Certifiable robustness and robust training for graph convolutional networks

Daniel Zügner; Stephan Günnemann

doi:10.1145/3292500.3330905

Certifiable robustness and robust training for graph convolutional networks

Daniel Zügner, Stephan Günnemann

Informatics 26 - Associate Professorship of Data Analytics and Machine Learning

Technical University of Munich

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

109 Scopus citations

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 semi-supervised 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.

Original language	English
Title of host publication	KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
Publisher	Association for Computing Machinery
Pages	246-256
Number of pages	11
ISBN (Electronic)	9781450362016
DOIs	https://doi.org/10.1145/3292500.3330905
State	Published - 25 Jul 2019
Event	25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019 - Anchorage, United States Duration: 4 Aug 2019 → 8 Aug 2019

Publication series

Name	Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Conference

Conference	25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019
Country/Territory	United States
City	Anchorage
Period	4/08/19 → 8/08/19

Access to Document

10.1145/3292500.3330905

Cite this

Zügner, D., & Günnemann, S. (2019). Certifiable robustness and robust training for graph convolutional networks. In KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 246-256). (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). Association for Computing Machinery. https://doi.org/10.1145/3292500.3330905

Zügner, Daniel ; Günnemann, Stephan. / Certifiable robustness and robust training for graph convolutional networks. KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2019. pp. 246-256 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining).

@inproceedings{b18f9e4200ca423186812a7acf013bfd,

title = "Certifiable robustness and robust training for graph convolutional networks",

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 semi-supervised 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.",

author = "Daniel Z{\"u}gner and Stephan G{\"u}nnemann",

note = "Publisher Copyright: {\textcopyright} 2019 Copyright held by the owner/author(s).; 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019 ; Conference date: 04-08-2019 Through 08-08-2019",

year = "2019",

month = jul,

day = "25",

doi = "10.1145/3292500.3330905",

language = "English",

series = "Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",

publisher = "Association for Computing Machinery",

pages = "246--256",

booktitle = "KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",

}

Zügner, D & Günnemann, S 2019, Certifiable robustness and robust training for graph convolutional networks. in KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Association for Computing Machinery, pp. 246-256, 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019, Anchorage, United States, 4/08/19. https://doi.org/10.1145/3292500.3330905

Certifiable robustness and robust training for graph convolutional networks. / Zügner, Daniel; Günnemann, Stephan.
KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2019. p. 246-256 (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Certifiable robustness and robust training for graph convolutional networks

AU - Zügner, Daniel

AU - Günnemann, Stephan

PY - 2019/7/25

Y1 - 2019/7/25

N2 - 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 semi-supervised 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.

AB - 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 semi-supervised 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.

UR - http://www.scopus.com/inward/record.url?scp=85071157650&partnerID=8YFLogxK

U2 - 10.1145/3292500.3330905

DO - 10.1145/3292500.3330905

M3 - Conference contribution

AN - SCOPUS:85071157650

T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

SP - 246

EP - 256

BT - KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

PB - Association for Computing Machinery

T2 - 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2019

Y2 - 4 August 2019 through 8 August 2019

ER -

Zügner D, Günnemann S. Certifiable robustness and robust training for graph convolutional networks. In KDD 2019 - Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery. 2019. p. 246-256. (Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining). doi: 10.1145/3292500.3330905

Certifiable robustness and robust training for graph convolutional networks

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this