Certified Robustness of Graph Classification against Topology Attack with Randomized Smoothing

Abstract

Graph classification has practical applications in diverse fields. Recent studies show that graph-based machine learning models are especially vulnerable to adversarial perturbations due to the non i.i.d nature of graph data. By adding or deleting a small number of edges in the graph, adversaries could greatly change the graph label predicted by a graph classification model. In this work, we propose to build a smoothed graph classification model with certified robustness guarantee. We have proven that the resulting graph classification model would output the same prediction for a graph under $l_0$ bounded adversarial perturbation. We also evaluate the effectiveness of our approach under graph convolutional network (GCN) based multi-class graph classification model.

arXiv commentAccepted to IEEE GLOBECOM 2020

References in corpus (12)

• Inductive Representation Learning on Large Graphs
• Adversarial Attacks on Neural Networks for Graph Data
• Certified Adversarial Robustness via Randomized Smoothing
• The Space of Transferable Adversarial Examples
• Security Evaluation of Pattern Classifiers under Attack
• Adversarial Attacks on Graph Neural Networks via Meta Learning
• Certified Adversarial Robustness with Additive Noise
• Adversarial Examples on Graph Data: Deep Insights into Attack and Defense
• Certifiable Robustness to Graph Perturbations
• A Unified Framework for Data Poisoning Attack to Graph-based Semi-supervised Learning
• Certified Robustness of Community Detection against Adversarial Structural Perturbation via Randomized Smoothing
• Graph Attentional Autoencoder for Anticancer Hyperfood Prediction