Pith

open record

sign in

arxiv: 2606.21072 · v1 · pith:53T3WR6M · submitted 2026-06-19 · cs.LG · cs.AI

An Efficient and Effective Architecture for Large-Scale Traffic Prediction via Geometry-Adaptive Square Partitioning

Yongfeng Su , Hongwen Li , Zijian Zhang , Ziquan Fang , Lu Chen , Christian S. Jensen This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-26 14:54 UTCgrok-4.3pith:53T3WR6Mrecord.jsonopen to challenge →

classification cs.LG cs.AI
keywords traffic predictionspatial partitioningscalable spatiotemporal modelslinear complexitysensor networksurban computinghierarchical modeling
0
0 comments X

The pith

SqLinear partitions traffic sensors into balanced near-square regions and replaces attention with linear interactions to cut error and training time at large scales.

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

The paper presents SqLinear for traffic prediction when the number of sensors reaches thousands and standard neural models slow down. It introduces a Square Partition algorithm that divides sensors into balanced non-overlapping near-square regions with mathematical guarantees on shape and coverage. A Hierarchical Linear Interaction module then models local dynamics inside each region and global links between regions using only linear operations. On four large datasets this yields lower average error than ten baselines and faster training, especially when the spatial size or forecast horizon increases.

Core claim

By grounding spatial partitioning in an algorithm with provable guarantees on aspect ratio, balance, and utilization, and by modeling spatiotemporal relations through a hierarchical linear scheme of linear complexity, SqLinear delivers both higher accuracy and lower runtime than attention-based or heuristic-partition baselines on large-scale traffic data.

What carries the argument

The Square Partition algorithm that produces balanced near-square regions with provable guarantees, together with the Hierarchical Linear Interaction module that exchanges information inside and across regions through lightweight linear layers.

If this is right

  • The method scales to larger numbers of sensors and longer prediction horizons while keeping or improving accuracy.
  • Provable balance and near-square shape reduce padding waste and boundary fragmentation compared with standard spatial indexes.
  • Linear-complexity interaction replaces quadratic attention yet still captures both local and global dependencies needed for traffic dynamics.
  • The architecture supports real-time urban-scale forecasting on datasets where existing models become impractical.

Where Pith is reading between the lines

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

  • Similar geometry-adaptive partitioning could be tested on other dense spatial sensor tasks such as air-quality or energy-demand forecasting.
  • The linear scaling opens the possibility of adding more hierarchy levels to handle city-wide networks without retraining from scratch.
  • Recomputing partitions periodically on streaming sensor data would test whether the method remains stable when the underlying sensor layout changes.

Load-bearing premise

The square regions created by the partition preserve the true traffic patterns and dependencies so that the linear module can model them without missing signals at the artificial boundaries.

What would settle it

Running the same experiments with grid or quadtree partitions in place of Square Partition and finding equal or lower MAE together with comparable or faster training times would undermine the claimed advantage of the geometry-adaptive squares.

Figures

Figures reproduced from arXiv: 2606.21072 by Christian S. Jensen, Hongwen Li, Lu Chen, Yongfeng Su, Zijian Zhang, Ziquan Fang.

Figure 1
Figure 1. Figure 1: Top: spatial partitioning quality; Bottom: training [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The architecture and workflow of SqLinear. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Low-rank measurement. iii) More importantly, unlike general-purpose spatial indexing structures designed for range and 𝑘NN queries, Square Partition is specifically tailored for tensorized trajectory forecasting. By enforc￾ing a capacity constraint on every leaf node, each partition can be directly mapped to a padding-free tensor in R 𝐶×𝑃×𝑑 , thereby align￾ing the partitioning objective with the computatio… view at source ↗
Figure 4
Figure 4. Figure 4: Performance comparison under long-horizon forecasting settings. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Accuracy-efficiency bubble plots under the stan [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Per-sample training time comparison. the per-sample training time under long-horizon forecasting set￾tings. On the largest CA dataset, SqLinear achieves the shortest per-sample training time across all prediction horizons and reduces per-sample training time by 46.78% on average compared with the most efficient baseline in each setting. Under the most chal￾lenging 672-step forecasting scenario, SqLinear co… view at source ↗
Figure 7
Figure 7. Figure 7: Ablation study. alternating axes instead of Eq. 6; (5) w/ MedianSplit uses the median split point instead of Eq. 7; (6) w/o Inter and (7) w/o Intra remove inter- and intra-patch interactions, respectively; (8) w/ Attn replaces the linearMLP with standard attention. As shown in [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hyperparameter sensitivity analysis [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Partition strategy analysis. shallower models are sufficient for smaller graphs. Excessive depth can add unnecessary computation and may lead to overfitting on smaller datasets. The low-rank dimension 𝑘. A moderate low-rank dimension is enough to encode latent spatial relations. Increasing 𝑘 initially improves representation capacity, but overly large values introduce redundancy and increase memory usage w… view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of learned spatial dependencies. [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of partitioning results. (down to 0.16). These measurements confirm that Square Partition realizes the utilization and balance guarantees of the analysis while remaining the most compact among capacity-balanced partitions in practice. D.5 Case Study (RQ7) To provide an intuitive understanding of SqLinear, we present a case study visualizing the partitioning and the patterns learned by the me… view at source ↗
read the original abstract

Traffic prediction is a core task in intelligent transportation systems and urban-scale decision making. Despite the effectiveness of mainstream neural-network based methods, their deployment in real-world settings with thousands of traffic sensors is jeopardized severely by their poor computational scalability. To address this, the community has attempted to incorporate spatial database partitioning techniques (e.g., Grid, Quadtree, and K-D Tree) to improve model scalability. However, these approaches rely on handcrafted geometric heuristics and often produce irregular or imbalanced data partitions, leading to boundary fragmentation, excessive padding overheads, and degraded model accuracy. In this paper, we propose SqLinear, an efficient and effective architecture for large-scale traffic prediction. First, we design Square Partition, a geometry-adaptive algorithm that partitions massive traffic sensors into balanced, non-overlapping, and near-square spatial regions. Unlike existing heuristic-based designs, Square Partition is theoretically grounded and provides provable guarantees on aspect ratio, balance, and partition utilization, establishing a high-quality foundation for downstream spatiotemporal modeling. Next, we propose a Hierarchical Linear Interaction (HLI) module that abandons the costly attention mechanisms commonly used in Transformer-based spatio-temporal models. HLI efficiently captures both local intra-region dynamics and global inter-region dependencies through a lightweight linear interaction scheme, enabling effective spatiotemporal modeling with linear computational complexity. Extensive experiments on four large-scale traffic datasets and 10 baselines show that SqLinear reduces MAE by 2.30% on average under standard setting and by 5.81% under extreme scalability settings, while reducing training runtime by 13.27%--30.84% in spatial- and horizon-scaling scenarios.

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

2 major / 1 minor

Summary. The paper proposes SqLinear for large-scale traffic prediction. It introduces a Square Partition algorithm that adaptively divides traffic sensors into balanced, non-overlapping, near-square regions with provable guarantees on aspect ratio, balance, and partition utilization. This is paired with a Hierarchical Linear Interaction (HLI) module that models intra-region and inter-region spatiotemporal dependencies using a lightweight linear scheme with linear complexity, replacing attention mechanisms. Experiments across four large-scale datasets and 10 baselines report average MAE reductions of 2.30% (standard) and 5.81% (extreme scalability), plus training runtime reductions of 13.27%–30.84% in scaling scenarios.

Significance. If the central claims hold, the work offers a meaningful advance for scalable traffic prediction in settings with thousands of sensors, where attention-based models struggle with compute. The explicit theoretical guarantees on the partitioning step are a clear strength, as is the design of a linear-complexity alternative to attention for both local and global modeling. The reported efficiency and accuracy gains, if reproducible, could support practical deployment in intelligent transportation systems.

major comments (2)
  1. [Square Partition and HLI sections] The manuscript states that Square Partition 'establishes a high-quality foundation for downstream spatiotemporal modeling' because of its geometric guarantees, yet provides no analysis or ablation showing that the induced region boundaries preserve traffic flow continuity or that HLI can model the resulting intra- and inter-region dynamics without new artifacts. This link is load-bearing for the accuracy claims.
  2. [Abstract and theoretical claims] No proof sketches, formal statements of the guarantees, or dataset statistics appear to support the 'provable guarantees' and empirical results stated in the abstract; without these the soundness of the theoretical and experimental claims cannot be fully assessed.
minor comments (1)
  1. [Abstract] The abstract and introduction would benefit from explicit citation of the four datasets (sizes, time horizons, sensor counts) to allow readers to contextualize the scalability claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential contribution of SqLinear to scalable traffic prediction. We address each major comment below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Square Partition and HLI sections] The manuscript states that Square Partition 'establishes a high-quality foundation for downstream spatiotemporal modeling' because of its geometric guarantees, yet provides no analysis or ablation showing that the induced region boundaries preserve traffic flow continuity or that HLI can model the resulting intra- and inter-region dynamics without new artifacts. This link is load-bearing for the accuracy claims.

    Authors: We agree that an explicit analysis linking the geometric guarantees to preserved traffic flow continuity and artifact-free HLI modeling would strengthen the paper. The current design prioritizes aspect ratio, balance, and utilization to reduce boundary fragmentation, and the empirical gains versus baselines provide indirect support. However, we will add a new ablation subsection with boundary-effect experiments (e.g., comparing predictions near vs. far from boundaries) and visualizations of intra- and inter-region interactions to directly address this concern. revision: yes

  2. Referee: [Abstract and theoretical claims] No proof sketches, formal statements of the guarantees, or dataset statistics appear to support the 'provable guarantees' and empirical results stated in the abstract; without these the soundness of the theoretical and experimental claims cannot be fully assessed.

    Authors: Formal statements of the guarantees (Theorems 1–3 on aspect ratio, balance, and utilization) and their proofs appear in Section 3.2 and the appendix; dataset statistics are in Table 1. To improve accessibility from the abstract, we will insert a concise proof sketch and key dataset descriptors (sensor counts, time spans) into the revised abstract and add a theorem box in the main text. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical gains independent of partition definitions

full rationale

The paper introduces Square Partition with claimed provable guarantees on aspect ratio, balance and utilization, followed by an HLI module whose performance is measured via experiments on four datasets against 10 baselines. No equations, self-definitions or fitted parameters are shown reducing the reported MAE reductions (2.30% average, 5.81% extreme) or runtime savings to quantities defined by the partition itself or by self-citation chains. The geometric guarantees address only spatial properties; the accuracy claims rest on external experimental validation rather than tautological derivation from the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of a geometry-adaptive partitioning procedure that simultaneously satisfies balance, non-overlap, and near-square shape with provable bounds, plus the assumption that linear interactions suffice to model the resulting spatiotemporal dependencies. No free parameters, axioms, or invented entities are explicitly named in the abstract.

axioms (1)
  • domain assumption Existence of a geometry-adaptive partitioning procedure that simultaneously satisfies balance, non-overlap, and near-square shape with provable bounds
    Abstract states the Square Partition is theoretically grounded and provides provable guarantees; this premise is required for the downstream modeling claims.

pith-pipeline@v0.9.1-grok · 5846 in / 1302 out tokens · 18610 ms · 2026-06-26T14:54:01.158426+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

72 extracted references · 6 canonical work pages · 2 internal anchors

  1. [1]

    Evangelos C Alexopoulos. 2010. Introduction to multivariate regression analysis. Hippokratia14, Suppl 1 (2010), 23

    2010

  2. [2]

    Bang An, Xun Zhou, Amin Vahedian, Nick Street, Jinping Guan, and Jun Luo

  3. [3]

    InIEEE International Conference on Data Mining

    LISA: Learning-Integrated Space Partitioning Framework for Traffic Acci- dent Forecasting on Heterogeneous Spatiotemporal Data. InIEEE International Conference on Data Mining. 11–20

  4. [4]

    Lei Bai, Lina Yao, Can Li, Xianzhi Wang, and Can Wang. 2020. Adaptive graph convolutional recurrent network for traffic forecasting. InNeural Information Processing Systems. 17804–17815

    2020

  5. [5]

    Jon Louis Bentley. 1975. Multidimensional binary search trees used for associative searching.Commun. ACM18, 9 (1975), 509–517

    1975

  6. [6]

    Friedman

    Jon Louis Bentley and Jerome H. Friedman. 1979. Data Structures for Range Searching.Comput. Surveys11, 4 (1979), 397–409

    1979

  7. [7]

    George EP Box and David A Pierce. 1970. Distribution of residual autocorrelations in autoregressive-integrated moving average time series models.Journal of the American statistical Association65, 332 (1970), 1509–1526

    1970

  8. [8]

    Defu Cao, Yujing Wang, Juanyong Duan, Ce Zhang, Xia Zhu, Congrui Huang, Yunhai Tong, Bixiong Xu, Jing Bai, Jie Tong, and Qi Zhang. 2020. Spectral Tem- poral Graph Neural Network for Multivariate Time-series Forecasting.Neural Information Processing Systems(2020), 17766–17778

    2020

  9. [9]

    Peng Chen, Yingying ZHANG, Yunyao Cheng, Yang Shu, Yihang Wang, Qingsong Wen, Bin Yang, and Chenjuan Guo. 2024. Pathformer: Multi-scale Transformers with Adaptive Pathways for Time Series Forecasting. InInternational Conference on Learning Representations

    2024

  10. [10]

    Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui Hsieh

  11. [11]

    InACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Cluster-GCN: An efficient algorithm for training deep and large graph convolutional networks. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 257–266

  12. [12]

    Zheng Dong, Renhe Jiang, Haotian Gao, Hangchen Liu, Jinliang Deng, Qingsong Wen, and Xuan Song. 2024. Heterogeneity-Informed Meta-Parameter Learning for Spatiotemporal Time Series Forecasting. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 631–641

    2024

  13. [13]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al . 2020. An Image is Worth 16x16 Words: Trans- formers for Image Recognition at Scale. InInternational Conference on Learning Representations. 1–22

    2020

  14. [14]

    James Durbin. 1959. Efficient estimation of parameters in moving-average models. Biometrika46, 3/4 (1959), 306–316

    1959

  15. [15]

    Yuchen Fang, Yuxuan Liang, Bo Hui, Zezhi Shao, Liwei Deng, Xu Liu, Xinke Jiang, and Kai Zheng. 2025. Efficient Large-Scale Traffic Forecasting with Trans- formers: A Spatial Data Management Perspective. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 307–317

    2025

  16. [16]

    Jensen, Xiaofang Zhou, and Kai Zheng

    Yuchen Fang, Hao Miao, Yuxuan Liang, Liwei Deng, Yue Cui, Ximu Zeng, Yuyang Xia, Yan Zhao, Torben Bach Pedersen, Christian S. Jensen, Xiaofang Zhou, and Kai Zheng. 2025. Unraveling Spatio-Temporal Foundation Models via the Pipeline Lens: A Comprehensive Review.arXiv preprint arXiv:2506.01364

    work page arXiv 2025

  17. [17]

    Yuchen Fang, Yanjun Qin, Haiyong Luo, Fang Zhao, Bingbing Xu, Liang Zeng, and Chenxing Wang. 2023. When spatio-temporal meet wavelets: Disentangled traffic forecasting via efficient spectral graph attention networks. InIEEE International Conference on Data Engineering. 517–529

    2023

  18. [18]

    Friedman, Jon Louis Bentley, and Raphael Ari Finkel

    Jerome H. Friedman, Jon Louis Bentley, and Raphael Ari Finkel. 1977. An Al- gorithm for Finding Best Matches in Logarithmic Expected Time.ACM Trans. Math. Software3, 3 (1977), 209–226

    1977

  19. [19]

    Han Gao, Xu Han, Jiaoyang Huang, Jian-Xun Wang, and Liping Liu. 2022. Patchgt: Transformer over non-trainable clusters for learning graph representations. In Learning on Graphs Conference. 1–27

    2022

  20. [20]

    Shengnan Guo, Youfang Lin, Ning Feng, Chao Song, and Huaiyu Wan. 2019. Attention based spatial-temporal graph convolutional networks for traffic flow forecasting. InAAAI Conference on Artificial Intelligence. 922–929

    2019

  21. [21]

    Jindong Han, Weijia Zhang, Hao Liu, Tao Tao, Naiqiang Tan, and Hui Xiong

  22. [22]

    InProceedings of the VLDB Endowment

    BigST: Linear Complexity Spatio-Temporal Graph Neural Network for Traffic Forecasting on Large-Scale Road Networks. InProceedings of the VLDB Endowment. 1081–1090

  23. [23]

    Xiaoxue Han, Zhuo Feng, and Yue Ning. 2024. A topology-aware graph coars- ening framework for continual graph learning.Neural Information Processing Systems37 (2024), 132491–132523

    2024

  24. [24]

    Jiahao Ji, Jingyuan Wang, Chao Huang, Junjie Wu, Boren Xu, Zhenhe Wu, Junbo Zhang, and Yu Zheng. 2023. Spatio-temporal self-supervised learning for traffic flow prediction. InAAAI Conference on Artificial Intelligence. 4356 – 4364

    2023

  25. [25]

    Jiahao Ji, Wentao Zhang, Jingyuan Wang, and Chao Huang. 2025. Seeing the Un- seen: Learning Basis Confounder Representations for Robust Traffic Prediction. InACM International Conference on Information and Knowledge Management. 577–588

    2025

  26. [26]

    Guangyin Jin, Yuxuan Liang, Yuchen Fang, Zezhi Shao, Jincai Huang, Junbo Zhang, and Yu Zheng. 2023. Spatio-temporal graph neural networks for predictive learning in urban computing: A survey.IEEE Transactions on Knowledge and Data Engineering(2023), 5388–5408

    2023

  27. [27]

    Weiyang Kong, Kaiqi Wu, Sen Zhang, and Yubao Liu. 2025. GraphSparseNet: A Novel Method for Large Scale Traffic Flow Prediction.Proceedings of the VLDB Endowment18, 7 (2025), 2295–2307

    2025

  28. [28]

    Xiangjie Kong, Wenfeng Zhou, Guojiang Shen, Wenyi Zhang, Nali Liu, and Yao Yang. 2023. Dynamic graph convolutional recurrent imputation network for spatiotemporal traffic missing data.Knowledge-Based Systems261 (2023), 110188

    2023

  29. [29]

    Dilfira Kudrat, Zongxia Xie, Yanru Sun, Tianyu Jia, and Qinghua Hu. 2025. Patch- wise Structural Loss for Time Series Forecasting.arXiv preprint arXiv:2503.00877 (2025)

    work page arXiv 2025

  30. [30]

    Shiyong Lan, Yitong Ma, Weikang Huang, Wenwu Wang, Hongyu Yang, and Pyang Li. 2022. DSTAGNN: Dynamic spatial-temporal aware graph neural network for traffic flow forecasting. InInternational Conference on Machine Learning. 11906–11917

    2022

  31. [31]

    Fuxian Li, Jie Feng, Huan Yan, Guangyin Jin, Fan Yang, Funing Sun, Depeng Jin, and Yong Li. 2023. Dynamic graph convolutional recurrent network for traffic prediction: Benchmark and solution.ACM Transactions on Knowledge Discovery from Data(2023), 1–21

    2023

  32. [32]

    Yaguang Li, Rose Yu, Cyrus Shahabi, and Yan Liu. 2018. Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting. InInternational Conference on Learning Representations

    2018

  33. [33]

    Zhonghang Li, Lianghao Xia, Jiabin Tang, Yong Xu, Lei Shi, Long Xia, Dawei Yin, and Chao Huang. 2024. UrbanGPT: Spatio-Temporal Large Language Models. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 5351–5362

    2024

  34. [34]

    Yuxuan Liang, Haomin Wen, Yuqi Nie, Yushan Jiang, Ming Jin, Dongjin Song, Shirui Pan, and Qingsong Wen. 2024. Foundation models for time series analysis: A tutorial and survey. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 6555–6565

    2024

  35. [35]

    Chenxi Liu, Kethmi Hirushini Hettige, Qianxiong Xu, Cheng Long, Shili Xiang, Gao Cong, Ziyue Li, and Rui Zhao. 2025. ST-LLM+: Graph Enhanced Spatio- Temporal Large Language Models for Traffic Prediction.IEEE Transactions on Knowledge and Data Engineering37, 8 (2025), 4846–4859

    2025

  36. [36]

    Chenxi Liu, Sun Yang, Qianxiong Xu, Zhishuai Li, Cheng Long, Ziyue Li, and Rui Zhao. 2024. Spatial-temporal large language model for traffic prediction. In IEEE International Conference on Mobile Data Management. 31–40

    2024

  37. [37]

    Hangchen Liu, Zheng Dong, Renhe Jiang, Jiewen Deng, Jinliang Deng, Quanjun Chen, and Xuan Song. 2023. Spatio-temporal adaptive embedding makes vanilla transformer sota for traffic forecasting. InACM International Conference on Information and Knowledge Management. 4125–4129

    2023

  38. [38]

    Xu Liu, Yutong Xia, Yuxuan Liang, Junfeng Hu, Yiwei Wang, Lei Bai, Chao Huang, Zhenguang Liu, Bryan Hooi, and Roger Zimmermann. 2024. LargeST: A benchmark dataset for large-scale traffic forecasting. InNeural Information Processing Systems. 75354–75371

    2024

  39. [39]

    Zhi Liu, Yang Chen, Feng Xia, Jixin Bian, Bing Zhu, Guojiang Shen, and Xiangjie Kong. 2023. TAP: Traffic Accident Profiling via Multi-Task Spatio-Temporal Graph Representation Learning.ACM Transactions on Knowledge Discovery from Data17, 4 (2023), 1–25

    2023

  40. [40]

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin transformer: Hierarchical vision transformer using shifted windows. InIEEE/CVF International Conference on Computer Vision. 10012–10022

    2021

  41. [41]

    Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101(2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  42. [42]

    Tinghui Luo, Ziquan Fang, Kaixuan Duan, Lu Chen, Panpan Feng, and Mingfan Lu. 2025. Towards Online Spatio-Temporal Prediction: A Knowledge Distillation Driven Continual Learning Approach. InIEEE International Conference on Data Engineering. 2642–2655

    2025

  43. [43]

    Tengfei Lyu, Weijia Zhang, Jinliang Deng, and Hao Liu. 2025. AutoSTF: Decou- pled Neural Architecture Search for Cost-Effective Automated Spatio-Temporal Forecasting. InACM SIGKDD Conference on Knowledge Discovery and Data Min- ing. 985–996

    2025

  44. [44]

    Moin Hussain Moti, Panagiotis Simatis, and Dimitris Papadias. 2022. Waffle: A workload-aware and query-sensitive framework for disk-based spatial indexing. Proceedings of the VLDB Endowment16, 4 (2022), 670–683

    2022

  45. [45]

    Yuqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. 2023. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers. In International Conference on Learning Representations

    2023

  46. [46]

    Namuk Park and Songkuk Kim. 2022. How do vision transformers work?arXiv preprint arXiv:2202.06709(2022)

    work page Pith review arXiv 2022

  47. [47]

    Hanan Samet. 1984. The quadtree and related hierarchical data structures.ACM Computing Surveys (CSUR)16, 2 (1984), 187–260

    1984

  48. [48]

    2006.Foundations of multidimensional and metric data structures

    Hanan Samet. 2006.Foundations of multidimensional and metric data structures. Morgan Kaufmann

    2006

  49. [49]

    Jensen, and Xueqi Cheng

    Zezhi Shao, Fei Wang, Yongjun Xu, Wei Wei, Chengqing Yu, Zhao Zhang, Di Yao, Tao Sun, Guangyin Jin, Xin Cao, Gao Cong, Christian S. Jensen, and Xueqi Cheng

  50. [50]

    Su et al

    Exploring Progress in Multivariate Time Series Forecasting: Comprehensive Benchmarking and Heterogeneity Analysis.IEEE Transactions on Knowledge and Data Engineering37, 1 (2025), 291–305. Su et al

    2025

  51. [51]

    Zezhi Shao, Zhao Zhang, Fei Wang, Wei Wei, and Yongjun Xu. 2022. Spatial- temporal identity: A simple yet effective baseline for multivariate time series forecasting. InACM International Conference on Information and Knowledge Management. 4454–4458

    2022

  52. [52]

    Zezhi Shao, Zhao Zhang, Fei Wang, and Yongjun Xu. 2022. Pre-training Enhanced Spatial-temporal Graph Neural Network for Multivariate Time Series Forecasting. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 1567–1577

    2022

  53. [53]

    Zezhi Shao, Zhao Zhang, Wei Wei, Fei Wang, Yongjun Xu, Xin Cao, and Chris- tian S Jensen. 2022. Decoupled dynamic spatial-temporal graph neural network for traffic forecasting. InProceedings of the VLDB Endowment. 2733–2746

    2022

  54. [54]

    Waldo R Tobler. 1970. A computer movie simulating urban growth in the Detroit region.Economic geography46 (1970), 234–240

    1970

  55. [55]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Neural Information Processing Systems30 (2017)

    2017

  56. [56]

    Senzhang Wang, Jiannong Cao, and S Yu Philip. 2020. Deep learning for spatio- temporal data mining: A survey.IEEE transactions on knowledge and data engi- neering34, 8 (2020), 3681–3700

    2020

  57. [57]

    Yuxuan Wang, Haixu Wu, Jiaxiang Dong, Yong Liu, Mingsheng Long, and Jianmin Wang. 2024. Deep time series models: A comprehensive survey and benchmark. arXiv preprint arXiv:2407.13278(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  58. [58]

    Zonghan Wu, Shirui Pan, Guodong Long, Jing Jiang, Xiaojun Chang, and Chengqi Zhang. 2020. Connecting the dots: Multivariate time series forecasting with graph neural networks. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 753–763

    2020

  59. [59]

    Zonghan Wu, Shirui Pan, Guodong Long, Jing Jiang, and Chengqi Zhang. 2019. Graph wavenet for deep spatial-temporal graph modeling. InInternational Joint Conference on Artificial Intelligence. 1907–1913

    2019

  60. [60]

    Zhenda Xie, Yutong Lin, Zhuliang Yao, Zheng Zhang, Qi Dai, Yue Cao, and Han Hu. 2021. Self-supervised learning with swin transformers.arXiv preprint arXiv:2105.04553(2021)

    work page arXiv 2021

  61. [61]

    Haiqiang Yang, Zihan Li, and Yashuai Qi. 2024. Predicting traffic propagation flow in urban road network with multi-graph convolutional network.Complex & Intelligent Systems10, 1 (2024), 23–35

    2024

  62. [62]

    Yueyang Yao, Xingyuan Dai, and Yisheng Lv. 2025. Leveraging Heterogeneous Experts with Advantageous Pattern Memory Learning for Traffic Prediction. In IEEE International Conference on Data Engineering. 3342–3355

    2025

  63. [63]

    Chin-Chia Michael Yeh, Yujie Fan, Xin Dai, Uday Singh Saini, Vivian Lai, Prince Osei Aboagye, Junpeng Wang, Huiyuan Chen, Yan Zheng, Zhongfang Zhuang, et al. 2024. RPMixer: Shaking Up Time Series Forecasting with Random Projections for Large Spatial-Temporal Data. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3919–3930

    2024

  64. [64]

    Bing Yu, Haoteng Yin, and Zhanxing Zhu. 2018. Spatio-Temporal Graph Con- volutional Networks: A Deep Learning Framework for Traffic Forecasting. In International Joint Conference on Artificial Intelligence. 3634–3640

    2018

  65. [65]

    Yuan Yuan, Jingtao Ding, Jie Feng, Depeng Jin, and Yong Li. 2024. UniST: A Prompt-Empowered Universal Model for Urban Spatio-Temporal Prediction. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4095–4106

    2024

  66. [66]

    Zhuoning Yuan, Xun Zhou, and Tianbao Yang. 2018. Hetero-ConvLSTM: A deep learning approach to traffic accident prediction on heterogeneous spatio- temporal data. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 984–992

    2018

  67. [67]

    Zhihao Zeng, Ziquan Fang, Yuting Huang, Qilong Wang, Lu Chen, and Yunjun Gao. 2025. Heterogeneous-Aware Traffic Prediction: A Privacy-Preserving Feder- ated Learning Framework . InIEEE International Conference on Data Engineering. 419–432

    2025

  68. [68]

    Cheng Zhang, Haocheng Wan, Xinyi Shen, and Zizhao Wu. 2022. Patchformer: An efficient point transformer with patch attention. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 11799–11808

    2022

  69. [69]

    Weijia Zhang, Le Zhang, Jindong Han, Hao Liu, Yanjie Fu, Jingbo Zhou, Yu Mei, and Hui Xiong. 2024. Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Networks. InACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4302–4313

    2024

  70. [70]

    Yiji Zhao, Zihao Zhong, Ao Wang, Haomin Wen, Ming Jin, Yuxuan Liang, Huaiyu Wan, and Hao Wu. 2026. FaST: Efficient and Effective Long-Horizon Forecasting for Large-Scale Spatial-Temporal Graphs via Mixture-of-Experts. InProceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    2026

  71. [71]

    Chuanpan Zheng, Xiaoliang Fan, Cheng Wang, and Jianzhong Qi. 2020. Gman: A graph multi-attention network for traffic prediction. InAAAI Conference on Artificial Intelligence. 1234–1241

    2020

  72. [72]

    Qi Zheng, Zihao Yao, and Yaying Zhang. 2025. ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting. InAAAI Conference on Artificial Intelligence, Vol. 39. 13419–13427. SqLinear Appendix Directory Appendix A Related Work11 Appendix B Algorithm12 Appendix C Theoretical Analysis12 C.1 Analysis of Spatial Partition . . . . ....

    2025