Research map / Label-efficient and robust learning / Semi-supervised learning
Learning Rare Category Classifiers on a Tight Labeling Budget
Ravi Teja Mullapudi, Fait Poms, William R. Mark, Deva Ramanan, Kayvon Fatahalian
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 pseudo negative labels (machine-labeled easy negatives) when updating the deep representation (ablation) | Ours (active semi-supervised rare category learner) · iNaturalist (50 rare categories) | mean F1 | Improved |
| add background splitting to handle extreme imbalance (ablation) | Ours · iNaturalist (50 rare categories) | mean F1 | Improved |
| add frequent linear model updates between deep feature updates (B=10,Q=5 vs B=50,Q=1) (ablation) | Ours · iNaturalist (50 rare categories) | mean F1: N = 10, B = 50, Q = 1 -> N = 10, B = 10, Q = 5 | Improved |
| replace easy-negative-only pseudo labels with label propagation pseudo positives and negatives (DeepProp) | Ours · iNaturalist (50 categories) and Places (20 categories) | mean F1 | Hurt |
| replace easy-negative-only pseudo labels with deep model predictions as pseudo positives and negatives (KD-Semi) | Ours · iNaturalist (50 categories) and Places (20 categories) | mean F1 | Hurt |
| raise KD-Semi confidence threshold for pseudo positives from 0.90 to 0.98 | KD-Semi · iNaturalist (50 categories) | false positive rate of automatic positives: 0.90 -> 0.95 -> 0.98 | Mixed |
| add deep feature updating (Ours) over frozen features (Tropel) with a deep model trained on pseudo negatives | Tropel · iNaturalist (50 categories) and Places (20 categories) | mean F1 | Improved |
| train a deep model only on human labels (Tropel-Deep) instead of a linear model on frozen features | Tropel · iNaturalist (50 categories) | mean F1 | Hurt |
| start from self-supervised SwAV representations (ImageNet or iNaturalist) instead of supervised ImageNet pretraining | Ours · iNaturalist (50 categories) | mean F1 | Hurt |
| add deep feature updating to self-supervised representations (Ours-Self vs Tropel-Self) | Tropel-Self · iNaturalist (50 categories) | mean F1 | Improved |
Closest papers by meaning
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.89)
- Automatically Discovering and Learning New Visual Categories with Ranking Statistics (ICLR 2020 · similarity 0.88)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.87)
- Learning from Label Proportions: Bootstrapping Supervised Learners via Belief Propagation (ICLR 2024 · similarity 0.87)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 2025 · similarity 0.86)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 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.
