Certifiable Robustness and Robust Training for Graph Convolutional Networks
D. Zügner, and S. Günnemann. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining - KDD '19, (2019)arXiv: 1906.12269.
DOI: 10.1145/3292500.3330905
Abstract
Recent works show that Graph Neural Networks (GNNs) are highly non-robust with respect to adversarial attacks on both the graph structure and the node attributes, making their outcomes unreliable. We propose the first method for certifiable (non-)robustness of graph convolutional networks with respect to perturbations of the node attributes1. We consider the case of binary node attributes (e.g. bag-of-words) and perturbations that are L0-bounded. If a node has been certified with our method, it is guaranteed to be robust under any possible perturbation given the attack model. Likewise, we can certify non-robustness. Finally, we propose a robust semisupervised training procedure that treats the labeled and unlabeled nodes jointly. As shown in our experimental evaluation, our method significantly improves the robustness of the GNN with only minimal effect on the predictive accuracy.
Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining - KDD '19
pages
246--256
language
en
file
Zügner, Günnemann - Certifiable Robustness and Robust Training for Graph Convolutional Networks.pdf:C\:\\Users\\Admin\\Documents\\Research\\_Paperbase\\Graph Embeddings\\Zügner, Günnemann - Certifiable Robustness and Robust Training for Graph Convolutional Networks.pdf:application/pdf
%0 Journal Article
%1 zugner_certifiable_2019
%A Zügner, Daniel
%A Günnemann, Stephan
%D 2019
%J Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining - KDD '19
%K Adversarial_Attacks GNN Node_Embeddings
%P 246--256
%R 10.1145/3292500.3330905
%T Certifiable Robustness and Robust Training for Graph Convolutional Networks
%U http://arxiv.org/abs/1906.12269
%X Recent works show that Graph Neural Networks (GNNs) are highly non-robust with respect to adversarial attacks on both the graph structure and the node attributes, making their outcomes unreliable. We propose the first method for certifiable (non-)robustness of graph convolutional networks with respect to perturbations of the node attributes1. We consider the case of binary node attributes (e.g. bag-of-words) and perturbations that are L0-bounded. If a node has been certified with our method, it is guaranteed to be robust under any possible perturbation given the attack model. Likewise, we can certify non-robustness. Finally, we propose a robust semisupervised training procedure that treats the labeled and unlabeled nodes jointly. As shown in our experimental evaluation, our method significantly improves the robustness of the GNN with only minimal effect on the predictive accuracy.
@article{zugner_certifiable_2019,
abstract = {Recent works show that Graph Neural Networks (GNNs) are highly non-robust with respect to adversarial attacks on both the graph structure and the node attributes, making their outcomes unreliable. We propose the first method for certifiable (non-)robustness of graph convolutional networks with respect to perturbations of the node attributes1. We consider the case of binary node attributes (e.g. bag-of-words) and perturbations that are L0-bounded. If a node has been certified with our method, it is guaranteed to be robust under any possible perturbation given the attack model. Likewise, we can certify non-robustness. Finally, we propose a robust semisupervised training procedure that treats the labeled and unlabeled nodes jointly. As shown in our experimental evaluation, our method significantly improves the robustness of the GNN with only minimal effect on the predictive accuracy.},
added-at = {2020-02-21T16:09:44.000+0100},
author = {Zügner, Daniel and Günnemann, Stephan},
biburl = {https://www.bibsonomy.org/bibtex/2e1ec0a50a07eff93e44296ed65cf4c84/tschumacher},
doi = {10.1145/3292500.3330905},
file = {Zügner, Günnemann - Certifiable Robustness and Robust Training for Graph Convolutional Networks.pdf:C\:\\Users\\Admin\\Documents\\Research\\_Paperbase\\Graph Embeddings\\Zügner, Günnemann - Certifiable Robustness and Robust Training for Graph Convolutional Networks.pdf:application/pdf},
interhash = {baa3fd329376cbbf0b67c851d5fa89ba},
intrahash = {e1ec0a50a07eff93e44296ed65cf4c84},
journal = {Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining - KDD '19},
keywords = {Adversarial_Attacks GNN Node_Embeddings},
language = {en},
note = {arXiv: 1906.12269},
pages = {246--256},
timestamp = {2020-02-21T16:09:44.000+0100},
title = {Certifiable {Robustness} and {Robust} {Training} for {Graph} {Convolutional} {Networks}},
url = {http://arxiv.org/abs/1906.12269},
urldate = {2019-12-10},
year = 2019
}