[20240527_LabSeminar_Huy]Meta-Graph.pptx

Spatio-Temporal Meta-Graph Learning for Traffic Forecasting

1. Quang-Huy Tran Network Science Lab Dept. of Artificial Intelligence The Catholic University of Korea E-mail: [email protected] 2024-05-27 Spatio -Temporal Meta-Graph Learning for Traffic Forecasting Renhe Jiang et al. AAAI-23: The Thirty-Seventh AAAI Conference on Artificial Intelligence

2. OUTLINE MOTIVATION INTRODUCTION METHODOLOGY EXPERIMENT & RESULT CONCLUSION

3. MOTIVATION Challenges in traffic forecasting: Heterogeneity: traffic condition differs over roads and time. Non-stationarity : there are X factors, such as accidents, congestions . Overview

4. MOTIVATION Representation of gra ph structure (graph construction): Heuristic: Pre-defined empirically by a metric (distance, similarity, etc ). Graph structure learning: Static view (Adaptive graph): Given parameter Dynamic view (Momentary graph): Given parameter Overview

5. INTRODUCTION Propose a novel Meta-Graph Learner for spatiotemporal graph (STG) learning explicitly disentangles the heterogeneity in space and time. Present a generic Meta-Graph Convolutional Recurrent Network ( MegaCRN ): relies on observational data to be robust. adaptive to any traffic situation, from normal to non-stationary. Proposed newly-published dataset (EXPY-TKY) that has larger scale and more complex incident situations.

6. METHODOLOGY Problem Setting and Graph Structure Learning Problem formulation: Adaptive graph: Momentum graph: Adaptive graph at different window time.

7. METHODOLOGY Main Architecture - MegaCRN

8. METHODOLOGY Graph Convolutional Recurrent Unit Apply the Graph Convolutional Network in form of Gated Recurrent Unit (GRU). Apply in Encoder – Decoder architecture.

9. METHODOLOGY Spatio -Temporal Meta-Graph Learner Meta-Node Bank : memory networks for node embedding inject the memorizing and distinguishing capabilities into spatio -temporal graph learning. Meta-graph: alternative to adaptive and momentary graphs to feed back into GCRU. HyperNetwork NH: using a one network to generate the weights for another network. denote the number of memory items and the dimension of each item. [1] Ha, D., Dai, A., & Le, Q. V. (2016). Hypernetworks. arXiv preprint arXiv:1609.09106.

10. METHODOLOGY Meta-Graph Convolutional Recurrent Network MegaCRN learns node-level prototypes of traffic patterns in Meta-Node Bank for updating the auxiliary graph adaptively based on the observation. To distinguishing power for diverse scenarios on different roads over time, contrastive loss is represented:

11. EXPERIMENT AND RESULT EXPERIMENT SETTINGs Measurement: Mean absolute error (MAE). Mean absolute percentage error (MAPE). Root mean square error (RMSE). Dataset: METR-LA, PEMS-BAY, and EXPY-TKY. METR- LA, PEMS-BAY: R oad traffic datasets from California . EXPY-TKY: The traffic speed information and the corresponding traffic incident in Tokyo.

12. EXPERIMENT AND RESULT EXPERIMENT – BASELINE [1] Yu, B., Yin, H., & Zhu, Z. (2017). Spatio -temporal graph convolutional networks: A deep learning framework for traffic forecasting. arXiv preprint arXiv:1709.04875. [2] Wu, Z., Pan, S., Long, G., Jiang, J., & Zhang, C. (2019). Graph wavenet for deep spatial-temporal graph modeling. arXiv preprint arXiv:1906.00121. [3] Wu, Z., Pan, S., Long, G., Jiang, J., Chang, X., & Zhang, C. (2020, August). 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 (pp. 753-763). [4] Xu, M., Dai, W., Liu, C., Gao, X., Lin, W., Qi, G. J., & Xiong, H. (2020). Spatial-temporal transformer networks for traffic flow forecasting. arXiv preprint arXiv:2001.02908. [5] Zheng, C., Fan, X., Wang, C., & Qi, J. (2020, April). Gman : A graph multi-attention network for traffic prediction. In Proceedings of the AAAI conference on artificial intelligence (Vol. 34, No. 01, pp. 1234-1241). [6] Li, Y., Yu, R., Shahabi , C., & Liu, Y. (2017). Diffusion convolutional recurrent neural network: Data-driven traffic forecasting. arXiv preprint arXiv:1707.01926.. [7] Cao, D.; Wang, Y.; Duan, J.; Zhang, C.; Zhu, X.; Huang, C.;Tong , Y.; Xu, B.; Bai, J.; Tong, J.; et al. 2020. Spectral temporal graph neural network for multivariate time-series forecasting. Advances in Neural Information Processing Systems, 33: 17766–17778. [8] Bai, L., Yao, L., Li, C., Wang, X., & Wang, C. (2020). Adaptive graph convolutional recurrent network for traffic forecasting. Advances in neural information processing systems, 33, 17804-17815. [9] Ye, J., Sun, L., Du, B., Fu, Y., & Xiong, H. (2021, May). Coupled layer-wise graph convolution for transportation demand prediction. In Proceedings of the AAAI conference on artificial intelligence (Vol. 35, No. 5, pp. 4617-4625). [10] Shang, C., Chen, J., & Bi, J. (2021). Discrete graph structure learning for forecasting multiple time series. arXiv preprint arXiv:2101.06861. [11] Lee, H., Jin, S., Chu, H., Lim, H., & Ko, S. (2021). Learning to remember patterns: pattern matching memory networks for traffic forecasting. arXiv preprint arXiv:2110.10380. SOTA model : Naive : Historical Average. Fully Convolutional : STGCN[1], GraphWaveNet [2], MTGNN[3]. Transformer-based : STTN[4], GMAN[5]. GCRN-based : DCRNN[6], StemGNN [7], AGCRN[8], CCRNN[9], GTS[10], PM- MemNet [11]. Graph Structure learner: Adaptive : GraphWaveNet , MTGNN, CCRNN, GTS. Momentary : StemGNN .

13. EXPERIMENT AND RESULT RESULT – Overall Perfor mance

14. EXPERIMENT AND RESULT RESULT – Visualization

15. EXPERIMENT AND RESULT RESULT – Visualization

16. CONCLUSION Propose Meta-Graph Convolutional Recurrent Network ( MegaCRN ) along with a novel spatiotemporal graph structure learning mechanism. Generate a brand-new traffic dataset(EXPY-TKY) from large-scale car GPS records and collect the corresponding traffic incident information. Visualization demonstrated the capability to disentangle the time and nodes with different patterns as well as the adaptability to incident situations.