Intrinsic Defenses Against Backdoor Attacks in High-Order Graph Neural Networks via Semantic and Outlier-Guided Subgraph Policies
Bibliographic record
Abstract
Graph Neural Networks (GNNs) are increasingly used in sensitive areas such as healthcare and finance, where reliability and security are critical. However, their vulnerability to adversarial backdoor attacks creates serious risks. These attacks insert harmful triggers during training to force the model to produce incorrect, targeted outputs during inference. Although several defense strategies have been proposed, designing GNN systems that are naturally resistant to such attacks remains a significant challenge. In this work, we leverage the enhanced expressiveness of higher-order GNNs and a substructure learning approach to propose a new, robust system that includes two built-in defense mechanisms. The first is a substructure extraction method that uses cosine similarity to measure the semantic alignment between connected nodes. Edges between nodes that are semantically different are marked as potential triggers and are removed. As a result, the substructures are formed by more consistent neighborhoods and meaningful relationships, and the model becomes more resistant to backdoor attacks that violate graph homophily. The second mechanism is an outlier detection-based approach that uses clustering to identify dominant and cohesive substructures within the graph. Nodes that differ significantly from these core structures are marked as potential triggers. By isolating and filtering out such anomalies, the model can reduce the success rate of backdoor attacks that introduce unnatural or deceptive graph elements. We evaluate our approach on standard datasets and under various state-of-the-art attack scenarios. Results show a significant reduction in backdoor attack success rates while maintaining high clean accuracy. These findings demonstrate the potential of embedding structure-aware defense mechanisms directly into GNN systems. This project's source code is publicly available at https://github.com/AbhiJeet70/SPROUT_GNN and permanently archived with DOI: 10.5281/zenodo. 17095064.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.011 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".