arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2008.11416cs.LGstat.ML
收藏

抗扰动的节点表示学习

Learning Node Representations against Perturbations

Shanghai Jiao Tong University(上海交通大学);University of Technology Sydney(悉尼科技大学)

Xu Chen, Yuangang Pan, Ivor Tsang, Ya Zhang

AI总结:

针对图神经网络节点表示易受输入扰动影响的问题,提出兼具稳定性与可识别性的无监督抗扰动框架SIGNNAP,可适配多种GNN骨干,在六类基准的节点分类任务中验证了有效性。

摘要:

近年来,图神经网络(GNN)在节点表示学习中取得了显著性能,其成功的关键因素之一是节点表示的平滑性特性。尽管如此,大多数GNN模型对图输入的扰动较为脆弱,可能学习到不可靠的节点表示。本文研究了如何在GNN中学习抗扰动的节点表示。具体而言,我们认为节点表示在输入受到轻微扰动时应保持稳定,且来自不同结构的节点表示应具备可区分性,这两点分别被称为节点表示的稳定性和可识别性。为此,我们提出了一种名为Stability-Identifiability GNN Against Perturbations(SIGNNAP,抗扰动稳定性-可识别性图神经网络)的新型模型,以无监督方式学习可靠的节点表示。SIGNNAP通过对比目标将稳定性和可识别性形式化,并借助现有GNN骨干网络保留平滑性。该方法是一个通用框架,可适配多种其他骨干模型(如GCN、GraphSage和GAT)。在节点分类的直推式和归纳式两种学习设置下,我们在六个基准数据集上开展了大量实验,结果证明了该方法的有效性。代码和数据可在线获取:https://github.com/xuChenSJTU/SIGNNAP-master-online

英文摘要:

Recent graph neural networks (GNN) has achieved remarkable performance in node representation learning. One key factor of GNN's success is the \emph{smoothness} property on node representations. Despite this, most GNN models are fragile to the perturbations on graph inputs and could learn unreliable node representations. In this paper, we study how to learn node representations against perturbations in GNN. Specifically, we consider that a node representation should remain stable under slight perturbations on the input, and node representations from different structures should be identifiable, which two are termed as the \emph{stability} and \emph{identifiability} on node representations, respectively. To this end, we propose a novel model called Stability-Identifiability GNN Against Perturbations (SIGNNAP) that learns reliable node representations in an unsupervised manner. SIGNNAP formalizes the \emph{stability} and \emph{identifiability} by a contrastive objective and preserves the \emph{smoothness} with existing GNN backbones. The proposed method is a generic framework that can be equipped with many other backbone models (e.g. GCN, GraphSage and GAT). Extensive experiments on six benchmarks under both transductive and inductive learning setups of node classification demonstrate the effectiveness of our method. Codes and data are available online:~\url{https://github.com/xuChenSJTU/SIGNNAP-master-online}

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑