Research map / Graph learning / GNN architectures
Revisiting Pre-Propagation GNNs: Robust Diffusion Operators and Hidden-State Re-Propagation
Zichao Yue, Zhiru Zhang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add robust diffusion operator (ablation) | SIGN · pokec | test accuracy: 81.02 -> 83.10 | Improved |
| add robust diffusion operator (ablation) | GAMLP · pokec | test accuracy: 78.25 -> 81.28 | Improved |
| add robust diffusion operator (ablation) | HOGA · pokec | test accuracy: 82.11 -> 82.65 | Improved |
| add hidden-state re-propagation (HRP) (ablation) | SIGN with robust op. · pokec | test accuracy: 83.10 -> 83.85 | Improved |
| add hidden-state re-propagation (HRP) (ablation) | HOGA with robust op. · pokec | test accuracy: 82.65 -> 84.50 | Improved |
| add HRP (ablation) | GARNN with robust op. · pokec | test accuracy: 83.19 -> 84.63 | Improved |
| use HRP instead of no label strategy (ablation) | GARNN · roman-empire | test accuracy: 82.06 -> 85.53 | Improved |
| use HRP instead of label-as-input (ablation) | GAMLP · roman-empire | test accuracy: 80.75 -> 83.08 | Improved |
| add label-as-input (ablation) | GAMLP · roman-empire | test accuracy: 81.28 -> 80.75 | Hurt |
| add label propagation (ablation) | GARNN · roman-empire | test accuracy: 82.06 -> 82.62 | No clear effect |
| apply robust operators and HRP | SIGN · Computer | test accuracy: 84.2 -> 91.7 | Improved |
| apply robust operators and HRP | GAMLP · amazon-ratings | test accuracy: 52.2 -> 52.2 | No clear effect |
Closest papers by meaning
- A Generalized Neural Diffusion Framework on Graphs (AAAI 2024 · similarity 0.89)
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.89)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.88)
- Towards Pre-trained Graph Condensation via Optimal Transport (NeurIPS 2025 · similarity 0.88)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.88)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (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.
