Research map / Label-efficient and robust learning / Test-time adaptation
Pseudo-label test-time adaptation
Uses pseudo-labels, feedback and proxy data for robust adaptation on noisy streams.
90 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · 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.
20 papers in 2023–24 and 67 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 18.2% to 38.1%.
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 · TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive? (NeurIPS 2021 · 76 citations)
- 2022 · NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation (NeurIPS 2022 · 30 citations)
- 2023 · Robust Test-Time Adaptation in Dynamic Scenarios (CVPR 2023 · 117 citations)
- 2024 · Improved Self-Training for Test-Time Adaptation (CVPR 2024 · 15 citations)
- 2025 · Gradient Alignment Improves Test-Time Adaptation for Medical Image Segmentation (AAAI 2025 · 7 citations)
- 2026 · NEO — No-Optimization Test-Time Adaptation through Latent Re-Centering (ICLR 2026 · 1 citation)
Earliest papers here
- TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive? (NeurIPS 2021 · 76 citations)
- NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation (NeurIPS 2022 · 30 citations)
- Test Time Adaptation via Conjugate Pseudo-labels (NeurIPS 2022 · 24 citations)
What has been tried
45 papers here report 539 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 5%, hurt 17%, 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 |
|---|---|---|---|---|
| Use PMA with π = 0.9 rather than disabling PMA | ImageNet · top-1 accuracy | 57.2 -> 58.3 | Helped | Improved Self-Training for Test-Time Adaptation (CVPR 2024 · 15 citations) |
| Add entropy-gradient alignment to the consistency objective | Average across domains A–E · DSC | 69.39 -> 71.53 | Helped | Gradient Alignment Improves Test-Time Adaptation for Medical Image Segmentation (AAAI 2025 · 7 citations) |
| Add paired-view consistency | CIFAR10-C, CIFAR100-C, and ImageNet-C gradual (level 5) · error rate | 11.3 -> 8.8; 25.9 -> 23.9; 37.4 -> 30.8 | Helped | Domain-Specific Block Selection and Paired-View Pseudo-Labeling for Online Test-Time Adaptation (CVPR 2024 · 7 citations) |
| Add csID entropy minimization to TENT | CIFAR-10-C · accuracy | 79.38 -> 85.04 | Helped | Unified Entropy Optimization for Open-Set Test-Time Adaptation (CVPR 2024 · 7 citations) |
| add GCAL to the stem layer | CIFAR-10-C · prediction error (%) | 30.8 -> 17.1 | Helped | L-TTA: Lightweight Test-Time Adaptation Using a Versatile Stem Layer (NeurIPS 2024 · 6 citations) |
| add conditional diffusion synthetic-data generation for fine-tuning | ImageNet-C · accuracy | 29.7 -> 30.4; 40.0 -> 40.1; 44.2 -> 44.6 | Helped | Everything to the Synthetic: Diffusion-driven Test-time Adaptation via Synthetic-Domain Alignment (CVPR 2025 · 5 citations) |
| Vary the number of dropout inferences N across 5, 10, and 15 | Continual CIFAR100-C · MAE (%) | No clear effect | AETTA: Label-Free Accuracy Estimation for Test-Time Adaptation (CVPR 2024 · 5 citations) | |
| add PLPD-based selection and weighting to entropy-based adaptation | ImageNet-C severity 5, mild · accuracy (%) | 47.35 -> 48.60 | Helped | Entropy is not Enough for Test-Time Adaptation: From the Perspective of Disentangled Factors (ICLR 2024 · 5 citations) |
| replace pseudo-label-based layer importance with gradient-based uncertainty from KL to a uniform distribution | CIFAR-10C, CIFAR-100C, ImageNet-C, CTTA severity 5 · mean error (%) | 17.9 -> 15.5; 33.8 -> 30.1; 68.6 -> 60.1 | Helped | PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation (AAAI 2025 · 3 citations) |
| Add ODP-based filtering | ImageNet-R and ImageNet-K, SH and NINCO · classification error rate (%) | 57.4 -> 56.1; 57.5 -> 56.7; 67.3 -> 66.0; 67.4 -> 66.8 | Helped | Test-Time Adaptation on Noisy Data via Model-Pruning-Based Filtering and Flatness-Aware Entropy Minimization (AAAI 2025 · 3 citations) |
| Replace iterative optimization with closed-form inference | Task 1 natural distribution shift; Task 2 corruption robustness; Task 3 fine-grained categorization · Accuracy | 66.74 -> 66.53; 28.68 -> 28.56; 71.05 -> 70.76 | Mixed | Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · 3 citations) |
| Add activation discrepancy to entropy-only CMA prompt adaptation | ImageNet-C level 5 · accuracy (%) | 44.9 -> 63.4 | Helped | Test-Time Model Adaptation with Only Forward Passes (ICML 2024 · 3 citations) |
What reviewers will expect
Datasets and baselines named in the 34 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 · 12 of 34 papers
- CIFAR10-C · 8 of 34 papers
- CIFAR100-C · 8 of 34 papers
- ImageNet-R · 7 of 34 papers
- ImageNet · 6 of 34 papers
- ImageNet-Sketch · 5 of 34 papers
- ImageNet-A · 4 of 34 papers
- PACS · 3 of 34 papers
Baselines
- TENT · 16 of 34 papers
- SAR · 11 of 34 papers
- CoTTA · 8 of 34 papers
- EATA · 7 of 34 papers
- TPT · 6 of 34 papers
- RoTTA · 5 of 34 papers
- SoTTA · 5 of 34 papers
- Source · 4 of 34 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 87 papers; from OpenAlex and the papers' affiliations).
- Chinese Academy of Sciences · 6 papers
- Korea Advanced Institute of Science and Technology · 4 papers
- Tsinghua University · 3 papers
- University of Chinese Academy of Sciences · 3 papers
- The Hong Kong Polytechnic University · 2 papers
- Nokia Bell Labs · 2 papers
- Huazhong University of Science and Technology · 2 papers
- Tencent · 2 papers
- South China University of Technology · 2 papers
- University of Science and Technology of China · 2 papers
Use it
Download BibTeX of 31 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)
- Extrapolation and single-sample adaptation (57 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).