Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarially Robust Neural Architecture Search for Graph Neural Networks
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 |
|---|---|---|---|
| search architectures using the robustness metric rather than omitting it (ablation) | G-RNA · Cora, Mettack, 25% changed edges | accuracy: 41.20±1.53 -> 67.23±1.66 | Helped |
| search architectures using the robustness metric rather than omitting it (ablation) | G-RNA · CiteSeer, Mettack, 25% changed edges | accuracy: 43.76±2.23 -> 61.33±1.35 | Helped |
| search architectures using the robustness metric rather than omitting it (ablation) | G-RNA · PubMed, Mettack, 25% changed edges | accuracy: 73.60±0.25 -> 85.82±0.12 | Helped |
| use graph structure mask operations, including NFS, in searched architectures (ablation) | G-RNA · Cora, Mettack, 5% perturbation | accuracy decrease | Helped |
| use GIN or SAGE to distinguish self-messages from neighbor-messages (ablation) | G-RNA architectures · Cora, Mettack, 5% perturbation | accuracy decrease | Helped |
| add skip connections (ablation) | G-RNA architectures · Cora, Mettack, 5% perturbation | accuracy decrease | Helped |
| increase the number of message-passing layers (ablation) | G-RNA architectures · Cora, Mettack, 5% perturbation | accuracy decrease | Hurt |
| use LSTM rather than other layer aggregation operations (ablation) | G-RNA architectures · Cora, Mettack, 5% perturbation | accuracy decrease | No clear effect |
| use G-RNA rather than GCN-Jaccard | node classification · Cora, Mettack, 25% changed edges | accuracy: 56.14±1.45 -> 67.23±1.66 | Helped |
| use G-RNA rather than Pro-GNN | node classification · CiteSeer, Mettack, 25% changed edges | accuracy: 55.73±2.04 -> 61.33±1.35 | Helped |
| use G-RNA rather than GASSO | node classification · PubMed, Mettack, 25% changed edges | accuracy: 83.25±1.26 -> 85.82±0.12 | Helped |
| use G-RNA rather than GCN-NoGraph | node classification · Cora and CiteSeer under structural perturbations | accuracy | Helped |
Closest papers by meaning
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.90)
- Bounding the Expected Robustness of Graph Neural Networks Subject to Node Feature Attacks (ICLR 2024 · similarity 0.90)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.90)
- A Simple and Yet Fairly Effective Defense for Graph Neural Networks (AAAI 2024 · similarity 0.89)
- If You Want to Be Robust, Be Wary of Initialization (NeurIPS 2024 · similarity 0.89)
- Adversarial Robust Generalization of Graph Neural Networks (ICML 2025 · similarity 0.88)
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.