Adversarial Defense Framework for Graph Neural Network

TL;DR

This paper introduces DefNet, a novel adversarial defense framework for GNNs that enhances robustness by identifying vulnerabilities and employing contrastive learning, significantly improving resistance to attacks across multiple datasets.

Contribution

The paper presents a new defense framework, DefNet, with strategies like dual-stage aggregation and contrastive learning, addressing GNN vulnerabilities and improving robustness against adversarial attacks.

Findings

DefNet improves GNN robustness against various attacks.

Effective across multiple GNN architectures and datasets.

Enhances security of graph-based learning models.

Abstract

Graph neural network (GNN), as a powerful representation learning model on graph data, attracts much attention across various disciplines. However, recent studies show that GNN is vulnerable to adversarial attacks. How to make GNN more robust? What are the key vulnerabilities in GNN? How to address the vulnerabilities and defense GNN against the adversarial attacks? In this paper, we propose DefNet, an effective adversarial defense framework for GNNs. In particular, we first investigate the latent vulnerabilities in every layer of GNNs and propose corresponding strategies including dual-stage aggregation and bottleneck perceptron. Then, to cope with the scarcity of training data, we propose an adversarial contrastive learning method to train the GNN in a conditional GAN manner by leveraging the high-level graph representation. Extensive experiments on three public datasets demonstrate the effectiveness of DefNet in improving the robustness of popular GNN variants, such as Graph Convolutional Network and GraphSAGE, under various types of adversarial attacks.