2 papers across 2 sessions
We formalize the over-squashing phenomenon in spatiotemporal graph neural networks and analyze how it affects information propagation across the spatial and temporal dimensions.