1 paper
Andreas Roth, Thomas Liebig
Our study reveals new theoretical insights into over-smoothing and feature over-correlation in graph neural networks. Specifically, we demonstrate that with increased depth, node r…