How to Build a Graph-Based Deep Learning Architecture in Traffic Domain: A Survey (2005.11691v6)

Abstract: In recent years, various deep learning architectures have been proposed to solve complex challenges (e.g. spatial dependency, temporal dependency) in traffic domain, which have achieved satisfactory performance. These architectures are composed of multiple deep learning techniques in order to tackle various challenges in traffic tasks. Traditionally, convolution neural networks (CNNs) are utilized to model spatial dependency by decomposing the traffic network as grids. However, many traffic networks are graph-structured in nature. In order to utilize such spatial information fully, it's more appropriate to formulate traffic networks as graphs mathematically. Recently, various novel deep learning techniques have been developed to process graph data, called graph neural networks (GNNs). More and more works combine GNNs with other deep learning techniques to construct an architecture dealing with various challenges in a complex traffic task, where GNNs are responsible for extracting spatial correlations in traffic network. These graph-based architectures have achieved state-of-the-art performance. To provide a comprehensive and clear picture of such emerging trend, this survey carefully examines various graph-based deep learning architectures in many traffic applications. We first give guidelines to formulate a traffic problem based on graph and construct graphs from various kinds of traffic datasets. Then we decompose these graph-based architectures to discuss their shared deep learning techniques, clarifying the utilization of each technique in traffic tasks. What's more, we summarize some common traffic challenges and the corresponding graph-based deep learning solutions to each challenge. Finally, we provide benchmark datasets, open source codes and future research directions in this rapidly growing field.

• The paper introduces a standardized formulation for modeling traffic networks as graphs, capturing spatial and temporal dependencies effectively.
• It systematically evaluates deep learning techniques like GNNs, TCNs, and RNNs, demonstrating improved performance over traditional grid-based methods.
• It offers practical guidelines, benchmark datasets, and open-source codes to facilitate reproducibility and further research in intelligent transportation systems.

Survey on Graph-Based Deep Learning Architectures in the Traffic Domain

This paper offers an exhaustive survey of graph-based deep learning architectures specifically applied within the traffic domain. It focuses on the transition from traditional grid-based modeling to graph-based approaches that exploit the inherent graph structure of many traffic networks, such as road and subway systems. The survey meticulously breaks down various architectures, emphasizing how Graph Neural Networks (GNNs) are integrated with other deep learning techniques like Temporal Convolution Networks (TCN), Recurrent Neural Networks (RNNs), Sequence-to-Sequence models, and Generative Adversarial Networks (GANs).

Key Contributions

1. Problem Formulation and Graph Construction:
   • The paper provides a standardized approach for defining traffic problems as graph-based spatial-temporal forecasting tasks. It outlines the method for constructing traffic graphs using data collected from sensors, GPS trajectories, ride-hailing order logs, and transaction records from public transportation systems. This formulation recognizes the unique isomorphic nature of traffic networks and helps to systematically capture spatial dependencies through adjacency matrices.
2. Deep Learning Techniques:
   • A systematic exploration is provided on various deep learning techniques and their applicability in traffic forecasting. It includes spectral graph convolution networks and diffusion graph convolution networks, demonstrating their superiority in processing the spatial complexity of traffic networks over conventional grid-based methods.
   • Temporal dependencies are addressed through the integration of RNNs and TCNs, which offer more efficient and stable sequence modeling capabilities.
3. Challenges and Solutions:
   • The paper identifies persistent challenges in traffic forecasting, including spatial locality, multi-relationships, global connectivity, and temporal dependencies. It highlights how graph-based methods can better capture complex network dynamics and presents various adaptations of deep learning techniques to tackle these challenges effectively.
4. Practical Implementation:
   • It gathers benchmark datasets and open-source codes for reproducibility, facilitating experimental validation of proposed models and accelerating future research and development in intelligent transportation systems.

Results and Implications

Graph-based deep learning architectures demonstrate improved prediction accuracy and model efficiency for traffic forecasting tasks due to their ability to leverage spatial structures inherently present in traffic networks. The marriage of GNNs with temporal learning models like RNNs and TCNs, offers a comprehensive solution to modeling spatiotemporal dependencies more effectively than traditional methods. Furthermore, the practical guidelines for graph construction and model selection extend significant utility to real-world traffic management applications.

Future Research Directions

The survey suggests several unexplored and promising research areas, including the application of graph-based techniques to traffic incident detection, transfer learning across different cities, and deeper integration with newer deep learning frameworks like reinforcement learning. It also notes the potential of further development of GNN branches, such as graph auto-encoders and recurrent graph neural networks, which remain underrepresented in traffic research.

Conclusion

Overall, this paper stands as a significant resource for researchers interested in the intersection of graph theory, deep learning, and transportation systems. It provides detailed analysis, theoretical insights, and practical methods for leveraging graph-based deep learning models to advance intelligent transportation systems, optimize traffic prediction tasks, and ultimately improve urban mobility and safety.