pith. sign in

arxiv: 2501.12281 · v2 · submitted 2025-01-21 · 💻 cs.LG

MoGERNN: An Inductive Traffic Predictor for Unobserved Locations

Qishen Zhou , Yifan Zhang , Michail A. Makridis , Anastasios Kouvelas , Yibing Wang , Simon Hu This is my paper

Pith reviewed 2026-05-23 04:46 UTC · model grok-4.3

classification 💻 cs.LG
keywords traffic predictiongraph neural networksinductive learningmixture of expertsspatio-temporal modelingroad networkssensor networks
0
0 comments X

The pith

MoGERNN uses a mixture of graph experts with sparse gating to predict traffic states at locations without any sensors.

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

The paper presents MoGERNN as an inductive model that learns from a partially observed road network and forecasts traffic at completely unobserved nodes. A mixture-of-graph-experts component with sparse gating routes nodes to specialized aggregators that capture varied spatial dependencies, while an encoder-decoder handles the combined spatial-temporal patterns. The approach is designed to remain effective when sensors are added or removed, without requiring full retraining. Experiments on two real datasets show the model beats baselines at both observed and unobserved sites and tracks congestion evolution in unsensored areas. This addresses the practical reality that traffic sensor coverage is always incomplete and changes over time.

Core claim

MoGERNN is an inductive spatio-temporal graph model whose Mixture of Graph Experts (MoGE) with sparse gating learns heterogeneous spatial dependencies from the observed subgraph alone and generalizes those dependencies to predict traffic states at unobserved nodes, while an encoder-decoder architecture integrates spatial and temporal information for the full prediction task.

What carries the argument

Mixture of Graph Experts (MoGE) with sparse gating that dynamically routes nodes to specialized graph aggregators.

If this is right

  • Traffic managers can obtain congestion forecasts for roads that have never had sensors installed.
  • Adding or removing sensors does not force complete model retraining while performance stays competitive.
  • Prediction accuracy remains stable across different densities of available sensors.
  • Ablation tests confirm that both the mixture-of-experts routing and the encoder-decoder structure contribute to the reported gains.

Where Pith is reading between the lines

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

  • The same inductive routing mechanism could be tested on other networked prediction problems such as power-grid load or epidemic spread where only partial node observations exist.
  • If the learned experts correspond to distinct traffic regimes, the model might reveal interpretable clusters of road behavior that planners could use directly.
  • Extending the sparse gating to include temporal experts could allow the same architecture to handle non-stationary traffic patterns without extra modules.

Load-bearing premise

A sparse-gating mixture of graph experts can extract heterogeneous spatial dependencies from the observed part of the network and apply those same dependencies to nodes that have never been observed.

What would settle it

Run MoGERNN on a dataset where ground-truth traffic measurements exist at locations treated as unobserved during training; if the model's error at those locations is not lower than strong transductive baselines, the generalization claim fails.

Figures

Figures reproduced from arXiv: 2501.12281 by Anastasios Kouvelas, Michail A. Makridis, Qishen Zhou, Simon Hu, Yibing Wang, Yifan Zhang.

Figure 1
Figure 1. Figure 1: An illustration of forecasting unobserved nodes under a dynamic sensing network: (a) an example of sensor [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The overview of the proposed MoGERNN. • Distance-based weighted aggregator: 𝑋 (𝑙+1) 𝑗 = Í 𝑖:A𝑖 𝑗>0 𝑋 (𝑙) 𝑖 ·A𝑖 𝑗 Í 𝑖:A𝑖 𝑗>0 A𝑖 𝑗 • Mean aggregator: 𝑋 (𝑙+1) 𝑗 = mean𝑖:A𝑖 𝑗>0 𝑋 (𝑙) 𝑖 • Max pooling aggregator: 𝑋 (𝑙+1) 𝑗 = max𝑖:A𝑖 𝑗>0 𝑋 (𝑙) 𝑖 • Min pooling aggregator: 𝑋 (𝑙+1) 𝑗 = min𝑖:A𝑖 𝑗>0 𝑋 (𝑙) 𝑖 • Diffusion convolution aggregator: 𝑋 (𝑙+1) 𝑗 = Í𝐾−1 𝑘=0 ( Í 𝑖 (D−1 𝑂 A)𝑘 𝑖 𝑗 𝑋 (𝑙) 𝑖 W(𝑙) 𝑘,𝑂 + Í 𝑖 (D−1 𝐼 A⊤) … view at source ↗
Figure 3
Figure 3. Figure 3: Map presentation of prediction performance in an evening-peak time point (2012-05-24 17:30) of METR-LA. [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Predicting results for METR-LA. (a) show the results of unobserved locations, including three virtual sensors [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Model performance under different ratios of VS to AAS. The first row of x-ticks labels indicates the number of [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
read the original abstract

Given a partially observed road network, how can we predict the traffic state of interested unobserved locations? Traffic prediction is crucial for advanced traffic management systems, with deep learning approaches showing exceptional performance. However, most existing approaches assume sensors are deployed at all locations of interest, which is impractical due to financial constraints. Furthermore, these methods are typically fragile to structural changes in sensing networks, which require costly retraining even for minor changes in sensor configuration. To address these challenges, we propose MoGERNN, an inductive spatio-temporal graph model with two key components: (i) a Mixture of Graph Experts (MoGE) with sparse gating mechanisms that dynamically route nodes to specialized graph aggregators, capturing heterogeneous spatial dependencies efficiently; (ii) a graph encoder-decoder architecture that leverages these embeddings to capture both spatial and temporal dependencies for comprehensive traffic state prediction. Experiments on two real-world datasets show MoGERNN consistently outperforms baseline methods for both observed and unobserved locations. MoGERNN can accurately predict congestion evolution even in areas without sensors, offering valuable information for traffic management. Moreover, MoGERNN is adaptable to the changes of sensor network, maintaining competitive performance even compared to its retrained counterpart. Tests performed with different numbers of available sensors confirm its consistent superiority, and ablation studies validate the effectiveness of its key modules. The code of this work is publicly available at: https://github.com/ZJU-TSELab/MoGERNN.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The paper introduces MoGERNN, an inductive spatio-temporal graph neural network for traffic state prediction at unobserved locations in partially observed road networks. It uses a Mixture of Graph Experts (MoGE) module with sparse gating to route nodes to specialized aggregators for heterogeneous spatial dependencies, combined with a graph encoder-decoder to model spatio-temporal patterns. Experiments on two real-world datasets claim consistent outperformance over baselines for both observed and unobserved nodes, plus adaptability to sensor network changes without full retraining, supported by ablation studies and tests with varying sensor counts. Code is released publicly.

Significance. If the empirical claims hold under rigorous validation, the work addresses a practically important gap in traffic forecasting by enabling inductive generalization to unsensored locations and robustness to sensor reconfiguration. This could reduce deployment costs for traffic management systems. Public code release supports reproducibility and is a positive contribution.

minor comments (3)
  1. [Abstract] Abstract: the performance claims reference 'two real-world datasets' and 'baseline methods' without naming them or reporting key metrics (e.g., MAE, RMSE) or dataset sizes; this should be expanded for immediate clarity even if details appear later.
  2. The description of the sparse gating mechanism in the MoGE component would benefit from an explicit equation or pseudocode showing how the routing probabilities are computed and how sparsity is enforced.
  3. Figure captions and axis labels should be checked for completeness so that results for unobserved locations are immediately interpretable without reference to the main text.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive review and recommendation of minor revision. The summary accurately reflects the paper's focus on inductive prediction for unobserved locations and adaptability to sensor network changes. No major comments are listed in the report.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces MoGERNN as a new inductive architecture (MoGE with sparse gating plus encoder-decoder) and supports its claims via experiments on two real-world datasets showing outperformance on observed and unobserved nodes. No equations, parameter-fitting steps, or derivation chain appear in the supplied text. The generalization claim rests on empirical results rather than any self-definitional mapping, fitted-input prediction, or self-citation reduction. This is the expected honest non-finding for an architecture-plus-experiments paper whose central assertions are externally falsifiable.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, mathematical axioms, or invented entities beyond the high-level model name. No numerical constants, lemmas, or new physical quantities are mentioned.

pith-pipeline@v0.9.0 · 5807 in / 1216 out tokens · 22787 ms · 2026-05-23T04:46:03.509117+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Proceedings of the AAAI Conference on Artificial Intelligence 34, 3187–3194

    Kriging convolutional networks. Proceedings of the AAAI Conference on Artificial Intelligence 34, 3187–3194. doi:10.1609/aaai.v34i04.5716. Aw, A., Rascle, M.,

    work page doi:10.1609/aaai.v34i04.5716

  2. [2]

    Transportation Research Part C: Emerging Technologies 143, 103820

    A novel reinforced dynamic graph convolutional network model with data imputation for network-wide traffic flow prediction. Transportation Research Part C: Emerging Technologies 143, 103820. doi:https://doi.org/10.1016/j.trc.2022.103820. Cini, A., Marisca, I., Alippi, C.,

    work page doi:10.1016/j.trc.2022.103820 2022

  3. [3]

    Deng, D., Shahabi, C., Demiryurek, U., Zhu, L., Yu, R., Liu, Y .,

    Using physics-informed regularization to improve extrapolation capabilities of neural networks, in: Fourth Workshop on Machine Learning and the Physical Sciences (NeurIPS 2021). Deng, D., Shahabi, C., Demiryurek, U., Zhu, L., Yu, R., Liu, Y .,

    work page 2021

  4. [4]

    Transportation Research Part C: Emerging Technologies 108, 12–28

    An effective spatial-temporal attention based neural network for traffic flow prediction. Transportation Research Part C: Emerging Technologies 108, 12–28. doi: https://doi.org/10.1016/j.trc.2019.09.008. Fesser, L., D’Amico-Wong, L., Qiu, R.,

    work page doi:10.1016/j.trc.2019.09.008 2019

  5. [5]

    arXiv preprint arXiv:2306.09478

    Understanding and mitigating extrapolation failures in physics-informed neural networks. arXiv preprint arXiv:2306.09478 . 17 Frejo, J.R.D., Camacho, E.F.,

    work page arXiv

  6. [6]

    IEEE Transactions on Intelligent Transportation Systems 13, 1556–1565

    Global versus local mpc algorithms in freeway traffic control with ramp metering and variable speed limits. IEEE Transactions on Intelligent Transportation Systems 13, 1556–1565. doi: 10.1109/TITS.2012.2195493. Gilmer, J., Schoenholz, S.S., Riley, P.F., Vinyals, O., Dahl, G.E., 2017a. Neural message passing for quantum chemistry, in: International confere...

    work page doi:10.1109/tits.2012.2195493 2012

  7. [7]

    Proceedings of the AAAI Conference on Artificial Intelligence 35, 8146–8154

    Dpm: A novel training method for physics- informed neural networks in extrapolation. Proceedings of the AAAI Conference on Artificial Intelligence 35, 8146–8154. doi: 10.1609/aaai.v35i9.16992. Kipf, T.N., Welling, M., 2017a. Semi-supervised classification with graph convolutional networks, in: International Conference on Learning Representations. URL: htt...

    work page doi:10.1609/aaai.v35i9.16992

  8. [8]

    IEEE Transactions on Intelligent Transportation Systems 25, 2966–2975

    A variational bayesian inference-based en-decoder framework for traffic flow prediction. IEEE Transactions on Intelligent Transportation Systems 25, 2966–2975. doi: 10.1109/TITS.2023.3276216. Lee, H., Ko, S.,

    work page doi:10.1109/tits.2023.3276216 2023

  9. [9]

    IEEE Transactions on Intelligent Transportation Systems 23, 18962–18974

    Bayesian kernelized matrix factorization for spatiotemporal traffic data imputation and kriging. IEEE Transactions on Intelligent Transportation Systems 23, 18962–18974. doi: 10.1109/TITS.2022.3161792. Li, S., Cui, Y ., Zhao, Y ., Yang, W., Zhang, R., Zhou, X.,

    work page doi:10.1109/tits.2022.3161792 2022

  10. [10]

    St-moe: Spatio-temporal mixture- of-experts for debiasing in traffic prediction, in: Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, Association for Computing Machinery, New York, NY , USA. p. 1208–1217. doi:10.1145/3583780.3615068. Li, Y ., Yu, R., Shahabi, C., Liu, Y .,

    work page doi:10.1145/3583780.3615068

  11. [11]

    Spatio-temporal adaptive embedding makes vanilla transformer sota for traffic forecasting, in: Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, Association for Computing Machinery, New York, NY , USA. p. 4125–4129. doi:10.1145/3583780.3615160. 18 Liu, M., Zeng, A., Chen, M., Xu, Z., Lai, Q., Ma, L., Xu, Q.,

    work page doi:10.1145/3583780.3615160

  12. [12]

    Transportation Research Part C: Emerging Technologies 149, 104066

    An adaptive framework for real-time freeway traffic estimation in the presence of cavs. Transportation Research Part C: Emerging Technologies 149, 104066. doi:https://doi.org/10.1016/j.trc.2023.104066. Mei, H., Li, J., Liang, Z., Zheng, G., Shi, B., Wei, H.,

    work page doi:10.1016/j.trc.2023.104066 2023

  13. [13]

    1223–1228

    Uncertainty-aware traffic prediction under missing data, in: 2023 IEEE International Conference on Data Mining (ICDM), pp. 1223–1228. doi:10.1109/ICDM58522.2023.00152. Payne, H.J.,

    work page doi:10.1109/icdm58522.2023.00152 2023

  14. [14]

    Journal of Computational Physics 378, 686–707

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics 378, 686–707. doi: https://doi.org/10. 1016/j.jcp.2018.10.045. Richards, P.I.,

    work page 2018

  15. [15]

    Transportation Research Record 2559, 90–100

    Use of speed measurements for highway traffic state estimation: Case studies on ngsim data and highway a20, netherlands. Transportation Research Record 2559, 90–100. doi: 10.3141/2559-11. Roth, A., Liebig, T.,

    work page doi:10.3141/2559-11

  16. [16]

    Forecasting unobserved node states with spatio-temporal graph neural networks, in: 2022 IEEE International Conference on Data Mining Workshops (ICDMW), pp. 740–747. doi: 10.1109/ICDMW58026.2022.00101. Saito, H., McKenna, S.A., Zimmerman, D., Coburn, T.C.,

    work page doi:10.1109/icdmw58026.2022.00101 2022

  17. [17]

    Spatial-temporal identity: A simple yet effective baseline for multivariate time series forecasting, in: Proceedings of the 31st ACM International Conference on Information & Knowledge Management, Association for Computing Machinery, New York, NY , USA. p. 4454–4458. doi:10.1145/3511808.3557702. Shaygan, M., Meese, C., Li, W., Zhao, X.G., Nejad, M.,

    work page doi:10.1145/3511808.3557702

  18. [18]

    Transportation Research Part C: Emerging Technologies 145, 103921

    Traffic prediction using artificial intelligence: Review of recent advances and emerging opportunities. Transportation Research Part C: Emerging Technologies 145, 103921. doi:https://doi.org/10.1016/j.trc.2022. 103921. Shi, R., Mo, Z., Di, X.,

    work page doi:10.1016/j.trc.2022 2022

  19. [19]

    Proceedings of the AAAI Conference on Artificial Intelligence 35, 540–547

    Physics-informed deep learning for traffic state estimation: A hybrid paradigm informed by second-order traffic models. Proceedings of the AAAI Conference on Artificial Intelligence 35, 540–547. doi:10.1609/aaai.v35i1.16132. 19 Shi, R., Mo, Z., Huang, K., Di, X., Du, Q.,

    work page doi:10.1609/aaai.v35i1.16132

  20. [20]

    IEEE Transactions on Intelligent Transportation Systems 23, 11688–11698

    A physics-informed deep learning paradigm for traffic state and fundamental diagram estimation. IEEE Transactions on Intelligent Transportation Systems 23, 11688–11698. doi: 10.1109/TITS.2021.3106259. Song, C., Lin, Y ., Guo, S., Wan, H.,

    work page doi:10.1109/tits.2021.3106259 2021

  21. [21]

    Proceedings of the AAAI Conference on Artificial Intelligence 34, 914–921

    Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting. Proceedings of the AAAI Conference on Artificial Intelligence 34, 914–921. doi:10.1609/aaai.v34i01.5438. Sun, Y ., Work, D.B.,

    work page doi:10.1609/aaai.v34i01.5438

  22. [22]

    IEEE Transac- tions on Control of Network Systems 5, 968–980

    Scaling the kalman filter for large-scale traffic estimation. IEEE Transac- tions on Control of Network Systems 5, 968–980. doi:10.1109/TCNS.2017.2668898. Wackernagel, H.,

    work page doi:10.1109/tcns.2017.2668898 2017

  23. [23]

    Springer Berlin Heidelberg, Berlin, Heidelberg

    Ordinary Kriging. Springer Berlin Heidelberg, Berlin, Heidelberg. pp. 74–81. Wang, H., Zhang, R., Cheng, X., Yang, L., 2022a. Hierarchical traffic flow prediction based on spatial-temporal graph convolutional network. IEEE Transactions on Intelligent Transportation Systems 23, 16137–16147. doi: 10.1109/TITS.2022.3148105. Wang, S., Wu, H., Shi, X., Hu, T.,...

    work page doi:10.1109/tits.2022.3148105 2022

  24. [24]

    Transportation Research Part B: Methodological 39, 141–167

    Real-time freeway traffic state estimation based on extended kalman filter: a general approach. Transportation Research Part B: Methodological 39, 141–167. doi:https://doi.org/10.1016/j.trb.2004.03.003. Wang, Y ., Zhao, M., Yu, X., Hu, Y ., Zheng, P., Hua, W., Zhang, L., Hu, S., Guo, J., 2022b. Real-time joint traffic state and model parameter estimation ...

    work page doi:10.1016/j.trb.2004.03.003 2004

  25. [25]

    Whitham, G.B.,

    1109/TITS.2020.2995546. Whitham, G.B.,

    work page arXiv 2020

  26. [26]

    Wu, Y ., Zhuang, D., Labbe, A., Sun, L., 2021a

    Timesnet: Temporal 2d-variation modeling for general time series analysis, in: The Eleventh International Conference on Learning Representations. Wu, Y ., Zhuang, D., Labbe, A., Sun, L., 2021a. Inductive graph neural networks for spatiotemporal kriging, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 4478–4485. Wu, Y ., Zhuang, D., ...

    work page arXiv

  27. [27]

    Connecting the dots: Multivariate time series forecasting with graph neural networks, in: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Association for Computing Machinery, New York, NY , USA. p. 753–763. doi:10.1145/3394486.3403118. 20 Xiong, L., Chen, X., Huang, T.K., Schneider, J., Carbonell, J.G.,

    work page doi:10.1145/3394486.3403118

  28. [28]

    Temporal collaborative filtering with bayesian probabilistic tensor factorization, in: Proceedings of the 2010 SIAM international conference on data mining, SIAM. pp. 211–222. Xu, D., Tang, Y ., Peng, H., Guo, H., Xuan, Q.,

    work page 2010

  29. [29]

    IEEE Internet of Things Journal , 1–1doi:10.1109/JIOT.2024.3427429

    Traffic state estimation of road sections without detectors based on multi-source causal interpretation graph. IEEE Internet of Things Journal , 1–1doi:10.1109/JIOT.2024.3427429. Xu, K., Hu, W., Leskovec, J., Jegelka, S.,

    work page doi:10.1109/jiot.2024.3427429 2024

  30. [30]

    Transportation Research Part B: Methodological 167, 99–117

    A traffic flow dependency and dynamics based deep learning aided approach for network-wide traffic speed propagation prediction. Transportation Research Part B: Methodological 167, 99–117. doi: https://doi.org/10.1016/j.trb. 2022.11.009. Yang, H., Yu, W., Zhang, G., Du, L.,

    work page doi:10.1016/j.trb 2022

  31. [31]

    IEEE Transactions on Intelligent Transportation Systems 23, 4927–4943

    Deep learning on traffic prediction: Methods, analysis, and future directions. IEEE Transactions on Intelligent Transportation Systems 23, 4927–4943. doi: 10.1109/TITS.2021.3054840. Yu, B., Yin, H., Zhu, Z.,

    work page doi:10.1109/tits.2021.3054840 2021

  32. [32]

    Yuan, Y ., Zhang, Z., Yang, X.T., Zhe, S.,

    1109/TITS.2021.3131333. Yuan, Y ., Zhang, Z., Yang, X.T., Zhe, S.,

    work page arXiv 2021

  33. [33]

    Transportation Research Part B: Methodological 146, 88–110

    Macroscopic traffic flow modeling with physics regularized gaussian process: A new insight into machine learning applications in transportation. Transportation Research Part B: Methodological 146, 88–110. doi: https://doi.org/10. 1016/j.trb.2021.02.007. Zhang, H.M.,

    work page 2021

  34. [34]

    IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 21, 3848–3858

    T-gcn: A temporal graph convolutional network for traffic prediction. IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 21, 3848–3858. doi:10.1109/TITS.2019.2935152. Zheng, C., Fan, X., Wang, C., Qi, J., Chen, C., Chen, L.,

    work page doi:10.1109/tits.2019.2935152 2019

  35. [35]

    Increase: Inductive graph representation learning for spatio-temporal kriging, in: Proceedings of the ACM Web Con- ference 2023, Association for Computing Machinery, New York, NY , USA. p. 673–683. doi:10.1145/3543507.3583525. 21

    work page doi:10.1145/3543507.3583525 2023