On Generalization of Graph Autoencoders with Adversarial Training

TL;DR

This paper investigates whether adversarial training enhances the generalization of graph autoencoders, demonstrating through extensive experiments that L1 and L2 adversarial training improve performance in link prediction, node clustering, and anomaly detection.

Contribution

It formulates L1 and L2 adversarial training methods for GAE and VGAE, and empirically shows their effectiveness across multiple graph tasks.

Findings

Adversarial training improves graph autoencoder generalization.

L1 and L2 adversarial training boost performance in key tasks.

Enhanced robustness against adversarial perturbations.

Abstract

Adversarial training is an approach for increasing model's resilience against adversarial perturbations. Such approaches have been demonstrated to result in models with feature representations that generalize better. However, limited works have been done on adversarial training of models on graph data. In this paper, we raise such a question { does adversarial training improve the generalization of graph representations. We formulate L2 and L1 versions of adversarial training in two powerful node embedding methods: graph autoencoder (GAE) and variational graph autoencoder (VGAE). We conduct extensive experiments on three main applications, i.e. link prediction, node clustering, graph anomaly detection of GAE and VGAE, and demonstrate that both L2 and L1 adversarial training boost the generalization of GAE and VGAE.