Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta-augmentation and meta-objectives
Improves meta-learning generalization through task augmentation, contrastive objectives and target-model transfer.
60 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Zero-/few-shot and meta-learning, keyword cluster “maml · meta algorithms · meta meta”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 6 tracked benchmarks.
13 papers in 2023–24 and 6 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.3% to 3.4%.
Open questions the data points to
- miniImageNet 5-way 5-shot (accuracy): best went from 82.53 (up to 2024) to 92.14 in 2025 — Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 0 citations)
- miniImageNet 5-way 1-shot (accuracy): best went from 68.03 (up to 2024) to 76.51 in 2025 — Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 0 citations)
- Papers report MAML on Mini-ImageNet 5-way 5-shot (accuracy) anywhere from 51.95 to 70.02: comparisons against it depend on whose number you take.
- Papers report MAML on Omniglot 20-way 1-shot (accuracy) anywhere from 87.4 to 97.21: comparisons against it depend on whose number you take.
No longer reported by papers here: Omniglot 20-way 1-shot (accuracy, last in 2023); Omniglot 20-way 5-shot (accuracy, last in 2023); Omniglot 5-way 1-shot (accuracy, last in 2023).
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 · Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations)
- 2018 · A Simple Neural Attentive Meta-Learner (ICLR 2018 · 875 citations)
- 2019 · Task Agnostic Meta-Learning for Few-Shot Learning (CVPR 2019 · 490 citations)
- 2020 · Meta-Learning without Memorization (ICLR 2020 · 102 citations)
- 2021 · Meta-Transfer Learning for Low-Resource Abstractive Summarization (AAAI 2021 · 33 citations)
- 2022 · Meta-Learning with Fewer Tasks through Task Interpolation (ICLR 2022 · 11 citations)
- 2023 · Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · 14 citations)
- 2024 · Context-Aware Meta-Learning (ICLR 2024 · 3 citations)
- 2025 · Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · 0 citations)
- 2026 · Universal Algorithm-Implicit Learning (ICML 2026 · 0 citations)
Earliest papers here
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations)
- Meta Networks (ICML 2017 · 211 citations)
- Neural Program Meta-Induction (NeurIPS 2017 · 0 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.
miniImageNet 5-way 5-shot · accuracy (higher is better)
17 papers, 2017–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 63.11 | Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations) |
| 2018 | 68.88 | A Simple Neural Attentive Meta-Learner (ICLR 2018 · 875 citations) |
| 2020 | 78.64 | MATE: Plugging in Model Awareness to Task Embedding for Meta Learning (NeurIPS 2020 · 4 citations) |
| 2021 | 82.53 | Towards Enabling Meta-Learning from Target Models (NeurIPS 2021 · 0 citations) |
| 2025 | 92.14 | Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 0 citations) |
miniImageNet 5-way 1-shot · accuracy (higher is better)
15 papers, 2017–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 48.7 | Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations) |
| 2018 | 55.71 | A Simple Neural Attentive Meta-Learner (ICLR 2018 · 875 citations) |
| 2020 | 62.08 | MATE: Plugging in Model Awareness to Task Embedding for Meta Learning (NeurIPS 2020 · 4 citations) |
| 2021 | 68.03 | Towards Enabling Meta-Learning from Target Models (NeurIPS 2021 · 0 citations) |
| 2025 | 76.51 | Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 0 citations) |
Omniglot 20-way 1-shot · accuracy (higher is better)
9 papers, 2017–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 95.8 | Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations) |
| 2018 | 97.64 | A Simple Neural Attentive Meta-Learner (ICLR 2018 · 875 citations) |
| 2023 | 97.77 | Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · 14 citations) |
Omniglot 20-way 5-shot · accuracy (higher is better)
7 papers, 2017–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 98.9 | Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations) |
| 2018 | 99.36 | A Simple Neural Attentive Meta-Learner (ICLR 2018 · 875 citations) |
Omniglot 5-way 1-shot · accuracy (higher is better)
5 papers, 2017–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2017 | 98.7 | Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations) |
| 2018 | 99.07 | A Simple Neural Attentive Meta-Learner (ICLR 2018 · 875 citations) |
tieredImageNet 5-way 5-shot · accuracy (higher is better)
5 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 82.48 | Meta-Learning with Adaptive Hyperparameters (NeurIPS 2020 · 6 citations) |
| 2021 | 86.91 | Towards Enabling Meta-Learning from Target Models (NeurIPS 2021 · 0 citations) |
| 2025 | 87.05 | Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · 0 citations) |
What has been tried
54 papers here report 604 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 9%, hurt 16%, 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 |
|---|---|---|---|---|
| Replace fixed ELMES class embeddings with learnable class embeddings | Few-shot benchmarks · accuracy | 85.5\mathbf{±.1} -> 85.9\mathbf{±.1}; 63.3±.3 -> 66.3\mathbf{±.2} | Mixed | Context-Aware Meta-Learning (ICLR 2024 · 3 citations) |
| Add semantic classifier to any-way MAML | miniImageNet, 3-way 1-shot · accuracy | 63.15 (0.52) -> 64.25 (1.42) | Helped | Any-Way Meta Learning (AAAI 2024 · 2 citations) |
| vary ConML subset sampling number from K=1 to K=16 | miniImageNet 5-way 1-shot · accuracy (%) | 56.25 -> 57.59 | Helped | Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · 0 citations) |
| Train CAMeLU with ImageNet-964 + MSCOCO + Fungi instead of ImageNet-964 alone | Cross-domain few-shot classification · accuracy | No clear effect | Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · 0 citations) |
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 |
|---|---|---|---|---|
| MAML | Mini-ImageNet 5-way 5-shot | accuracy | 51.95 – 70.02 | 4 papers, e.g. ICML 2022 and CVPR 2023 |
| MAML | Omniglot 20-way 1-shot | accuracy | 87.4 – 97.21 | 3 papers, e.g. ICML 2022 and CVPR 2023 |
| MAML | Mini-ImageNet 5-way 1-shot | accuracy | 32.93 – 48.7 | 2 papers, e.g. ICML 2022 and CVPR 2019 |
| MetaOptNet | Mini-ImageNet 5-way 1-shot | accuracy | 53.23 – 62.64 | 2 papers, e.g. CVPR 2023 and ICML 2021 |
What reviewers will expect
Datasets and baselines named in the 5 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- miniImageNet · 4 of 5 papers
- TieredImageNet · 3 of 5 papers
- CUB · 3 of 5 papers
- Fungi · 2 of 5 papers
- MSCOCO · 2 of 5 papers
- CIFAR-fs · 2 of 5 papers
- Aircraft · 2 of 5 papers
- meta-iNat · 2 of 5 papers
Baselines
- ProtoNet · 3 of 5 papers
- P>M>F · 2 of 5 papers
- CAML · 2 of 5 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 19 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 5 papers
- Stanford University · 2 papers
Use it
Download BibTeX of 29 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
- Meta-learning analysis and benchmarks (50 papers)
- Normalization and unsupervised meta-learning (39 papers)
Elsewhere in the topic
- Generative feature synthesis for zero-shot (117 papers)
- Zero-shot robustness and transfer (101 papers)
- Robust fine-tuning of zero-shot models (92 papers)
- Transductive and semantic-dictionary zero-shot (90 papers)
- Embedding adaptation for few-shot classification (71 papers)
- Fine-tuning strategies for few-shot (67 papers)
- Fine-grained and cross-domain few-shot baselines (59 papers)
- Alignment-based few-shot metric learning (38 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).