Research map / Label-efficient and robust learning / Noisy labels
Contrastive learning on noisy data
Applies contrastive and diffusion-based training to noisy, imbalanced or weakly labeled recognition.
55 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Noisy labels, keyword cluster “noisy labels · robust · clean”
State History Tried Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
14 papers in 2023–24 and 20 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 11.6% to 16.4%.
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 · Unsupervised Learning from Noisy Networks with Applications to Hi-C Data (NeurIPS 2016)
- 2017 · Learning From Noisy Large-Scale Datasets With Minimal Supervision (CVPR 2017 · 481 citations)
- 2018 · Deep Unsupervised Saliency Detection: A Multiple Noisy Labeling Perspective (CVPR 2018 · 195 citations)
- 2019 · Unequal-Training for Deep Face Recognition With Long-Tailed Noisy Data (CVPR 2019 · 134 citations)
- 2020 · SuperLoss: A Generic Loss for Robust Curriculum Learning (NeurIPS 2020 · 41 citations)
- 2021 · Partially View-Aligned Representation Learning With Noise-Robust Contrastive Loss (CVPR 2021 · 172 citations)
- 2022 · The Devil Is in the Labels: Noisy Label Correction for Robust Scene Graph Generation (CVPR 2022 · 113 citations)
- 2023 · Twin Contrastive Learning With Noisy Labels (CVPR 2023 · 115 citations)
- 2024 · Negative Pre-aware for Noisy Cross-Modal Matching (AAAI 2024 · 18 citations)
- 2025 · Learning with Noisy Triplet Correspondence for Composed Image Retrieval (CVPR 2025 · 9 citations)
- 2026 · ConeSep: Cone-based Robust Noise-Unlearning Compositional Network for Composed Image Retrieval (CVPR 2026 · 1 citation)
Earliest papers here
- Unsupervised Learning from Noisy Networks with Applications to Hi-C Data (NeurIPS 2016)
- Learning From Noisy Large-Scale Datasets With Minimal Supervision (CVPR 2017 · 481 citations)
- Deep Unsupervised Saliency Detection: A Multiple Noisy Labeling Perspective (CVPR 2018 · 195 citations)
What has been tried
41 papers here report 382 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 8%, hurt 14%, 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 |
|---|---|---|---|---|
| Add confidence weight w | Flickr30K, 60% noise, image-to-text · R@1 | 78.2 -> 83.0 | Helped | Negative Pre-aware for Noisy Cross-Modal Matching (AAAI 2024 · 18 citations) |
| Add CSC to the SURE variant with CRL, RegMixup, and FMFP | CIFAR-100 failure prediction · accuracy | 79.17±0.30 -> 80.49±0.18 | Helped | SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · 15 citations) |
| Add GMM filtering | CIRR validation, 20% noise · Avg(R@5, Rsubset@1) | 78.75 -> 79.69 | Helped | Learning with Noisy Triplet Correspondence for Composed Image Retrieval (CVPR 2025 · 9 citations) |
| replace Lsr with L | MoGR, SNR=0 · NRMSE(%) | 0.69 -> 25.45 | Hurt | Unsupervised Deep Unrolling Networks for Phase Unwrapping (CVPR 2024 · 6 citations) |
| Add selective division, adaptive learning and ACCA (full RoDA) vs. CE only | Office-Home 20% noise, 12 tasks · mAP (Best) | 49.47 vs 32.99 | Helped | RoDA: Robust Domain Alignment for Cross-Domain Retrieval Against Label Noise (AAAI 2025 · 5 citations) |
| Add CGP (coarse-grained pruning) | FERV39k · WAR | 42.59 -> 46.13 | Helped | D2SP: Dynamic Dual-Stage Purification Framework for Dual Noise Mitigation in Vision-based Affective Recognition. (CVPR 2025 · 5 citations) |
| Add SIVC soft-label estimation instead of hard labels | Flickr30K, 40% noise · Image-Text R@1 | 74.8 -> 77.7 | Helped | TSVC: Tripartite Learning with Semantic Variation Consistency for Robust Image-Text Retrieval (AAAI 2025 · 4 citations) |
| Add the k-nearest neighbor graph for label propagation | Cora, 5% IND noise and near OOD noise · F1 | 74.87 -> 78.36 | Helped | ROG_PL: Robust Open-Set Graph Learning via Region-Based Prototype Learning (AAAI 2024 · 3 citations) |
| add AE&ER soft thresholding | ISIC2016&ph2 · Dice | 0.9165 -> 0.9359 | Helped | Minding Fuzzy Regions: A Data-driven Alternating Learning Paradigm for Stable Lesion Segmentation (CVPR 2025 · 2 citations) |
| Use iMRF label correction instead of network predictions as posterior | CIFAR-100, LIOND, TinyImageNet open-set ρ=40% · top-1 accuracy (%) | 53.56 -> 56.54 | Helped | Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · 2 citations) |
| Increase coarse-to-fine refinement from S=0 to S=3 | Middlebury2021 · ORD | 0.313 -> 0.305 | Helped | Self-Distilled Depth Refinement with Noisy Poisson Fusion (NeurIPS 2024 · 1 citation) |
| Add noisy identification and rectification | Six benchmark datasets, noisy scenarios · clustering performance | Helped | Automatically Identify and Rectify: Robust Deep Contrastive Multi-view Clustering in Noisy Scenarios (ICML 2025 · 0 citations) |
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 34 papers; from OpenAlex and the papers' affiliations).
- National University of Singapore · 3 papers
- A*STAR · 2 papers
- Sichuan University · 2 papers
- Fudan University · 2 papers
- State University of New York at Albany · 2 papers
Use it
Download BibTeX of 37 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)
- Noise-aware loss and pairwise methods (66 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).