Research map / Label-efficient and robust learning / OOD and anomaly detection
Uncertainty and misclassification detection
Detects errors and OOD inputs using baseline scores, generative models and outlier mixing.
56 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in OOD and anomaly detection, keyword cluster “ood detection · distribution ood · distribution detection”
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.
19 papers in 2023–24 and 23 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 9.0% to 8.9%.
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 · A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (ICLR 2017 · 1,577 citations)
- 2020 · CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (NeurIPS 2020 · 266 citations)
- 2021 · Multidimensional Uncertainty-Aware Evidential Neural Networks (AAAI 2021 · 22 citations)
- 2022 · Mitigating Neural Network Overconfidence with Logit Normalization (ICML 2022 · 74 citations)
- 2023 · Implicit Identity Leakage: The Stumbling Block to Improving Deepfake Detection Generalization (CVPR 2023 · 202 citations)
- 2024 · Transcending Forgery Specificity with Latent Space Augmentation for Generalizable Deepfake Detection (CVPR 2024 · 152 citations)
- 2025 · Early Concept Drift Detection via Prediction Uncertainty (AAAI 2025 · 15 citations)
- 2026 · GEM-FI: Gated Evidential Mixtures with Fisher Modulation (ICML 2026 · 0 citations)
Earliest papers here
- A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (ICLR 2017 · 1,577 citations)
- CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances (NeurIPS 2020 · 266 citations)
- Your classifier is secretly an energy based model and you should treat it like one (ICLR 2020 · 81 citations)
What has been tried
40 papers here report 426 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 9%, hurt 19%, 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 within-domain and cross-domain latent augmentation | Average across CDF-v1, CDF-v2, DFDCP, DFDC · AUC | 0.755 -> 0.825 | Helped | Transcending Forgery Specificity with Latent Space Augmentation for Generalizable Deepfake Detection (CVPR 2024 · 152 citations) |
| Replace Ei-kMeans with Adaptive PU-index Bucketing | 9 datasets, 3 classifiers · average accuracy | Helped | Early Concept Drift Detection via Prediction Uncertainty (AAAI 2025 · 15 citations) | |
| Add a patch-shuffled image branch to the original-image branch | Validation set, ablation groups 2 and 4 · total mean accuracy | 0.858 -> 0.913 | Helped | D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · 13 citations) |
| Train with AutoAugOOD-generated negative pairs instead of omitting negative data | MNIST, CIFAR-10, MVTecAD (mean) · AUROC (%) | 77.2 -> 97.3 | Helped | Universal Novelty Detection Through Adaptive Contrastive Learning (CVPR 2024 · 10 citations) |
| use only Xception representation rather than the full FakeDiffer information sources | FF++ C23; train F2F+FS+NT, test DF C23 · AUC | 98.47 -> 93.35 | Hurt | FakeDiffer: Distributional Disparity Learning on Differentiated Reconstruction for Face Forgery Detection (AAAI 2025 · 5 citations) |
| Add selective ReLU removal to knowledge distillation | CIFAR-10 OOD average · MMC | 76.81 -> 76.48 | Helped | Simplification Is All You Need against Out-of-Distribution Overconfidence (CVPR 2025 · 5 citations) |
| set training hyperparameter δ from 0.5 to 0.875 | CIFAR10 test · accuracy (%) | 93.74 -> 94.57 | Helped | Credal Deep Ensembles for Uncertainty Quantification (NeurIPS 2024 · 5 citations) |
| Add hyper-opinion to multinomial-opinion without opinion projection | Flower-102 · AUROC | 67.23 -> 97.61 | Helped | Hyper-opinion Evidential Deep Learning for Out-of-Distribution Detection (NeurIPS 2024 · 4 citations) |
| Add EXP (alternative exponential parameterization) to EDL | CIFAR-10 · test accuracy | 83.55 ± 0.6 -> 88.59 ± 0.4 | Helped | Uncertainty Estimation by Density Aware Evidential Deep Learning (ICML 2024 · 2 citations) |
| increase the number of non-singleton focal sets to the full power set rather than using a limited budget | classification; CIFAR-10 reported for the K study · test accuracy | Hurt | Random-Set Neural Networks (ICLR 2025 · 0 citations) | |
| Increase the tuning-split size | CIFAR-10 and CIFAR-100 · FPR at 95% TPR | Helped | A Data-Driven Measure of Relative Uncertainty for Misclassification Detection (ICLR 2024 · 0 citations) | |
| use dynamic magnitude T(τ) rather than fixed T with cross-entropy | CIFAR100, Old FD · AURC | 99.60 -> 94.33 | Helped | Typicalness-Aware Learning for Failure Detection (NeurIPS 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 14 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 6 of 14 papers
- CIFAR-100 · 4 of 14 papers
- SVHN · 3 of 14 papers
- TinyImageNet · 3 of 14 papers
- Places365 · 2 of 14 papers
- iNaturalist · 2 of 14 papers
- ImageNet-1k · 2 of 14 papers
- MNIST · 2 of 14 papers
Baselines
- MSP · 3 of 14 papers
- Energy · 2 of 14 papers
- ReAct · 2 of 14 papers
- Deep Ensembles · 2 of 14 papers
- MC Dropout · 2 of 14 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 42 papers; from OpenAlex and the papers' affiliations).
- Harbin Institute of Technology · 3 papers
- Nanyang Technological University · 3 papers
- Chinese Academy of Sciences · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- The Chinese University of Hong Kong · 2 papers
- Princeton University · 2 papers
- Mila · 2 papers
- Purdue University · 2 papers
- University of Science and Technology of China · 2 papers
Use it
Download BibTeX of 35 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
- Provable and feature-based OOD detection (122 papers)
- Spurious-aware OOD detection (109 papers)
- Out-of-distribution generalization theory (71 papers)
- Robust fine-tuning for OOD generalization (56 papers)
Elsewhere in the topic
- Unified multi-class anomaly detection (86 papers)
- Anomaly detection with contaminated data (64 papers)
- Deep one-class and tabular anomaly detection (50 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).