Pith. sign in

REVIEW 2 major objections 4 minor 42 references

SPARNet: Continual Test-Time Adaptation via Sample Partitioning Strategy and Anti-Forgetting Regularization

T0 review · 2 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read SPARNet claims that splitting test samples by confidence and constraining important weights lets a model adapt across a stream of corruptions without forgetting, reaching average errors of 16.0% on CIFAR10-C, 30.6% on CIFAR100-C, and…

desk verdict SPARNet's headline gain over CoTTA is driven almost entirely by a source-data-requiring regularizer that contradicts its own source-free framing; the proposed modules alone are not competitive. read the letter →

arxiv 2501.00818 v1 pith:YS3GDW3A submitted 2025-01-01 cs.CV

Xinru Meng , Han Sun , Jiamei Liu , Ningzhong Liu , Huiyu Zhou This is my paper · ORCID
classification cs.CV
keywords continualtest-timeadaptationsamplepartitioninggeneralizedentropyminimizationmeanteacherconsistencyanti-forgettingregularizationcatastrophicforgettingCIFAR-10-CImageNet-C
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

The paper tries to show that a deployed classifier can keep improving through a long, unlabeled stream of corrupted test images, even though the corruption type changes without warning, if the adaptation is guided by a confidence split and a weight-protection regularizer. The proposed SPARNet framework separates each batch into low-entropy (reliable) samples that are trained with a softened entropy loss and high-entropy (unreliable) samples that are used only for mean-teacher consistency between augmented views. On top of both, an anti-forgetting term penalizes changes to parameters that the original model depends on most. On the continual CIFAR10-C, CIFAR100-C, and ImageNet-C benchmarks, the authors report average error rates that beat the continual TTA baselines they compare against, with the ablation on CIFAR10-C showing the anti-forgetting term as the single largest source of gain. If true, this points to a practical recipe for keeping models stable under long-term distribution shift without labels.

What carries the argument

The load-bearing object is the entropy-based sample partition plus the parameter-importance vector. A threshold $E_0$ on prediction entropy splits each batch into reliable and unreliable groups; the reliable group drives adaptation through generalized entropy minimization with a temperature-scaled softmax, while the unreliable group drives representation learning through symmetric cross-entropy consistency between the student and a mean teacher whose weights are an exponential moving average of the student. The anti-forgetting term uses the sensitivity of the network's output function to each parameter, computed over source-domain samples before deployment, to build an importance weight $\Omega_i$; the regularizer then penalizes the squared change of each parameter scaled by that weight. This design is what lets the model adapt long-term without drifting.

What would settle it

Train a separate run of SPARNet in which the parameter importances $\Omega_i$ are computed from unlabeled target-domain data (the first available batches) or from random initialization instead of the source-domain set $Q$, keeping everything else fixed. If the average CIFAR10-C error rises back toward the 27.3% of the no-regularizer ablation, the method's stated source-free premise is falsified in practice.

Watch

Extended reading notes

Core claim

SPARNet's central claim is that continual test-time adaptation is best handled by treating the confidence of the model's own predictions as a switch: reliable samples receive a generalized entropy-minimization loss with softened logits, unreliable samples receive a symmetric cross-entropy consistency loss between a student and a mean-teacher model, and a memory-aware-synapses style regularizer computed once before deployment keeps the most important parameters close to their source values. The authors report that this combination yields average error rates of 16.0% on CIFAR10-C, 30.6% on CIFAR100-C, and 67.3% on ImageNet-C in the continual setting, outperforming CoTTA, RoTTA, and AR-TTA in their runs. The ablation in Table IV attributes the largest single improvement on CIFAR10-C to the anti-forgetting regularizer: adding it to GEM plus SCE drops the average error from 27.3% to 16.2%.

Load-bearing premise

The anti-forgetting regularizer requires access to source-domain samples to compute parameter importances before deployment, which sits uneasily with the paper's claim that source data is unavailable for privacy reasons; without those samples, the largest single component of the reported gain cannot be computed.

Editorial extensions

If this is right

  • If SPARNet's results hold, continual TTA systems can stop discarding low-confidence test samples and instead use them as consistency-training data without contaminating the model.
  • The large ablation gain of the anti-forgetting term suggests that preventing drift of important parameters matters more than the choice of pseudo-labeling loss for long adaptation streams.
  • A fixed entropy threshold $E_0 = 0.4\ln C$ and precomputed importance weights make the method attractive for edge deployment, since the per-step updates are standard backpropagation.
  • The reported margins over CoTTA, RoTTA, and AR-TTA indicate that a simple two-branch loss plus weight protection is sufficient to beat more complex memory-buffer or robust-optimization machinery.

Reading between the lines

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

  • A natural next test is whether the importance vector could be estimated from a tiny held-out source set or synthesized prototypes, which would keep the anti-forgetting gain while respecting the source-privacy constraint; the paper itself does not explore this.
  • Because the partition threshold and the regularization strength are dataset-specific hyperparameters, the method's practical advantage may depend on how much tuning is allowed; a fair benchmark would fix them across all corruption types.
  • The same recipe could apply to semantic segmentation or object detection under continuous domain shift, where pseudo-label noise is more severe and per-pixel confidence partitioning would be the analogue of the sample split.
  • One could test whether the anti-forgetting regularizer alone, without sample partitioning, explains most of the improvement; the Table IV ablation suggests yes for CIFAR10-C, though the paper reports only one combination order.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper proposes SPARNet, a continual test-time adaptation method that splits incoming test samples into reliable and unreliable groups by prediction entropy. Reliable samples are trained with a generalized entropy-minimization loss, while unreliable samples are used for mean-teacher consistency learning with augmentation-averaged pseudo-labels and symmetric cross-entropy. To prevent catastrophic forgetting, the method adds a Memory-Aware-Synapses-style regularizer that penalizes changes to parameters deemed important for the source task. Experiments on CIFAR10-C, CIFAR100-C, and ImageNet-C report average error rates of 16.0%, 30.6%, and 67.3%, respectively, which the paper claims outperform CoTTA, RoTTA, and AR-TTA. An ablation on CIFAR10-C attributes the largest error reduction to the anti-forgetting regularizer.

Significance. If the reported results were obtained under the stated source-free continual test-time adaptation setting and the comparisons were fair, SPARNet would be a useful contribution: the sample-partitioning idea is simple, the combination of GEM and mean-teacher consistency is reasonable, and the ablation study is informative. The paper also addresses a real problem, namely catastrophic forgetting during long-term adaptation. However, the central claim is undermined by the fact that the dominant component, the anti-forgetting regularizer, requires access to source-domain samples to compute parameter importance, which contradicts the paper's stated privacy motivation and makes the comparison with source-free baselines not apples-to-apples. In addition, the reported gains are not accompanied by error bars, and key hyperparameters are tuned per dataset on the same evaluation benchmark, so the quantitative claims are not yet established at the level expected for this venue.

major comments (2)
  1. [Section III.E, Eq. (9); Section IV.A; Table IV] The anti-forgetting regularizer is computed from Omega_i = (1/|Q|) * sum_{q in Q} || d[f_0(q)]^2 / d theta_i ||_2, where Q is explicitly the source domain data set, and Section IV.A reports using 2000 source samples to calculate Omega. This directly contradicts the abstract and introduction, which state that 'the source data is unavailable due to privacy concerns,' and Section III.A, which defines the setting as 'without using source domain data during test-time.' Computing Omega once before deployment is still access to source-domain samples, and the paper itself acknowledges that 'regularization-based methods inevitably use the original distribution samples.' More importantly, Table IV shows that this regularizer is the dominant contributor to the reported gain: GEM alone gives 30.6, GEM+SCE gives 27.3, but adding Reg drops the error to 16.2, and the full method gives 16.0. Since CoTTA, RoTTA, and AR-TTA do not use source-domain samples, the comparisons in Tables I-III are not on equal footing. As written, the paper's headline claim for a source-free continual TTA setting is not supported; the main experimental gain comes from a component that violates the stated setting.
  2. [Section IV.A and IV.C; Tables I-III] The quantitative claim of state-of-the-art performance is fragile. No error bars or multiple-seed results are reported, and the hyperparameters lambda and beta are set separately for each dataset (1.8, 1.8, 0.3 and 1, 1, 0.005, respectively) on the same evaluation benchmark used for the final tables. The sensitivity analysis in Section IV.C is only shown for CIFAR10-C, so the reader cannot assess how the selected values affect CIFAR100-C and ImageNet-C. Moreover, the margin over CoTTA on CIFAR10-C is 0.2 percentage points (16.0 vs 16.2), which is likely within run-to-run variability for this type of experiment. The central claim that SPARNet outperforms prior work therefore needs stronger empirical support, preferably with error bars and with hyperparameters chosen on a validation split or shown to be stable across datasets.
minor comments (4)
  1. [Section IV.B, Table III] For ImageNet-C, AR-TTA is reported only as a mean value (68.0) with no per-corruption entries; the reader cannot verify whether the comparison is on exactly the same corruption sequence and severity as the other rows.
  2. [Section IV.A] The implementation details contain several grammatical errors, e.g., 'we using the Adam optimizer' and 'we using the SGD optimizer,' which should be corrected.
  3. [Section IV.C, Figs. 2 and 3] The sensitivity figures for lambda and beta are described only by their captions; the axes and exact metric are not described in the text, making the sensitivity claim hard to evaluate from the figures alone.
  4. [Section III.D, Eq. (4)] The notation for augmentation-averaged teacher predictions in Eq. (4) is somewhat unclear: the sum over i of aug_i(x) is written without an explicit average normalization, and the teacher predictions y_tu should be defined as the normalized average, not the unnormalized sum.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SPARNet is a composition of externally cited components; the reported results are empirical measurements, not predictions derived from the method's own assumptions.

full rationale

SPARNet's derivation chain does not reduce to its inputs. The sample-partitioning threshold E0 is inherited from EATA [5]; the generalized entropy loss is taken from DomainAdaptor [32] with temperature scaling from [33]; the mean-teacher consistency loss follows CoTTA [1] and uses symmetric cross-entropy from [39]; and the anti-forgetting regularizer is MAS [31]. All of these are external prior works, none is a self-citation by the present authors, and none is justified solely by a claim unique to this paper. The paper's headline numbers are empirical error rates measured on CIFAR10-C, CIFAR100-C, and ImageNet-C, not quantities derived from the framework's equations, so there is no fitted parameter being renamed as a prediction. The ablation in Table IV is an empirical attribution of gains, not a constructional equivalence: the regularizer's large contribution is measured, not assumed. Hyperparameters lambda, beta, and E0 are tuned on the same benchmarks, which is a benchmarking and overfitting concern rather than circularity. One genuine consistency problem, located at Section I ('the source data is unavailable due to privacy concerns'), Section III.E Eq. (9) ('Q denotes the source domain data set'), and Section IV.A ('We use 2000 samples to calculate Omega'), is that computing the regularizer requires source-domain samples, contradicting the stated source-free setting and making comparisons with source-free baselines unfair. That is a correctness and experimental-validity risk, not a circular-reasoning step, so it does not change the circularity score.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The method is an empirical combination of existing losses, so the ledger is dominated by domain assumptions about pseudo-label reliability and source-data access. The most consequential is the undisclosed requirement of 2000 source samples for Eq. (9), which also creates an unfair advantage over baselines. The three loss terms in Eq. (10) introduce no new theoretical entities; all components come from the cited literature.

free parameters (7)
  • Entropy threshold E0 = 0.4 * ln(C)
    Splits reliable and unreliable samples in Eq. (1); adopted from EATA rather than tuned, but directly controls the partitioning.
  • GEM temperature scaling strength s = 1
    Controls the dynamic temperature in Eq. (3); set to 1 following [33], affecting the GEM loss magnitude.
  • Regularization weight lambda = 1.8 (CIFAR10-C), 1.8 (CIFAR100-C), 0.3 (ImageNet-C)
    Trade-off weight in Eq. (10); per-dataset tuning on the evaluation benchmark.
  • Regularization weight beta = 1 (CIFAR10-C), 1 (CIFAR100-C), 0.005 (ImageNet-C)
    Trade-off weight for the anti-forgetting term in Eq. (10); per-dataset tuning on the evaluation benchmark.
  • Teacher EMA smoothing factor alpha = not specified
    Needed in Eq. (6) to update teacher weights; value is omitted, blocking exact replication.
  • Number of stochastic augmentations N = 32
    Number of augmented views used to compute the teacher's augmentation-averaged pseudo-label in Eq. (4).
  • Number of source samples for Omega = 2000
    Number of source-domain samples used to compute parameter importance in Eq. (9).
assumptions (4)
  • domain assumption Low-entropy predictions are trustworthy on corrupted target data
    The entire partitioning strategy in Eq. (1) assumes that high-confidence softmax outputs are reliable under domain shift; the paper provides no correctness calibration for this.
  • domain assumption Source-domain samples are available once before deployment
    Eq. (9) averages gradients over source dataset Q; if source data is truly unavailable, the anti-forgetting term cannot be computed, contradicting the paper's privacy motivation.
  • domain assumption Mean-teacher augmentation-averaged pseudo-labels are accurate enough to supervise unreliable samples
    Eqs. (4)-(5) use teacher predictions as training targets; if these are wrong, consistency learning amplifies errors rather than reducing them.
  • domain assumption A fixed entropy threshold E0 works across all corruption types
    E0 = 0.4 ln C is set once for all 15 corruptions and all domains; no adaptive threshold is used.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SPARNet: Continual Test-Time Adaptation via Sample Partitioning Strategy and Anti-Forgetting Regularization." pith.science (2026). https://pith.science/paper/YS3GDW3A

@misc{pith2026250100818,
  author       = {Pith},
  title        = {Pith review of: SPARNet: Continual Test-Time Adaptation via Sample Partitioning Strategy and Anti-Forgetting Regularization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YS3GDW3A}},
  note         = {Machine review of arXiv:2501.00818}
}
read the original abstract

Test-time Adaptation (TTA) aims to improve model performance when the model encounters domain changes after deployment. The standard TTA mainly considers the case where the target domain is static, while the continual TTA needs to undergo a sequence of domain changes. This encounters a significant challenge as the model needs to adapt for the long-term and is unaware of when the domain changes occur. The quality of pseudo-labels is hard to guarantee. Noisy pseudo-labels produced by simple self-training methods can cause error accumulation and catastrophic forgetting. In this work, we propose a new framework named SPARNet which consists of two parts, sample partitioning strategy and anti-forgetting regularization. The sample partition strategy divides samples into two groups, namely reliable samples and unreliable samples. According to the characteristics of each group of samples, we choose different strategies to deal with different groups of samples. This ensures that reliable samples contribute more to the model. At the same time, the negative impacts of unreliable samples are eliminated by the mean teacher's consistency learning. Finally, we introduce a regularization term to alleviate the catastrophic forgetting problem, which can limit important parameters from excessive changes. This term enables long-term adaptation of parameters in the network. The effectiveness of our method is demonstrated in continual TTA scenario by conducting a large number of experiments on CIFAR10-C, CIFAR100-C and ImageNet-C.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 36 canonical work pages

  1. [1]

    Continual test -time domain adaptation,

    Wang, Q., Fink, O., Van Gool, L., Dai, D., “Continual test -time domain adaptation,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition., pp. 7201 –7211 (2022)

    work page 2022

  2. [2]

    Improving robustness against common corruptions by covariate shift adaptation,

    Schneider, S., Rusak, E., Eck, L., Bringmann, O., Brendel, W., Bethge, M., “Improving robustness against common corruptions by covariate shift adaptation,” Advances in neural information processing systems., 33, 11539–11551 (2020)

    work page 2020

  3. [3]

    Tent: Fully test -time adaptation by entropy minimization,

    Wang, D., Shelhamer, E., Liu, S., Olshausen, B., Darrell, T., “Tent: Fully test -time adaptation by entropy minimization,” arXiv preprint arXiv:2006.10726 (2020)

    arXiv 2020

  4. [4]

    Evaluation of Test-Time Adaptation Under Computational Time Constraints

    Alfarra, M., Itani, H., Pardo, A., Alhuwaider, S., Ramazanova, M., P ́erez, J.C., Cai, Z., M ̈uller, M., Ghanem, B., “Revisiting test time adaptation under online evaluation,” arXiv preprint arXiv:2304.04795 (2023)

    work page Pith review arXiv 2023

  5. [5]

    Efficient test -time model adaptation without forgetting,

    Niu, S., Wu, J., Zhang, Y., Chen, Y. , Zheng, S., Zhao, P., Tan, M., “Efficient test -time model adaptation without forgetting,” International conference on machine learning., pp. 16888–16905. PMLR (2022)

    work page 2022

  6. [6]

    Adversarial continual learning,

    Ebrahimi, S., Meier, F., Calandra, R., Darrell, T., Rohrbach, M., “Adversarial continual learning,” Computer Vision –ECCV 2020: 16th European Conference, Glasgow, UK, August 23 –28, 2020, Proceedings, Part XI 16., pp. 386–402. Springer (2020)

    work page 2020

  7. [7]

    Catastrophic interference in connectionist networks: The sequential learning problem,

    McCloskey, M., Cohen, N.J., “Catastrophic interference in connectionist networks: The sequential learning problem,” Psychology of learning and motivation. , vol. 24, pp. 109 –165. Elsevier (1989)

    work page 1989

  8. [8]

    Continual lifelong learning with neural networks: A review,

    Parisi, G.I., Kemker, R., Part, J.L., Kanan, C., Wermter, S., “Continual lifelong learning with neural networks: A review,” Neural networks., 113, 54–71 (2019)

    work page 2019

Show all 42 references
  1. [9]

    Instance weighting for domain adaptation in nlp,

    Jiang, J., Zhai, C., “Instance weighting for domain adaptation in nlp,” ACL (2007)

    2007

  2. [10]

    Instance weighting for neural machine translation domain adaptation,

    Wang, R., Utiyama, M., Liu, L., Chen, K., Sumita, E., “Instance weighting for neural machine translation domain adaptation,” Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing., pp. 1482–1488 (2017)

    2017

  3. [11]

    Boosting for transfer learning with multiple sources,

    Yao, Y., Doretto, G., “Boosting for transfer learning with multiple sources,” 2010 IEEE computer society conference on computer vision and pattern recognition ., pp. 1855–1862. IEEE (2010)

    2010

  4. [12]

    Domain -adversarial training of neural networks,

    Ganin, Y., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., March, M., Lempitsky, V., “Domain -adversarial training of neural networks,” Journal of machine learning research., 17(59), 1–35 (2016)

    2016

  5. [13]

    Adversarial discriminative domain adaptation,

    Tzeng, E., Hoffman, J., Saenko, K., Darrell, T., “Adversarial discriminative domain adaptation,” Proceedings of the IEEE conference on computer vision and pattern recognition., pp. 7167–7176 (2017)

    2017

  6. [14]

    Homm: Higher -order moment matching for unsupervised domain adaptation,

    Chen, C., Fu, Z., Chen, Z., Jin, S., Cheng, Z., Jin, X., Hua, X.S., “Homm: Higher -order moment matching for unsupervised domain adaptation,” Proceedings of the AAAI conference on artificial intelligence., vol. 34, pp. 3422–3429 (2020)

    2020

  7. [15]

    Deep coral: Correlation alignment for deep domain adaptation,

    Sun, B., Saenko, K., “Deep coral: Correlation alignment for deep domain adaptation,” Computer Vision–ECCV 2016., Workshops: Amsterdam, The Netherlands, October 8 -10 and 15 -16, 2016, Proceedings, Part III 14. pp. 443–450. Springer (2016)

    2016

  8. [16]

    Mind the class weight bias: Weighted maximum mean discrepancy for unsupervised domain adaptation,

    Yan, H., Ding, Y., Li, P., Wang, Q., Xu, Y., Zuo, W., “Mind the class weight bias: Weighted maximum mean discrepancy for unsupervised domain adaptation,” Proceedings of the IEEE conference on computer vision and pattern recognition ., pp. 2272–2281 (2017)

    2017

  9. [17]

    Contrastive adaptation network for unsupervised domain adaptation,

    Kang, G., Jiang, L., Yang, Y., Hauptmann, A.G., “Contrastive adaptation network for unsupervised domain adaptation,” Proceedings of the IEEE/CVF conference on computer vision and pattern recognition., pp. 4893–4902 (2019)

    2019

  10. [18]

    Contrastive learning and self -training for unsupervised domain adaptation in semantic segmentation,

    Marsden, R.A., Bartler, A., D ̈obler, M., Yang, B., “Contrastive learning and self -training for unsupervised domain adaptation in semantic segmentation,” 2022 International Joint Conference on Neural Networks (IJCNN)., pp. 1–8. IEEE (2022)

    2022

  11. [19]

    Parameter-free online test -time adaptation,

    Boudiaf, M., Mueller, R., Ben Ayed, I., Bertinetto, L., “Parameter-free online test -time adaptation,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition., pp. 8344–8353 (2022)

    2022

  12. [20]

    Overcoming catastrophic forgetting in neural networks,

    Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A.A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., et al., “Overcoming catastrophic forgetting in neural networks,” Proceedings of the national academy of sciences., 114(13), 3521–3526 (2017)

    2017

  13. [21]

    Towards stable test -time adaptation in dynamic wild world,

    Niu, S., Wu, J., Zhang, Y., Wen, Z., Chen, Y., Zhao, P., Tan, M., “Towards stable test -time adaptation in dynamic wild world,” arXiv preprint arXiv:2302.12400 (2023)

    2023 arXiv

  14. [22]

    Robust mean teacher for continual and gradual test-time adaptation,

    DÖbler, M., Marsden, R.A., Yang, B., “Robust mean teacher for continual and gradual test-time adaptation,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition., pp. 7704–7714 (2023)

    2023

  15. [23]

    Improving test -time adaptation via shif t-agnostic weight regularization and nearest source prototypes,

    Choi, S., Yang, S., Choi, S., Yun, S., “Improving test -time adaptation via shif t-agnostic weight regularization and nearest source prototypes,” European Conference on Computer Vision ., pp. 440–458. Springer (2022)

    2022

  16. [24]

    Ar -tta: A simple method for real -world continual test -time adap tation,

    SÓjka, D., Cygert, S., Twardowski, B., Trzci ́nski, T., “Ar -tta: A simple method for real -world continual test -time adap tation,” Proceedings of the IEEE/CVF International Conference on Computer Vision., pp. 3491–3495 (2023)

    2023

  17. [25]

    Robust test -time adaptation in dynamic scenarios,

    Yuan, L., Xie, B., Li, S., “Robust test -time adaptation in dynamic scenarios,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition ., pp. 15922 –15932 (2023)

    2023

  18. [26]

    A continual learning survey: Defying forgetting in classification tasks,

    De Lange, M., Aljundi, R., Masana, M., Parisot, S., Jia, X., Leonardis, A., Slabaugh, G., Tuytelaars, T., “A continual learning survey: Defying forgetting in classification tasks,” IEEE transactions on pattern analysis and machine intelligence 44(7)., 3366–3385 (2021)

    2021

  19. [27]

    icarl: Incremental classifier and representation learning,

    Rebuffi, S.A., Kolesnikov, A., Sperl, G., Lampert, C.H., “icarl: Incremental classifier and representation learning,” Proceedings of the IEEE conference on Computer Vision and Pattern Recognition., pp. 2001–2010 (2017)

    2017

  20. [28]

    The task rehearsal method of life-long learning: Overcoming impoverished data,

    Silver, D.L., Mercer, R.E., “The task rehearsal method of life-long learning: Overcoming impoverished data,” Advances in Artificial Intelligence: 15th Conference of the Canadian Society for Computational Studies of Intelligence. , AI 2002 Calgary, Canada, May 27–29, 2002 Proce...

    2002

  21. [29]

    Continual learning through synaptic intelligence,

    Zenke, F., Poole, B., Ganguli, S., “Continual learning through synaptic intelligence,” International conference on machine learning., pp. 3987–3995. PMLR (2017)

    2017

  22. [30]

    Learning without forgetting,

    Li, Z., Hoiem, D., “Learning without forgetting,” IEEE transactions on pattern analysis and machine intelligence 40(12)., 2935–2947 (2017)

    2017

  23. [31]

    Memory aware synapses: Learning what (not) to forget,

    Aljundi, R., Babiloni, F., Elhoseiny, M., Rohrbach, M., Tuytelaars, T., “Memory aware synapses: Learning what (not) to forget,” Proceedings of the European conference on computer vision (ECCV)., pp. 139–154 (2018)

    2018

  24. [32]

    Domainadaptor: A novel approach to test-time adaptation,

    Zhang, J., Qi, L., Shi, Y., Gao, Y., “Domainadaptor: A novel approach to test-time adaptation,” Proceedings of the IEEE/CVF International Conference on Computer Vision ., pp. 18971–18981 (2023)

    2023

  25. [33]

    Reducing the teacher-student gap via spherical knowledge distillation,

    Guo, J., Chen, M., Hu, Y., Zhu, C., He, X., Cai, D., “Reducing the teacher-student gap via spherical knowledge distillation,” arXiv preprint arXiv:2010.07485 (2020)

    2020 arXiv

  26. [34]

    Temporal ensembling for semi -supervised learning,

    Laine, S., Aila, T., “Temporal ensembling for semi -supervised learning,” arXiv preprint arXiv:1610.02242 (2016)

    2016 arXiv

  27. [35]

    Mean teachers are better role models: Weight -averaged consistency targets improve semi-supervised deep learning results,

    Tarvainen, A., Valpola, H., “Mean teachers are better role models: Weight -averaged consistency targets improve semi-supervised deep learning results,” Advances in neural information processing systems., 30 (2017)

    2017

  28. [36]

    Unsupervised data augmentation for consistency training,

    Xie, Q., Dai, Z., Hovy, E., Luong, T., Le, Q., “Unsupervised data augmentation for consistency training,” Advances in neural information processing systems., 33, 6256–6268 (2020)

    2020

  29. [37]

    Virtual adversarial training: a regularization method for supervised and semi-supervised learning,

    Miyato, T., Maeda, S.i., Koyama, M., Ishii, S., “Virtual adversarial training: a regularization method for supervised and semi-supervised learning,” IEEE transactions on pattern analysis and machine intelligence., 41(8), 1979–1993 (2018)

    2018

  30. [38]

    A survey on deep semi-supervised learning,

    Yang, X., Song, Z., King, I., Xu, Z., “A survey on deep semi-supervised learning,” IEEE Transactions on Knowledge and Data Engineering 35(9)., 8934–8954 (2022)

    2022

  31. [39]

    Symmetric cross entropy for robust learning with noisy labels,

    Wang, Y., Ma, X., Chen, Z., Luo, Y., Yi, J., Bailey, J., “Symmetric cross entropy for robust learning with noisy labels,” Proceedings of the IEEE/CVF international conference o n computer vision., pp. 322–330 (2019)

    2019

  32. [40]

    Wide residual networks,

    Zagoruyko, S., Komodakis, N., “Wide residual networks,” arXiv preprint arXiv:1605.07146 (2016)

    2016 arXiv

  33. [41]

    Aggregated residual transformations for deep neural networks,

    Xie, S., Girshick, R., Doll ́ar, P., Tu, Z., He, K., “Aggregated residual transformations for deep neural networks,” Proceedings of the IEEE conference on computer vision and pattern recognition., pp. 1492–1500 (2017)

    2017

  34. [42]

    Robustbench: a standardized adversarial robustness benchmark,

    Croce, F., Andriushchenko, M., Sehwag, V., Debenedetti, E., Flammarion, N., Chiang, M., Mittal, P., Hein, M., “Robustbench: a standardized adversarial robustness benchmark,” arXiv preprint arXiv:2010.09670 (2020)

    2020 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.