- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The authors compare triggers produced by existing graph backdoor attacks and find that nodes in corresponding positions often have unusually similar features. They use this observation, together with local attachment patterns, to build MTFD. Their attack, FRGBA, constructs each trigger from the victim node's features and limits similarity across triggers while preserving similarity to the victim. On four citation networks, tests with GCN, GraphSAGE, and GAT show that MTFD suppresses the baseline attacks, whereas FRGBA retains average attack success rates above 90% with clean accuracy close to the corresponding clean models.
Critical strengths
- Figure 2 identifies unusually high similarity between corresponding nodes across triggers, and Tables 2 and 7 show that MTFD can use this pattern to detect or suppress the evaluated attacks.
- FRGBA retains 90.06%-96.70% average ASR under MTFD across four datasets and three GNNs, while the ablation in Figure 5 supports the contribution of its cross-trigger similarity constraint.
Critical weaknesses
- FRGBA's single-node triggers still satisfy CN(v,t)=0 and Exit(v,t)=0, so the reported results do not establish that it reduces the structural footprints defined in Section 3.2.
- Missing baseline ASR and CA in Tables 3 and 6, unclear handling of trigger sizes and label permissions, and unspecified MTFD thresholds and loss-weight updates make the comparisons difficult to assess, especially given the drop in ASR from 94.55% to 48.83% when p_t changes in Table 10.
Questions for authors
- What fraction of FRGBA triggers is detected at each stage of MTFD, given that they still satisfy CN=0 and Exit=0?
- How do the baseline attacks compare under the defenses in Table 3 when trigger sizes, poisoning budgets, and label permissions are explicitly accounted for?
- Which data and rules were used to set MTFD's thresholds, p_v, p_t, and loss weights, and what are the standard deviations of the main results?
- How does removing victim-feature fusion alone affect ASR and detection when normalization and both similarity constraints are retained?
6 comments
We thank the reviewer for raising this concern regarding the fairness of the experimental comparisons. To provide a more comprehensive evaluation of FRGBA's robustness against existing defenses, we conducted additional experiments with GTA, UGBA, DPGBA, and CNGBA under five defense methods: Prune, Prune+LD, OD, RIGBD, and LoSplit.
Regarding the experimental settings, to control for the effects of trigger size and poisoning budget, we followed the same setup as Table 2, using single-node triggers and a uniform poisoning budget of 1% for all attack methods. Regarding label-modification permissions, we retained the original label settings of each attack method to preserve their respective attack configurations and avoid altering the characteristics of the baseline methods through additional label constraints.
Table: Attack performance under five existing defenses on Cora (ASR (%) / CA (%)).
| Attack | Prune | Prune+LD | OD | RIGBD | LoSplit |
|---|---|---|---|---|---|
| GTA | 25.20 / 72.15 | 39.52 / 57.43 | 76.35 / 81.97 | 57.81 / 82.84 | 36.80 / 83.21 |
| UGBA | 86.92 / 84.44 | 88.89 / 76.30 | 64.71 / 85.93 | 75.86 / 82.96 | 58.33 / 84.07 |
| DPGBA | 45.31 / 82.22 | 32.29 / 70.06 | 81.08 / 83.46 | 42.70 / 81.60 | 39.93 / 83.33 |
| CNGBA | 89.76 / 86.05 | 89.05 / 86.05 | 76.78 / 83.02 | 100.00 / 85.80 | 1.79 / 84.94 |
| FRGBA | 96.53 / 82.10 | 96.15 / 79.14 | 91.32 / 83.08 | 96.18 / 82.59 | 94.45 / 83.70 |
The additional results on Cora demonstrate that FRGBA maintains strong attack effectiveness under existing defenses. Across the 20 attack-defense combinations involving the four baseline attacks, FRGBA achieves a higher ASR in 19 cases. Specifically, under Prune, Prune+LD, OD, and LoSplit, FRGBA outperforms the best-performing baseline by 6.77, 7.10, 10.24, and 36.12 percentage points, respectively. The only exception is under RIGBD, where FRGBA achieves a slightly lower ASR than CNGBA. These additional results further support the relative robustness of FRGBA under different defense settings.
We thank the reviewer for the questions regarding parameter selection. All similarity thresholds in our method are computed based on the corresponding reference feature similarity distributions, while the associated percentile parameters are selected based on preliminary experimental observations and empirical analysis.
Specifically, for MTFD, our analysis in Section 3.3 reveals abnormally high feature similarities among trigger nodes across different triggers, which differ noticeably from the similarity distribution of clean-node pairs. Based on this observation, MTFD adopts the p-th percentile of the clean-node pairwise similarity distribution as the similarity detection threshold. In addition, each candidate node is required to have at least M high-similarity matches to reduce false positives caused by incidental similarities among benign nodes.
To further examine cross-trigger similarity across datasets, we report three statistics: CC 99th, the 99th percentile of pairwise cosine similarities between clean-node pairs; TT 1st, the 1st percentile of similarities between trigger-node pairs at the same structural positions across different triggers; and TT above CC 99th, the proportion of such pairs exceeding CC 99th.
Table 1: Cross-trigger feature similarity statistics across four datasets.
| Dataset | Attack | CC 99th | TT 1st | TT above CC 99th |
|---|---|---|---|---|
| Cora | GTA | 0.5968 | 0.9799 | 1.0000 |
| UGBA | 0.4607 | 0.8286 | ||
| DPGBA | 0.9999 | 1.0000 | ||
| CNGBA | 0.9130 | 1.0000 | ||
| FRGBA | 0.1910 | 0.0000 | ||
| CiteSeer | GTA | 0.6765 | 0.9686 | 1.0000 |
| UGBA | 0.9999 | 1.0000 | ||
| DPGBA | 0.9819 | 1.0000 | ||
| CNGBA | 0.8571 | 1.0000 | ||
| FRGBA | 0.2792 | 0.0000 | ||
| PubMed | GTA | 0.7196 | 0.9863 | 1.0000 |
| UGBA | 0.3021 | 0.7173 | ||
| DPGBA | 1.0000 | 1.0000 | ||
| CNGBA | 0.9337 | 1.0000 | ||
| FRGBA | 0.0448 | 0.0000 | ||
| ogbn-arxiv | GTA | 0.9498 | 0.9996 | 1.0000 |
| UGBA | 0.9295 | 0.9815 | ||
| DPGBA | 1.0000 | 1.0000 | ||
| CNGBA | 0.9983 | 1.0000 | ||
| FRGBA | 0.6806 | 0.0013 |
For FRGBA, the victim–trigger similarity lower bound and the cross-trigger similarity upper bound are computed based on the feature similarity distributions of real edges between clean nodes and pairwise similarities among clean reference nodes, respectively. The percentile parameters and are empirically selected through preliminary experiments to balance attack effectiveness and the strength of the similarity constraints. The attack-loss weight is kept fixed, while the two similarity-constraint weights are dynamically adjusted according to their constraint violations to balance attack effectiveness and trigger stealthiness.
We thank the reviewer for the concern regarding the reliability of the experimental results. Table 2 reports the average attack performance of FRGBA across three target GNN architectures (GCN, GraphSAGE, and GAT), while Table 5 in Appendix H.1 provides the corresponding results for each architecture to illustrate performance variations across different models. The original experiments were conducted over five independent runs. To further evaluate the stability of the results, we conducted five new independent runs under the same experimental settings and report the mean and standard deviation of ASR and CA for FRGBA on each of the three target architectures. These results provide further insight into FRGBA's performance and run-to-run variability across different model architectures.
Table 2: FRGBA performance over five independent runs (mean ± standard deviation; ASR (%) and CA (%)).
| Dataset | Defense | Metric | GCN | GraphSAGE | GAT |
|---|---|---|---|---|---|
| Cora | None | ASR | 99.65 ± 0.30 | 89.58 ± 1.04 | 98.96 ± 0.52 |
| CA | 83.45 ± 0.43 | 82.34 ± 0.85 | 84.20 ± 1.75 | ||
| MTFD | ASR | 99.48 ± 0.00 | 87.85 ± 1.50 | 97.57 ± 1.31 | |
| CA | 83.21 ± 0.21 | 83.21 ± 0.77 | 83.82 ± 0.43 | ||
| CiteSeer | None | ASR | 99.75 ± 0.22 | 89.48 ± 1.54 | 99.87 ± 0.22 |
| CA | 74.80 ± 0.35 | 71.39 ± 0.31 | 74.60 ± 1.06 | ||
| MTFD | ASR | 99.62 ± 0.00 | 89.22 ± 1.16 | 100.00 ± 0.00 | |
| CA | 74.90 ± 0.35 | 71.19 ± 0.63 | 74.10 ± 0.30 | ||
| PubMed | None | ASR | 97.94 ± 0.56 | 89.51 ± 0.46 | 96.85 ± 0.27 |
| CA | 85.20 ± 0.08 | 85.91 ± 0.11 | 84.02 ± 0.13 | ||
| MTFD | ASR | 97.00 ± 0.52 | 88.43 ± 0.57 | 96.27 ± 0.65 | |
| CA | 84.91 ± 0.08 | 85.80 ± 0.31 | 83.80 ± 0.21 | ||
| ogbn-arxiv | None | ASR | 90.41 ± 0.46 | 80.14 ± 0.69 | 99.51 ± 0.07 |
| CA | 65.03 ± 0.37 | 65.91 ± 0.42 | 64.85 ± 0.09 | ||
| MTFD | ASR | 90.28 ± 0.03 | 80.24 ± 0.42 | 99.57 ± 0.04 | |
| CA | 65.53 ± 0.08 | 65.84 ± 0.09 | 65.03 ± 0.07 |
We thank the reviewer for this question. We agree that the single-node leaf triggers adopted by FRGBA still satisfy both CN = 0 and Exit = 0, and therefore cannot evade the first-stage structural screening of MTFD. It is important to note that the first stage primarily identifies structural candidates and does not imply that these candidates will ultimately be confirmed as triggers.
One possible way to alter these structural conditions would be to introduce additional edges between trigger nodes and other nodes in the original graph. However, such modifications would increase the number of injected edges and introduce greater perturbations to the original graph topology. Therefore, FRGBA adopts a minimal single-node trigger to limit structural modifications, rather than introducing additional connections solely to evade the first-stage screening.
By introducing a cross-trigger similarity constraint, FRGBA reduces abnormally high feature similarities between corresponding nodes across different triggers. Consequently, although FRGBA triggers enter the first-stage candidate set, they are less likely to be confirmed during the second-stage similarity verification. Thus, FRGBA's resistance to MTFD primarily comes from weakening similarity-based detection in the second stage, rather than evading structural screening in the first stage. We have revised the manuscript to clarify that the single-node trigger design aims to limit structural perturbations rather than evade structural detection.
To quantitatively demonstrate this mechanism, we provide additional stage-wise detection results for FRGBA across four datasets. Specifically, we report the proportion of true triggers included in the first-stage candidate set and the proportion finally identified after second-stage verification. Both proportions use the total number of injected triggers as the denominator.
Table: Stage-wise detection of FRGBA triggers by MTFD (identified triggers / total triggers).