Towards Mechanistic Interpretability of Graph Transformers via Attention Graphs

TL;DR

This paper introduces Attention Graphs, a tool for understanding how Graph Transformers process information by analyzing attention patterns, revealing that learned attention structures often differ from original graphs and vary across models.

Contribution

It presents Attention Graphs as a novel interpretability method linking message passing in GNNs to self-attention in Transformers, with extensive analysis on different graph types.

Findings

Attention Graphs reveal diverse information flow patterns.

Learned attention structures often do not match original graph structures.

Different Transformer variants can achieve similar performance with distinct attention patterns.

Abstract

We introduce Attention Graphs, a new tool for mechanistic interpretability of Graph Neural Networks (GNNs) and Graph Transformers based on the mathematical equivalence between message passing in GNNs and the self-attention mechanism in Transformers. Attention Graphs aggregate attention matrices across Transformer layers and heads to describe how information flows among input nodes. Through experiments on homophilous and heterophilous node classification tasks, we analyze Attention Graphs from a network science perspective and find that: (1) When Graph Transformers are allowed to learn the optimal graph structure using all-to-all attention among input nodes, the Attention Graphs learned by the model do not tend to correlate with the input/original graph structure; and (2) For heterophilous graphs, different Graph Transformer variants can achieve similar performance while utilising distinct information flow patterns. Open source code: https://github.com/batu-el/understanding-inductive-biases-of-gnns