pith. sign in

arxiv: 2606.02670 · v3 · pith:EGMGGRCZnew · submitted 2026-06-01 · 💻 cs.LG · cs.AI

Anomalies in Multivariate Time Series Benchmarks Are Mostly Univariate

Marc Pinet (LIG) , Julien Cumin , Samuel Berlemont , Dominique Vaufreydaz (LIG) This is my paper

Pith reviewed 2026-06-28 15:27 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords multivariate time seriesanomaly detectionbenchmarksunivariate deviationscross-channel correlationsdiagnostic frameworkchannel-independent models
0
0 comments X

The pith

Anomalies in standard multivariate time series benchmarks almost always show up as deviations in single channels.

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

The paper introduces a per-segment diagnostic framework to test whether labeled anomalies in multivariate time series require cross-channel modeling or can be explained by individual channel behavior. Applied to eight common benchmarks, the framework finds that every instance of changed cross-channel correlation is accompanied by at least one channel deviating from its own normal history, and on six benchmarks the majority of anomaly segments are univariate on 89 to 100 percent of their timesteps. Synthetic data with controlled phase shifts and noise confirm that the same framework correctly flags purely cross-channel anomalies when they are engineered to exist. Direct comparison of channel-dependent and channel-independent versions of a recent detector shows no performance gain from cross-channel modeling on the real benchmarks.

Core claim

No cross-channel rupture occurs without an accompanying univariate deviation across a range of reasonable thresholds. On six of the eight benchmarks, at least half of the labeled anomaly segments deviate univariately on 89 percent to 100 percent of their timesteps. The same diagnostic correctly identifies engineered cross-channel anomalies in synthetic data, yet channel-dependent models show no measurable advantage over channel-independent ones on the real benchmarks.

What carries the argument

The per-segment diagnostic framework that classifies each labeled anomaly segment by the presence of univariate channel deviation, cross-channel correlation change, or both.

If this is right

  • Channel-dependent models bring no measurable gain on these benchmarks compared with channel-independent ones.
  • Current public MTSAD benchmarks cannot validate the value of cross-channel modeling.
  • New evaluation sets containing anomalies whose structure genuinely spans multiple channels are required.

Where Pith is reading between the lines

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

  • Effort spent on cross-channel components in MTSAD models may be misdirected for the domains represented by these benchmarks.
  • Univariate anomaly detection methods may already be sufficient for the majority of cases these benchmarks contain.
  • Real-world multivariate anomaly problems outside these eight datasets could still benefit from cross-channel modeling if they contain different anomaly structures.

Load-bearing premise

The diagnostic correctly separates univariate deviations from cross-channel changes when run on the actual benchmark data.

What would settle it

Finding even one benchmark in which a substantial fraction of labeled anomaly segments exhibit cross-channel rupture with no univariate deviation at the thresholds used in the study.

Figures

Figures reproduced from arXiv: 2606.02670 by Dominique Vaufreydaz (LIG), Julien Cumin, Marc Pinet (LIG), Samuel Berlemont.

Figure 1
Figure 1. Figure 1: Sensitivity of the CROSS-CHANNEL count to both thresholds τz and τρ, aggregated across the eight benchmarks for each of the three correlation methods. Each cell reports the number of datasets containing at least one CROSS-CHANNEL segment, followed by the total count of such segments in parentheses. 1 2 3 4 5 6 7 8 9 Anomalous segment length (timesteps) 0 1000 2000 3000 4000 5000 Count [PITH_FULL_IMAGE:fig… view at source ↗
Figure 2
Figure 2. Figure 2: Distribution of segment lengths on SWAN-SF. All labeled anomalies are point-wise (length 1) or near-point-wise, with only two segments exceeding Lmin = 10. ments do not threaten the main claim of Section 4, being too short for the cross-channel test. MSL and SMAP The NASA benchmarks differ from the others. All channels in MSL and SMAP are binary (with the exception of one channel each), which limits the di… view at source ↗
Figure 5
Figure 5. Figure 5: Synthetic protocols that generate cross-channel-only anomalies. NPROLL has its noise removed for clearer view. For each anomalous segment A, a subset SA ⊂ {1, . . . , C} of channels is selected to be corrupted, uniformly at random. The size |SA| is itself drawn uniformly in {1, . . . , C − 1}. Both extremes are excluded because corrupting zero channels would leave the segment unaltered, while corrupting al… view at source ↗
Figure 6
Figure 6. Figure 6: Reconstruction losses from LinearAE (CI and CD) on an NPROLL anomaly segment. CI captures the boundaries but not the whole segment whereas CD fully captures it. Latent dimension is 16 and window size is 64 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

Many recent multivariate time series anomaly detection (MTSAD) models incorporate cross-channel modeling, under the implicit assumption that the structure of anomalies may be spread across multiple channels. We evaluate this assumption on eight widely used public benchmarks by introducing a per-segment diagnostic framework that flags, for each labeled anomaly, whether at least one channel deviates individually from its normal history, whether the cross-channel correlation structure changes, or both. The framework shows that no cross-channel rupture occurs without an accompanying univariate deviation across a range of reasonable thresholds. A complementary metric also reveals that on six of the eight benchmarks, at least half of the labeled anomaly segments deviate univariately on 89% to 100% of their timesteps, reaching 100% on three of these datasets. To verify that our framework captures cross-channel structure when present, we construct synthetic data of phase-shifted sinusoidal channels with shared noise. Each anomalous segment is altered through one of two channel-wise corruptions that preserve the per-channel marginal distribution while breaking cross-channel structure, and our framework correctly characterizes these segments as cross-channel-only. On these data, channel-dependent (CD) models successfully exploit the cross-channel signal whereas channel-independent (CI) ones fail. The CI/CD comparison of a recent SOTA detector on real benchmarks further confirms that CD modeling brings no measurable gain. We conclude that current MTSAD benchmarks are unsuitable for validating cross-channel modeling capabilities, and we call for the development of more structurally diverse evaluation sets. The code for this study is publicly available.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The manuscript evaluates the assumption underlying many MTSAD models that anomalies involve cross-channel structure. On eight public benchmarks it applies a per-segment diagnostic framework that classifies each labeled anomaly segment by the presence of univariate channel deviations, changes in cross-channel correlation structure, or both. The framework finds no cross-channel ruptures without accompanying univariate deviations across thresholds; a complementary per-timestep metric shows that on six benchmarks at least half the anomaly segments deviate univariately on 89–100 % of their timesteps. A synthetic dataset of phase-shifted sinusoids with controlled channel-wise corruptions (preserving marginals while breaking cross-channel dependence) is used to verify that the framework correctly flags cross-channel-only anomalies. Channel-dependent vs. channel-independent model comparisons on both synthetic and real data show no measurable benefit from cross-channel modeling, leading to the conclusion that current benchmarks are unsuitable for validating such capabilities.

Significance. If the central empirical findings hold, the work identifies a structural limitation in existing MTSAD benchmarks that could redirect dataset construction and model evaluation practices. Credit is due for the synthetic control experiment that isolates cross-channel effects while preserving per-channel marginals, the public release of code, and the consistency of results across thresholds. These elements make the claims more falsifiable and reproducible than typical benchmark critiques.

minor comments (2)
  1. [Abstract and §3] Abstract and §3: the phrase 'a range of reasonable thresholds' should be accompanied by the concrete numerical values (or a table) actually used for the univariate deviation and correlation-change detectors so that readers can assess sensitivity without consulting the released code.
  2. [§4] §4 (synthetic data): the precise parameters of the phase-shifted sinusoids and the two channel-wise corruption operators should be stated explicitly (e.g., shift amounts, noise variance, corruption magnitude) rather than left to the code repository.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary of our manuscript, the recognition of the synthetic control experiment, code release, and threshold consistency, and the recommendation for minor revision. No major comments were listed in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper's central claims rest on an explicitly defined per-segment diagnostic framework applied directly to eight public benchmark datasets and validated on independently constructed synthetic data (phase-shifted sinusoids with controlled channel-wise corruptions that preserve marginals). No equations, fitted parameters, or self-citations are used to derive the reported metrics; the framework is tested for its ability to detect cross-channel structure when present, and the real-data results follow from direct measurement of labeled segments. This structure is self-contained against external benchmarks and does not reduce any prediction or uniqueness claim to its own inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the operational definitions of univariate deviation and cross-channel correlation change plus the choice of thresholds used to flag them; these are domain assumptions rather than derived quantities.

free parameters (1)
  • univariate deviation threshold
    A range of reasonable thresholds is used to determine individual-channel deviation; exact values are not data-fitted but selected as reasonable.
axioms (1)
  • domain assumption An anomaly segment can be classified by the presence of univariate deviation, cross-channel correlation change, or both.
    This classification is the core of the per-segment diagnostic framework described in the abstract.

pith-pipeline@v0.9.1-grok · 5817 in / 1370 out tokens · 43065 ms · 2026-06-28T15:27:16.632327+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

41 extracted references · 27 canonical work pages · 1 internal anchor

  1. [1]

    Ad- vances in Time-Series Anomaly Detection: Algorithms, Benchmarks, and Evaluation Measures

    J. Paparrizos, P. Boniol, Q. Liu, and T. Palpanas. “Ad- vances in Time-Series Anomaly Detection: Algorithms, Benchmarks, and Evaluation Measures”. In:KDD. KDD ’25: The 31st ACM SIGKDD Conference on Knowl- edge Discovery and Data Mining. Toronto, ON, Canada: ACM, Aug. 3, 2025, pp. 6151–6161. DOI: 10.1145/ 3711896.3736565

    work page arXiv 2025

  2. [2]

    Graph Neural Network-Based Anomaly Detection in Multivariate Time Series

    A. Deng and B. Hooi. “Graph Neural Network-Based Anomaly Detection in Multivariate Time Series”. In: AAAI. V ol. 35. 5. 2021, pp. 4027–4035.DOI: 10.1609/ aaai.v35i5.16523

    2021

  3. [3]

    Multivariate Time- Series Anomaly Detection via Graph Attention Net- work

    H. Zhao, Y . Wang, J. Duan, C. Huang, D. Cao, Y . Tong, B. Xu, J. Bai, J. Tong, and Q. Zhang. “Multivariate Time- Series Anomaly Detection via Graph Attention Net- work”. In: ICDM. 2020 IEEE International Conference on Data Mining (ICDM). IEEE, Nov. 2020, pp. 841–850. DOI: 10.1109/ICDM50108.2020.00093

    work page doi:10.1109/icdm50108.2020.00093 2020

  4. [4]

    CATCH: Channel-Aware Multivariate Time Series Anomaly Detection via Frequency Patch- ing

    X. Wu, X. Qiu, Z. Li, Y . Wang, J. Hu, C. Guo, H. Xiong, and B. Yang. “CATCH: Channel-Aware Multivariate Time Series Anomaly Detection via Frequency Patch- ing”. In: ICLR. 2025

    2025

  5. [5]

    XCTFormer: Leveraging Cross-Channel and Cross-Time Dependencies for En- hanced Time-Series Analysis

    I. Zexer and O. Azencot. “XCTFormer: Leveraging Cross-Channel and Cross-Time Dependencies for En- hanced Time-Series Analysis”. In: TMLR (Mar. 18, 2026)

    2026

  6. [6]

    A Time Series Is Worth 64 Words: Long-Term Forecast- ing with Transformers

    Y . Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam. “A Time Series Is Worth 64 Words: Long-Term Forecast- ing with Transformers”. In: ICLR. 2023

    2023

  7. [7]

    The Capacity and Robustness Trade-Off: Revisiting the Channel Indepen- dent Strategy for Multivariate Time Series Forecasting

    L. Han, H.-J. Ye, and D.-C. Zhan. “The Capacity and Robustness Trade-Off: Revisiting the Channel Indepen- dent Strategy for Multivariate Time Series Forecasting”. In: IEEE TKDE 36.11 (Nov. 2024), pp. 7129–7142. DOI: 10.1109/TKDE.2024.3400008

    work page doi:10.1109/tkde.2024.3400008 2024

  8. [8]

    X. Qiu, H. Cheng, X. Wu, J. Lu, J. Hu, C. Guo, C. S. Jensen, and B. Yang. A Comprehensive Survey of Deep Learning for Multivariate Time Series Forecasting: A Channel Strategy Perspective. Jan. 27, 2026. DOI: 10. 48550 / arXiv . 2502 . 10721. URL: http : / / arxiv . org/abs/2502.10721 (visited on 04/11/2026). Pre- published

    work page arXiv 2026

  9. [9]

    DCde- tector: Dual Attention Contrastive Representation Learn- ing for Time Series Anomaly Detection

    Y . Yang, C. Zhang, T. Zhou, Q. Wen, and L. Sun. “DCde- tector: Dual Attention Contrastive Representation Learn- ing for Time Series Anomaly Detection”. In: KDD. Aug. 6, 2023, pp. 3033–3045. DOI: 10.1145/3580305. 3599295

    work page doi:10.1145/3580305 2023

  10. [10]

    KAN-AD: Time Series Anomaly Detection with Kolmogorov–Arnold Networks

    Q. Zhou, C. Pei, F. Sun, H. Jing, Z. Gao, H. Zhang, G. Xie, D. Pei, and J. Li. “KAN-AD: Time Series Anomaly Detection with Kolmogorov–Arnold Networks”. In: ICML. V ol. 267. Proceedings of Machine Learning Re- search. PMLR, 2025, pp. 79136–79149

    2025

  11. [11]

    CrossAD: Time Series Anomaly Detection with Cross-Scale Associations and Cross- Window Modeling

    B. Li, Q. Shentu, Y . Shu, H. Zhang, M. Li, N. Jin, B. Yang, and C. Guo. “CrossAD: Time Series Anomaly Detection with Cross-Scale Associations and Cross- Window Modeling”. In: NeurIPS. 2025

    2025

  12. [12]

    Towards a General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders

    Q. Shentu, B. Li, K. Zhao, Y . Shu, Z. Rao, L. Pan, B. Yang, and C. Guo. “Towards a General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders”. In: ICLR. 2025

    2025

  13. [13]

    CiTranGAN: Channel-Independent Based-Anomaly De- tection for Multivariate Time Series Data

    X. Chen, T. Li, Z. Ma, J. Chen, J. Guo, and Z. Liu. “CiTranGAN: Channel-Independent Based-Anomaly De- tection for Multivariate Time Series Data”. In: Elec- tronics 14.9 (May 2, 2025), p. 1857. DOI: 10.3390/ electronics14091857

    2025

  14. [14]

    The Distance Corre- lation T-test of Independence in High Dimension

    G. J. Sz ´ekely and M. L. Rizzo. “The Distance Corre- lation T-test of Independence in High Dimension”. In: J. Multivar. Anal. 117 (May 2013), pp. 193–213. DOI: 10.1016/j.jmva.2013.02.012

    work page doi:10.1016/j.jmva.2013.02.012 2013

  15. [15]

    PatchAD: A Lightweight Patch-Based MLP-Mixer for Time Series Anomaly Detection

    Z. Zhong, Z. Yu, Y . Yang, W. Wang, K. Yang, and C. L. P. Chen. “PatchAD: A Lightweight Patch-Based MLP-Mixer for Time Series Anomaly Detection”. In: IEEE TBD 11 (2025), pp. 3460–3473. DOI: 10.1109/ TBDATA.2025.3596745

    work page arXiv 2025

  16. [16]

    Cluster-Aware Causal Mixer for Online Anomaly Detection in Multivariate Time Series

    M. M. N. Murad and Y . Yilmaz. “Cluster-Aware Causal Mixer for Online Anomaly Detection in Multivariate Time Series”. In: ICML. 2026. DOI: 10.48550/arXiv. 2506.00188

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv 2026

  17. [17]

    DBAD: A Dual-Branch Time Series Anomaly Detec- tion Method Based on Transformer and MLP

    Y . Sun, N. Zhang, C. Zhang, M. Wang, and C. Shi. “DBAD: A Dual-Branch Time Series Anomaly Detec- tion Method Based on Transformer and MLP”. In: AICI. AICI 2025: 2025 International Conference on Artifi- cial Intelligence and Computational Intelligence. Kuala 9 AUTHOR VERSION Lumpur Malaysia: ACM, Feb. 14, 2025, pp. 440–446. DOI: 10.1145/3730436.3730510

    work page doi:10.1145/3730436.3730510 2025

  18. [18]

    arXiv preprint arXiv:2408.14763 , year=

    M. Wang, Z. Xie, B. Chen, H. Liu, and J. Kwok. “Chan- nel Matters: Estimating Channel Influence for Multivari- ate Time Series”. In: NeurIPS. 2026. DOI: 10.48550/ arXiv.2408.14763

    work page arXiv 2026

  19. [19]

    Current time series anomaly detection benchmarks are flawed and are creating the illusion of progress.IEEE Transactions on Knowledge and Data En- gineering, page 1 ˆa=C“1, 2021

    R. Wu and E. Keogh. “Current Time Series Anomaly Detection Benchmarks Are Flawed and Are Creating the Illusion of Progress”. In: IEEE TKDE (2021), pp. 1–1. DOI: 10.1109/TKDE.2021.3112126

    work page doi:10.1109/tkde.2021.3112126 2021

  20. [20]

    In: Proceed- ings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    Y . Su, Y . Zhao, C. Niu, R. Liu, W. Sun, and D. Pei. “Robust Anomaly Detection for Multivariate Time Se- ries through Stochastic Recurrent Neural Network”. In: KDD. KDD ’19: The 25th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. Anchorage AK USA: ACM, July 25, 2019, pp. 2828–2837. DOI: 10.1145/3292500.3330672

    work page doi:10.1145/3292500.3330672 2019

  21. [21]

    TimeSeAD: Benchmarking Deep Multivariate Time-Series Anomaly Detection

    D. Wagner, T. Michels, F. C. F. Schulz, A. Nair, M. Rudolph, and M. Kloft. “TimeSeAD: Benchmarking Deep Multivariate Time-Series Anomaly Detection”. In: TMLR (2023)

    2023

  22. [22]

    Position: Quo Vadis, Unsupervised Time Series Anomaly Detection?

    M. S. Sarfraz, M.-Y . Chen, L. Layer, K. Peng, and M. Koulakis. “Position: Quo Vadis, Unsupervised Time Series Anomaly Detection?” In: ICML. V ol. 235. Pro- ceedings of Machine Learning Research. PMLR, 2024, pp. 43461–43476

    2024

  23. [23]

    In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C

    Q. Liu and J. Paparrizos. “The Elephant in the Room: To- wards a Reliable Time-Series Anomaly Detection Bench- mark”. In: NeurIPS. V ol. 37. Curran Associates, Inc., 2024, pp. 108231–108261. DOI: 10.52202/079017- 3437

    work page doi:10.52202/079017- 2024

  24. [24]

    mTSBench: Bench- marking Multivariate Time Series Anomaly Detection and Model Selection at Scale

    X. Zhou, C. Brif, and I. Lourentzou. “mTSBench: Bench- marking Multivariate Time Series Anomaly Detection and Model Selection at Scale”. In: TMLR (2026)

    2026

  25. [25]

    An Evaluation of Anomaly Detection and Diagno- sis in Multivariate Time Series

    A. Garg, W. Zhang, J. Samaran, R. Savitha, and C. -S. Foo. “An Evaluation of Anomaly Detection and Diagno- sis in Multivariate Time Series”. In:IEEE TNNLS 33.6 (June 2022), pp. 2508–2517. DOI: 10 . 1109 / TNNLS . 2021.3105827

    work page arXiv 2022

  26. [26]

    Anomaly Detectors for Multivariate Time Series: The Proof of the Pudding is in the Eating

    P. Wenig, S. Schmidl, and T. Papenbrock. “Anomaly Detectors for Multivariate Time Series: The Proof of the Pudding is in the Eating”. In: ICDEW. 2024, pp. 96–101. DOI: 10.1109/ICDEW61823.2024.00018

    work page doi:10.1109/icdew61823.2024.00018 2024

  27. [27]

    Ma- trix Profile I: All Pairs Similarity Joins for Time Series: A Unifying View That Includes Motifs, Discords and Shapelets

    C.-C. M. Yeh, Y . Zhu, L. Ulanova, N. Begum, Y . Ding, H. A. Dau, D. F. Silva, A. Mueen, and E. Keogh. “Ma- trix Profile I: All Pairs Similarity Joins for Time Series: A Unifying View That Includes Motifs, Discords and Shapelets”. In: ICDM. 2016 IEEE 16th International Conference on Data Mining (ICDM). Barcelona, Spain: IEEE, Dec. 2016, pp. 1317–1322. DOI...

    work page doi:10.1109/icdm 2016

  28. [28]

    Mea- suring and Testing Dependence by Correlation of Dis- tances

    G. J. Sz ´ekely, M. L. Rizzo, and N. K. Bakirov. “Mea- suring and Testing Dependence by Correlation of Dis- tances”. In: Ann. Statist. 35.6 (Dec. 2007), pp. 2769–

    2007

  29. [29]

    DOI: 10.1214/009053607000000505

    work page doi:10.1214/009053607000000505

  30. [30]

    Im- proved distance correlation estimation

    B. E. Monroy-Castillo, M. A. J´acome, and R. Cao. “Im- proved distance correlation estimation”. In: Applied In- telligence 55.4 (2025), p. 263. DOI: 10.1007/s10489- 024-05940-x

    work page doi:10.1007/s10489- 2025

  31. [31]

    Rethinking Channel Dependence for Multivariate Time Series Forecasting: Learning from Leading Indicators

    L. Zhao and Y . Shen. “Rethinking Channel Dependence for Multivariate Time Series Forecasting: Learning from Leading Indicators”. In: ICLR. 2024

    2024

  32. [32]

    To- wards a Rigorous Evaluation of Time-Series Anomaly Detection

    S. Kim, K. Choi, H.-S. Choi, B. Lee, and S. Yoon. “To- wards a Rigorous Evaluation of Time-Series Anomaly Detection”. In: AAAI. V ol. 36. 7. June 30, 2022, pp. 7194–7201. DOI: 10.1609/aaai.v36i7.20680

    work page doi:10.1609/aaai.v36i7.20680 2022

  33. [33]

    Anomaly Detection in Time Series: A Comprehensive Evaluation

    S. Schmidl, P. Wenig, and T. Papenbrock. “Anomaly Detection in Time Series: A Comprehensive Evaluation”. In: PVLDB 15.9 (May 2022), pp. 1779–1797. DOI: 10. 14778/3538598.3538602

    work page arXiv 2022

  34. [34]

    Moritz, F

    S. Moritz, F. Rehbach, S. Chandrasekaran, M. Rebolledo, and T. Bartz-Beielstein. GECCO Industrial Challenge 2018 Dataset: A Water Quality Dataset for the ’Internet of Things: Online Anomaly Detection for Drinking Water Quality’ Competition at the Genetic and Evolutionary Computation Conference 2018, Kyoto, Japan. Zenodo, Feb. 1, 2018. DOI: 10.5281/ZENODO.3884398

    work page doi:10.5281/zenodo.3884398 2018

  35. [35]

    Hundman, V

    K. Hundman, V . Constantinou, C. Laporte, I. Colwell, and T. Soderstrom. “Detecting Spacecraft Anomalies Us- ing LSTMs and Nonparametric Dynamic Thresholding”. In: KDD. KDD ’18: The 24th ACM SIGKDD Interna- tional Conference on Knowledge Discovery and Data Mining. London United Kingdom: ACM, July 19, 2018, pp. 387–395. DOI: 10.1145/3219819.3219845

    work page doi:10.1145/3219819.3219845 2018

  36. [36]

    Practical Approach to Asynchronous Multivariate Time Series Anomaly Detection and Localization

    A. Abdulaal, Z. Liu, and T. Lancewicki. “Practical Approach to Asynchronous Multivariate Time Series Anomaly Detection and Localization”. In: KDD. KDD ’21: The 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. Virtual Event Singapore: ACM, Aug. 14, 2021, pp. 2485–2494. DOI: 10.1145/ 3447548.3467174

    work page arXiv 2021

  37. [37]

    In: Bue, A.D., Canton, C., Pont-Tuset, J., Tommasi, T

    J. Goh, S. Adepu, K. N. Junejo, and A. Mathur. “A Dataset to Support Research in the Design of Secure Wa- ter Treatment Systems”. In:CRITIS. V ol. 10242. Lecture Notes in Computer Science. Cham: Springer Interna- tional Publishing, 2017, pp. 88–99. DOI: 10.1007/978- 3-319-71368-7_8

    work page doi:10.1007/978- 2017

  38. [38]

    Multivariate Time Series Dataset for Space Weather Data Analytics

    R. A. Angryk, P. C. Martens, B. Aydin, D. Kempton, S. S. Mahajan, S. Basodi, A. Ahmadzadeh, X. Cai, S. Filali Boubrahimi, S. M. Hamdi, M. A. Schuh, and M. K. Georgoulis. “Multivariate Time Series Dataset for Space Weather Data Analytics”. In: Sci. Data 7.1 (July 10, 2020), p. 227. DOI: 10.1038/s41597-020-0548-x

    work page doi:10.1038/s41597-020-0548-x 2020

  39. [39]

    In: Proceedings of the 3rd international workshop on cyber-physical systems for smart water networks

    C. M. Ahmed, V . R. Palleti, and A. P. Mathur. “WADI: A Water Distribution Testbed for Research in the Design of Secure Cyber Physical Systems”. In: CySWater. CPS Week ’17: Cyber Physical Systems Week 2017. Pitts- burgh Pennsylvania: ACM, Apr. 21, 2017, pp. 25–28. DOI: 10.1145/3055366.3055375

    work page doi:10.1145/3055366.3055375 2017

  40. [40]

    Tsay, Aaron Elmore, and Michael J

    J. Paparrizos, P. Boniol, T. Palpanas, R. S. Tsay, A. El- more, and M. J. Franklin. “V olume under the Surface: A New Accuracy Evaluation Measure for Time-Series 10 AUTHOR VERSION Anomaly Detection”. In: PVLDB 15.11 (July 2022), pp. 2774–2787. DOI: 10.14778/3551793.3551830

    work page doi:10.14778/3551793.3551830 2022

  41. [41]

    PUPAE: Intuitive and Actionable Explanations for Time Series Anoma- lies

    A. Der, C.-C. M. Yeh, Y . Zheng, J. Wang, Z. Zhuang, L. Wang, W. Zhang, and E. Keogh. “PUPAE: Intuitive and Actionable Explanations for Time Series Anoma- lies”. In: SDM. 2024, pp. 37–45. DOI: 10 . 1137 / 1 . 9781611978032.5. 11

    2024