Pith. sign in

REVIEW 3 major objections 5 minor 37 references

CTA: Cross-Task Alignment for Better Test Time Training

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A contrastive alignment step lets a self-supervised encoder serve as the feature extractor for a frozen classifier during test-time training, improving corrupted-image accuracy over prior TTT methods.

desk verdict A genuinely novel two-encoder TTT idea with strong results, but the main alignment loss contains a dimension mismatch that makes it undefined as written; this is fixable but must be addressed before the paper can be reproduced. read the letter →

arxiv 2507.05221 v2 pith:UOWD6SRB submitted 2025-07-07 cs.CV cs.AI

Samuel Barbeau , Pedram Fekri , David Osowiechi , Ali Bahri , Moslem Yazdanpanah , Masih Aminbeidokhti , Christian Desrosiers This is my paper
classification cs.CVcs.AI
keywords test-timetrainingadaptationcontrastivelearningrepresentationalignmentdistributionshiftSimCLRCIFAR-10-Cself-supervised
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

CTA's central claim is that the conflict between the supervised main task and the self-supervised auxiliary task in test-time training can be removed by separating them into two encoders and aligning them. The paper trains a SimCLR-style encoder and a supervised encoder on the same source data, then uses a contrastive cross-encoder loss to pull the self-supervised features toward the frozen supervised features. After this alignment, the self-supervised encoder can act as the feature extractor for the frozen classifier, and updating only the self-supervised encoder with the contrastive loss at test time improves accuracy on CIFAR-10-C, CIFAR-100-C, and TinyImageNet-C. A reader should care because the method needs no special architecture or chosen update layer, and it beat four recent test-time training baselines on those benchmarks.

What carries the argument

The load-bearing mechanism is the cross-encoder contrastive alignment loss $\mathcal{L}_{\mathrm{align}}$ (Eq. 4), which treats each augmented view of an image as a positive pair with the frozen supervised encoder's feature vector for the same image, and all other batch features as negatives. This student-teacher setup transfers the supervised feature geometry into the self-supervised encoder while the SimCLR objective $\mathcal{L}_{\mathrm{con}}$ remains the encoder's native task. At test time only $\mathcal{L}_{\mathrm{con}}$ is used, so the updated features stay in the same contrastive space that was aligned to the classifier during source training.

What would settle it

A reader could settle the claim by measuring the distance between the feature distributions of $\pi(g(\cdot))$ and $f(\cdot)$ on a target corruption after 20 adaptation iterations: CTA predicts this distance stays small or shrinks, while an unaligned shared-backbone model drifts. If the distance grows while accuracy still rises, the alignment story is wrong even if the benchmark numbers hold; alternatively, running the same two-stage alignment with a non-contrastive self-supervised task and seeing the gains vanish would show the effect is specific to SimCLR rather than alignment.

Watch

Extended reading notes

Core claim

The discovery, on the paper's own terms, is that the decision boundary of the supervised classifier can be transferred into the self-supervised encoder through alignment, without ever training that encoder on labels. Concretely, after minimizing $\mathcal{L}_{\mathrm{align}}$ (Eq. 4) between the two encoders on source data, the test-time model $h(\pi(g(\cdot)))$ uses the self-supervised encoder as the feature extractor and the frozen classifier as the head; test-time adaptation with $\mathcal{L}_{\mathrm{con}}$ updates all parameters of $\pi(g(\cdot))$. The paper reports that this aligned self-supervised encoder beats the supervised encoder it was aligned to on two of three source datasets and is more robust on all three corrupted benchmarks, with the full CTA pipeline reaching 87.42 on CIFAR-10-C, 63.95 on CIFAR-100-C, and 41.76 on TinyImageNet-C. The claim is that alignment avoids gradient interference while preserving the intrinsic shift-robustness of self-supervised representations.

Load-bearing premise

The whole method assumes that the decision boundary learned by the supervised encoder during source alignment continues to separate the classes after test-time SimCLR updates on corrupted images, so the frozen classifier still works on the adapted features.

Editorial extensions

If this is right

  • A pretrained supervised model can be turned into a TTT model by duplicating it, training the copy with a self-supervised loss, and aligning the copy to the original, with no change to the architecture.
  • The test-time adapter is the self-supervised encoder alone; the classifier stays frozen, so adaptation updates do not need to be filtered by layer choice or entropy signals.
  • Because the encoder never sees the cross-entropy loss, the method preserves the self-supervised representation's robustness under large corruptions, which the paper argues explains the gains over standard multi-task TTT.
  • On the benchmarks tested, the full CTA pipeline outperforms the baselines on average across all corruption types on CIFAR-10-C and on most corruption types on CIFAR-100-C and TinyImageNet-C, with gains of 4.51, 3.29, and 4.27 points over the previous best ReC-TTT.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension the paper leaves implicit is testing whether the same alignment transfers when the self-supervised task is replaced with masked image modeling or other objectives; the method's architecture-agnostic claim predicts the gain should persist.
  • The inverse temperature behavior (larger $\tau$ for alignment, smaller for test-time adaptation) suggests the teacher should be smoother than the student during adaptation; that asymmetry could be studied as a general principle for alignment-based adaptation.
  • If the representation-space alignment is the true cause of the gains, then CTA should work as a plug-in for any off-the-shelf frozen classifier, not only the ResNet50 fine-tuned here; a reader could test this by applying the two-stage alignment to a ViT or a detection backbone.
  • The paper's DBI and centroid-distance evidence implies that test-time contrastive updates push the self-supervised features closer to the frozen classifier's space, so a sharper test would be whether the distance stays bounded under much longer adaptation than the 20 iterations they report.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes CTA (Cross-Task Alignment), a test-time training method that trains a supervised encoder and a self-supervised encoder on the source domain, then aligns the self-supervised encoder's representations to the supervised encoder's representations with a contrastive loss. At test time, the self-supervised encoder is used as the feature extractor for the frozen classifier and is updated with the SimCLR contrastive loss. The authors report consistent average accuracy gains over prior TTT methods on CIFAR-10-C, CIFAR-100-C, and TinyImageNet-C, and include an ablation (CTA-C) that isolates the alignment component. The central claimed mechanism is that alignment transfers the supervised decision boundary into the self-supervised encoder while preserving the intrinsic robustness of self-supervised learning.

Significance. If the results hold, CTA would be a practically useful TTT method because it does not require a specialized architecture and can be applied to a pretrained backbone by duplicating it and fine-tuning the copy. The CTA-C ablation is a strength: it isolates the effect of the alignment stage and suggests that the gains over ReC-TTT are not solely due to using a self-supervised objective. The evaluation covers three standard corruption benchmarks and compares with recent methods. However, the formulation of the central alignment loss is inconsistent as written, the experimental protocol is incompletely specified, and some hyperparameters appear to be selected using target-domain accuracy. These issues must be resolved before the reported state-of-the-art claims can be accepted.

major comments (3)
  1. [Sec. 3.4, Eq. (4)] The alignment loss as written cannot be instantiated. In Sec. 3.2, h(·) is the classifier, so w_i = h(f(x_i)) has dimension equal to the number of classes C (10, 100, or 200), whereas Sec. 3.1 states that π(g(x_i)) preserves the encoder dimension d. The cosine similarity in Eq. (4) therefore compares vectors of different dimensions and the loss is undefined. The surrounding text (Sec. 3.4 and Fig. 1) and the dimension-preservation statement in Sec. 3.1 indicate that the intended target is f(x_i), the supervised encoder's feature vector, not the classifier output. Please correct the definition of w_i and revise the 'decision boundary distillation' claims in the Abstract, Sec. 1, Sec. 3.5, and Sec. 6 accordingly: the construction as stated aligns feature spaces, not classifier outputs. Because every downstream result in the paper depends on this loss, the implementation choice must be stated explicitly.
  2. [Sec. 4.2 and Sec. 5.4] The final accuracy numbers in Tables 2, 4, and 5 do not state how many test-time adaptation iterations were used. Section 4.2 gives batch size and learning rate only; Section 5.4 discusses 20 iterations and Figure 3, but never states whether the reported tables correspond to a fixed iteration count, the best iteration, or some other protocol. In addition, Tables 4 and 5 contain no standard deviations for CTA or the comparison methods, and in Table 2 the CTA-C column has no variance. Please report the exact adaptation protocol and run-to-run or batch-level variability for all main results.
  3. [Sec. 5.1, Table 1] The alignment temperature and test-time temperature appear to be selected using target-domain accuracy. Table 1 reports accuracy on the corrupted test sets as the alignment temperature varies, and Sec. 5.1 states that τ=0.5 was chosen for the final experiments and that smaller τ values were found to be better for test-time training. Since test-time training has no labels or validation set on the target distribution, this selection protocol can inflate the reported gains relative to methods whose hyperparameters are fixed on source data. Please clarify whether any target-domain information was used to choose τ_align and τ_con, and if so, provide a source-only or held-out validation procedure, or report sensitivity results that do not depend on target labels.
minor comments (5)
  1. [Introduction, Sec. 3.5] The phrase 'extract layer hyperparameter' in the Introduction and 'update-layer hyperparameter' in Sec. 3.5 should be made consistent; the current wording is confusing.
  2. [References] The bibliography contains duplicate entries for the same works: [3] and [4] are both ReC-TTT, [15] and [16] are both TTT++, and [21] and [22] are both NC-TTT. These should be merged.
  3. [Sec. 5.5, Table 2] The sentence 'CTA-C demonstrates an average loss of 4.48% compared to CTA' should say '4.48 percentage points lower' to avoid confusion with the training loss.
  4. [Sec. 5.2, Table 2] The sentence about being 'on par with NC-TTT for the smallest amount of variability (i.e., ±0.22)' is unclear because Table 2 reports per-corruption standard deviations rather than a single average variability for the full benchmark; please clarify what quantity is being compared.
  5. [Fig. 2 caption] The caption refers to 'the second column' and 'the third column' but the figure layout appears to be a grid of rows and columns; please make the description unambiguous.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: CTA's alignment and test-time adaptation are defined by independent equations and evaluated on external benchmarks.

full rationale

The paper's central chain is: (i) pretrain a supervised encoder h(f(·)) and a separate SimCLR encoder π(g(·)); (ii) align π(g(·)) to the supervised encoder features with the contrastive loss L_align in Eq. (4); (iii) at test time update only π(g(·)) with the SimCLR loss L_con and classify with the frozen classifier h(·). Each step is specified by its own loss function and is evaluated on held-out CIFAR-10-C, CIFAR-100-C and TinyImageNet-C benchmarks against external baselines. The alignment loss directly optimizes feature similarity, so the later use of the aligned SSL encoder as a feature extractor is the intended mechanism rather than a renamed or fitted version of the reported accuracy numbers. Hyperparameters such as τ_align and τ_con are empirical choices, not fitted parameters that are then relabeled as predictions. The many citations to prior work from the same research group appear as related-work comparisons and baselines (e.g., ClusT3, NC-TTT, ReC-TTT), not as load-bearing evidence for CTA's own mechanism; no uniqueness theorem or prior result by the authors is invoked to forbid alternative designs. The paper's claim that the decision boundary is distilled into the SSL encoder may be under-justified, and the dimensional consistency of Eq. (4) is a possible implementation concern, but these are correctness or clarity issues, not circularity. No equation reduces to another by construction, and no self-citation carries the central claim, so the appropriate circularity score is 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. Its assumptions are all domain-level or method-specific: SimCLR as the auxiliary task, cross-encoder alignment as a mechanism to transfer the decision boundary, and the persistence of that transfer under test-time updates. Hyperparameters such as temperatures and learning rates are chosen from prior work or target-set exploration.

free parameters (4)
  • alignment temperature tau_align = 0.5
    Chosen after ablation on target benchmarks (Sec. 5.1); used in L_align (Eq. 4).
  • contrastive temperature tau_con = 0.01
    Used for source SimCLR training and test-time updates; selected based on target performance (Sec. 5.1).
  • test-time learning rate = 1e-6
    Fixed from prior work [21]; affects adaptation magnitude and final accuracy.
  • number of adaptation iterations = not reported
    Final results in Tables 2, 4, and 5 do not state how many iterations are used; Sec. 5.4 discusses a 20-iteration analysis.
assumptions (4)
  • domain assumption SimCLR contrastive loss is an effective self-supervised auxiliary task for test-time training.
    Adopted from TTT++ [16]; not derived in this paper.
  • ad hoc to paper Cross-encoder contrastive alignment (Eq. 4) transfers the supervised decision boundary into the self-supervised encoder without destroying SSL robustness.
    This mechanism is the core of CTA and is supported only by empirical DBI/distance analysis.
  • domain assumption The frozen supervised classifier h remains compatible with aligned self-supervised features under target distribution shifts.
    Required for using h(pi(g(.))) at test time; validated empirically but not guaranteed.
  • domain assumption CIFAR-C benchmarks are representative of real distribution shifts.
    Standard benchmarks used in prior TTT works.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CTA: Cross-Task Alignment for Better Test Time Training." pith.science (2026). https://pith.science/paper/UOWD6SRB

@misc{pith2026250705221,
  author       = {Pith},
  title        = {Pith review of: CTA: Cross-Task Alignment for Better Test Time Training},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UOWD6SRB}},
  note         = {Machine review of arXiv:2507.05221}
}
read the original abstract

Deep learning models have demonstrated exceptional performance across a wide range of computer vision tasks. However, their performance often degrades significantly when faced with distribution shifts, such as domain or dataset changes. Test-Time Training (TTT) has emerged as an effective method to enhance model robustness by incorporating an auxiliary unsupervised task during training and leveraging it for model updates at test time. In this work, we introduce CTA (Cross-Task Alignment), a novel approach for improving TTT. Unlike existing TTT methods, CTA does not require a specialized model architecture and instead takes inspiration from the success of multi-modal contrastive learning to align a supervised encoder with a self-supervised one. This process enforces alignment between the learned representations of both models, thereby mitigating the risk of gradient interference, preserving the intrinsic robustness of self-supervised learning and enabling more semantically meaningful updates at test-time. Experimental results demonstrate substantial improvements in robustness and generalization over the state-of-the-art on several benchmark datasets.

Figures

Figures reproduced from arXiv: 2507.05221 by the authors.

Figure 1
Figure 1. Overview of CTA. During the source training stage (a), two distinct models h(f(·)) and g(·) are trained using a common dataset but with different objectives, namely cross-entropy LCE and contrastive learning Lcon respectively. During the alignment stage (b), the self-supervised model π(g(·)) is trained using source data to match the feature distribution of the frozen supervised encoder f(·) using the contrastive los… view at source ↗
Figure 2
Figure 2. Comparative visualization of domain representations as t-SNE plots between the Y model and CTA for CIFAR10 and CIFAR10- [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Evolution of accuracy (%) as a function of the number [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 21 canonical work pages

  1. [1]

    Swad: Domain generalization by seeking flat minima

    Junbum Cha, Sanghyuk Chun, Kyungjae Lee, Han-Cheol Cho, Seunghyun Park, Yunsung Lee, and Sungrae Park. Swad: Domain generalization by seeking flat minima. Advances in Neural Information Processing Systems, 34: 22405–22418, 2021. 1, 2

    work page 2021

  2. [2]

    A simple framework for contrastive learning of visual representations.arXiv preprint arXiv:2002.05709,

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learning of visual representations.arXiv preprint arXiv:2002.05709,

    arXiv 2002

  3. [3]

    ReC-TTT: Contrastive Feature Reconstruction for Test-Time Training

    Marco Colussi, Sergio Mascetti, Jose Dolz, and Christian Desrosiers. ReC-TTT: Contrastive feature reconstruction for test-time training.arXiv preprint arXiv:2411.17869, 2024. 2

    work page Pith review arXiv 2024

  4. [4]

    Rec-ttt: Contrastive feature reconstruction for test-time training, 2024

    Marco Colussi, Sergio Mascetti, Jose Dolz, and Christian Desrosiers. Rec-ttt: Contrastive feature reconstruction for test-time training, 2024. 1, 4, 5, 6, 7

    work page 2024

  5. [5]

    Imagenet: A large-scale hierarchical image database

    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, pages 248–255, 2009. 5

    2009

  6. [6]

    Test-time training with masked autoencoders.Advances in Neural Information Processing Systems, 35:29374–29385,

    Yossi Gandelsman, Yu Sun, Xinlei Chen, and Alexei Efros. Test-time training with masked autoencoders.Advances in Neural Information Processing Systems, 35:29374–29385,

  7. [7]

    Raphael Gontijo-Lopes, Yann Dauphin, and Ekin D. Cubuk. No one representation to rule them all: Overlapping features of training methods, 2022. 5

    work page 2022

  8. [8]

    Clust3: Information invariant test- time training

    Gustavo A Vargas Hakim, David Osowiechi, Mehrdad Noori, Milad Cheraghalikhani, Ali Bahri, Ismail Ben Ayed, and Christian Desrosiers. Clust3: Information invariant test- time training. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 6136–6145, 2023. 1, 2, 4, 6

    work page 2023

Show all 37 references
  1. [9]

    Deep residual learning for image recognition.arXiv preprint arXiv:1512.03385, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition.arXiv preprint arXiv:1512.03385, 2015. 5

    2015 arXiv

  2. [10]

    Benchmarking neu- ral network robustness to common corruptions and perturba- tions.Proceedings of the International Conference on Learn- ing Representations, 2019

    Dan Hendrycks and Thomas Dietterich. Benchmarking neu- ral network robustness to common corruptions and perturba- tions.Proceedings of the International Conference on Learn- ing Representations, 2019. 5

    2019

  3. [11]

    Distilling the knowledge in a neural network, 2015

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network, 2015. 4

    2015

  4. [12]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report, University of Toronto, 2009. 5

    2009

  5. [13]

    Tiny imagenet visual recognition challenge, 2015

    Ya Le and Xuan Yang. Tiny imagenet visual recognition challenge, 2015. 5

    2015

  6. [14]

    Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation

    Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation. InInternational conference on machine learning, pages 6028–6039. PMLR, 2020. 1

    2020

  7. [15]

    Ttt++: When does self-supervised test-time training fail or thrive? Advances in Neural Information Processing Systems, 34: 21808–21820, 2021

    Yuejiang Liu, Parth Kothari, Bastien Van Delft, Baptiste Bellot-Gurlet, Taylor Mordan, and Alexandre Alahi. Ttt++: When does self-supervised test-time training fail or thrive? Advances in Neural Information Processing Systems, 34: 21808–21820, 2021. 2

    2021

  8. [16]

    Ttt++: When does self-supervised test-time training fail or thrive? InThirty-Fifth Conference on Neural Information Processing Systems, 2021

    Yuejiang Liu, Parth Kothari, Bastien Germain van Delft, Baptiste Bellot-Gurlet, Taylor Mordan, and Alexandre Alahi. Ttt++: When does self-supervised test-time training fail or thrive? InThirty-Fifth Conference on Neural Information Processing Systems, 2021. 1, 2, 3, 5, 6, 8

    2021

  9. [17]

    Accuracy on the line: on the strong correlation between out-of-distribution and in- distribution generalization

    John P Miller, Rohan Taori, Aditi Raghunathan, Shiori Sagawa, Pang Wei Koh, Vaishaal Shankar, Percy Liang, Yair Carmon, and Ludwig Schmidt. Accuracy on the line: on the strong correlation between out-of-distribution and in- distribution generalization. InInternational Conferen...

    2021

  10. [18]

    Evaluating prediction-time batch normalization for robust- ness under covariate shift.arXiv preprint arXiv:2006.10963,

    Zachary Nado, Shreyas Padhy, D Sculley, Alexander D’Amour, Balaji Lakshminarayanan, and Jasper Snoek. Evaluating prediction-time batch normalization for robust- ness under covariate shift.arXiv preprint arXiv:2006.10963,

    2006 arXiv

  11. [19]

    Tipi: Test time adaptation with transforma- tion invariance

    A Tuan Nguyen, Thanh Nguyen-Tang, Ser-Nam Lim, and Philip HS Torr. Tipi: Test time adaptation with transforma- tion invariance. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24162– 24171, 2023. 2

    2023

  12. [20]

    Tttflow: Unsupervised test-time training with normalizing flow

    David Osowiechi, Gustavo A Vargas Hakim, Mehrdad Noori, Milad Cheraghalikhani, Ismail Ben Ayed, and Chris- tian Desrosiers. Tttflow: Unsupervised test-time training with normalizing flow. InProceedings of the IEEE/CVF Win- ter Conference on Applications of Computer Vision, pag...

    2023

  13. [21]

    Vargas Hakim, Mehrdad Noori, Milad Cheraghalikhani, Ali Bahri, Moslem Yazdan- panah, Ismail Ben Ayed, and Christian Desrosiers

    David Osowiechi, Gustavo A. Vargas Hakim, Mehrdad Noori, Milad Cheraghalikhani, Ali Bahri, Moslem Yazdan- panah, Ismail Ben Ayed, and Christian Desrosiers. Nc-ttt: A noise constrastive approach for test-time training. In***,

  14. [22]

    Nc-ttt: A noise constrastive approach for test-time training

    David Osowiechi, Gustavo A Vargas Hakim, Mehrdad Noori, Milad Cheraghalikhani, Ali Bahri, Moslem Yazdan- panah, Ismail Ben Ayed, and Christian Desrosiers. Nc-ttt: A noise constrastive approach for test-time training. InPro- ceedings of the IEEE/CVF Conference on Computer Visio...

    2024

  15. [23]

    Watt: Weight average test time adaptation of clip

    David Osowiechi, Mehrdad Noori, Gustavo Vargas Hakim, Moslem Yazdanpanah, Ali Bahri, Milad Cheraghalikhani, Sahar Dastani, Farzad Beizaee, Ismail Ayed, and Christian Desrosiers. Watt: Weight average test time adaptation of clip. InAdvances in Neural Information Processing Syst...

    2024

  16. [24]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, p...

    2021

  17. [25]

    An overview of multi-task learning in deep neural networks, 2017

    Sebastian Ruder. An overview of multi-task learning in deep neural networks, 2017. 1, 8

    2017

  18. [26]

    Adapting visual category models to new domains

    Kate Saenko, Brian Kulis, Mario Fritz, and Trevor Darrell. Adapting visual category models to new domains. InCom- puter Vision–ECCV 2010: 11th European Conference on Computer Vision, Heraklion, Crete, Greece, September 5- 11, 2010, Proceedings, Part IV 11, pages 213–226. Springer,

    2010

  19. [27]

    V ogt, Philip H

    Yuge Shi, Imant Daunhawer, Julia E. V ogt, Philip H. S. Torr, and Amartya Sanyal. How robust is unsupervised represen- tation learning to distribution shift?, 2022. 5

    2022

  20. [28]

    Test- time prompt tuning for zero-shot generalization in vision- language models

    Manli Shu, Weili Nie, De-An Huang, Zhiding Yu, Tom Goldstein, Anima Anandkumar, and Chaowei Xiao. Test- time prompt tuning for zero-shot generalization in vision- language models. InAdvances in Neural Information Pro- cessing Systems, pages 14274–14289. Curran Associates, Inc....

    2022

  21. [29]

    Test-time training with self- supervision for generalization under distribution shifts

    Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. Test-time training with self- supervision for generalization under distribution shifts. In International conference on machine learning, pages 9229–

  22. [30]

    Yu Sun, Xiaolong Wang, Liu Zhuang, John Miller, Moritz Hardt, and Alexei A. Efros. Test-time training with self- supervision for generalization under distribution shifts. In ICML, 2020. 5

    2020

  23. [31]

    Unbiased look at dataset bias

    Antonio Torralba and Alexei A Efros. Unbiased look at dataset bias. InCVPR 2011, pages 1521–1528. IEEE, 2011. 1

    2011

  24. [32]

    Clipartt: Light-weight adaptation of clip to new domains at test time,

    Gustavo Adolfo Vargas Hakim, David Osowiechi, Mehrdad Noori, Milad Cheraghalikhani, Ali Bahri, Moslem Yazdan- panah, Ismail Ben Ayed, and Christian Desrosiers. Clipartt: Light-weight adaptation of clip to new domains at test time,

  25. [33]

    Deep hashing network for unsupervised domain adaptation

    Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep hashing network for unsupervised domain adaptation. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 5018–5027, 2017. 1

    2017

  26. [34]

    Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726,

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Ol- shausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726,

    2006 arXiv

  27. [35]

    Gradient surgery for multi-task learning, 2020

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning, 2020. 1, 8

    2020

  28. [36]

    Domain generalization with mixstyle.arXiv preprint arXiv:2104.02008, 2021

    Kaiyang Zhou, Yongxin Yang, Yu Qiao, and Tao Xi- ang. Domain generalization with mixstyle.arXiv preprint arXiv:2104.02008, 2021. 1

    2021 arXiv

  29. [37]

    Domain generalization: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence,

    Kaiyang Zhou, Ziwei Liu, Yu Qiao, Tao Xiang, and Chen Change Loy. Domain generalization: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence,

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.