Adversarial Training for Graph Neural Networks: Pitfalls, Solutions, and New Directions

Lukas Gosch, Simon Geisler, Daniel Sturm, Bertrand Charpentier, Daniel Zügner +1 more
6/27/2023
cs.LG

No official code found. Best implementation: saper0/adversarial_training (13 stars, MIT, last commit 2023-12-08).

Abstract

Despite its success in the image domain, adversarial training did not (yet) stand out as an effective defense for Graph Neural Networks (GNNs) against graph structure perturbations. In the pursuit of fixing adversarial training (1) we show and overcome fundamental theoretical as well as practical limitations of the adopted graph learning setting in prior work; (2) we reveal that more flexible GNNs based on learnable graph diffusion are able to adjust to adversarial perturbations, while the learned message passing scheme is naturally interpretable; (3) we introduce the first attack for structure perturbations that, while targeting multiple nodes at once, is capable of handling global (graph-level) as well as local (node-level) constraints. Including these contributions, we demonstrate that adversarial training is a state-of-the-art defense against adversarial structure perturbations.

View on arXivView PDF

Code Implementations(1)

saper0/adversarial_trainingCommunity60%
MIT

Codebase used to generate the results for NeurIPS23 "Adversarial Training for Graph Neural Networks: Pitfalls, Solutions, and New Directions" paper.

133C++, CudaDec 8, 20232 years agoMIT

Similar Papers

Unifying Adversarial Perturbation for Graph Neural Networks

92% match2025

Adversarial Training for Graph Neural Networks via Graph Subspace Energy Optimization

89% match2024

Improving Graph Neural Network Training, Defense and Hypergraph Partitioning via Adversarial Robustness Evaluation

88% match2024

Relaxing Graph Transformers for Adversarial Attacks

88% match2024

Key Principles of Graph Machine Learning: Representation, Robustness, and Generalization

88% match2026

Disttack: Graph Adversarial Attacks Toward Distributed GNN Training

88% match1 repo22024

Cite this paper

@article{gosch2023adversarial,
  title  = {Adversarial Training for Graph Neural Networks: Pitfalls, Solutions, and New Directions},
  author = {Lukas Gosch and Simon Geisler and Daniel Sturm and Bertrand Charpentier and Daniel Zügner and Stephan Günnemann},
  year   = {2023},
  eprint = {2306.15427},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2306.15427v2},
  journal = {NEURIPS 2023 2023}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2306.15427.svg)](https://codeofpaper.com/paper/2306.15427)

Discussion