可以最终模拟本地结构的图卷积

Nov, 2020

Graph convolutions that can finally model local structure

Rémy Brossard, Oriel Frigo, David Dehaene

TL;DR我们提出了一种很简单的方法来修改GIN卷积，以便使网络可以检测到很小的环，这对于在化学上下文中对图形子结构进行分析的下游任务非常重要。在真实的分子属性数据集上进行测试，我们的模型始终改善了大型多任务数据集的性能，无论是在全局还是在每个任务的设置中。

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 netw