Pith. sign in

REVIEW 3 major objections 4 minor 55 references

ReLATE+: Unified Framework for Adversarial Attack Detection, Classification, and Resilient Model Selection in Time-Series Classification

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read ReLATE+ claims that a detect-classify-select pipeline can choose near-Oracle time-series classifiers while cutting computation by 77.68%.

desk verdict The attack detection and group classification are worth reading, but the headline near-Oracle model-selection results are unverified because the evaluation protocol selects the deployed model using test-set labels, and the overhead section is unfinished. read the letter →

arxiv 2508.19456 v1 pith:MJXFKO4K submitted 2025-08-26 cs.CR

classification cs.CR
keywords time-seriesclassificationadversarialattacksmodelselectiondatasetsimilarityFourierandwaveletdetectioncomputationaloverheadreductionmultivariatetimeseriesresilientmachinelearning
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

ReLATE+ is a framework for multivariate time-series classification that tries to make deep-learning model selection cheap and attack-aware. The central claim is that a three-stage pipeline—detect whether incoming data is clean, partially attacked, or fully attacked; classify any detected attack into one of two structural families; then select the best-performing models from the benchmark dataset whose CNN embeddings are most similar—can match the Oracle, the best model found by exhaustive training, almost exactly. Across six multivariate time-series datasets, the paper reports an average 77.68% reduction in computational overhead while staying within 2.02% of Oracle performance and beating random model selection by 13.28%. If this holds, real-time and resource-constrained systems can deploy resilient classifiers without retraining a large model zoo on every new data stream.

What carries the argument

The load-bearing mechanism is dataset-level embedding similarity: a small one-dimensional CNN trained on the training split of each dataset yields feature embeddings, which are L2-normalized and compared with cosine similarity. The most similar benchmark dataset's top three models—ranked by accuracy for clean data and by attack success rate for attacked data—are transferred to the new dataset. Two auxiliary mechanisms support this transfer: a max-of-two Fourier and Wavelet detector with a 13% threshold that splits data into clean, partial, or full attack, and an XGBoost attack-group classifier that maps attacks into two coarse families so the similarity search can be restricted to plausible attack conditions.

What would settle it

Run the framework on a held-out dataset that is embedding-similar to one repository dataset but is known to favor a different model family; if the transferred top-three models underperform random selection, the similarity-transfer premise fails. A direct check is to compare the framework's top-three ranking with the true ranking obtained from exhaustive training on many held-out datasets and measure how often the true best model is absent from the chosen three.

Watch

Extended reading notes

Core claim

The core discovery is that model performance on an unseen time-series dataset can be predicted well enough by the performance of models on the most embedding-similar dataset in a precomputed benchmark. ReLATE+ formalizes this as a detect-classify-select loop: Fourier and Wavelet anomaly scores are combined by taking their maximum to label data as clean, partially attacked, or fully attacked; an XGBoost classifier trained on the benchmark assigns detected attacks to either an iterative-gradient family or an optimization-and-decision family; and a lightweight CNN trained on each dataset produces L2-normalized embeddings compared by cosine similarity to find the nearest benchmark dataset. The top three models from that dataset are then evaluated, and the best among them is deployed. In leave-one-out experiments on six multivariate time-series datasets, the paper reports that this procedure lands within 2.02% of the Oracle, improves on random model selection by 13.28% on average, and reduces total computational overhead by 77.68%.

Load-bearing premise

The load-bearing premise is that when two datasets produce similar feature embeddings, the models that worked best on the old dataset will also work best on the new one.

Editorial extensions

If this is right

  • Practitioners can avoid training all candidate models on each new dataset; only the top three models from the most similar known dataset need to be evaluated.
  • The detection and attack-group stages are lightweight enough that the end-to-end overhead saving of 77.68% survives in streaming settings where retraining cycles are impractical.
  • Exact attack identity is not required: grouping attacks into iterative-gradient versus optimization-and-decision families is enough to guide resilient model selection.
  • The similarity transfer appears to work across domains, since the six benchmark datasets cover different application areas and the reported gap to Oracle stays near 2%.
  • The framework turns a model repository into a reusable asset: each new dataset enriches the benchmark and can improve future selections without retraining earlier entries.

Reading between the lines

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

  • The reported savings should grow with repository size, because ReLATE+ evaluates a fixed top three models while Oracle cost scales with the number of candidate models; this scaling property is not tested in the paper.
  • A natural stress test is a truly out-of-distribution dataset with no close embedding match: the framework should degrade toward random selection rather than toward Oracle, and the paper does not report this regime.
  • The max-of-two Fourier and Wavelet detector implies that any attack leaving no detectable trace in either spectral or time-frequency domain would evade Module 1; the epsilon-0.01 ablation already hints at this boundary.
  • The attack-group abstraction suggests a testable design principle for defenses: perturbation geometry, not the named attack, is what determines which model family survives, which could simplify future robustness benchmarks.
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

3 major / 4 minor

Summary. ReLATE+ is a framework for multivariate time-series classification under adversarial attacks that combines three stages: (1) detection of clean, fully attacked, or partially attacked inputs using Fourier and Wavelet transforms; (2) classification of detected attacks into two groups via an XGBoost classifier trained on the performance benchmark database; and (3) selection of the most resilient model by matching the incoming dataset's CNN-based embeddings to the most similar dataset in the benchmark repository and reusing that dataset's top three models. The paper reports average computational overhead reduction of 77.68%, accuracy within 2.02% of the Oracle, and a 13.28% improvement over random model selection across six UEA datasets and seven adversarial attacks.

Significance. If the reported performance held under a deployment-relevant protocol, the detect-classify-select pipeline would be a useful practical contribution: it addresses the real cost of exhaustive model training and testing in time-series classification under adversarial threat. The paper has some commendable components: an explicit pseudo-code algorithm, an epsilon ablation, a comparison of three similarity metrics, and a clear separation of clean, fully attacked, and partially attacked cases. However, the central quantitative claims are not established because the final model is selected using test-set labels. In addition, the overhead section contains an unresolved placeholder, and no code or data availability statement is provided, so the experimental results cannot be independently audited. The framework's core idea is plausible, but the evidence as presented is insufficient to support the headline numbers.

major comments (3)
  1. [Section V.A.3 / Algorithm 1] Step 2 of Module 3 and Algorithm 1 (lines 8, 17, and 36) select the deployed model by evaluating the three transferred candidates on the test portion of the incoming dataset (testin) and keeping the best performer. This is test-label leakage: the random baseline averages over 1,000 uniformly chosen models without seeing test labels, Oracle is the best of all 14 models on the test set, and ReLATE+ is effectively the best of the top-3 candidates on the test set. The reported 2.02% gap to Oracle and 13.28% improvement over random selection are therefore upper bounds for any deployment-time system. The authors should either select the model using only the validation portion and report test performance of that single selected model, or report the distribution (e.g., min, median, max) of the top-3 candidates instead of only the best.
  2. [Section V.B] The overhead analysis contains an unresolved placeholder: 'Even when including ReLATE+’s internal framework costs, the overall reduction remains high at X% on average.' The abstract and conclusion state a 77.68% average overhead reduction, but the only per-case reductions reported in Section V.B are 81.07%, 73.21%, and 75.73%, which average to about 76.67%, and the overall figure is never computed. As written, the 77.68% claim is not reproducible from the paper and must be either corrected with the actual computation or removed.
  3. [Section III.E.2 / Section V.A] The detection threshold T = 13% is tuned on validation sets 'within the PBD' and then evaluated in a leave-one-out rotation over the same six datasets in Section V.A. It is not stated whether the validation sets used for threshold selection overlap with the validation sets of datasets that later play the role of the incoming dataset; if they do, the detection rates in Tables II-IV are optimistically biased. Additionally, Algorithm 1 loop over all D in B does not explicitly exclude the incoming dataset's own precomputed PBD entry from the similarity and top-3 ranking; the text says the remaining datasets form the drive, but the algorithm and the description should state this exclusion explicitly, since otherwise self-similarity would trivially select the incoming dataset's own best models.
minor comments (4)
  1. [Figures 2 and 4] There are typos and naming inconsistencies: Figure 2 uses 'ReLATE' in the legend while the framework is 'ReLATE+', and Figure 4 labels the y-axis 'Attack Sucess Rate' instead of 'Attack Success Rate'.
  2. [Table III] The last column header is formatted as 'A WR' and should read 'AWR' for ArticularyWordRecognition; this typo also appears in the table's row entries.
  3. [Section V.B] The statement that Module 2 'requires only 3% of the Oracle’s cost' is presented without any measurement basis; the paper should specify how this percentage is computed or cite the corresponding timing measurement.
  4. [Overall] No code, data, or reproducibility statement is provided, which makes it impossible to verify the reported tables and figures; a public artifact or detailed per-dataset results would be needed for a serious journal.

Circularity Check

1 steps flagged · score 6.0 of 10

ReLATE+ reports near-Oracle performance by selecting the best of three candidate models using test-set labels, so the central accuracy/ASR claim is an oracle over the top-3 rather than a deployment-time prediction.

  1. fitted input called prediction [Section III.D Step 2 (Module 3); Algorithm 1 lines 9, 17, 36; reported in Section V.A.3]
    "Next, we assess the performance of the selected models on the new dataset by training them on its training portion and evaluating their accuracy and resilience on the test portion. ... The best model is then deployed on the new dataset for real-time use."

    The deployed model is chosen by evaluating the three transferred candidates on the test portion of the incoming dataset, and the reported 'ReLATE+ performance' is that test-selected best. Oracle is defined as the best of all 14 models on the same test data, while the random baseline is a single uniform draw averaged over 1,000 draws with no test access. Thus the headline results (within 2.02% of Oracle and 13.28% better than random) are not a deployment-time prediction; they are an oracle over the top-3 candidate set. The discrete model choice is fit to the test labels and then scored on those same labels, so the central performance claim is inflated by construction rather than derived from the similarity-based selection alone.

full rationale

ReLATE+ is primarily an empirical benchmark paper, and most of its pipeline is not circular: Module 0 builds a Performance Benchmark Database from explicit train/validation/test splits, Modules 1 and 2 use transform-based detection and XGBoost, and Module 3's similarity transfer (custom CNN embeddings plus cosine similarity) is compared against DTW and Wasserstein. No load-bearing uniqueness theorem is imported from the authors' prior ReLATE paper; the self-citations [15,34] are background and not the argument. The one significant circularity is in Module 3 Step 2 / Algorithm 1: the 'best model' is selected by evaluating the three transferred candidates on the test portion of the incoming dataset, and this test-selected best is then reported as ReLATE+'s performance. Because Oracle is the test-best of all 14 models and the random baseline is a single no-test-label draw, the reported 2.02%-of-Oracle and 13.28%-over-random numbers are an oracle-over-top-3 bound, not a deployable selection result. This is a fitted-input-called-prediction issue, not a self-citation circularity. Separately, the unresolved 'X%' placeholder in Section V.B means the 77.68% average overhead reduction is not fully supported by the reported text, but that is a missing-support concern rather than a circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The framework relies on a small set of domain assumptions rather than a mathematical derivation. The only fitted scalar is the detection threshold T=13%, tuned on validation. No new physical or mathematical entities are introduced.

free parameters (3)
  • detection threshold T = 13%
    Chosen via validation experiments over the PBD to balance false positives and false negatives; applied to the max(Fourier, Wavelet) detection ratio. This threshold is a hand-tuned scalar that directly controls the clean/partial/full classification.
  • epsilon perturbation strength = 0.1
    Selected for all attacks that have adjustable strength; the ablation shows sensitivity to this value, and the text inconsistently reports 0.01 vs 0.001 in the ablation section.
  • top-k models = 3
    The number of candidate models selected from the most similar dataset is set to three without justification or sensitivity analysis.
assumptions (4)
  • domain assumption Fourier and Wavelet anomaly scores with a fixed threshold separate clean, partially attacked, and fully attacked inputs.
    Section III.B defines the three-way classification; the transformation-based detection is assumed to be informative across domains.
  • domain assumption Cosine similarity between CNN embeddings of datasets predicts relative model performance transferability.
    Section III.D Step 1 uses this to select the top-3 models from the most similar dataset; no direct validation of transferability is provided beyond the reported results.
  • domain assumption XGBoost attack-group classifier trained on the PBD generalizes to held-out datasets and random partial attack mixes.
    Section III.C trains on PBD datasets; rotation experiments evaluate on held-out datasets but no external validation is provided.
  • domain assumption The six UEA datasets constitute a representative cross-domain benchmark for time-series classification.
    Section IV says five HAR and one motion dataset; the paper calls this cross-domain, but domain diversity is limited.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ReLATE+: Unified Framework for Adversarial Attack Detection, Classification, and Resilient Model Selection in Time-Series Classification." pith.science (2026). https://pith.science/paper/MJXFKO4K

@misc{pith2026250819456,
  author       = {Pith},
  title        = {Pith review of: ReLATE+: Unified Framework for Adversarial Attack Detection, Classification, and Resilient Model Selection in Time-Series Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MJXFKO4K}},
  note         = {Machine review of arXiv:2508.19456}
}
read the original abstract

Minimizing computational overhead in time-series classification, particularly in deep learning models, presents a significant challenge due to the high complexity of model architectures and the large volume of sequential data that must be processed in real time. This challenge is further compounded by adversarial attacks, emphasizing the need for resilient methods that ensure robust performance and efficient model selection. To address this challenge, we propose ReLATE+, a comprehensive framework that detects and classifies adversarial attacks, adaptively selects deep learning models based on dataset-level similarity, and thus substantially reduces retraining costs relative to conventional methods that do not leverage prior knowledge, while maintaining strong performance. ReLATE+ first checks whether the incoming data is adversarial and, if so, classifies the attack type, using this insight to identify a similar dataset from a repository and enable the reuse of the best-performing associated model. This approach ensures strong performance while reducing the need for retraining, and it generalizes well across different domains with varying data distributions and feature spaces. Experiments show that ReLATE+ reduces computational overhead by an average of 77.68%, enhancing adversarial resilience and streamlining robust model selection, all without sacrificing performance, within 2.02% of Oracle.

Figures

Figures reproduced from arXiv: 2508.19456 by the authors.

Figure 1
Figure 1. Performance of state-of-the-art deep learning models [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. ReLATE+ Framework incoming data as clean (Case 1), fully attacked (Case 2), or partially attacked (Case 3). For this classification task, we apply both Wavelet Transform and Fourier Transform, selected for their complementary strengths in detecting different types of adversarial perturbations. The structure of incoming time￾series data, whether perturbations are localized or spread throughout, is typically unknown. … view at source ↗
Figure 3
Figure 3. Overall Case Explanations 2) Step 2: ReLATE+ Performance Assessment: Here, we find the dataset with the highest similarity score and select its top three performing DL models (ranked by test performance recorded in Module 0: Step 1 and 2). This approach minimizes computational overhead by eliminating the need to retrain or test all models on the incoming dataset. Next, we assess the performance of the selected model… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: shows results for Case 2 across all attack types, along with their average performance across datasets, provid￾ing a comprehensive assessment of ReLATE+’s robustness un￾der different adversarial conditions [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: ReLATE+ Case 3 ASR results Oracle overhead is determined by evaluating all models for each case. This process involves training each model and applying all adversarial attacks, ensuring a comprehensive assessment of their performance. Since ReLATE+ focuses on choosing …
Figure 6
Figure 6. Figure 6: Similarity metric performance comparison [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 43 canonical work pages

  1. [1]

    A robust framework for evaluation of unsupervised time-series anomaly detec- tion,

    O. Gungor, A. Rios, P. Mudgal, N. Ahuja, and T. Rosing, “A robust framework for evaluation of unsupervised time-series anomaly detec- tion,” in International Conference on Pattern Recognition . Springer, 2024, pp. 48–64

  2. [2]

    A review and evaluation of elastic distance functions for time series clustering,

    C. Holder, M. Middlehurst, and A. Bagnall, “A review and evaluation of elastic distance functions for time series clustering,” Knowledge and Information Systems, vol. 66, no. 2, pp. 765–809, 2024

  3. [3]

    Deep learning for time series classification: a review,

    H. Ismail Fawaz, G. Forestier, J. Weber, L. Idoumghar, and P.-A. Muller, “Deep learning for time series classification: a review,” Data mining and knowledge discovery, vol. 33, no. 4, pp. 917–963, 2019

  4. [4]

    Rigorous evaluation of machine learning-based intrusion detection against adversarial attacks,

    O. Gungor, E. Li, Z. Shang, Y . Guo, J. Chen, J. Davis, and T. Rosing, “Rigorous evaluation of machine learning-based intrusion detection against adversarial attacks,” in 2024 IEEE International Conference on Cyber Security and Resilience (CSR) . IEEE, 2024, pp. 152–158

  5. [5]

    Cyberattack event logs classification using deep learning with semantic feature analysis,

    A. Alzu’bi, O. Darwish, A. Albashayreh, and Y . Tashtoush, “Cyberattack event logs classification using deep learning with semantic feature analysis,” Computers & Security , vol. 150, p. 104222, 2025

  6. [6]

    Towards accurate run-time hardware-assisted stealthy malware detection: a lightweight, yet effective time series cnn-based approach,

    H. Sayadi et al., “Towards accurate run-time hardware-assisted stealthy malware detection: a lightweight, yet effective time series cnn-based approach,” Cryptography, vol. 5, no. 4, p. 28, 2021

  7. [7]

    Multiple classification of eeg signals and epileptic seizure diagnosis with combined deep learning,

    M. Varlı and H. Yılmaz, “Multiple classification of eeg signals and epileptic seizure diagnosis with combined deep learning,” Journal of Computational Science, vol. 67, p. 101943, 2023

  8. [8]

    Convolutional neural networks for agricul- tural land use classification from sentinel-2 image time series,

    A.-M. Sim ´on S´anchez et al., “Convolutional neural networks for agricul- tural land use classification from sentinel-2 image time series,” Remote Sensing, vol. 14, no. 21, p. 5373, 2022

Show all 55 references
  1. [9]

    Deep convolutional and LSTM recurrent neural networks for multimodal wearable activity recognition,

    F. J. Ordo ˜nez and D. Roggen, “Deep convolutional and LSTM recurrent neural networks for multimodal wearable activity recognition,” Sensors, vol. 16, no. 1, p. 115, 2016

  2. [10]

    Multi-Scale Temporal Dependent Recurrent Convolutional Neural Network for Financial Time-Series Classification,

    G. Liu, X. Wang, and R. Li, “Multi-Scale Temporal Dependent Recurrent Convolutional Neural Network for Financial Time-Series Classification,” arXiv preprint arXiv:1911.09359 , 2019

  3. [11]

    Learning spectro- temporal features with 3d convolutional neural networks for speech emotion recognition,

    J. Kim, K. P. Truong, G. Englebienne, and V . Evers, “Learning spectro- temporal features with 3d convolutional neural networks for speech emotion recognition,” arXiv preprint arXiv:1708.05071 , 2017

  4. [12]

    A survey on machine learning for recurring concept drifting data streams,

    A. L. Su ´arez-Cetrulo et al., “A survey on machine learning for recurring concept drifting data streams,” Expert Systems with Applications , vol. 213, p. 118934, 2023

  5. [13]

    Rocket: exceptionally fast and accurate time series classification using random convolutional kernels,

    A. Dempster et al., “Rocket: exceptionally fast and accurate time series classification using random convolutional kernels,” Data Mining and Knowledge Discovery, vol. 34, no. 5, pp. 1454–1495, 2020

  6. [14]

    Adversarial-hd: Hyperdimen- sional computing adversarial attack design for secure industrial internet of things,

    O. Gungor, T. Rosing, and B. Aksanli, “Adversarial-hd: Hyperdimen- sional computing adversarial attack design for secure industrial internet of things,” in Proceedings of Cyber-Physical Systems and Internet of Things Week 2023, 2023, pp. 1–6

  7. [15]

    Relate: Resilient learner selection for multivariate time-series classification against adversarial attacks,

    C. I. Kocal, O. Gungor, A. Tartz, T. Rosing, and B. Aksanli, “Relate: Resilient learner selection for multivariate time-series classification against adversarial attacks,” arXiv preprint arXiv:2503.07882 , 2025

  8. [16]

    Transferability in machine learning: from phe- nomena to black-box attacks using adversarial samples,

    N. Papernot et al. , “Transferability in machine learning: from phe- nomena to black-box attacks using adversarial samples,” arXiv preprint arXiv:1605.07277, 2016

  9. [17]

    Adversarial attacks on time series,

    F. Karim, S. Majumdar, and H. Darabi, “Adversarial attacks on time series,” IEEE Transactions on Neural Networks and Learning Systems , vol. 33, no. 10, pp. 5354–5365, 2022

  10. [18]

    How deep learning sees the world: A survey on adversarial attacks & defenses,

    J. C. Costa et al. , “How deep learning sees the world: A survey on adversarial attacks & defenses,” IEEE Access, 2024

  11. [19]

    Explainable multivariate time series classification: a deep neural network which learns to attend to important variables as well as time intervals,

    T.-Y . Hsieh, S. Wang, Y . Sun, and V . Honavar, “Explainable multivariate time series classification: a deep neural network which learns to attend to important variables as well as time intervals,” in Proceedings of the 14th ACM international conference on web search and data...

  12. [20]

    Higher-order cross-structural embedding model for time series analysis,

    G. Lin, C. Shen, and A. Lin, “Higher-order cross-structural embedding model for time series analysis,” arXiv preprint arXiv:2410.22984, 2024

  13. [21]

    Exploiting multi-channels deep convolutional neural networks for multivariate time series classification,

    Y . Zheng et al. , “Exploiting multi-channels deep convolutional neural networks for multivariate time series classification,” Frontiers of Com- puter Science, vol. 10, pp. 96–112, 2016

  14. [22]

    Fast classification of univariate and multivariate time series through shapelet discovery,

    J. Grabocka et al., “Fast classification of univariate and multivariate time series through shapelet discovery,” Knowledge and information systems , vol. 49, pp. 429–454, 2016

  15. [23]

    Benchmarking multivariate time series classification algorithms,

    A. P. Ruiz et al., “Benchmarking multivariate time series classification algorithms,” arXiv preprint arXiv:2007.13156 , 2020

  16. [24]

    Multivariate times series classification through an interpretable representation,

    F. J. Bald ´an and J. M. Ben ´ıtez, “Multivariate times series classification through an interpretable representation,” Information Sciences, vol. 569, pp. 596–614, 2021

  17. [25]

    Deep learning for time series classification: a review,

    H. I. Fawaz, G. Forestier, J. Weber, L. Idoumghar, and P.-A. Muller, “Deep learning for time series classification: a review,” Data Mining and Knowledge Discovery , vol. 33, no. 4, pp. 917–963, 2019

  18. [26]

    Validation and model selection: Three similarity measures compared,

    R. E. Marks, “Validation and model selection: Three similarity measures compared,” Complexity Economics, vol. 2, no. 1, pp. 41–61, 2013

  19. [27]

    A test of relative similarity for model selection in generative models,

    W. Bounliphone, E. Belilovsky, M. B. Blaschko, I. Antonoglou, and A. Gretton, “A test of relative similarity for model selection in generative models,” arXiv preprint arXiv:1511.04581 , 2015

  20. [28]

    Multivariate sample similarity measure for feature selection with a resemblance model,

    T. A. Assegie et al., “Multivariate sample similarity measure for feature selection with a resemblance model,” International Journal of Electrical and Computer Engineering , vol. 13, no. 3, pp. 3359–3366, 2023

  21. [29]

    The wasserstein-fourier distance for stationary time series,

    E. Cazelles, A. Robert, and F. Tobar, “The wasserstein-fourier distance for stationary time series,” IEEE Transactions on Signal Processing , vol. 69, pp. 709–721, 2020

  22. [30]

    Dynamic time warping for time series curve similarity,

    F. Xie, C. Wang, and X. Li, “Dynamic time warping for time series curve similarity,” in Proceedings of the International Conference on Information Science and Technology , 2020

  23. [31]

    Adversarial attacks on multivariate time series,

    S. Harford et al., “Adversarial attacks on multivariate time series,” arXiv preprint arXiv:2004.00410, 2020

  24. [32]

    On the susceptibility and robustness of time series models through adversarial attack and defense,

    A. H. Galib and B. Bashyal, “On the susceptibility and robustness of time series models through adversarial attack and defense,” arXiv preprint arXiv:2301.03703, 2023

  25. [33]

    Benchmarking adversarial attacks and defenses for time-series data,

    S. A. Siddiqui et al. , “Benchmarking adversarial attacks and defenses for time-series data,” in International Conference on Neural Information Processing. Springer, 2020, pp. 544–554

  26. [34]

    Stewart: Stacking ensemble for white-box adversarial attacks towards more resilient data-driven predictive maintenance,

    O. Gungor, T. Rosing, and B. Aksanli, “Stewart: Stacking ensemble for white-box adversarial attacks towards more resilient data-driven predictive maintenance,” Computers in Industry , vol. 140, p. 103660, 2022

  27. [35]

    On detecting adversarial perturbations,

    J. H. Metzen et al. , “On detecting adversarial perturbations,” in Inter- national Conference on Learning Representations (ICLR) , 2017

  28. [36]

    Feature squeezing: Detecting adversarial examples in deep neural networks,

    W. Xu, D. Evans, and Y . Qi, “Feature squeezing: Detecting adversarial examples in deep neural networks,” in 25th Annual Network and Distributed System Security Symposium (NDSS) , 2018

  29. [37]

    A simple unified framework for detecting out-of- distribution samples and adversarial attacks,

    K. Lee et al. , “A simple unified framework for detecting out-of- distribution samples and adversarial attacks,” in Advances in Neural Information Processing Systems (NeurIPS) , vol. 31, 2018

  30. [38]

    Radar: Recurrent autoencoder based detector for adversarial examples on temporal ehr,

    S. Harford, J. Attia, and Z. C. Lipton, “Radar: Recurrent autoencoder based detector for adversarial examples on temporal ehr,” inProceedings of the ACM Conference on Health, Inference, and Learning (CHIL) . ACM, 2020, pp. 162–172

  31. [39]

    From pixels to trajectory: Universal adversarial example detection via temporal imprints,

    H. Xu et al., “From pixels to trajectory: Universal adversarial example detection via temporal imprints,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 15 226–15 235

  32. [40]

    Time series classification from scratch with deep neural networks: A strong baseline,

    Z. Wang et al., “Time series classification from scratch with deep neural networks: A strong baseline,” in 2017 International joint conference on neural networks (IJCNN) . IEEE, 2017, pp. 1578–1585

  33. [41]

    Lstm fully convolutional networks for time series classification,

    F. Karim et al. , “Lstm fully convolutional networks for time series classification,” IEEE access, vol. 6, pp. 1662–1669, 2017

  34. [42]

    Deep gated recurrent and convolutional network hybrid model for univariate time series classification,

    N. Elsayed et al. , “Deep gated recurrent and convolutional network hybrid model for univariate time series classification,” arXiv preprint arXiv:1812.07683, 2018

  35. [43]

    An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,

    S. Bai et al. , “An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,” arXiv preprint arXiv:1803.01271, 2018

  36. [44]

    Multivariate lstm-fcns for time series classification,

    F. Karim et al. , “Multivariate lstm-fcns for time series classification,” Neural networks, vol. 116, pp. 237–245, 2019

  37. [45]

    Inceptiontime: Finding alexnet for time series classification,

    H. Ismail Fawaz et al., “Inceptiontime: Finding alexnet for time series classification,” Data Mining and Knowledge Discovery , vol. 34, no. 6, pp. 1936–1962, 2020

  38. [46]

    Integration of residual network and convolutional neural network along with various activation functions and global pooling for time series classification,

    X. Zou et al., “Integration of residual network and convolutional neural network along with various activation functions and global pooling for time series classification,” Neurocomputing, vol. 367, pp. 39–45, 2019

  39. [47]

    Omni-scale cnns: a simple and effective ker- nel size configuration for time series classification,

    W. Tang et al. , “Omni-scale cnns: a simple and effective ker- nel size configuration for time series classification,” arXiv preprint arXiv:2002.10061, 2020

  40. [48]

    Xcm: An explainable convolutional neural network for multivariate time series classification,

    K. Fauvel, T. Lin, V . Masson, ´E. Fromont, and A. Termier, “Xcm: An explainable convolutional neural network for multivariate time series classification,” Mathematics, vol. 9, no. 23, p. 3137, 2021

  41. [49]

    Towards a novel perspective on adversarial examples driven by frequency,

    Z. Zhang et al., “Towards a novel perspective on adversarial examples driven by frequency,” arXiv preprint arXiv:2404.10202 , 2024. 13

  42. [50]

    A hybrid wavelet-fourier method for next-generation conditional diffusion models,

    A. Kiruluta and A. Lemos, “A hybrid wavelet-fourier method for next-generation conditional diffusion models,” arXiv preprint arXiv:2504.03821, 2025

  43. [51]

    Xgboost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . ACM, 2016, pp. 785–794

  44. [52]

    Uea multivariate time series classification repository,

    A. Bagnall et al., “Uea multivariate time series classification repository,” https://www.timeseriesclassification.com/dataset.php, 2018

  45. [53]

    The great multivariate time series classification bake off: A review and experimental evaluation of recent algorithmic advances,

    A. P. Ruiz, M. Flynn, J. Large, M. Middlehurst, and A. Bagnall, “The great multivariate time series classification bake off: A review and experimental evaluation of recent algorithmic advances,” Data Mining and Knowledge Discovery , vol. 35, pp. 401–449, 2021

  46. [54]

    A global averaging method for dynamic time warping, with applications to clustering,

    F. Petitjean, A. Ketterlin, and P. Ganc ¸arski, “A global averaging method for dynamic time warping, with applications to clustering,” Pattern Recognition, vol. 44, no. 3, pp. 678–693, 2011

  47. [55]

    Wasserstein discriminant analysis,

    R. Flamary and N. Courty, “Wasserstein discriminant analysis,” Machine Learning, vol. 107, no. 12, pp. 2505–2531, 2017

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.