Research map / Label-efficient and robust learning / Semi-supervised learning
Doubly-Robust Self-Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · mini-ImageNet100 5% labeled | top-1 accuracy: 4.27 -> 4.89 | Helped |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · mini-ImageNet100 20% labeled | top-1 accuracy: 13.94 -> 13.50 | Hurt |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · mini-ImageNet100 100% labeled | top-1 accuracy: 34.33 -> 34.01 | No clear effect |
| add pseudo-labels to labeled-only training | Labeled Only · mini-ImageNet100 50% labeled | top-1 accuracy: 20.69 -> 24.9 | Helped |
| use pseudo-labels only instead of labeled only | Labeled Only · mini-ImageNet100 100% labeled | top-1 accuracy: 31.07 -> 28.95 | Hurt |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · nuScenes 1/16 labels | mAP: 11.60 -> 12.30 | Helped |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · nuScenes 1/24 labels | NDS: 17.32 -> 18.33 | Helped |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · nuScenes 1/4 labels | mAP: 28.36 -> 27.48 | Hurt |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · nuScenes 1/16 labels, Car | per-class mAP: 48.8 -> 51.5 | Helped |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · nuScenes 1/16 labels, Traffic Cone | per-class mAP: 4.0 -> 4.5 | Helped |
| add doubly robust loss to Labeled + Pseudo | Labeled + Pseudo · nuScenes 1/16 labels, Trailer | per-class mAP: 5.7 -> 5.2 | Hurt |
| use constant alpha_t=1 instead of naive baseline (ablation) | Naive Labeled + Pseudo · ImageNet-100 30% GTs | top-1 accuracy: 28.01 -> 28.43 | No clear effect |
Closest papers by meaning
- Debiased Self-Training for Semi-Supervised Learning (NeurIPS 2022 · similarity 0.89)
- Interactive Self-Training With Mean Teachers for Semi-Supervised Object Detection (CVPR 2021 · similarity 0.88)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.87)
- Unlabeled Data Improves Adversarial Robustness (NeurIPS 2019 · similarity 0.87)
- DualNet: Robust Self-Supervised Stereo Matching with Pseudo-Label Supervision (AAAI 2025 · similarity 0.86)
- Collapse-Proof Non-Contrastive Self-Supervised Learning (ICML 2025 · 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.