Short-Range Oversquashing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mishayev, Yaaqov, Sverdlov, Yonatan, Amir, Tal, Dym, Nadav
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915637489565696
author Mishayev, Yaaqov
Sverdlov, Yonatan
Amir, Tal
Dym, Nadav
author_facet Mishayev, Yaaqov
Sverdlov, Yonatan
Amir, Tal
Dym, Nadav
contents Message Passing Neural Networks (MPNNs) are widely used for learning on graphs, but their ability to process long-range information is limited by the phenomenon of oversquashing. This limitation has led some researchers to advocate Graph Transformers as a better alternative, whereas others suggest that it can be mitigated within the MPNN framework, using virtual nodes or other rewiring techniques. In this work, we demonstrate that oversquashing is not limited to long-range tasks, but can also arise in short-range problems. This observation allows us to disentangle two distinct mechanisms underlying oversquashing: (1) the bottleneck phenomenon, which can arise even in low-range settings, and (2) the vanishing gradient phenomenon, which is closely associated with long-range tasks. We further show that the short-range bottleneck effect is not captured by existing explanations for oversquashing, and that adding virtual nodes does not resolve it. In contrast, transformers do succeed in such tasks, positioning them as the more compelling solution to oversquashing, compared to specialized MPNNs.
format Preprint
id arxiv_https___arxiv_org_abs_2511_20406
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Short-Range Oversquashing
Mishayev, Yaaqov
Sverdlov, Yonatan
Amir, Tal
Dym, Nadav
Machine Learning
Artificial Intelligence
Message Passing Neural Networks (MPNNs) are widely used for learning on graphs, but their ability to process long-range information is limited by the phenomenon of oversquashing. This limitation has led some researchers to advocate Graph Transformers as a better alternative, whereas others suggest that it can be mitigated within the MPNN framework, using virtual nodes or other rewiring techniques. In this work, we demonstrate that oversquashing is not limited to long-range tasks, but can also arise in short-range problems. This observation allows us to disentangle two distinct mechanisms underlying oversquashing: (1) the bottleneck phenomenon, which can arise even in low-range settings, and (2) the vanishing gradient phenomenon, which is closely associated with long-range tasks. We further show that the short-range bottleneck effect is not captured by existing explanations for oversquashing, and that adding virtual nodes does not resolve it. In contrast, transformers do succeed in such tasks, positioning them as the more compelling solution to oversquashing, compared to specialized MPNNs.
title Short-Range Oversquashing
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.20406