Research map / Label-efficient and robust learning / OOD and anomaly detection
Out-of-distribution generalization theory
Analyzes and improves OOD generalization via invariance, sharpness and causal learning.
71 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in OOD and anomaly detection, keyword cluster “ood detection · distribution ood · distribution detection”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
33 papers in 2023–24 and 16 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 15.6% to 6.2%.
Open questions the data points to
- Papers report ERM on Terra Incognita (accuracy) anywhere from 42.6 to 49.8: comparisons against it depend on whose number you take.
No longer reported by papers here: Terra Incognita (accuracy (%), 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
- 2021 · DecAug: Out-of-Distribution Generalization via Decomposed Feature Representation and Semantic Augmentation (AAAI 2021 · 60 citations)
- 2022 · OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalization (CVPR 2022 · 80 citations)
- 2023 · Harnessing Out-Of-Distribution Examples via Augmenting Content and Style (ICLR 2023 · 10 citations)
- 2024 · Can Biases in ImageNet Models Explain Generalization? (CVPR 2024 · 21 citations)
- 2025 · In Search of Forgotten Domain Generalization (ICLR 2025 · 2 citations)
- 2026 · Boosting for Predictive Sufficiency (ICLR 2026 · 0 citations)
Earliest papers here
- DecAug: Out-of-Distribution Generalization via Decomposed Feature Representation and Semantic Augmentation (AAAI 2021 · 60 citations)
- Learning Causal Semantic Representation for Out-of-Distribution Prediction (NeurIPS 2021 · 47 citations)
- Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalization (NeurIPS 2021 · 44 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.
Terra Incognita · accuracy (%) (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 56.4 | Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalization (NeurIPS 2021 · 44 citations) |
What has been tried
62 papers here report 680 measured changes (a component added, removed or swapped, compared with a baseline): helped 63%, no clear effect 11%, hurt 20%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| increase the number of test domains used for oracle model selection from K=1 to K=3 | DomainNet and NICO++ · oracle-selection leakage | -35%; -38%; -42%; -79% | Helped | Rethinking the Evaluation Protocol of Domain Generalization (CVPR 2024 · 13 citations) |
| enable neural feature selector s(·) | CS-MNIST-COCO, ID · accuracy | 90.34 -> 91.72 | Helped | Diagnosing and Rectifying Fake OOD Invariance: A Restructured Causal Approach (AAAI 2024 · 5 citations) |
| use variance constraint instead of mean constraint | CIFAR10-C · ECE | 8.70±{0.2} -> 8.90±{0.2} | No clear effect | MaxEnt Loss: Constrained Maximum Entropy for Calibration under Out-of-Distribution Shift (AAAI 2024 · 4 citations) |
| Apply IRM to all three losses instead of center prediction loss | USPTO-50k_M mol-size OOD · Top-1 accuracy | 53.5% -> 53.7% ID Top-1; OOD Top-1: 36.9% -> 38.0% | Hurt | RetroOOD: Understanding Out-of-Distribution Generalization in Retrosynthesis Prediction (AAAI 2024 · 4 citations) |
| Use layers 4–1 rather than layer4 alone | CIFAR-10 · FPR95 | 23.50 -> 18.31 | Helped | Neuron Activation Coverage: Rethinking Out-of-distribution Detection and Generalization (ICLR 2024 · 4 citations) |
| add Label Smoothing during fine-tuning before weight averaging | Avg OOD (IN-V2, IN-R, IN-A, IN-S, ObjectNet) · accuracy | 63.0 -> 63.9 | Helped | Spurious Feature Diversification Improves Out-of-distribution Generalization (ICLR 2024 · 2 citations) |
| increase D-BAT hypothesis count from K=2 to K=5 | Waterbirds-CC · worst-group accuracy | 57.1 (± 3.7) -> 57.1 (± 3.7) | No clear effect | Unraveling the Key Components of OOD Generalization via Diversification (ICLR 2024 · 1 citation) |
| Increase the AHA labeling budget from 100 to 2000 | CIFAR-10-C + Texture · OOD Acc. | 79.77 -> 90.46 | Helped | AHA: Human-Assisted Out-of-Distribution Generalization and Detection (NeurIPS 2024 · 1 citation) |
| Increase image resolution from 32×32 to 64×64 | ImageNet-100; OOD datasets · % OOD performance retained | negli.~(0.002); 0.315 | No clear effect | What Variables Affect Out-of-Distribution Generalization in Pretrained Models? (NeurIPS 2024 · 1 citation) |
| Use an MLP selector instead of a linear selector | CelebA (ResNet18) · WG error | 49.6 (± 0.3) -> 47.7 (± 0.2) | Helped | OOD-Chameleon: Is Algorithm Selection for OOD Generalization Learnable? (ICML 2025 · 0 citations) |
| Add entropy regularization to the encoder | ImageNet-100; 8 OOD datasets average · OOD generalization error (%) | 44.56 -> 41.85 | Helped | Controlling Neural Collapse Enhances Out-of-Distribution Detection and Transfer Learning (ICML 2025 · 0 citations) |
| Vary hidden-layer size | ESS, CGSS, SST-2, and MNIST OOD · accuracy and F1 | Mixed | Factor Decorrelation Enhanced Data Removal from Deep Predictive Models (NeurIPS 2025 · 0 citations) |
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 |
|---|---|---|---|---|
| ERM | Terra Incognita | accuracy | 42.6 – 49.8 | 2 papers, e.g. CVPR 2022 and NeurIPS 2021 |
| ERM | Colored MNIST | accuracy | 27.3 – 29.9 | 2 papers, e.g. ICML 2022 and CVPR 2022 |
| ERM | PACS | accuracy (average) | 79.05 – 85.7 | 2 papers, e.g. AAAI 2021 and ICLR 2023 |
What reviewers will expect
Datasets and baselines named in the 12 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CelebA · 3 of 12 papers
- Colored-MNIST · 2 of 12 papers
- CIFAR-10 · 2 of 12 papers
- ImageNet-A · 2 of 12 papers
- ImageNet-R · 2 of 12 papers
Baselines
- Random selection · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 49 papers; from OpenAlex and the papers' affiliations).
- Huawei · 4 papers
- Tsinghua University · 4 papers
- City University of Hong Kong · 3 papers
- RIKEN · 3 papers
- Tencent · 3 papers
- University of Wisconsin - Madison · 3 papers
- University of Cambridge · 2 papers
- Peking University · 2 papers
- National University of Singapore · 2 papers
- Zhejiang University · 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
- Provable and feature-based OOD detection (122 papers)
- Spurious-aware OOD detection (109 papers)
- Uncertainty and misclassification detection (56 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).