Research map / Label-efficient and robust learning / Noisy labels
Instance-dependent noise modeling
Estimates feature-dependent noise transitions and correction for instance-level label noise.
106 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Noisy labels, keyword cluster “label noise · transition · instance”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
30 papers in 2023–24 and 24 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 24.8% to 19.7%.
Open questions the data points to
- CIFAR-100 Sym-20% (accuracy): best went from 69.54 (up to 2024) to 92.67 in 2025 — Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · 0 citations)
No longer reported by papers here: Clothing1M (accuracy, last in 2024); CIFAR-10 IDN-20% (accuracy, last in 2024); CIFAR-10 IDN-40% (accuracy, last in 2024); CIFAR-10 Sym-20% (accuracy, last in 2024); CIFAR-10 Sym-50% (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 · Estimating the class prior and posterior from noisy positives and unlabeled data (NeurIPS 2016 · 83 citations)
- 2018 · CleanNet: Transfer Learning for Scalable Image Classifier Training With Label Noise (CVPR 2018 · 497 citations)
- 2019 · Graph Convolutional Label Noise Cleaner: Train a Plug-And-Play Action Classifier for Anomaly Detection (CVPR 2019 · 594 citations)
- 2020 · Distilling Effective Supervision From Severe Label Noise (CVPR 2020 · 147 citations)
- 2021 · Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (AAAI 2021 · 112 citations)
- 2022 · Instance-Dependent Label-Noise Learning With Manifold-Regularized Transition Matrix Estimation (CVPR 2022 · 69 citations)
- 2023 · Identifiability of Label Noise Transition Matrix (ICML 2023 · 6 citations)
- 2024 · Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · 32 citations)
- 2025 · Rethinking Pseudo-Label Guided Learning for Weakly Supervised Temporal Action Localization from the Perspective of Noise Correction (AAAI 2025 · 8 citations)
- 2026 · Stochastic Order Learning: An Approach to Rank Estimation Using Noisy Data (ICML 2026 · 0 citations)
Earliest papers here
- Estimating the class prior and posterior from noisy positives and unlabeled data (NeurIPS 2016 · 83 citations)
- Loss factorization, weakly supervised learning and label noise robustness (ICML 2016 · 41 citations)
- CleanNet: Transfer Learning for Scalable Image Classifier Training With Label Noise (CVPR 2018 · 497 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 · accuracy (higher is better)
12 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 80.67 | Using Trusted Data to Train Deep Networks on Labels Corrupted by Severe Noise (NeurIPS 2018 · 114 citations) |
CIFAR-10 IDN-20% · accuracy (higher is better)
5 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 87.79 | Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (AAAI 2021 · 112 citations) |
| 2022 | 92.26 | Instance-Dependent Label-Noise Learning With Manifold-Regularized Transition Matrix Estimation (CVPR 2022 · 69 citations) |
CIFAR-100 Sym-20% · accuracy (higher is better)
5 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 65.4 | Are Anchor Points Really Indispensable in Label-Noise Learning? (NeurIPS 2019 · 188 citations) |
| 2024 | 69.54 | Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · 15 citations) |
| 2025 | 92.67 | Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · 0 citations) |
CIFAR-10 IDN-40% · accuracy (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 82.98 | Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (AAAI 2021 · 112 citations) |
| 2022 | 85.94 | Instance-Dependent Label-Noise Learning With Manifold-Regularized Transition Matrix Estimation (CVPR 2022 · 69 citations) |
CIFAR-10 Sym-20% · accuracy (higher is better)
4 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 89.63 | Are Anchor Points Really Indispensable in Label-Noise Learning? (NeurIPS 2019 · 188 citations) |
| 2020 | 90.37 | Dual T: Reducing Estimation Error for Transition Matrix in Label-noise Learning (NeurIPS 2020 · 104 citations) |
| 2024 | 91.1 | Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · 15 citations) |
CIFAR-10 Sym-50% · accuracy (higher is better)
4 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 83.4 | Are Anchor Points Really Indispensable in Label-Noise Learning? (NeurIPS 2019 · 188 citations) |
| 2024 | 85.08 | Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · 15 citations) |
What has been tried
90 papers here report 866 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 6%, hurt 17%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add negative learning to standard noisy-label training | CIFAR100N symmetric 20% noise · test accuracy | 35.50 -> 54.70 | Helped | Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · 32 citations) |
| Add contrastive learning | CIFAR80N Sym-20% · test accuracy (%) | 73.17 -> 77.23 | Helped | Unlocking the Power of Open Set: A New Perspective for Open-Set Noisy Label Learning (AAAI 2024 · 14 citations) |
| combine label correction followed by filtering rather than use no cleaning | symmetric noise 20%, 40%; asymmetric noise 20%, 40% · accuracy | 47.6 -> 74.3; 45.3 -> 69.8; 51.9 -> 76.3; 37.7 -> 73.4 | Helped | Which Is More Effective in Label Noise Cleaning, Correction or Filtering? (AAAI 2024 · 14 citations) |
| Add selection criterion cr1 | CIFAR-100 (q=0.1) · accuracy | 73.12% -> 84.07% | Helped | CroSel: Cross Selection of Confident Pseudo Labels for Partial-Label Learning (CVPR 2024 · 11 citations) |
| Add CALA to the pseudo-label retraining baseline | THUMOS14 · AVG mAP (%) | 35.8 -> 37.1 (+1.3) | Helped | Rethinking Pseudo-Label Guided Learning for Weakly Supervised Temporal Action Localization from the Perspective of Noise Correction (AAAI 2025 · 8 citations) |
| add information-bottleneck representation extraction (LIB) | Espgame, Corel 5k, VOC 2007 · AP | 0.1499 -> 0.1934; 0.1543 -> 0.1997; 0.3612 -> 0.4622 | Helped | Theory-Inspired Deep Multi-View Multi-Label Learning with Incomplete Views and Noisy Labels (CVPR 2025 · 6 citations) |
| Remove the instance-dependent penalty | medical (partial multi-label) · AveragePrecision | .821±.036 -> .793±.020 | Hurt | Limited-Supervised Multi-Label Learning with Dependency Noise (AAAI 2024 · 5 citations) |
| Add NMTune regularization to MLP tuning | 14 vision ID tasks · average accuracy | 76.87 -> 77.63; 77.58 -> 78.43; 82.51 -> 84.16; 88.53 -> 89.48; 87.23 -> 88.57 | Helped | Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks (ICLR 2024 · 5 citations) |
| add the policy model (X2) | FashionMNIST SYM 20% and 80%, ASYM 20% and 40%, IDN 20% and 40% · test accuracy | 92.1 -> 92.1; 75.8 -> 71.5; 89.6 -> 91.4; 82.4 -> 88.7; 91.3 -> 91.0; 90.3 -> 90.4 | Mixed | Learning Causal Transition Matrix for Instance-dependent Label Noise (AAAI 2025 · 1 citation) |
| Add ASC to MV-Softmax | LFW · accuracy | 99.12 -> 99.13 | Helped | RepFace: Refining Closed-Set Noise with Progressive Label Correction for Face Recognition (AAAI 2025 · 1 citation) |
| Add collaborative pseudo-labeling | CIFAR-100 · accuracy | 68.40% -> 71.04% | Helped | Pre-Trained Vision-Language Models as Noisy Partial Annotators (AAAI 2025 · 0 citations) |
| Add the noisy-sample loss L_N to L_C | CIFAR-100 Sym. 60% · accuracy | 75.1 -> 79.5 | Helped | Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · 0 citations) |
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 |
|---|---|---|---|---|
| Forward | Clothing1M | accuracy | 71.79 – 79.03 | 2 papers, e.g. NeurIPS 2019 and NeurIPS 2018 |
What reviewers will expect
Datasets and baselines named in the 14 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 14 papers
- CIFAR-100 · 5 of 14 papers
- SVHN · 3 of 14 papers
- MNIST · 2 of 14 papers
- FashionMNIST · 2 of 14 papers
- Clothing1M · 2 of 14 papers
Baselines
- SCE · 2 of 14 papers
- DivideMix · 2 of 14 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 54 papers; from OpenAlex and the papers' affiliations).
- Carnegie Mellon University · 6 papers
- Korea Advanced Institute of Science and Technology · 3 papers
- Southern University of Science and Technology · 3 papers
- The University of Tokyo · 3 papers
- University of Melbourne · 3 papers
- Xi’an Jiaotong University · 2 papers
- Shanxi University · 2 papers
- Zhejiang University · 2 papers
- Tsinghua University · 2 papers
- Swiss Federal Institute of Technology · 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
- Loss correction and noise robustness theory (77 papers)
- Crowdsourced label aggregation (23 papers)
Elsewhere in the topic
- Sample selection and co-training (86 papers)
- Noise-aware loss and pairwise methods (66 papers)
- Contrastive learning on noisy data (55 papers)
- Noisy cross-modal hashing and multi-label (49 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).