Research map / Label-efficient and robust learning / Semi-supervised learning
Holistic semi-supervised pipelines
Combines consistency, augmentation and big self-supervised pretraining for semi-supervised classification.
125 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Semi-supervised learning, keyword cluster “semi · examples · classification”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
30 papers in 2023–24 and 22 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 18.9% to 13.9%.
Open questions the data points to
- Papers report StyleMatch on DigitsDG, 5 labels per class (accuracy (%)) anywhere from 55.7 to 71.9: comparisons against it depend on whose number you take.
- Papers report StyleMatch on DigitsDG, 10 labels per class (accuracy (%)) anywhere from 65.9 to 78.4: comparisons against it depend on whose number you take.
No longer reported by papers here: CIFAR-10, 250 labels (error rate, last in 2024); CIFAR-10, 4000 labels (error rate, last in 2022); CIFAR-100, 10000 labels (error rate, last in 2023); CIFAR-100, 400 labels (error rate, last in 2024); STL-10, 1000 labels (error rate, last in 2024); SVHN, 1000 labels (error rate, last in 2022).
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 · SAND: Semi-Supervised Adaptive Novel Class Detection and Classification over Data Stream (AAAI 2016 · 170 citations)
- 2017 · Temporal Ensembling for Semi-Supervised Learning (ICLR 2017 · 1,481 citations)
- 2018 · Smooth Neighbors on Teacher Graphs for Semi-Supervised Learning (CVPR 2018 · 279 citations)
- 2019 · Label Propagation for Deep Semi-Supervised Learning (CVPR 2019 · 690 citations)
- 2020 · Unsupervised Data Augmentation for Consistency Training (NeurIPS 2020 · 1,598 citations)
- 2021 · Adaptive Consistency Regularization for Semi-Supervised Transfer Learning (CVPR 2021 · 117 citations)
- 2022 · Dist-PU: Positive-Unlabeled Learning From a Label Distribution Perspective (CVPR 2022 · 48 citations)
- 2023 · PromptCAL: Contrastive Affinity Learning via Auxiliary Prompts for Generalized Novel Category Discovery (CVPR 2023 · 79 citations)
- 2024 · Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · 20 citations)
- 2025 · Taste More, Taste Better: Diverse Data and Strong Model Boost Semi-Supervised Crowd Counting (CVPR 2025 · 8 citations)
- 2026 · From Human Labels to Literature: Semi-Supervised Learning of NMR Chemical Shifts at Scale (ICML 2026 · 0 citations)
Earliest papers here
- SAND: Semi-Supervised Adaptive Novel Class Detection and Classification over Data Stream (AAAI 2016 · 170 citations)
- Supervised and Semi-Supervised Text Categorization using LSTM for Region Embeddings (ICML 2016 · 126 citations)
- Deconstructing the Ladder Network Architecture (ICML 2016 · 64 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.
CIFAR-10, 250 labels · error rate (lower is better)
6 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 11.08 | MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019 · 595 citations) |
| 2020 | 4.8 | Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (NeurIPS 2020 · 21 citations) |
| 2024 | 4.6 | OTMatch: Improving Semi-Supervised Learning with Optimal Transport (ICML 2024 · 0 citations) |
CIFAR-10, 4000 labels · error rate (lower is better)
5 papers, 2018–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 18.57 | Semi-Supervised Learning via Compact Latent Space Clustering (ICML 2018 · 41 citations) |
| 2019 | 6.24 | MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019 · 595 citations) |
| 2020 | 4.4 | Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (NeurIPS 2020 · 21 citations) |
| 2022 | 3.84 | Semi-supervised Learning with Multi-Head Co-Training (AAAI 2022 · 24 citations) |
CIFAR-100, 10000 labels · error rate (lower is better)
4 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 25.88 | MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019 · 595 citations) |
| 2020 | 21.35 | Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (NeurIPS 2020 · 21 citations) |
| 2023 | 19.92 | CHMATCH: Contrastive Hierarchical Matching and Robust Adaptive Threshold Boosted Semi-Supervised Learning (CVPR 2023 · 14 citations) |
CIFAR-100, 400 labels · error rate (lower is better)
4 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 43.36 | Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (NeurIPS 2020 · 21 citations) |
| 2022 | 42.71 | Hyperspherical Consistency Regularization (CVPR 2022 · 19 citations) |
| 2023 | 36.57 | CHMATCH: Contrastive Hierarchical Matching and Robust Adaptive Threshold Boosted Semi-Supervised Learning (CVPR 2023 · 14 citations) |
STL-10, 1000 labels · error rate (lower is better)
4 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 10.18 | MixMatch: A Holistic Approach to Semi-Supervised Learning (NeurIPS 2019 · 595 citations) |
| 2020 | 4.01 | Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (NeurIPS 2020 · 21 citations) |
SVHN, 1000 labels · error rate (lower is better)
4 papers, 2018–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 5.69 | Semi-Supervised Learning via Compact Latent Space Clustering (ICML 2018 · 41 citations) |
| 2020 | 1.96 | Unsupervised Semantic Aggregation and Deformable Template Matching for Semi-Supervised Learning (NeurIPS 2020 · 21 citations) |
What has been tried
110 papers here report 1,086 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 6%, hurt 14%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Tune hyperparameters on the target dataset rather than transfer them from TissueMNIST | AIROGS · balanced accuracy | 61.75 -> 77.14 | Helped | Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · 20 citations) |
| Add representation calibration with a targeted classifier | CIFAR-10 · All accuracy | 70.6 -> 94.4 | Helped | Targeted Representation Alignment for Open-World Semi-Supervised Learning (CVPR 2024 · 18 citations) |
| Add same-domain FBC | PACS, 5 labels per class · accuracy (%) | 73.4 -> 76.0 | Helped | Towards Generalizing to Unseen Domains with Few Labels (CVPR 2024 · 10 citations) |
| Add unknown-aware graph regularization to FixMatch | CIFAR-100 uncorrelated · error rate | 29.4 -> 23.6 | Helped | Unknown-Aware Graph Regularization for Robust Semi-supervised Learning from Uncurated Data (AAAI 2024 · 9 citations) |
| Add VSSM to the supervised regression setup | UCF-QNRF, 5% labeled · MAE | 115.35 -> 110.41 | Helped | Taste More, Taste Better: Diverse Data and Strong Model Boost Semi-Supervised Crowd Counting (CVPR 2025 · 8 citations) |
| Add RandMixUp to the supervised module | CIFAR-10, 40 labels · accuracy | 38.72 -> 47.93; 9.21% | Helped | Scaling Up Semi-supervised Learning with Unconstrained Unlabelled Data (AAAI 2024 · 7 citations) |
| Add ASM to the baseline, omitting APKT | 4C-A, 10% labeled · mAP | 53.1 -> 57.3 (+4.2) | Helped | Anatomical Knowledge Mining and Matching for Semi-supervised Medical Multi-structure Detection (AAAI 2025 · 3 citations) |
| Add instance-level similarity loss | 10-class RFF · overall accuracy | 21.10 -> 73.95 | Helped | Open-world Radio Frequency Fingerprint Identification via Augmented Semi-supervised Learning (AAAI 2025 · 3 citations) |
| Add variable selection network (VSN) | Nine tabular datasets, 20% label noise · F1 score | 8.770; 2.037; 4.201; 1.412; 0.877; 3.307; 3.344; -4.017; 4.947 | Mixed | Geodesic Flow Kernels for Semi-Supervised Learning on Mixed-Variable Tabular Dataset (AAAI 2025 · 2 citations) |
| add extractor-based local structure extraction | Amazon and DBLP · Fscore | 0.856 -> 0.873; 0.586 -> 0.613 | Helped | SLRL: Semi-Supervised Local Community Detection Based on Reinforcement Learning (AAAI 2025 · 2 citations) |
| Add ARC to the supervised regression model | UTKFace, 250 labels · MAE | 9.42 -> 7.06 | Helped | RankUp: Boosting Semi-Supervised Regression with an Auxiliary Ranking Classifier (NeurIPS 2024 · 2 citations) |
| Use decoupled discriminative learning instead of classifier-based or prototype-based pseudo-labeling | 2D-3D cross-modal retrieval · MAP | Helped | DREAM: Decoupled Discriminative Learning with Bigraph-aware Alignment for Semi-supervised 2D-3D Cross-modal Retrieval (AAAI 2025 · 1 citation) |
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 |
|---|---|---|---|---|
| StyleMatch | DigitsDG, 5 labels per class | accuracy (%) | 55.7 – 71.9 | 2 papers, e.g. CVPR 2024 and CVPR 2025 |
| StyleMatch | DigitsDG, 10 labels per class | accuracy (%) | 65.9 – 78.4 | 2 papers, e.g. CVPR 2024 and CVPR 2025 |
| ReMixMatch | CIFAR-100, 2500 labels | error rate | 26.03 – 27.43 | 2 papers, e.g. ICML 2024 and NeurIPS 2020 |
What reviewers will expect
Datasets and baselines named in the 10 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Office-Home · 2 of 10 papers
Baselines
- Supervised · 2 of 10 papers
- FixMatch · 2 of 10 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 52 papers; from OpenAlex and the papers' affiliations).
- Hong Kong University of Science and Technology · 3 papers
- Harbin Institute of Technology · 3 papers
- Korea Advanced Institute of Science and Technology · 2 papers
- Northeastern University · 2 papers
- Nanjing University · 2 papers
- Anhui University · 2 papers
- University of California, Los Angeles · 2 papers
- Carnegie Mellon University · 2 papers
- Mohamed bin Zayed University of AI · 2 papers
- Southeast University · 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
- Self-training theory (86 papers)
- Safe and realistic semi-supervised learning (70 papers)
- GAN-based semi-supervised learning (61 papers)
Elsewhere in the topic
- Confidence-based pseudo-label selection (134 papers)
- Semi-supervised medical segmentation (50 papers)
- Consistency training for semi-supervised segmentation (49 papers)
- Pseudo-labeling for specialized tasks (48 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).