Graph convolutions that can finally model local structure

TL;DR

This paper introduces a simple modification to graph neural networks that significantly improves their ability to detect local structures like small cycles, enhancing performance on chemistry-related tasks.

Contribution

A straightforward correction to GIN convolution that enables better detection of local graph structures with minimal computational overhead.

Findings

Improved detection of small cycles in graphs.

Enhanced performance on molecule property prediction datasets.

Consistent gains over baseline models across multiple tasks.

Abstract

Despite quick progress in the last few years, recent studies have shown that modern graph neural networks can still fail at very simple tasks, like detecting small cycles. This hints at the fact that current networks fail to catch information about the local structure, which is problematic if the downstream task heavily relies on graph substructure analysis, as in the context of chemistry. We propose a very simple correction to the now standard GIN convolution that enables the network to detect small cycles with nearly no cost in terms of computation time and number of parameters. Tested on real life molecule property datasets, our model consistently improves performance on large multi-tasked datasets over all baselines, both globally and on a per-task setting.