Research map / Label-efficient and robust learning / Domain adaptation and generalization
Augmentation and mixture-of-experts generalization
Improves open-domain generalization via augmentation, meta-learning and sparse experts.
51 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Domain adaptation and generalization, keyword cluster “domain generalization · dg · unseen”
State History Tried Baselines 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.
16 papers in 2023–24 and 15 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 4.7% to 4.7%.
Open questions the data points to
- Papers report Baseline on Gaze360→EyeDiap (angular error) anywhere from 9 to 11.42: comparisons against it depend on whose number you take.
- Papers report Baseline on Gaze360→MPIIGaze (angular error) anywhere from 8.18 to 9.89: comparisons against it depend on whose number you take.
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 · Learning Attributes Equals Multi-Source Domain Generalization (CVPR 2016 · 175 citations)
- 2019 · Learning Robust Representations by Projecting Superficial Statistics Out (ICLR 2019 · 90 citations)
- 2020 · Regularized Fine-Grained Meta Face Anti-Spoofing (AAAI 2020 · 201 citations)
- 2021 · RobustNet: Improving Domain Generalization in Urban-Scene Segmentation via Instance Selective Whitening (CVPR 2021 · 326 citations)
- 2022 · Style Neophile: Constantly Seeking Novel Styles for Domain Generalization (CVPR 2022 · 103 citations)
- 2023 · Rethinking Domain Generalization for Face Anti-Spoofing: Separability and Alignment (CVPR 2023 · 100 citations)
- 2024 · CFPL-FAS: Class Free Prompt Learning for Generalizable Face Anti-spoofing (CVPR 2024 · 79 citations)
- 2025 · Improving Generalization for AI-Synthesized Voice Detection (AAAI 2025 · 7 citations)
- 2026 · Causal Decoupling Domain Generalization for Remote Sensing Change Detection (AAAI 2026 · 2 citations)
Earliest papers here
- Learning Attributes Equals Multi-Source Domain Generalization (CVPR 2016 · 175 citations)
- Learning Robust Representations by Projecting Superficial Statistics Out (ICLR 2019 · 90 citations)
- Compact Feature Learning for Multi-Domain Image Classification (CVPR 2019 · 29 citations)
What has been tried
43 papers here report 509 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 6%, hurt 13%, 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 prompt-text matched (PTM) supervision to the prompt-learning baseline | Protocol 2 average on all sub-protocols · HTER | 8.11 -> 7.50 | Helped | CFPL-FAS: Class Free Prompt Learning for Generalizable Face Anti-spoofing (CVPR 2024 · 79 citations) |
| Add Lsem | PACS · H-score | 92.56 -> 99.70 | Helped | Unknown Prompt the only Lacuna: Unveiling CLIP's Potential for Open Domain Generalization (CVPR 2024 · 17 citations) |
| Use MS-SSIM instead of L1, L2, or SSIM for the frontalization loss | Cross-dataset gaze estimation · angular error (degrees) | L1 7.28, 7.40, 6.76, 7.47; L2 7.12, 8.25, 7.18, 8.19; SSIM 6.78, 8.59, 6.33, 7.56; MS-SSIM 5.72, 6.97, 7.18, 7.38 | Mixed | Gaze from Origin: Learning for Generalized Gaze Estimation by Embedding the Gaze Frontalization Process (AAAI 2024 · 8 citations) |
| add reconstruction module | LibriSeVoc seen vocoders · EER (%) | 20.21 -> 19.16 | Mixed | Improving Generalization for AI-Synthesized Voice Detection (AAAI 2025 · 7 citations) |
| Use the gaze-target rotation and translation alignment function rather than offset, linear, affine, or homography mappings | MPIIFaceGaze + EyeDiap average · angular error (°) | 6.95, 6.87, 6.81, 6.80 -> 6.76 | Helped | Gaze Label Alignment: Alleviating Domain Shift for Gaze Estimation (AAAI 2025 · 4 citations) |
| add State Style Randomization (SSR) | PACS · average accuracy | 89.4 -> 90.4; 1.0% | Helped | DGFamba: Learning Flow Factorized State Space for Visual Domain Generalization (AAAI 2025 · 3 citations) |
| Add a domain-specific prompt to the domain-invariant prompt | Camelyon17 · A-Acc | 85.94 -> 86.24 | Helped | Memory-Efficient Prompt Tuning for Incremental Histopathology Classification (AAAI 2024 · 3 citations) |
| add the dual-branch adapter | SDGOD Avg. · mAP | 59.8 -> 60.7 | Helped | Towards Single-Source Domain Generalized Object Detection via Causal Visual Prompts (NeurIPS 2025 · 2 citations) |
| Add Robust-PT initialization | Robust UDA tasks across perturbation budgets · accuracy | Helped | TAROT: Towards Essentially Domain-Invariant Robustness with Theoretical Justification (CVPR 2025 · 1 citation) | |
| Train on a subset of queried test-domain classes rather than excluding the test domain | Clipart and Sketch · balanced top-1 accuracy | Clipart: 27.4 -> 17.1; Sketch: 30.1 -> 19.2 | Hurt | When and How Does CLIP Enable Domain and Compositional Generalization? (ICML 2025 · 0 citations) |
| add original PatchMix | DomainNet Real+Painting, Painting · ACC All | 34.5 -> 34.8 | Helped | HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts (ICLR 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 |
|---|---|---|---|---|
| Baseline | Gaze360→EyeDiap | angular error | 9 – 11.42 | 2 papers, e.g. AAAI 2023 and AAAI 2022 |
| Baseline | Gaze360→MPIIGaze | angular error | 8.18 – 9.89 | 2 papers, e.g. AAAI 2023 and AAAI 2022 |
| Baseline | ETH-XGaze→EyeDiap | angular error | 7.74 – 8.52 | 2 papers, e.g. AAAI 2022 and AAAI 2023 |
| Baseline | ETH-XGaze→MPIIGaze | angular error | 8.13 – 8.8 | 2 papers, e.g. AAAI 2022 and AAAI 2023 |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- DomainNet · 3 of 8 papers
- OfficeHome · 2 of 8 papers
- ImageNet · 2 of 8 papers
- ImageNet-Sketch · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 31 papers; from OpenAlex and the papers' affiliations).
- Nankai University · 2 papers
- Beihang University · 2 papers
- Peng Cheng Laboratory · 2 papers
- Hikvision Research Institute · 2 papers
- Google · 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
- Optimization-based domain generalization (101 papers)
- Foundation-model domain generalization (56 papers)
- Generalizable person re-identification (11 papers)
Elsewhere in the topic
- Conditional distribution alignment (162 papers)
- Cross-domain few-shot with transformers (116 papers)
- Feature alignment for unsupervised adaptation (105 papers)
- Adversarial domain adaptation (95 papers)
- Geometric alignment and synthetic-to-real adaptation (93 papers)
- Source-free domain adaptation (92 papers)
- Adaptation for semantic segmentation (81 papers)
- Statistical alignment and hypothesis testing (80 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).