Pith. sign in

REVIEW 5 major objections 5 minor 28 references

Temporal Attention Evolutional Graph Convolutional Network for Multivariate Time Series Forecasting

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A graph rebuilt at each time step from recent patterns forecasts traffic better than fixed-graph models, with the clearest gain at the 60-minute horizon.

desk verdict The architecture is a sensible recombination of existing pieces and the METR-LA 60-minute numbers are eye-catching, but the paper's own PEMS-BAY rows contradict its 'superior performance' claim and no code or seeds are provided. read the letter →

arxiv 2505.00302 v1 pith:QQVKTIRS submitted 2025-05-01 cs.LG

Xinlong Zhao , Liying Zhang , Tianbo Zou , Yan Zhang This is my paper · ORCID
classification cs.LG
keywords multivariatetimeseriesforecastinggraphconvolutionalnetworkmulti-headself-attentionevolvablestructuredynamicspatialdependencytemporalcausalitytrafficlearning
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 tries to establish that a multivariate forecaster gains accuracy when the graph describing inter-variable dependence is rebuilt at each time period from the series' own temporal features, rather than fixed after learning. The proposed TAEGCN pairs masked multi-head self-attention (TMSA) with a GRU-driven Evolvable Graph Construction module (EGC): attention produces per-time-step features, EGC turns them into a fresh adjacency matrix, and graph convolution uses that matrix to forecast. On METR-LA the model reports the lowest errors among seven baselines at 15, 30, and 60 minutes; on PEMS-BAY it leads at 15 and 30 minutes and is essentially level with the strongest baseline at 60. The clearest single margin is METR-LA at 60 minutes, where MAE falls from Graph-WaveNet's 3.53 to 3.19 and MAPE from 10.01% to 8.73%. The reason to care is that road networks, financial markets, and power systems all have correlations that change over hours, while most graph forecaster baselines assume one static graph.

What carries the argument

The machinery is the Evolvable Graph Construction (EGC) module's time-varying adjacency matrix, the table of edge strengths between nodes. EGC assumes the graph is constant inside a short time interval and evolves between intervals; a Gated Recurrent Unit with reset and update gates maintains a node state initialized from static node features, and the adjacency entry between nodes is computed from the pair of states by a multi-layer perceptron, then gated by a second multi-layer perceptron through a sigmoid. The temporal module TMSA, causal masked multi-head self-attention with window sizes 1, 3, 6, and 12 across layers, supplies the per-time-step node features from which the graph is built, and the fully connected layer after attention keeps every layer's output at the same time length. The role of this machinery is to make spatial dependency a function of the recent temporal pattern, so a 60-minute forecast can use different neighborhoods than a 15-minute forecast.

What would settle it

Use a traffic dataset with logged external events and compare TAEGCN against Graph-WaveNet only in windows right after those events; if TAEGCN shows no extra advantage there, or if its learned adjacency matrix does not change when the event happens, the dynamic graph is not capturing the true driver of spatial change.

Watch

Extended reading notes

Core claim

The central claim is that temporal and spatial structure can be made consistent by construction: causal masked attention produces node features that still carry every time step, a gated recurrent unit evolves node states from those features, an adjacency matrix is built from pairwise states, and graph convolution consumes it. The paper reports that this joint scheme beats fixed-graph and self-learned-graph baselines on both traffic datasets at most horizons, with the biggest gap at 60 minutes on METR-LA. The ablation is the load-bearing evidence: replacing the dynamic graph with a fixed adjacency matrix costs more accuracy than replacing the temporal attention with a plain TCN, so the evolving graph is contributing beyond extra capacity.

Load-bearing premise

The load-bearing premise is that all spatial dependencies needed for accurate forecasts can be learned from the historical series alone, with no weather, incident, or holiday information required.

Editorial extensions

If this is right

  • Because the graph is learned from the series alone, TAEGCN does not require a precomputed road-network distance matrix, so the method applies to domains without a known map.
  • The reported MAE advantage over Graph-WaveNet on METR-LA grows from 0.05 at 15 minutes to 0.24 at 30 minutes and 0.34 at 60 minutes, so the benefit of a dynamic graph is largest at the longest horizon tested.
  • The ablations imply that the dynamic graph module contributes more than temporal attention in the paper's setup: on METR-LA, removing EGC raises MAE by 0.17 at 30 minutes and 0.26 at 60 minutes, while removing TMSA raises it by 0.11 and 0.12.
  • The reported margin is dataset-dependent: on METR-LA the 60-minute MAE advantage over Graph-WaveNet is 0.34, while on PEMS-BAY at 60 minutes the model is essentially tied, with MAE 1.98 versus 1.95.

Reading between the lines

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

  • Porting EGC to financial or energy series with regime-switching correlations is a natural next step; because EGC needs no external map, any sufficiently long multivariate series is a candidate domain.
  • The modularity of the design suggests a benchmark in which TMSA is replaced by a cheaper causal temporal encoder while EGC is kept, isolating how much of the gain is due to graph evolution rather than attention.
  • A practical implication the paper does not pursue is that the evolving adjacency matrix itself is a diagnostic of changing network structure, so the model could be used to detect when two sensors or variables start or stop moving together.
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

5 major / 5 minor

Summary. The paper proposes TAEGCN, a spatio-temporal graph convolutional network for multivariate time series forecasting. The model stacks spatio-temporal layers, each with a Temporal Multi-head Self-Attention (TMSA) module that uses causal masks of increasing window sizes to extract multi-scale temporal features, an Evolvable Graph Construction (EGC) module that updates the adjacency matrix through a GRU based on temporal features, and a graph convolutional network for spatial processing. The authors evaluate TAEGCN on METR-LA and PEMS-BAY traffic datasets at 15-, 30-, and 60-minute horizons, reporting improvements over baselines such as Graph-WaveNet, and include ablation studies for the TMSA and EGC modules.

Significance. The problem of learning dynamic graph structure for multivariate time series is timely, and the proposed combination of causal masked self-attention with a GRU-driven adjacency update is a reasonable design. If the reported METR-LA results are reproducible (e.g., MAE 3.19 versus 3.53 at 60 minutes), TAEGCN would be a competitive forecaster, and the ablation study credits both TMSA and EGC with positive contributions. However, the significance is limited by the absence of error bars or statistical tests, the lack of code and data release, the close reliance of the EGC module on reference [22], and the paper's own PEMS-BAY results contradicting the claimed superior performance. The contribution is therefore incremental and requires substantially stronger empirical evidence.

major comments (5)
  1. [Section 3.3, Table 2] The abstract and Section 3.3 claim superior performance of TAEGCN on both datasets, but the PEMS-BAY rows of Table 2 contradict this claim: at 30 minutes TAEGCN ties Graph-WaveNet on MAE (1.63) and is worse on MAPE (3.73% versus 3.67%), and at 60 minutes TAEGCN is worse on MAE (1.98 versus 1.95) while tying on MAPE (4.63%). The statement in Section 3.3 that TAEGCN exhibits performance gains over Graph-WaveNet across both datasets in the 15- to 30-minute horizon is therefore false, and the central empirical claim rests solely on METR-LA. The results need to be reframed and the PEMS-BAY outcomes discussed in full.
  2. [Sections 3.3 and 3.4, Tables 2 and 3] No standard deviations, confidence intervals, or statistical significance tests are reported for any of the forecast errors, and no code or data are released. The baseline numbers in Table 2 appear to be reproduced from prior publications (for instance, the Graph-WaveNet values match the original paper) rather than generated by retraining the baselines under the same protocol as TAEGCN. Given the small differences on PEMS-BAY, it is impossible to determine whether the reported improvements are meaningful or stem from noise or implementation choices, so the empirical contribution is not verifiable as presented.
  3. [Section 2.3, Equations (3)-(6)] The EGC module is largely adopted from reference [22]: the GRU equations are explicitly said to be the same as in [22], and the initialization of node embeddings followed by an MLP to produce the adjacency matrix is also the scheme of [22]. The paper does not identify any new component in EGC beyond its integration with TMSA, and the 'random stage' mentioned in Section 2.3 is never defined or represented in the equations. The claimed novelty of an 'evolvable graph structure learning method' is therefore not established, and the authors must clarify the difference from [22].
  4. [Section 2.2 and Figure 2] The TMSA mask is described both as a local window with sizes 1, 3, 6, and 12 and as enforcing temporal causality, but these are not equivalent: a local window centered on a time step can include future values. The paper does not state whether the mask explicitly zeros out future positions, nor does it explain how the output length stays equal to the input length for all window sizes and layer depths. Since causality is fundamental to the forecasting setup, this ambiguity needs to be resolved.
  5. [Section 3.1] The hyperparameter specification is incomplete: the text reports the learning rate, L2 weight, batch size, and epoch count, but not the number of attention heads, hidden dimensions, number of spatio-temporal layers, or the static embedding dimension Cs. The statement that these were determined by grid search is not supported by the grid ranges or the selection criterion, which prevents other researchers from reproducing the reported performance.
minor comments (5)
  1. [Section 2.2, reference [21]] The multi-head self-attention mechanism is attributed to reference [21], but that reference (Elinas et al.) is about variational inference for graph convolutional networks; the attention mechanism should be attributed to Vaswani et al. [23], which is already in the reference list.
  2. [Figure 2 caption] The caption 'each part adopt 1,3,6,12 windows respectively' is grammatically unclear; please specify whether 1, 3, 6, 12 are mask window sizes, numbers of heads, or something else.
  3. [Section 3.4] The ablation replaces TMSA with 'a conventional TCN' and EGC with 'a standard GCN', but the details of these replacement architectures are not provided, making it hard to interpret the source of the performance degradation.
  4. [Section 3.6] The heatmap discussion refers to discrete time indices ('time 3', 'times 5 and 6', 'time 7'), while Figure 7 appears to show continuous series; please clarify how these time points are defined.
  5. [Table 2 caption] The caption contains a typo: 'compasion' should be 'comparison'.

Circularity Check

1 steps flagged · score 2.0 of 10

One non-load-bearing renaming of the cited EGC module; the forecast evaluation itself is not circular.

  1. renaming known result [Section 1 Contributions and Section 2.3, Eqs. (3)-(6)]
    "The paper introduces a novel evolvable graph structure learning method, wherein the graph structure is dynamically updated at each training iteration based on different time periods associated with each node. ... The definition of GRU is same as[22], a module for evolving graph representation"

    The contribution statement credits TAEGCN with 'a novel evolvable graph structure learning method,' but the method's core is explicitly imported: 'The definition of GRU is same as[22]', and Equations (3)-(6) reproduce the GRU-based evolutionary adjacency construction of reference [22] (reset/update gates, static-node MLP initialization, MLP edge scorer, and mask). The claimed novelty is therefore the same module under a new name, with no first-principles derivation offered. This is a genuine renaming, but it is not load-bearing for the forecast numbers: the module is a component trained on held-in data and evaluated on a chronological test split, so the empirical comparison does not reduce to the novelty claim.

full rationale

The paper contains no self-citations, no parameter fitted to the test set and then called a prediction, and no uniqueness theorem imported from the authors' prior work. The load-bearing empirical claim is a standard supervised-learning comparison on METR-LA and PEMS-BAY with a chronological 70/10/20 split, so the forecast results are not equivalent to any training input by construction. The only circularity-adjacent issue is in the contributions: the EGC module is advertised as 'novel' while Section 2.3 states its GRU definition is 'same as[22]' and Eqs. (3)-(6) reproduce reference [22]'s evolutionary graph learner. That is a renamed known result, but it is a component imported into the architecture, not a prediction derived from the paper's own output, so it does not force the empirical results. Separately, the prose claim of gains over Graph-WaveNet 'across both datasets in the 15-minute to 30-minute horizon' is not supported by Table 2 at PEMS-BAY 30 min (MAE tie, MAPE worse) and 60 min (MAE worse); this inconsistency concerns reporting accuracy, not derivation circularity. Overall score reflects one minor attribution/renaming step; the central forecasting chain is not circular.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim depends on a set of hand-tuned hyperparameters and architectural choices that are only partially reported. The most consequential domain assumption is that the full spatial dependency structure is observable from the traffic series themselves, without external covariates. The model introduces no new physical or conceptual entities beyond learned adjacency matrices and node representations.

free parameters (6)
  • learning rate = 1e-4
    Set by grid search on the validation split per Section 3.1; affects convergence and final accuracy.
  • mask window sizes per layer = 1, 3, 6, 12
    Chosen by hand to widen the receptive field across layers; no sensitivity analysis is reported.
  • L2 regularization weight = 1e-4
    Selected during tuning per Section 3.1; not ablated or justified independently.
  • batch size and epoch count = 8, 40
    Determined by grid search per Section 3.1; no convergence curves are shown.
  • hidden dimensions and number of spatio-temporal layers = not specified
    The paper omits these, yet they materially affect model capacity and are free choices.
  • static node embedding dimension Cs = not specified
    Learned representation dimension for graph initialization in Eq. (5); not reported or ablated.
assumptions (6)
  • standard math GCN is a valid first-order approximation of ChebNet on the learned adjacency matrix.
    Invoked in Section 2.1 and 2.3 without proof; standard in the graph neural network literature.
  • domain assumption Traffic sensor readings aggregated into 5-minute windows preserve the underlying dynamics at 15, 30, and 60-minute horizons.
    Standard preprocessing for METR-LA and PEMS-BAY inherited from reference [8].
  • ad hoc to paper The graph structure is constant within a time interval and evolves only between intervals.
    Explicitly assumed in Section 2.3 to justify the GRU-based evolution; if violated, the EGC adjacency is mis-specified.
  • domain assumption External factors such as weather, incidents, and holidays need not be modeled because the multivariate time series itself is information-rich.
    Stated in Section 2.3 as the reason for constructing graphs only from historical series; this is load-bearing for the claim that the learned graph captures true dependencies.
  • ad hoc to paper The mask with increasing window sizes preserves temporal causality and produces output length equal to input length.
    Described as a design goal in Section 2.2, but the exact masking procedure on attention weights is not formally specified.
  • standard math The GRU gating equations from reference [22] are assumed stable and effective for evolving node representations.
    The paper states 'The definition of GRU is same as [22]' and relies on it without further justification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Temporal Attention Evolutional Graph Convolutional Network for Multivariate Time Series Forecasting." pith.science (2026). https://pith.science/paper/QQVKTIRS

@misc{pith2026250500302,
  author       = {Pith},
  title        = {Pith review of: Temporal Attention Evolutional Graph Convolutional Network for Multivariate Time Series Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QQVKTIRS}},
  note         = {Machine review of arXiv:2505.00302}
}
read the original abstract

Multivariate time series forecasting enables the prediction of future states by leveraging historical data, thereby facilitating decision-making processes. Each data node in a multivariate time series encompasses a sequence of multiple dimensions. These nodes exhibit interdependent relationships, forming a graph structure. While existing prediction methods often assume a fixed graph structure, many real-world scenarios involve dynamic graph structures. Moreover, interactions among time series observed at different time scales vary significantly. To enhance prediction accuracy by capturing precise temporal and spatial features, this paper introduces the Temporal Attention Evolutional Graph Convolutional Network (TAEGCN). This novel method not only integrates causal temporal convolution and a multi-head self-attention mechanism to learn temporal features of nodes, but also construct the dynamic graph structure based on these temporal features to keep the consistency of the changing in spatial feature with temporal series. TAEGCN adeptly captures temporal causal relationships and hidden spatial dependencies within the data. Furthermore, TAEGCN incorporates a unified neural network that seamlessly integrates these components to generate final predictions. Experimental results conducted on two public transportation network datasets, METR-LA and PEMS-BAY, demonstrate the superior performance of the proposed model.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 18 canonical work pages

  1. [22]

    In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp

    Ye, J., Liu, Z., Du, B., Sun, L., Li, W., Fu, Y., Xiong, H.: Learning the evolutionary and multi-scale graph structure for multivari- ate time series forecasting. In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 2296–2306 (2022)

    work page 2022

  2. [1]

    Natural computing 18, 747–756 (2019)

    Guo, F., Ren, L., Jin, Y., Ding, Y.: A dynamic SVR–ARMA model with improved fruit fly algorithm for the nonlinear fiber stretching process. Natural computing 18, 747–756 (2019)

    work page 2019

  3. [2]

    Statistics and Decision Mak- ing 11(11), 23–25 (2009)

    Liu, Q., Liu, J.: Research on arima-based multivariate time series neural network fore- casting model. Statistics and Decision Mak- ing 11(11), 23–25 (2009)

    work page 2009

  4. [3]

    In: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pp

    Lee, J.-G., Roh, Y., Song, H., Whang, S.E.: Machine learning robustness, fairness, and their convergence. In: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pp. 4046–4047 (2021)

    work page 2021

  5. [4]

    Neural Computation 9(8), 1735–1780 (1997)

    Hochreiter, S., Schmidhuber, J.J.N.C.: Long short-term memory. Neural Computation 9(8), 1735–1780 (1997)

    work page 1997

  6. [5]

    arXiv preprint arXiv:1612.01022 (2016)

    Wu, Y., Tan, H.: Short-term traffic flow fore- casting with spatial-temporal correlation in a hybrid deep learning framework. arXiv preprint arXiv:1612.01022 (2016)

    arXiv 2016

  7. [6]

    In: The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, pp

    Lai, G., Chang, W.-C., Yang, Y., Liu, H.: Modeling long-and short-term temporal pat- terns with deep neural networks. In: The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, pp. 95–104 (2018)

    work page 2018

  8. [7]

    Machine Learning 108, 1421–1441 (2019)

    Shih, S.-Y., Sun, F.-K., Lee, H.-y.: Tempo- ral pattern attention for multivariate time 11 series forecasting. Machine Learning 108, 1421–1441 (2019)

    work page 2019

Show all 28 references
  1. [8]

    arXiv preprint arXiv:2101.06861 (2021)

    Shang, C., Chen, J., Bi, J.: Discrete graph structure learning for forecasting multiple time series. arXiv preprint arXiv:2101.06861 (2021)

    2021 arXiv

  2. [9]

    arXiv preprint arXiv:1707.01926 (2017)

    Li, Y., Yu, R., Shahabi, C., Liu, Y.: Dif- fusion convolutional recurrent neural net- work: Data-driven traffic forecasting. arXiv preprint arXiv:1707.01926 (2017)

    2017 arXiv

  3. [10]

    arXiv preprint arXiv:1906.00121 (2019)

    Wu, Z., Pan, S., Long, G., Jiang, J., Zhang, C.: Graph wavenet for deep spatial- temporal graph modeling. arXiv preprint arXiv:1906.00121 (2019)

    2019 arXiv

  4. [11]

    Advances in neural information processing systems 29 (2016)

    Defferrard, M., Bresson, X., Vandergheynst, P.: Convolutional neural networks on graphs with fast localized spectral filtering. Advances in neural information processing systems 29 (2016)

    2016

  5. [12]

    In: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp

    Wu, Z., Pan, S., Long, G., Jiang, J., Chang, X., Zhang, C.: Connecting the dots: Mul- tivariate time series forecasting with graph neural networks. In: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 753–763 (2020)

    2020

  6. [13]

    Con- trol and Decision 37(3), 645–653 (2022)

    Gu, Z., Chen, C., Zheng, J., et al.: Traffic flow prediction based on spatio-temporal graph convolutional recurrent neural network. Con- trol and Decision 37(3), 645–653 (2022)

    2022

  7. [14]

    Advances in neural information processing systems 33, 17804–17815 (2020)

    Bai, L., Yao, L., Li, C., Wang, X., Wang, C.: Adaptive graph convolutional recurrent network for traffic forecasting. Advances in neural information processing systems 33, 17804–17815 (2020)

    2020

  8. [15]

    In: Proceedings of the AAAI Con- ference on Artificial Intelligence, vol

    Zheng, C., Fan, X., Wang, C., Qi, J.: Gman: A graph multi-attention network for traffic prediction. In: Proceedings of the AAAI Con- ference on Artificial Intelligence, vol. 34, pp. 1234–1241 (2020)

    2020

  9. [16]

    arXiv preprint arXiv:2012.03363 (2020)

    Pan, C., Chen, S., Ortega, A.: Spatio- temporal graph scattering transform. arXiv preprint arXiv:2012.03363 (2020)

    2020 arXiv

  10. [17]

    In: Proceedings of the AAAI Conference on Artificial Intelli- gence, vol

    Liang, Y., Xia, Y., Ke, S., Wang, Y., Wen, Q., Zhang, J., Zheng, Y., Zimmermann, R.: Airformer: Predicting nationwide air quality in china with transformers. In: Proceedings of the AAAI Conference on Artificial Intelli- gence, vol. 37, pp. 14329–14337 (2023)

    2023

  11. [18]

    IEEE Transac- tions on Intelligent Transportation Systems 23(2), 1009–1018 (2020)

    Guo, K., Hu, Y., Qian, Z., Sun, Y., Gao, J., Yin, B.: Dynamic graph convolution network for traffic forecasting based on latent network of laplace matrix estimation. IEEE Transac- tions on Intelligent Transportation Systems 23(2), 1009–1018 (2020)

    2020

  12. [19]

    In: ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp

    Roddenberry, T.M., Navarro, M., Segarra, S.: Network topology inference with graphon spectral penalties. In: ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 5390–5394 (2021). IEEE

    2021

  13. [20]

    arXiv preprint arXiv:2103.03036 14, 1–1 (2021)

    Zhu, Y., Xu, W., Zhang, J., Liu, Q., Wu, S., Wang, L.: Deep graph structure learning for robust representations: A survey. arXiv preprint arXiv:2103.03036 14, 1–1 (2021)

    2021 arXiv

  14. [21]

    Advances in neural information processing systems 33, 18648–18660 (2020)

    Elinas, P., Bonilla, E.V., Tiao, L.: Variational inference for graph convolutional networks in the absence of graph data and adversar- ial settings. Advances in neural information processing systems 33, 18648–18660 (2020)

    2020

  15. [23]

    Advances in neural information processing systems 30 (2017)

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

    2017

  16. [24]

    IEEE Transactions on Intelli- gent Transportation Systems 14(2), 871–882 (2013) 12

    Lippi, M., Bertini, M., Frasconi, P.: Short- term traffic flow forecasting: An experimental comparison of time-series analysis and super- vised learning. IEEE Transactions on Intelli- gent Transportation Systems 14(2), 871–882 (2013) 12

    2013

  17. [25]

    Advances in neural information processing systems 27 (2014)

    Sutskever, I., Vinyals, O., Le, Q.V.: Sequence to sequence learning with neural networks. Advances in neural information processing systems 27 (2014)

    2014

  18. [26]

    arXiv preprint arXiv:1609.03499 12 (2016)

    Van Den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalch- brenner, N., Senior, A., Kavukcuoglu, K., et al.: Wavenet: A generative model for raw audio. arXiv preprint arXiv:1609.03499 12 (2016)

    2016 arXiv

  19. [27]

    arXiv preprint arXiv:1803.07294 (2018)

    Zhang, J., Shi, X., Xie, J., Ma, H., King, I., Yeung, D.-Y.: Gaan: Gated attention net- works for learning on large and spatiotempo- ral graphs. arXiv preprint arXiv:1803.07294 (2018)

    2018 arXiv

  20. [28]

    arXiv preprint arXiv:1709.04875 (2017) 13

    Yu, B., Yin, H., Zhu, Z.: Spatio-temporal graph convolutional networks: A deep learn- ing framework for traffic forecasting. arXiv preprint arXiv:1709.04875 (2017) 13

    2017 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.