Główne pojęcia
The author proposes the COOL model to capture high-order spatio-temporal relationships in traffic forecasting by utilizing heterogeneous graphs and self-attention decoders.
Streszczenie
The paper introduces the COOL model for traffic forecasting, addressing the limitations of existing methods in capturing composite spatio-temporal relationships and diverse transitional patterns. By leveraging prior and posterior information, as well as self-attention mechanisms, COOL outperforms competitive baselines on benchmark datasets.
Statystyki
Nodes: 325 (PEMS-BAY), 170 (PEMS08), 207 (METR-LA), 883 (PEMS07)
Edges: 2369 (PEMS-BAY), 295 (PEMS08), 1515 (METR-LA), 866 (PEMS07)
Time Steps: 52,116 (PEMS-BAY), 17,856 (PEMS08), 34,272 (METR-LA), 28,224 (PEMS07)
Cytaty
"Existing methods remain sub-optimal due to their tendency to model temporal and spatial relationships independently."
"Our proposed COOL provides state-of-the-art performance compared with competitive baselines."