Research map / Label-efficient and robust learning / Test-time adaptation
Realistic and stable test-time adaptation
Tackles layer selection, long-horizon drift and sequential streams in practical test-time adaptation.
71 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Test-time adaptation, keyword cluster “time · adaptation · test”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 1 tracked benchmark.
28 papers in 2023–24 and 36 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 25.5% to 20.5%.
Open questions the data points to
- ImageNet-R (accuracy): best went from 55.1 (up to 2024) to 70.2 in 2025 — Learning to Generate Gradients for Test-Time Adaptation via Test-Time Training Layers (AAAI 2025 · 4 citations)
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
- 2020 · Learning Loss for Test-Time Augmentation (NeurIPS 2020 · 38 citations)
- 2021 · Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021 · 283 citations)
- 2022 · Contrastive Test-Time Adaptation (CVPR 2022 · 262 citations)
- 2023 · Back to the Source: Diffusion-Driven Adaptation To Test-Time Corruption (CVPR 2023 · 70 citations)
- 2024 · Test-Time Domain Adaptation by Learning Domain-Aware Batch Normalization (AAAI 2024 · 24 citations)
- 2025 · A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · 4 citations)
- 2026 · PriorGuide: Test-Time Prior Adaptation for Simulation-Based Inference (ICLR 2026 · 0 citations)
Earliest papers here
- Learning Loss for Test-Time Augmentation (NeurIPS 2020 · 38 citations)
- Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021 · 283 citations)
- Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization (NeurIPS 2021 · 83 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.
ImageNet-R · accuracy (higher is better)
4 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 38.9 | Test-Time Training with Masked Autoencoders (NeurIPS 2022 · 36 citations) |
| 2024 | 55.1 | Cross-Device Collaborative Test-Time Adaptation (NeurIPS 2024 · 8 citations) |
| 2025 | 70.2 | Learning to Generate Gradients for Test-Time Adaptation via Test-Time Training Layers (AAAI 2025 · 4 citations) |
What has been tried
48 papers here report 560 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 8%, hurt 18%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add the SSL branch and joint training | iWildCam · accuracy | 68.7 -> 70.5 | Helped | Test-Time Domain Adaptation by Learning Domain-Aware Batch Normalization (AAAI 2024 · 24 citations) |
| replace Robust BN with Balanced BN | CIFAR10-C GLI-TTA-F (I.F.=100) · classification error (%) | 43.48 -> 29.00 | Helped | Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · 20 citations) |
| add batch norm updates to ProxyTTA-fast | all reported datasets · average improvement over pretrained model | 44.53% | Helped | Test-Time Adaptation for Depth Completion (CVPR 2024 · 11 citations) |
| use 100% rather than 25% of source features as prototype memory | ScanObjectNN-C · classification error (%) | 43.91 -> 43.75 | Helped | Backpropagation-free Network for 3D Test-time Adaptation (CVPR 2024 · 8 citations) |
| Add the learnable temperature Tl to knowledge reprogramming | ImageNet-C Gaussian, early adaptation with 960 samples · accuracy (%) | 22.2% -> 53.2% | Helped | Cross-Device Collaborative Test-Time Adaptation (NeurIPS 2024 · 8 citations) |
| Use single-layer rather than single-block selection | Domainbed averaged settings · accuracy | 67.64 -> 68.57 | Helped | A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · 4 citations) |
| Add attention bootstrapping | Kinetics50-C · accuracy | Helped | Attention Bootstrapping for Multi-Modal Test-Time Adaptation (AAAI 2025 · 4 citations) | |
| Add Confident Distribution Optimizer | DIABETE · F1 | 51.26±1.69 -> 59.27±0.96 | Helped | Fully Test-time Adaptation for Tabular Data (AAAI 2025 · 4 citations) |
| add MGG gradient optimization | ImageNet-C severity 5 · accuracy | 70.0 -> 71.3 | Helped | Learning to Generate Gradients for Test-Time Adaptation via Test-Time Training Layers (AAAI 2025 · 4 citations) |
| Replace the entropy matching loss ℓmatch with the filtered and weighted ℓmatch++ loss | ImageNet-C severity 5, 15-corruption average · accuracy | Helped | Protected Test-Time Adaptation via Online Entropy Matching: A Betting Approach (NeurIPS 2024 · 4 citations) | |
| add TEMA to current test-batch normalization statistics | CIFAR-10-C, CIFAR-100-C, and ImageNet-C continual; Avg. across batch sizes · error rate | 37.38 -> 21.61; 56.06 -> 39.80; 82.89 -> 72.47 | Helped | Unraveling Batch Normalization for Realistic Test-Time Adaptation (AAAI 2024 · 3 citations) |
| Add prediction difference-based sample selection | ImageNet-C · Avg. Err. | 54.7 -> 53.8 | Helped | Effortless Active Labeling for Long-Term Test-Time Adaptation (CVPR 2025 · 2 citations) |
What reviewers will expect
Datasets and baselines named in the 23 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ImageNet-C · 8 of 23 papers
- ImageNet-R · 5 of 23 papers
- PACS · 3 of 23 papers
- CIFAR10-C · 3 of 23 papers
- CIFAR100-C · 3 of 23 papers
- VLCS · 2 of 23 papers
- OfficeHome · 2 of 23 papers
- ImageNet-Sketch · 2 of 23 papers
Baselines
- TENT · 11 of 23 papers
- SAR · 10 of 23 papers
- EATA · 8 of 23 papers
- CoTTA · 6 of 23 papers
- DeYO · 6 of 23 papers
- LAME · 4 of 23 papers
- FOA · 4 of 23 papers
- T3A · 3 of 23 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 64 papers; from OpenAlex and the papers' affiliations).
- South China University of Technology · 4 papers
- Peking University · 4 papers
- Nanyang Technological University · 4 papers
- EPFL · 4 papers
- Korea Advanced Institute of Science and Technology · 4 papers
- Amazon · 3 papers
- Swiss Federal Institute of Technology Lausanne · 3 papers
- The Chinese University of Hong Kong · 2 papers
- University of California, Los Angeles · 2 papers
- University of Washington · 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
- Test-time adaptation of vision-language models (79 papers)
- Extrapolation and single-sample adaptation (57 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).