Pith. sign in

REVIEW 4 major objections 6 minor 60 references

Causal and Local Correlations Based Network for Multivariate Time Series Classification

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read CaLoNet builds a per-sample transfer-entropy causal graph between dimensions, adds CBAM and sparse-attention local features, and classifies with a GIN; it reports the highest average accuracy (0.770) among 17 methods on 21 UEA datasets.

desk verdict A legitimate but under-specified assembly of known components whose main empirical claim is undermined by an internally inconsistent ablation and an unreported transfer-entropy graph. read the letter →

arxiv 2411.18008 v1 pith:XNG7FMGG submitted 2024-11-27 cs.LG cs.AIstat.MEstat.ML

classification cs.LGcs.AIstat.MEstat.ML
keywords multivariatetimeseriesclassificationtransferentropycausalgraphneuralnetworkGINsparseself-attentionCBAMUEAdatasets
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

This paper proposes CaLoNet, an end-to-end network for classifying multivariate time series that treats the dimensions of a series as nodes in a graph. The authors argue that prior methods neglect either spatial correlations among dimensions or local correlations among features, and that making spatial correlations explicit as a causal graph should improve classification. CaLoNet constructs a per-sample adjacency matrix from transfer entropy, extracts local feature correlations with CBAM and sparse self-attention, and runs a GIN over the graph before an MLP predicts the label. On 21 UEA datasets the paper reports an average accuracy of 0.770, the highest among the 17 compared methods, while noting that the difference from the second-ranked method is not statistically significant.

What carries the argument

The central object is the causal correlation matrix $M$ built from transfer entropy: $M_{i,j}=C_{T_i,T_j}=TE_{T_i\to T_j}-TE_{T_j\to T_i}$ if this exceeds threshold $c$, else 0. It turns each multivariate time series into a directed graph whose nodes are dimensions and whose edge weights are information-transfer strengths. The second machinery is the local correlation network: the series is split into non-overlapping four-step chunks, embedded, refined by CBAM channel/spatial attention, and passed through a log-sparse self-attention layer that restricts dot products to cells at exponential steps, cutting complexity from $O(L^2)$ to $O(L\log L)$. The third is GIN's sum-based neighborhood aggregation $h_v^{(k)}=MLP_k((1+\epsilon_k)h_v^{(k-1)}+\sum_{u\in N(v)}h_u^{(k-1)})$, which propagates node features along the causal graph. The work these pieces do: the matrix supplies explicit, interpretable spatial structure; the attention network supplies long-range local features; the GIN combines the two into graph-level embeddings for an MLP classifier.

What would settle it

Fix the architecture and rerun the 21 UEA experiments while varying the transfer-entropy estimator, history lengths, threshold $c$, and single-sample averaging; if accuracy moves as much as the reported ablation gaps or matches a random-graph control, the causal graph is not the load-bearing ingredient.

Watch

Extended reading notes

Core claim

The paper's central claim is that explicitly representing between-dimension spatial correlations as a causal graph improves multivariate time series classification, and that fusing that graph with local temporal features gives CaLoNet an edge over prior methods. For each sample, transfer entropy from dimension $T_j$ to $T_i$ is computed, the directed strength $C_{T_i,T_j}=TE_{T_i\to T_j}-TE_{T_j\to T_i}$ is thresholded to form an $n\times n$ adjacency matrix, and this matrix is the graph on which a GIN aggregates node features produced by a CBAM-plus-sparse-self-attention local correlation network. Class labels come from an MLP on the resulting graph embedding. On 21 UEA datasets CaLoNet obtains the highest average accuracy of 0.770, wins on 5 datasets, and ranks second in the critical difference diagram behind SMATE; a Wilcoxon test against SMATE gives p=0.9709, so the paper presents the result as competitive rather than significantly better.

Load-bearing premise

The approach assumes that the directed information flow measured by transfer entropy from a single sample of each series is a stable and task-relevant description of how the dimensions relate; if that graph is noisy or dataset-dependent, the claimed benefit of the causal structure collapses.

Editorial extensions

If this is right

  • Explicit causal graphs give the model an interpretable, per-sample representation of which dimensions influence which, so a predicted class can be examined alongside directed edges such as 'pen-tip force drives x and y velocity' in CharacterTrajectories.
  • Removing either component hurts: CaLoNet averages 0.755 versus 0.748 for local correlations alone and 0.693 for causal correlations alone, a 6.2% gain over the causal-only variant.
  • The log-sparse attention mechanism lets the local correlation network scale to long series while the GIN's sparse graph keeps the spatial side cheap.
  • On the epoch curves reported for three datasets, accuracy and loss stabilize after about 30 epochs, indicating that the combined architecture converges within the 50-epoch budget.

Reading between the lines

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

  • Because the paper includes no control that replaces the causal adjacency with a random or correlation-only graph, the ablation does not isolate whether the transfer-entropy graph itself is the source of the gain; a reader should treat 'causal graph helps' as plausible but unproven.
  • The unspecified details of the transfer-entropy estimator, history lengths, threshold, and single-sample estimation make the graph construction a free design choice that could shift the reported margins substantially across datasets.
  • The same per-sample causal graph could be reused as a fixed structure for forecasting or anomaly detection, a direction beyond classification that the paper does not explore.
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 / 6 minor

Summary. This paper proposes CaLoNet, a deep network for multivariate time series classification. It constructs a per-sample causal correlation matrix among dimensions using transfer entropy, thresholds it to define a graph, extracts local correlation features via an embedding layer with CBAM and sparse self-attention, and uses a GIN followed by an MLP for classification. Experiments on 21 UEA datasets compare CaLoNet with 16 baselines and report the highest average accuracy (0.770), with additional ablations, epoch curves, and visualizations of causal graphs and local attention.

Significance. If the experimental claims are reproducible, the contribution is a reasonably novel combination of transfer-entropy-based graph construction with local-correlation feature extraction for MTS classification, evaluated on standard public benchmarks. Strengths include comparison with 16 methods on 21 datasets, release of code, and visualizations of both the causal graphs and the local correlation maps. The main weaknesses are that the core graph construction is under-specified (estimator, history lengths, threshold), the ablation results contradict the main table, and no variance or statistical detail is reported; these currently prevent the empirical claims from being fully supported.

major comments (4)
  1. [Section 3.2, Algorithm 1, Eq. (4)] The core causal graph construction is not reproducible as written. The text never specifies the transfer entropy estimator (e.g., discretization method or kNN estimator), the history lengths k and l, or the threshold c. Since the graph is computed per sample from a single realization (Algorithm 1, lines 2-10), the reliability of the adjacency matrix for short series (e.g., PenDigits length 8, Libras length 45) is questionable. The authors should state the estimator, parameters, threshold selection procedure, and ideally assess the sensitivity of the final accuracy to these choices.
  2. [Tables 2 and 3] The ablation results are inconsistent with the main results for the same model. Table 2 (column 17) reports AWR=0.983, AF=0.414, SWJ=0.533, whereas Table 3 reports CaLoNet AWR=0.960, AF=0.333, SWJ=0.400. Moreover, on the 10 datasets in Table 3, CaLoNet (0.755) is only 0.7 points above Only LCP (0.748), with 5 wins out of 10; this does not establish that the causal graph contributes materially to performance. The authors need to reconcile the two tables or explain the discrepancy in protocols.
  3. [Section 4.2, Table 2] The average accuracy is computed without stating how N/A entries are handled, and the 'Win' counts treat missing baseline results differently across methods; for example, RT(100%) and RT(20%) are missing for CT, JV, and SAD, and several methods have missing entries for EC and MI. If the average ignores missing entries, it is not comparable across methods. The authors should report per-method averages over the same dataset subset or explicitly state the imputation or omission rule.
  4. [Sections 4.1.2 and 4.2] No repeated runs, standard deviations, or confidence intervals are reported for CaLoNet or the deep baselines, and the Wilcoxon test is reported only as a single p-value with no description of the samples used. The text states p=0.9709 and statistic -0.0364, which is unusual for a rank-sum test. The paper should include variance over multiple seeds and a properly described paired significance test, or explicitly justify why this is not needed.
minor comments (6)
  1. [Section 3.3.2, Eq. (8)] The sparse attention mechanism is described as O(log L) and O(L log L) with no formal definition of the sparse bias matrix B; please define the pattern precisely and connect it to Fig. 7.
  2. [Section 3.4, Eq. (9)] The neighbor sum is written as 'P h^{(k-1)}_u µ∈N(v)', which mixes a sum with a set condition; the notation should be corrected to Σ_{u∈N(v)} h^{(k-1)}_u.
  3. [Table 1] The abbreviation for ArticularyWordRecognition is printed as 'A WR' and elsewhere as AWR; consistent naming would improve readability.
  4. [Section 3.3.2 and Figure 6] There are typographical issues in the text and figure ('MatMal', 'Sof tM axis'), and the description of the dot-product score says 'multiplying Q and V' where it should be Q and K.
  5. [Section 4.3] The ablation study reports only 10 of the 21 datasets without explaining why the remaining datasets were omitted; this should be clarified.
  6. [Section 5] The acknowledged limitation that the static graph structure cannot capture dynamic changes should also be discussed in the experimental section as a potential source of error for datasets with time-varying relationships.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CaLoNet's reported accuracies are measured on held-out test data, and the transfer-entropy graph is an unsupervised input rather than a fitted predictor.

full rationale

The paper does not claim to derive a mathematical theorem, and its central empirical claim is that the CaLoNet architecture achieves competitive classification accuracy on UEA datasets. That accuracy is evaluated on held-out test splits, so the result is not forced by construction. The transfer-entropy-based causal graph is computed from each sample's dimensions via Eqs. (1)-(4) before any label information is used; it is an input representation, not a fitted parameter that has been optimized to reproduce the test labels. The GIN update in Eq. (9) and the cross-entropy loss in Eq. (10) are standard supervised components, and the graph itself is not equivalent to the output label. The authors do cite several of their own prior works (e.g., Refs. [1], [26], [57], [58]), but these are used as background or as baselines, not as the load-bearing justification for the claimed prediction. No uniqueness theorem from the authors' prior work is invoked, and no ansatz is smuggled in via self-citation: CBAM, SSA, and GIN are adopted from independent external literature. The under-specification of the transfer-entropy estimator, history lengths, and threshold c is a reproducibility and soundness concern, not a circularity concern, because those choices do not make the prediction equal to its own input. The inconsistency between Tables 2 and 3 for the same model also affects the strength of the empirical evidence, but it does not demonstrate that the derivation reduces to its inputs. Overall, no step in the paper makes a predicted quantity equivalent by construction to a fitted quantity or to a self-cited premise.

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

The method rests on standard deep learning modules plus a transfer entropy graph; the main unaccounted-for degrees of freedom are the transfer entropy estimation details and threshold c, which determine the graph, and the unstated architecture hyperparameters. No new physical entities are introduced.

free parameters (3)
  • threshold c = not reported
    Eq. (4) and Algorithm 1 line 4 keep edges only when the transfer entropy difference exceeds c; the value and tuning procedure are not given in Section 4.1, so the graph structure can depend on an unstated fitted constant.
  • TE history lengths k and l = not reported
    Eq. (1) defines transfer entropy using past vectors x_t^(k) and y_t^(l); no lag lengths or estimator (binning or kernel) are specified, affecting every edge weight.
  • architecture hyperparameters (embedding width, SSA layers and heads, GIN layers, MLP sizes) = not reported
    Section 3 lists modules but no dimensions, layer counts, or regularization; these are chosen by hand and influence accuracy.
assumptions (4)
  • domain assumption Transfer entropy difference C_{X,Y} = TE_{X to Y} minus TE_{Y to X} indicates causal influence between dimensions.
    Section 3.2 treats transfer entropy as causality, but transfer entropy measures predictive information and is not proof of causation; the causal interpretation is assumed.
  • domain assumption A fixed per-sample graph computed from one time series realization is a stable input for GNN classification.
    Section 3.2 constructs the adjacency from a single MTS sample; no variance or significance testing of transfer entropy estimates is provided.
  • domain assumption The log-sparse attention mask captures the local correlations needed for classification.
    Section 3.3.2 adopts the log-sparse pattern from [52] without validating that this sparsity preserves discriminative local features for these datasets.
  • domain assumption Baseline accuracies from different original papers are directly comparable.
    Section 4.2 imports baseline numbers from external papers without re-running them under identical protocols or reporting standard deviations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal and Local Correlations Based Network for Multivariate Time Series Classification." pith.science (2026). https://pith.science/paper/XNG7FMGG

@misc{pith2026241118008,
  author       = {Pith},
  title        = {Pith review of: Causal and Local Correlations Based Network for Multivariate Time Series Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XNG7FMGG}},
  note         = {Machine review of arXiv:2411.18008}
}
read the original abstract

Recently, time series classification has attracted the attention of a large number of researchers, and hundreds of methods have been proposed. However, these methods often ignore the spatial correlations among dimensions and the local correlations among features. To address this issue, the causal and local correlations based network (CaLoNet) is proposed in this study for multivariate time series classification. First, pairwise spatial correlations between dimensions are modeled using causality modeling to obtain the graph structure. Then, a relationship extraction network is used to fuse local correlations to obtain long-term dependency features. Finally, the graph structure and long-term dependency features are integrated into the graph neural network. Experiments on the UEA datasets show that CaLoNet can obtain competitive performance compared with state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2411.18008 by the authors.

Figure 1
Figure 1. Examples of spatial correlations and local correlations in MTSs. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overall structure of CaLoNet. and yt+1 represent the future state. In addition, x (k) t = [xt , xt−1, ..., xt−k+1] and y (l) t = [yt , yt−1, ..., yt−l+1]. T Ey→x = Xp  xt+1, x (k) t , y (l) t  log2 p  xt+1 | x (k) t , y (l) t  p  xt+1 | x (k) t  = Xp  xt+1, x (k) t , y (l) t  log2 p  xt+1 | x (k) t , y (l) t  − Xp  xt+1, x (k) t  log2 p  xt+1 | x (k) t  =H (Xt+1 | Xt) − H (Xt+1 | Xt , Yt), (1) In Eq.(1… view at source ↗
Figure 3
Figure 3. Example of a causal correlation matrix. causal correlation is significant. An example of a causal correlation matrix is shown in Fig.3, in which a 6×6 causal correlations matrix has been obtained for an MTS with six dimensions using this procedure. Mi,j =  CTi,Tj , CTi,Tj > c 0, otherwise (4) Transfer entropy is a method used to quantify the causal relationships between MTSs. It is based on the concept of informati… view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Local correlation network. second pass [49], which moves the temporal blocks within the window. This solves the problem of global features being restricted to the local window partition. Each patch can then interact with other patches in a new window. 3.3.1. CBAM layer…
Figure 5
Figure 5. Figure 5: CBAM block. attention feature map is generated by the one-dimensional CNN, it is com￾puted as follows, where f a denotes a one-dimensional convolution operation with filter size a, and E ′ are features from the channel attention layer. E ′′ = sigmoid(f a ([AvgPool(E ′ …
Figure 6
Figure 6. Figure 6: SSA mechanism. focus on the features of a particular week. Therefore, we restrict the cells to two adjacent layers. That is, we only allow each layer of cells to participate in the dot product operation for those cells with previous exponential steps. Thus, the SSA lay…
Figure 7
Figure 7. Figure 7: Principle of the SSA mechanism. The SSA is computed by only allowing each [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Node embedding process. correlations. The input MTS is converted into a feature matrix H ∈ Fn×d, where d is the calculated number of features, as introduced in Section 3.2. Matrix H can be viewed as a graph feature matrix with n nodes, which correspond to the six varia…
Figure 9
Figure 9. Figure 9: Critical difference diagram for the proposed method and baseline methods. [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: Pairwise accuracy comparison between CaLoNet and Only LCP or Only CCP. [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: Accuracy at different numbers of epochs. [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]
Figure 12
Figure 12. Figure 12: Loss at different numbers of epochs. The BasicMotions dataset was generated by four students who wore smartwatches while performing four activities: walking, resting, running, and badminton. The watches collected data from 3D accelerometers and 3D gyroscopes. Each par…
Figure 13
Figure 13. Figure 13: Heatmap of different samples from the BasicMotions dataset. [PITH_FULL_IMAGE:figures/full_fig_p026_13.png]
Figure 14
Figure 14. Figure 14: Causal correlation-based graph structures from various datasets. [PITH_FULL_IMAGE:figures/full_fig_p027_14.png]
Figure 15
Figure 15. Figure 15: Heatmaps for various datasets. 5. Conclusion In this paper, we proposed CaLoNet, a new end-to-end deep-learning model for exploring local and causal correlations. CaLoNet first leverages pairwise spatial correlations between dimensions based on causal correlations to …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 51 canonical work pages

  1. [1]

    C. Ji, M. Du, Y. Hu, S. Liu, L. Pan, X. Zheng, Time series classification based on temporal features, Applied Soft Computing 128 (2022) 109494

  2. [2]

    H. Ma, W. Li, X. Zhang, S. Gao, S. Lu, Attnsense: Multi-level attention mechanism for multimodal human activity recognition., in: IJCAI, 2019, pp. 3109–3115

  3. [3]

    Strodthoff, C

    N. Strodthoff, C. Strodthoff, Detecting and interpreting myocardial in- farction using fully convolutional neural networks, Physiological mea- surement 40 (1) (2019) 015001

  4. [4]

    Y. Li, M. Chen, X. Jiang, L. Liu, B. Han, L. Zhang, S. Wei, An atrial fibrillation detection algorithm based on lightweight design architecture and feature fusion strategy, Biomed. Signal Process. Control. 91 (2024) 106016

  5. [5]

    J. H. Cochrane, Time series for macroeconomics and finance, Manuscript, University of Chicago 15 (2005) 16

  6. [6]

    Y. Yu, X. Zeng, X. Xue, J. Ma, Lstm-based intrusion detection system for vanets: A time series classification approach to false message detec- tion, IEEE Transactions on Intelligent Transportation Systems 23 (12) (2022) 23906–23918

  7. [7]

    R. Zuo, G. Li, B. Choi, S. S. Bhowmick, D. N. yin Mah, G. L. Wong, Svp-t: A shape-level variable-position transformer for multivariate time series classification, in: AAAI Conference on Artificial Intelligence, 2023

  8. [8]

    H. Liu, X. Liu, D. Yang, Z. Liang, H. Wang, Y. Cui, J. Gu, Todynet: Temporal dynamic graph neural network for multivariate time series classification, ArXiv abs/2304.05078 (2023)

Show all 60 references
  1. [9]

    L. Ye, E. Keogh, Time series shapelets: a new primitive for data mining, in: Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, 2009, pp. 947–956. 29

  2. [10]

    Z. Duan, H. Xu, Y. Huang, J. Feng, Y. Wang, Multivariate time series forecasting with transfer entropy graph, Tsinghua Science and Technol- ogy 28 (1) (2022) 141–149

  3. [11]

    D. Cao, Y. Wang, J. Duan, C. Zhang, X. Zhu, C. Huang, Y. Tong, B. Xu, J. Bai, J. Tong, et al., Spectral temporal graph neural network for multivariate time-series forecasting, Advances in neural information processing systems 33 (2020) 17766–17778

  4. [12]

    M. Chen, Y. Li, L. Zhang, L. Liu, B. Han, W. Shi, S. Wei, Elimination of random mixed noise in ecg using convolutional denoising autoencoder with transformer encoder, IEEE Journal of Biomedical and Health In- formatics 28 (2024) 1993–2004

  5. [13]

    O. B. Bahri, P. Li, S. F. Boubrahimi, S. M. Hamdi, Shapelet-based tem- poral association rule mining for multivariate time series classification, 2022 IEEE International Conference on Big Data (Big Data) (2022) 242–251

  6. [14]

    C. Yang, X. Wang, L. Yao, G. Long, G. Xu, Dyformer: A dynamic transformer-based architecture for multivariate time series classification, Inf. Sci. 656 (2023) 119881

  7. [15]

    Z. Xiao, H. Xing, R. Qu, L. Feng, S. Luo, P. Dai, B. Zhao, Y. Dai, Densely knowledge-aware network for multivariate time series classifica- tion, IEEE Transactions on Systems, Man, and Cybernetics: Systems 54 (2024) 2192–2204

  8. [16]

    Y. Wang, Y. Zhao, L. Petzold, Enhancing transformer efficiency for multivariate time series classification, ArXiv abs/2203.14472 (2022)

  9. [17]

    Cheng, Q

    M. Cheng, Q. Liu, Z. Liu, Z. Li, Y. Luo, E. Chen, Formertime: Hierarchi- cal multi-scale representations for multivariate time series classification, Proceedings of the ACM Web Conference 2023 (2023)

  10. [18]

    N. M. Foumani, C. W. Tan, G. I. Webb, M. Salehi, Improving position encoding of transformers for multivariate time series classification, Data Mining and Knowledge Discovery (2023) 1–27

  11. [19]

    Y. Chen, X. Zhou, Z. Xing, Z. Liu, M. Xu, Cass: A channel-aware self-supervised representation learning framework for multivariate time 30 series classification, in: International Conference on Database Systems for Advanced Applications, 2022

  12. [20]

    Z. Wang, T. Jiang, Z. Xu, J. Zhang, J. Gao, Irregularly sampled mul- tivariate time series classification: A graph learning approach, IEEE Intelligent Systems 38 (2023) 3–11

  13. [21]

    Y. Wang, Y. Xu, J. Yang, M. Wu, X. Li, L. Xie, Z. Chen, Fully-connected spatial-temporal graph for multivariate time series data, in: AAAI Con- ference on Artificial Intelligence, 2023

  14. [22]

    Y. Wang, M. Wu, X. Li, L. Xie, Z. Chen, Multivariate time-series rep- resentation learning via hierarchical correlation pooling boosted graph neural network, IEEE Transactions on Artificial Intelligence 5 (2024) 321–333

  15. [23]

    A. P. Ruiz, M. Flynn, J. Large, M. Middlehurst, A. Bagnall, The great multivariate time series classification bake off: a review and experimen- tal evaluation of recent algorithmic advances, Data Mining and Knowl- edge Discovery 35 (2) (2021) 401–449

  16. [24]

    Abanda, U

    A. Abanda, U. Mori, J. A. Lozano, A review on distance based time series classification, Data Mining and Knowledge Discovery 33 (2) (2019) 378–412

  17. [25]

    G´ orecki, M

    T. G´ orecki, M. Luczak, Multivariate time series classification with para- metric derivative dynamic time warping, Expert Systems with Applica- tions 42 (5) (2015) 2305–2312

  18. [26]

    C. Ji, Y. Hu, S. Liu, L. Pan, B. Li, X. Zheng, Fully convolutional net- works with shapelet features for time series classification, Information Sciences 612 (2022) 835–847

  19. [27]

    Sch¨ afer, U

    P. Sch¨ afer, U. Leser, Multivariate time series classification with weasel+ muse, arXiv preprint arXiv:1711.11343 (2017)

  20. [28]

    M. G. Baydogan, G. Runger, Time series representation and similarity based on local autopatterns, Data Mining and Knowledge Discovery 30 (2016) 476–509. 31

  21. [29]

    LeCun, Y

    Y. LeCun, Y. Bengio, G. Hinton, et al., Deep learning. nature, 521 (7553), 436-444, Google Scholar Google Scholar Cross Ref Cross Ref (2015) 25

  22. [30]

    Z. Chen, Y. Liu, J. Zhu, Y. Zhang, R. Jin, X. He, J. Tao, L. Chen, Time- frequency deep metric learning for multivariate time series classification, Neurocomputing 462 (2021) 221–237

  23. [31]

    Zheng, Q

    Y. Zheng, Q. Liu, E. Chen, Y. Ge, J. L. Zhao, Time series classification using multi-channels deep convolutional neural networks, in: Web-Age Information Management: 15th International Conference, W AIM 2014, Macau, China, June 16-18, 2014. Proceedings 15, Springer, 2014, pp. 298–310

  24. [32]

    A. M. Tripathi, R. D. Baruah, Multivariate time series classification with an attention-based multivariate convolutional neural network, in: 2020 International Joint Conference on Neural Networks (IJCNN), IEEE, 2020, pp. 1–8

  25. [33]

    Huang, C

    Z. Huang, C. Yang, X. Chen, X. Zhou, G. Chen, T. Huang, W. Gui, Functional deep echo state network improved by a bi-level optimization approach for multivariate time series classification, Applied Soft Com- puting 106 (2021) 107314

  26. [34]

    Zhang, Y

    X. Zhang, Y. Gao, J. Lin, C.-T. Lu, Tapnet: Multivariate time series classification with attentional prototypical network, in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 34, 2020, pp. 6845– 6852

  27. [35]

    Z. Xiao, X. Xu, H. Xing, S. Luo, P. Dai, D. Zhan, Rtfn: a robust tem- poral feature network for time series classification, Information sciences 571 (2021) 65–86

  28. [36]

    J. Zuo, K. Zeitouni, Y. Taher, Smate: Semi-supervised spatio-temporal representation learning on multivariate time series, in: 2021 IEEE In- ternational Conference on Data Mining (ICDM), IEEE, 2021, pp. 1565– 1570

  29. [37]

    D. Yang, H. Chen, Y. Song, Z. Gong, Granger causality for multivariate time series classification, in: 2017 IEEE international conference on big knowledge (ICBK), IEEE, 2017, pp. 103–110. 32

  30. [38]

    Zha, K.-H

    D. Zha, K.-H. Lai, K. Zhou, X. Hu, Towards similarity-aware time- series classification, in: Proceedings of the 2022 SIAM International Conference on Data Mining (SDM), SIAM, 2022, pp. 199–207

  31. [39]

    M. Liu, S. Ren, S. Ma, J. Jiao, Y. Chen, Z. Wang, W. Song, Gated transformer networks for multivariate time series classification, arXiv preprint arXiv:2103.14438 (2021)

  32. [40]

    Karim, S

    F. Karim, S. Majumdar, H. Darabi, S. Harford, Multivariate lstm-fcns for time series classification, Neural networks 116 (2019) 237–245

  33. [41]

    J. Hu, L. Shen, G. Sun, Squeeze-and-excitation networks, in: Proceed- ings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

  34. [42]

    R. Chen, X. Yan, S. Wang, G. Xiao, Da-net: Dual-attention network for multivariate time series classification, Information Sciences 610 (2022) 472–487

  35. [43]

    Y. Hao, H. Cao, A new attention mechanism to classify multivariate time series, in: Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, 2020

  36. [44]

    B. Hong, Z. Yan, Y. Chen, et al., Long memory gated recurrent unit for time series classification, in: Journal of Physics: Conference Series, Vol. 2278, IOP Publishing, 2022, p. 012017

  37. [45]

    Zhang, H

    W. Zhang, H. Zhang, Z. Zhao, P. Tang, Z. Zhang, Attention to both global and local features: A novel temporal encoder for satellite image time series classification, Remote Sensing 15 (3) (2023) 618

  38. [46]

    C. W. Granger, Investigating causal relations by econometric models and cross-spectral methods, Econometrica: journal of the Econometric Society (1969) 424–438

  39. [47]

    Seth, Granger causality, Scholarpedia 2 (7) (2007) 1667

    A. Seth, Granger causality, Scholarpedia 2 (7) (2007) 1667

  40. [48]

    Schreiber, Measuring information transfer, Physical review letters 85 (2) (2000) 461

    T. Schreiber, Measuring information transfer, Physical review letters 85 (2) (2000) 461. 33

  41. [49]

    Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, B. Guo, Swin transformer: Hierarchical vision transformer using shifted windows, in: Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 10012–10022

  42. [50]

    S. Woo, J. Park, J.-Y. Lee, I. S. Kweon, Cbam: Convolutional block attention module, in: Proceedings of the European conference on com- puter vision (ECCV), 2018, pp. 3–19

  43. [51]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, Attention is all you need, Advances in neural information processing systems 30 (2017)

  44. [52]

    S. Li, X. Jin, Y. Xuan, X. Zhou, W. Chen, Y.-X. Wang, X. Yan, Enhanc- ing the locality and breaking the memory bottleneck of transformer on time series forecasting, Advances in neural information processing sys- tems 32 (2019)

  45. [53]

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, W. Zhang, In- former: Beyond efficient transformer for long sequence time-series fore- casting, in: Proceedings of the AAAI conference on artificial intelligence, Vol. 35, 2021, pp. 11106–11115

  46. [54]

    K. Xu, W. Hu, J. Leskovec, S. Jegelka, How powerful are graph neural networks?, arXiv preprint arXiv:1810.00826 (2018)

  47. [55]

    Feremans, B

    L. Feremans, B. Cule, B. Goethals, Petsc: pattern-based embedding for time series classification, Data Mining and Knowledge Discovery 36 (3) (2022) 1015–1061

  48. [56]

    Y. Chen, B. Hu, E. Keogh, G. E. Batista, Dtw-d: time series semi- supervised learning from a single example, in: Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining, 2013, pp. 383–391

  49. [57]

    M. Du, Y. Wei, X. Zheng, C. Ji, Multi-feature based network for multi- variate time series classification, Inf. Sci. 639 (2023) 119009

  50. [58]

    M. Du, Y. Wei, Y. Hu, X. Zheng, C. Ji, Multivariate time series clas- sification based on fusion features, Expert Systems with Applications (2024). 34

  51. [59]

    Wilcoxon, Individual comparisons by ranking methods, Biometrics 1 (1945) 196–202

    F. Wilcoxon, Individual comparisons by ranking methods, Biometrics 1 (1945) 196–202

  52. [60]

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-cam: Visual explanations from deep networks via gradient-based localization, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 618–626. 35

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.