One for All: Universal Topological Primitive Transfer for Graph Structure Learning

Neural Information Processing Systems 

The non-Euclidean geometry inherent in graph structures fundamentally impedes cross-graph knowledge transfer.