Relational Attention: Generalizing Transformers for Graph-Structured Tasks

TL;DR

This paper introduces Relational Attention, a generalization of transformers that incorporates edge information for improved reasoning over graph-structured data, outperforming existing graph neural networks on various tasks.

Contribution

The paper proposes a novel relational transformer that updates edge vectors, enhancing the model's ability to reason over graphs compared to traditional set-based transformers.

Findings

Relational transformer significantly outperforms state-of-the-art graph neural networks.

It demonstrates superior performance on the CLRS Algorithmic Reasoning Benchmark.

The approach leverages the greater expressivity of graphs over sets.

Abstract

Transformers flexibly operate over sets of real-valued vectors representing task-specific entities and their attributes, where each vector might encode one word-piece token and its position in a sequence, or some piece of information that carries no position at all. But as set processors, transformers are at a disadvantage in reasoning over more general graph-structured data where nodes represent entities and edges represent relations between entities. To address this shortcoming, we generalize transformer attention to consider and update edge vectors in each transformer layer. We evaluate this relational transformer on a diverse array of graph-structured tasks, including the large and challenging CLRS Algorithmic Reasoning Benchmark. There, it dramatically outperforms state-of-the-art graph neural networks expressly designed to reason over graph-structured data. Our analysis demonstrates that these gains are attributable to relational attention's inherent ability to leverage the greater expressivity of graphs over sets.