<ul><li>Dynamic Stiefel Graph Neural Network (DST-SGNN) proposed for efficient spatio-temporal time series forecasting.</li><li>DST-SGNN addresses challenges in accurately forecasting spatio-temporal time series by balancing effectiveness and efficiency in modeling dynamic relations.</li><li>Novel techniques like Stiefel Graph Spectral Convolution (SGSC), Stiefel Graph Fourier Transform (SGFT), and Linear Dynamic Graph Optimization on Stiefel Manifold (LDGOSM) introduced in DST-SGNN.</li><li>Extensive experiments on seven datasets demonstrate that DST-SGNN outperforms existing methods in forecasting spatio-temporal time series with lower computational costs.</li></ul>

A Dynamic Stiefel Graph Neural Network for Efficient Spatio-Temporal Time Series Forecasting

Discover more