Research map / Graph learning / GNN architectures
Adversarial Training for Graph Neural Networks: Pitfalls, Solutions, and New Directions
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add adversarial training (LR-BCD) with self-training to GPRGNN (ablation) | GPRGNN · Citeseer, LR-BCD eps=0.25 | robust accuracy: +18.4 -> +24.8 (pp vs GCN) | Helped |
| use learnable-diffusion GPRGNN instead of GCN under adversarial training (ablation) | GCN · Citeseer, LR-BCD eps=0.25 | robust accuracy: +10.9 -> +24.8 (pp vs GCN) | Helped |
| use GPRGNN diffusion instead of GAT under adversarial training (ablation) | GAT · Citeseer, LR-BCD eps=0.25 | robust accuracy: +8.4 -> +24.8 | Helped |
| adversarial training (robust diffusion) vs SoftMedian GDC | SoftMedian GDC · Citeseer, LR-BCD eps=0.25 | robust accuracy: +16.2 -> +24.8 | Helped |
| adversarial training with robust diffusion on clean accuracy | GPRGNN · Citeseer | clean accuracy: +1.7 pp vs GCN | No clear effect |
| adversarial training without self-training in transductive setting (ablation) | GCN · Cora transductive | robust accuracy | Hurt |
| add adversarial training on top of self-training in transductive setting (ablation) | GCN · Cora transductive | robust accuracy | No clear effect |
| adversarial training with robust diffusion across datasets | GPRGNN · Cora ML, Cora, Pubmed, SBM, Squirrel, WikiCS | robust accuracy eps=0.25: 57.0 -> 74.4 (Cora ML) | Helped |
| adversarially train with local constraints (LR-BCD) instead of PR-BCD (ablation) | GPRGNN · Citeseer, LR-BCD eval eps=0.25 | robust accuracy: +24.8 vs +23.2 (LR-BCD eval); +22.9 vs +26.3 (PR-BCD eval) | Mixed |
| train without local constraints (PR-BCD) (ablation) | GPRGNN · Karate Club / spectral analysis | learned filter | Hurt |
| use transductive memorization defense | any GNN · transductive node classification | robustness | Helped |
Closest papers by meaning
- Boosting the Adversarial Robustness of Graph Neural Networks: An OOD Perspective (ICLR 2024 · similarity 0.92)
- Value at Adversarial Risk: A Graph Defense Strategy against Cost-Aware Attacks (AAAI 2024 · similarity 0.92)
- If You Want to Be Robust, Be Wary of Initialization (NeurIPS 2024 · similarity 0.91)
- On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (NeurIPS 2025 · similarity 0.90)
- A Simple and Yet Fairly Effective Defense for Graph Neural Networks (AAAI 2024 · similarity 0.90)
- Adversarial Robust Generalization of Graph Neural Networks (ICML 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.