduction (ablation) | D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 6.71 -> 6.37 | Improved |
| Add reprediction while repeating the second stage, without reducing the learning rate (ablation) | D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 6.37 -> 6.23 | Improved |
| Reduce the learning rate while repeating the second stage, without reprediction (ablation) | D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 6.71 -> 5.94 | Improved |
| Add both reprediction and learning-rate reduction to repeated second-stage training (ablation) | D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 5.94 -> 5.78 | Improved |
| Set α=0.2 rather than α=0.1 with β=0.03 (ablation) | R2-D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 5.78 -> 5.44 | Improved |
| Vary α from 0.1 through 0.5 with β=0.03 (ablation) | R2-D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 5.78; 5.44; 5.81; 5.90; 6.11 | Mixed |
| Vary β from 0.01 through 0.05 with α=0.1 (ablation) | R2-D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 5.62; 5.75; 5.78; 5.83; 5.76 | No clear effect |
| Replace forward KL with reverse KL as the consistency loss (ablation) | R2-D2 on CIFAR-10 · CIFAR-10 test, 4,000 labeled | Error rate (%): 5.78 -> 8.06 | Hurt |
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.
- HybridNet — HybridNet: Classification and Reconstruction Cooperation for Semi-Supervised Learning (ECCV 2018) · 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
- All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training (CVPR 2021 · similarity 0.85)
- Realistic Evaluation of Deep Semi-Supervised Learning Algorithms (NeurIPS 2018 · similarity 0.84)
- Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling (NeurIPS 2026 · similarity 0.84)
- Non-Stationary Predictions May Be More Informative: Exploring Pseudo-Labels with a Two-Phase Pattern of Training Dynamics (ICML 2025 · similarity 0.83)
- SemiReward: A General Reward Model for Semi-supervised Learning (ICLR 2024 · similarity 0.82)
- Contrastive Credibility Propagation for Reliable Semi-supervised Learning (AAAI 2024 · similarity 0.82)
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.
