Research map / Graph learning / GNN architectures
Diffusion Improves Graph Learning
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 |
|---|---|---|---|
| replace adjacency with sparsified graph diffusion (GDC, PPR or heat kernel) | GCN, GAT, JK, GIN, ARMA · CORA, CITESEER, PUBMED, COAUTHOR CS, AMZ COMP, AMZ PHOTO node classification | accuracy | Helped |
| apply GDC to the input graph | DCSBM, spectral clustering, DeepWalk, DGI · six datasets, clustering | clustering accuracy: up to 30 percentage points (spectral clustering) | Helped |
| sparsify the diffusion matrix (top-k / threshold) to an optimal degree (ablation) | GCN+GDC · CORA, CITESEER, AMZ COMP | accuracy | Helped |
| use random-walk transition matrix Trw instead of symmetric Tsym (ablation) | GCN+GDC · CORA, CITESEER, AMZ COMP | accuracy | Hurt |
| add self-loops to Tsym (ablation) | GCN+GDC · CORA, CITESEER, AMZ COMP | accuracy | No clear effect |
| learn diffusion coefficients (AdaDIF) instead of fixed PPR coefficients (ablation) | GCN, JK, ARMA + GDC · CORA, CITESEER, AMZ COMP | accuracy | Hurt |
| apply GDC | GCN · DD graph classification | accuracy: 2.5 percentage points | Helped |
| apply GDC | GNN · PPI inductive | accuracy | No clear effect |
| apply GDC | GNN · link prediction | link prediction performance | No clear effect |
| apply GDC at sparser label rates | GCN, JK, ARMA · CORA, 5-60 labels per class | accuracy | Helped |
Closest papers by meaning
- Adaptive Diffusion in Graph Neural Networks (NeurIPS 2021 · similarity 0.93)
- Simple Spectral Graph Convolution (ICLR 2021 · similarity 0.93)
- Learning to Approximate Adaptive Kernel Convolution on Graphs (AAAI 2024 · similarity 0.91)
- A Generalized Neural Diffusion Framework on Graphs (AAAI 2024 · similarity 0.90)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.90)
- SBGD: Improving Graph Diffusion Generative Model via Stochastic Block Diffusion (ICML 2025 · similarity 0.89)
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.