search

flyingtango / DiGCN

Implement of DiGCN, NeurIPS-2020
MIT License
45 stars 8 forks source link

Question about equation 4 #5

Closed JhuoW closed 2 years ago

JhuoW commented 2 years ago

Dear author, I'm a little confused about Eq.4. You adopt symmetric normalized directed Laplacian proposed by Chung F (2005), however, It is not clear to me why the graph filter in Eq 4 can reflect directed property? Since it is a symmetric matrix.

flyingtango commented 2 years ago

Hi,

Thank you for your interest in our work. We consider that the directed graph Laplacian proposed by Chung has already encodes directed information, such as this form satisfying Cheeger inequality. Although the final form is symmetric, the eigenvector matrix for performing normilizaton varies with the directed graph structure.

Regards, Zekun

JhuoW commented 2 years ago

Thanks for your explanation.