2023.02.01

Our paper on "Spatio-Temporal Meta-Graph Learning for Traffic Forecasting" has been accepted to the top international conference on artificial intelligence, AAAI2023.

InfoTech
Connected Advanced Development Div.
TOYOTA MOTOR CORPORATION
 Jiawei Yong (Project Leader)
 Yasumasa Kobayashi,
 Shintaro Fukushima (Group Manager)

Toyota Motor Corporation announces that a research paper on traffic prediction, as a joint research achievement with Assistant Professor Renhe Jiang, the University of Tokyo, has been accepted to the top international conference on artificial intelligence, AAAI2023.

●Presentation Highlights

• -  In order to develop a traffic prediction method that achieves high accuracy, high efficiency, and high interpretability, we designed a Meta-graph Convolutional Recurrent Network (MegaCRN) by combining Meta-Graph Learner and Encoder-Decoder Graph Convolutional Recurrent Network.
• -  On benchmark datasets METR-LA, PEMS-BAY, and Toyota's original dataset EXPY-TKY, our proposed method achieved higher prediction accuracy than existing methods in most accuracy evaluation metrics for road speed prediction. Additionally, our proposed method has relatively fewer model parameters compared to existing methods, resulting in significant reduction in training time.
• -  Various applications enhanced by this proposed method, including implementation in real-world services, are expected in the future.

●Presentation Info

• Date: 9th Feb 2023
• Conference: AAAI2023 (The 37th AAAI Conference on Artificial Intelligence)
• Title: Spatio-Temporal Meta-Graph Learning for Traffic Forecasting
• Pre-print: https://arxiv.org/abs/2211.14701

Please refer to the press release in the attached PDF for more details.