Citation notice #6714 · 2026-07-11 03:19:17.190701+00:00

Identifying Backdoored Graphs in Graph Neural Network Training: An Explanation-Based Approach with Novel Metrics

Correction Crossref Open

cites Graph Neural Networks: A Review of Methods and Applications,, which carries a correction notice dated 2024-01-09. One-hop deterministic notice: the citation edge exists in the Pith bibliography graph; no model judged whether the citation was load-bearing.

This is not a judgment on the citing paper.

Citing paper Event page Original DOI Notice DOI File a formal challenge All reference changes

01Evidence

Raw extraction · bibliography line · bibliography index 50

Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C., Sun, M.: Graph neural networks: A review of methods and applications. AI Open1, 57– 81 (2020). https://doi.org/https://doi.org/10.1016/j.aiopen.2021.01.001, https://www.sciencedirect.com/science/article/pii/S2666651021000012 1 Explanation-Based Identification of Backdoored Training Graphs 1 Appendix A Backdoor Detection Results of Various Explainers As stated in our main paper, GNNExplainer fails as a method for reverse- engineering backdoor triggers. To test whether this issue is restricted to GN- NExplainer, we also explored the effectiveness of two other explainers – PGEx- plainer [25], known for its parameterized probabilistic graphical model approach in interpreting complex machine learning models, and CaptumExplainer [19], recognized for its comprehensive suite of neural network interpretability tools, including advanced algorithms like Integrated Gradients and Deconvolution. Fig. S1: An example of a backdoored sample from each dataset, after applying the mask generated by CaptumExplainer (top) and PGExplainer (bottom). These expla- nations were generated using the same hyperparameters as in Figure 2.

Parser render (TeX stripped for reading; raw above is the evidence)

Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C., Sun, M.: Graph neural networks: A review of methods and applications. AI Open1, 57– 81 (2020). https://doi.org/https://doi.org/10.1016/j.aiopen.2021.01.001, https://www.sciencedirect.com/science/article/pii/S2666651021000012 1 Explanation-Based Identification of Backdoored Training Graphs 1 Appendix A Backdoor Detection Results of Various Explainers As stated in our main paper, GNNExplainer fails as a method for reverse- engineering backdoor triggers. To test whether this issue is restricted to GN- NExplainer, we also explored the effectiveness of two other explainers – PGEx- plainer [25], known for its parameterized probabilistic graphical model approach in interpreting complex machine learning models, and CaptumExplainer [19], recognized for its comprehensive suite of neural network interpretability tools, including advanced algorithms like Integrated Gradients and Deconvolution. Fig. S1: An example of a backdoored sample from each dataset, after applying the mask generated by CaptumExplainer (top) and PGExplainer (bottom). These expla- nations were generated using the same hyperparameters as in Figure 2

02Event

Type: Correction
Source: Crossref
Original DOI: 10.1016/j.aiopen.2021.01.001
Notice DOI: 10.1016/j.aiopen.2024.01.002
Date: 2024-01-09
Title: Erratum regarding Declaration of Competing Interest statements in previously published articles
Reasons: ['Erratum']
Work: Graph Neural Networks: A Review of Methods and Applications, (2020) AI Open

Schema constants (for re-runners): correction · crossref

03Dispute this notice

If this citation does not depend on the flagged claim, or the event is wrong, say so. Disputes are public. For a signed challenge against the paper itself, use the formal challenge form.