PatchSTG: Scalable Spatiotemporal Graph Transformers for Traffic Forecasting on Irregular Sensor Networks

In three linesPatchSTG, a patch-based spatiotemporal Transformer, improves traffic forecasting on irregular sensor networks. The model partitions sensors into balanced patches and uses dual attention (intra-patch and inter-patch) to reduce complexity from quadratic to near-linear. Evaluation on Rhode Island traffic data shows competitive performance with improved computational efficiency.

Summary generated by Claude — human-verified