Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Transductive and semantic-dictionary zero-shot
Uses unlabeled test data and learned semantic dictionaries for zero-shot and CLIP transfer.
90 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 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.
18 papers in 2023–24 and 39 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 8.7% to 22.2%.
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 · Latent Embeddings for Zero-Shot Classification (CVPR 2016 · 702 citations)
- 2017 · Low-Rank Embedded Ensemble Semantic Dictionary for Zero-Shot Learning (CVPR 2017 · 135 citations)
- 2018 · Towards Universal Representation for Unseen Action Recognition (CVPR 2018 · 107 citations)
- 2019 · F-VAEGAN-D2: A Feature Generating Framework for Any-Shot Learning (CVPR 2019 · 609 citations)
- 2020 · Domain-Aware Visual Bias Eliminating for Generalized Zero-Shot Learning (CVPR 2020 · 178 citations)
- 2021 · Light Field Super-Resolution With Zero-Shot Learning (CVPR 2021 · 73 citations)
- 2022 · Siamese Contrastive Embedding Network for Compositional Zero-Shot Learning (CVPR 2022 · 98 citations)
- 2023 · Pic2Word: Mapping Pictures to Words for Zero-Shot Composed Image Retrieval (CVPR 2023 · 135 citations)
- 2024 · Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · 1,178 citations)
- 2025 · FoundationStereo: Zero-Shot Stereo Matching (CVPR 2025 · 98 citations)
- 2026 · MSME: A Multi-Stage Multi-Expert Framework for Zero-Shot Stance Detection (AAAI 2026 · 1 citation)
Earliest papers here
- Latent Embeddings for Zero-Shot Classification (CVPR 2016 · 702 citations)
- Concepts Not Alone: Exploring Pairwise Relationships for Zero-Shot Video Activity Recognition (AAAI 2016 · 84 citations)
- Low-Rank Embedded Ensemble Semantic Dictionary for Zero-Shot Learning (CVPR 2017 · 135 citations)
What has been tried
68 papers here report 759 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 5%, hurt 20%, 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 large-scale pseudo-labeled unlabeled images without strong perturbations | Sintel · AbsRel | 0.492 -> 0.481 | Helped | Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · 1,178 citations) |
| Use DepthAnythingV2-L instead of DINOv2-L | Middlebury training set · BP-2 | 2.46 -> 1.97 | Helped | FoundationStereo: Zero-Shot Stereo Matching (CVPR 2025 · 98 citations) |
| add dense training objective | SPair-71k standard subset · PCK@0.10 | 74.6 -> 78.3 | Helped | Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence (CVPR 2024 · 48 citations) |
| add anchor-loss regularization to self-training | COCO, box prompt · mIoU | 58.88 -> 79.65 | Helped | Improving the Generalization of Segmentation Foundation Model under Distribution Shift via Weakly Supervised Adaptation (CVPR 2024 · 36 citations) |
| Use Dirichlet rather than diagonal-covariance Gaussian data fitting | Zero-shot, 11 datasets (average) · accuracy | 49.7 -> 65.8 | Helped | Transductive Zero-Shot and Few-Shot CLIP (CVPR 2024 · 27 citations) |
| Add monocular context without retraining the context backbone | Booster (Q) and Middlebury 2014 (H) · bad > 2 All (%) | 17.84 -> 15.85; 11.15 -> 14.96 | Mixed | Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail (CVPR 2025 · 26 citations) |
| Add self-distillation loss (LSD) | Sketchy Ext1 · mAP | 0.165 -> 0.548 | Helped | Asymmetric Mutual Alignment for Unsupervised Zero-Shot Sketch-Based Image Retrieval (AAAI 2024 · 14 citations) |
| Use aggregation degrees {1, 3, 5} rather than a single aggregation degree | MVTec AD · pixel-AUROC | 94.6 -> 97.3 | Helped | MuSc: Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images (ICLR 2024 · 7 citations) |
| Add retrieved data to few-shot finetuning | Nine datasets (4-, 8-, and 16-shot) · mean accuracy | +1.4; +0.3; +0.2 | Mixed | Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · 5 citations) |
| replace learned diverse residual filters with a 5×5 high-pass filter | ImageNet-1k vs ProGAN, SDXL, DALLE 3, MJ v6, Firefly · AUC | 0.986 -> 0.913 | Hurt | Forensic Self-Descriptions Are All You Need for Zero-Shot Detection, Open-Set Source Attribution, and Clustering of AI-generated Images (CVPR 2025 · 5 citations) |
| Add the depth module to the flow module | KITTI 00-10, nuScenes, Argoverse, GTA · ATE | 174.24 -> 123.42; 12.54 -> 8.40; 9.42 -> 5.71; 25.38 -> 12.37 | Helped | ZeroVO: Visual Odometry with Minimal Assumptions (CVPR 2025 · 5 citations) |
| replace three-branch CFE capsule construction with direct feature concatenation | VG150 Predicate classification · zR@50 | 37.7 -> 33.3 | Hurt | Navigating the Unseen: Zero-shot Scene Graph Generation via Capsule-Based Equivariant Features (CVPR 2025 · 4 citations) |
What reviewers will expect
Datasets and baselines named in the 20 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ImageNet · 4 of 20 papers
- StanfordCars · 4 of 20 papers
- Food101 · 4 of 20 papers
- FGVCAircraft · 3 of 20 papers
- ImageNet-A · 2 of 20 papers
- ImageNetV2 · 2 of 20 papers
- ImageNet-R · 2 of 20 papers
- ImageNet-Sketch · 2 of 20 papers
Baselines
- Aeroblade · 2 of 20 papers
- CLIP · 2 of 20 papers
- Selective-IGEV · 2 of 20 papers
- IGEV-Stereo · 2 of 20 papers
- NMRF · 2 of 20 papers
- RAFT-Stereo · 2 of 20 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 57 papers; from OpenAlex and the papers' affiliations).
- Beijing University of Posts and Telecommunications · 2 papers
- University of Oxford · 2 papers
- Google · 2 papers
- Boston University · 2 papers
- The Chinese University of Hong Kong · 2 papers
- Shanghai Jiao Tong University · 2 papers
- NVIDIA · 2 papers
- EPFL · 2 papers
Use it
Download BibTeX of 38 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
- Generative feature synthesis for zero-shot (117 papers)
- Zero-shot robustness and transfer (101 papers)
- Robust fine-tuning of zero-shot models (92 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).