Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Generative feature synthesis for zero-shot
Synthesizes unseen-class features and aligns semantic embeddings to reduce seen-class bias.
117 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Zero-/few-shot and meta-learning, keyword cluster “zero shot · zsl · seen”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
16 papers in 2023–24 and 17 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 7.8% to 9.7%.
Open questions the data points to
- Papers report TF-VAEGAN on CUB (H) anywhere from 56 to 71: comparisons against it depend on whose number you take.
- Papers report TF-VAEGAN on SUN (H) anywhere from 40 to 53.7: comparisons against it depend on whose number you take.
No longer reported by papers here: aPY (H, last in 2023); AWA1 (H, last in 2024).
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 · Zero-Shot Learning via Joint Latent Similarity Embedding (CVPR 2016 · 362 citations)
- 2017 · Semantic Autoencoder for Zero-Shot Learning (CVPR 2017 · 878 citations)
- 2018 · Feature Generating Networks for Zero-Shot Learning (CVPR 2018 · 1,080 citations)
- 2019 · Generalized Zero- and Few-Shot Learning via Aligned Variational Autoencoders (CVPR 2019 · 760 citations)
- 2020 · Fine-Grained Generalized Zero-Shot Learning via Dense Attribute-Based Attention (CVPR 2020 · 253 citations)
- 2021 · Contrastive Embedding for Generalized Zero-Shot Learning (CVPR 2021 · 286 citations)
- 2022 · MSDN: Mutually Semantic Distillation Network for Zero-Shot Learning (CVPR 2022 · 154 citations)
- 2023 · Progressive Semantic-Visual Mutual Adaption for Generalized Zero-Shot Learning (CVPR 2023 · 72 citations)
- 2024 · Visual-Augmented Dynamic Semantic Prototype for Generative Zero-Shot Learning (CVPR 2024 · 39 citations)
- 2025 · ZeroHAR: Sensor Context Augments Zero-Shot Wearable Action Recognition (AAAI 2025 · 4 citations)
- 2026 · Counterfactual-Driven Zero-Shot Classifier Expansion (AAAI 2026 · 3 citations)
Earliest papers here
- Zero-Shot Learning via Joint Latent Similarity Embedding (CVPR 2016 · 362 citations)
- Semi-Supervised Vocabulary-Informed Learning (CVPR 2016 · 120 citations)
- Transductive Zero-Shot Recognition via Shared Model Space Learning (AAAI 2016 · 109 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.
CUB · H (higher is better)
25 papers, 2018–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 73.2 | Transductive Unbiased Embedding for Zero-Shot Learning (CVPR 2018 · 233 citations) |
| 2021 | 73.5 | Dual Progressive Prototype Network for Generalized Zero-Shot Learning (NeurIPS 2021 · 22 citations) |
| 2023 | 77.3 | Deconstructed Generation-Based Zero-Shot Model (AAAI 2023 · 6 citations) |
AWA2 · H (higher is better)
20 papers, 2018–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 77.4 | Transductive Unbiased Embedding for Zero-Shot Learning (CVPR 2018 · 233 citations) |
| 2020 | 89.1 | Self-Supervised Domain-Aware Generative Network for Generalized Zero-Shot Learning (CVPR 2020 · 71 citations) |
| 2023 | 90.4 | Bi-Directional Distribution Alignment for Transductive Zero-Shot Learning (CVPR 2023 · 32 citations) |
SUN · H (higher is better)
18 papers, 2018–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 41.7 | Transductive Unbiased Embedding for Zero-Shot Learning (CVPR 2018 · 233 citations) |
| 2020 | 52.8 | Self-Supervised Domain-Aware Generative Network for Generalized Zero-Shot Learning (CVPR 2020 · 71 citations) |
| 2022 | 75.2 | Distinguishing Unseen From Seen for Generalized Zero-Shot Learning (CVPR 2022 · 45 citations) |
aPY · H (higher is better)
12 papers, 2018–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 22.6 | Zero-Shot Visual Recognition Using Semantics-Preserving Adversarial Embedding Networks (CVPR 2018) |
| 2019 | 56.2 | Generalized Zero-Shot Recognition Based on Visually Semantic Embedding (CVPR 2019 · 116 citations) |
AWA1 · H (higher is better)
10 papers, 2019–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 63.4 | Dual Adversarial Semantics-Consistent Network for Generalized Zero-Shot Learning (NeurIPS 2019 · 65 citations) |
| 2020 | 87.7 | Self-Supervised Domain-Aware Generative Network for Generalized Zero-Shot Learning (CVPR 2020 · 71 citations) |
| 2023 | 89.1 | Bi-Directional Distribution Alignment for Transductive Zero-Shot Learning (CVPR 2023 · 32 citations) |
CUB GZSL · H (higher is better)
10 papers, 2018–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 54 | Feature Generating Networks for Zero-Shot Learning (CVPR 2018 · 1,080 citations) |
| 2021 | 75.9 | Hardness Sampling for Self-Training Based Transductive Zero-Shot Learning (CVPR 2021 · 23 citations) |
What has been tried
105 papers here report 1,116 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 5%, hurt 15%, 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 |
|---|---|---|---|---|
| add VDKL visual prior | CUB · CZSL accuracy | 84.9 -> 86.8 | Helped | Visual-Augmented Dynamic Semantic Prototype for Generative Zero-Shot Learning (CVPR 2024 · 39 citations) |
| Add ID²SD and O²DBD | CUB · H | 54.7 -> 56.5 | Helped | Data Distribution Distilled Generative Model for Generalized Zero-Shot Recognition (AAAI 2024 · 26 citations) |
| Add FLPT prompt tuning instead of hand-crafted prompts | AWA2 · H | 90.57 -> 93.24 | Helped | Data-Free Generalized Zero-Shot Learning (AAAI 2024 · 18 citations) |
| Add the attribute prior rather than use no prior | UT-Zappos · AUC | 32.3 -> 33.4 | Helped | Revealing the Proximate Long-Tail Distribution in Compositional Zero-Shot Learning (AAAI 2024 · 17 citations) |
| Add the retrieval module | UT-Zappos closed-world · AUC | 39.4 -> 40.6 | Helped | Retrieval-Augmented Primitive Representations for Compositional Zero-Shot Learning (AAAI 2024 · 15 citations) |
| add diverse semantics from external class names and hierarchy taxonomy enhancement | CUB dissimilar · GZSL harmonic mean | 5.06 -> 47.44 | Helped | Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · 14 citations) |
| Increase the number of category anchors | News and review datasets, 5-way 5-shot · accuracy | Helped | Liberating Seen Classes: Boosting Few-Shot and Zero-Shot Text Classification via Anchor Generation and Classification Reframing (AAAI 2024 · 7 citations) | |
| add Stage I sensor-context training before Stage II | 18-dataset average · accuracy | 9.7 -> 45.6 | Helped | ZeroHAR: Sensor Context Augments Zero-Shot Wearable Action Recognition (AAAI 2025 · 4 citations) |
| add the GPA module to the backbone-only model (M=8) | NUS-Wide ZSL · mAP | 28.6 -> 32.4 | Helped | Epsilon: Exploring Comprehensive Visual-Semantic Projection for Multi-Label Zero-Shot Learning (AAAI 2025 · 3 citations) |
| Add the diffusion discriminator | AWA2, CUB, and SUN · ZSL Top-1 accuracy | 79.9 -> 82.3; 83.7 -> 84.1; 74.7 -> 75.1 | Helped | ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · 2 citations) |
| fine-tune the feature extraction network | AWA2 · H | 64.7 -> 68.2 | Helped | Class Semantic Attribute Perception Guided Zero-Shot Learning (AAAI 2025 · 1 citation) |
| add class diversity promotion | CUB GZSL · H | 55.2 -> 57.4 | Helped | GenZSL: Generative Zero-Shot Learning Via Inductive Variational Autoencoder (ICML 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 |
|---|---|---|---|---|
| TF-VAEGAN | CUB | H | 56 – 71 | 6 papers, e.g. CVPR 2021 and CVPR 2023 |
| TF-VAEGAN | SUN | H | 40 – 53.7 | 3 papers, e.g. CVPR 2021 and CVPR 2023 |
| CE-GZSL | CUB | H | 53 – 65.3 | 3 papers, e.g. AAAI 2022 and AAAI 2024 |
| TF-VAEGAN | SUN (GZSL) | H | 43 – 53.7 | 2 papers, e.g. AAAI 2024 and CVPR 2021 |
| SynC-struct | AwA generalized ZSL | AUSUC | 0.533 – 0.583 | 2 papers, e.g. CVPR 2017 and CVPR 2017 |
| E-PGN | AWA1 | H | 65.3 – 71.2 | 2 papers, e.g. AAAI 2021 and ICLR 2021 |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- AWA2 · 3 of 8 papers
- CUB · 3 of 8 papers
- SUN · 3 of 8 papers
Baselines
- CLIP · 3 of 8 papers
- CoOp · 3 of 8 papers
- CE-GZSL · 2 of 8 papers
- HSVA · 2 of 8 papers
- f-VAEGAN · 2 of 8 papers
- TF-VAEGAN · 2 of 8 papers
- VADS · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 33 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 3 papers
- Nanjing University of Science and Technology · 2 papers
- Beihang University · 2 papers
- Northwestern Polytechnical University · 2 papers
- Zhejiang University · 2 papers
- The University of Sydney · 2 papers
- Huazhong University of Science and Technology · 2 papers
- Australian National University · 2 papers
- Mohamed bin Zayed University of Artificial Intelligence · 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
- Zero-shot robustness and transfer (101 papers)
- Robust fine-tuning of zero-shot models (92 papers)
- Transductive and semantic-dictionary zero-shot (90 papers)
Elsewhere in the topic
- Embedding adaptation for few-shot classification (71 papers)
- Fine-tuning strategies for few-shot (67 papers)
- Meta-augmentation and meta-objectives (60 papers)
- Fine-grained and cross-domain few-shot baselines (59 papers)
- Meta-learning analysis and benchmarks (50 papers)
- Normalization and unsupervised meta-learning (39 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).