Global Minima, Recoverability Thresholds, and Higher-Order Structure in GNNS
This work provides theoretical and empirical insights into GNN behavior for researchers in graph machine learning, though it is incremental in nature.
The paper analyzes Graph Neural Network (GNN) performance using random graph theory, showing that heavy-tailed degree distributions enhance GNNs and that SAGE and Graph Transformer perform well on noisy edge data, but no architecture handles noisy feature data effectively.
We analyze the performance of graph neural network (GNN) architectures from the perspective of random graph theory. Our approach promises to complement existing lenses on GNN analysis, such as combinatorial expressive power and worst-case adversarial analysis, by connecting the performance of GNNs to typical-case properties of the training data. First, we theoretically characterize the nodewise accuracy of one- and two-layer GCNs relative to the contextual stochastic block model (cSBM) and related models. We additionally prove that GCNs cannot beat linear models under certain circumstances. Second, we numerically map the recoverability thresholds, in terms of accuracy, of four diverse GNN architectures (GCN, GAT, SAGE, and Graph Transformer) under a variety of assumptions about the data. Sample results of this second analysis include: heavy-tailed degree distributions enhance GNN performance, GNNs can work well on strongly heterophilous graphs, and SAGE and Graph Transformer can perform well on arbitrarily noisy edge data, but no architecture handled sufficiently noisy feature data well. Finally, we show how both specific higher-order structures in synthetic data and the mix of empirical structures in real data have dramatic effects (usually negative) on GNN performance.