Proposes a method to forecast spatial-temporal data with limited training data.
problem Forecasting with nodes having no temporal training data.
method Temporal data augmentation and spatial graph topology learning.
result Improves forecasting performance on nodes without training data.