Research map / Label-efficient and robust learning / Test-time adaptation
Extrapolation and single-sample adaptation
Generalizes to new domains or tasks from single test samples with uncertainty estimates.
57 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Test-time adaptation, keyword cluster “time · adaptation · test”
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 17 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 17.3% to 9.7%.
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 · Selective Classification for Deep Neural Networks (NeurIPS 2017 · 176 citations)
- 2018 · DiverseNet: When One Right Answer Is Not Enough (CVPR 2018 · 21 citations)
- 2019 · Fast Context Adaptation via Meta-Learning (ICML 2019 · 177 citations)
- 2020 · You Only Train Once: Loss-Conditional Training of Deep Networks (ICLR 2020 · 47 citations)
- 2021 · Overfitting for Fun and Profit: Instance-Adaptive Data Compression (ICLR 2021 · 30 citations)
- 2022 · RIO: Rotation-Equivariance Supervised Learning of Robust Inertial Odometry (CVPR 2022 · 29 citations)
- 2023 · Cloud-Device Collaborative Adaptation to Continual Changing Environments in the Real-World (CVPR 2023 · 20 citations)
- 2024 · Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · 59 citations)
- 2025 · Bridging the Gap for Test-Time Multimodal Sentiment Analysis (AAAI 2025 · 12 citations)
- 2026 · Lost in Time? A Meta-Learning Framework for Time-Shift-Tolerant Physiological Signal Transformation (AAAI 2026 · 1 citation)
Earliest papers here
- Selective Classification for Deep Neural Networks (NeurIPS 2017 · 176 citations)
- Resource Constrained Structured Prediction (AAAI 2017 · 4 citations)
- DiverseNet: When One Right Answer Is Not Enough (CVPR 2018 · 21 citations)
What has been tried
42 papers here report 444 measured changes (a component added, removed or swapped, compared with a baseline): helped 64%, no clear effect 12%, 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 TTSP style projection | I&C&M to O · HTER (%) | +3.40 | Helped | Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · 59 citations) |
| use dense rather than sparse training samples | nuScenes Original validation · mAP | 61.5 -> 86.4 | Helped | Localization Is All You Evaluate: Data Leakage in Online Mapping Datasets and How to Fix It (CVPR 2024 · 22 citations) |
| Add masked-autoencoder reconstruction loss alongside regression loss | INTER → Lyft (short-term) · mADE6 | 0.674 -> 0.391 | Helped | T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memory (CVPR 2024 · 18 citations) |
| Add contrastive adaptation | all five domain shifts · accuracy | Helped | Bridging the Gap for Test-Time Multimodal Sentiment Analysis (AAAI 2025 · 12 citations) | |
| add the MAE reconstruction task | IAM · WER | 11.06% -> 10.21% | Helped | MetaWriter: Personalized Handwritten Text Recognition Using Meta-Learned Prompt Tuning (CVPR 2025 · 6 citations) |
| Add adaptive-k | Fraud Dataset · AUC | 0.8218 -> 0.8233 | Helped | Online Fraud Detection via Test-Time Retrieval-Based Representation Enrichment (AAAI 2025 · 3 citations) |
| Use sample-wise loss weights based on simulated exit layers (SWM) | GLUE tasks, performance-efficiency trade-off · task performance | Helped | COSEE: Consistency-Oriented Signal-Based Early Exiting via Calibrated Sample Weighting Mechanism (AAAI 2025 · 2 citations) | |
| Train the shared model on data from all robot platforms instead of a single platform | Wheeled, handled, legged, and aerial platforms · ATE | Wheeled: 7.51 -> 6.17; Handled: 6.64 -> 4.32; Legged: 3.64 -> 1.46; Aerial: 3.84 -> 3.32 | Helped | Tartan IMU: A Light Foundation Model for Inertial Positioning in Robotics (CVPR 2025 · 2 citations) |
| train LPN with one latent gradient step rather than no latent optimization | Pattern, 100 inference gradient steps · exact match accuracy | 67.5% -> 99.5% | Helped | Searching Latent Program Spaces (NeurIPS 2025 · 2 citations) |
| Use only CNN DM expert features | Summarization average · ROUGE-2 | 23.7171 -> 23.7180 | No clear effect | Fusing Models with Complementary Expertise (ICLR 2024 · 2 citations) |
| Add HGNN to capture global user dynamic interactions | Christianity and Douban · MAP@100 | 0.1965 -> 0.2673; 0.1092 -> 0.1417 | Helped | Ghidorah: Towards Robust Multi-Scale Information Diffusion Prediction via Test-Time Training (AAAI 2025 · 1 citation) |
| use weighted flips with quadratic interpolation | 23 ImageNet-scale datasets · MAE | 6.14 -> 5.75 | Helped | The Entropy Enigma: Success and Failure of Entropy Minimization (ICML 2024 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- Norm · 2 of 9 papers
- Tent · 2 of 9 papers
- EATA · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 36 papers; from OpenAlex and the papers' affiliations).
- Beihang University · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- Peking University · 2 papers
- National University of Singapore · 2 papers
- Korea Advanced Institute of Science and Technology · 2 papers
- Carnegie Mellon University · 2 papers
- Google · 2 papers
- University of Amsterdam · 2 papers
Use it
Download BibTeX of 36 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
- Test-time adaptation of vision-language models (79 papers)
- Realistic and stable test-time adaptation (71 papers)
- Pseudo-label test-time adaptation (90 papers)
Elsewhere in the topic
- Robust optimization for distribution shift (34 papers)
- Characterizing out-of-distribution failures (28 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).