GAS-PFN: Graph-Aware Spatiotemporal Prior-Data Fitted Networks for Zero-Shot Traffic Forecasting
arXiv cs.AI · 28 Oct 2026, 03:31 AEST
The Graph-Aware Spatiotemporal Prior-Data Fitted Network (GAS-PFN) enhances traffic forecasting by integrating graph-structured dependencies. It achieves significant reductions in mean squared error compared to existing models.
Key facts
- Traffic forecasting requires modeling complex temporal dynamics and the spatial dependencies imposed by road topology.
- To combine graph-aware traffic modeling with zero-shot adaptation, we introduce the Graph-Aware Spatiotemporal Prior-Data Fitted Network (GAS-PFN).
- Across six benchmarks at the 90-minute horizon, GAS-PFN reduces MSE by 34.0% on average relative to TimePFN, its closest PFN-based predecessor.
- It also reduces MSE by 27.7% on average relative to four recent industry-developed general-purpose zero-shot forecasters.
Read aloud
The Graph-Aware Spatiotemporal Prior-Data Fitted Network GAS-PFN enhances traffic forecasting by integrating graph-structured dependencies. It achieves significant reductions in mean squared error compared to existing models. Traffic forecasting requires modeling complex temporal dynamics and the spatial dependencies imposed by road topology. To combine graph-aware traffic modeling with zero-shot adaptation, we introduce the Graph-Aware Spatiotemporal Prior-Data Fitted Network GAS-PFN. Across six benchmarks at the ninety-minute horizon, GAS-PFN reduces MSE by thirty four point zero percent on average relative to TimePFN, its closest PFN-based predecessor. It also reduces MSE by twenty seven point seven percent on average relative to four recent industry-developed general-purpose zero-shot forecasters.
Sources in this cluster
- GAS-PFN: Graph-Aware Spatiotemporal Prior-Data Fitted Networks for Zero-Shot Traffic Forecasting
arXiv cs.AI · 28 Oct 2026, 03:31 AEST