Research map / Label-efficient and robust learning / Self-supervised and contrastive
Representation evaluation and trade-offs
Evaluates universality, label efficiency and discriminative structure of learned representations.
37 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Self-supervised and contrastive, keyword cluster “supervised contrastive · contrastive self · negative”
State History Tried Checklist 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.
16 papers in 2023–24 and 5 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 4.9% to 1.6%.
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
- 2017 · Learning Discrete Representations via Information Maximizing Self-Augmented Training (ICML 2017 · 291 citations)
- 2018 · An efficient framework for learning sentence representations (ICLR 2018 · 426 citations)
- 2020 · On Mutual Information Maximization for Representation Learning (ICLR 2020 · 274 citations)
- 2021 · Learning and Evaluating Representations for Deep One-Class Classification (ICLR 2021 · 105 citations)
- 2022 · TreeMoCo: Contrastive Neuron Morphology Representation Learning (NeurIPS 2022 · 4 citations)
- 2023 · Protein Representation Learning by Geometric Structure Pretraining (ICLR 2023 · 40 citations)
- 2024 · DenoSent: A Denoising Objective for Self-Supervised Sentence Representation Learning (AAAI 2024 · 6 citations)
- 2025 · Principled Data Augmentation for Learning to Solve Quadratic Programming Problems (NeurIPS 2025 · 0 citations)
- 2026 · Towards Hierarchy–Uniformity Equilibrium: Recovering Semantic Depth in Hypergraph Contrastive Learning (ICML 2026)
Earliest papers here
- Learning Discrete Representations via Information Maximizing Self-Augmented Training (ICML 2017 · 291 citations)
- Self-Paced Learning: An Implicit Regularization Perspective (AAAI 2017 · 48 citations)
- An efficient framework for learning sentence representations (ICLR 2018 · 426 citations)
What has been tried
29 papers here report 332 measured changes (a component added, removed or swapped, compared with a baseline): helped 70%, no clear effect 6%, hurt 21%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add denoising objective | Average STS · Spearman correlation (×100) | 77.96 -> 79.33 | Helped | DenoSent: A Denoising Objective for Self-Supervised Sentence Representation Learning (AAAI 2024 · 6 citations) |
| Increase weight decay | MNIST-on-CIFAR-10 with p_drop=1 · digit classification accuracy | Hurt | Investigating the Benefits of Projection Head for Representation Learning (ICLR 2024 · 3 citations) | |
| use Full Mask instead of 80-10-10 masking | downstream tasks · performance | Helped | CODE REPRESENTATION LEARNING AT SCALE (ICLR 2024 · 1 citation) | |
| Add label smoothing without input averaging | FMNIST-INR, 1 View · test accuracy | 68.30 ± 0.62 -> 69.15 ± 0.51 | Helped | Improved Generalization of Weight Space Networks via Augmentations (ICML 2024 · 1 citation) |
| apply node dropping, edge perturbation, feature masking (GraphCL-style augmentations) | LP and QP · mean relative objective error (%) | e.g., LP 10% Drop node 8.618 vs None 8.539; LP 100% Drop node 4.654 vs 2.784 | Hurt | Principled Data Augmentation for Learning to Solve Quadratic Programming Problems (NeurIPS 2025 · 0 citations) |
| replace GPT-2-XL backbone with T5-Small | FEVER dev · label accuracy | Hurt | Unsupervised Pretraining for Fact Verification by Language Model Distillation (ICLR 2024 · 0 citations) | |
| use multiple balancing iterations instead of one | CIFAR-10, CIFAR-100, and STL-10 zero-shot classification · average per-class recall | Helped | The Benefits of Balance: From Information Projections to Variance Reduction (NeurIPS 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 12 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 3 of 12 papers
- MNIST · 2 of 12 papers
- CIFAR-100 · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 21 papers; from OpenAlex and the papers' affiliations).
- University of Cambridge · 3 papers
- Mila · 2 papers
- IBM · 2 papers
- Amazon · 2 papers
Use it
Download BibTeX of 31 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
- Contrastive versus non-contrastive theory (74 papers)
- Inductive bias and minimal representations (66 papers)
- Empirical contrastive visual learning (38 papers)
Elsewhere in the topic
- Invariance-based self-supervised learning (168 papers)
- Pretext-task visual pre-training (136 papers)
- Data-efficient and robust self-supervision (122 papers)
- Self-supervision in supervised pipelines (121 papers)
- Self-supervised neuroimaging representations (98 papers)
- Self-supervised speech and brain decoding (56 papers)
- Masked autoencoder pretraining (48 papers)
- Self-supervised speech representations (24 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).