| 注册
首页|期刊导航|郑州大学学报(工学版)|结构-特征协同防御的图神经网络

结构-特征协同防御的图神经网络

韩继辉 石玉鹏 黄子奇 张安琳 黄道颖

郑州大学学报(工学版)2026,Vol.47Issue(4):134-142,9.
✕
郑州大学学报(工学版)2026,Vol.47Issue(4):134-142,9.DOI:10.13705/j.issn.1671-6833.2026.04.010

结构-特征协同防御的图神经网络

A Graph Neural Network for Structure-Feature Collaborative Defense

韩继辉 1石玉鹏 1黄子奇 2张安琳 3黄道颖1

作者信息

  • 1. 郑州轻工业大学 计算机与人工智能学院,河南 郑州 450001
  • 2. 北方信息控制研究院集团有限公司,江苏 南京 211153
  • 3. 郑州轻工业大学 工程训练中心,河南 郑州 450001
  • 折叠

摘要

Abstract

To address the degradation of node representations in graph neural networks with complex perturbation environments,a structure-feature collaborative defense graph neural network named SFCoRobustGNN was pro-posed.Structurally,a sparse attention mechanism that integrated structure priors to dynamically suppress anomalous edges was introduced.Feature-wise,a channel gating mechanism was combined with a nonlinear feature mixing module(FeatureMixPro)to enhance the model's adaptability to feature perturbations.A collaborative dual-pathway defense was achieved through adversarial training and a multi-objective optimization strategy.Experiments on multi-ple benchmark datasets,including Cora and Citeseer,demonstrated that the proposed method outperformed most of mainstream baseline metods with various intensities of structure perturbations(5%—40%)and feature attacks(ε=0.01-0.10),showing significant improvement in node classification accuracy.On the large-scale ogbn-products dataset,it maintained an accuracy of 71.82%even with a 20%MetaAttack structure perturbation,demonstrating its strong scalability.Ablation studies validated the effectiveness and synergistic effects of each module.The pro-posed method effectively mitigated performance degradation with complex perturbations and exhibited excellent gen-eralization.

关键词

图神经网络/鲁棒性/结构扰动/特征扰动/稀疏注意力

Key words

graph neural networks/robustness/structure perturbation/feature perturbation/sparse attention

分类

信息技术与安全科学

引用本文复制引用

韩继辉,石玉鹏,黄子奇,张安琳,黄道颖..结构-特征协同防御的图神经网络[J].郑州大学学报(工学版),2026,47(4):134-142,9.

基金项目

河南省科技攻关项目(252102211089,232102210064) (252102211089,232102210064)

郑州大学学报(工学版)

1671-6833

访问量0
|
下载量0
段落导航相关论文