Question 15
Why do we need specialized deep learning architectures for graphs?
The flattened adjacency matrix does not change with ordering of nodes
Graphs have variable-sized structures and irregular connectivity patterns
Feed-forward networks are permutation equivariant functions
All of these