Research map / Label-efficient and robust learning
OOD and anomaly detection: research map
836 papers on OOD and anomaly detection 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 2 clusters and 8 approaches. The busiest year so far is 2026.
What it is
A model should notice when an input is unlike anything it was trained on, instead of confidently giving a wrong answer. Out-of-distribution (OOD) detection flags such inputs; anomaly detection finds rare defects or unusual events when examples of them are scarce or missing.
Terms you will see on this page:
- OOD detection
- deciding whether a test input comes from a different distribution than the training data.
- Anomaly detection
- finding rare, unusual samples, often learning only from normal ones.
- One-class learning
- training on normal data only and flagging whatever does not fit.
Start reading with the most cited papers here:
- A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (ICLR 2017 · 1,577 citations)
- CutPaste: Self-Supervised Learning for Anomaly Detection and Localization (CVPR 2021 · 1,006 citations)
Within Label-efficient and robust learning, its share grew from 7.4% in 2023–24 to 8.5% in 2025–26 (212 → 259 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore OOD and anomaly detection 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):
- Unified multi-class anomaly detection rising: 8% → 22% of the topic (18 → 56 papers)
- Anomaly detection with contaminated data rising: 8% → 12% of the topic (16 → 32 papers)
- Spurious-aware OOD detection rising: 15% → 18% of the topic (32 → 46 papers)
- Provable and feature-based OOD detection falling: 28% → 15% of the topic (59 → 39 papers)
- Out-of-distribution generalization theory falling: 16% → 6% of the topic (33 → 16 papers)
Papers from 2025–2026 that are already cited most:
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · 103 citations)
- On the Generalization of Handwritten Text Recognition Models (CVPR 2025 · 9 citations)
- Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection (CVPR 2025 · 8 citations)
Approaches and key papers
Out-of-distribution detection · 534 papers
Keywords: ood detection, distribution ood, distribution detection
Approaches in this cluster:
- Provable and feature-based OOD detection (167 papers)
Designs OOD scores and outlier training with theoretical guarantees using typical features and prototypes. - Spurious-aware OOD detection (158 papers)
Addresses spurious correlations and scaling in OOD detection with gradients and prototypes. - Out-of-distribution generalization theory (81 papers)
Analyzes and improves OOD generalization via invariance, sharpness and causal learning. - Uncertainty and misclassification detection (64 papers)
Detects errors and OOD inputs using baseline scores, generative models and outlier mixing. - Robust fine-tuning for OOD generalization (64 papers)
Fine-tunes vision-language models while preserving pretrained robustness using energy and calibration objectives.
Most cited and most cited since 2024:
- A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (ICLR 2017 · 1,577 citations)
- Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks (ICLR 2018 · 634 citations)
- Transcending Forgery Specificity with Latent Space Augmentation for Generalizable Deepfake Detection (CVPR 2024 · 152 citations)
- Debiasing Multimodal Sarcasm Detection with Contrastive Learning (AAAI 2024 · 35 citations)
Anomaly detection · 302 papers
Keywords: anomaly detection, normal, anomalies
Approaches in this cluster:
- Unified multi-class anomaly detection (152 papers)
Trains one anomaly detector across classes using reconstruction, cross-modal constraints and class-generalizable features. - Anomaly detection with contaminated data (87 papers)
Handles noisy training data and distribution shifts using clustering, ensembles and outlier exposure. - Deep one-class and tabular anomaly detection (63 papers)
Applies one-class objectives, self-supervision and reinforcement learning to tabular and general data.
Most cited and most cited since 2024:
- CutPaste: Self-Supervised Learning for Anomaly Detection and Localization (CVPR 2021 · 1,006 citations)
- Deep One-Class Classification (ICML 2018 · 847 citations)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · 198 citations)
- PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection (CVPR 2024 · 135 citations)
Related topics in Label-efficient and robust learning
- Noisy labels (576)
- Zero-/few-shot and meta-learning (1,029)
- Test-time adaptation (437)
- 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)
