Work

Explaining GNN Explanations with Edge Gradients

Year: 2025

Type: conference-paper

Abstract: In recent years, the remarkable success of graph neural networks (GNNs) on graph-structured data has prompted a surge of methods for explaining GNN predictions. However, the state-of-the-art for GNN explainability remains in flux. Different comparisons find mixed results for different methods, with many explainers struggling on more complex GNN architectures and tasks. This presents an urgent need for a more careful theoretical analysis of competing GNN explanation methods. In this work we take a closer look at GNN explanations in two different settings: input-level explanations, which produce explanatory subgraphs of the input graph, and layerwise explanations, which produce explanatory subgraphs of the computation graph. We establish the first theoretical connections between the popular perturbation-based and classical gradient-based methods, as well as point out connections between other recently proposed methods. At the input level, we demonstrate conditions under which GNNExplainer can be approximated by a simple heuristic based on the sign of the edge gradients. In the layerwise setting, we point out that edge gradients are equivalent to occlusion search for linear GNNs. Finally, we demonstrate how our theoretical results manifest in practice with experiments on both synthetic and real datasets.

Source: ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

DOI: 10.1145/3711896.3736947

Authors: Jesse He, Akbar Rafiey, Gal Mishne, Yusu Wang

Institutions: University of California San Diego, New York University

Language: English

Topic: Explainable Artificial Intelligence (XAI)

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funders: National Science Foundation, Universitas Brawijaya

Awards: G2081960379, Collaborative Research: AF: Small: Graph Analysis: Integrating Metric and Topological Perspectives, AI Institute for Learning-Enabled Optimization at Scale (TILOS), Collaborative Research: CIF: Medium:Data Manifolds and Local Geometry in Deep Networks, Collaborative Research: EnCORE: Institute for Emerging CORE Methods in Data Science