Research map / Federated learning and privacy / Differential privacy
Differentially Private Decoupled Graph Convolutions for Multigranular Topology Protection
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 |
|---|---|---|---|
| Use decoupled graph convolution instead of standard graph convolution | DP graph learning under edge GDP · Squirrel and Chameleon | test accuracy (%): 35.15 $\pm$ 0.47 -> 38.18 $\pm$ 1.48; 49.47 $\pm$ 0.88 -> 53.83 $\pm$ 1.11 | Helped |
| Use decoupled graph convolution instead of standard graph convolution | DP graph learning under N1 GDP · Facebook, Computers, Photo | test accuracy (%): 51.57 $\pm$ 0.28 -> 80.94 $\pm$ 0.27; 77.50 $\pm$ 0.20 -> 83.49 $\pm$ 0.29; 81.27 $\pm$ 0.31 -> 88.38 $\pm$ 0.44 | Helped |
| Use decoupled graph convolution instead of standard graph convolution | DP graph learning under N1 GDP · Pubmed and Cora | test accuracy (%): 85.72 $\pm$ 0.11 -> 84.33 $\pm$ 0.40; 64.29 $\pm$ 0.80 -> 59.98 $\pm$ 0.81 | Hurt |
| Increase the protected neighborhood granularity k | DPDGC · k-neighbor GDP | test accuracy (%): N1 to N5 to N25: Facebook 80.94 $\pm$ 0.27, 76.84 $\pm$ 0.36, 68.66 $\pm$ 0.32 | Hurt |
| Use a graph model rather than DP-MLP | DPDGC · cSBM with strong graph-topology information | test accuracy | Helped |
| Use a graph model rather than DP-MLP | DPDGC · cSBM with weak graph-topology information | test accuracy | Hurt |
| Use decoupled graph convolution rather than standard graph convolution | DPDGC vs GAP · edge GDP with small privacy budget ε | utility | Hurt |
| Use decoupled graph convolution rather than standard graph convolution | DPDGC vs GAP · edge GDP with ε≥4 | utility: ε≥4 | Helped |
| Replace standard graph convolution A H with decoupled A W^(A) | DPDGC vs GAP · k-neighbor GDP | required DP noise dependence on k: GAP: 4DLα/(2s^2); DPDGC: kα/(2s^2) | Helped |
| Replace standard graph convolution A H with decoupled A W^(A) | DPDGC vs GAP · node GDP | required DP noise dependence on maximum degree: GAP: 4DLα/(2s^2); DPDGC: 2Dα/(2s^2) | Helped |
Closest papers by meaning
- Going Deeper into Locally Differentially Private Graph Neural Networks (ICML 2025 · similarity 0.91)
- Poincaré Differential Privacy for Hierarchy-Aware Graph Embedding (AAAI 2024 · similarity 0.88)
- Can Graph Neural Networks Expose Training Data Properties? An Efficient Risk Assessment Approach (NeurIPS 2024 · similarity 0.87)
- GRAIN: Exact Graph Reconstruction from Gradients (ICLR 2025 · similarity 0.87)
- Chasing Fairness in Graphs: A GNN Architecture Perspective (AAAI 2024 · similarity 0.87)
- FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks (ICML 2025 · 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.