Revisiting Neighborhood Aggregation in Graph Neural Networks for Node Classification using Statistical Signal Processing

TL;DR

This paper critically examines neighborhood aggregation in GNNs for node classification, revealing flaws in existing models under common assumptions and proposing a signal processing perspective for improved design.

Contribution

It introduces a statistical signal processing approach to analyze neighborhood aggregation, highlighting conceptual flaws in benchmark GNN models under certain assumptions.

Findings

Identifies flaws in current GNN models with edge-independent labels

Provides a new perspective for designing more efficient GNNs

Offers insights into the limitations of existing benchmark models

Abstract

We delve into the issue of node classification within graphs, specifically reevaluating the concept of neighborhood aggregation, which is a fundamental component in graph neural networks (GNNs). Our analysis reveals conceptual flaws within certain benchmark GNN models when operating under the assumption of edge-independent node labels, a condition commonly observed in benchmark graphs employed for node classification. Approaching neighborhood aggregation from a statistical signal processing perspective, our investigation provides novel insights which may be used to design more efficient GNN models.