Research map / Label-efficient and robust learning / Noisy labels
Noise-aware loss and pairwise methods
Reduces noise effects through pairwise similarity, open-set noise and noise detection.
66 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Noisy labels, keyword cluster “noisy labels · robust · clean”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
15 papers in 2023–24 and 25 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 12.4% to 20.5%.
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
- 2018 · Iterative Learning With Open-Set Noisy Labels (CVPR 2018 · 320 citations)
- 2019 · Learning From Noisy Labels by Regularized Estimation of Annotator Confusion (CVPR 2019 · 248 citations)
- 2020 · Peer Loss Functions: Learning from Noisy Labels without Knowing Noise Rates (ICML 2020 · 97 citations)
- 2021 · Meta Label Correction for Noisy Label Learning (AAAI 2021 · 199 citations)
- 2022 · UniCon: Combating Label Noise Through Uniform Selection and Contrastive Learning (CVPR 2022 · 177 citations)
- 2023 · Fine-Grained Classification With Noisy Labels (CVPR 2023 · 56 citations)
- 2024 · Learning Discriminative Dynamics with Label Corruption for Noisy Label Detection (CVPR 2024 · 16 citations)
- 2025 · Optimized Gradient Clipping for Noisy Label Learning (AAAI 2025 · 6 citations)
- 2026 · Is the Information Bottleneck Robust Enough? Towards Label-Noise Resistant Information Bottleneck Learning (AAAI 2026 · 2 citations)
Earliest papers here
- Iterative Learning With Open-Set Noisy Labels (CVPR 2018 · 320 citations)
- Masking: A New Perspective of Noisy Supervision (NeurIPS 2018 · 170 citations)
- Learning From Noisy Labels by Regularized Estimation of Annotator Confusion (CVPR 2019 · 248 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.
Clothing1M · test accuracy (higher is better)
8 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 73.39 | Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels (ICML 2021 · 21 citations) |
| 2022 | 74.98 | UniCon: Combating Label Noise Through Uniform Selection and Contrastive Learning (CVPR 2022 · 177 citations) |
| 2023 | 87.06 | Delving into Noisy Label Detection with Clean Data (ICML 2023) |
What has been tried
48 papers here report 489 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 8%, hurt 17%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add the clustering loss to a randomly initialized dynamics encoder | CIFAR-10 Asym. · F1 score (%) | 93.8 ± 0.17 -> 93.2 ± 0.11 | Hurt | Learning Discriminative Dynamics with Label Corruption for Noisy Label Detection (CVPR 2024 · 16 citations) |
| Add the CA strategy to AUM-only weighting | CIFAR-10 symmetric noise 0.5 · accuracy (%) | 67.9 -> 82.1 | Helped | Learning with Noisy Labels Using Hyperspherical Margin Weighting (AAAI 2024 · 12 citations) |
| Add loss processing to RML | CIFAR-10 and CIFAR-100 with instance-dependent, symmetric, and pairflip noise · test accuracy (%) | CIFAR-10: 88.47 ± 0.37, 86.73 ± 0.33, 87.75 ± 0.71 -> 90.73 ± 0.44, 89.13 ± 0.21, 90.14 ± 0.34; | Helped | Regroup Median Loss for Combating Label Noise (AAAI 2024 · 12 citations) |
| Use a smaller ε, such as ε = 0.01 | CIFAR-10, Sym-50% and Sym-80% · test accuracy | Mixed | Optimized Gradient Clipping for Noisy Label Learning (AAAI 2025 · 6 citations) | |
| Increase the CT significance level from 0.01 to 0.05 or 0.10 | CIFAR-10/CIFAR-100 noisy labels · accuracy | No clear effect | Enhanced Sample Selection with Confidence Tracking: Identifying Correctly Labeled Yet Hard-to-Learn Samples in Noisy Data (AAAI 2025 · 5 citations) | |
| Add penalty term Lp | UCI (50% noise) · accuracy (%) | 79.80 -> 92.15 | Helped | Noisy Label Calibration for Multi-View Classification (AAAI 2025 · 5 citations) |
| Add weak/strong augmentation | CIFAR-10 symmetric 90% noise · test accuracy | 82.3±2.64 -> 93.5±0.24 | Helped | Revisiting Interpolation for Noisy Label Correction (AAAI 2025 · 5 citations) |
| Add logit adjustment | CIFAR-10, noise 0.5, imbalance 0.01 · accuracy (best) | 72.96 -> 86.71 | Helped | Robust Logit Adjustment for Learning with Long-Tailed Noisy Data (AAAI 2025 · 5 citations) |
| increase the number of perturbations | noise rectification experiments · rectification performance | Helped | NoiseGPT: Label Noise Detection and Rectification through Probability Curvature (NeurIPS 2024 · 5 citations) | |
| use dual-space learning with clean-sample OVA and prototype losses rather than warm-up only | CIFAR80N Sym-80% · Avg. | 45.03 -> 54.82 | Helped | Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 2025 · 3 citations) |
| use binary rather than continuous rescaling | SMS and CoLA, 30% uniform noise · F1 / Matthews correlation | SMS: 82.3±3.1 -> 15.3±24.7^{\Downarrow}; CoLA: 51.8±2.1 -> 0.0±0.0^{\Downarrow} | Hurt | Learning from Noisy Labels via Self-Taught On-the-Fly Meta Loss Rescaling (AAAI 2025 · 2 citations) |
| reduce the sample-selection iterations from three to one | CIFAR-100, symmetric 40% noise · test accuracy | 69.94 ± 0.30 -> 69.53 \(±\) 0.10 | Hurt | Enhancing Sample Selection Against Label Noise by Cutting Mislabeled Easy Examples (NeurIPS 2025 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 16 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-100 · 8 of 16 papers
- CIFAR-10 · 7 of 16 papers
- CIFAR-100N · 3 of 16 papers
- WebVision · 3 of 16 papers
- Food-101N · 3 of 16 papers
- ILSVRC12 · 3 of 16 papers
- CIFAR-10N · 2 of 16 papers
- TinyImageNet · 2 of 16 papers
Baselines
- DivideMix · 6 of 16 papers
- Co-teaching · 4 of 16 papers
- UNICON · 3 of 16 papers
- GCE · 3 of 16 papers
- ELR+ · 3 of 16 papers
- CE · 2 of 16 papers
- HMW · 2 of 16 papers
- DISC · 2 of 16 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 40 papers; from OpenAlex and the papers' affiliations).
- Mohamed bin Zayed University of Artificial Intelligence · 3 papers
- Southeast University · 2 papers
- Hong Kong Baptist University · 2 papers
- Wuhan University · 2 papers
- Google · 2 papers
- The University of Sydney · 2 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
- Sample selection and co-training (86 papers)
- Contrastive learning on noisy data (55 papers)
- Noisy cross-modal hashing and multi-label (49 papers)
Elsewhere in the topic
- Instance-dependent noise modeling (106 papers)
- Loss correction and noise robustness theory (77 papers)
- Crowdsourced label aggregation (23 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).