AdaST: adaptive coupling framework for spatiotemporal forecasting (arXiv:2609.36119v1)
AdaST is a new spatiotemporal forecasting framework that adaptively modulates spatial and temporal modeling by factorizing inputs into components with different coupling patterns using heterogeneity-aware experts, processing them with role-aligned modules, and integrating results via a correlation-informed adaptive recomposer. The arXiv:2609.36119v1 paper reports that AdaST substantially outperforms state-of-the-art baselines across extensive experiments, addressing issues from unknown and heterogeneous coupling to suboptimal spatial modeling.