Pith. sign in

REVIEW 1 major objections 4 minor 36 references

A three-family taxonomy and shift-pattern map that turns continual test-time adaptation into a coherent research program.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review

2026-07-14 15:37 UTC pith:OMWCYM6F

load-bearing objection Solid, timely CTTA survey that actually organizes the field; the three-family taxonomy and shift-pattern catalog are the real contributions, and the re-tabulated numbers are already caveated. the 1 major comments →

arxiv 2607.08164 v2 pith:OMWCYM6F submitted 2026-07-09 cs.CV

Continual Test-Time Adaptation in Computer Vision: Methods, Benchmarks, and Future Directions

Sarthak Kumar Maharana , Shambhavi Mishra , Yunbei Zhang , Shuaicheng Niu , Taki Hasan Rafi , Jihun Hamm , Marco Pedersoli , Jose Dolz
show 1 more author
Yunhui Guo
This is my paper
classification cs.CV
keywords continual test-time adaptationcatastrophic forgettingerror accumulationentropy minimizationteacher-studentdomain shift patternsparameter-efficient adaptationfoundation-model adaptation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

When models leave the lab they meet streams of unlabeled data whose statistics keep changing. Continual Test-Time Adaptation (CTTA) is the problem of updating a pretrained model on that stream without ever seeing source data again, without labels, and without knowing when the domain changes. The survey's central claim is that the growing literature can be organized by what each method actually changes: the loss (optimization-based), a thin subset of parameters (parameter-efficient), or the network itself via added modules (architecture-based). Alongside the taxonomy the authors map the distinct temporal patterns used in evaluation—abrupt structured sequences, gradual transitions, mixed and imbalanced batches, dynamic durations, and recurring domains—and show why methods that look strong under one pattern often fail under another. Comparative numbers on standard corruption and segmentation suites then let a reader see which design choices actually control forgetting and error accumulation. The result is a practical roadmap for choosing methods today and for the harder problems of foundation models and black-box APIs tomorrow.

Core claim

The authors establish that CTTA methods fall cleanly into three families—optimization-based (entropy minimization, pseudo-labeling, parameter restoration), parameter-efficient (normalization-layer or selective-parameter updates), and architecture-based (teacher-student, adapters, visual prompts, masked modeling)—and that this taxonomy, together with a careful analysis of continual domain-shift patterns, supplies a usable framework both for evaluating existing work and for designing methods that remain stable under non-stationary streams.

What carries the argument

The hierarchical taxonomy of CTTA methods (three families defined by what is adapted) paired with the catalogue of continual domain-shift patterns (CSC, gradual, PTTA, CDC, recurring). Together they organize the literature and expose which failure modes each design choice is intended to control.

Load-bearing premise

That the experimental numbers compiled from original papers remain directly comparable across methods even though batch sizes, learning-rate schedules, number of rounds, and backbone details often differ.

What would settle it

A single controlled re-implementation of the main methods under identical batch size, learning rate, multi-round CSC protocol and backbone that reverses the relative ranking reported in the survey's tables would show the compiled comparisons are not reliable.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X LinkedIn Reddit HN

If this is right

  • Practitioners can match method family to hardware and shift pattern: teacher-student when memory is abundant, adapters or prompts on transformers, selective layers or forward-only methods on edge devices.
  • Benchmarks that mix CDC with temporal correlation become necessary; single-round CSC alone is no longer a sufficient stress test.
  • Normalization-layer methods are ruled out for LayerNorm architectures, pushing future work toward model-agnostic adapters and prompts.
  • The same three-family lens extends naturally to foundation models, VLMs and black-box APIs, giving a concrete agenda for those emerging settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The taxonomy itself can serve as a diagnostic: if a new method cannot be placed cleanly into one family, it is likely inventing a genuinely new adaptation axis.
  • Because error accumulation and forgetting are treated as structural rather than incidental, the survey implies that any future CTTA objective must include an explicit anti-forgetting term, not merely a better self-training loss.
  • The gap between academic CSC numbers and realistic mixed-shift streams suggests that reported gains may shrink sharply once domain durations and label balance become uncontrolled.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 4 minor

Summary. This survey formally defines Continual Test-Time Adaptation (CTTA) for computer vision, distinguishes it from standard continual learning and single-domain TTA, and analyzes the temporal structure of domain shifts under named protocols (CSC, gradual/CGS/CCC, PTTA, CDC, recurring). It proposes a hierarchical taxonomy of methods into three families—optimization-based (entropy minimization, pseudo-labeling, topological consistency, parameter restoration), parameter-efficient (normalization-layer adaptation, adaptive parameter selection), and architecture-based (teacher-student, adapters, visual prompting, masked modeling)—reviews representative algorithms, and compiles comparative results on CIFAR-C, ImageNet-C, and Cityscapes→ACDC. Limitations and future directions (foundation models, black-box adaptation, multi-modal and LLM settings, theory) are discussed, with an accompanying public repository.

Significance. CTTA has grown rapidly since CoTTA (2022) yet lacked a dedicated, in-depth survey that treats continual non-stationary shifts as the primary object of study. The three-family taxonomy and the explicit catalog of domain-shift patterns (CSC through recurring CDC) supply a usable organizational and evaluation roadmap that is already more granular than prior TTA surveys. The manuscript is careful about protocol heterogeneity (§7.3, §7.6.4), flags when BN-centric methods fail on LayerNorm backbones (Table 7), and includes a broader-impact statement. These strengths make the paper a solid reference for both method designers and practitioners, even though it does not introduce new algorithms or machine-checked proofs.

major comments (1)
  1. Tables 4–7 re-tabulate mean errors/mIoU from original papers (or a few unified re-runs) that differ in batch size, learning-rate schedule, single- vs multi-round CSC, and occasional backbone swaps. Although §7.3 and §7.6.4 acknowledge that many conclusions are protocol-dependent, the tables themselves still present a single “Mean” column without per-row protocol tags. Adding a compact protocol column (or a short appendix table) would make the comparability caveat operational rather than only textual, and would strengthen the claim that the taxonomy remains useful even when individual numbers shift.
minor comments (4)
  1. Figure 3 and the surrounding text place a few methods (e.g., CoTTA, RMT, ViDA) under multiple branches; a brief note that multi-category membership is intentional would reduce reader confusion.
  2. Notation for the source model (θ_S vs. f_θS) and for the Bernoulli restoration mask m is introduced more than once; a single consolidated notation paragraph early in §2 would improve consistency.
  3. The repository URL is given in the abstract and introduction; ensuring that the linked page already contains the compiled tables and taxonomy diagram at acceptance would aid reproducibility.
  4. A handful of very recent 2026 citations appear only in the future-directions section; verifying that they are publicly available (or marking them as concurrent) would avoid citation opacity.

Circularity Check

0 steps flagged

No circularity: survey taxonomy and domain-shift patterns are descriptive organizational claims, not derivations that reduce to their own inputs.

full rationale

This is a literature survey whose central contribution is a hierarchical taxonomy of existing CTTA methods into three families (optimization-based, parameter-efficient, architecture-based) plus an enumeration of continual domain-shift patterns (CSC, gradual, PTTA, CDC, recurring). These are definitional groupings of published work, not equations, fitted parameters, uniqueness theorems, or first-principles predictions. No self-definitional loops appear (e.g., no quantity is defined via the quantity it is later said to predict). No parameters are fitted to data and then re-presented as predictions. Self-citations of the authors’ own prior CTTA papers (PALM, DPCore, OT-VP, etc.) exist and are included as ordinary members of the taxonomy alongside many independent works (CoTTA, TENT, EATA, RMT, etc.); they are not load-bearing justifications for the taxonomy itself. Comparative tables recompile numbers from heterogeneous original papers, with the manuscript itself explicitly qualifying comparability limits (§7.3, §7.6.4). Because the paper advances no quantitative derivation that could collapse by construction, the circularity score is 0 and the steps list is empty.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 2 invented entities

As a survey the paper inherits the standard closed-set covariate-shift assumption of the CTTA literature and the operational constraints (source-free, single-pass, unknown boundaries). No free parameters are fitted; the taxonomy categories are invented organizational entities without independent empirical status beyond the papers they group.

axioms (3)
  • domain assumption Source and target share the same closed label space; only covariate shift (p(x) changes, p(y|x) fixed) is considered.
    Stated explicitly in §2.4 and Table 2; semantic/open-set shift is declared out of scope.
  • domain assumption Adaptation must be performed online with a single forward-backward pass per batch and without access to source data or task boundaries.
    Core problem definition in §2.1 and comparison with standard continual learning in Table 1.
  • domain assumption Entropy of the predictive distribution is a usable unsupervised proxy for adaptation quality under mild shifts.
    Inherited from TENT and used as the base objective for the optimization-based family (§4.1.1).
invented entities (2)
  • Three-family hierarchical taxonomy (optimization-based / parameter-efficient / architecture-based) no independent evidence
    purpose: Organize the growing set of CTTA algorithms into mutually intelligible groups for comparison and future design.
    Introduced in §4 and Figure 3; purely organizational, no claim of ontological novelty.
  • Named continual domain-shift patterns (CSC, CGS, CCC, PTTA, CDC, Recurring) no independent evidence
    purpose: Catalog the temporal structures under which methods are evaluated so that robustness claims can be qualified.
    Systematized in §3 from existing protocols; useful nomenclature rather than a new physical entity.

reviewed 2026-07-14 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Continual Test-Time Adaptation in Computer Vision: Methods, Benchmarks, and Future Directions." pith.science (2026). https://pith.science/paper/OMWCYM6F

@misc{pith2026260708164,
  author       = {Pith},
  title        = {Pith review of: Continual Test-Time Adaptation in Computer Vision: Methods, Benchmarks, and Future Directions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OMWCYM6F}},
  note         = {Machine review of arXiv:2607.08164}
}
Share X LinkedIn Reddit HN
read the original abstract

Deep neural nets achieve remarkable performance when training and test data share the same distribution, but this assumption frequently breaks in real-world deployment, where data undergoes continual distributional shifts. Continual Test-Time Adaptation (CTTA) addresses this challenge by adapting pretrained models to non-stationary target distributions on-the-fly, without access to source data or labeled targets, while mitigating two critical failure modes: catastrophic forgetting of source knowledge and error accumulation from noisy pseudo-labels over extended time horizons. In this comprehensive survey, we formally define the CTTA problem, analyze the diverse continual domain shift patterns that characterize different evaluation protocols, and propose a hierarchical taxonomy that categorizes existing methods into three families: optimization-based strategies (entropy minimization, pseudo-labeling, parameter restoration), parameter-efficient methods (normalization layer adaptation, adaptive parameter selection), and architecture-based approaches (teacher-student frameworks, adapters, visual prompting, masked modeling). We systematically review representative methods within each category and present comparative benchmarks and experimental results across standard evaluation settings. Finally, we discuss the limitations of current approaches and highlight emerging research directions, including the adaptation of foundation models and black-box systems, thereby providing a roadmap for future research in robust continual test-time adaptation.

Figures

Figures reproduced from arXiv: 2607.08164 by Jihun Hamm, Jose Dolz, Marco Pedersoli, Sarthak Kumar Maharana, Shambhavi Mishra, Shuaicheng Niu, Taki Hasan Rafi, Yunbei Zhang, Yunhui Guo.

Figure 1
Figure 1. Figure 1: Comparison of adaptation paradigms under distribution shift. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Growth of continual test-time adaptation and related research [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hierarchical taxonomy of representative CTTA methods. Methods are organized into three main [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Timeline of methods from 2020 to 2025. The field evolved from foundational TTA works (TTT, [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of estimating BatchNorm statistics and feature modulation at test-time. For an input [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Teacher-Student Framework: The teacher model is updated via an exponential moving average [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Adapters: Parallel branch architecture for CTTA. Lightweight adapter modules are inserted along [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visual Prompting: Input space adaptation for CTTA. The model backbone remains completely [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Masked Modeling: Uncertainty-guided reconstruction for CTTA. [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: ImageNet-C corruption examples. The 15 corruption types are grouped into four categories: Noise (Gaussian, Shot, Impulse), Blur (Defocus, Glass, Motion, Zoom), Weather (Snow, Frost, Fog, Bright￾ness), and Digital (Contrast, Elastic, Pixelate, JPEG). 27 [PITH_FULL_IMAGE:figures/full_fig_p027_10.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 19 linked inside Pith

  1. [1]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    Pith/arXiv arXiv

  2. [2]

    Towards open world recognition

    Abhijit Bendale and Terrance Boult. Towards open world recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 1893–1902,

    1902

  3. [3]

    On the opportunities and risks of foundation models.arXiv preprint arXiv:2108.07258,

    36 Published in Transactions on Machine Learning Research (07/2026) Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. On the opportunities and risks of foundation models.arXiv preprint arXiv:2108.07258,

    Pith/arXiv arXiv 2026

  4. [4]

    Parameter-free online test-time adaptation.2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Malik Boudiaf, Romain Mueller, Ismail Ben Ayed, and Luca Bertinetto. Parameter-free online test-time adaptation.2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 8334–8343,

    2022

  5. [5]

    The intriguing role of module criticality in the generalization of deep networks.arXiv preprint arXiv:1912.00528,

    Niladri S Chatterji, Behnam Neyshabur, and Hanie Sedghi. The intriguing role of module criticality in the generalization of deep networks.arXiv preprint arXiv:1912.00528,

    Pith/arXiv arXiv 1912

  6. [6]

    On tiny episodic memories in continual learning.arXiv preprint arXiv:1902.10486,

    Arslan Chaudhry, Marcus Rohrbach, Mohamed Elhoseiny, Thalaiyasingam Ajanthan, Puneet K Dokania, Philip HS Torr, and Marc’Aurelio Ranzato. On tiny episodic memories in continual learning.arXiv preprint arXiv:1902.10486,

    Pith/arXiv arXiv 1902

  7. [7]

    Imagenet: A large-scale hierarchical image database

    37 Published in Transactions on Machine Learning Research (07/2026) Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee,

    2026

  8. [8]

    A survey on in-context learning

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, Xu Sun, Lei Li, and Zhifang Sui. A survey on in-context learning. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics,

    2024

  9. [9]

    Deep think with confidence.arXiv preprint arXiv:2508.15260,

    Yichao Fu, Xuewei Wang, Yuandong Tian, and Jiawei Zhao. Deep think with confidence.arXiv preprint arXiv:2508.15260,

    Pith/arXiv arXiv

  10. [10]

    MIT press,

    38 Published in Transactions on Machine Learning Research (07/2026) Ian Goodfellow, Yoshua Bengio, and Aaron Courville.Deep learning. MIT press,

    2026

  11. [11]

    An empirical investigation of catastrophic forgetting in gradient-based neural networks.arXiv preprint arXiv:1312.6211,

    Ian J Goodfellow, Mehdi Mirza, Da Xiao, Aaron Courville, and Yoshua Bengio. An empirical investigation of catastrophic forgetting in gradient-based neural networks.arXiv preprint arXiv:1312.6211,

    Pith/arXiv arXiv

  12. [12]

    Mixnorm: Test-time adaptation through online normalization estimation.arXiv preprint arXiv:2110.11478,

    Xuefeng Hu, Gokhan Uzunbas, Sirius Chen, Rui Wang, Ashish Shah, Ram Nevatia, and Ser-Nam Lim. Mixnorm: Test-time adaptation through online normalization estimation.arXiv preprint arXiv:2110.11478,

    Pith/arXiv arXiv

  13. [13]

    Pcotta: Continual test-time adaptation for multi-task point cloud understanding

    Jincen Jiang, Qianyu Zhou, Yuhang Li, Xinkui Zhao, Meili Wang, Lizhuang Ma, Jian Chang, Jian J Zhang, and Xuequan Lu. Pcotta: Continual test-time adaptation for multi-task point cloud understanding. Advances in Neural Information Processing Systems, 37:96229–96253, 2024a. 39 Published in Transactions on Machine Learning Research (07/2026) JincenJiang, Qia...

    2026

  14. [14]

    Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation

    40 Published in Transactions on Machine Learning Research (07/2026) Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation. InInternational conference on machine learning, pp. 6028–

    2026

  15. [15]

    Continual test-time adaptation for end-to-end speech recognition on noisy speech

    Guan-Ting Lin, Wei-Ping Huang, and Hung yi Leeg. Continual test-time adaptation for end-to-end speech recognition on noisy speech. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics,

    2024

  16. [16]

    Ettrl: Balancing exploration and exploitation in llm test-time reinforcement learning via entropy mechanism.arXiv preprint arXiv:2508.11356, 2025a

    Jia Liu, ChangYi He, YingQiao Lin, MingMin Yang, FeiYang Shen, and ShaoGuo Liu. Ettrl: Balancing exploration and exploitation in llm test-time reinforcement learning via entropy mechanism.arXiv preprint arXiv:2508.11356, 2025a. Jiaming Liu, Ran Xu, Senqiao Yang, Renrui Zhang, Qizhe Zhang, Zehui Chen, Yandong Guo, and Shanghang Zhang. Continual-mae: Adapti...

    Pith/arXiv arXiv

  17. [17]

    Ctta-t: Continual test-time adaptation for text understanding via teacher-student with a domain-aware and generalized teacher.arXiv preprint arXiv:2512.18321, 2025b

    Tianlun Liu, Zhiliang Tian, Zhen Huang, Xingzhi Zhou, Wanlong Yu, Tianle Liu, Feng Liu, and Dongsheng Li. Ctta-t: Continual test-time adaptation for text understanding via teacher-student with a domain-aware and generalized teacher.arXiv preprint arXiv:2512.18321, 2025b. David Lopez-Paz and Marc’Aurelio Ranzato. Gradient episodic memory for continual lear...

    arXiv

  18. [18]

    On-demand test-time adaptation for edge devices.arXiv preprint arXiv:2505.00986,

    Xiao Ma, Young D Kwon, and Dong Ma. On-demand test-time adaptation for edge devices.arXiv preprint arXiv:2505.00986,

    Pith/arXiv arXiv

  19. [19]

    Palm: Pushing adaptive learning rate mecha- nisms for continual test-time adaptation

    Sarthak Kumar Maharana, Baoming Zhang, and Yunhui Guo. Palm: Pushing adaptive learning rate mecha- nisms for continual test-time adaptation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp. 19378–19386, 2025a. Sarthak Kumar Maharana, Baoming Zhang, Leonid Karlinsky, Rogerio Feris, and Yunhui Guo. Batclip: Bimodal online test...

    Pith/arXiv arXiv

  20. [20]

    Semantic anchor transport: Robust test-time adaptation for vision-language models,

    41 Published in Transactions on Machine Learning Research (07/2026) Shambhavi Mishra, Julio Silva-Rodriguez, Ismail Ben Ayed, Marco Pedersoli, and Jose Dolz. Semantic anchor transport: Robust test-time adaptation for vision-language models,

    2026

  21. [21]

    Effective restoration of source knowledge in continual test time adaptation

    Fahim Faisal Niloy, Sk Miraj Ahmed, Dripta S Raychaudhuri, Samet Oymak, and Amit K Roy-Chowdhury. Effective restoration of source knowledge in continual test time adaptation. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 2091–2100,

    2091

  22. [22]

    Lifelong domain adaptive 3d human pose estimation

    42 Published in Transactions on Machine Learning Research (07/2026) Qucheng Peng, Hongfei Xue, Pu Wang, and Chen Chen. Lifelong domain adaptive 3d human pose estimation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pp. 8358–8366,

    2026

  23. [23]

    If your data distribution shifts, use self-learning.arXiv preprint arXiv:2104.12928,

    Evgenia Rusak, Steffen Schneider, George Pachitariu, Luisa Eck, Peter Gehler, Oliver Bringmann, Wieland Brendel, and Matthias Bethge. If your data distribution shifts, use self-learning.arXiv preprint arXiv:2104.12928,

    Pith/arXiv arXiv

  24. [24]

    Empirical analysis of the hessian of over-parametrized neural networks.arXiv preprint arXiv:1706.04454,

    Levent Sagun, Utku Evci, V Ugur Guney, Yann Dauphin, and Leon Bottou. Empirical analysis of the hessian of over-parametrized neural networks.arXiv preprint arXiv:1706.04454,

    Pith/arXiv arXiv

  25. [25]

    Ecotta: Memory-efficient continual test-time adaptation via self-distilled regularization

    43 Published in Transactions on Machine Learning Research (07/2026) Junha Song, Jungsoo Lee, In So Kweon, and Sungha Choi. Ecotta: Memory-efficient continual test-time adaptation via self-distilled regularization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 11920–11929,

    2026

  26. [26]

    Less is more: Pseudo-label filtering for continual test-time adaptation.arXiv preprint arXiv:2406.02609,

    Jiayao Tan, Fan Lyu, Chenggong Ni, Tingliang Feng, Fuyuan Hu, Zhang Zhang, Shaochuang Zhao, and Liang Wang. Less is more: Pseudo-label filtering for continual test-time adaptation.arXiv preprint arXiv:2406.02609,

    Pith/arXiv arXiv

  27. [27]

    Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971,

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971,

    Pith/arXiv arXiv

  28. [28]

    Instance normalization: The missing ingredient for fast stylization.arXiv preprint arXiv:1607.08022,

    Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Instance normalization: The missing ingredient for fast stylization.arXiv preprint arXiv:1607.08022,

    Pith/arXiv arXiv

  29. [29]

    Test-time adaptation for geospatial point cloud semantic segmentation with distinct domain shifts.ISPRS Journal of Photogrammetry and Remote Sensing, 229: 422–435, 2025d

    44 Published in Transactions on Machine Learning Research (07/2026) Puzuo Wang, Wei Yao, Jie Shao, and Zhiyi He. Test-time adaptation for geospatial point cloud semantic segmentation with distinct domain shifts.ISPRS Journal of Photogrammetry and Remote Sensing, 229: 422–435, 2025d. Qin Wang, Olga Fink, Luc Van Gool, and Dengxin Dai. Continual test-time d...

    2026

  30. [30]

    Beyond model adaptation at test time: A survey.arXiv preprint arXiv:2411.03687,

    Zehao Xiao and Cees GM Snoek. Beyond model adaptation at test time: A survey.arXiv preprint arXiv:2411.03687,

    Pith/arXiv arXiv

  31. [31]

    Robust test-time adaptation in dynamic scenarios

    45 Published in Transactions on Machine Learning Research (07/2026) Longhui Yuan, Binhui Xie, and Shuang Li. Robust test-time adaptation in dynamic scenarios. InCVPR,

    2026

  32. [32]

    Wide residual networks.arXiv preprint arXiv:1605.07146,

    Sergey Zagoruyko and Nikos Komodakis. Wide residual networks.arXiv preprint arXiv:1605.07146,

    Pith/arXiv arXiv

  33. [33]

    Right question is already half the answer: Fully unsupervised llm reasoning incentivization.arXiv preprint arXiv:2504.05812, 2025a

    Qingyang Zhang, Haitao Wu, Changqing Zhang, Peilin Zhao, and Yatao Bian. Right question is already half the answer: Fully unsupervised llm reasoning incentivization.arXiv preprint arXiv:2504.05812, 2025a. YufeiZhang, YichengXu, HongxinWei, ZhipingLin, XiaofengZou, CenChen, andHuipingZhuang. Analytic continualtest-timeadaptationformulti-modalitycorruption....

    Pith/arXiv arXiv 1929

  34. [34]

    Learning to reason without external rewards.arXiv preprint arXiv:2505.19590, 2025b

    Xuandong Zhao, Zhewei Kang, Aosong Feng, Sergey Levine, and Dawn Song. Learning to reason without external rewards.arXiv preprint arXiv:2505.19590, 2025b. Jiahuan Zhou, Chao Zhu, Zhenyu Cui, Zichen Liu, Xu Zou, and Gang Hua. Class-aware domain knowledge fusion and fission for continual test-time adaptation. InThe Thirty-ninth Annual Conference on Neural I...

    Pith/arXiv arXiv

  35. [35]

    TTRL: Test-time reinforcement learning

    46 Published in Transactions on Machine Learning Research (07/2026) Yuxin Zuo, Kaiyan Zhang, Li Sheng, Shang Qu, Ganqu Cui, Xuekai Zhu, Haozhan Li, Yuchen Zhang, Xinwei Long, Ermo Hua, Biqing Qi, Youbang Sun, Zhiyuan Ma, Lifan Yuan, Ning Ding, and Bowen Zhou. TTRL: Test-time reinforcement learning. InThe Thirty-ninth Annual Conference on Neural Informatio...

    2026

  36. [36]

    While this capability addresses genuine and pressing challenges in robust deployment, it also introduces a set of ethical, safety, and societal considerations that we discuss below

    47 Published in Transactions on Machine Learning Research (07/2026) A Broader Impact Statement Our comprehensive survey on continual test-time adaptation (CTTA) focuses on self-adapting models to continual target or test distributions. While this capability addresses genuine and pressing challenges in robust deployment, it also introduces a set of ethical...

    2026

This paper was first reviewed by grok-4.5 on July 14, 2026.