Research map / Label-efficient and robust learning
Zero-/few-shot and meta-learning: research map
1,029 papers on Zero-/few-shot and meta-learning in Label-efficient and robust learning, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 3 clusters and 11 approaches. The busiest year so far is 2023.
What it is
How can a model recognise a class it has seen only a few times, or never? Few-shot learning works from a handful of labelled examples, zero-shot learning from a description of the class, and meta-learning trains a model across many small tasks so that it learns new ones quickly.
Terms you will see on this page:
- N-way K-shot
- a test task with N new classes and K labelled examples of each.
- Support and query sets
- the few labelled examples given for a task, and the examples to be classified.
- MAML
- a meta-learning method that finds a starting point from which a few gradient steps solve a new task.
Start reading with the most cited papers here:
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations)
- Prototypical Networks for Few-shot Learning (NeurIPS 2017 · 5,178 citations)
Within Label-efficient and robust learning, its share shrank from 7.2% in 2023–24 to 5.8% in 2025–26 (206 → 176 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Zero-/few-shot and meta-learning in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Transductive and semantic-dictionary zero-shot rising: 9% → 22% of the topic (18 → 39 papers)
- Zero-shot robustness and transfer rising: 15% → 28% of the topic (31 → 49 papers)
- Generative feature synthesis for zero-shot rising: 8% → 10% of the topic (16 → 17 papers)
- Alignment-based few-shot metric learning falling: 9% → 2% of the topic (18 → 4 papers)
- Normalization and unsupervised meta-learning falling: 7% → 2% of the topic (14 → 4 papers)
Papers from 2025–2026 that are already cited most:
- FoundationStereo: Zero-Shot Stereo Matching (CVPR 2025 · 98 citations)
- PertEval-scFM: Benchmarking Single-Cell Foundation Models for Perturbation Effect Prediction (ICML 2025 · 40 citations)
- vesselFM: A Foundation Model for Universal 3D Blood Vessel Segmentation (CVPR 2025 · 23 citations)
- Few-shot Personalized Scanpath Prediction (CVPR 2025 · 6 citations)
Approaches and key papers
Zero-shot recognition · 536 papers
Keywords: zero shot, zsl, seen
Approaches in this cluster:
- Generative feature synthesis for zero-shot (200 papers)
Synthesizes unseen-class features and aligns semantic embeddings to reduce seen-class bias. - Zero-shot robustness and transfer (106 papers)
Studies and improves generalization of foundation models to new groups, domains and data. - Robust fine-tuning of zero-shot models (99 papers)
Improves robustness and prompt ensembles of zero-shot models, with theory for zero-shot prediction. - Transductive and semantic-dictionary zero-shot (131 papers)
Uses unlabeled test data and learned semantic dictionaries for zero-shot and CLIP transfer.
Most cited and most cited since 2024:
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · 1,178 citations)
- Feature Generating Networks for Zero-Shot Learning (CVPR 2018 · 1,080 citations)
- FoundationStereo: Zero-Shot Stereo Matching (CVPR 2025 · 98 citations)
- Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence (CVPR 2024 · 48 citations)
Few-shot classification · 310 papers
Keywords: fsl, shot classification, support
Approaches in this cluster:
- Embedding adaptation for few-shot classification (105 papers)
Adapts feature embeddings with set functions and feature-wise transforms, including cross-domain settings. - Fine-tuning strategies for few-shot (79 papers)
Revisits partial fine-tuning, auxiliary data and causal interventions for few-shot generalization. - Fine-grained and cross-domain few-shot baselines (80 papers)
Uses baselines, feature reconstruction and prototypical refinement for fine-grained or cross-domain few-shot tasks. - Alignment-based few-shot metric learning (46 papers)
Learns dynamic alignment, affiliation and task-tailored embeddings between support and query samples.
Most cited and most cited since 2024:
- Prototypical Networks for Few-shot Learning (NeurIPS 2017 · 5,178 citations)
- Dynamic Few-Shot Visual Learning Without Forgetting (CVPR 2018 · 1,151 citations)
- Cross-Layer and Cross-Sample Feature Optimization Network for Few-Shot Fine-Grained Image Classification (AAAI 2024 · 55 citations)
- APSeg: Auto-Prompt Network for Cross-Domain Few-Shot Semantic Segmentation (CVPR 2024 · 34 citations)
Meta-learning algorithms · 183 papers
Keywords: maml, meta algorithms, meta meta
Approaches in this cluster:
- Meta-augmentation and meta-objectives (70 papers)
Improves meta-learning generalization through task augmentation, contrastive objectives and target-model transfer. - Meta-learning analysis and benchmarks (67 papers)
Analyzes meta-learned representations and builds benchmarks and Bayesian or convex-optimization learners. - Normalization and unsupervised meta-learning (46 papers)
Adapts batch normalization for tasks and builds unsupervised or hard-sample meta-learning.
Most cited and most cited since 2024:
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (ICML 2017 · 6,288 citations)
- Meta-Learning With Differentiable Convex Optimization (CVPR 2019 · 1,317 citations)
- FREE: Faster and Better Data-Free Meta-Learning (CVPR 2024 · 5 citations)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · 4 citations)
Related topics in Label-efficient and robust learning
- Noisy labels (576)
- Test-time adaptation (437)
- OOD and anomaly detection (836)
- Recognition and long-tail (6,306)
- Semi-supervised learning (897)
- Self-supervised and contrastive (1,267)
- Domain adaptation and generalization (2,012)
- Knowledge distillation (783)
