Capturing Graphs with Hypo-Elliptic Diffusions Csaba Toth Mathematical Institute University of Oxford

Convolutional layers within graph neural networks operate by aggregating information about local neighbourhood structures; one common way to encode such substructures is through random walks.