Pith. sign in

REVIEW 4 major objections 6 minor 50 references

STDCformer: A Transformer-Based Model with a Spatial-Temporal Causal De-Confounding Strategy for Crowd Flow Prediction

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper argues that crowd flow prediction should be modeled as the causal effect of the past on the future, and builds a transformer that de-confounds spatial and temporal attributes to estimate that effect.

desk verdict A solid gated-fusion transformer for crowd flow wrapped in an unsupported causal backdoor story; refereeing should force the causal claims to be either derived or removed. read the letter →

arxiv 2412.02942 v1 pith:PFOTNHUX submitted 2024-12-04 cs.AI

Silu He , Peng Shen , Pingzhen Xu , Qinyao Luo , Haifeng Li This is my paper · ORCID
classification cs.AI
keywords crowdflowpredictionspatial-temporalcausalde-confoundingbackdooradjustmentcross-timeattentiontransformerout-of-distributiongeneralizationgraphnetworks
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

The paper proposes to view crowd flow prediction not as a direct learned map from past to future observations, but as three linked processes — encoding, cross-time mapping, and decoding — and argues that the key is learning a representation space where the past-to-future mapping is causal rather than merely correlational. It claims that each spatial-temporal token’s spatial attributes (where) and temporal attributes (when) act as confounders that bias ordinary training, and introduces a spatial-temporal backdoor adjustment that reweights spatial and temporal representations during fusion. The resulting model, STDCformer, is reported to beat STGNN and ST-transformer baselines on two New York City taxi-based crowd flow datasets and to generalize zero-shot from Manhattan to Brooklyn. If correct, this suggests that explicitly modeling confounders in urban flow data improves both prediction accuracy and out-of-distribution robustness.

What carries the argument

The central object is the spatial-temporal backdoor adjustment identity, $P(Y|\operatorname{do}(X)) = P(Y|X, C=C_S)\,P(C=C_S) + P(Y|X, C=C_T)\,P(C=C_T)$, where the confounders are the spatial attributes (‘where’) and temporal attributes (‘when’) of each spatial-temporal token. The identity is made operative in the Spatial-Temporal De-Confounded Fusion module, which computes scalar weights $P(C_S)=\operatorname{Sigmoid}(C_S+C_T)$, $P(C_T)=1-P(C_S)$ from learned confounder embeddings and uses them to gate the sum of the spatial-attention and temporal-attention representations. This gated fusion is what simulates intervention. The Spatial-Temporal Embedding (STE) and Cross-Time Attention are the complementary machinery that let the model query the past-to-future relationship using those same confounder representations.

What would settle it

On a spatial-temporal dataset where the true causal effect is known (for example, a synthetic process with an intervention that sets past flow), compare STDCformer’s estimated $P(Y|\operatorname{do}(X))$ with the ground-truth intervention distribution; if the estimate is no closer than an associative baseline, the de-confounding claim collapses. A cheaper test would shuffle the auxiliary confounder features while keeping the flow data identical: if the weights are causal, prediction error on shifted test distributions should grow, whereas a pure attention gate would be insensitive.

Watch

Extended reading notes

Core claim

The paper’s central claim is that the right objective for spatial-temporal prediction is the de-confounded causal effect $P(Y|\operatorname{do}(X))$ rather than the associative fit $P(Y|X)$. It argues that each spatial-temporal token carries background attributes — spatial attributes $S$ such as location, functionality, and safety, and temporal attributes $T$ such as time of day, holiday, and weather — that influence both historical and future observations, making them confounders of the past-to-future mapping. To remove their bias, the paper proposes a spatial-temporal backdoor adjustment that partitions the confounders into $C=\{C_S, C_T\}$ and implements the adjustment in a fusion module as $H = P(C_S)\cdot STR'_S + P(C_T)\cdot STR'_T$ with $P(C_S)=\operatorname{Sigmoid}(C_S+C_T)$ and $P(C_T)=1-P(C_S)$. This defines the Spatial-Temporal De-Confounded (STDC) representation space. Cross-time attention between future and past spatial-temporal embeddings then answers how to map from past to future in that space. Empirically the paper reports state-of-the-art MAE and RMSE on two New York City taxi-derived crowd flow datasets and stronger zero-shot transfer from Manhattan to Brooklyn than transformer and STGNN baselines.

Load-bearing premise

The causal story assumes that the chosen spatial and temporal auxiliary features are the only confounders of the past-to-future relationship, so the learned weights estimate the true intervention effect rather than merely acting as data-dependent attention gates.

Editorial extensions

If this is right

  • If the causal graph is correct, training with the backdoor adjustment should reduce spurious correlations between past and future flow, which is the stated reason for the reported gains in IID accuracy.
  • The learned confounder weights carry interpretable physical meaning: the model raises the spatial weight for low-traffic, park-like zones and lowers it for busy mixed-function districts, so the weights can be inspected to explain predictions.
  • Cross-time attention exposes how each region maps past to future: short-horizon matches attend to the most recent past, while longer horizons shift attention to historically similar segments, giving a per-region explanation of the mapping.
  • Zero-shot transfer from Manhattan to Brooklyn is reported to work better than transformer baselines, suggesting that the de-confounded representation retains usable structure across spatial distributions.
  • The encode–map–decode decomposition frames spatial-temporal prediction as representation learning plus causal mapping, a skeleton the paper claims can be instantiated beyond crowd flow.

Reading between the lines

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

  • The soft Sigmoid gating is not a stratified backdoor estimator: it collapses the confounder distribution into a single scalar trade-off between spatial and temporal weights, so the causal interpretation rests on how faithfully the auxiliary features represent the true confounders.
  • A straightforward next test is to compare the de-confounding weights against inverse-propensity reweighting computed from known confounders on a synthetic spatial-temporal dataset, where the true intervention effect is computable.
  • Because the paper’s zero-shot OOD test moves between two boroughs of the same city with the same taxi infrastructure, a stronger test would transfer across cities or sensor modalities; if the advantage disappears there, the de-confounding benefit may be task-specific.
  • The same backdoor-adjustment skeleton might transfer to traffic, epidemic, or air-quality forecasting by swapping the auxiliary confounder features, which is a testable extension the paper does not run.
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. STDCformer proposes a transformer architecture for crowd flow prediction in which the past-to-future mapping is decomposed into encoding, cross-time mapping, and decoding. The paper's headline contribution is a 'Spatial-Temporal Backdoor Adjustment' strategy that is claimed to estimate the causal effect P(Future | do(Past)) by categorizing confounders into temporal and spatial types, and to learn a de-confounded representation space. The architecture adds spatial and temporal confounder embeddings, a spatial-temporal de-confounded fusion gating, and a cross-time attention module. The authors report state-of-the-art or competitive IID results on two new NYC taxi datasets and better zero-shot OOD transfer than two transformer baselines, along with ablations, hyperparameter analyses, and qualitative interpretations of learned weights.

Significance. If the causal claims were valid, the paper would offer a principled way to bring backdoor adjustment into spatiotemporal forecasting and would contribute two new publicly described benchmarks with auxiliary covariates. The empirical scaffolding is substantial: two real-world datasets, a Transformer architecture, ablations, and a zero-shot transfer experiment are all presented. However, the central causal derivation is not sound. Equation (2) is not the standard backdoor adjustment, and Equation (6) implements what is effectively a learned gated fusion rather than a causal estimator. No identification conditions are stated, and the interpretation of the learned weights in Section 4.6.3 is circular. The empirical gains are modest and not consistent across all metrics, so the paper's contribution rests on the causal framing, which is not supported.

major comments (4)
  1. [Section 3.1.1, Eq. (2)] The proposed 'spatial-temporal backdoor adjustment' is not the standard backdoor adjustment formula. For a confounder set C, the backdoor adjustment is P(Y|do(X)) = Σ_c P(Y|X, C=c)P(C=c) when C is discrete. Equation (2) instead writes P(Y|do(X)) = P(Y|X, C=C_S)P(C=C_S) + P(Y|X, C=C_T)P(C=C_T). Even if S and T are the only confounders, the adjustment should condition on the joint value (S=s, T=t) and sum or integrate over the joint strata; adding two marginal terms with the notation 'C=C_S' is not a special case of the formula. The paper provides no identification theorem showing that the right-hand side equals the interventional distribution for the graph in Figure 6. The central claim of the abstract, that the model 'estimates the de-confounding causal effect of historical data on future data,' is therefore unsupported.
  2. [Section 3.3.2, Eq. (6)] The implemented fusion in Eq. (6) is a sigmoid-gated convex combination of spatial and temporal attention outputs: H = P(C_S)·STR'_S + P(C_T)·STR'_T, with P(C_T)=1−P(C_S). No argument is given that the sigmoid of a summed embedding equals P(C=C_S), nor that the attention outputs STR'_S and STR'_T equal the conditional outcome terms P(Y|X,C=C_S) and P(Y|X,C=C_T). The text asserts this correspondence without derivation. In practice this is standard gated fusion of two representation streams conditioned on auxiliary features; calling it a causal effect estimator is not justified.
  3. [Section 4.6.3] The interpretation of learned P(C_S) weights as evidence of successful de-confounding is circular. The weights are deterministic functions of the learned confounder embeddings through the Sigmoid in Eq. (6) and are trained end-to-end to minimize MAE. The manuscript then uses these same weights to conclude that the model compensates for 'the confounding bias in the original data' and that 'the de-confounding process adjusts this bias.' No external ground truth, no causal validation experiment, and no comparison of the learned weights to known confounding distributions are provided. The zero-shot OOD result is the only independent evidence, but it does not by itself establish that the adjustment identifies a causal effect rather than improving shift robustness through feature conditioning.
  4. [Section 3.1.1 and Figure 6] The assumed causal graph is not defended. The paper states in Section 3.1 that confounders are often unobservable, yet the method rests on the assumption that the measured auxiliary features (Table 2) exactly represent the complete set of confounders and that no unobserved confounders exist. No argument is given that the backdoor criterion is satisfied for the graph in Figure 6, nor that the proxy features (e.g., POI counts, crime statistics) adequately capture the proposed constructs. As written, the model's causal interpretation collapses if unobserved confounders exist or if the proxies are incomplete, and the manuscript does not address this threat.
minor comments (6)
  1. [Section 3.3.1, Eq. (3)] In Eq. (3), the temporal confounder is written as C_T = Conv_{s→d}(T), but T ∈ R^{1×t}; the subscript appears to be a typo and should be Conv_{t→d}(T).
  2. [Section 3.4, Eq. (7)] The dimension of the attention map A_MAP is stated as R^{n×T_p×T_f}; with Q from the future STE and K from the past STE, the attention matrix should have shape R^{T_f×n×T_p} (or a flattened variant). Please correct the dimension statement.
  3. [Section 4.4.2, Table 4] Table 4 reports MAPE, but the metric is not defined in the text and the baseline comparisons in Table 3 use only MAE and RMSE; please define MAPE and state how it is computed for aggregate flow.
  4. [References] Several in-text citations appear to be mismatched with the reference list. For example, the citation [23] in Section 2.2.1 for self-supervised deconfounding does not match the listed 'STG4Traffic' reference, and [30] is cited as node2vec in the text but appears as MTGNN in the list. A careful renumbering and cross-check of all citations is needed.
  5. [Figures and text] The manuscript contains frequent typos and formatting artifacts, including 'di fferent', 'wights', 'f uture', and inconsistent spacing in equations and citations. A thorough proofread is recommended before resubmission.
  6. [Section 4.4.1, Table 3] The claimed state-of-the-art result is not uniform: PDFormer achieves a lower MAE than STDCformer for inflow on MHT (14.64 vs. 14.81). The text acknowledges this candidly, but the contribution statement in Section 1 should be made consistent with the full set of results.

Circularity Check

2 steps flagged · score 6.0 of 10

The 'spatial-temporal backdoor adjustment' is implemented as a learned gated fusion of spatial and temporal attention outputs, and the fitted gating weights are then interpreted as evidence of successful de-confounding.

  1. renaming known result [Section 3.3.2, Eq. (6) (with Eq. (2) in Section 3.1.1)]
    "For each S T Ti j, to obtain P(C = CS ) and P(C = CT ), CS i and CT j are summed and passed through a Sigmoid function, mapping the weight of the two confounding factors to the range from 0 to 1. Finally, after fusion, the final spatial-temporal representation H∈ RTP×n×d is obtained for all STTs in the input data, as shown by Eq. 6: P(CS ) = Sigmoid(CS + CT ); P(CT ) = 1− P(C = CS ); H = P(CS )· S T R′_S + P(CT )· S T R′_T"

    The backdoor estimand announced in Eq. (2) is implemented as this exact weighted sum: the two terms P(Y|X,C=CS) and P(Y|X,C=CT) are declared to be the spatial and temporal attention outputs, and P(C=CS), P(C=CT) are declared to be the sigmoid of learned embeddings. Thus the 'de-confounded causal effect' is, by construction, the same gated fusion of spatial and temporal representations used by ordinary adaptive-fusion ST transformers, and the causal label adds no independent constraint. No identification theorem shows Eq. (2) equals the standard backdoor formula for the graph in Fig. 6, which would require summing over joint strata P(Y|X,C=(s,t))P(C=(s,t)). The empirical gain of the architecture does not certify that the learned H is an interventional quantity.

  2. fitted input called prediction [Section 4.6.3, 'Physical Meaning of Confounders']
    "To explore the physical meaning of the temporal and spatial confounders learned by the model, we analyzed the weights output from the spatial-temporal de-confounding layers. Here, P (CS ) represents the weight of the spatial confounder in the STTs."

    P(CS) is a trainable parameter produced by a sigmoid of confounder embeddings and optimized solely by the prediction loss (Eq. 8). The section then treats the fitted values of P(CS) as evidence that the model removes confounding bias by reweighting and as a physical measurement of confounder distribution. This is a fitted parameter being read back as validation of the causal claim: no causal ground truth, placebo intervention, or identification test is used. Observations such as P(CS)≥0.5 or higher weights for parks describe the learned gating weights themselves, so they cannot independently confirm de-confounding.

full rationale

The paper's central causal claim reduces to a reweighted sum of the model's own spatial and temporal attention outputs. In Eq. (2) the paper writes P(Y|do(X)) as two marginal terms involving CS and CT, but the standard backdoor adjustment for the graph in Fig. 6 would need a sum over joint strata of C=(S,T); the paper gives no identification argument. In Eq. (6) those terms are then replaced by sigmoid-weighted attention outputs, so the 'de-confounded representation' H is exactly the gated fusion that the model is trained to produce. The fitted gating weights are later interpreted in Section 4.6.3 as physical evidence of successful de-confounding, which is a fitted input being used as its own validation. The IID and zero-shot OOD experiments are genuine external benchmarks and provide independent evidence of predictive performance and generalization, but they do not test the causal identification or the claim that the learned weights equal confounder distributions. There is also a minor self-citation ([50]) used only as supporting consistency, not as load-bearing evidence. Because the headline causal mechanism is, by construction, the learned gated fusion itself, the paper receives a partial circularity score of 6.

Assumptions & free parameters 6 free parameters · 4 assumptions · 3 invented entities

The paper's causal claims rest on an asserted causal graph, a two-category confounder partition, and hand-picked auxiliary features. The backdoor adjustment in Eq. 2 is not a standard backdoor adjustment, and Eq. 6 reduces it to a learned sigmoid gate. The remaining free parameters are standard architecture hyperparameters chosen by validation.

free parameters (6)
  • De-confounding fusion weight P(CS) / P(CT) = learned per STT via sigmoid(CS+CT)
    Eq. 6 defines the entire "backdoor adjustment" as a convex combination of spatial and temporal attention outputs with a learned sigmoid weight; no causal identification is performed.
  • Number of confounder categories k = 2
    Eq. 2 replaces the general sum over k confounder strata with exactly two hand-chosen categories, temporal and spatial, based on "when" and "where".
  • Laplacian eigenvalue dimension = 8
    Section 4.3 sets this to 8; ablation in Section 4.5.2 selects it by validation.
  • Number of STDC attention blocks = 5 for MHT, 6 for BKL
    Section 4.3 and Section 4.5.2; chosen per dataset by validation.
  • Hidden representation dimension = 64
    Section 4.3 and Section 4.5.2; selected from [32,64,128].
  • Number of attention heads = 8 for MHT, 4 for BKL
    Section 4.5.2; selected from [1,2,4,8].
assumptions (4)
  • domain assumption S and T are the only confounders of the past-to-future effect; no unobserved confounding exists.
    Section 3.1.1 introduces the causal graph in Fig. 6 and applies backdoor adjustment without testing unconfoundedness.
  • domain assumption The backdoor criterion holds for X=past representation, Y=future representation with adjustment set C={S,T}.
    Section 3.1.1 states Eq. 2 as if it were a backdoor adjustment, but the graphical conditions for the backdoor criterion are never shown.
  • ad hoc to paper The two-category partition into temporal and spatial confounders is complete and sufficient.
    Section 3.1.1 argues "when" and "where" are the minimal descriptive set of human movement, but provides no evidence that other factors are not confounders.
  • domain assumption Auxiliary features (POI, crime, weather, housing) faithfully represent the true confounders C_S and C_T after learned encoders.
    Section 3.3.1 constructs C_S and C_T from hand-picked features; there is no validation that these embeddings align with the true confounders.
invented entities (3)
  • Spatial confounder C_S
    purpose: Latent variable representing region attributes; used to compute fusion weights and STE.
    Learned from hand-picked spatial features; no falsifiable prediction outside the model. Section 4.6.3 interprets its weights post hoc.
  • Temporal confounder C_T
    purpose: Latent variable representing time-window attributes; used for fusion weights and STE.
    Learned from temporal features; no external handle or testable implication.
  • Spatial-Temporal Token (STT)
    purpose: Conceptual unit to justify confounder analysis; not an empirical discovery.
    Introduced in Section 3.1.1 as an analogy to text tokens.

how reviews work

0 comments
Cite this review

Pith. "Pith review of STDCformer: A Transformer-Based Model with a Spatial-Temporal Causal De-Confounding Strategy for Crowd Flow Prediction." pith.science (2026). https://pith.science/paper/PFOTNHUX

@misc{pith2026241202942,
  author       = {Pith},
  title        = {Pith review of: STDCformer: A Transformer-Based Model with a Spatial-Temporal Causal De-Confounding Strategy for Crowd Flow Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PFOTNHUX}},
  note         = {Machine review of arXiv:2412.02942}
}
abstract

Existing works typically treat spatial-temporal prediction as the task of learning a function $F$ to transform historical observations to future observations. We further decompose this cross-time transformation into three processes: (1) Encoding ($E$): learning the intrinsic representation of observations, (2) Cross-Time Mapping ($M$): transforming past representations into future representations, and (3) Decoding ($D$): reconstructing future observations from the future representations. From this perspective, spatial-temporal prediction can be viewed as learning $F = E \cdot M \cdot D$, which includes learning the space transformations $\left\{{E},{D}\right\}$ between the observation space and the hidden representation space, as well as the spatial-temporal mapping $M$ from future states to past states within the representation space. This leads to two key questions: \textbf{Q1: What kind of representation space allows for mapping the past to the future? Q2: How to achieve map the past to the future within the representation space?} To address Q1, we propose a Spatial-Temporal Backdoor Adjustment strategy, which learns a Spatial-Temporal De-Confounded (STDC) representation space and estimates the de-confounding causal effect of historical data on future data. This causal relationship we captured serves as the foundation for subsequent spatial-temporal mapping. To address Q2, we design a Spatial-Temporal Embedding (STE) that fuses the information of temporal and spatial confounders, capturing the intrinsic spatial-temporal characteristics of the representations. Additionally, we introduce a Cross-Time Attention mechanism, which queries the attention between the future and the past to guide spatial-temporal mapping.

Figures

Figures reproduced from arXiv: 2412.02942 by the authors.

Figure 1
Figure 1. The characteristics of Spatial-Temporal Tokens in crowd flow prediction task. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Difference between historical and future flow under different time and region sampling. of STTs by fusing time and spatial confounders, which preserve the global positional and structural information of the STTs in both time and space dimensions. To better capture the dynamic transformation relationship between past and future, which is a pair of cross-time entities, we introduce a Cross-Time-Attention (CTA) mechani… view at source ↗
Figure 3
Figure 3. The fusion of STDC strategy and ST Transformer framework. (a) ST Transformer framework. (b) ST Transformer with STDC Strategy. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (18 more)
Figure 4
Figure 4. Figure 4: The framework of STGNNs. Currently, mainstream spatial-temporal graph prediction frameworks can be divided into two categories: Spatial￾temporal Graph Neural Networks (STGNNs) [4, 27, 28] and Spatial-temporal Transformers (ST Transformers) [22, 29]. As shown in [PITH_…
Figure 5
Figure 5. Figure 5: The formulation of crowd flow prediction. (a) Existing perspective. (b) A novel decomposing perspective proposed in this paper. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The causal graph and underlying spatial-temporal prediction and the intervention operation, where [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: The prediction skeleton of STDCformer which fuses the information of temporal confounders and spatial confounders. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: The overall structure of STDCformer. Where the information of spatial confounders and its flow are marked as [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: The architecture of Spatial-temporal De-Confounded attention block. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: The architecture of Cross-Time attention block. [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Study area. Auxiliary data. To learn the confounders’ representations of STTs, this study collected auxiliary data to charac￾terize the background characteristics of the STTs. To comprehensively and accurately introduce auxiliary information 16 [PITH_FULL_IMAGE:figur…
Figure 12
Figure 12. Figure 12: The results of ablation experiments. (a) MHT dataset. (b) BKL dataset. [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: The results of hyperparameter analysis. 4.6. Analytical Experiments and Interpretation To further investigate and explain the reasons behind the model’s strong performance, in addition to the zero￾shot transfer experiment for spatial OOD regions (Section 4.4.2), we co…
Figure 14
Figure 14. Figure 14: The comparison of prediction accuracy of Inflow and Outflow. (a) MHT dataset. (b) BKL dataset. [PITH_FULL_IMAGE:figures/full_fig_p022_14.png]
Figure 15
Figure 15. Figure 15: Models’ prediction accuracy across different time windows and regions in MHT dataset. 4.6.2. Prediction Visualization In general, all models perform well in capturing the upward and downward trends of crowd flow. However, STDCformer prefer to make predicted values wit…
Figure 16
Figure 16. Figure 16: Visualization of prediction result for Zone 48. [PITH_FULL_IMAGE:figures/full_fig_p023_16.png]
Figure 17
Figure 17. Figure 17: Visualization of prediction result for Zone164. [PITH_FULL_IMAGE:figures/full_fig_p023_17.png]
Figure 18
Figure 18. Figure 18: Trend variation of crowd flow and P (CS ) over time. (a) Zone 68. (b) Zone 88. (c) Zone 164 3. Observation 3: For regions in closer clusters, the values of their P (CS ) are also closer. Guess: The weight of CS for a region is significantly associated with its spatial…
Figure 20
Figure 20. Figure 20: The distribution of average value of [PITH_FULL_IMAGE:figures/full_fig_p025_20.png]
Figure 21
Figure 21. Figure 21: The flow variation of Zone164 and corresponding cross-time attention. [PITH_FULL_IMAGE:figures/full_fig_p026_21.png]
Figure 22
Figure 22. Figure 22: The flow variation of Zone 125 and corresponding cross-time attention. [PITH_FULL_IMAGE:figures/full_fig_p026_22.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 40 canonical work pages

  1. [1]

    H. Go, S. Park, A study on deep learning model based on global–local structure for crowd flow prediction, Scientific Reports 14 (1) (2024) 12623. doi:10.1038/s41598-024-63310-6 . URL https://doi.org/10.1038/s41598-024-63310-6

    work page doi:10.1038/s41598-024-63310-6 2024

  2. [2]

    P. Xie, T. Li, J. Liu, S. Du, X. Yang, J. Zhang, Urban flow prediction from spatiotemporal data using machine learning: A survey, Information Fusion 59 (2020) 1–12

    work page 2020

  3. [3]

    J. Wang, J. Jiang, W. Jiang, C. Li, W. X. Zhao, Libcity: An open library for tra ffic prediction (2021). doi:10.1145/3474717.3483923

    arXiv 2021

  4. [4]

    G. Jin, Y . Liang, Y . Fang, Z. Shao, J. Huang, J. Zhang, Y . Zheng, Spatio-temporal graph neural networks for predictive learning in urban computing: A survey, IEEE Transactions on Knowledge and Data Engineering 36 (10) (2024) 5388–5408. doi:10.1109/TKDE.2023. 3333824

    doi:10.1109/tkde.2023 2024

  5. [5]

    H. Liu, C. Zhu, D. Zhang, Q. Li, Attention-based spatial-temporal graph convolutional recurrent networks for tra ffic forecasting, in: Interna- tional Conference on Advanced Data Mining and Applications, Springer, 2023, pp. 630–645

    work page 2023

  6. [6]

    Y . Li, R. Yu, C. Shahabi, Y . Liu, Diffusion convolutional recurrent neural network: Data-driven traffic forecasting (2018)

    work page 2018

  7. [7]

    J. Zhu, X. Han, H. Deng, C. Tao, L. Zhao, P. Wang, T. Lin, H. Li, Kst-gcn: A knowledge-driven spatial-temporal graph convolutional network for traffic forecasting, IEEE Transactions on Intelligent Transportation Systems 23 (9 doi:10.1109/tits.2021.3136287. URL https://doi.org/10.1109/TITS.2021.3136287

    work page arXiv

  8. [8]

    H. Yan, X. Ma, Z. Pu, Learning dynamic and hierarchical tra ffic spatiotemporal features with transformer, IEEE Transactions on Intelligent Transportation Systems 23 (11) (2022) 22386–22399. doi:10.1109/TITS.2021.3102983

    arXiv 2022

Show all 50 references
  1. [9]

    X. Yin, G. Wu, J. Wei, Y . Shen, H. Qi, B. Yin, Multi-stage attention spatial-temporal graph networks for traffic prediction, Neurocomputing 428 (2021) 42–53. doi:https://doi.org/10.1016/j.neucom.2020.11.038. URL https://www.sciencedirect.com/science/article/pii/S0925231220318312

    2021 doi

  2. [10]

    L. Zhao, Y . Song, C. Zhang, Y . Liu, P. Wang, T. Lin, M. Deng, H. Li, T-gcn: A temporal graph convolutional network for traffic prediction, IEEE Transactions on Intelligent Transportation Systems 21 (9) (2019) 3848–3858

    2019

  3. [11]

    Z. Pan, Y . Liang, W. Wang, Y . Yu, Y . Zheng, J. Zhang, Urban tra ffic prediction from spatio-temporal data using deep meta learning, in: Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery and data mining, 2019, pp. 1720–1730

    2019

  4. [12]

    S. Guo, Y . Lin, N. Feng, C. Song, H. Wan, Attention based spatial-temporal graph convolutional networks for tra ffic flow forecasting, in: Proceedings of the AAAI conference on artificial intelligence, V ol. 33, 2019, pp. 922–929

    2019

  5. [13]

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

    2019

  6. [14]

    K. Guo, Y . Hu, Y . Sun, S. Qian, J. Gao, B. Yin, Hierarchical graph convolution network for traffic forecasting, in: Proceedings of the AAAI conference on artificial intelligence, V ol. 35, 2021, pp. 151–159

    2021

  7. [15]

    M. Li, Z. Zhu, Spatial-temporal fusion graph neural networks for tra ffic flow forecasting, Proceedings of the AAAI Conference on Artificial Intelligence 35 (5) (2021) 4189–4196. URL https://ojs.aaai.org/index.php/AAAI/article/view/16542

    2021

  8. [16]

    B. Yu, H. Yin, Z. Zhu, Spatio-temporal graph convolutional networks: a deep learning framework for tra ffic forecasting (2018)

    2018

  9. [17]

    M. Xu, W. Dai, C. Liu, X. Gao, W. Lin, G.-J. Qi, H. Xiong, Spatial-temporal transformer networks for traffic flow forecasting, arXiv preprint arXiv:2001.02908

    2001 arXiv

  10. [18]

    Jiang, C

    J. Jiang, C. Han, W. X. Zhao, J. Wang, Pdformer: Propagation delay-aware dynamic long-range transformer for tra ffic flow prediction, in: AAAI, AAAI Press, 2023

    2023

  11. [19]

    Q. Luo, S. He, X. Han, Y . Wang, H. Li, Lsttn: A long-short term transformer-based spatiotemporal neural network for traffic flow forecasting, Knowledge-Based Systems 293 (2024) 111637. doi:https://doi.org/10.1016/j.knosys.2024.111637. URL https://www.sciencedirect.com/science...

    2024

  12. [20]

    Huang, H

    B. Huang, H. Dou, Y . Luo, J. Li, J. Wang, T. Zhou, Adaptive spatiotemporal transformer graph network for traffic flow forecasting by iot loop detectors, IEEE Internet of Things Journal 10 (2) (2022) 1642–1653. 27 Silu He, Peng Shen, Pingzhen Xu, Qinyao Luo, Haifeng Li / Infor...

    2022

  13. [21]

    J. Ma, R. Guo, C. Chen, A. Zhang, J. Li, Deconfounding with networked observational data in a dynamic environment (2021). doi: 10.1145/3437963.3441818. URL https://doi.org/10.1145/3437963.3441818

    2021

  14. [22]

    Y . Xia, Y . Liang, H. Wen, X. Liu, K. Wang, Z. Zhou, R. Zimmermann, Deciphering spatio-temporal graph forecasting: A causal lens and treatment, Advances in Neural Information Processing Systems 36

  15. [23]

    J. Ji, W. Zhang, J. Wang, Y . He, C. Huang, Self-supervised deconfounding against spatio-temporal shifts: Theory and modeling, arXiv preprint arXiv:2311.12472

    arXiv

  16. [24]

    X. Luo, C. Zhu, D. Zhang, Q. Li, Stg4tra ffic: A survey and benchmark of spatial-temporal graph neural networks for traffic prediction, arXiv preprint arXiv:2307.00495

    arXiv

  17. [25]

    Rahmani, A

    S. Rahmani, A. Baghbani, N. Bouguila, Z. Patterson, Graph neural networks for intelligent transportation systems: A survey, IEEE Transac- tions on Intelligent Transportation Systems 24 (8) (2023) 8846–8885. doi:10.1109/TITS.2023.3257759

    2023

  18. [26]

    B. Gu, J. Zhan, S. Gong, W. Liu, Z. Su, M. Guizani, A spatial-temporal transformer network for city-level cellular tra ffic analysis and prediction, IEEE Transactions on Wireless Communications 22 (12) (2023) 9412–9423. doi:10.1109/TWC.2023.3270441

    2023

  19. [27]

    A. Liu, Y . Zhang, An e fficient spatial-temporal transformer with temporal aggregation and spatial memory for tra ffic forecasting, Expert Systems with Applications 250 (2024) 123884. doi:https://doi.org/10.1016/j.eswa.2024.123884. URL https://www.sciencedirect.com/science/ar...

    2024

  20. [28]

    Zhang, Y

    J. Zhang, Y . Zheng, D. Qi, R. Li, X. Yi, Dnn-based prediction model for spatio-temporal data, in: Proceedings of the 24th ACM SIGSPATIAL international conference on advances in geographic information systems, 2016, pp. 1–4

    2016

  21. [29]

    Z. Wu, S. Pan, G. Long, J. Jiang, X. Chang, C. Zhang, Connecting the dots: Multivariate time series forecasting with graph neural networks, in: Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery and data mining, 2020, pp. 753–763

    2020

  22. [30]

    Grover, J

    A. Grover, J. Leskovec, node2vec: Scalable feature learning for networks, in: Proceedings of the 22nd ACM SIGKDD international confer- ence on Knowledge discovery and data mining, 2016, pp. 855–864

    2016

  23. [31]

    J. W. C. Lint, S. Hoogendoorn, H. Zuvlen, Freeway travel time prediction with state-space neural networks: Modeling state-space dynamics with recurrent neural networks, Transportation Research Record 1811. doi:10.3141/1811-04

    doi

  24. [32]

    Z. Lv, J. Xu, K. Zheng, H. Yin, P. Zhao, X. Zhou, Lc-rnn: A deep learning model for tra ffic speed prediction, in: IJCAI, 2018, p. 27th

    2018

  25. [33]

    R. Fu, Z. Zhang, L. Li, Using lstm and gru neural network methods for tra ffic flow prediction, in: 31st Youth Academic Annual Conference of Chinese Association of Automation (Y AC), IEEE, 2016, pp. 324–328

    2016

  26. [34]

    Prabowo, W

    A. Prabowo, W. Shao, H. Xue, P. Koniusz, F. D. Salim, Because every sensor is unique, so is every pair: Handling dynamicity in tra ffic forecasting, in: Proceedings of the 8th ACM/IEEE Conference on Internet of Things Design and Implementation, 2023, pp. 93–104

    2023

  27. [35]

    B. Pu, J. Liu, Y . Kang, J. Chen, S. Y . Philip, Mvstt: A multiview spatial-temporal transformer network for tra ffic-flow forecasting, IEEE transactions on cybernetics 54 (3) (2022) 1582–1595

    2022

  28. [36]

    Zheng, X

    C. Zheng, X. Fan, C. Wang, J. Qi, Gman: A graph multi-attention network for tra ffic prediction (2020). URL https://ojs.aaai.org/index.php/AAAI/article/view/5477

    2020

  29. [37]

    P. Deng, J. Liu, X. Wang, X. Jia, Y . Zhao, M. Wang, X. Dai, Stctn: a spatio-temporal causal representation learning method based on temporal bias adjustment and spatial causal transition, Chinese Journal of Computers 46 (12) (2023) 2535–2550. URL https://link.cnki.net/urlid/1...

    2023

  30. [38]

    C. Ge, S. Song, G. Huang, Causal intervention for human trajectory prediction with cross attention mechanism, in: Proceedings of the AAAI Conference on Artificial Intelligence, V ol. 37, 2023, pp. 658–666

    2023

  31. [39]

    X. Luo, W. Yin, Z. Li, Spatio-temporal graph neural network with hidden confounders for causal forecast, in: Special Track on AI for Socio-Ecological Welfare at ICCBR2024 (ICCBR AI Track’24), 2024

    2024

  32. [40]

    Y . Zhao, P. Deng, J. Liu, X. Jia, M. Wang, Causal conditional hidden markov model for multimodal traffic prediction, in: Proceedings of the AAAI Conference on Artificial Intelligence, V ol. 37, 2023, pp. 4929–4936

    2023

  33. [41]

    P. Deng, Y . Zhao, J. Liu, X. Jia, M. Wang, Spatio-temporal neural structural causal models for bike flow prediction, in: Proceedings of the AAAI conference on artificial intelligence, V ol. 37, 2023, pp. 4242–4249

    2023

  34. [42]

    J. Liu, H. Lin, X. Wang, L. Wu, S. Garg, M. M. Hassan, Reliable trajectory prediction in scene fusion based on spatio-temporal structure causal model, Information Fusion 107 (2024) 102309. doi:https://doi.org/10.1016/j.inffus.2024.102309. URL https://www.sciencedirect.com/scie...

    2024

  35. [43]

    B. Jing, D. Zhou, K. Ren, C. Yang, Casper: Causality-aware spatiotemporal graph neural networks for spatiotemporal time series imputation, in: the 33rd ACM International Conference on Information and Knowledge Management (CIKM ’24), 2024

    2024

  36. [44]

    L. Yao, Z. Chu, S. Li, Y . Li, J. Gao, A. Zhang, A survey on causal inference, ACM Transactions on Knowledge Discovery from Data (TKDD) 15 (5) (2021) 1–46

    2021

  37. [45]

    J. Zeng, G. Zhang, C. Rong, J. Ding, J. Yuan, Y . Li, Causal learning empowered od prediction for urban planning, in: Proceedings of the 31st ACM International Conference on Information and Knowledge Management, 2022, pp. 2455–2464

    2022

  38. [46]

    S. Deng, H. Rangwala, Y . Ning, Robust event forecasting with spatiotemporal confounder learning, in: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2022, pp. 294–304

    2022

  39. [47]

    Takeuchi, R

    K. Takeuchi, R. Nishida, H. Kashima, M. Onishi, Causal e ffect estimation on hierarchical spatial graph data, in: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 2145–2154

    2023

  40. [48]

    J. Ma, Y . Dong, Z. Huang, D. Mietchen, J. Li, Assessing the causal impact of covid-19 related policies on outbreak dynamics: A case study in the us, in: Proceedings of the ACM Web Conference 2022, 2022, pp. 2678–2686

    2022

  41. [49]

    Pearl, D

    J. Pearl, D. Mackenzie, The Book of Why: The New Science of Cause and E ffect, Basic Books, Inc., 2018

    2018

  42. [50]

    S. He, Q. Luo, R. Du, L. Zhao, G. He, H. Fu, H. Li, Stgc-gnns: A gnn-based tra ffic prediction framework with a spatial–temporal granger causality graph, Physica A: Statistical Mechanics and its Applications 623 (2023) 128913. doi:https://doi.org/10.1016/j.physa. 2023.128913. ...

    2023

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.