Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

MoSSDA: A Semi-Supervised Domain Adaptation Framework for Multivariate Time-Series Classification using Momentum Encoder

T0 review · 4 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 two-stage momentum-encoder framework, MoSSDA, learns domain-invariant and class-discriminative time-series features and reports state-of-the-art semi-supervised domain adaptation results.

desk verdict A solid combination of known ideas for time-series SSDA, but the state-of-the-art claim rests on baseline comparisons that are not yet verified. read the letter →

arxiv 2508.08280 v1 pith:LKCHZ3JL submitted 2025-08-01 cs.LG cs.AIcs.CV

Seonyoung Kim , Dongil Kim This is my paper
classification cs.LGcs.AIcs.CV
keywords semi-superviseddomainadaptationmultivariatetime-seriesclassificationmomentumencodercontrastivelearningmixupmaximummeandiscrepancyshifttwo-stagetraining
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

MoSSDA is a semi-supervised domain adaptation method for multivariate time-series classification. In the realistic setting where a fully labeled source domain is available but the target domain supplies labels for only a small fraction of its data, the paper aims to learn representations that are both domain-invariant and class-discriminative without relying on data augmentation. The method splits training into two stages: first it trains a domain-invariant encoder with an MMD loss and a mixup-enhanced supervised contrastive loss built around a momentum encoder, then it freezes that encoder and trains a classifier on labeled source and target features. The paper reports that MoSSDA outperforms six semi-supervised domain adaptation baselines across six datasets, three backbone architectures, and three unlabeled ratios, with an average rank of 1.06 to 1.11.

What carries the argument

The load-bearing mechanism is the positive contrastive module, a supervised contrastive loss whose positive pairs include mixup-interpolated features of same-class samples from both source and target domains. A momentum encoder, a copy of the online encoder updated by exponential moving average rather than backpropagation, provides stable feature keys for the contrastive loss. An MMD loss on the same encoder enforces domain invariance, and a two-stage schedule separates gradient flow: the encoder and contrastive module are trained first, and only then is the classifier trained on frozen features.

What would settle it

Re-run the six baselines with their official implementations on the same six datasets, same domain splits, same three backbones, and same unlabeled ratios; MoSSDA's central claim fails if its average rank moves far from 1 or if its margins over the best baseline shrink to noise.

Watch

Extended reading notes

Core claim

The paper's central claim is that decoupling representation learning from classifier training, and replacing augmentation with mixup in a momentum-encoder contrastive module, produces time-series features that transfer well under domain shift. The domain-invariant encoder minimizes maximum mean discrepancy between source and target feature distributions, while the positive contrastive module pulls together labeled samples of the same class from both domains and their mixup interpolations; the momentum encoder supplies stable, slowly changing feature keys so the contrastive dictionary does not shift abruptly between gradient steps. The classifier is then trained on frozen features using cross-entropy on all labeled data. According to the paper, this combination yields the best target-domain accuracy and F1 across ResNet18, CNN, and TCN backbones, and the ablation studies attribute the largest gains to the contrastive module and the two-stage separation.

Load-bearing premise

The claim of consistent superiority rests on the fairness of the baseline comparison: the six comparison methods were re-implemented with time-series augmentations by the authors, and no code or configuration files for those runs are provided, so if the baselines were undertuned the reported margins would overstate MoSSDA's advantage.

Editorial extensions

If this is right

  • Semi-supervised time-series domain adaptation can be done without data augmentation, which matters because temporal order can be destroyed by image-style transforms.
  • The framework is backbone-agnostic: the same two-stage recipe improves ResNet18, CNN, and TCN encoders, so it can likely be dropped into other time-series encoders as well.
  • With as little as 5% labeled target data, MoSSDA keeps a stable rank, whereas baselines that depend on pseudo-labeling or augmentation degrade at unlabeled ratio 0.95.
  • Ablations indicate that the positive contrastive module and the two-stage separation are the main contributors; removing the contrastive loss collapses F1-score by large margins.

Reading between the lines

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

  • If the reported margins hold, augmentation-free and decoupled training is a stronger recipe for time-series semi-supervised domain adaptation than augment-then-regularize, suggesting future methods for sequential data should first test their augmentations for temporal-structure damage.
  • Because the framework treats source and target positives symmetrically and needs no domain discriminator, it may extend to multi-source and online domain shift settings where stable feature keys are especially valuable.
  • A testable extension is to push the unlabeled ratio below 0.7 to see whether the momentum encoder's stability keeps the rank gap or whether the labeled-target signal becomes too weak.
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

4 major / 5 minor

Summary. The paper proposes MoSSDA, a two-stage semi-supervised domain adaptation framework for multivariate time-series classification. Stage 1 trains a domain-invariant encoder with an MMD loss and a mixup-enhanced supervised contrastive loss, using a momentum encoder for feature stability; Stage 2 freezes the encoder and trains a classifier on labeled source and target data. The method is evaluated on six datasets, three backbone architectures, and three target unlabeled ratios, with Tables 1-3 reporting averaged accuracy and F1, Table 4 giving average ranks, and an ablation study decomposing the contribution of each module. The central claim is that MoSSDA achieves state-of-the-art SSDA performance for multivariate time-series classification, with average ranks of 1.06-1.11 across all settings.

Significance. If the reported results are reproducible, this paper would provide a useful and systematically evaluated SSDA baseline for time-series classification, with broader backbone and dataset coverage than most prior work. Strengths include the decoupled two-stage design, the inclusion of full per-scenario results in the supplement, an ablation study covering all major components, and a provided code link. The significance is currently tempered by three issues: the baseline comparisons are not established as fair or reproducible, the main comparison tables lack variance information, and the specific role of the momentum encoder is underspecified. The broad empirical contribution is therefore promising, but the headline state-of-the-art claim is not yet fully supported.

major comments (4)
  1. [Experiments, Benchmark Methods; Supplementary S.3.2] The central 'state-of-the-art' claim, quantified by the average ranks in Table 4, rests entirely on comparisons with six baseline methods whose implementations, as run, are not reproducible. The paper states that all benchmark methods were adapted by replacing image-specific augmentations with time-series-specific augmentations, but no per-method hyperparameters, training schedules, tuning budgets, or official code versions are given. The reader cannot determine whether the baselines were tuned to the same degree as MoSSDA or whether their official implementations would produce the same numbers. Please provide runnable code or detailed configuration files for all baselines, document the hyperparameter search range and budget per method, and use official implementations where available.
  2. [Tables 1-3; Table 4] Several baseline columns exhibit patterns characteristic of collapsed or degenerate training, which would inflate MoSSDA's rank advantage. CDAC is nearly invariant to the unlabeled ratio in many settings, for example 0.4550 on MFD across Tables 1-3 and values near 0.15 on HAR/HHAR, and DST shows abrupt non-monotonic drops such as WISDM/CNN at u=0.9 (0.5091 accuracy at u=0.7 to 0.3365 at u=0.9 in Table 2). If these implementations did not train successfully, the comparison is not informative. Please show learning curves or validation checks for the baselines, and re-run any method whose accuracy does not respond to the labeled-target fraction.
  3. [Methodology, Eq. (3); Positive Contrastive Module] The role of the momentum encoder is underspecified. Eq. (3) defines a supervised contrastive loss on features z, but the text does not state whether the positive and negative features come from the online encoder h_q, the momentum encoder h_m, or a queue of momentum features; the two encoders are described but never connected to the loss. The ablation study (Tables 5-6 and S1-S4) removes the entire contrastive loss but never ablates the momentum encoder alone, so the paper's title claim that the momentum encoder is responsible for the gains is not demonstrated. Please specify the feature sources in Eq. (3) and add an ablation that replaces the momentum encoder with a direct-gradient encoder.
  4. [Tables 1-3] The main comparison tables report only point estimates. Without standard deviations or confidence intervals over repeated runs or domain splits, the reported margins cannot be distinguished from noise, especially on high-variance datasets such as PTBXL and WISDM where the supplementary tables show MoSSDA standard deviations up to about ±0.15. Please report mean±std (or equivalent) for every method in the main tables, and state the number of seeds or domain-pair folds used to compute each statistic.
minor comments (5)
  1. [Abstract and Methodology, Eq. (2)] The abstract says the method works 'without data augmentation,' but Eq. (2) uses mixup, which is commonly considered a form of augmentation; please rephrase to 'without transformation-based augmentation' to avoid a direct contradiction.
  2. [Experimental Results, t-SNE paragraph] In the t-SNE discussion, the phrase 'the other methods could not distinguish between classes 2, 3, and 5' is ambiguous because CLDA is also mentioned as separating Class 0; please specify exactly which methods are being compared in each sentence.
  3. [Table 4] The table header uses 'OURS'; please use 'MoSSDA' consistently for readability.
  4. [References] The reference list contains duplicate entries for Eldele et al. 2021a and 2021b with the same arXiv identifier; please consolidate or correct these entries.
  5. [Experimental Results, heading] There is a typo in the section heading 'Visulaization with t-SNE'; it should be 'Visualization with t-SNE.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical comparison; the claimed SOTA result rests on external benchmarks and standard loss components, not on a self-referential derivation.

full rationale

MoSSDA's contribution is a training framework whose losses (MMD, supervised contrastive with mixup, cross-entropy) are standard components with external provenance (Khosla et al. 2020; He et al. 2020; Zhang et al. 2017); no predictive quantity is derived from a fitted parameter of the same quantity. The only self-citation is Saito et al. 2019 (co-authored by D. Kim) in the introduction and related work as general support for SSDA effectiveness and as a description of the MME baseline; it is not load-bearing for any claimed result. The SOTA claim is an empirical ranking against six published methods on six public datasets, which is externally falsifiable. The main weakness—baselines adapted with custom time-series augmentations and no per-method tuning budget—is a comparison-fairness/correctness risk, not a circularity risk, because the comparison is not constructed to equal the method's own outputs. Accordingly no circular step can be exhibited by quotation and reduction; score 0.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The paper is purely empirical; no new physical entity or mathematical object is introduced. The main assumptions are standard for SSDA, while the free parameters are common hyperparameters that are not fitted to test data. No parameter-free derivation is attempted.

free parameters (5)
  • τ (temperature) = 0.5
    Temperature in supervised contrastive loss, fixed by hand in Implementation.
  • m (momentum coefficient) = 0.999
    EMA update rate for momentum encoder, fixed by hand in Implementation.
  • α (mixup Beta parameter) = 1
    Shape parameter for Beta distribution in feature mixup, fixed by hand in Implementation.
  • λ_mmd = 0.5
    Weight for MMD loss in total loss, fixed by hand.
  • λ_ctr = 0.5
    Weight for contrastive loss in total loss, fixed by hand.
assumptions (3)
  • domain assumption Source and target share the same label space Y and differ only in input distribution P(X), with covariate shift and no label shift.
    Stated in Problem Formulation; the method and evaluation assume this standard SSDA setting.
  • domain assumption Feature-level mixup preserves class identity, so the interpolated feature zmix can be used as a positive sample of the same class.
    Used in Positive Contrastive Module Eq. (2); this is an unproved heuristic borrowed from mixup literature.
  • domain assumption MMD with a linear kernel is sufficient to align source and target feature distributions for the datasets considered.
    The paper uses a linear kernel for simplicity and says other kernels are possible, but no kernel comparison is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MoSSDA: A Semi-Supervised Domain Adaptation Framework for Multivariate Time-Series Classification using Momentum Encoder." pith.science (2026). https://pith.science/paper/LKCHZ3JL

@misc{pith2026250808280,
  author       = {Pith},
  title        = {Pith review of: MoSSDA: A Semi-Supervised Domain Adaptation Framework for Multivariate Time-Series Classification using Momentum Encoder},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LKCHZ3JL}},
  note         = {Machine review of arXiv:2508.08280}
}
read the original abstract

Deep learning has emerged as the most promising approach in various fields; however, when the distributions of training and test data are different (domain shift), the performance of deep learning models can degrade. Semi-supervised domain adaptation (SSDA) is a major approach for addressing this issue, assuming that a fully labeled training set (source domain) is available, but the test set (target domain) provides labels only for a small subset. In this study, we propose a novel two-step momentum encoder-utilized SSDA framework, MoSSDA, for multivariate time-series classification. Time series data are highly sensitive to noise, and sequential dependencies cause domain shifts resulting in critical performance degradation. To obtain a robust, domain-invariant and class-discriminative representation, MoSSDA employs a domain-invariant encoder to learn features from both source and target domains. Subsequently, the learned features are fed to a mixup-enhanced positive contrastive module consisting of an online momentum encoder. The final classifier is trained with learned features that exhibit consistency and discriminability with limited labeled target domain data, without data augmentation. We applied a two-stage process by separating the gradient flow between the encoders and the classifier to obtain rich and complex representations. Through extensive experiments on six diverse datasets, MoSSDA achieved state-of-the-art performance for three different backbones and various unlabeled ratios in the target domain data. The Ablation study confirms that each module, including two-stage learning, is effective in improving the performance. Our code is available at https://github.com/seonyoungKimm/MoSSDA

Figures

Figures reproduced from arXiv: 2508.08280 by the authors.

Figure 1
Figure 1. Overview of MoSSDA framework. In the decou￾pled two-step framework, the first step involves updating the components (colored in navy): The following part is used to update the classifier (colored in purple). The weights in the other steps were not updated for each step. Ns is the number of source samples. Each sample, Xsrc i ∈ R D×T , is a multivariate time-series instance with D vari￾ables (channels) and T time ste… view at source ↗
Figure 2
Figure 2. t-SNE visualization learned on HHAR 1 to 6 DA [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. emg2tendon: From sEMG Signals to Tendon Control in Musculoskeletal Hands

    cs.RO 2025-07 unverdicted novelty 6.0 of 10

    The authors present emg2tendon, a 193-subject, 370-hour dataset linking sEMG recordings to MyoSuite MyoHand tendon activations, with baseline and diffusion-based regression models.

Reference graph

Works this paper leans on

52 extracted references · 38 canonical work pages · cited by 1 Pith paper

  1. [1]

    L.; et al

    Anguita, D.; Ghio, A.; Oneto, L.; Parra, X.; Reyes-Ortiz, J. L.; et al. 2013. A public domain dataset for human activity recognition using smartphones. In Esann, volume 3, 3--4

    work page 2013

  2. [2]

    Z.; and Koltun, V

    Bai, S.; Kolter, J. Z.; and Koltun, V. 2018. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271

    arXiv 2018

  3. [3]

    Berthelot, D.; Roelofs, R.; Sohn, K.; Carlini, N.; and Kurakin, A. 2021. Adamatch: A unified approach to semi-supervised learning and domain adaptation. arXiv preprint arXiv:2106.04732

    arXiv 2021

  4. [4]

    Chang, C.; Chan, C.-T.; Wang, W.-Y.; Peng, W.-C.; and Chen, T.-F. 2024. TimeDRL: Disentangled Representation Learning for Multivariate Time-Series. In 2024 IEEE 40th International Conference on Data Engineering (ICDE), 625--638. IEEE

    work page 2024

  5. [5]

    Chang, Y.; Mathur, A.; Isopoussu, A.; Song, J.; and Kawsar, F. 2020. A systematic study of unsupervised domain adaptation for robust human-activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 4(1): 1--30

    work page 2020

  6. [6]

    Chen, B.; Jiang, J.; Wang, X.; Wan, P.; Wang, J.; and Long, M. 2022. Debiased self-training for semi-supervised learning. Advances in Neural Information Processing Systems, 35: 32424--32437

    work page 2022

  7. [7]

    Chen, X.; and He, K. 2021. Exploring simple siamese representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 15750--15758

    2021

  8. [8]

    Chen, Y.; Yan, X.; Yang, Y.; Zhang, J.; Zhang, J.; Pan, L.; and Li, J. 2024. Disentangling domain and general representations for time series classification. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, 3834--3842

    work page 2024

Show all 52 references
  1. [9]

    Cheng, L.; and Pan, S. J. 2014. Semi-supervised domain adaptation on manifolds. IEEE transactions on neural networks and learning systems, 25(12): 2240--2249

    2014

  2. [10]

    Z.; Yang, Y.; Webb, G

    Darban, Z. Z.; Yang, Y.; Webb, G. I.; Aggarwal, C. C.; Wen, Q.; and Salehi, M. 2024. DACAD: Domain adaptation contrastive learning for anomaly detection in multivariate time series. arXiv preprint arXiv:2404.11269

    2024 arXiv

  3. [11]

    Deng, F.; Tu, S.; and Xu, L. 2021. Multi-source unsupervised domain adaptation for ECG classification. In 2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 854--859. IEEE

    2021

  4. [12]

    Eldele, E.; Ragab, M.; Chen, Z.; Wu, M.; Kwoh, C.-K.; and Li, X. 2023. Contrastive domain adaptation for time-series via temporal mixup. IEEE Transactions on Artificial Intelligence, 5(3): 1185--1194

    2023

  5. [14]

    K.; Li, X.; and Guan, C

    Eldele, E.; Ragab, M.; Chen, Z.; Wu, M.; Kwoh, C. K.; Li, X.; and Guan, C. 2021 b . Time-series representation learning via temporal and contextual contrasting. arXiv preprint arXiv:2106.14112

    2021 arXiv

  6. [15]

    Eldele, E.; Ragab, M.; Chen, Z.; Wu, M.; Kwoh, C.-K.; Li, X.; and Guan, C. 2022. ADAST: Attentive cross-domain EEG-based sleep staging framework with iterative self-training. IEEE Transactions on Emerging Topics in Computational Intelligence, 7(1): 210--221

    2022

  7. [16]

    Fawaz, H. I. 2020. Deep learning for time series classification. arXiv preprint arXiv:2010.00567

    2020 arXiv

  8. [17]

    Ganin, Y.; and Lempitsky, V. 2015. Unsupervised domain adaptation by backpropagation. In International conference on machine learning, 1180--1189. PMLR

    2015

  9. [18]

    L.; Amaral, L

    Goldberger, A. L.; Amaral, L. A.; Glass, L.; Hausdorff, J. M.; Ivanov, P. C.; Mark, R. G.; Mietus, J. E.; Moody, G. B.; Peng, C.-K.; and Stanley, H. E. 2000. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. circul...

    2000

  10. [19]

    Grill, J.-B.; Strub, F.; Altch \'e , F.; Tallec, C.; Richemond, P.; Buchatskaya, E.; Doersch, C.; Avila Pires, B.; Guo, Z.; Gheshlaghi Azar, M.; et al. 2020. Bootstrap your own latent-a new approach to self-supervised learning. Advances in neural information processing systems...

    2020

  11. [20]

    He, H.; Queen, O.; Koker, T.; Cuevas, C.; Tsiligkaridis, T.; and Zitnik, M. 2023. Domain adaptation for time series under feature and label shifts. In International conference on machine learning, 12746--12774. PMLR

    2023

  12. [21]

    He, K.; Fan, H.; Wu, Y.; Xie, S.; and Girshick, R. 2020. Momentum contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 9729--9738

    2020

  13. [22]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

    2016

  14. [23]

    Iglesias, G.; Talavera, E.; Gonz \'a lez-Prieto, \'A .; Mozo, A.; and G \'o mez-Canaval, S. 2023. Data augmentation techniques in time series domain: a survey and taxonomy. Neural Computing and Applications, 35(14): 10123--10145

    2023

  15. [24]

    V.; and Zhang, Z

    Ilbert, R.; Hoang, T. V.; and Zhang, Z. 2024. Data augmentation for multivariate time series classification: An experimental study. In 2024 IEEE 40th International Conference on Data Engineering Workshops (ICDEW), 128--139. IEEE

    2024

  16. [25]

    Jin, X.; Park, Y.; Maddix, D.; Wang, H.; and Wang, Y. 2022. Domain adaptation for time series forecasting via attention sharing. In International Conference on Machine Learning, 10280--10297. PMLR

    2022

  17. [26]

    Khosla, P.; Teterwak, P.; Wang, C.; Sarna, A.; Tian, Y.; Isola, P.; Maschinot, A.; Liu, C.; and Krishnan, D. 2020. Supervised contrastive learning. Advances in neural information processing systems, 33: 18661--18673

    2020

  18. [27]

    H.; Ngo, B

    Kim, J. H.; Ngo, B. H.; Park, J. H.; Kwon, J. E.; Lee, H. S.; and Cho, S. I. 2022. Distilling and Refining Domain-Specific Knowledge for Semi-Supervised Domain Adaptation. In BMVC, 606

    2022

  19. [28]

    R.; Weiss, G

    Kwapisz, J. R.; Weiss, G. M.; and Moore, S. A. 2011. Activity recognition using cell phone accelerometers. ACM SigKDD Explorations Newsletter, 12(2): 74--82

    2011

  20. [29]

    K.; Zimmer, D.; and Sextro, W

    Lessmeier, C.; Kimotho, J. K.; Zimmer, D.; and Sextro, W. 2016. Condition monitoring of bearing damage in electromechanical drive systems by using motor current signals of electric motors: A benchmark data set for data-driven classification. In PHM society European conference,...

    2016

  21. [30]

    Li, J.; Li, G.; Shi, Y.; and Yu, Y. 2021 a . Cross-domain adaptive clustering for semi-supervised domain adaptation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2505--2514

    2021

  22. [31]

    Li, K.; Liu, C.; Zhao, H.; Zhang, Y.; and Fu, Y. 2021 b . Ecacl: A holistic framework for semi-supervised domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision, 8578--8587

    2021

  23. [32]

    Li, Y.; Li, K.; Chen, C.; Zhou, X.; Zeng, Z.; and Li, K. 2021 c . Modeling temporal patterns with dilated convolutions for time-series forecasting. ACM Transactions on Knowledge Discovery from Data (TKDD), 16(1): 1--22

    2021

  24. [33]

    Liu, Q.; and Xue, H. 2021. Adversarial Spectral Kernel Matching for Unsupervised Time Series Domain Adaptation. In IJCAI, 2744--2750

    2021

  25. [34]

    Long, M.; Cao, Z.; Wang, J.; and Jordan, M. I. 2018. Conditional adversarial domain adaptation. Advances in neural information processing systems, 31

    2018

  26. [35]

    Mishra, S.; Saenko, K.; and Saligrama, V. 2021. Surprisingly simple semi-supervised domain adaptation with pretraining and consistency. arXiv preprint arXiv:2101.12727

    2021 arXiv

  27. [36]

    Ott, F.; R \"u gamer, D.; Heublein, L.; Bischl, B.; and Mutschler, C. 2022. Domain adaptation for time-series classification to mitigate covariate shift. In Proceedings of the 30th ACM international conference on multimedia, 5934--5943

    2022

  28. [37]

    L.; Foo, C.-S.; Chen, Z.; Wu, M.; Kwoh, C.-K.; and Li, X

    Ragab, M.; Eldele, E.; Tan, W. L.; Foo, C.-S.; Chen, Z.; Wu, M.; Kwoh, C.-K.; and Li, X. 2023. Adatime: A benchmarking suite for domain adaptation on time series data. ACM Transactions on Knowledge Discovery from Data, 17(8): 1--18

    2023

  29. [38]

    Saito, K.; Kim, D.; Sclaroff, S.; Darrell, T.; and Saenko, K. 2019. Semi-supervised domain adaptation via minimax entropy. In Proceedings of the IEEE/CVF international conference on computer vision, 8050--8058

    2019

  30. [39]

    Shi, Y.; Ying, X.; and Yang, J. 2022. Deep unsupervised domain adaptation with time series sensor data: A survey. Sensors, 22(15): 5507

    2022

  31. [40]

    H.; Narui, H.; and Ermon, S

    Shu, R.; Bui, H. H.; Narui, H.; and Ermon, S. 2018. A dirt-t approach to unsupervised domain adaptation. arXiv preprint arXiv:1802.08735

    2018 arXiv

  32. [41]

    Singh, A. 2021. Clda: Contrastive learning for semi-supervised domain adaptation. Advances in neural information processing systems, 34: 5089--5101

    2021

  33. [42]

    S.; Kj rgaard, M

    Stisen, A.; Blunck, H.; Bhattacharya, S.; Prentow, T. S.; Kj rgaard, M. B.; Dey, A.; Sonne, T.; and Jensen, M. M. 2015. Smart devices are different: Assessing and mitigatingmobile sensing heterogeneities for activity recognition. In Proceedings of the 13th ACM conference on em...

    2015

  34. [43]

    Sun, S.; et al. 2024. CaudiTS: Causal disentangled domain adaptation of multivariate time series. In Forty-first International Conference on Machine Learning

    2024

  35. [44]

    Thill, M.; Konen, W.; and B \"a ck, T. 2020. Time series encodings with temporal convolutional networks. In International Conference on Bioinspired Methods and Their Applications, 161--173. Springer

    2020

  36. [45]

    Wagner, P.; Strodthoff, N.; Bousseljot, R.; Samek, W.; and Schaeffter, T. 2022. PTB-XL, a large publicly available electrocardiography dataset (version 1.0. 3), 2022. URL https://doi. org/10.13026/kfzx-aw45

    2022 doi

  37. [46]

    I.; Samek, W.; and Schaeffter, T

    Wagner, P.; Strodthoff, N.; Bousseljot, R.-D.; Kreiseler, D.; Lunze, F. I.; Samek, W.; and Schaeffter, T. 2020. PTB-XL, a large publicly available electrocardiography dataset. Scientific data, 7(1): 1--15

    2020

  38. [47]

    R.; and Cook, D

    Wilson, G.; Doppa, J. R.; and Cook, D. J. 2020. Multi-source deep domain adaptation with weak supervision for time-series sensor data. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, 1768--1778

    2020

  39. [48]

    Q.; Chao, W.-L.; and Lim, S.-N

    Yang, L.; Wang, Y.; Gao, M.; Shrivastava, A.; Weinberger, K. Q.; Chao, W.-L.; and Lim, S.-N. 2021. Deep co-training with task decomposition for semi-supervised domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision, 8906--8916

    2021

  40. [49]

    Yoon, J.; Kang, D.; and Cho, M. 2022. Semi-supervised domain adaptation via sample-to-sample self-distillation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 1978--1987

    2022

  41. [50]

    N.; and Lopez-Paz, D

    Zhang, H.; Cisse, M.; Dauphin, Y. N.; and Lopez-Paz, D. 2017. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412

    2017 arXiv

  42. [51]

    Zhang, W.; Liu, Q.; Cong, F. O. W.; Ragab, M.; and Foo, C.-S. 2024. Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 23912--23921

    2024

  43. [52]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  44. [53]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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