Abstract: Existing spatio-temporal prediction networks that rely on recurrent neural networks face significant parallelization challenges, leading to high computational costs and prolonged training ...