Over-Smoothing Effect of Graph Convolutional Networks: Spectral and Topological Analysis with Practical Remedies
Author: Fang Sun
Venue: IEEE SoutheastCon 2026, Track 6 - AI and Predictive Modeling
Abstract: Graph convolutional networks (GCNs) are effective for node classification but their depth is limited by the over-smoothing effect, where node representations converge as layers increase. This work provides spectral and topological analysis of this phenomenon along with practical remedies.