Unifying approach to uniform expressivity of graph neural networks

TL;DR

This paper introduces Template GNNs (T-GNNs), a unified framework that enhances GNN expressivity by aggregating over graph templates, linking their power to a new logic and unifying previous GNN variants.

Contribution

The paper formalizes T-GNNs, introduces GML(T) logic, and establishes their equivalence, providing a unifying analysis of GNN expressivity beyond traditional methods.

Findings

T-GNNs generalize existing GNN architectures.

A new logic GML(T) characterizes T-GNN expressivity.

Standard GNN variants are special cases of T-GNNs.

Abstract

The expressive power of Graph Neural Networks (GNNs) is often analysed via correspondence to the Weisfeiler-Leman (WL) algorithm and fragments of first-order logic. Standard GNNs are limited to performing aggregation over immediate neighbourhoods or over global read-outs. To increase their expressivity, recent attempts have been made to incorporate substructural information (e.g. cycle counts and subgraph properties). In this paper, we formalize this architectural trend by introducing Template GNNs (T-GNNs), a generalized framework where node features are updated by aggregating over valid template embeddings from a specified set of graph templates. We propose a corresponding logic, Graded template modal logic (GML(T)), and generalized notions of template-based bisimulation and WL algorithm. We establish an equivalence between the expressive power of T-GNNs and GML(T), and provide a unifying approach for analysing GNN expressivity: we show how standard AC-GNNs and its recent variants can be interpreted as instantiations of T-GNNs.