Frequency-domain multi-scale graph learning with information-theoretic constraint for spatio-temporal prediction

Shun Wang, Yong Zhang, Xuanqi Lin, Guangyu Huo, Xinglin Piao, Yongli Hu, Baocai Yin

Abstract: Highlights•We introduce a frequency-domain multi-scale graph learning framework that unifies temporal decomposition and adaptive spatial modeling for spatiotemporal prediction.•Our method incorporates a scale independence constraint and a wavelet-weighted loss to theoretically and practically enhance multi-scale feature discrimination.•Extensive experiments show that FMSGL consistently surpasses state-of-the-art baselines on complex spatiotemporal datasets.

External IDs:dblp:journals/pr/WangZLHPHY26