Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Fine-tuning strategies for few-shot
Revisits partial fine-tuning, auxiliary data and causal interventions for few-shot generalization.
67 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Zero-/few-shot and meta-learning, keyword cluster “fsl · shot classification · support”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
21 papers in 2023–24 and 10 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 10.2% to 5.7%.
No longer reported by papers here: miniImageNet 5-way 5-shot (accuracy, last in 2023); miniImageNet 5-way 1-shot (accuracy, last in 2023); tieredImageNet 5-way 1-shot (accuracy, last in 2023); tieredImageNet 5-way 5-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 · One-Shot Imitation Learning (NeurIPS 2017 · 267 citations)
- 2018 · Learning to Compare: Relation Network for Few-Shot Learning (CVPR 2018 · 372 citations)
- 2019 · LEARNING TO PROPAGATE LABELS: TRANSDUCTIVE PROPAGATION NETWORK FOR FEW-SHOT LEARNING (ICLR 2019 · 517 citations)
- 2020 · Interventional Few-Shot Learning (NeurIPS 2020 · 122 citations)
- 2021 · Prototype Completion With Primitive Knowledge for Few-Shot Learning (CVPR 2021 · 162 citations)
- 2022 · ContrastNet: A Contrastive Learning Framework for Few-Shot Text Classification (AAAI 2022 · 103 citations)
- 2023 · The Unreasonable Effectiveness of Few-shot Learning for Machine Translation (ICML 2023 · 24 citations)
- 2024 · Meta-Exploiting Frequency Prior for Cross-Domain Few-Shot Learning (NeurIPS 2024 · 4 citations)
- 2025 · Few-shot Personalized Scanpath Prediction (CVPR 2025 · 6 citations)
- 2026 · CAOS: Conformal Aggregation of One-Shot Predictors (ICML 2026 · 0 citations)
Earliest papers here
- One-Shot Imitation Learning (NeurIPS 2017 · 267 citations)
- Few-Shot Learning Through an Information Retrieval Lens (NeurIPS 2017 · 65 citations)
- Learning to Compare: Relation Network for Few-Shot Learning (CVPR 2018 · 372 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)
11 papers, 2018–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 76.7 | TADAM: Task dependent adaptive metric for improved few-shot learning (NeurIPS 2018 · 194 citations) |
| 2021 | 85.63 | Reinforced Attention for Few-Shot Learning and Beyond (CVPR 2021 · 48 citations) |
| 2022 | 86.08 | FeLMi : Few shot Learning with hard Mixup (NeurIPS 2022 · 2 citations) |
miniImageNet 5-way 1-shot · accuracy (higher is better)
10 papers, 2018–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 58.5 | TADAM: Task dependent adaptive metric for improved few-shot learning (NeurIPS 2018 · 194 citations) |
| 2019 | 59.46 | LEARNING TO PROPAGATE LABELS: TRANSDUCTIVE PROPAGATION NETWORK FOR FEW-SHOT LEARNING (ICLR 2019 · 517 citations) |
| 2021 | 75.58 | Reinforced Attention for Few-Shot Learning and Beyond (CVPR 2021 · 48 citations) |
tieredImageNet 5-way 1-shot · accuracy (higher is better)
7 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 57.53 | LEARNING TO PROPAGATE LABELS: TRANSDUCTIVE PROPAGATION NETWORK FOR FEW-SHOT LEARNING (ICLR 2019 · 517 citations) |
| 2021 | 81.04 | Prototype Completion With Primitive Knowledge for Few-Shot Learning (CVPR 2021 · 162 citations) |
tieredImageNet 5-way 5-shot · accuracy (higher is better)
7 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 72.85 | LEARNING TO PROPAGATE LABELS: TRANSDUCTIVE PROPAGATION NETWORK FOR FEW-SHOT LEARNING (ICLR 2019 · 517 citations) |
| 2021 | 88.13 | ECKPN: Explicit Class Knowledge Propagation Network for Transductive Few-Shot Learning (CVPR 2021 · 68 citations) |
| 2023 | 88.75 | ProtoDiff: Learning to Learn Prototypical Networks by Task-Guided Diffusion (NeurIPS 2023 · 5 citations) |
What has been tried
57 papers here report 608 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 6%, hurt 20%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add contrastive loss to classification-loss training | OSIE · ScanMatch | 0.361 -> 0.375 | Helped | Few-shot Personalized Scanpath Prediction (CVPR 2025 · 6 citations) |
| Add proposed cross-attention attribute encoding to domain-specific prompts | Office-Home and Mini-DomainNet, 1-shot · H-score | 62.96 -> 64.04; 67.27 -> 67.57 | Helped | OSLoPrompt: Bridging Low-Supervision Challenges and Open-Set Domain Generalization in CLIP (CVPR 2025 · 4 citations) |
| Add the feature reconstruction prior | 5-way 1-shot, 8-domain average · accuracy (%) | 53.41 -> 54.16 | Helped | Meta-Exploiting Frequency Prior for Cross-Domain Few-Shot Learning (NeurIPS 2024 · 4 citations) |
| include consistency in task selection | Omniglot · accuracy | 72.18 -> 74.33 | Helped | Towards Few-Shot Adaptation of Foundation Models via Multitask Finetuning (ICLR 2024 · 1 citation) |
| Add the FLMI term to retrieval with class balancing and quality scoring | ImageNet 16-shot · accuracy (%) | 72.46 ± 0.03 -> 72.52 ± 0.11 | Helped | COBRA: COmBinatorial Retrieval Augmentation for Few-Shot Adaptation (CVPR 2025 · 0 citations) |
| update the temporal dynamic generation module during adaptation | Sth-Else · accuracy | Hurt | Learning Causal Domain-Invariant Temporal Dynamics for Few-Shot Action Recognition (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 10 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Mini-ImageNet · 3 of 10 papers
- MNIST · 2 of 10 papers
- CIFAR-10 · 2 of 10 papers
- CelebA · 2 of 10 papers
Baselines
- Random · 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 31 papers; from OpenAlex and the papers' affiliations).
- University of Chinese Academy of Sciences · 3 papers
- Google · 2 papers
- University of Wisconsin - Madison · 2 papers
- Nanyang Technological University · 2 papers
- Chinese Academy of Sciences · 2 papers
- Institute of Computing Technology · 2 papers
- University of California, Santa Barbara · 2 papers
Use it
Download BibTeX of 36 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
- Embedding adaptation for few-shot classification (71 papers)
- Fine-grained and cross-domain few-shot baselines (59 papers)
- Alignment-based few-shot metric learning (38 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)
- Meta-augmentation and meta-objectives (60 papers)
- Meta-learning analysis and benchmarks (50 papers)
- Normalization and unsupervised meta-learning (39 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).