Generalizable Insights for Graph Transformers in Theory and Practice

TL;DR

This paper introduces the Generalized-Distance Transformer, a versatile Graph Transformer architecture, and provides extensive empirical and theoretical insights into its design choices, demonstrating strong performance across diverse large-scale graph tasks.

Contribution

The paper presents the GDT architecture that unifies recent advancements in GTs and offers a comprehensive analysis of its expressivity and empirical performance across multiple domains.

Findings

Design choices that improve generalization across tasks

Strong few-shot transfer performance without fine-tuning

Empirical validation on over eight million graphs

Abstract

Graph Transformers (GTs) have shown strong empirical performance, yet current architectures vary widely in their use of attention mechanisms, positional embeddings (PEs), and expressivity. Existing expressivity results are often tied to specific design choices and lack comprehensive empirical validation on large-scale data. This leaves a gap between theory and practice, preventing generalizable insights that exceed particular application domains. Here, we propose the Generalized-Distance Transformer (GDT), a GT architecture using standard attention that incorporates many advancements for GTs from recent years, and develop a fine-grained understanding of the GDT's representation power in terms of attention and PEs. Through extensive experiments, we identify design choices that consistently perform well across various applications, tasks, and model scales, demonstrating strong performance in a few-shot transfer setting without fine-tuning. Our evaluation covers over eight million graphs with roughly 270M tokens across diverse domains, including image-based object detection, molecular property prediction, code summarization, and out-of-distribution algorithmic reasoning. We distill our theoretical and practical findings into several generalizable insights about effective GT design, training, and inference.