Research map / Label-efficient and robust learning
Semi-supervised learning: research map
897 papers on Semi-supervised learning 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 8 approaches. The busiest year so far is 2023.
What it is
Labelling data is expensive, but unlabeled data is plentiful. Semi-supervised learning trains with a small labelled set plus a large unlabeled one, mostly by letting the model label the unlabeled data itself (pseudo-labelling) and by asking for consistent predictions under different augmentations.
Terms you will see on this page:
- Pseudo-labelling
- using the model's confident predictions on unlabeled data as training labels.
- Consistency regularisation
- requiring the same prediction for two perturbed versions of an input.
- Self-training
- repeatedly training a model on its own pseudo-labels.
Start reading with the most cited papers here:
- Unsupervised Data Augmentation for Consistency Training (NeurIPS 2020 · 1,598 citations)
- Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results (NeurIPS 2017 · 1,557 citations)
Within Label-efficient and robust learning, its share held steady from 5.5% in 2023–24 to 5.2% in 2025–26 (159 → 158 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Semi-supervised learning 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):
- Confidence-based pseudo-label selection rising: 23% → 32% of the topic (37 → 50 papers)
- Semi-supervised medical segmentation rising: 9% → 15% of the topic (15 → 24 papers)
- Pseudo-labeling for specialized tasks rising: 9% → 11% of the topic (15 → 18 papers)
- Safe and realistic semi-supervised learning falling: 14% → 6% of the topic (22 → 10 papers)
- Self-training theory falling: 14% → 8% of the topic (22 → 13 papers)
Papers from 2025–2026 that are already cited most:
- Relaxed Class-consensus Consistency for Semi-supervised Semantic Segmentation (AAAI 2025 · 7 citations)
Approaches and key papers
Pseudo-labels and semi-supervised segmentation · 451 papers
Keywords: pseudo labels, segmentation, pseudo labeling
Approaches in this cluster:
- Confidence-based pseudo-label selection (192 papers)
Selects reliable pseudo-labels using uncertainty, local variance and bias-adaptive classifiers. - Consistency training for semi-supervised segmentation (92 papers)
Uses cross pseudo supervision and consistency with unreliable-label handling for segmentation. - Semi-supervised medical segmentation (82 papers)
Combines foundation models, density-aware contrast and variance analysis for medical image segmentation. - Pseudo-labeling for specialized tasks (85 papers)
Applies pseudo-label teachers to mirror detection, referring expressions, motion and rotation regression.
Most cited and most cited since 2024:
- Semi-Supervised Semantic Segmentation With Cross Pseudo Supervision (CVPR 2021 · 1,204 citations)
- Semi-Supervised Semantic Segmentation With Cross-Consistency Training (CVPR 2020 · 981 citations)
- Adaptive Bidirectional Displacement for Semi-Supervised Medical Image Segmentation (CVPR 2024 · 110 citations)
- Constructing and Exploring Intermediate Domains in Mixed Domain Semi-supervised Medical Image Segmentation (CVPR 2024 · 45 citations)
Semi-supervised classification · 446 papers
Keywords: semi, examples, classification
Approaches in this cluster:
- Holistic semi-supervised pipelines (160 papers)
Combines consistency, augmentation and big self-supervised pretraining for semi-supervised classification. - Self-training theory (101 papers)
Analyzes self-training, contrastive learning and unlabeled data with theoretical guarantees. - Safe and realistic semi-supervised learning (98 papers)
Evaluates and debiases use of unlabeled samples, including moderately confident ones. - GAN-based semi-supervised learning (87 papers)
Uses generative adversarial models and perturbation regularization to exploit unlabeled data.
Most cited and most cited since 2024:
- Unsupervised Data Augmentation for Consistency Training (NeurIPS 2020 · 1,598 citations)
- Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results (NeurIPS 2017 · 1,557 citations)
- A Semi-supervised Nighttime Dehazing Baseline with Spatial-Frequency Aware and Realistic Brightness Constraint (CVPR 2024 · 74 citations)
- Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · 20 citations)
Related topics in Label-efficient and robust learning
- Noisy labels (576)
- Zero-/few-shot and meta-learning (1,029)
- Test-time adaptation (437)
- OOD and anomaly detection (836)
- Recognition and long-tail (6,306)
- Self-supervised and contrastive (1,267)
- Domain adaptation and generalization (2,012)
- Knowledge distillation (783)
