Do We Really Need Adaptive Global Spatial Attention for Traffic Forecasting?
arXiv:2607.12462
The paper investigates whether transformer-style attention is necessary for extracting global spatial information in traffic forecasting, showing that a simple uniform full-range mixing operator achieves comparable accuracy with lower computational complexity.
Abstract
Existing traffic forecasting models commonly focus on extracting spatial dependencies, particularly global spatial information, which characterizes the representations obtained through interactions between each node and all nodes across the traffic network. However, the underlying mechanism by which global information is modeled and extracted remains insufficiently investigated. Whether global information must be extracted by high-degree-of-freedom adaptive attention or can be captured by a simple global aggregation operator remains unclear. For this purpose, we design a controlled ablation framework that replaces only the spatial mixing module to test attention-based global interaction. Across six traffic benchmarks, standard spatial attention yields relative MAE changes of to compared with uniform full-range mixing, and we observe no consistent advantage for standard spatial attention, while uniform full-range mixing reduces node-scale spatial-mixing complexity from to . We further propose a hypothesized model that decomposes spatial attention into a row-uniform global background and a non-uniform residual. The residual shows dataset-dependent effects. Overall, uniform full-range mixing provides a strong global spatial baseline, while the non-uniform attention residual is not consistently beneficial across datasets.
Revised manuscript; added experiments and updated figures and tables. 16 pages, 3 figures, results unchanged. Source code available at https://github.com/uuesti/U-Trans