Research map / Label-efficient and robust learning
Noisy labels: research map
576 papers on Noisy labels in Label-efficient and robust learning, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 2 clusters and 7 approaches. The busiest year so far is 2024.
What it is
Real datasets contain wrong labels, from crowd workers, web scraping or automatic tagging. Deep networks memorise these mistakes, so this topic studies how to train well anyway: modelling how labels get corrupted, using losses that are robust to noise, or selecting the examples that are probably labelled correctly.
Terms you will see on this page:
- Label noise
- training labels that are wrong, at random or depending on the example.
- Noise transition matrix
- the probabilities with which each true class is mislabelled as each other class.
- Sample selection
- training mainly on examples the model believes are correctly labelled.
Start reading with the most cited papers here:
- Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels (NeurIPS 2018 · 1,467 citations)
- Making Deep Neural Networks Robust to Label Noise: A Loss Correction Approach (CVPR 2017 · 1,463 citations)
Within Label-efficient and robust learning, its share held steady from 4.2% in 2023–24 to 4.0% in 2025–26 (121 → 122 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Noisy labels in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Noise-aware loss and pairwise methods rising: 12% → 20% of the topic (15 → 25 papers)
- Noisy cross-modal hashing and multi-label rising: 9% → 15% of the topic (11 → 18 papers)
- Loss correction and noise robustness theory rising: 14% → 20% of the topic (17 → 24 papers)
- Sample selection and co-training falling: 20% → 6% of the topic (24 → 7 papers)
- Instance-dependent noise modeling falling: 25% → 20% of the topic (30 → 24 papers)
Papers from 2025–2026 that are already cited most:
- Data Augmentation of Contrastive Learning is Estimating Positive-incentive Noise (ICML 2026 · 17 citations)
- Robust Self-Paced Hashing for Cross-Modal Retrieval with Noisy Labels (AAAI 2025 · 17 citations)
- ROLL: Robust Noisy Pseudo-label Learning for Multi-View Clustering with Noisy Correspondence (CVPR 2025 · 14 citations)
Approaches and key papers
Training robustly on noisy labels · 337 papers
Keywords: noisy labels, robust, clean
Approaches in this cluster:
- Sample selection and co-training (108 papers)
Separates clean from noisy samples using small-loss selection, co-teaching and joint optimization. - Noise-aware loss and pairwise methods (100 papers)
Reduces noise effects through pairwise similarity, open-set noise and noise detection. - Contrastive learning on noisy data (72 papers)
Applies contrastive and diffusion-based training to noisy, imbalanced or weakly labeled recognition. - Noisy cross-modal hashing and multi-label (57 papers)
Handles noisy labels in cross-modal retrieval and multi-label classification via robust hashing and loss handling.
Most cited and most cited since 2024:
- Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels (NeurIPS 2018 · 1,467 citations)
- Co-teaching: Robust training of deep neural networks with extremely noisy labels (NeurIPS 2018 · 1,182 citations)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · 20 citations)
- L2B: Learning to Bootstrap Robust Models for Combating Label Noise (CVPR 2024 · 20 citations)
Modelling how labels get corrupted · 239 papers
Keywords: label noise, transition, instance
Approaches in this cluster:
- Instance-dependent noise modeling (130 papers)
Estimates feature-dependent noise transitions and correction for instance-level label noise. - Loss correction and noise robustness theory (79 papers)
Corrects losses and analyzes how overfitting and subclass structure affect noisy training. - Crowdsourced label aggregation (30 papers)
Infers true labels from noisy annotators using worker confusion and ability models.
Most cited and most cited since 2024:
- Making Deep Neural Networks Robust to Label Noise: A Loss Correction Approach (CVPR 2017 · 1,463 citations)
- Graph Convolutional Label Noise Cleaner: Train a Plug-And-Play Action Classifier for Anomaly Detection (CVPR 2019 · 594 citations)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · 32 citations)
- Dual Self-Paced Cross-Modal Hashing (AAAI 2024 · 28 citations)
Related topics in Label-efficient and robust learning
- Zero-/few-shot and meta-learning (1,029)
- Test-time adaptation (437)
- OOD and anomaly detection (836)
- Recognition and long-tail (6,306)
- Semi-supervised learning (897)
- Self-supervised and contrastive (1,267)
- Domain adaptation and generalization (2,012)
- Knowledge distillation (783)
