Aggregation Buffer: Revisiting DropEdge with a New Parameter Block

TL;DR

This paper analyzes DropEdge's limitations in GNNs and introduces Aggregation Buffer, a new parameter block that enhances robustness and addresses structural issues across various datasets.

Contribution

We propose Aggregation Buffer, a novel parameter block that improves GNN robustness and overcomes DropEdge's limitations, compatible with any GNN architecture.

Findings

Consistent performance improvements on multiple datasets

Addresses degree bias and structural disparity issues

Theoretically explains DropEdge's limitations

Abstract

We revisit DropEdge, a data augmentation technique for GNNs which randomly removes edges to expose diverse graph structures during training. While being a promising approach to effectively reduce overfitting on specific connections in the graph, we observe that its potential performance gain in supervised learning tasks is significantly limited. To understand why, we provide a theoretical analysis showing that the limited performance of DropEdge comes from the fundamental limitation that exists in many GNN architectures. Based on this analysis, we propose Aggregation Buffer, a parameter block specifically designed to improve the robustness of GNNs by addressing the limitation of DropEdge. Our method is compatible with any GNN model, and shows consistent performance improvements on multiple datasets. Moreover, our method effectively addresses well-known problems such as degree bias or structural disparity as a unifying solution. Code and datasets are available at https://github.com/dooho00/agg-buffer.