Research map / Graph learning / GNN architectures
Information Obfuscation of Graph Neural Networks
Peiyuan Liao, Han Zhao, Keyulu Xu, Tommi Jaakkola, Geoffrey J. Gordon, Stefanie Jegelka, Ruslan Salakhutdinov
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Train ChebNet with GAL using the Wasserstein objective at λ=0.5 rather than the Bose and Hamilton baseline | ChebNet on Movielens-1M · Movielens-1M | Gender-F1: 0.667 -> 0.593 | Improved |
| Train ChebNet with GAL using the Wasserstein objective at λ=0.5 rather than the Bose and Hamilton baseline | ChebNet on Movielens-1M · Movielens-1M | RMSE: 0.874 -> 0.872 | Improved |
| Use the Wasserstein objective instead of Total Variation | GCN on CiteSeer · CiteSeer | Macro-F1: 0.785 -> 0.644 | Improved |
| Use the Wasserstein objective instead of Total Variation | GCN on CiteSeer · CiteSeer | AUC: 0.867 -> 0.814 | Hurt |
| Increase λ from 0 to 0.5 (ablation) | ChebNet-TV on Movielens-1M · Movielens-1M | Gender-F1: 0.692 -> 0.617 | Improved |
| Increase λ from 0 to 0.5 (ablation) | ChebNet-TV on Movielens-1M · Movielens-1M | RMSE: 0.852 -> 0.881 | Hurt |
| Increase λ from 0.5 to 4 (ablation) | ChebNet-TV on Movielens-1M · Movielens-1M | Gender-F1: 0.617 -> 0.505 | Improved |
| Increase λ from 0.5 to 4 (ablation) | ChebNet-TV on Movielens-1M · Movielens-1M | RMSE: 0.881 -> 1.280 | Hurt |
| Increase λ (ablation) | GAL on Movielens-1M · Movielens-1M | downstream task performance | Hurt |
| Increase λ (ablation) | GAL on Movielens-1M · Movielens-1M | adversary performance | Improved |
| Use the Wasserstein distance rather than Total Variation | GAL · Cora | training stability | Improved |
| Obfuscate neighborhood-level embeddings | GAL on Movielens-1M · Movielens-1M neighborhood attacks | adversary performance | Improved |
Closest papers by meaning
- Towards More Practical Adversarial Attacks on Graph Neural Networks (NeurIPS 2020 · similarity 0.88)
- Unsupervised Adversarially Robust Representation Learning on Graphs (AAAI 2022 · similarity 0.88)
- Value at Adversarial Risk: A Graph Defense Strategy against Cost-Aware Attacks (AAAI 2024 · similarity 0.88)
- Can Graph Neural Networks Expose Training Data Properties? An Efficient Risk Assessment Approach (NeurIPS 2024 · similarity 0.87)
- Rethinking Independent Cross-Entropy Loss For Graph-Structured Data (ICML 2024 · similarity 0.87)
- On provable privacy vulnerabilities of graph representations (NeurIPS 2024 · similarity 0.87)
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.
