Research map / Label-efficient and robust learning / Recognition and long-tail
Learning from unlabeled and positive-unlabeled data
Trains classifiers from positive-unlabeled or multiple unlabeled datasets with consistent risk estimators.
240 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Recognition and long-tail, keyword cluster “label · classification · classifier”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
53 papers in 2023–24 and 52 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 4.2% to 4.0%.
Open questions the data points to
- Papers report Plain CNN on TinyImageNet (accuracy) anywhere from 52.2 to 63.7: comparisons against it depend on whose number you take.
No longer reported by papers here: CIFAR-10 (accuracy, last in 2024); CIFAR-100 (accuracy, last in 2024).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Equality of Opportunity in Supervised Learning (NeurIPS 2016 · 2,184 citations)
- 2017 · Revisiting Classifier Two-Sample Tests (ICLR 2017 · 58 citations)
- 2018 · Between-Class Learning for Image Classification (CVPR 2018 · 197 citations)
- 2019 · Classification-Reconstruction Learning for Open-Set Recognition (CVPR 2019 · 473 citations)
- 2020 · Generative-Discriminative Feature Representations for Open-Set Recognition (CVPR 2020 · 204 citations)
- 2021 · Learning Placeholders for Open-Set Recognition (CVPR 2021 · 243 citations)
- 2022 · Crowd Counting in the Frequency Domain (CVPR 2022 · 93 citations)
- 2023 · Contrastive Open Set Recognition (AAAI 2023 · 28 citations)
- 2024 · Multi-Class Support Vector Machine with Maximizing Minimum Margin (AAAI 2024 · 20 citations)
- 2025 · Multi-Granularity Open Intent Classification via Adaptive Granular-Ball Decision Boundary (AAAI 2025 · 5 citations)
- 2026 · Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification (NeurIPS 2026 · 1 citation)
Earliest papers here
- Equality of Opportunity in Supervised Learning (NeurIPS 2016 · 2,184 citations)
- A Probabilistic Collaborative Representation Based Approach for Pattern Classification (CVPR 2016 · 301 citations)
- Fashion Style in 128 Floats: Joint Ranking and Classification Using Weak Data for Feature Extraction (CVPR 2016 · 161 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
CIFAR-10 · accuracy (higher is better)
7 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 90.77 | Self-PU: Self Boosted and Calibrated Positive-Unlabeled Training (ICML 2020 · 26 citations) |
| 2021 | 92.6 | Learning Placeholders for Open-Set Recognition (CVPR 2021 · 243 citations) |
| 2023 | 94.6 | Contrastive Open Set Recognition (AAAI 2023 · 28 citations) |
| 2024 | 96.6 | Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation (CVPR 2024 · 6 citations) |
CIFAR-100 · accuracy (higher is better)
5 papers, 2023–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 73 | Contrastive Open Set Recognition (AAAI 2023 · 28 citations) |
| 2024 | 89.8 | Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation (CVPR 2024 · 6 citations) |
What has been tried
194 papers here report 1,874 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 8%, hurt 19%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Increase the margin regularizer exponent p | Six datasets shown in Figure 4 · test accuracy | Mixed | Multi-Class Support Vector Machine with Maximizing Minimum Margin (AAAI 2024 · 20 citations) | |
| Add novelty-aware sampling to random selection | CIFAR100 · All accuracy | 67.28 -> 69.33 | Helped | Active Generalized Category Discovery (CVPR 2024 · 18 citations) |
| Update label-to-prototype assignment at τ′=10 rather than τ′=1 | CIFAR-100 · accuracy | Helped | Hyperspherical Classification with Dynamic Label-to-Prototype Assignment (CVPR 2024 · 10 citations) | |
| add the TAU-side DC loss | CIFAR10 · AUROC | 92.3 -> 95.6 | Helped | All Beings Are Equal in Open Set Recognition (AAAI 2024 · 9 citations) |
| add rationale-assisted fine-tuning | PHATE test — Violence · F1 | 52.3±2 -> 55.0±3 | Helped | Spanning the Spectrum of Hatred Detection: A Persian Multi-Label Hate Speech Dataset with Annotator Rationales (AAAI 2024 · 7 citations) |
| add regrouping technique | TaxS, TaxH, TaxZ · Accuracy (%) | 87.611 -> 89.451; 92.471 -> 92.597; 95.615 -> 96.161 | Helped | RR-PU: A Synergistic Two-Stage Positive and Unlabeled Learning Framework for Robust Tax Evasion Detection (AAAI 2024 · 6 citations) |
| Add meta-disambiguation to nnPU with Lcont | CIFAR-10 · accuracy | 91.2 -> 96.6 | Helped | Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation (CVPR 2024 · 6 citations) |
| Add hierarchical representation learning | StackOverflow, SNIPS, and BANKING at 25% known classes · F1-All | 79.72 -> 84.43; 68.94 -> 72.20; 71.55 -> 75.19 | Helped | Multi-Granularity Open Intent Classification via Adaptive Granular-Ball Decision Boundary (AAAI 2025 · 5 citations) |
| intra-modal consistency weighting instead of pseudo-label averaging | SET1 four splits, task-aware · cluster accuracy on unlabeled test | e.g. split I 56.9 vs 54.8 (PL avg.) | Helped | Novel Class Discovery in Chest X-rays via Paired Images and Text (AAAI 2024 · 5 citations) |
| Remove the ranker | Figure-reported ablation settings · recommendation performance | Hurt | From Pairwise to Ranking: Climbing the Ladder to Ideal Collaborative Filtering with Pseudo-Ranking (AAAI 2025 · 4 citations) | |
| Add inflated argmax to subbagging | Fashion-MNIST test · maximum instability | 0.966 -> 0.000 | Helped | Building a stable classifier with the inflated argmax (NeurIPS 2024 · 2 citations) |
| Replace L1 loss with KL loss | CIFAR-10, 1 label/category · Top-1 accuracy | 38.22 -> 35.23 | Hurt | Rethinking Guidance Information to Utilize Unlabeled Samples: A Label Encoding Perspective (ICML 2024 · 1 citation) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| Plain CNN | TinyImageNet | accuracy | 52.2 – 63.7 | 2 papers, e.g. CVPR 2021 and AAAI 2023 |
What reviewers will expect
Datasets and baselines named in the 32 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 7 of 32 papers
- CIFAR-100 · 6 of 32 papers
- ImageNet-1k · 4 of 32 papers
- MNIST · 3 of 32 papers
- Stanford Cars · 3 of 32 papers
- FGVC-Aircraft · 3 of 32 papers
- ImageNet-100 · 2 of 32 papers
- EuroSAT · 2 of 32 papers
Baselines
- k-means · 2 of 32 papers
- UNO+ · 2 of 32 papers
- ORCA · 2 of 32 papers
- GCD · 2 of 32 papers
- DCCL · 2 of 32 papers
- GPC · 2 of 32 papers
- SimGCD · 2 of 32 papers
- InfoSieve · 2 of 32 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 105 papers; from OpenAlex and the papers' affiliations).
- Carnegie Mellon University · 5 papers
- Nanyang Technological University · 4 papers
- Nanjing University · 4 papers
- Chongqing University · 3 papers
- Zhejiang University · 3 papers
- Microsoft · 3 papers
- Peking University · 3 papers
- Singapore University of Technology and Design · 3 papers
- University of California, Berkeley · 3 papers
- Alibaba Group · 3 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Partial and multi-label learning (345 papers)
- Loss design and classifier theory (273 papers)
- Group-robust classification (251 papers)
- Confidence calibration objectives (129 papers)
- Deep active learning (76 papers)
Elsewhere in the topic
- Multilingual language representation pretraining (507 papers)
- Contrastive representation theory (495 papers)
- Dense features and foundation priors (482 papers)
- Correspondence and causal representation learning (447 papers)
- Data augmentation and generalization bias (409 papers)
- Recommendation representation learning (207 papers)
- Margin-based face recognition losses (124 papers)
- Supervised person re-identification (108 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).