Research map / Graph learning / GNN architectures
Characterizing the Influence of Graph Elements
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 |
|---|---|---|---|
| Remove edges with negative estimated influence to improve SGC | SGC · Cora, Pubmed, Citeseer | accuracy: 81.0 { $\pm$ 0.0} -> 82.0 { $\pm$ 0.0}; 78.9 { $\pm$ 0.0} -> 79.7 { $\pm$ 0.0}; 71.9 { $\pm$ 0.1} -> 73.7 { $\pm$ 0.0} | Helped |
| Remove edges with highest positive estimated influence | GCN using an SGC surrogate · Cora, Citeseer, Pubmed; edge elimination rates 1%, 3%, 5% | accuracy: Cora: 81.10% -> 77.3%, 74.2%, 72.8%; Citeseer: 70.07% -> 69.3%, 67.4%, 64.7%; Pubmed: 79.80% -> 69.3%, 65.2%, 64.1% | Hurt |
| Remove training nodes ranked by positive estimated influence | GCN · Cora, Citeseer, Pubmed; node removal rates 5%, 10%, 15% | accuracy: Cora: 81.10% -> 74.7%, 59.8%, 57.9%; Citeseer: 70.07% -> 69.5%, 65.5%, 56.1%; Pubmed: 79.80% -> 79.0%, 77.2%, 75.2% | Mixed |
| Decrease the L2 regularization strength | SGC influence estimation · Citeseer | Spearman correlation | Hurt |
| Remove multiple edges simultaneously rather than estimate their influence individually | SGC influence estimation · Cora | influence estimation accuracy | Hurt |
Closest papers by meaning
- Influence Functions for Edge Edits in Non-Convex Graph Neural Networks (NeurIPS 2025 · similarity 0.92)
- Influence Functions in Deep Learning Are Fragile (ICLR 2021 · similarity 0.89)
- Towards Robust Influence Functions with Flat Validation Minima (ICML 2025 · similarity 0.85)
- Feature Distribution on Graph Topology Mediates the Effect of Graph Convolution: Homophily Perspective (ICML 2024 · similarity 0.85)
- Multi-Label Node Classification with Label Influence Propagation (ICLR 2025 · similarity 0.83)
- Bounding the Expected Robustness of Graph Neural Networks Subject to Node Feature Attacks (ICLR 2024 · similarity 0.83)
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.