What is the motivation of the graph filter in the Graph Convolutional Networks?

68 Views Asked by At

What is the motivation of the graph filter in the Graph Convolutional Networks? For example in the paper

In the Eq.~(3): they perform the convolution the singal $x$ with $g_\theta$ as:

$g_\theta * x = U g_\theta U^Tx$.

What it the role of $g_\theta$ here?Thanks

As I know from signal process convolution means "reverse, shift, multiply and sum". And for graph, what is the intuition for convolution?

1

There are 1 best solutions below

0
On BEST ANSWER

I think the graph filter make a role as localization the node with its neighborhoods. To make them share the properties with it. And nice paper recently explain this trick