Regularized Adaptive Graph Convolution for Scalable Traffic Forecasting
Traffic prediction on large-scale road networks faces scalability challenges due to the quadratic complexity of traditional graph convolution. Researchers propose a Regularized Adaptive Graph Convolution (RAGC) model featuring an Efficient Cosine Operator (ECO) to reduce computational cost while maintaining accuracy. The method addresses limitations of existing approximation, compression, and spatial partitioning techniques that often trade efficiency for precision. RAGC aims to enable practical deployment of spatial-temporal graph convolutional networks on extensive urban road networks.
Key facts
- Traffic prediction is critical for travel planning and urban management.
- Spatial-Temporal Graph Convolutional Networks (STGCNs) achieve advanced performance but have quadratic complexity.
- Large-scale road networks limit scalability of traditional graph convolution.
- Existing solutions include approximation, compression, or spatial partitioning.
- These methods often fail to achieve sufficient efficiency or compromise accuracy.
- RAGC introduces the Efficient Cosine Operator (ECO) for graph convolution.
- The model is designed for scalability on large road networks.
- The paper is available on arXiv with ID 2506.07179.
Entities
Institutions
- arXiv